Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
7 Evaluation of Final Reports of Individual Studies INTRODUCTION THE STRUCTURE OF THE ecological monitoring program was segmented, rather than integrated, in that the research teams worked independently of each other. The purpose of ]:TTRI's solicitation in March 1982 was to attract sub- contracting researchers to develop and conduct separate ecological monitoring studies that would determine whether low-level, long-term electric and mag- netic fields (EMFs) and gradients produced by the ELF communications sys- tem would affect vegetation or wildlife in and near the system area or other- wise result in changes in individual organisms or their communities. Eventu- ally physiologic, developmental, behavioral, and ecological aspects of predom- inant organisms in upland, riverine, and wetland habitats near the Navy's ELF transmitting facilities were monitored for possible effects of EMFs produced by the Navy's antennas. The organisms and ecological relationships selected for monitoring were chosen because they were judged to be important to their ecosystems and to be of interest to local residents (Zapotosky and Gauger 1993~. The monitoring program studies were designed to compare data collected at control sites with data collected at treatment sites. As discussed in Chapter 2, the paired sites were intended to have matched environmental factors but to be dissimilar in the magnitude of their exposure to the 76-Hz EMFs gener 32
EVALUATION OF FINAL REPORTS 33 ated by the communications system antennas. Sites exposed to those 76-Hz EMFs were established by locating treatment sites near or within the rights-of- way for the antennas; control sites had to be far enough from the communica- tions system that EMF intensities resulting from antennas would be substan- tially lower than those at treatment sites. However, control sites had to be close enough to have environmental factors similar to those of their matched treatment sites. Siting criteria called for intensities of the 76-Hz EMFs at treatment sites to be at least 10 times those at control sites, for intensities of the 76-Hz EMFs at treatment sites to be at least 10 times those of the 60-Hz EMFs at treatment sites and control sites, and for intensities of the 60-Hz EMFs at treatment sites to be within 10 times those at control sites (Haradem et al. 19941. This chapter presents evaluations of the ~ ~ final ecological reports on the following topics: wetlands, slime mold, Wisconsin birds, Michigan birds, small vertebrates, litter decomposition and microflora, upland flora, aquatic ecosystems, pollinating insects, soil arthropods and earthworms, and soil amebas. The committee combined its discussion of Wisconsin and Michigan birds in this chapter. The committee used the following criteria to evaluate the reports: adherence to the original project proposal, adequacy of selection of species and response variables, adequacy of experiment design and imple- mentation (including biologic and ecological sampling techniques, physical measurements, and statistical power), responsiveness to reviewers' comments while studies were being conducted, presentation of results (including consid- eration of alternative analyses or hypotheses and interpretation), and appropri- ateness of conclusions (including validity and uncertainties). The committee found that the various criteria did not warrant the same amount of discussion for each study. WETLANDS PROJECT PROPOSAE The authors of the wetlands final report (Guntenspergen et al. 1989) pointed out that wetlands in the upper Midwest are sensitive ecosystems and are common near the ELF communications system sites, especially the Wis- consin location. They pointed out that past studies on effects of ELF EMFs indicated that plant membranes might be affected by electromagnetic radiation.
34 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM Therefore, they proposed to look for changes in plant competitive ability through measurement of plant or ecosystem functions that were related to membrane functions. Possible changes included leaf diffusion resistance, foliar nutrient content, changes in functions related to transport of water and nutri- ents across membranes, and decomposition by microorganisms, all of which depend heavily on secretion and adsorption through membranes. The basic null hypothesis of the wetland-monitoring project was that ELF EMFs resulting from the operation of the Navy antenna have no effect on selected ecosystem variables. SYSTEM, SITE, AND SPECIES SELECTION Much of the initial, pilot-study year of this project was spent in establish- ing study sites and testing methods. Stearns et al (1982) described five north- ern wetland vegetation types in Wisconsin: northern conifer swamp, shrub wetland, emergent marsh, northern sedge meadow, and open bog. Because the northern conifer swamp was common near the ELF location in Wisconsin and offered all life forms-including trees, shrubs, herbs and nonvascular plants it was chosen as the ecosystem type for monitoring. The authors of the proposal recognized the heterogeneity of the region near the ELF antenna, as well as the heterogeneity among stands of the same type of wetland. That created a problem in site selection but was addressed through identification of sites with close similarities in vegetational composi- tion and environmental characteristics. The former was determined through use of contingency tables and similarity indexes (such as Sorensen's Index), and the latter through measurement of soil-water temperature and pH, cation concentrations, and redox potential. in general, it was thought that sites with similar vegetation would have similar, but certainly not identical, environ- ments. More than 200 potential sites were selected from aerial photographs; these were reduced to SO sites for priority-setting, and then 15 sites were selected for measurement of 60- and 76-Hz EMFs in potential study plots, one per site. Eleven sites were eventually chosen for initial summer (1983) stud- ~es. Location of "similar" sites was based on relative EMF intensity. Electric-field strength measured by lITR] was used to establish intensity gradi- ents; these were called background, intermediate, antenna, and ground to correspond to location of sites relative to ELF facilities. No true control could be established, because the ELF communications system was already function
EVALUATION OF FINAL REPORTS 35 ing, so the background sites were considered as control sites. In most cases, three sites were used for each exposure scenario, but only two ground sites were used within one large peatiand. Although the selection of study sites satisfied the criteria established by IITR} and a gradient of exposure from treatment sites to control sites existed, the three treatment sites varied among themselves in terms of EMF intensity. For example, 1987 field measurements showed that the plots within the antenna treatment sites ranged from 0.053 to 0.196 V/m in electric-field intensity and from 6.! to 19.S mG in magnetic- f~eld intensity. Selection of the northern conifer forest wetland allowed use of tree, shrub, and herb species, when appropriate, for determination of response variables associated with plants. Final experimental species were not selected until 1985. Labrador tea (Ledum groeniandicum) was selected as the primary species for measuring stomata! resistance after other species were tried; for instance, the leaf anatomy of black spruce (Picea mariana) made stomata! measurement difficult. Labrador tea is a common shrub throughout the conifer wetlands of Wisconsin. Labrador tea leaves replaced pure cellulose sheets in the decomposition studies for testing of "natural plant materials" and to im- prove within-site measurement consistency. Black spruce, Labrador tea, the shrub leather leaf (Chamaedaphne calyculata), and the herb false solomon's seal (Smilacina trifolia) were used for foliar nutrient content; this permitted comparisons across life forms. Nitrogen fixation studies, initially on alder and then on moss and peat, were dropped during the study period. RESPONSE VARIABLES Response variables chosen for the wetland studies and used throughout the study period all were related to membrane-associated functions. The choice was based on a National Research Council (NRC 1977) report and other stud- ies on effects of EMFs that indicated that biologic and ecological responses to EMFs were most likely in functions associated with membranes. Stomatal resistance was chosen as a response variable because it is associated with water transport across membranes and all vascular plants could potentially be influ- enced through this leaf function. Nutrient content of leaves was examined because it is closely related to nutrient transport across membranes in root cells. Decomposition was examined because it is associated with across-mem- brane secretion of enzymes by microorganisms and adsorption of decomposed cellulose and other organic compounds. Other processes could have been
36 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM examined, such as growth rates or species composition changes, but they were not, because they were considered likely to result in too much variability within sites and were considered to be long-term response variables. Nitrogen fixation was examined for a couple of years, but that effort was eventually dropped from the program because the method was not reliable. EXPERIMENT DESIGN The wetland monitoring study was conducted from 1983 through 1987. The first year was used for selecting sites, developing and evaluating protocol, and beginning preliminary sampling. After experimentation with a rectangular plot design in the first year in which all the experimental measurements were made in a regular array throughout the plot, the study team settled on a 70 x 15-m rectangular plot for all experiments, oriented with the long axis parallel to the closest antenna. Six square subplots were designated in the rectangular plot with centers (where shallow groundwater wells were placed) 10 m apart. All measurements of response variables were taken within these subplots; all selected species were within each subplot. Environmental data were collected monthly from May to September (the frostfree period). ELF EMFs were measured once a year and assumed to be constant over the year. The antennas were capable of operating at full strength from 1985 through 1987, when most of the established protocols were in place; however, the antennas were not on full-time during this period, and it is not certain to what extent the off periods were taken into account in the study. That might be important for response variables that respond instanta- neously or in a short term. Biologic Sampling The final report presented three biologic measurements used to determine possible effects of ELF EMFs on wetland ecosystems in the vicinity of the Wisconsin transmitter facility (Guntenspergen et al. 1989~: stomata! resistance, foliar nutrients, and decomposition. Stomatal Resistance The primary results of stoma/al-resistance tests were based on responses of Labrador tea, a common species in every study bog. Other species were tested both because of ease of measurement and because
EVALUATION OF FINAL REPORTS 37 of response to different light levels. Spruce and smilacina were dropped be- cause of measurement difficulties. The equipment used for measurement of stomata! resistance was a null-balance diffusive-resistance porometer, which measures the rate of water-vapor diffusion through the stomata. It is standard equipment for such measurement, and the method is easily replicated if need- ed. Stomatal resistance was measured in leather leaf and Labrador tea under different light intensities. Labrador tea, least responsive to differences in light especially at low intensities, was selected as the species for testing the re- sponse of plant stomata! resistance to ELF EMFs. That choice essentially eliminated light as one of the independent environmental variables that might have to be considered a covariate in later statistical analyses. Selection of only one species, however, implied the assumption that all other species would respond to ELF EMFs in a similar fashion. if different species had different stoma/al-resistance responses to light levels, might this indicate different re- sponses to other external variables? That is partially addressed by the results, which include data on leather leaf, which was dropped as a consequence of tests on light intensities. Measurements were made during four periods at all I! study sites in August and September 1986 and 1987. Measurements were made over several days in each sample period, and there was an attempt to stratify measurements to cover all environmental variables, especially light intensity and cloud cover, for the background, intermediate, antenna, and ground sites. The number of samples taken was doubled between 1986 and 1987 to allow for resolving 20% differences in means at p = 0.05 with an 80% probability. The increase in samples increased variability in the measurements of external environmental conditions. That situation is commonly found in field sampling. An increase in sampling frequency might reduce the statistical significance of the results because it increases variability in sample measurements. Foliar Nutrients Changes in foliar nutrient concentrations in plants grow- ing in a relatively nutrient-poor environment were considered a possible indi- cator of the condition of various plant biochemical pathways. The initial spe- cies for foliar-nutrient sampling were a shrub (leather leaf), an herb (smilaci- na), two sedges, and a tree (black spruce). Labrador tea was substituted for the sedges because destructive harvesting of these sedge species might have damaged their limited populations. Sampling periods were based on phenol- ogy; thus, herbs were sampled earlier in the season than shrubs. Sample size was increased several times over the early years of the study to improve the
38 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM power of the statistical analysis, eventually reaching 396 (6 x 6 x ~ i) sam- ples per species per sample date. Only current-year foliar tissue was collected. Standard methods were used for preparation and analysis for calcium, magne- sium, and potassium. Samples from 1987 were also analyzed for manganese and phosphorus. For quality control, National Bureau of Standards (NBS) standards were analyzed with field samples, and spikes of known-cation stan- dard solutions were added to field and NBS samples. Decomposition Two approaches were used for decomposition studies. The first used pure cellulose as a substrate, and the second used Labrador tea leaves. The use of pure cellulose was intended to provide a uniform substrate for decomposition. If preweighed samples of cellulose were placed in a f~ber- glass bag and inserted vertically into the peat, less variation was expected among all samples. However, the cellulose became soft, adhered to the bag, and could not be retrieved fully for posttreatment weighing. Labrador tea leaves were used instead of cellulose. Bags with about the same amount of leaf material were mixed and randomly selected for placement in the bogs. The use of Labrador tea leaves resulted in less within-group variance. The decomposition bags were allowed several months for incubation in situ. The bags were retrieved, and the cellulose or leaf samples were removed, cleaned of foreign material, dried, and weighed. Weight loss indicated decom- position. The duration of incubation varied from 4 to 12 months. Environmental Characteristics The primary environmental characteristics measured during the study, excluding ELF-EMF exposure levels, were those of the interstitial water with- in each wetland site. The characteristics of the bog water were considered to influence decomposition rates and root activity. Shallow groundwater wells were placed in the center of each of the six subplots in the study plot at each site. Water-quality measures were depth to water table, depth to anaerobic zone, reduction-oxidation potential, specific conductance, temperature, pH, and calcium, magnesium, and potassium contents. Dissolved organic carbon was measured in 1984 only. Water samples were prepared with standard methods. No data in the wetlands report indicate regular measurement of ambient temperature, rainfall, or other external climatologic conditions. Measurements
EVALUATION OF FINAL REPORTS 39 of water quality, light intensity and leaf temperature made during various sampling periods appear to have been considered sufficient for evaluating the influence of the external conditions. Statistical Methods Statistical methods were chosen to test the null hypothesis that ELF EMFs resulting from the operation of the Navy antenna have no effect on selected ecosystem variables. Researchers primarily used nested analysis-of- variance (ANOVA) models to examine treatment and control groups. Stepwise-multiple-regression models were used to explain the variance in the dependent biologic variables (stomata! resistance, foliar nutrients, and decom- position rates) on the basis of environmental variables (such as water-quality data) and ELF-EMF data. Significance levels (p=0.05) of the two models were compared. In a few instances, the models did not agree, and lack of significance within one mode! was selected as the appropriate test of the re- sponse variable (i.e., it was not significantly influenced by ELF EMFs). Reviewers of the project over the period of the study sometimes questioned the use of stepwise multiple regression because of the interdependence of the variables, even those considered independent. In ANOVA, the variables were usually considered covariates; in multiple regression, they were treated as independent. Treating variables in this manner is a common practice in eco- logical studies, the understanding being that no variables in an ecosystem are truly independent. The use of stepwise regression attempts to alleviate this concern. To show the relationship of environmental variables to decomposition rates, all environmental data collected during the incubation period were sub- jected to principal-components analysis (PCA), an approach that reduces the number of independent variables to a few "composite" variables. The princi- pal components representing environmental data were then regressed (in step- wise fashion) against decomposition rates. Several times during the study period, the number of samples was in- creased to increase the power of the statistical analysis. That was done mostly for the foliar nutrient analysis. Other levels of sampling were considered sufficient, after preliminary studies, to achieve the established level of confi- dence at a 0.05 significance level.
40 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM Quality Assurance and Quality Control Quality assurance and quality control, especially as related to chemical analysis, were addressed in the methodology discussion of the final report. Methods selected for stomata! resistance were standard and repeatable. Decomposition-rate methods were straightforward. The research design estab- lished for this study helped to avoid pseudoreplication and allowed sound ANOVA of data from study sites. Exposure Assessment UTR} provided ELF-EMF exposure data at the background, intermediate, antenna, and ground sites. ITTR] measured ELF EMFs annually at each sam- ple plot at the study locations. The 1987 annual report (Guntenspergen et al. 1988) made it obvious that the investigators took into account the spatial EMF gradient resulting from antenna operation and used it in regression analyses. The ELF communications system was not operated continuously; study sites were therefore not consistently exposed to ELF EMFs. Different por- tions of the antennas were turned on and off several times each day, with varying modulations, frequencies, current intensities, and phase angles. The Navy provided ITTR} and researchers with detailed logs of antenna activity. The final wetlands report does not indicate whether the antennas were on or off during field measurements or whether information on the antenna operation was used in data analyses. Response variables would likely have varied in their sensitivity to antenna operations. For example, annual decomposition rates could be related to the annual EMF measurements, but stomata! resistance might have an immediate cellular response to external factors, such as light, temperature, and ELF EMFs. Use of short-term response variables, such as stomata! resistance, could have created analytic difficulties. Field measurements of these very short-term response variables would have had to be timed in coordination with antenna operations to ensure measurement of possible EMF influences that might not be observable when the antenna is off. In 1987, stomata! resistance was measured in August, the same month that I]:TRI measured exposure levels at each study plot. There is some evi- dence in the 1987 annual report (p. 17) that the researchers examined the on- off status of the Wisconsin transmitting facility relative to their field sampling.
EVALUATION OF FINAL REPORTS 41 If these measurements occurred at the same time, the exposure assessment of this study took into account the need to measure short-term responses while the antenna was actually on. The final report does not address the coordination of field measurements with antenna activity. If the measurements were not coordinated, the use of a short-term response variable, such as stomata! resis- tance, is questionable. RESPONSE TO REVIEW The researchers received comments from reviewers starting with the 1983 annual report (Stearns et al. 1984~. Common comments were related to site characterization, use of specific response variables and timing of measure- ments, and level of sampling for adequate statistical analysis. The study design was modified to accommodate some of the critics' comments; others were addressed through explanations of why changes were not made. For example, a criticism of site selection was lack of extensive soil data. The researchers characterized similar sites as those with similar peat sub- strates, but then emphasized measurement of interstitial water at each site as more appropriate for studying substrate composition. The use of stomata! conductance as a response variable was acceptable to the reviewers, but they were concerned that because this variable is so closely influenced by light and temperature, it needed to be measured at the same time of day under similar conditions of light and temperature to yield useful comparisons. The researchers selected the species with the least re- sponse to light variation because it was impossible to make all measurements at the same time of day or even on the same day. The use of foliar composition was questioned because it is not considered appropriate for determining soil nutrient availability in agricultural systems. The researchers pointed out, however, that foliar composition is commonly used in natural ecosystems as an indicator of plant condition and therefore was appropriate for this study. Although the final statistical approaches did not fulfill all the requirements of the reviewers, the study design was altered in some cases to increase sample size. Inappropriate uses of statistical terms, pointed out by reviewers, also were corrected in the final report. In most cases, the investigators did respond to reviewers' comments, thus producing an improved final document.
42 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM PRESENTATION OF RESULTS Alternative Hypotheses Results were presented in different forms: graphical presentations with bar graphs (often showing standard errors), tables with nested ANOVA analy- ses, and tables with stepwise-multiple-regression analyses. The initial proposal suggested that ANOVA would be the appropriate statistical approach to test the null hypothesis of no difference between or within treatment and control sites. Development of data sets also demonstrated to the researchers that there was a need to attempt to explain the variances of the biologic responses be- tween and within treatments, in addition to assessing them. Although no alter- native hypotheses were presented, use of stepwise multiple regression to assess the significance of independent environmental variables indicates possible consideration of an additional hypothesis that variances in dependent biologic responses are explained no more by natural external environmental factors than by ELF EMFs. Interpretation Interpretation of results was based primarily on comparison of the two statistical approaches to the empirical data. If the ANOVA models indicated that there was no greater difference between study sites than within study sites at the 0.05 significance level, the researchers interpreted that as conclusive. They checked their interpretation by applying the multiple-regression models that used environmental characteristics (sometimes considering biologic vari- ables as independent, such as leaf nutrients for stoma/al-resistance compari- sons) and ELF-EMF characteristics (in most cases in a PCA form). In many cases, neither environmental nor ELF-EMF principal components accounted for much of the variance of the response variable. in a few cases, a signifi- cant correlation for a response variable was found using the regression mode! that was not found using the ANOVA model. The researchers interpreted the mode! showing no significant correlation to be correct. That raises the question of the level of confidence selected to demonstrate statistical significance of the ANOVA and multiple-regression models. For this study, a significance level of 0.05 was chosen. Considering the variability of the ecosystems being stud- ied, this is probably appropriate, although many statisticians might consider it no more than an indication that the variance of the results depends on chance.
EVALUATION OF FINAL REPORTS How Well Researchers' Conclusions Were Supported 43 The general conclusion of the researchers was that ELF EMFs generated by the communications system had no measurable effect on the biologic func- tions used as surrogates of wetland ecosystem responses. The functions stud- ied were stomata! resistance, foliar nutrient content, and decomposition rates. All those functions are in some way influenced by across-membrane movement of water, nutrients, or enzymes. Using nested ANOVA to test the significance of variance in responses between and within treatments and the control, they concluded that the few occurrences of significance were at a level not much greater than would be due to chance (about 5% for this study). Using stepwise multiple regression to explain the variance of biologic responses, primarily to see whether ELF-EMF principal components might be among the most signif~- cant independent variables in the development of the coefficient of determina- tion, they concluded that environmental variables explained more of the vari- ance in the biologic-response variables than did ELF EMFs and not much at that. There were very few statistically significant ANOVA tests, and the multiple-regression analyses of these same variables were not statistically significant. The researchers did not attribute biologic importance to the results from any one test that showed a statistically significant response to ELF EMFs; they noted that such results are expected because of chance alone when many similar tests are performed, as was the case in this study. Instead, they looked for statistically significant effects attributable to ELF EMFs that were repeated over months or years. This appears to be an appropriate approach because response variables were statistically significant in few cases and were always the results of only one test Because the experiments were straightforward, essentially well designed, and properly analyzed, the conclusions were well supported by the data. Although none of the studies was simple, the methods were sound and repro- ducible and gave the researchers confidence in their conclusions. COMMITTEE CRITIQUE This study was properly designed from the beginning, although minor changes might be necessary to improve it. Some constraints were placed on the study because of prior antenna operations and complexity of the landscape. It also had built into its design the ability to modify the procedures when necessary to improve statistical power or to substitute a material that would produce more-consistent results. For example, the number of leaf samples
44 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM taken for foliar nutrient content was increased several times during the first few years of the study, and substitution of Labrador tea {eaves for pure cellu- lose represented not only use of a natural component of the ecosystem, in addition to a cellulose source, but also a material that gave more consistent responses. Initial establishment of study sites was compromised in that the Navy's ELF facility was already in operation at the time of the first response measure- ments, so no true controls could be established. It was never explained why the background locations, which were intended to represent the low end of ELF-EMF exposure, had to be near the antenna. Was that prerequisite estab- lished to ensure that ambient conditions would be similar for all sites? Were there no northern conifer bogs at some distance from the ELF antenna with all the same characteristics as those selected as exposure sites? The research- ers considered 200 sites on the basis of aerial photographic comparisons. How far from the antenna were some of these sites, and could some have been used as better "control" sites? Researchers justified their choice of response variables on the basis that they represented biologic functions that were associated with across-membrane processes, membrane functions being known to respond to EMFs. Other eco- system functions were not considered in the proposal or in continued develop- ment of the study over the first few years. A substudy on nitrogen fixation, also associated with membrane functions, was attempted but discontinued. For example, other growth processes might have been considered. In the decompo- sition study, results indicating statistically significant differences in rates of decomposition were apparently compromised by the differential growth of moss over the decomposition bags. Bags at sites with higher ELF-EMF inten- sities had almost total moss cover; those in the background were only partially covered. Although growth of moss was not an initial response variable, might not this phenomenon, albeit discovered well into the study, have triggered some controlled studies on growth of nonvascular plants in the bog? Because the moss-growth phenomenon was not controlled, there was no way for the study to address the potential relevance of this occurrence. The use of a short-term response variable, stomata! resistance, might have created an analytic problem relative to exposure assessment. The on-off cycle of the antenna might have influenced the temporal response of this vari- able, but antenna operations do not appear to have been factored into the analysis or interpretation. However, the spatial variance of exposure levels within treatment and control sites was taken into account. If biologic field measurements were coordinated with ELF-EMF exposure measurements (this
EVALUATION OF FINAL REPORTS 45 might have been possible in August 1987), then this study could have related specific exposures to short-term response variables. In light of the basic soundness of the design and analysis, the results, interpretation, and conclusions of this study are in all likelihood appropriate and acceptable. They have demonstrated that, according to the responses of the variables monitored, wetlands, an ecosystem characterized by water through which ELF EMFs readily pass, appear not to have been significantly influ- enced by ELF EMFs. SLIME MOLD PROJECT PROPOSAL The hypothesis tested in this program was that exposure to weak ELF EMFs generated by the Wisconsin antenna would induce observable physio- logic alterations in Physarum (a type of slime mold). Physarum cells were directly exposed to ELF EMFs generated by the Wisconsin antenna and to a laboratory simulation of the ELF EMFs at the Wisconsin ground terminal site. SPECIES SELECTION Physarum polycephalum was selected for two main reasons. First, the results of laboratory-based research had indicated that weak ELF EMFs, similar to what might be associated with the Navy ELF communications sys- tem, had effects on Physarum. Exposure to such fields had been reported to lengthen, reversibly, the cell cycle and to lower, reversibly, the respiration rate and the adenosine triphosphate (ATP) concentration of the cells. (Brayman et al. (1985) attempted to replicate the results of the original laboratory study but could not. The committee was not asked and did not attempt to determine why.) Second, the natural habitat of Physarum is the forest floor, where it functions with other organisms in recycling organic material, so its relevance to the ecological monitoring program is clear. Both of those considerations point to this particular study as one of key importance in the program. Although the research did not involve ecological monitoring itself, effects of ELF EMFs on the variables measured could have important ecological implications. Among the studies in the program, this appears to be the only one in which axenic cultures of a single organism were used. The use of
46 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM axen~c cultures has advantages and limitations. Substantial conclusions about ecological effects cannot easily be made on the basis of a single organism, but effects of an environmental variable are more easily identified and clearly attributable when the study involves a single organism. SELECTION OF RESPONSE VARIABLES Laboratory studies of possible ELF-EMF effects on slime mold initiated in 1972 had included three measurements: length of the mitotic cycle, rate of respiration, and ATP content. Mitotic-cycle measurements were included in the original proposal and in experimental results reported up to the 1986 annu- al report, but were eventually replaced with the ATP determination. Outside data had suggested that ATP content might be a more sensitive indicator of an ELF-EMF effect. Early reports indicate substantial effects on the length of the mitotic cycle, but, according to the 1987 annual report (Goodman and Green- baum 1988) an increase in the length of the control cell cycle caused research- ers to examine their handling procedures. In response to questioning by the committee (E. Goodman, University of Wisconsin-Parkside, memorandum, 1996), the principal investigator stated that "the onset of an effect required at least 120 days . . . [and] . . . another 60 days of experimentation was usually performed." Because of "the abbreviated season at the Wisconsin Test Facil- ity, this type of experiment wasn't deemed to be feasible." According to the final report (Goodman and Greenebaum 1990, p. 41), "weather conditions in northern Wisconsin prevented experiments from being carried out for more than 140 days" because cultures would not grow in the cold temperatures from October through May. Only respiration and ATP content were included in the final report. Both respiration (in microliters of oxygen consumed per minute per milligram of protein) and ATP content (in nanomoles per milligram of protein in extracts made in Tris-borate buffer, pH 9.2, 9SC°) were deemed suitable for the desired investigation and were examined in the laboratory and at study sites. Those choices are justified by virtue of the earlier studies, in which effects on the two variables were reported and for which the methods were well established. in addition, they represent basic biochemical and physiologic processes at the cellular level, and any statistically significant effects would have undeniable ecological implications. The same could be said for effects on the mitotic cycle, if they had been included.
EVALUATION OF FINAL REPORTS EXPERIMENT DESIGN AND IMPLEMENTATION 47 The use of axenic cultures required an experiment design for the study sites quite different from those of the other projects. Pure cultures were main- tained and exposed in polyethylene chambers (7 x 4 x 2.25 in.), themselves placed individually in protective boxes in a hole about 20 in. on each side covered by plywood to protect the system from foraging animals. Even when placed in the ground near the transmitter, the cells inside the chamber did not receive an ELF-EMF exposure comparable to that which would be experienced by cells free in the soil at the same location. To simu- late the ELF EMFs, two stainIess-stee} electrodes, placed 6 in. apart and about 0.25 in. from the bottom of the polyethylene chamber, were connected to copper collector plates buried in the ground about ~ m from each hole along a line with the predominant electric field. The procedure also exposed the cells to the environmental temperature and to some extent other environmental factors- for example, humidity and barometric pressure but not others, such as soil moisture. Two duplicate cultures (chambers) were maintained and exposed at each site; the second served as a backup in the case of contamination of the first. Biologic Sampling Techniques Samples were placed at three sites: one control site 7 miles east of the nearest antenna element, one treatment site near the west ground element, and one treatment site near (about 30 ft from) the overhead cables of the antenna (outside the right of way). Cells were allowed to grow on a solid (nutrient agar) medium in the chambers provided. A portion of each exposed culture was inoculated onto fresh medium each week (subculturing to maintain continuing exposure) at a study-site laboratory. The remaining part of the exposed culture was trans- ported by air to the laboratory at the University of Wisconsin-Parkside, where measurements were made according to the same protocols as used for cultures exposed in the laboratory. Biologic measurements involved the cellular amounts of ATP and respira- lion rates. ATP in cell extracts was measured in cell extracts with the firefly luciferase method, which is highly specific and reliable. Respiration was mea- sured with an oxygen electrode, which is also reliable.
48 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM On return to the laboratory, macroplasmodia from each chamber were scraped from the agar surface, placed in 125-mL ErIenmeyer flasks containing 25 mL of half-strength growth medium, and incubated overnight with shaking. The final report states that, after washing and resuspension in full-strength medium, ATP and respiration rates were measured, usually within 40-72 h of removal from the exposure sites (Goodman and Greenebaum 1990, p. 5~. A similar delay occurred with the cultures exposed in the laboratory, but it was about 9 h less because transportation time was eliminated. Because the effects of ELF EMFs on this organism previously reported were stated to be reversible, the protocol, which involved maintenance of cells for 2-3 days before analysis, seems to compromise a firm conclusion. If the EMF has only a small effect, ATP content and respiration rates might well return to control levels during the time of study. In response to committee questioning (E. Goodman, University of Wisconsin-Parkside, memorandum, 1996), the principal investigator confirmed that the effect is reversible but stated that "our data suggest that at least 3-4 weeks is required for an effect to dissipate." The final report cites earlier work by the same researchers which found that the lengthened mitotic cycle returned to control levels after 3-4 weeks. However, it seems unlikely that the perturbation of a metabolite like ATP, which turns over rapidly in the cell, would persist long after the condi- tions are changed. In any event, it seems that such a problem could have been circumvented by freezing the cells immediately after removing them from the chambers and measuring their ATP content later. When asked about this, the researcher stated (in the same memorandum) that the 2-day period was needed to convert the exposed cultures from macroplasmodia, which were "not conducive to the types of biochemical tests we performed," to microplasmodia. Furthermore, because "the same cultures were used for ATP analysis as well as [respiration] measurements, freezing the ceils did not appear to be a viable option." Respi- ration measurements are made with living cells, and freezing could well have resulted in some damage that would compromise respiration measurements. That would not be so for ATP, however, because it is a chemical substance. In our opinion, the design of the experiments is problematic. For the procedure to be validated, an experimental control should probably include a repeat of the investigators' earlier reported effects of ELF EMFs on ATP content and respiration rates in Physarum showing that the effect did not degrade. To be sure, the protocol for the earlier experiments might also have included the long delay, but it is not mentioned in the final report. The com
EVALUATION OF FINAL REPORTS 49 mittee wonders whether the variability, which is repeatedly alluded to, might be due in part to the variation in the time from exposure to measurement, assuming that recovery is taking place during that time. Physical Measurements Other than ELF-EMF exposure, the only physical characteristic measured was temperature. Temperature was estimated at each site by placing a battery- operated Dickson monitor, calibrated in the laboratory, inside the protective box of one of the cultures at each site. These monitors are accurate to within one degree Fahrenheit and operated satisfactorily except when water got into the chamber. The mean weekly temperature was calculated by averaging the daily high and low at each site; soil temperature was also recorded hourly by TTTR] starting in July 1987. Statistical Methods Data on both study site cultures and cultures maintained and exposed in the laboratory were evaluated with analysis of variance (ANOVA) techniques. The independent variables used for laboratory data included Replication of measurements. Time in microplasmodia suspension culture. Intensity of EMF exposure. Time of EMF exposure. For study site data, three independent variables were used for the ANOVA: Exposure type. Exposure site. Time of EMF exposure. Study site data were also analyzed using multivariate linear regression; values derived from hourly measurements of the electric-field component of the ELF EMFs, current density, and temperature were included in this analy- sis. For each site, these data were averaged over the periods between meta- bolic measurements. A multivariate linear-regression procedure was used to predict ATP content and respiration rate as functions of the electric-field
50 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM component of the ELF EMFs, current density, and temperature. The period of EMF exposure was included as a predictor variable. Pearson moment corre- lations were calculated. Those methods are probably as good as any alternatives. Exposure Assessment Because the currents and ELF EMFs in the individual containers, both in the laboratory and at the study site, were adjusted to mimic conditions at the soil surface, the measured exposures are probably known better for this project than for many of the others. At the study site, the exposures resulted from the operation of the Wisconsin transmitting facility; exposures are well documented in the final report. in laboratory experiments, no attempt was made to duplicate any of the fluctuations in ELF-EMF levels that occurred at the antenna and ground treatment sites. PRESENTATION OF RESULTS The hypothesis tested was that exposure to weak ELF EMFs like those produced by the Wisconsin antenna would result in observable physiologic alterations in Physarum. The results are presented in the final report in seven tables and 22 figures with commentary that guides the reader to the conclu- sions and interpretations provided by the authors (Goodman and Greenebaum 1990, pp. ll-40~. No statistically significant differences between EMF-ex- posed cells and control cells in either study site or laboratory experiments were observed. Tables ~ and 2 of the final report (pp. 12-13) consider the effect of the amount of time that a culture was maintained out of the ELF-EMF environ- ment before analysis, which is related to the issue discussed above the long time between termination of exposure and measurements. Table 2 indicates that the mean value for respiration decreased with the number of hours elapsed since exposure to ELF EMFs an apparently statistically significant effect but no effect on ATP content was found. A concern with the results presented is that Table 2 lists the results of analyses performed after 4S, 72, and 96 h in submerged shake culture. if there is a degradation of an ELF-EMF effect, the largest part of it would be expected in the first 24 h. At 48 h, the investiga- tors would be looking only at the tail of that phenomenon. Another worry is
EVALUATION OF FINAL REPORTS 51 that the effect seen, that respiration decreases with time out of the ELF EMF, would mean that the supposed treatment effect is to increase respiration the opposite of what was reported in earlier laboratory experiments (exposures at higher field strengths). The authors state that, because of the observed decrease in respiration with time out of the exposure environment, only the 48-h results were used in later statistical analyses. That is suggested as eliminating concern about the issue, which it might not do. in addition, for study-site cultures, it is hard to reconcile that declaration with the description in the methods section, which says that the "analyses were performed within 40 to 72 hours after removing the cultures from the Wisconsin Test Facility." How can the shorter time allow for 48 h in liquid suspension? For the laboratory cultures, there were unexplained statistically signifi- cant differences in the measured values of ATP and respiration rate from year to year. All data before 1985 were omitted from consideration, as is reason- able because the antenna operation was intermittent. But data thereafter also showed differences from year to year for unknown reasons. Study site data on respiration (but not ATP content) are stated to vary significantly with time, and this is discussed by the authors on the basis of aging cultures. There is no comment on how values differ from year to year, but the conclusions section (p. 42) states that "the field and tIaboratory] values obtained for Physarum's [rate of respiration] and ATP were similar to those we have previ- ously published." The authors conclude that there is no indication that the ELF-EMF expo- sures used resulted in any "extensive big-effects" (p. 41~. They then discuss why this was so if their earlier studies did show an effect. Differences cited include the earlier studies' use of submerged (liquid) cultures for longer times (more than 180 days whereas the current study kept them on agar for about 140 days) and other differences in culture methods. Most important, it seems, is that the original laboratory experiments were performed at intensities 5-10 times higher than those in the current study. CONCLUSIONS AND COMMITTEE CRITIQUE Validity These experiments were carried out by an experienced and dedicated team of scientists. The organism selected has a unique place in the evaluation
52 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM of ELF-EMF effects: it has already been reported as one in which physiologic changes occur after exposure. The research team was expert in the mainte- nance of cultures and measurement of biologic changes. They worked closely with the team of engineers in setting and recording exposure levels. Uncertainties The committee finds no uncertainties in the collection of data, but has a question about exposure and how the data are analyzed in relation to it. The committee also has three major concerns about the design of the experiments. First, as described on p. 25 of the final report, "there are some dramatic differences in the field intensities during weeks 8-9, ~ l-13 and 17.5-~.5. " These were not attributable to the antenna, which was fully operational. On being questioned and in the final report itself, the principal investigator indi- cated that the reason for the variation was not known but speculated that it might have been due to any of several factors. Such might include differences in exposure cells, differences in the setting of the control potentiometer, changes in growth cell conductivity due to growth of the mold, and local changes in soil conductivity (E. Goodman, memorandum, 1996~. The data on times of high intensity do not appear to be treated separately from those on low intensities. Might some different conclusion have been reached if they had been treated separately? Also, exposure data are plotted starting in the middle of week 6, whereas ATP content levels are shown starting at week 2. What exposure values were used for those early weeks? Second, if ELF EMFs have an effect on the respiration rate of cells and on cellular concentrations of ATP and the effect is reversible (as it is stated to be on the basis of earlier results when effects were recorded), the return to the original state should begin at the time that exposure to ELF EMFs ends. In going from state A to state B in a mode! system, the rate of change usually is maximal at the outset and the final state is approached asymptotically. Supposed effects in this system are not measured until much later-between 40 and 72 h after the exposure is stopped, according to the methods section, or after 4S, 72, and 96 h in liquid suspension culture, according to the results section. That compromises the conclusions substantially. The authors do not discuss or refer to the fact that ATP is a product of respiration and that the two are thus expected to be related. Third, ELF-EMF intensities only 0.1-0.2 times those previously shown to have an effect were used. As already noted, the use of this organism was
EVALUATION OF FINAL REPORTS 53 of special value because of the earlier studies, but those experiments were not duplicated. However, the authors still refer to them as valid and correct. It can be argued that the objective of the monitoring program was to determine whether the emission of the Wisconsin facility would cause any physiologic effects and that higher intensities were therefore not relevant. Such a view is invalid; if there is an effect at a higher intensity, the most-important objective should be to determine the nature of that effect, from which one can evaluate the possible effects at the lower intensity. Moreover, it is widely agreed that there are variations between individual organisms and between species. Some test subjects are killed by doses considerably lower than the dose at which 50% of the subjects die (LDso), and there are species differences. Slime mold might not respond to fields only 0.! times those of the original study, but some other species might be responsive at such fields. Any effect at any dose is of potential interest because it could lead to an understanding of mechanism. Another consideration is that in experimental science one of the major techniques for establishing cause and effect is to determine dose-response relationships, but this was not done here. The researchers have measured only one point on a putative curve- an unsatisfactory experimental approach and the response to ELF-EMF exposure reported earlier was not replicated. In the ecological monitoring program, the technical aspects of imposing fields were presumably well managed, so it would be good to verify the earlier results and proceed from them. That has not been done, and without a dose-response relationship it is difficult to say that there is no effect. WISCONSIN BIRDS AND MICHIGAN BIRDS PROJECT PROPOSAL The Navy's original monitoring plan called for research on bird popula- tions that focused on potential effects of ELF EMFs on bird migration and orientation (lITR] 1976, pp. 37-411. Those subjects were given priority be- cause existing data indicated that at least some birds use geomagnetic fields for orientation and navigation. Several kinds of monitoring programs were sug- gested. The two given the most attention were radar tracking to assess the number and trajectory of birds migrating near the antennas, from which infer- ences could be drawn about whether birds are disoriented by antenna-gener- ated EMFs or actively avoid them, and conventional ground counting to deter- mine whether breeding migratory birds avoid ELF EMFs or recruitment of
5 4 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM new breeding birds is reduced over the years because of effects on reproduc- tive success. An additional suggestion was longitudinal studies of marked individuals to make inferences about effects of the antenna on survivorship, reproduction and site fidelity; this approach was never seriously pursued because it requires impractical sample sizes. Initially, ITTRT accepted a proposal (dated 1982-1983) that emphasized radar tracking to determine effects on migrating birds. The proposal also included some ground counting. Peer-review comments in 1984 criticized the design and execution of the project, and it was terminated. A proposal from G. I. Niemi and I. M Hanowski, of the University of Minnesota, Duluth, dated May 1984, was accepted as a replacement. It em- phasized ground counting of birds. The investigators proposed to use ground censuses along sample transects to determine whether there were differences in species richness, relative density, and relative frequency of breeding birds between treatment areas adjacent to the antenna and control areas away from the antenna. The proposal recognized that four factors needed to be accounted for in an analysis of variance in various abundance measures: habitat type, region (Michigan or Wisconsin), ELF-EMF intensity, and right-of-way (ROW) clearing effects. Region was handled by separate analyses in Wisconsin and Michigan, as is appropriate. Habitat type was handled in Wisconsin by statistical models. In Michigan, attempts to match habitats on treatment and control sites resulted in all the control sites' being clustered to the southeast of the antenna. That creates what is known as lack of interspersion and results in a statistical flaw known as pseudoreplication (see Chapter 4 for a more-detailed discussion). ELF-EMF intensity was used to establish treatment and control sites; but in later analyses of the response variables, year served as a surrogate of ELF- EMF intensity. EI-F-EMF intensity varied within and across years, so the power of statistical tests is difficult to assess. In dealing with ROW clearing effects, the design gambit was to eliminate "Ethe ROW] factor from the sam- pling design by eliminating this zone of habitat [near the ROW] from sam- pling." (Niemi and Hanowski 1984, p. 2~. Treatment sites were 125 m from the edge of the ROW clearing and hence 150 m from the antenna. Conse- quently, the high-intensity treatment sites had magnetic field intensities less than 0. ~ times those under the antenna. The switch in project teams shifted the focus of bird population studies from direct, immediate impacts on bird navigation and migration to indirect, potentially delayed impacts on local bird abundance.
EVALUATION OF FINAL REPORTS SPECIES SEEECTION 55 The investigators were not intentionally selective about bird species. They counted all birds detectable from line transects, using conventional tran- sect methods. The approach is well-established, standard procedure. The technique is most sensitive to conspicuous birds, like singing adult males, and least sensitive to inconspicuous birds, like silent, subordinate young or females on the nest. Although different species, sexes, and age classes are differen- tially detectable and habitat differences influence detectability, it is unlikely that these considerations would contribute to artifactual treatment effects in the statistical analysis of this particular data set. RESPONSE VARIABEES The response variables chosen were abundance indexes of individual species and aggregate indexes of species richness. Local abundance results from the integration of demographic processes (birth, immigration, death, and emigration), some of which can operate on fairly long time scales in long-lived organisms. These abundance indexes are basically sound for assessing effects that involve birds' perception and avoidance of EMFs. The indexes would have little power to detect small demographic changes with effects on abun- dance that involve long lag times. For instance, if adult breeding birds were recruited primarily from locally produced offspring and nestling production decreased by around 10%, the effects on bird abundance would be severe in the long run but might not be detectable by the end of the study, because of the slow turnover of adults. Other projects on nesting success of tree swal- lows (see "Small Vertebrates" below) are aimed at these kinds of effects. Another possibility is that demographic rates compensate to maintain local abundance. For instance, birds that die or depart after long exposure might be replaced by cryptic subordinates ("floaters") from the surrounding area. If the EMF changed the treatment sites into demographic "sinks" without dramatically lowering population densities, that would be missed. EXPERIMENT DESIGN AND IMPEEMENTATION The method of walking along transects and counting birds by sight and
56 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM sound is well established. The investigators used great care and sophistication. Daily and seasonal timing were appropriate to the objectives. The location of the treatment transects, 150 m from the antenna, is problematic. It is well known that ROW clearing through forest habitat affects bird populations, but the exact mechanisms and magnitudes over distance from the edge are ill understood. EMF strength and ROW effects on habitat are greatly confounded. The treatment transects were placed parallel to the antenna at a distance of 150 m ("near" the antenna) in the hope that this would "allow us to separate effects of electromagnetic fields on bird species and communities from effects due to direct habitat changes along the ROW" (Hanowski et al. 1991, p. l). The investigators chose to have the treatment transects at 150 m with the knowledge that magnetic fields at that distance were still more than 10 times as high as magnetic fields at control transects and in the hope that any unmeasured ROW edge effects were negligible. The 10: ~ criterion was established by ITTRI. An unfortunate consequence of that gambit is that the magnetic field in the air at the treatment transects, although more than 10 times that of the control transect, is less than 0. ~ times the magnetic field under the antenna (see Table G-6 of the 1989 engineering report, where magnetic flux at 150 m is less than 5% of the flux under the wire). Figure 3-! shows the distance decay function for magnetic flux. The treatment transects are off the hump and out on the tail of the magnetic field. It is apparent that, given the TTTR] 10: ! criterion, the treatment transects at 150 m could just as well be called control transects for a hypothetical treatment transect under the antenna. Although they could often detect birds up to 100 m from the transect, the investigators' sensitivity must have been relatively low at ranges over 50 m. Consequently, comparisons between treatment and reference transects are testing for effects of the combined ELF EMF and ROW clearing at roughly 50-250 m (more probably 100-200 m) from the wire. Either effect, if it ex- ists, should be greatly diminished at this distance. In principle, ROW edge effects and ELF-EMF effects could be decoupled temporally in Michigan, where there are data on conditions after ROW clear- ing but before antenna activation. Treatment transects could have been placed nearer the antenna to determine the incremental effects of the field, beyond ROW effects. Even so, if ROW edge effects developed slowly, they would still be confounded with the onset of antenna operation. Proper design would also require the establishment of matching ROW effects at the control sites. There is no easy solution to the problem. At this point, it is essential to make explicit how the experimental design limits the interpretation of the
EVALUATION OF FINAL REPORTS 57 results. The generally negative results do not show that EMFs associated with the Navy's ELF antenna have no effect on bird populations. They indicate no consistent, measurable effects beyond roughly 50-100 m, where magnetic fields in the air have dropped to less than 10% of their magnitude under the antenna. Although control sites were selected by a seemingly adequate randomiza- tion procedure, all the control sites in Michigan were southeast of the antenna and treatment sites (Hanowski et al. 1994, Figure I). That might have re- sulted from efforts to relocate control sites so that they would better match the habitats found on the treatment sites. In any case, treatment and control sites were not interspersed. Less important, treatment transects were oriented mostly north-south and control transects mostly east-west. In Wisconsin, treatment and control sites were not interspersed, but the controls at least surrounded the antenna, and compass orientations were more varied within groups. The inability to intersperse treatment and control sites "at random" raises subtle questions about pseudoreplication, but it does not seem to lead to obvious misinterpretations of the results. Physical Measurements Seemingly adequate measurements of habitat structure were used to establish statistical control (Wisconsin) or to match habitats between treatment and control sites (Michigan). Physical factors, such as microclimate, were assumed to be randomized across sites. Statistical Methods The Investigators used repeated-measures analysis of variance (ANOVA) and analysis of covariance (ANCOVA) (to incorporate habitat differences in Wisconsin) on species abundances and aggregate indexes over time on treat- ment and control transects. The between-subject factor was location (treatment versus control), and the within-subject factor was year. It was expected that abundance indexes would fluctuate from year to year. The main question was whether these fluctuations were different on treatment and control sites. Sta- tistically significant interactions between factors would indicate that changes in bird abundance over time were not similar in treatment and control areas. The plan was to use the technique of multiple contrasts to incorporate informa
58 mO .a' ~ ~ C,) 0 ~m .= to ~ I c' a' I ~in m-- c ~ ~ I: in C E .- = of .E l OQ L o LO Cad U) o o em, E / Ct a~ Ct ._ o ._ LL O ~ Ln ~ CO 4 - f~ ° m ~_ o ._ Q cn o o LO ~ ._ o .C~ , ~ _ ~: ~4 C~ o ~._ 1D 0 lD 0 6 ° LO O C~ (ssne6!11!w) 1uaJJn~ euua~u~ ~o saJadw~ OOl Jad A~!suea xnl~ o!~oU6eW a, o · _ Ct S~ Ct ~q o 4 - Ct o ._ o o S~ ;^ c~ · - c~ · - · o~
EVALUATION OF FINAL REPORTS 59 lion on antenna power levels if the screening based on years turned up statisti- cally significant interactions. The strategy of screening for treatment-by-year interactions in an ANOVA (with repeated measures over years) seems reason able, but the somewhat erratic pattern of antenna power within and across years raises concerns. Information on ELF-EMF strength was not used effec- tively. Changes over time unrelated to antenna operation, such as localized drought or localized insect populations, could lead to false-positive results (type ~ errors) if they occurred at either the treatment or control site, but not both. The possibility is increased by the lack of interspersion between treat- ment and control sites, and it seems to have happened in Michigan (see, for instance, Figure 6 in Hanowski et al. 19941. Several statistically significant treatment-by-year interactions are dismissed because they appear to be unre- lated to antenna activity. False-negative results (type TT errors) could result if variations in power make year a weak surrogate for ELF-EMF strength. Inspection of the data suggests that this is not a problem with this data set. Some sort of formal before-and-after, control-and-intervention (BACH) analysis would be more appropriate. Although the investigators conscientiously estimated the power of their statistical tests, it is impossible to estimate their power accurately because of the somewhat messy relationship between year and field strength and the pseudoreplication problem. Quality Assurance, Quality Control, and Replicability The quality-control and quality-assurance efforts seem to exceed the general standards for this sort of work. Peer reviews were generally very positive about the yearly reports, and the investigators addressed the few questions that were raised. Questions about habitat differences across treat- ments in Michigan resulted in relocation of control sites to match treatment sites better (investigators attempted to deal with this problem in Wisconsin statistically, by using ANCOVA to incorporate habitat differences explicitly). The severe lack of interspersion might have been due in part to this relocation; investigators might have had to cluster their control sites in this relatively restricted area to find habitats that matched the treatment sites. It is Impossible to gather this sort of observational data "blind." One has to trust that investigators do not consciously or unconsciously introduce biases. These investigators have made a conscientious effort to reduce the risks by using experienced field workers and rotating them between treatment and control locations.
60 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM Exposure Assessment TTTR! made adequate spatial maps of ELF EMFs at full-power antenna operation. It is impossible to determine the exposure levels for highly mobile adult birds. The relationship between the magnetic-field strength and distance from the antenna (as shown in Figure 3-~) means that the average exposure is higher than that expected at the average distance, but average distances for actively foraging adults are unknown. The hypothesis tested was not about exposure effects themselves, but about changes in bird abundance in the space near the antenna, where ELF EMFs are high, at least on the average, com- pared with that in space farther away. Although it is not essential that individ- ual dosages be known, better spatial maps of average fields integrated over time would have been useful. RESPONSE TO REVIEW UTR! responded to the critical reviews of the initial project by canceling it. Peer reviews of the replacement project were generally very positive. Reviews of the first annual reports from this project expressed concern that ROW effects on habitat were not eliminated. None of the peer reviews ex- pressed concern about the low intensities of the magnetic field at the treatment sites. PRESENTATION OF RESULTS The reports are very well written, facilitating critical review. With the few exceptions already noted, each decision is clearly explained and justified. The analyses and conclusions are explained clearly. Reasonable alternative hypotheses are raised and adequately considered. CONCLUSIONS Many bird species and several aggregate community indexes were com- pared across locations and years and subjected to statistical tests. About as many comparisons were statistically significant at the conventional level of 0.05 as one would expect to occur by chance. In addition, apparent false- positive results were considered case by case and dismissed. The investigators
EVALUATION OF FINAL REPORTS 61 interpreted the results reasonably: "We found no convincing evidence that overall breeding bird distribution or abundance was affected by electromag- netic fields produced by the ELF antenna" (Hanowski et al. 1994, p. 27~. That is a reasonable interpretation, as long as it is kept in mind how far from the antenna the investigators were looking for effects and the potential insensi- tivity to demographic impacts that might take a long time to become detectable in abundance effects. The conclusion should not be generalized beyond the low magnetic-field strengths found at the distances at which censuses were taken (less than 10 mG and less than 10% of the value beneath the antenna). An analogy might help to reinforce this message. Imagine that the magnetic-field strength function shown in Figure 3-! was actually a cross section of a hillside and that one is interested in knowing whether the hill had an effect on the density of birds in the landscape. Then, imagine that we found that the density of birds living t00-200 units from the peak was about the same as the density more than 1,000 units away. One could not conclude that the hill has no effect on the population, but, one could conclude that no effect on the index of abundance is apparent across low elevations. One could also reasonably conclude that there are not widespread effects over the land- scape, at least in the short term. Any effects that do exist, even if locally severe, are concentrated in a relatively small area. There are always questions about how far results like these can be gener- alized. The goal of the project was not to discover generalities about birds and EMFs, but rather to assess the impacts of particular transmitting facilities on bird abundances. It is always possible that even at the ELF sites seemingly small effects on populations, mediated through modest reductions in reproduc- tion and recruitment, will slowly compound and only later become apparent. At present, there is no evidence of statistically significant, widespread, short- term impacts of EMFs associated with the ELF antennas on bird populations. The studies would not have detected any effects within roughly 50-100 m of the antennas. SMALL VERTEBRATES PROJECT PROPOSAE This project attempted to ascertain whether the operation of the Navy's ELF antenna system in Republic, Michigan, had effects on small vertebrates chickadees, tree swallows, chipmunks, and deer mice in an ecological con- text. It is clear that even a partial answer to this question could have exhaust
62 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM ed all the resources allocated to the entire program. The investigators used their scientific judgment and the state of the art in research on EMF biologic effects to select four species and a few important variables to examine, and they divided their overall effort into six substudies that were completed: · Fecundity, mortality, growth, and maturation in tree swallows. Growth and maturation in deer mice. · Homing in tree swallows. · Homing in small mammals. · Development in tree swallows. Maximal aerobic metabolism in chickadees and deer mice. Three other planned substudies were eliminated in March 1989 because of budgetary constraints associated with a negotiated increase in the salaries of nonfaculty employees: on small-mammal communities, small-mammal par- ental care, and tree swallow incubation. FECUNDITY, MORTALITY, GROWTH, AND MATURATION IN TREE SWALLOWS This large substudy examined several aspects of reproduction in tree swallows, including number of eggs per clutch, hatching success within a clutch, fledgling success, rate of growth and probability of mortality of eggs, young, and nests. It was a generally well-conducted substudy with meticulous attention to experimental detail. Some of the variables were direct measures (such as, egg weight and volume), and others were indirect measures that resulted from fitting the data to a mode! function and then comparing fitted parameters in the mode! function across conditions. The substudy also included a creative nestling-swapping experiment (in 1990 and 1991 only) to separate the effects of EMF exposure during the period in ovo and compare it with the effects of exposure of nestlings. A number of control and sham permutations were included in the designs. The investigators concluded that no effects were consistently attributable to the operation of the ELF antenna system. Three major questions arose in this substudy. Two statistical power and integration of exposure data are common to the entire set of substudies and are discussed below. The third, peculiar to this substudy, was related to the
EVALUATION OF FINAL REPORTS 63 nonlinear least-squares fitting technique used to fit the growth data to nonlinear functions. The investigators used the NONLIN module of SYSTAT, which contains robust implementations of a steepest-descent (also known as the Newton-Gauss method) minimization method and the slow but relentless sim- plex method. The investigators do not provide details, so it is not clear which method they used. It is also not clear whether they tested various starting conditions in parameter space to ascertain whether they had reached the true minimum or the algorithm had been "fooled" by a local minimum. Trapping by local minimums is a problem that can occur with either algorithm, but the steepest-descent methods are particularly susceptible to this problem. One wonders, when one sees isolated large differences in mean values (as, for example, in the 1985 test in Figure 17 or the 1988 test in Figure 64), whether they reflect true variability or an artifact of the fitting process. GROWTH AND MATURATION IN DEER MICE This substudy terminated in 1991 because of technical problems. The design was similar to that of the substudy on avian growth and maturation. It produced one of the few positive effects that is not "explained away" in the general discussion: a pattern of earlier eye-opening in deer mice from test plots during periods of full operation than from control plots. The meaning of that effect, if any, for later development, health, or reproductive fitness of deer mice is unclear. All other measures were interpreted as negative by the investigators a reasonable conclusion in light of the few data that were ob- tained. HOMING IN TREE SWAEEOWS This substudy addressed the homing success of tree swallows by measur- ing the proportion of animals that, on displacement, successfully returned home and the time required for the return, which was around 4-5 hours. The flight paths for test-plot and control-plot animals were of comparable length and required the animals to fly over one leg of the antenna system. This was a carefully conducted substudy. The investigation paid attention to Endings that could result from possible artifacts due to peculiarities of the release site. The investigators observed differences in success and speed of homing that depended on the origin of the birds (from control or test plots).
64 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM They carefully examined whether the differences, which were seen every year, were due to plot differences or antenna operation, and they concluded that antenna operation was not related to the observed differences. This substudy exemplifies the internal inconsistencies that arise when the distinction between short-term and long-term effects is ignored and the exposure-assessment data are not integrated with biologic design consider- ations. Although the possibility of short-term and long-term effects is ac- knowledged (final report, p. 150), the acknowledgment does not translate into a clear separation into two hypotheses that require separate experiment de- signs. That is evident in the tortuous reasoning (pp. 152-155) that leads to the investigators' conclusions that the antenna system did not have an effect on homing behavior; their reasoning freely mixes potential short-term and {ong- term effects. The investigators also did not indicate which of the following three distinct hypotheses they were testing, their statements on p. 150 notwith- standing: a. If the hypothesis is that the long-term exposure to the electric and magnetic fields produced by the Navy ELF system causes a {ong-term effect in homing ability in birds, it is not relevant to have the birds fly over the antenna system; there should be differences between test-plot birds and control-plot birds released from several sites where the flight path does not intersect the antennas. b. In contrast, if the hypothesis is that test-plot and control-plot birds are comparable (that is, there are no long-term effects) and the homing effect results from antenna operation, it is less important whether there are plot differences, and it is all-important to ascertain whether the antenna leg that intersects the flight path was active during the 4-5 hours that the birds were in flight. Even in years of full antenna operation, there were down periods, and in some cases the legs of the antennas were rotated in their activity. c. Finally, if the hypothesis is that long-term operation of the antenna system causes some long-term change in homing activity in birds and that the effect is altered or magnified by acute exposure resulting from flight over the antenna, the experiment design that was used is appropriate. However, it is still important to ascertain whether the antenna leg that intersects the flight path was active during the 4-5 hours that the birds were in flight. Not only could that lead to exposure misclassification (and result in a bias of the results toward the null hypothesis), but it would confound attempts to clarify whether the observed, consistent differences between test-plot birds and control-plot birds are the result only of plot differences or of interaction between plot
EVALUATION OF FINAL REPORTS 65 (chronic exposure) and antenna activity (acute exposure). In Table 39, the investigators identified (footnotes C and D) some conditions in which they knew that the antenna was on for only part of the flights and that the propor- tion of time that it was on varied with each individual bird. In addition, they acknowledge (p. lLSO) that the antenna operation is highly variable, nearly hourly. HOMING IN SMALE MAMMALS This substudy was patterned after the tree swallow homing study but used deer mice and chipmunks. Considerable creativity was exercised in what must have been a difficult project. Although the data are generally consistent with the conclusion that the antenna system did not affect homing behavior in deer mice and chipmunks, the investigators acknowledge (pp. 163-164) that the sample sizes were often inadequate. For that reason, this substudy cannot be considered informative. DEVELOPMENT IN TREE SWAEEOWS This substudy characterized the normal embryologic development in tree swallows. Except for the common questions of statistical power and of whether the antenna was active in the period when the eggs were in the nest before collection it is an excellent substudy. The investigators used the well- accepted developmental framework provided for the chick by Hamilton and Hamburger in their landmark compendium and staging, and they adapted it to the different developmental timeframe of the tree swallow. They used ac- cepted, noncontroversial histologic methods and provided good rationales for the judgments that visual evaluation of embryos always entails. The results are generally negative, although a few isolated points reached statistical significance. However, the investigators' argument that the pattern of changes is neither consistent nor compelling for an antenna effect is con- vincing. Some specific aspects of their discussions are not so straightforward. In their discussion and comparison with other species, they suggest that differ- ences in waveforms between air and the egg-mother environment might have played a role in the general confusion present in this literature. They refer to work by Martin, who deliberately exposed eggs to different waveforms. Although the investigators might have a case with respect to electric fields, the
66 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM waveform from a magnetic field will not be altered by the egg-mother environ- ment. Far more relevant to a comparison with other species is other work by A. H. Martin (Martin 1989; not cited in their report) that showed, in the same laboratory under the same conditions, consistent differences that were directly attributable to which strain of chick eggs (either Arbor Acre or White Leg- horn) was used in the experiments. If the use of different strains of chicks can lead to different results, why not the use of different species? MAXIMAL AEROBIC METABOLISM IN CHICKADEES AND DEER MICE This substudy sought to measure the peak metabolic rates (PMRs) of chickadees and deer mice during winter. The authors used the "helox" tech- nique, which takes advantage of the higher thermal conductivity of helium than of nitrogen. For each measurement, the animals are given a helium-oxygen mixture to breathe while they are held at a preset low temperature. The helox mixture causes an increased loss of heat from the body, which, in turn, trig- gers an acute thermoregulatory response. The motivation of this approach appears sound ecologically, although the enthusiasm for this single measure as an index of "physiologic health" and ability to cope with "stress" appears to be more widely accepted by the authors of the quoted textbooks than by the entire community of physiologists. The substudy is problematic. Thermoregulatory responses cannot be equated with "stress" in the broad sense, in that, depending on the species, thermoregulatory responses might not share the hormonal profile of stressors. For example, the general stress response in most mammals involves the ACTH-adrenal steroid axis and sympathetic activation leading to releases of norepinephrine. A thyroid hormone response is sometimes but not always found. In contrast, the thermoregulatory response in birds is associated with an absence of the norepinephrine response and the main hormonal mediator is thyroid hormone. Even if one considers PMR on its own merits, the measurement must be done carefully. As outlined by the investigators (p. iS9), a true peak can be ascertained only by taking several measurements and showing a dropoff at temperatures above and below the peak. The peak temperature must be deter- mined for each individual. The investigators tested each temperature only once a day and needed a minimum of 3-5 days to determine the PMR. Adap- tation to acute temperature stresses has been reported, although it is not clear whether that presents a problem in this context.
EVALUATION OF FINAL REPORTS 67 A more important problem is that the investigators were not able to obtain complete, high-quality data on all the animals. They had to develop a rating scheme for "data quality." In addition, although 86-~% of the deer mice were able to complete all measures at acceptable quality, the success rate in chickadees was only 37-40%. Of the birds, 16-20% died during the initial transition to captivity, and an additional 34% of the ones that were left died during the multiday testing procedure. A physiologist wonders about a proce- dure to examine "physiologic health" in ostensibly healthy animals that results in the death of one-third of the animals. A number of analyses revealed plot differences and operation-period differences but not plot-operation interactions. On the basis of the analyses and the feeling that only the high-quality data classes should be considered definitive (p. 205), the investigators concluded that no differences could be attributed to antenna activity. COMMON LIMITATIONS OF SUBSTUDIES Integration of Exposure Assessment Into Experimental Biologic Design and Statistical Treatment of Data An Important limitation in all the substudies was the failure to integrate the exposure-assessment data into the biologic design. It is regrettable because most of the substudies collected data carefully. The limitation seems to have arisen in part because of a lack of broadly EMF-knowledgeable biologists at the project-definition and project-overview stages. This is also true for the other projects in the program. Although many publications have arisen out of the Navy's ecological monitoring program, the only two investigators that had and continue to have distinguished positions in the EMF-effects community are E. Goodman and B. Greenebaum. At the time of the 1977 National Research Council report (NRC 1977), it was known that the exposure metric that might be relevant to biologic re- sponses to EMF, if any such responses exist, had not been ascertained. No one had been able to define the biologically relevant "dose" metric. That situation continues. Some investigators examine the time-weighted average field exposure (TWA), others peak exposure, and others intermittence in field exposure as candidates for the biologically active field metric. Also unknown is the duration of exposure that is required for a response of an organism or the time required after exposure ends for the effects of EMF exposure, if any, to disappear; the relevant time constants have not been
68 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM defined. However, that does not mean that it is impossible to conduct useful experiments or that specific hypotheses about exposures and responses cannot be articulated and tested. The series of experiments presented in this final report (except the homing experiments) failed to ask that question, so the experiment design and statistical analyses are muddled and, in some cases (as in the PMR work), internally inconsistent. In particular, the statistical analyses use, as a measure of "dose," a three- leve! stratification of exposure based on the fraction of time that the antenna system was operational. That has several implicit assumptions that are seldom made explicit. The arbitrary definition of dose, which assumes that the frac- tion of time that the animals are exposed to any ELF EMFs is the relevant variable, implies little dependence on the level of operation (which for mag- netic fields is proportional to the current in the antenna). If the investigators believed that level of antenna activity was important, the obvious metric would be the product of the time of exposure and the field intensity; this was not used. It should be a truism that the times of operation of the antenna are impor- tant only if they coincided with times when the biologically relevant processes were taking place. For example, in the developmental studies of tree swallow eggs, it does not matter whether the antenna was on or off for the entire year, but only whether the antenna was active while the eggs were developing inside the mother or during the 3-4 days after the appearance of the last egg, when the embryos were being incubated before the clutch was collected and ana- lyzed. Consider the following three scenarios: a. For clutch 1 in 1988 the antenna was active for 2 of the 4 days of maternal incubation during this level 1 year. b. For clutch 2 in 198S, the antenna was active for 4 of the 4 days of maternal incubation during this level 1 year. c. For clutch 3 in 1991 the antenna was (because of repairs or other downtime) active for O of the 4 days of maternal incubation during this full- operation year. In those scenarios, the exposure assignment should be Clutch 1: level r (as defined by the researchers). Clutch 2: full. Clutch 3: inactive. Instead, the classifications based on the investigators' exposure system are
EVALUATION OF FINAL REPORTS Clutch 1: level I. Clutch 2: inactive. Clutch 3: full. 69 The resulting exposure misclassification tends to bias the results, in general, toward the null hypothesis (no effect), whether that is true or not. This is potentially correctable in that the investigators know when they col- lected a specific clutch and the Navy knows when the antenna was active. There is no evidence that this question was asked. The analysis by Beaver et al. (1994) intrinsically also assumes that the biologic time constant for onset of an effect is instantaneous and the constant for decay of an effect essentially infinite. That {cads to the following internal inconsistencies in the experiments: a. By the design of the PMR studies, the animals (from control plots and test plots) were already acclimated to the winter temperature. The helox procedure used to measure maximal metabolic rate can define the acute re- sponse in already-acclimated animals. Presumably, the hypothesis is that EMFs from the ELF antenna system could impair the acute response to cold, which is mediated primarily by a release of norepinephrine in mammals but not in birds, in which the main mediator is output of thyroid hormone (for example, see Prosser 19731. However, it must also be assumed (it is not stated) that EMF exposure affects chronically only the acute response and not the acclimation process. if it did not affect the acclimation process, the ani- mals from test plots and control plots would differ in metabolic capacity on the day they were caught, but no data were collected on that day. Why assume that EMF exposure chronically affects a fast, transient, minute-to-minute thermoregulatory mechanism (acute response to cold stress), and not also test whether the EMF response in this physiologically fast process might itself be fast and transient? Why not assume that (or at least test whether) nearly-con- tinuous EMF exposure (during the full-operation years) affects the acclimation process? b. Along the same lines, the animals were held in a holding area for at least a day before any measurements were conducted. The investigators report performing only one measurement per day and holding the animals for 2.5 weeks for retesting. Under the infinite-decay-constant assumption, the procedures are appropriate and the results obtained are consistent with the hypothesis: there are no differences in retesting measurements from short term to long term. But if a metabolic effect had a I-day decay time constant, which is not an unreasonable value for active physiologic responses known to
70 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM result from magnetic field exposure (for example, measured activation and decay of ODC in vitro or alteration of heart-rate variability in humans), then the animals from both control and test plots would have decayed back to baseline by the time of the first measurement, and no differences would be found. The comparisons of first-day rates to attempt to analyze short-term effects would already be too late to detect short time constants, and the consid- erable amount of work in the comparisons would be uninformative. Experiments with Statistical Power Too Low to Yield Meaningful Information The investigators decreased the sample size, and hence the power of the studies early in the process. in fact, the power that they settled for (a 70% chance of detecting a difference of a given magnitude) is below biologic stan- dards. That makes it much less likely that effects, if any were present, would be detected. As a project peer reviewer who was otherwise favorably im- pressed as he followed the project from year to year notes, it is not clear that the project would have been found meritorious enough to be funded if the lowered power had been proposed from the start. When combined with as- sumptions about information that would have been obtained in the projects that were dropped, it is clear that the information obtained is severely limited and statistically biased toward not finding an effect even if one was present. That is important for the evaluation of this program for the stated goals of the Navv's ELF monitoring program, and it raises questions about the manage- ment of the project. To allow investigators to drop the nominal power of some of the data collected, and hence of the statistical tests, to 70% is distressing; to allow designs that lead to statistical power of 30% or even less is unconscionable (see, for example, Beaver et al. 1994; pp. 49-5 1, swallow fecundity; pp. 72- 73, swallow body-mass growth and age at maximum; pp. 82-84, swallow tarsus-length variables; pp. 92-93, swallow ulna; pp. 93, 96, swallow wing; pp. 102-103, swallow landmark events; pp. 120-123, swallow growth in rela- tion to electric-field or magnetic-field intensity; and pp. 137-140, deer mouse landmarks). An accurate and sober appraisal of the consequences of this approach were clearly spelled out by a peer reviewer for the small-vertebrates project (letter dated February 15, 1990, p. 2, first paragraph): Approach Used in Statistical or Modeling Efforts. The statistical treat ment of the data continues to be excellent, although the need to drop the
EVALUATION OF FINAL REPORTS 71 power of the treatment from 90% to 70% for some data is sobering. While the reason for doing so is clear, it raises a question as to whether or not the final data will be effective in resolving potential ELF effects? doubt that the studies affected would have been endorsed had their original standard for statistical sufficiency been at the 70% certainty level. We'll just have to wait and see." [emphasis added] If the peer reviewer was disturbed by a drop from the 90% power "gold standard" to a goal of 70%, what are we to make of designs -and data with statistical power of, in some cases, less than 30%? In epidemiology and in clinical trials, where subject participation can be problematic (an approximate human equivalent of animal field trials), the standard is often relaxed to 80%. Below that, serious questions are asked as to whether a study is worth doing, as the peer reviewer questioned. The reason why studies that have low power and show no effects are not considered informative is that their results will fad! to exclude the null hypothe- sis in any useful way. For example, if a specific study with a 50%-power design indicates that an effect of a 35 % change in a variable was not found, there is an essentially even chance that a repetition of such a study could find an effect of up to 35 % and a hard-to-quantify but disturbingly nonzero chance that another repetition of the study could find an effect of more than 35% . There are rigorous ways to quantify such probabilities. These methods have become available to the nonstatistician only recently with the advent of inex- pensive computing power, and a research team should not be faulted for incor- porating in the early l980s a design that calls for early 199Os computing power. Where their judgment is called into question is in encouraging and performing studies that have been known since at least the early 1960s not to be informative. An alternative, nonmathematical way to illustrate the ramifications of studies with low power is to consider that the easiest and cheapest way to conclude that an effect is not present, even if it exists, is not to look for the effect ($0 spent). The next easiest way ($X spent) is to look for an effect in a cursory fashion, so that only a huge effect is likely to be observed. The situation does not change substantially if, instead of one variable in a cursory fashion, five variables are examined in an equally cursory fashion ($5X spent). The only conclusion possible is that for each of the five variables no huge effect is likely to be present, but little or no confidence attaches to the failure to observe an effect. Finally, if we say that $5X is all the money that is available, we can either conduct five cursory examinations or put all $5X into the examination
72 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM of one variable with appropriate power. The last approach has as a resource- limited conclusion that there could be an effect on any of the potentially infi- nite variables that were not examined. It also has the conclusion that the one variable that was examined properly could be categorized as showing "effect" or "no effect" with a high degree of statistical certainty. That is the approach that is almost universally applied in scientific research but surprisingly not by the TTTR! team. Scientific data are generally used only to accept a null hypothesis of no effect or to rule out the null hypothesis because an effect has been demonstrated. Hence, it makes sense to make the "effect-no effect" decision with the highest possible degree of statistical certainty. To the extent that some of the substudies produced data with statistical power ranging from 70% to less than 30%, such data cannot be considered to contribute to ascertaining whether the Navy's ELF antenna does or does not have an effect on neighboring small vertebrates. CONCLUSIONS The investigators distilled a number of questions of interest into hypothe- ses that were testable, at least in principle. Except on the studies of maximal aerobic metabolism, it is difficult to argue with their rationale and choice of variables to measure. Their choice of species seems reasonable and is well justified. They implemented generally sound programs to test their hypothe- ses. Apart from the lack of an appropriate and hypothesis-driven integration of the ELF-EMF exposure data with the biologic questions, their designs were generally good and the statistical analyses appropriate. However, several of the statistical analyses would have been different had the experiment design taken into account the complexities of EMF dose assessment. in general, the data from a given year were pooled and compared with those from other years; there was little attempt to ascertain the replicability of a given finding within the study except on a year-by-year basis. The investigators were generally responsive to the peer reviewers, one of whom submitted long, detailed critiques and appears responsible for several improvements and for the increase in the quality of exposition over the years. This was a multipart, complicated project, and the annual progress reports reveal the "growing pains" of the investigators, who sought and eventually found a format that was informative and readable. The presentation of the results is generally clear, and the discussion of the confounding variables that were identified was satisfactory. The peer reviewers consistently commended the investigators for their
EVALUATION OF FINAL REPORTS 73 attention to detail and for their candor in recognizing problems and pitfalls and their coming up with reasonable and constructive alternatives. However, after ascertaining that the variances of several of the variables of interest were higher than had been estimated, the investigators proposed (and apparently received ITTR} approval for) reducing the statistical power of their tests from 90% to 70%, instead of discarding some projects and increasing the sample size of the remaining projects to be able to meet the normal biologic standard of 80-90% power. In fact, when all the data were analyzed, many tests had powers of 30% or less. The investigators, after carefully analyzing their data and controlling for seasonal, yearly, and plot differences, reached the overall conclusion that the operation of the ELF antenna system did not lead to a consistent pattern of changes in the measured variables. There appears to have been some effort at integrating various findings and interpreting occasional positive, statistically significant findings as negative (for example, see the discussion on pp. 125- 126 of the final report). There was little consideration of alternative scenar- ios, for example, that because of lack of statistical power the null hypothesis could not be accepted or rejected. Another alternative scenario that was not considered was that biologic effects of EMFs might not have monotonic dose- response relationships. If the latter hypothesis had been considered, the inves- tigators might have been struck by how often the values obtained at the inter- mediate level of antenna activity differed from those at both the inactive and the fully active states, which were similar to each other (V-shaped dose-re- sponse curves). Even if the protocol had included a sound integration of exposure and biologic data, the limitations with respect to statistical power would lead to serious questions as to the appropriate extent of confidence in the investiga- tors' conclusions. Because of these limitations, it is not clear how informative the study results were in answering the basic question of the ELF monitoring program. Some of the additional analyses proposed in Table 5-! might yet provide informative results. LITTER DECOMPOSITION AND MICROFLORA PROJECT PROPOSAL Litter decomposition and associated microbial population dynamics con- tro! the flow of carbon and nutrients through soil, the ability of plants to take up nutrients through mycorrhizal symbioses, and the transmission of root
74 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM pathogens. Accordingly, this phase of the ELF monitoring program attempted to detect whether ELF EMFs from the Navy ELF communications system in Michigan altered those processes. If it did, there could be long-term conse- quences for plant-community structure and ecosystem processes, such as net primary productivity. The decomposition study had three parts: a litterbag decomposition experiment in which litter from red pine (Pinus resinosa), red oak (Quercus ruhra var. borealis), and red maple (Acer rubrum) was incubated in mesh bags in several places to detect the effect of the antenna after accounting for other factors that could also affect decomposition; an investigation of population densities of mycorrhizal-associated streptomycete populations; and an investi- gation of rates of disease progression through red pine plantations, the disease being infection with the root pathogen and wood-rotting fungus Armillaria. SPECIES AND SYSTEM SELECTION Litter Decay Red pine, red oak, and red maple are three dominant tree species of the northern hardwood forests of the Upper Peninsula of Michigan and in particu- lar those of the antenna site. Litter from these three species represents the extremes (pine and maple) and the middle (oak) of a continuum of litter qual- ity among species of the region, litter quality being defined as the sum of chemical properties of the material that affect decay processes. High-quality litter (maple) decays faster than low-quality litter (pine) because of higher initial nitrogen and water-soluble carbohydrate contents and lower lignin con- tents (Melillo et al. 1982~. The choice of these species to represent the contin- uum of leaf-litter types in the area is justified. MYcorrhizal-Associated Streptomycete Populations Streptomycetes are common actinomycetes associated with myccorhizal symbioses; the exact nature of the association remains unclear (Marx 1982, but see Richter et al. 1989~. Streptomycetes are important regulators of cai- cium oxalate, cellulose, and lignocellulose degradation (Crawford 197S, Knut- son et al. 1980, Antai and Crawford 1981), perhaps through the effect of their excretion of vitamins, amino acids, hormones, and enzymes on heterotrophic
EVALUATION OF FINAL REPORTS 75 decomposer populations (Richter et al. 1989~. Streptomycete populations are good choices as indicators of microbial-community activity. Root Pathogens Arn?illana is a heterotrophic fungus that decomposes lignin and cellulose in dead woody litter and live woody roots. Armillaria forms large, long-lived genes (Smith et al. 1992) whose detection and culture are straightforward. Armillaria is a suitable organism for studying the effects of environmental changes on heterotrophic decomposer and disease organisms. SELECTION OF RESPONSE VARIABLES Litter Decomposition Mass loss at various times during incubation was the only response vari- able selected. Mass loss integrates the activities of the entire microbial com- munity as influenced by litter quality and extrinsic environmental variables, such as climate, soil water potential, and in this case the potential effects of increased electromagnetic radiation. It is unfortunate that analyses were not also made of the release rate of nutrients, particularly nitrogen, which do not necessarily follow mass loss in a straightforward manner. Mass loss is only a crude measure of the release rate of nutrients from decomposing litter for plant uptake (Melillo et al. 1982, 1984~. The data required were in fact col- lected in the earlier years of the project but were later abandoned for logistic reasons. Streptomycetes The response variable chosen was the most probable numerical estimate of population density according to serial dilution in sterile cultures. This HA genes is a single individual genetically identical with other individuals in a clone. For example, a single aspen tree is a genes arising from a root stock common to surrounding trees.
76 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM variable is the standard response variable in soil-microorganism ecological studies. Different morphotypes of streptomycetes were isolated by exposure to different stains. Armil~aria Genets of Arn~illana were isolated from different portions of the red pine plantations, and seedlings infected by each genes were mapped. The mortality of the red pine seedling population infected by a particular genes was the response variable. EXPERIMENT DESIGN Biologic and Ecological Sampling Techniques The field design included the evaluation of the above response variables in two stand types (northern hardwood natural stand and red pine plantation) and three sites (overhead-antenna treatment, ground-antenna treatment, and control). A red pine plantation was sampled in each treatment site, and a northern hardwood stand was sampled in the overhead-antenna treatment and control sites. Thus, the experiment was pseudoreplicated (this will be dis- cussed later). Fresh material for each year's placement of new litterbags was obtained from one nearby red pine and northern hardwood stand to minimize variation due to initial differences in materials. Litterbags were put into place each December and collected monthly in May-November the following year for 9 consecutive years; no individual litterbag experiment lasted longer than a year. Soil for isolation of streptomycetes was collected at the same monthly intervals as litterbags. Red pine seedlings were assessed annually for mortality associated with Armillaria genets. Physical and Chemical Measurements intensities of the 76-Hz EMFs were mapped continuously at all sites, but the resulting data were never used in these experiments. The initial measure- ments used to determine treatment and control sites were the only measures of ELF-EMF exposure used.
EVALUATION OF FINAL REPORTS 77 Other climatic measures that could affect decay rates and microbial popu- lations included annual sums of air and soil degree days, total precipitation and storm frequencies, and evapotranspiration, assuming a soil water-holding capacity of 25 mm. Chemical measures of litter properties included initial concentrations of nitrogen, phosphorous, potassium, calcium, magnesium, and lignin in material collected each year and in monthly samples of incubated litter for the first 2 years only. Statistical Methods The main effect of interest the treatment effect due to the presence of ELF EMFs generated by the transmitting facility was pseudoreplicated (HurI- bert, 1984) in that there was only one site for each treatment. Therefore, the experimental data provide an estimate of variance of response variables within each treatment site but not the variance due to treatments across sites. Treat- ment is confounded with site. Sometimes pseudoreplication is necessary for logistical reasons, but apparently not in this case. However, when pseudoreplication is unavoidable, the generalization of treatment effects to other sites might be justified (with caution) if it can be demonstrated that the chosen sites for each single application of the treatment are not significantly different from each other at the outset and are at or very near the modal values of other environmental factors thought to affect the response variable of inter- est. Those two conditions were not considered, even though in many cases the data are available. The antennas were turned on in 1986; data on the physical and chemical measurements are available for 19SS. For the litterbag experi- ments, there were statistically significant differences in annual mass loss across the sites in 1985 even before the antenna was turned on (Bruhn et al. 1994, pp. 78-80, Figs. 7-91. Thus, the treatment effect of interest is imposed on a pre-existing site effect. This pseudoreplication problem underlies many other studies in the overall ecological monitoring program. The acceptance of all further conclusions must proceed with this caveat in mind, but the caveat is not clearly stated anywhere in the report. Because of covariation of site factors and treatments across sites, the effects of treatment, stand type, month, year, and (in the case of litter-mass loss) species were separated with an analysis of covariance (ANCOVA). Separate analyses were done for each response variable in each stand type; therefore, the overall effect in each stand type on the response variable of
78 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM interest is urdmown. Treatment (presence or absence of ELF EMFs) and year were combined into one variable, called "site-year." Month of sampling within a year was treated as a main effect. Replicates were three blocks or plots established within each stand at each site (the pseudoreplicate problem arises here) and replicates of numbers of samples taken of litterbag, soil, etc. within each block. For example, the effect of site-year and month on the percentage of original mass remaining at time t (I, t = 1 . . . 7 months sampled per year) was tested with the analysis of covariance using nine climatic variables and one interaction term as covariates (Bruhn et al. 1994, p. 70, Table 23~. The authors use the notation Xm for mass remaining at a given month m, but we will instead use the more standard notation I, where t is time measured in some meaningful units. The reasons for this will become apparent shortly. This analysis of covariance is wrong, in that it treats elapsed time within a year (successive sampling months) as independent. For example, the decay in, say, July is not independent of the decay in June, May, or any previous month since t = 0, because the residual material left after decay in any month is passed along to the next month. Similar considerations apply to years. For example, although a new litterbag experiment was begun each year with mate- rial collected afresh from the reference collection stands each year, it is possi- ble that the chemical quality of the material collected in year t + ~ is not independent of that collected in year t. Similarly, the populations of strepto- mycetes and especially the progression of Armiliaria infection in year t + certainly depends on the populations in year t, year t - ~ . . . year t - n. The data are therefore a time series, and the statistical tests used do not recog- nize this. The proper question to be asked is whether the presence or absence of the ELF EMFs alters the time series of data. An additional problem with this analysis of covariance is that it examines the effects of month, year, site, and covariates on the mean mass remaining across all months within a given year. That mean is the average of all consec- utive mass-Ioss data for months May . . . October. The mean of a monotoni- cally decreasing function from t = 0 to t = m months (see Eq. I, below) is meaningless. Sensitivity to and power to detect a statistically significant effect of the ELF EMFs on all response variables were considered, and the usual quality assurance and quality control procedures were used. However, the violations of some fundamental assumptions of ANOVA and ANCOVA and the pseudo- replicated nature of the design also affect the power tests. Therefore, any conclusions must be accepted cautiously, if at all, unless they are corroborated by alternative analyses.
EVALUATION OF FINAL REPORTS RESPONSE TO REVIEW 79 The litter-decomposition and microflora study was peer-reviewed annu- ally from 1983 to 1994. The comments of the reviewers seemed to be mixed. Suggestions that required minor changes in the experimental technique were usually followed, to the later satisfaction of the reviewers. More difficult problems involving theoretical issues of how to treat data were often not con- sidered or only partially considered, even when the reviewer directed the researchers' attention to specific references in the literature to guide them. Very few of those references were cited in the final report, and this leads to the conclusion that the researchers did not understand the issues. For exam ple, one reviewer repeatedly pointed out the utility of fitting decomposition models to a time series of mass-Ioss data and examining the effect of the antenna on values of the mode} parameters. The researchers responded by collecting such a time series of data but treating it improperly. Such problems remain in the final report and will be discussed below. PRESENTATION OF RESULTS Consideration of Alternative Analyses To critique the litterbag decomposition experiments, it is first necessary to consider some theory. Litterbag experiments generally rely on a standard first-order differential equation to describe decay rates (Olson 1963~: dX - = -kX, (3-1) dt which hypothesizes that the loss rate of a homogeneous material X is a con- stant, k, under a given set of conditions that are themselves assumed to be time-invariant. The solution to this equation is: X = Xe-kt, t O (3-2) where X is percentage of original mass remaining at time t. The authors chose not to pursue fitting the parameters of this equation to the data and evaluating changes in k as the response variable, because the first year's data
80 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM showed that mass losses had positive residuals (discrepancies of the observed data from the predicted values) from this equation in early spring and late autumn and negative residuals in summer. That indicates that k is not constant with respect to time measured in months. in particular, the equation overesti- mates decay in spring and autumn and underestimates decay in midsummer; the magnitude of these residuals is not discussed. The authors conclude that the mode} is not a valid description of the decay process in this study. They then postulate that might be due to changing environmental condi- tions during the decay process and therefore consider alternatives to include these environmental variables in Eq. 3-2: X = X e lilt - k2p t O Xt = XOe P. and (3-3) (3-4) Xt = XO(e kit + e-k2P) (3_5) where p is an environmental variable, such as precipitation. They then reject those approaches because they would become "overly complex," particularly for the "mission oriented objective of this research" (Bruhn et al. 1994, p.iS). Noting could be more wrong. First, what does it mean to say that Eq. 3-2 is the wrong mode! because of nonrandom distribution of residuals? That could happen for any of several reasons: (1) the mode! is fundamentally flawed as a description of the underlying biologic processes; (2) the units of the model, particularly those of the exponent, are not biologically reasonable; or (3) the model incorporates only some of the dynamics underlying the decay process, and others not modeled are responsible for the residuals. The authors accept reason 1 and do not consider reasons 2 and 3. Let us consider the latter two reasons, whose elucidation is the normal manner of proceeding in studies of this type. First, although time in calendar units (movement of the earth in its orbit) is the usual metric for k, in fact such units have little biologic meaning for microorganisms except as they reflect the integration of some controlling factors. Rather, time in this case might be measured as the cumulative sum of some climatic parameter, such as actual evapotranspiration or degree days. When that is done, the problems of nonrandom distribution of residuals often disappear (McClaugherty et al.
EVALUATION OF FINAL REPORTS 81 1985~. In fact, the presence of positive residuals under the cooler conditions of spring and fall and the negative residuals during the warmer summer sup- ports such an approach. That was not considered. Reason 3 is somewhat more serious, and in fact Eqs. 3-3 through 3-5 might represent attempts to resolve it. In fact, these equations are themselves the wrong approach. Rather, the more-appropriate mode! would be to replace k in Eq. 3-2 with a function describing a response surface of k with respect to various environmental variables: X = X e (f( A) (3-6) where f ~ ~ is a function describing a response surface of k in an e-dimensional space of environmental variables. Such an approach has been taken by Meente- meyer (1978), and Pastor and Post (1986~. Although the strength of the ELF EMFs could be one of the variables, perhaps a more-powerful neutral-mode! approach could have been taken (Caswell 19761. In a neutral-mode! approach, the hypothesis would be that the response surface completely describes the behavior of k without any considerations of the ELF EMFs. Alternatively, it could be hypothesized that the presence of the ELF EMFs alters the shape of the response surface f ~ ). The advantages of such an approach would be that it flows directly from Eq. 3-2, which has real biologic meaning, and that it would be predictive. it is not at all clear that it would have been more compli- cated than the approach taken. The ANCOVA approach taken is not compara- ble with any approach in the literature, nor is it predictive. It is an open question whether it is simpler. Interpretation The major conclusions reached through the ANCOVA of litter-mass loss, streptomycete populations, and Armitiaria infection rates of red pine seedlings are as follows: . Very small changes in decay rates of each of the three species were statistically detectable between antenna and control sites; these differences were not consistent from year to year. · No differences in streptomycete population of any morphotype between antenna and control sites were detectable.
82 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM · No differences in rate of spread of Armillaria infections of red pine seedlings were detectable, mainly because of high spatial variability in the spread of the disease. The question is, given the above-noted problems of experimental design, Are the conclusions presented by the researchers warranted? Would alterna- tive analyses support the conclusions (see Chapter 5~? Litter Decomposition The authors are correct in pointing out that a nonrandom distribution of residuals in the fits of data to Eq. 3-2 requires caution in the interpretation of the parameters of that equation. Nonetheless, k remains a useful first approximation of decay rates if the overall fit (r 2) is high and if the distribution of residuals follows the same pattern in all cases. The interpretation of differences in k would not then be biased by different and nonrandom distributions of residuals. We fitted Eq. 3-2 parameters to the mass-loss data for December i, 1992 to October 3l, 1993, as presented in Tables 3-5 (Bruhn et al. 1994, pp. 4446) to see whether this alternative meth- od of analysis corroborates their conclusions. December ~ was assumed as t = 0 in the absence of specific information in the report. Because small changes in k can have large consequences for mass loss, we also calculated half-lives (in days) of each species' litter for each treatment site: In(0.5) _ -0.693 ~ 5 ~ _ _ ohs k k . (3-7) The parameter values are presented in Table 3-1. First, we consider the proportion of variation explained by Eq. 3-2 (r 2) and the distribution of residuals. Because r 2 2 0.949, the unexplained varia- tion was always less than 5. ~ % and usually less than 2% . It seems that the exponential-decay model accounts for virtually all the variation across months for each species within a site. Therefore, the unexplained residuals, although admittedly not randomly distributed, are not serious. There is a pattern of slightly positive residuals during cooler weather in spring (the mode! overpre- dicts decay) that decline as the growing season warms (the mode} underpre- dicts decay), and occasionally return toward positive in fall in both the control and antenna sites. The pattern was common across all species and sites. Bio- logic reasons for this distribution of residuals have already been noted. Given the small residuals from Eq. 3-2 and their generally similar pattern, the use of k as a first approximation of decay rate poses no important problem. In
EVALUATION OF FINAL REPORTS 83 TABLE3 ~ Parameters Determined with Equation 3-2 for 1992-1993 Litter-Mass Loss Treatment k r2 I/2 life, days RMAP 0.001713 0.996 405 RMAH 0.001619 0.996 428 RMCP 0.001630 0.995 425 RMCH 0.001476 0.992 469 RMGP 0.001675 0.999 413 ROAP 0.000957 0.964 724 ROAH 0.000923 0.954 751 ROCP 0.000884 0.953 784 ROCH 0.000883 0.949 785 ROGP 0.000879 0.965 789 RPAP 0.000864 0.985 802 RPAH 0.000951 0.977 729 RPCP 0.000963 0.990 720 RPCH 0.000905 0.986 766 RPGP 0.000903 0.988 768 _ RM, red maple. RO, red oak. RP, red pine. AP, antenna plantation. AH, antenna hardwood. CP, control plantation. CH, control hardwood. GP, ground plantation. fact, the use of k implicitly treats the data as a time series and avoids the problems in this regard noted for the ANCOVA. Because of pseudoreplication, it is difficult to test for statistically signif~- cant differences in k between sites. This can be done by testing the difference in slopes of Eq. 3-l in each treatment against that of the control; it is not
~ 4 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM attempted in this report. Nonetheless, an examination of the values shows that species differences account for most of the variation and that differences with- in a species between sites are relatively small. Moreover, the direction of difference between control and antenna sites is inconsistent between species. That is, sometimes litter decays faster in the antenna sites than in the control sites and sometimes slower. Thus, the conclusions of the authors that the effect of the antenna on decay rates is much smaller than species differences and moreover is inconsis- tently expressed are borne out by the alternative analysis. The reason for the length of this alternative presentation is to show that such an alternative analy- sis, rejected by the authors, is possible and even results in simpler models than the ANCOVA used. Streptomycetes Insufficient data are presented to attempt a different anal ysis with, for example, a repeated-measures ANOVA or similar ANCOVA. However, inspection of the mean levels of micorrhizoplane streptomycetes and their coefficients of variation (CVs) for each month and each year (Bruhn et al. 1994, pp. 93-95, Tables 32-34) indicates that the means are rarely different in an absolute sense and the CVs are often in the range of 30-40% and some- times higher. The adjusted mean levels of micorrhizoplane streptomycetes (adjusted for covariance of climate variables between sites) are different in each of the study sites only in the third significant figure. Given such small differences in the absolute values of the means and the large variation within a sampling unit, it is doubtful whether any other analysis could detect a statis- tically significant treatment effect. Therefore, despite the problems with the experimental design and ANCOVA noted above, the conclusions might be robust to the relaxation of assumptions. ArmilIaria Arguments similar to those for levels of mycorrhizoplane streptomycetes could be made for Armiliaria infection rates. Inspection of the data in Table 42 of Bruhn et al. (1994, p. ~16) for the percentage of red pine seedling mortality caused by Armillaria in the years 1986-1993 shows high variability with each site and similar means. In fact, the range of data for the control stand encompasses the range of data on the overhead-antenna site for every year and nearly encompasses the range of data on the ground-antenna site for every year. Given that distribution of data, it is doubtful whether any other analysis could detect a statistically significant treatment effect. There- fore, despite the problems with the experiment design and ANCOVA noted above, the conclusions are probably robust to the relaxation of assumptions.
EVALUATION OF FINAL REPORTS CONCEUSIONS 85 The validity of these experiments for policy-making rests on whether the user of the results wishes to accept the problems involved in inferring a gen- eral effect of the antenna from a pseudoreplicated design and a design in which the progression of response variables is not treated as a time series of data. The user would have to accept the following propositions: . That the effects or lack thereof of the antenna detected in the three sites is not due to other site differences that existed before and during antenna operation and that these effects or lack thereof will be present on any other set of sites under the influence of the ELF EMF generated (the pseudoreplication problem). That the projection of effects into the future does not depend on the past state of the system (the time-series problem). . The danger is in committing a type II error accepting the null hypothesis (no effect) when it is false. It is not possible to calculate the probability of a type Al error, because it depends on an independent estimate of differences between treatment and control. Inasmuch as the treatment effect is confounded with the site effect, the differences between treatment and control cannot be attrib- uted solely to the antenna, the differences are not independent of pre-existing effects and confounding site effects that continued during the experiment. Given the small effects detected and the large variance within sites and between years, the effect of the antenna, if any, on decomposition processes and the microbial community is most likely much smaller than existing natural spatial and temporal variation in the forests. We will never know, however, whether the results are site-specific. It should also be noted that this provi- sionai conclusion of negligible effects of the antenna is due to the small differ- ences in means and the large variation in the data and not to the statistical, mathematical, and biologic validity of the analytic methods used. UPLAND FLORA PROJECT PROPOSAE According to Mroz et al. (1994), several studies (Wiewiorka 1990, Wiewiorka and Sarosiek 1987, Krizaj and Valencic 1989), which were inde
86 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM pendent of the Navy's ELF monitoring program, reported relationships differ- ent species between plant growth and EMF exposures. Accordingly, the effect of 76-Hz EMFs from overhead and ground antennas on tree growth, plant phenology, mycorrhizal symbioses, litterfall, and various environmental fac- tors that could affect growth were assessed from 1985 to 1993. SPECIES OR SYSTEM SELECTION Red pine (Pinus resinosa) plantations and northern hardwood stands were chosen for study. These forest cover types are the most common in the Upper Peninsula of Michigan and are therefore appropriate for determining potential effects of ELF EMFs on tree growth and associated environmental variables. SELECTION OF RESPONSE VARIABLES Atmospheric, soil, tree-growth, litterfall, mycorhizal, and phenologic response variables chosen for study are listed in Table 3-2 with sampling frequency. This is an extensive survey of tree-growth measurements and environmental factors that potentially affect it. The sampling frequency for each variable is appropriate and in many instances greater than normally used in ecological research. EXPERIMENT DESIGN Biologic an] Ecological Sampling Techniques The field design included the evaluation of the above response variables in two stand types (northern hardwood natural stand and red pine plantation) and three sites (overhead antenna, ground antenna, and control). A red pine plantation was sampled in each treatment site, and a northern hardwood stand was sampled in the overhead and control sites. Replicates were three plots established within each stand at each site. Thus, the experiment is pseudorep- licated; this will be discussed later. Physical and Chemical Measurements These are included in Table 3-2. Researchers assigned a 76-Hz magnetic-field dose to each tree on the basis of a weighted-average dose from
EVALUATION OF FINAL REPORTS 87 TABLE 3-2 Variables and Their Sampling intervals Tested for Significant Effects of 76-Hz EMFs in Upland Flora Studies (Mroz et al. 1994) Atmospheric Air temperature 2 m above ground (30 min) Photosynthetically active radiation (30 min) Relative humidity (30 min) Precipitation (cumulative, according to rain gauges) Soil Soil temperature 5 cm and 10 cm below surface (30 min) Soil moisture 5 cm and 10 cm below surface (3 h) Nutrients (0-15 cm mineral soil depth; total N and P; exchangeable Ca, K, and Mg; June and July) Tree Growth Diameter growth in hardwood species (! wit) Height growth in red pine (! wk) Litterfall Total mass by species (monthly or weekly, composited seasonally) Nutrient contents (N. P. Ca, K, and Mg; sampled as above) Mycorrhizue Total numbers by morphotype on red pine roots (monthly during growing season) Phenology Growth rates and timing of leaf and bud break on Trientalis borealis (twice a week) all the times the antennas were on or off. However, the actual dose to each tree was never the average, because of temporal variation during the growing season due to the antennas being on or off and spatial variation of the field across each site. These problems are discussed further in Chapter 4 in the section on use of exposure data by ecological monitoring teams. Statistical Methods The main effect of interest the effect of the ELF EMFs generated by the
BB EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM antennas was pseudoreplicated (HurIbert 1984), in that there was only one site for each treatment. Therefore, the experimental data provide an estimate of variance of response variables within each treatment at one site, but not the variance due to treatments across sites. Treatment is confounded with site. Sometimes pseudoreplication is necessary for logistical reasons, although that does not necessarily appear to be the case here. When pseudoreplication is unavoidable, the generalization of treatment effects to other sites might be justified (with caution) if it can be demonstrated that the chosen sites for each application of the treatment are not substantially different from each other at the outset. Accordingly, an extensive survey of a number of candidate sites was made at the outset. Sites were compared for diameter and species distribution of trees and soil characteristics listed in Table 3-2, and other soil characteris- tics, such as texture and volume of rock fragments that could affect tree growth and species composition. Three sites were chosen to minimize differ- ences in these variables at the outset to increase the probability of detecting an effect of the ELF EMFs. Analysis of variance (ANOVA) showed no statisti- cally significant differences between sites in any property except diameter distributions of red maple, which had a greater proportion of large-diameter trees at the overhead-antenna treatment site. For logistic reasons, climatic and soil temperatures were not measured before site selection. However, a great effort was made to minimize site differences before antenna operation. The careful site selection alleviated, but did not eliminate, the problem of pseudoreplication. The responses to the treatment were not evaluated with- in the context of variability across other sites that might be classified similarly. More important, the conclusions apply only to these cover types on these soils. Although the cover types and soils chosen are common in the area, other cover types (jack pine, hemlock, and balsam fir) and soil types (haplu- dalfs2 common on moraines) were not sampled. Therefore, extension of the conclusions to other upland ecosystems must be cautious. To separate the effects of the ELF EMFs from other environmental variables, the data were treated with ANCOVA with repeated measures. Variables that had no demonstrated correlation with the strength of the magnetic-flux density of the ELF EMFs but did correlate significantly with response variables of interest were selected as covariates. For most response 2Hapludalfs are soils of temperate climates with an accumulation of clay in the lower layers.
EVALUATION OF FINAL REPORTS 89 variables (soil nutrients, litterfall, and plant growth), ambient climatic vari- ables were used as covariates. The use of repeated-measures treatment of time is appropriate because the response in any year is not independent of responses in previous years. in the ANCOVA, the treatment-by-time interaction was examined for statistical significance. That is appropriate because it tests whether there are consistent differences in the effect of the ELF EMFs on the time series of data compared with controls. Because it tests differences in the shape and direction of the treatment responses, compared with controls, the treatment-by-time interaction term is independent of any site differences that existed before exposure to the ELF EMFs. The use of the treatment-by-time interaction term thus increases the power to detect statistically significant effects. Additional calculations of power and detection limits were also con- sidered for each variable. Tree-diameter growth was analyzed with a different method. The allom- etry (geometric relationship of the dimensions of one part of an organism to another) of diameter growth necessarily results in smaller increments in diame- ter with time regardless of limitations. The problem is to detect an effect of the ELF EMFs on the time course of diameter growth that is independent of other limiting factors. Accordingly, a model of diameter growth of hardwoods or shoot growth of red pine was constructed to predict expected growth given the selected tree species, size, site factors (mainly climatic), and presence of competitors. Residuals between predicted and observed growth were then tested for effects of the ELF EMFs by using the average strength of the field during the growing season (see comments above under "Physical and Chemical Measurements"~. Parameter values for the mode! were fitted to local data and tested against tree growth observed before exposure. There were no consistent differences between predicted and observed diameter growth across all plots and in 2 years for red oak (Quercus rubra var. borealis), paper birch (Betula papyrifera), red maple (Acer rubrum), and quaking aspen (Populus tremu- loides) or shoot growth of red pine. Details of model equations, parameter values, tests, and residual-effects testing are given in Mroz et al. (1994), Jones et se. (1991), end Reed et al. (1992, 19931. RESPONSE TO REVIEW The upland flora study underwent outside peer review for the years 1983- 1994. In general, the reviewers praised the strengths of the study and the ability and willingness of the researchers to deal with the very difficult ques
90 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM lions of experiment design. The use of the same reviewers during many con- secutive years allowed a degree of continuity and accountability for making improvements in experiment design or improving explanations of what was being done. The reviewers seemed on the whole to be satisfied that their earlier comments were being taken into account or that their concerns were being addressed in later elaborations in the annual reports. For example, one reviewer seriously questioned the utility of the tree-modeling approach in the early years of the study but later explicitly reversed himself and agreed with the approach taken when it was more fully explained. That these researchers appeared to take peer review seriously might have contributed to their later success in publishing the results in peer-reviewed journals. PRESENTATION OF RESULTS Consideration of Alternative Hypotheses Other factors that might explain differences in response variables between control and treatment sites (climate, soil nutrients, etc.) were included in the design by virtue of the ANCOVA with repeated measures. One could argue at length about whether the methods of measurement of these covariates were proper or whether other variables should also have been considered. The fact remains that a substantial amount of effort was put into measuring a broad suite of factors likely to influence growth other than the ELF EMFs and estab- lishing that there were negligible difference between sites in many of the response variables that were of interest before exposure. Interpretation The only variables that responded statistically significantly to antenna operation after consideration of covariate effects were in response to magnetic flux densities of 104 mG, with declining responses at higher fluxes: . Soil temperature at a depth of 10 cm in hardwood stands (control 0.5-~.5°C warmer than overhead-antenna treatment site during the first 3 years of antenna operation, with declining differences thereafter) · Increased diameter growth in aspen and red maple (about 0.14 cm/year maximal diameter response in aspen and 0.08 cm/year in red maple). · Increased shoot (height) growth in red pine (maximal annual re
EVALUATION OF FINAL REPORTS 91 spouses of 0.73 cm/year at the antenna-ground treatment site and 0.83 cm/year at the overhead-antenna treatment site). Any changes in other variables during the period of antenna operation were explained by concurrent changes in covariates (particularly climate variables, but also variables related to stand maturation), rather than any residual correla- tion with ELF-EMF strength. CONCLUSIONS Validity Among all the response variables measured (see Table 3-2 of this report), only four variables showed any response for which possible effects of the ELF EMFs could not be disregarded (soil temperature at a depth of 10 cm in the hardwood stand, diameter growth in aspen and red maple, and height growth in red pine at a narrow window of relatively low magnetic-field strengths). The responses of these variables are small and in fact did not persist as the magnetic-field strength increased. That is consistent with more-controlled experiments previously reported (Wiewiorka 1990, Wiewiorka and Sarosiek 1987, Krizaj and Valencic 1989~. No detrimental effects on plant growth, phenology, or nutrient uptake were noted; if anything, there was slight stimu- lation of growth of the three tree species within a narrow range of small magnetic-field strength. Given the few, small, and ephemeral effects ob- served, the researchers concluded that the ELF EMFs did not have any statisti- cally significant effect on vegetation and stand micrometeorology. That conclusion needs to be tempered by the fact that the magnetic-field strengths used to assess exposure the weighted average over the growing season were not the magnetic-field strengths to which the trees actually were exposed. In fact, the trees almost always were exposed to a magnetic field either higher or lower than the average. in most cases, the magnetic-field strength is close to one of two levels. Thus, average strength is not likely to correspond to an actual exposure level. Therefore, the conclusion that tree growth is not affected, or even slightly stimulated, by exposure to a 76-Hz magnetic-field intensity of 2.0 mG is weakened by the lack of exposure to such a field strength. It can be concluded that the total operation of the anten- nas had no major detrimental effect on upland flora beyond the background environmental fluctuations of weather and other variables.
92 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM Uncertainties The major uncertainty in the generalization of results lies in the pseudo- replicated nature of the experiment design. In part, that is an almost inevitable consequence of the site-selection criteria imposed to minimize site differences in as many variables as possible before antenna operation. It was probably extremely difficult to find three such sites with minimal differences in soils and vegetation, and they were then each divided into three replicates (or pseudoreplicates). Finding nine such sites to achieve true replication would be impossible. The demonstration that the sites were not different in almost every variable (except number of large red maples) before antenna operation only enhanced the detection of ELF-EMF effects if they existed at all The tradeoff is that, strictly speaking, the conclusions pertain only to the sites where the data were gathered. With some caution, they might be ex- tended to other sites that appear similar (northern hardwood stands and red pine plantations of these densities on sandy haplorthods3 of mixed mineralogy and frigid temperature regimes). However, the reliability of extending the conclusions to other sites is limited. Because of the pseudoreplication, there is no information on what the responses would be on other, albeit similar, sites. There is no direct evidence to determine whether ELF-EMF effects would be detected under other conditions. However, given the few, small, and ephemeral effects noted and the lack of any known mechanisms for the ELF EMFs to affect community structure and stand micrometeorology, this does not appear likely. . AQUATIC ECOSYSTEMS PROJECT PROPOSAL Potential effects of ELF EMFs on a stream ecosystem were examined by contrasting a stream site under the Michigan antenna and a paired control site downstream. The study had three main parts: examining the response of periphyton, analyzing the response of aquatic insects and their processing of leaf litter in the stream, and analyzing possible changes in fish abundance and movement as a result of exposure to ELF EMFs. 3Haplorthods are soils of colder climates with accumulations of iron and aluminum in lower layers.
EVALUATION OF FINAL REPORTS SPECIES AND SYSTEM SEEECTION 93 Periphyton or algal production is crucial to any stream and is a natural place to look for the effects of environmental perturbations. The researchers examined aggregate production values (in terms of chIorophyIl-a, cell volume, and organic matter) and the density of two dominant diatoms. Insects could be influenced by ELF EMFs directly or indirectly through effects on periphyton. Insects were sampled with baskets buried in the sub- strate that could then be removed and sorted. Insect activities could also be influenced, and these were examined in two studies. First, the movement rates of a dominant dragonfly naiad were quantified in the presence and ab- sence of ELF-EMF exposure. Second, the colonization of experimentally deployed alder leafpacks by insects was quantified as a function of ELF-EMF exposure, with measurements of leaf decay (quantified as a decay constant). The abundance, community composition, and movement of fish were examined in the control and treatment sites. Because fish behavior or physiol- ogy might be altered directly by ELF EMFs, it is prudent to look at their responses of all these kinds. The individual species selected for more-detailed studies were the dominant species and thus a likely choice. SEEECTION OF RESPONSE VARIABEES Periphyton Periphyton response was measured both through aggregate variables (such as total chIorophyIl) and through the densities of particular diatom spe- cies. These are apt response variables, although some indication of effect might be lost because of the aggregation of species into composite variables (see Carpenter et al. 1993~. Insect Response The response of insects was quantified at the level of functional groups, species abundance, community indexes, movement of dragonfly naiads, and colonization of leafpackets. The only questionable response variable concerns the heavy reliance on "functional groups," such as "collectors-gatherers"; previous studies of aquatic systems have shown that these aggregate variables conceal effects evident at the genus or species level (Carpenter et al. 19931.
94 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM However, because several analyses were also done at the species level, this criticism is not too serious. The interpretation of such community descriptors as diversity indexes and evenness indexes is obscure without dissection into species responses, but these community descriptors are not given much discus- sion in any event. Naiad movement was analyzed by regressing average dis- tance moved against ELF-EMF exposures. Average distance moved is a gener- ally poor summary statistic for dispersal rates or patterns, and a more sophisti- cated dispersal variable would have been preferable (e.g., mean squared dis- tance, which would give long-range dispersal greater weight, or the decay constant in an exponential probability distribution). Fish Abundance and Movement The analysis of fish response was to a large extent comparable with the analyses of insect response, with attention to abundances of particular species, and community structure. However, in this case, there was no aggregation of variables, and all the dominant fish were treated species by species. in addi- tion to counts of fish, the size structure and growth rates of fish were as- sessed; this enhances the value of the study because these are likely to be more sensitive indicators of any effect than are census data. Finally, the fish mov- ing past the treatment site and the control site were counted to see whether ELF EMFs inhibited movement in any way. Because ELF EMFs could alter movement, this was an excellent target for research, although, as discussed later, the questionable independence of control versus treatment sites compro- mised the results. EXPERIMENT DESIGN AND IMPLEMENTATION Only two sites were used: the treatment site and a control site ~ km downstream from the treatment site. The lack of replication is unfortunate, but the placement of the control site downstream of the treatment site makes the results of this study difficult to interpret. Such a placement means that one has to assume that what goes in one place does not influence fish, insects, and algae ~ km downstream an unjustifiable assumption. The researchers se- lected the two sites to match physical conditions. But if the two sites must be on the same stream, it would be far better to have the control site upstream so that "turning on the antenna" is less likely to influence the control sites di- rectly simply through mass movement of water, nutrients, and organisms downstream from the treatment site.
EVALUATION OF FINAL REPORTS 95 The streams were well characterized physically and chemically. Mea- surements included temperature, radiation, discharge, pH, dissolved oxygen, alkalinity, phosphorus, inorganic nitrogen, organic nitrogen, chloride, and silicate. The treatment and control sites were extremely well matched with respect to these physical attributes. Statistical methods were sound, and the researchers did a good job of testing whether the assumptions of before-and-after, control-and-impact (BACO were met and when they were not met, adjusting the analysis accord- ingly. PRESENTATION OF RESULTS Alternative hypotheses to ELF-EMF effects were examined where the BACT analyses indicated a statistically significant effect. For instance, possi- ble alternative explanations of increased periphyton production of chIorophyIl- a associated with ELF-EMF exposure were explicitly explored. Commend- ably, the power of the statistical tests was aptly examined; it was pointed out that only large changes in the fish community or biomass data could have been detected, given the variance and sampling frequency representative of the fish studies. CONCLUSIONS Validity The methods and analyses of this study were generally sound. However, three weaknesses detract from it: lack of replication (only one pair of sites was contrasted); placement of the control site downstream of the treatment site, which might be expected to show changes at the treatment site and there- by undermine such analyses as BACI; and the fact that many of the response variables were aggregate variables, which might be expected to be poor indica- tors of change (Carpenter et al. 1993~. Uncertainties By having such well-matched sites, the researchers minimized uncertain- ties due to fundamental site differences that had nothing to do with ELF EMFs. Unfortunately, because they placed the control site downstream of the treatment site, many of the findings of no difference between control and
96 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM treatment sites are suspect. The researchers recognize this; when discussing fish responses (Burton et al. 1994), they point out that "it is invalid to assume that the ELF antenna operations do not affect the assemblage at [the control sitel." The same uncertainty extends to other aspects of the ecological com- munity as well, such as the insects and periphyton, although the researchers do not address that possibility. The validity of the studies for policy-making is severely weakened by the placement of the control site downstream of the treatment site, even though the statistics are quite well done. if one dismisses the downstream problem, the evidence points mainly to a positive effect of ELF-EMF exposure on peri- phyton productivity and to no other major effects. It appears clear that the movements of dragonfly larvae, the colonization of leaf litter, and the move- ment of fish are not inhibited in any way by ELF EMFs. The ability to detect changes in the fish community was weak, as mentioned by the researchers, so only a very large change in fish biomass or abundance could have been de- tected, given the research design and inherent variability in the data. POLLINATING INSECTS PROJECT PROPOSAL The proposal is clearly stated and justified. [I bees are disoriented or damaged by ELF-EMF exposure, orientation to nests and foraging behavior are likely to change. If foraging is slower and less efficient, bees might reduce reproductive investment per progeny. Progeny might have higher mortality because of both direct and indirect ELF-EMF effects. To address the potential effects of ELF EMFs, the researchers generated eight specific, testable hypotheses. Possible mechanisms and response vari- ables are discussed for each hypothesis. Other factors that could affect the responses are considered later in the discussion. SPECIES SELECTION The choice of two species of leaf-cutting bees (Megachilidae) is fairly well justified on mechanistic grounds. The orientation and behavior of honey- bees have been shown to be affected by high-voltage transmission lines and fluctuations in the earth's magnetic field. Honeybees are known to have
EVALUATION OF FINAL REPORTS 97 magnetite particles in an abdominal organ that they use as a compass. The weak point, which is noted, is that "no data exist on the ability of megachilids to detect magnetic fields" (Strickler and Scriber. 1994, p. 5) or on whether they also contain magnetite. And the differences between fields resulting from the ELF antenna, fields from high-voltage transmission lines, and the earth's magnetic field are not discussed. Limiting the study to two bee species changed the research from the initial proposal to study a bee community and yet allowed for comparisons and reduced the risk inherent in studying one species in the field. Apparently, the two species were expected to respond similarly. Honeybees might have af- forded a better biologic assay of effects because of previous studies, but hon- eybees are rare in the study area and cannot overwinter there. However, some behavior studies could have been done with imported hives. The two megachilids were excellent bee species to study because they will use artificial trap nests placed out by the researchers. Therefore, nest availability and location could be controlled, and aspects of the nests and their contained progeny could be measured. Both species were relatively common in the area, so sample sizes were usually adequate. SEEECTION OF RESPONSE VARIABEES The decision to examine behavior, reproduction, and mortality, rather than population sizes, was excellent, given that population estimates are notori- ously Inaccurate. Instead, good measurements of specific responses that could affect population growth were made. The response variables were mostly based on previous results of studies of honeybees exposed to high-voltage transmission lines and generally were highly relevant and well justified. Honeybees show changes in orientation (waggle dances) and increased agitation in the presence of EMFs. For these bees, duration of leaf foraging was measured because it was more consistent than pollen and nectar foraging. Trip duration could be affected by such factors as leaf distance or quality, but these were not examined. Trip duration probably involves less-complex behav- ior than honeybee orientation and might be less likely to be disrupted by exposure to ELF EMFs. Orientation was measured as direction of nest choice. Honeybees exposed to high-voltage power transmission lines produce fewer cells and collect more propolis (a resinous substance in tree buds). Nest-building behavior was characterized by cell length, cells per nest, plug thickness, and leaves per nest. However, all those factors and reproductive
98 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM i: Investment per offspring depend on the sex of each offspring and are auto- correlated. Mothers under stress can vary the sex of their offspring. Sex ratios could not always be directly determined because of deaths and parasitism; therefore, they were estimated on the basis of cell positions; this adds some error. Overwintering mortality of honeybees has been shown to increase under high-voltage power transmission lines. Overwintering mortality was measured as the percentage of progeny dying and the percentage of nests with at least one death. Although each of those responses could change with ELF-EMF effects, they can also change in response to other variables- such as resources, tem- perature, or bee age and can be determined by a complex of factors that can vary spatially and temporally. The researchers could not estimate total reproductive output (an index of fitness) because females can make more than one nest and individual females were not marked and followed. EXPERIMENT DESIGN AND IMPLEMENTATION Biologic and Ecological Sampling Techniques Sites were chosen on the basis of required exposure criteria. However, the experimental design has two major flaws. First, the two replicates are actually pseudoreplicates because they are so close together; the researchers addressed this by using a nested design. Second, the treatment and control areas initially differed in many respects, including flower resources, and these probably changed independently in both time and space. Otherwise, great care was taken in standardizing methods and measure- ments. This often involved complex procedures. The methods of sampling (nest boxes) measurement are discussed in detail and could be replicated by new researchers. Observer bias was included in the analysis. The study is rigorous. The researchers provided i,152 nests per site, ensuring nonlimiting availability and good samples. Completed nests were immediately replaced with new trap nests. Overwintering mortality was measured on the basis of nests kept outside starting in 1989; they would not be affected by 60-Hz EMFs and microclimate indoors.
EVALUATION OF FINAL REPORTS Physical Measurements 99 Climatic measurements were obtained from the ecological monitoring program's upland flora study. However, these were not used in the final analyses a reasonable decision, given that local microclimate probably affects bee behavior more. Nest hutches were exposed to different temperatures and shading. Because of complex variation over time, measurements of this were not taken. Statistical Methods Data were analyzed by using the general linear models (GEM) procedure on a Statistical Analytical Systems (SAS) software package. Given that the design was unbalanced, with both nested and random effects, a mixed-model analysis of variance (ANOVA) using maximum likelihood or a before-and- after, control-and-impact (BACT) analysis would be more appropriate. Whe- ther the find interpretations of results would be different is unclear, given the high variation, confounding variation, and lack of true replicates. When distributions were nonnormal (for nest plugs), the researchers appropriately used a nonparametric factorial analysis of the data. The authors were aware of those statistical problems and addressed many of them. Their test for statistically significant ELF-EMF effects was based on a significant interaction between treatment-site response variables and levels of antenna operation, which would indicate that differences between the treat- ment and control sites were greater when the antenna was on full power than earlier, when it was on low or 50% power. However, this is problematic, as recognized, in that such an interaction could indicate either that the treatment and control sites respond differently to the different antenna levels or that the sites were different before antenna operation, but not after, for other reasons. For each hypothesis, minimal detectable differences and power of the tests were calculated and discussed in relation to the results. Quality Assurance and Quality Control in general, great care was taken in measurement. For example, measure- ments of nests were made in wire-mesh Faraday cages to minimize exposure of developing bees to electric fields during measurements. For consistency, 1
~ 00 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM only the first three foraging trips were measured. Observer identity was used as a variable in the ANOVA mode! to test for bias. Because sex ratios can vary with nest diameters and depths, data not standardized for these variables were excluded from the analysis. Exposure Assessment L Measurements of actual ELF-EMF exposures are reported for the treat- ment and control sites, with maximums and minimums given for one of the two treatment sites for June through August. The 76-Hz magnetic-flux densi- ties at the treatment site were 330 times stronger than at the control site. Exposure varied substantially between treatment sites, within treatment sites, and among years. The two treatment sites differed, in cumulative gauss- hours, by a factor of 2 (Strickler and Scriber 1994, Fig. 13~. Nest hutches differed within a treatment site by up to a factor of 100. Pretreatment years included low to 50% power (1983-19881. Mobile adult bees are subject to different exposures, but developing progeny are stationary and should have a consistent exposure, although this obviously varied between nests, sites, and years. The researchers argue that this variation in ELF EMFs is unlikely to be Important because the bees will show a threshold response to the large expo- sure differences between treatment and control sites. However, if the re- sponses are dose-dependent, this variation could confound the results. If there is a humped response curve (maximum at some intermediate level), ELF-EMF effects would be missed in the analysis. RESPONSE TO REVIEW Reviewers' suggestions to focus the initial study were heeded and con- tributed greatly to the success of this research. Other suggestions about meth- ods, such as shielding overwintering progeny from electric fields indoors and including observer bias as a variable, were also followed. Two major suggestions were not followed, for reasons that are not en- tirely clear. One was to use a BACT analysis with covariates. The other was to increase replication by placing fewer trap nests at many sites. The original design had four sites (with no replicates), but two of these were dropped. Given that ELF EMFs become ambient at 1.6 km, one reviewer suggested that
EVALUATION OFFINALREPORTS lO] control and treatment sites could be closer than 48 km apart, which could reduce some of the other environmental differences between sites. increased replicates and reduced variation could have allowed for stronger conclusions about the results. PRESENTATION OF RESULTS The authors are forthright about problems, such as unreliable data, which were eliminated from the analysis, and other confounding variables. A strength of this report is the clarity of discussion about problems and alterna- tives. Alternative hypotheses were often used to dismiss results that indicated potential ELF-EMF effects. The researchers' arguments are reasonable and soundly based on the data, but opposing arguments could be made for real ELF-EMF effects in most cases. Of eight parameters analyzed for one or both species, five showed statistically significant ELF-EMF effects for one or the other species, but only one, overwintering mortality, is interpreted as a possi- ble real ELF-EMF effect and this effect is regarded as ambiguous. Statistically significant effects on three parameters (cell length, leaf num- ber, and nest orientation) are interpreted as being caused by factors other than ELF EMFS. Cell lengths became more similar after full antenna operation for one species. The authors argue that ELF EMFS were not likely to have pre- vented the reduction in cell length (0.2 mm of ~ ~ . ~ mm) at the treatment site. Leaf number also became more similar after the antenna began operating at full power for one species; that is the opposite of what is expected if ELF EMFS are detrimental. The authors also argue that this difference (0.5 leaf/ cell) is trivial and unlikely to affect fitness or population growth. Nest orien- tation was argued as reflecting differences in local flower availability or shad- ing, rather than ELF-EMF exposure. Effects on nest thickness were not statistically significant, but the authors neted that the effect would have had to be very large to be detected with their test (hence, they dismissed this as a strong conclusion). They also cautioned against accepting the null hypothesis for trip duration and sex ratios, because of low power; that is commendable. In contrast, they accept the null hypothe- sis of no effects on offspring weight because statistical power was good. The ELF-EMF effect on overwintering mortality was statistically signifi- cant for one species. Mortality was lower at the treatment site at low antenna power but increased to the level of the control site after full antenna operation.
~ 02 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM The authors dismiss this as an ELF-EMF effect because mortality became increasingly similar and because of protocol changes, parasitism, and small sample sizes. However, mortality did increase after full antenna operation. Overwintering mortality was also found to differ significantly with ELF- EMF exposure in a nest-transplant experiment. Nests occupied at a treatment site but moved and overwintered at control site showed lower mortality than nests occupied and overwintered at the treatment site. In addition, the mortal- ity of transplanted nests was similar to that of the nests occupied and left at the control site. Although that is strong evidence for ELF-EMF effects, the au- thors argue that the effect is unlikely to be caused by ELF-EMF exposure. The sources of mortality were not emphasized in the researchers' report, because the sources were difficult to determine. For example, the researchers could not separate prepupal mortality in winter from that in summer, fall, or spring. They speculate that weather probably was important for much of the prepupal mortality. Because prepupal mortality varied greatly between years and sites, they decided to test the potential effects of ELF-EMFs on mortality after a bee had survived to the prepupal stage. That would reduce the varia- tion caused by site and weather differences. For this study, a major parasite was the cuckoo bee, Coelioxys, which is also a megachilid and could not be distinguished from the host in the prepupal stage. Therefore, both parasites and hosts are included in the mortality data. They argue that hosts and para- sites would probably have been similarly affected by ELF EMFs so this should not distort the results or interpretations. Overall, the authors conclude that ELF-EMF impacts are absent or at most minimal. CONCLUSIONS Validity Two major problems in the design seriously reduce the ability to detect statistically significant ELF-EMF effects: low replication (pseudoreplication) and confounding variables. The treatment and control sites initially differed in many factors, including flower resources, which also probably varied inde- pendently over time. Otherwise, the details of the study were well designed and well executed. Great care was taken in standardizing variables and in maintaining data qual- ity. A more appropriate statistical analysis could have been used, but it is
EVALUATION OF FINAL REPORTS 103 doubtful that different conclusions would be reached. Problems are discussed at length. Uncertainties The authors' conclusion that ELF-EMF effects are absent or minimal is uncertain because of the weak ability to detect effects. That results in a great- er likelihood of accepting a false null hypothesis (type II error) than of reject- ing a true null hypothesis (type T error or false positive). A type Il error would also be likely if the response curves were dose-dependent and either monotonic or hump-shaped. Therefore, the fact that the authors did find statistically significant effects requires careful consideration. The finding of increased overwintering mortality in two independent experiments makes an especially strong case for the existence of statistically significant ELF-EMF effects. The researchers' conclusion that ELF-EMF effects are absent or minimal might reflect the low power of the tests rather than the reality of no effects. Real effects would likely have been difficult to detect because of the small sample sizes and high variation in many factors. Therefore, the conclusion of "no effects" might, in fact, be based on the acceptance of a false null hypothe- sis (type TT error). A type lI error would also be likely if the response were dose~ependent and showed a monotonic or hump-shaped relationship to dose. To their credit, they discuss their reasons at length, but good reasons could be advanced for rejecting the null hypothesis in most cases. They also argue that the few statistically significant effects are small and would have little impact on populations. That is an erroneous argument because small differences over a long time can produce large changes in population sizes. Summary The authors' final conclusion that ELF-EMF effects are absent or mini- mal is questionable. The authors' explanations are inadequate for discounting potential ELF-EMF effects and for accepting the null hypothesis of no effect. Given the weak ability of the experiment design to detect ELF-EMF effects, any significant effects should be given careful consideration. Similar argu- ments were made by one reviewer of the final report who strongly feels that ELF-EMF effects were clearly demonstrated in this study. More independent
1 04 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM replicates would help to determine whether the statistically significant effects are caused by ELF EMFs or by some other factor. SOIL ARTHROPODS AND EARTHWORMS PROJECT PROPOSAL Soil macrofauna, such as arthropods and earthworms, control many of the decomposition processes critical to ecosystem function. It makes sense to examine the abundances, activity, and demographics of dominant soil animals. This study has six parts: soil and litter arthropod censuses, analyses of surface-active arthropod activity via pitfall traps, analyses of earthworm popu- lations sampled by square cores, analyses of growth and reproduction rates of earthworms incubated in soil bags, analyses of litter inputs sampled by litter traps, and analyses of litter-decomposition rates measured in litterbags. SPECIES AND SYSTEM SELECTION Soil and Litter Arthropods The species chosen for analysis were essentially the numerically dominant mites and collembolans (springtails) found in soil cores and litter samples. With no reason to choose a particular species, it is sensible to use the domi- nant species, because their frequency of occurrence makes them conducive to statistical analysis, compared with organisms that occur in only a few samples. Surface-Active Arthropods The researchers selected species on the basis of their abundance and commonness, which is reasonable. it would also have been desirable to pick species on the basis of relative uniformity in distribution over the areas of interest, so as to diminish the statistical problems of place-to-place and time- to-time variations. in addition, arthropods that actively forage on the soil surface could exhibit sublethal effects because of environmental perturbations (through altered behavior), and their activity patterns are potentially good indicators of ELF-EMF effects.
EVALUATION OF FINAL REPORTS Earthworm Field Populations 105 The researchers examined all nine earthworm species found at the treat- ment or control sites. Because they looked at all species, there was nothing arbitrary in the analysis, and the thoroughness is commendable. Earthworm Growth and Reproduction in Incubation Bags In addition to counting animals, it is a good idea to look for per capita differences in reproductive rates or growth. Using such demographic charac- teristics can yield far more sensitive indicators of ELF-EMF effects than waiting for population densities to reflect differences due to the activation of the antenna. Litter Inputs One can imagine ELF-EMFs influencing tree phenology, leaf production, and leaf abscission in a way that could alter litter inputs into forest soils. The measurement of litter inputs represents a sound research decision. Litter Decomposition Litter decay is certainly an appropriate system process to examine, espe- cially in the context of concordant measures of earthworm and other macro- fauna associated with decomposition. SEEECTION OF RESPONSE VARIABEES Most of the response variables involved counts or densities of individuals by species, an unassailable focus for analysis. Measures of community struc- ture, such as diversity and evenness indexes, were also examined; these aggre- gate indexes are difficult to interpret, and any changes in them would be impossible to understand without also analyzing effects species by species. However, these community indexes play a minor role in the analyses and are
~ 06 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM distracting only in that they add length to an already confusingly long docu- ment. A few response variables deserve comment. For the earthworm compo- nent, the vertical distribution of worms was examined as a possible indicator of changes in behavior or habits. The age distribution of worms and the size and cocoon production of earthworms were examined as functions of year (preoperational versus operational antenna years). in the earthwo~n~-incuba- tion experiments, the growth and reproductive status of "enclosed" earthworms were followed as a function of antenna operation and previous exposure. Because earthworms can live several years and might require 3 or more years to reach maturity, the attention to growth and reproductive rates in the earth- worm component of the study is commendable; these demographic rates should be more sensitive than absolute population numbers (which could re- flect long time lags and history). Leaf litter inputs were assessed by lumping grams of dry weight per square meter for basswood, maple, and all other species. Litter decomposition was assessed on the basis of the percentage of initial mass remaining. Mass loss is a crude measure of decomposition, and it would have been far better if a more-direct measurement of nutrient release had been obtained. EXPERIMENT DESIGN Biologic and Ecological Sampling Techniques The field design involved a single control site and a single treatment site for all components of the research. Replication is therefore impossible, and the only appropriate statistical analysis uses a before-and-after, control-and- impact (BACT) approach. Although the absence of replication might be un- avoidable, some aspects of the rationale are dubious. First, the two sites were not comparable: they had strikingly different earthworm and arthropod fauna. Second, for the earthworm-incubation experiments, the experiment design involved moving earthworms to a control site at which they did not occur naturally in any abundance (compared with either the treatment site or the sites from which they were taken). Thus, the experiment could be interpreted as an investigation of the effects on a species' growth and reproduction of trans- planting it out of its habitat. The detailed studies of earthworm size, age, and reproductive structure were confined to species that occurred only at the treat- ment site. That makes it virtually impossible to distinguish possible effects of antenna activation from effects of other temporally varying factors.
EVALUATION OF FINAL REPORTS Physical and Chemical Measurements 107 ELF EMFs were measured, and it was well demonstrated that the treat- ment site exhibited ELF-EMF levels at least 10 times those at the control site. The sites fulfilled TTTRI's other criteria as well. Statistical Methods The studies had four severe statistical flaws. First, when the BAC! analysis was used, the appropriateness of its assumptions of no serial autocorrelation and of additivity was never evaluated. Because the control and treatment sites clearly differed before the antenna was turned on, this is a serious problem. Second, often a simple ANCOVA was applied, although a repeated-measures approach is more appropriate. Third, the power of the tests and sampling scheme was low. For example, to sample surface-active arthro- pods, only 10 pitfall traps were used per site, an absurdly low number. In addition, given the large differences between sites and the variability in data, one questions how powerful the BAC! could be in detecting ELF-EMF effects. Fourth, for many of the earthworm analyses, the only data came from the test site, and no serious effort was made statistically to distinguish temporal varia- tion due to antenna activation from temporal variation due to other environ- mental variables. PRESENTATION OF RESULTS Consideration of Alternative Analyses In general, analyses did not include much consideration of alternative approaches. For example, numerous BACI tests were performed, some of which indicated statistically significant effects. No serious effort was made to determine whether those effects were due to ELF EMFs or to some other environmental characteristics associated with the difference between control and treatment sites. Interpretation The interpretations of data seem predisposed to conclude that ELF-EMF
~ 08 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM exposure had no statistically significant effect; several potentially significant results (as detected via BACI) were consistently dismissed. Biologically, that might be the correct conclusion, but the interpretation gave too little attention to the weakness of the experimental approach and the statistics being used relative to the variance in the data. Some experiments, such as the earthworm-incubation studies, were especially questionable in interpretation. This study cannot distinguish between a test of "natural" versus "unnatural habitat" and treatment versus control, because in the control site the earth- worm species being incubated was very rare. CONCLUSIONS The utility of this research for policy-making is compromised by a failure to ensure that BAC} was properly applied, by the problem of pseudoreplication and the fact that at best only one control site was compared with one treatment site (which differed substantially from the control site before antenna activa- tion), and by a statistical failure to disentangle temporal trends due to antenna activation from other temporally varying environmental factors. Any effects of ELF EMFs were small, compared with the total variation of the processes measured over the Il-year study; but the study included drought years (likely to be important to soil macrofauna), and it might not be consoling to learn that variation due to ELF EMFs is minor in comparison with the variation caused by a severe drought. SOIL AMEBAS PROJECT PROPOSAL This study, carried out over a period of more than 10 years starting in 1983, records data on populations of soil amebas from two treatment sites near the antenna (one next to the ground terminal and one under the antenna) and a control site about 15 km away, where 76-Hz EMFs were at most one-tenth as large. The studies focused on the ameba Acanthamoeba polyphaga and included counts of organisms in the soil (population sizes), growth-rate mea- surements in situ in culture vessels designed to match ELF-EMF exposures in the soil, and species present and determinations of genetic heterogeneity based on isoenzyme analyses.
EVALUATION OF FINAL REPORTS SPECIES SELECTION 109 Acanthamoeba polyphaga is one of the more common and already well- studied species of soil amebas that occur near the Michigan transmitting facil- ity. Soil amebas are micropredators, and variations in total counts are thought to reflect differences in quality and quantity of food available, especially bacte- ria. They are close to the bottom of the food chain and thus might be highly indicative of effects at that level. Measurements of total bacteria in the soil, a classically indeterminate value, were attempted with a modification of the acridine orange direct-counting technique; numbers were around lO9/g of soil but were highly variable, so attempts to use this technique to explain variance in soil-ameba numbers were abandoned. SELECTION OF RESPONSE VARIABLES The three principal response variables, as noted above, were counts of organisms in the soil, growth-rate measurements, and determinations of spe- cies present and the genetic heterogeneity. These seem well suited to address the overall study from an ecological vantage point, in that they embrace the static situation, the dynamic growth question, and possible genetic effects according to a well-established criterion. EXPERIMENT DESIGN AND IMPLEMENTATION Biologic Sampling Techniques The sampling techniques were well described and appear to have been carried out with professional competence. Similar experimental procedures were used for samples from the control and treatment sites. Physical Measurements and Sites The physical measurements of the fields and currents at both treatment and control sites were made in cooperation with ITTR] personnel and are well described in the report. Extensive measurements were also made of soil chemistry and moisture, the latter having a substantial effect on the biologic
~ ~ 0 EVALUATION OF ELF ECOLOGICAL MONITORING PROGRAM systems measured. Data on temperature were collected and are presented in full. The sites for sampling the organisms were selected in cooperation with IlTR} personnel. All sites had a similar 60-Hz EMF background, and the control site had 76-Hz EMF intensities no more than one-tenth those at the treatment sites. The ground treatment site was 39 m from the ground termi- nal, whereas the antenna treatment site was about 40 m from the north-south leg of the antenna. The control site was 15 hen south of the ground treatment site. Statistical Methods One-way analysis of variance (ANOVA) was used to detect differences in total-ameba and cyst counts at the three sites. The before-and-after, control-and-impact (BACT) analysis was used for the log maximal ameba counts and the maximal cyst counts and in the measurements of genetic diver- sity. CONCLUSIONS No population-density differences were found between antenna treatment, ground treatment, and control sites. In addition, growth rates of Acanthamoe- bapolyphaga did not differ between sites, and genetic-diversity studies failed to reveal differences between sites. The only statistically significant difference found was in conditions before and after antenna operation; there was a small but statistically significant difference in maximal population densities between the control site and ground treatment site, but by the same method of analysis no differences between the control site and the antenna treatment site or be- tween the antenna and ground treatment sites. The study appears to have been carried out carefully and competently. The inherent variability in the data was great and changes due to temperature and moisture were large, so small effects would not have been detectable. But the study results constitute convincing evidence that large-scale ELF effects did not occur over the time that the study was carried out. It should be noted that the antenna was fully operational only during the last years of the study and that even then there were down periods. Neverthe- less, this study was adequate for the questions being asked.
