Click for next page ( 58


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 57
3 Removal Estimation: Alternative Survey Design and Analysis Method Angler surveys that are well designed, soundly executed, and care- fully analyzed with modern statistical methods are crucial for providing high-quality information on total fisheries-related removals and related parameters (fishing effort) on which to base sound fisheries management decisions. As stated in Chapter 2, and now iterated, the important parameters to estimate from a recreational fishing survey are total recreational fishing effort, total recreational harvest (kept catch), and total recreational released catch. Effort (E) is often estimated from one survey and harvest per unit effort (HPUE) and released catch per unit effort (CPUE)Released from a second survey with total harvest (H = A + B1) estimated as: H = E HPUE and total released catch (CR) as: CR = E (CPUE)Released In addition, the fraction of the released catch that dies needs to be esti- mated in "hooking" mortality (MH) studies. This enables the estimation of total recreational fishing removals (R), which consists of the kept catch plus the released catch that dies as: R = H + CR MH 57

OCR for page 57
58 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS This then becomes the basis of stock assessment models. (See Chapter 1 for the definitions of harvest terms.) Harvest and total removals need to be measured for species and species complexes, specific spatial regions, and temporal periods, de- pending on the management needs involved. Further, total removals need to be assigned to age or size classes. Also, due to the very different nature of the for-hire and general fishing sectors, these sectors also have to be sampled separately using different methods for efficient estimation. SMALL- TO LARGE-SCALE SURVEYS FOR SOUND FISHERIES MANAGEMENT To estimate primarily angler effort and harvest, angler survey design has received much attention since the 1990s when the American Fisheries Society commissioned a symposium and a detailed monograph on the subject (Guthrie et al., 1991; Pollock et al., 1994). The traditional access and roving surveys developed in the 1960s (Robson, 1960, 1961; Malvestuto, 1983) for small water bodies (e.g., lakes, reservoirs, trout streams) were just not suitable for the larger spatial-scale surveys, which are so crucial in fisheries management. This is especially the case in marine fisheries management where the unit of management may range from coastal waters of a small state, to a region involving groups of states, or even up to the national level. One traditional survey is the access-point intercept survey. Robson and Jones (1989) developed a modification called the "bus route design" and applied it to a small regional-scale fishery in Lake Ontario tributaries in New York. Related access-point marine surveys at the regional scale are run in Texas and Oregon, among others that came under the mandate of this report (see Appendix B). Unfortunately, there are several prob- lems with using these designs, and without major modification and enhancement, these problems limit the usefulness of these surveys. There may be a large number of access points and some may be very small in size; often there is private access that cannot be sampled using only public access points, and the spatial scale may be so large that cost savings may be achieved by using an offsite contact method (e.g., telephone). Roving surveys using agents on foot or in boats also become impractical when it comes to larger spatial scales. An active area of research involves the design of complex surveys for even larger regional and national marine fisheries (Dauk and Schwarz, 2001; Lyle et al., 2002; Henry, 2002; Pollock, 2002; Volstad et

OCR for page 57
REMOVAL ESTIMATION 59 al., in press). Often these surveys require a design that uses one survey for effort and another survey for catch rate. Examples include the pairing of aerial surveys of effort with access surveys of catch or telephone surveys of effort with roving surveys of catch. These paired surveys are known as complemented surveys (Pollock et al., 1994). One example of a regional survey that uses an important comple- mented design (aerial and access) is the Georgia Strait Creel Survey. This survey has been run since 1980 by Fisheries and Oceans Canada for the Georgia Strait area near Vancouver, British Columbia. It uses aerial flights to estimate angler effort by taking aerial counts of boats fishing and expanding these counts. This effort estimate is combined with data collected by clerks stationed at access points to record catch rates of individual anglers to estimate total catch. Catch and effort statistics for this tidal sport fishery are calculated for each month and statistical area, and for individual species. According to survey results, catch of salmon species has shown serious declines since 1980 (Hardie et al., 1998; Dauk and Schwarz, 2001). Surveys with this design also are used by Michigan on many of its Great Lakes Surveys (Lockwood et al., 2001) and also in the Delaware River Creel Survey (Volstad et al., in press). The latter survey was designed to estimate catch for important anadromous species (e.g., shad, striped bass) in Delaware and Pennsylvania. Unfortunately, in many settings, there is a need for more information at much larger regional and even national scales that will require the abandonment of direct onsite estimation of fishing effort for total cost reasons. This suggests the possible use of telephoneaccess and telephonetelephone survey designs (Pollock et al., 1994). This com- mittee was formed because of a concern for the reliability of a large spatial-scale telephoneaccess survey, which is what the Marine Recre- ational Fisheries Statistics Survey (MRFSS) uses (Essig and Holliday, 1991; see Chapter 2). Another national survey run recently in Australia used a telephonetelephone survey design with anglers contacted repeat- edly using a panel diary approach (Henry, 2002; Lyle et al., 2002). The objectives of the Australian survey were to describe the characteristics of anglers (participation rates, sociodemographics); eval- uate effort and catch by species, mode, and region; assess economic impacts in terms of investment and expenditure associated with fishing; and evaluate awareness and attitudes to fishing-related matters. All salt- water and freshwater fishing activities were included within the scope of these surveys, which were comprised of the following components:

OCR for page 57
60 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS 1) A screening survey designed to identify fishing households and to invite anglers to participate in the follow-up diary survey 2) The diary survey in which fishing and expenditure activity was monitored over 12 months through regular telephone contact by survey interviewers 3) An attitudinal survey administered as a final telephone interview at the completion of the diary survey In general, an advantage of the use of telephone surveys is that one can obtain information on effort and catch rates for anglers not easily reachable in an onsite survey (typically an access survey). These could include night anglers and anglers fishing from private docks and jetties. However, a key concern is that effort and catch-rate data that are self- reported may contain large measurement errors. These errors may be due to willful deception, recall bias, prestige bias, or lack of knowledge (e.g., species identifications). Lyle et al. (2002) discuss these potential prob- lems and review the methods that they used to attempt to reduce these errors to a low level. In the Australian context, it was not feasible to go to the telephoneaccess design for cost reasons. It is widely known that there are tradeoffs between survey costs and the precision of the estimates, but it is also true that methods that reduce bias in the estimates may be much more expensive. Onsite catch-rate estimates are much more expensive than offsite self-reported catch-rate estimates (Pollock, 2002). An access survey for catch rate would get around these problems (Essig and Holliday, 1991), and this was an important reason for the current MRFSS design. What are some appropriate combinations of contact methods to use in particular situations? The spatial scale of fisheries management decisions will be a crucial component. For some local or regional fish- eries, the accessaccess surveys may be optimal; whereas, for other regional surveys, the aerialaccess design may be preferred, and at larger scales, the telephoneaccess (augmented with special studies) is often the only practical option for both the general angler and the for-hire sector. Telephonetelephone surveys, while useful in Australia, will not be useful in the U.S. marine setting to estimate removals for management decisions, as there is the need for an onsite interview component in all surveys. However, telephonetelephone surveys may be useful in special studies of night and private-access fishing because these modes cannot be well assessed in the MRFSS (Chapter 2). General questions that involve policy and economics also could employ telephone panel surveys

OCR for page 57
REMOVAL ESTIMATION 61 (Chapter 5). Augmentation of telephone contacts by internet surveys needs to be considered and will be discussed later in this chapter. ANGLER SURVEY FRAMES For estimation of removals and related parameters (effort and CPUE) for marine recreational fisheries, frame problems are extremely challenging. A frame is a set of units that are somehow linked to the population elements of interest. Estimation of a population characteristic is carried out by sampling units from the frame, identifying the population elements linked to the sampled units, and measuring the variable(s) of interest on the population elements. Two standard types of frames, also discussed in Chapter 2, are list frames and area frames. A list frame with known undercoverage, but that is inexpensive to sample, may be combined with an area frame or another complete list frame that is expensive to sample. Such surveys are called dual-frame surveys. To illustrate, consider the simplest dual-frame estimator, called the screening estimator (Hartley, 1962). The general idea is that the list frame is incomplete; whereas, the area frame is complete, and therefore, there are two components. The overlap domain (OL) is the list frame, and the nonoverlap domain (NOL) consists of those members of the area frame that are not on the list frame. Therefore, assuming simple random sampling in each frame, an estimate of the population total () would be the sum of the population estimates for the two domains: Y^ = Y^OL +Y^NOL where the usual population total estimator for the list frame is used for the overlap domain. All units that are on the list frame are screened out from the area frame, and only the remaining units are used in a standard estimator to get the nonoverlap domain estimator. There are many complications when dual frames are used in real surveys, but this illustrates the general principles. In some of the applications of most interest here, the complete frame would be a random digit dialing (RDD) telephone frame (instead of an area frame), and the incomplete list frame would be a telephone list frame from an angler license file that suffers from incompleteness.

OCR for page 57
62 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS Possible Frames for Effort Estimation The population characteristic of interest for effort estimates is total angler effort (e.g., number of angler days, number of angler trips). While the description of the unit of effort might vary somewhat among angler modes, the following discussion uses angler days as a surrogate for all of these units of effort. Angler effort can be assessed by either defining the population as all fishing days and then counting anglers active on those days or by defining the population as all anglers and then counting the days they fished. The first option is problematic because there is usually no simple way to count active anglers on a given day. There are a huge number of ways in which anglers can access the water, though this varies greatly from region to region. Sampling from area frames of coasts and coastal waters could be very inefficient, except in certain constrained waters (e.g., bays, estuaries) in which fishing effort could be assessed through aerial surveys, as mentioned earlier in this chapter, or other direct observations (e.g., bar crossings from the Columbia River in the Oregon Recreational Boat Survey; see Appendix B). Sampling from access-site list frames is used in some smaller regional surveys to get at effort through on-the-ground assessments, such as counting boat trailers or empty marina slips. However, many surveys cover such a large spatial area that this becomes completely impractical. Other difficulties with using access-site list frames are discussed further when considering CPUE estimation through angler intercepts. The second option of sampling the population of all marine anglers is currently problematic but offers the best hope for sound future surveys. It depends on the availability of a list (frame) of the population of all marine anglers. Such license file lists are available in some states but not others; in general, states in the northeastern United States (New Jersey northward) do not have saltwater licenses at all. At the inception of the MRFSS, license frames were not available in many states and the MRFSS designers were forced to use a different list frame. Through its RDD sample, the MRFSS uses a frame of all working landline telephone numbers in coastal counties. This frame suffers from overcoverage since not all households contain anglers, undercoverage since some anglers do not live in coastal counties or they live in coastal counties but do not have landline telephones (a problem likely to grow as more households move to only cellular telephones), and duplications since some anglers live in households with more than one working landline. Overcoverage leads to severe inefficiency in the RDD sampling effort. Undercoverage

OCR for page 57
REMOVAL ESTIMATION 63 in the coastal county frame may lead to serious bias since anglers from noncoastal counties are likely to have different effort characteristics than those from coastal counties. An attempt has been made to adjust for this potential bias using information collected via field intercepts in a pro- cedure much like a dual-frame survey; however, as mentioned in Chapter 2, this procedure is ad hoc and likely biased. Other list frames used in sampling the population of marine anglers include state- or regional-level licensing systems (Washington, Oregon, and California surveys use such frames). Licenses are linked directly to the angler population of interest, but license frames can suffer from overcoverage (e.g., due to out-of-date licensing information), under- coverage (due to license exemptions or poaching), and duplications. Overcoverage in the license frame is much less than with RDD so sampling is potentially far more efficient. Undercoverage is reduced if license exemptions are minimized. Undetected duplications could be problematic because anglers with more than one license listing may be more avid anglers and would be overrepresented in the sample. Clearly, there is a need for a complete angler registry in all states; these should be designed rigorously to minimize under- and overcoverage. If license frames suffer from substantial incompleteness, then dual-frame ap- proaches could be and should be used to adjust for this incompleteness rigorously, but this will be more expensive and make the surveys more complex than if a complete license (registry) file frame were available. In the for-hire sector, list frames of operators (based on licenses) are available and being used in telephone surveys in many regions of the country. The same issues of making sure that these lists have minimal under- and overcoverage problems are important. Frames for Catch per Unit Effort Estimation The population of interest for CPUE estimation is the population of angler days or trips. This population is, on occasion, accessed through an area frame of coastal waters, with roving boats visiting fishing vessels, but this is expensive and impractical in large spatial-scale surveys and also may be seen as intrusive. As a result, it can be difficult to count and measure the fish caught accurately. The population is accessed primarily through a list frame of site days, where sites are documented fishing access points. (Such site lists are used on occasion for assessing effort, as noted above.) Site days are selected with guidance from a "pressure matrix" that indicates expected fishing intensity across site days. Once a

OCR for page 57
64 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS site day is selected, field personnel visit the site on that day and attempt to intercept returning anglers. The field personnel have considerable latitude in how they go about intercepting anglers. Errors in estimating the expected fishing intensity and failure to account for expected fishing intensity in the estimation process can lead to both increased variance and bias in the CPUE estimates. The major problem with site list frames is undercoverage. Some public access points may be missed in the listing procedure, and private access points are not listed at all. Estimates of CPUE may be biased if anglers accessing the water from private access points or from little- known public access points differ in their fishing (e.g., fishing modes, areas and species targeted, effort and success rate) from those accessing the water from well-documented public access points. In the for-hire sector, access-point interviews of anglers also are required, and the same issue of inaccessible private marinas may apply. Since expertise on local geography, fishing modes, and species variation is critical, maintenance and sampling of access-point list frames for CPUE estimation is best done at a local level. Even with outstanding local expertise, access-point list frames have a number of potentially serious deficiencies, as outlined above, and need to be supplemented with area samples or other dual-frame techniques to get at CPUE for anglers not accessing the water from listed public access points. National Registry Frame This discussion of difficulties with existing frames means that, barring major advances in technology (such as remote sensing) that would allow assessment of fishing effort day by day, a much improved frame for interviewing anglers is needed. Use of the RDD approach in coastal counties is inefficient, potentially biased, and likely to grow even worse over time, but it is the only currently viable option in states without a complete registration of marine anglers to provide a license frame. A national registry database built on existing state angler licenses and augmented with new licenses would be an ideal frame for sampling marine anglers if it minimized duplications through rigorous and nationally consistent registration standards, minimized overcoverage with regular database updates, and minimized undercoverage by disallowing exemptions. Such a national registry database would yield considerable efficiency for sampling effort over the current RDD frame.

OCR for page 57
REMOVAL ESTIMATION 65 There would be enormous management benefits, cost and interview savings, and increased quality of the catch estimates obtained. Some states currently require a license to fish in marine waters but do not use the associated angler information to conduct effort or CPUE surveys. This happened in states where the license was developed principally as a means of revenue generation with little application to data collection. Because of the associated fee component, these licenses frequently have numerous exemptions, which reduce their usefulness for frame development and sampling. For example, in Florida,1 only anglers fishing from a boat in state waters (or traversing state waters to land fish caught in the exclusive economic zone) must buy a Florida saltwater fishing license. Anglers fishing from shore and those over 65 and under 16 are exempt and therefore would not be contacted if the license frame were used for data collection. Saltwater fishing license requirements vary by state, as do the exemptions. Therefore, many current license programs would need to be modified substantially to be suitable as a complete sampling frame. The recognized need for a national list frame of anglers is not new, and several previous reviews have offered similar recommendations (National Research Council, 2000), but there has been significant resis- tance from some states to federal involvement in this issue (Box 3.1). Some fear that the additional cost associated with purchasing a license will dissuade people from becoming anglers, and those that are now exempt from license fees likely will resist imposition of the fee if they are required to purchase a license. Further, in the northeast in particular, there appears to be a cultural aversion to the basic idea of saltwater licensing. Still, there are many reasons why a state-level saltwater angler license would benefit data-collection efforts. Cooperation between the federal and state governments on a mandatory salt-water angler registry (or license), with attention to eliminating exemptions in states with current saltwater licensing and with encouragement to other states to implement such licenses as quickly as possible, would lead to realization of those benefits. The national registry and state survey programs would need addi- tional funding to establish and maintain this type of database. However, there also would be large cost savings associated with sampling from this frame as compared with RDD, where a small proportion of the contacts reach an angler. An updated, complete registration list would greatly 1 Refer to Florida Fish and Wildlife Conservation Commission (2005) for a full list of Florida's exemptions.

OCR for page 57
66 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS Box 3.1 Lessons Learned from Boating Registration State fisheries agencies generally believe that the federal govern- ment lacks sufficient authority for requiring saltwater licenses for those who land their fish in state sovereignty waters. Because of the delicate balance of state and federal interests in marine fisheries, implementing a national saltwater fishing registry continues to be a contentious issue and significant political will may be needed. However, there are important lessons to be learned from recreational boating, most notably, boat regi- stration and numbering. Previous legislative actions for this sector can serve as model for the statefederal cooperation that will be needed in establishing a national angler registration. At one time, some states had recreational boating registration systems while others did not. Likewise, there were differences in how each state registered boats, the data they collected from owners, and the interval in which information was updated. The Federal Boating Act of 1958 (46 U.S.C. 527-527h) gave states the responsibility for registration and numbering of all undocumented boats after years of benign neglect by the U.S. Coast Guard. Furthermore, national standards for registration and numbering were instituted, including what data were to be collected from boat owners. Deficiencies in boat coverage for numbering and registration purposes were remedied in follow-up federal legislation (Federal Boat Safety Act of 1971 [46 U.S.C. 1451 et seq.]). This statute provided incentives in the form of additional funding for states that adopted uniform laws; states that failed to do so were penalized by having a federal numbering and registration system implemented in their respective state. Thus, to improve the quality and quantity of survey data on marine recreational fisheries, there is a need to establish national standards for existing and proposed state-level saltwater angler licenses or for even- tually generating a national universe of marine anglers. It is not auto- matically necessary to establish a national saltwater fishing license to be administered by the National Marine Fisheries Service (NMFS). There are notable differences here, and the words are important. Some states with saltwater licenses may only have to modify the types of data they collect or expand licensing coverage to anglers previously exempted; other states may need more convincing. Federal standards should deal with the exact types of data collected from anglers and should require that exemptions be eliminated or kept to an absolute minimum. improve efficiency both in terms of time and cost. It is not assumed that these savings would cover the entire cost of maintaining such a database. However, the benefit from the increased quality and quantity of the data

OCR for page 57
REMOVAL ESTIMATION 67 will be well worth the extra cost, especially if there is an associated increase in public confidence with the final estimates. Also, the creation of such a list will be essential to implementing some of the other recommendations found in this report. It is critical that the licensing requirements eliminate exemptions and noncompliance by segments of the fishing public. Significant efforts to enforce these registration requirements will be necessary. The statistical problems arising from any unavoidable incompleteness of the frame can be addressed in various ways, with the most important one being the use of a dual-frame approach. This will add additional expense so it is crucial to minimize undercoverage of the saltwater license frame. Also, the benefits associated with the angler list frame would be diminished if this list also included freshwater anglers. Including freshwater anglers in the same database would reduce the efficiency gained by the implementation of the registration--unless the data about each angler identifies them either as a freshwater angler, a saltwater angler, or both. OTHER SURVEY DESIGNS Panel Surveys A panel survey is another methodology that has been used in collecting recreational fisheries data. One example is the telephone diary panel survey used in Australia to assess recreational fishing (Henry, 2002; Lyle et al., 2002). This survey used multiple contact telephone interviews to get both fishing effort and harvest rate over a one-year period. Panel surveys should be considered for the telephone survey portion of the MRFSS (National Research Council, 2000). A rotating panel design, with membership in the panel lasting one year (six waves), might be a reasonable approach for the MRFSS. Panel surveys collect data from the same individuals at regular intervals of time. This design also is referred to as rotation sampling. The main purpose of such a design is that it produces more efficient estimates of change from one time period to the next. To see this, yt is defined as the parameter of interest at time t (e.g., total fishing effort in a given wave) and yt+ as the total in the next time period. These totals are esti- 1 mated by y^t and y^t , and the change (t^ +1 ,t+1) is:

OCR for page 57
72 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS methods to estimate the number released are needed. Use of observers in boat-based fisheries to get direct estimates of numbers released could be explored. However, there are problems with releases being different if observers are present. Also, released catch are usually not incorporated into catch estimates, even though there is the potential that high hooking mortality could result in high mortality of the released catch. Released fish mortality estimation from cage studies and tagging studies needs more attention because auxiliary data on depth caught and release condition are hard to collect but are important. For-Hire Sector Survey Design For some fisheries, the for-hire sector is responsible for taking most of the recreational catch, which is, in some cases, the majority of the total catch (Coleman et al., 2004). There are at least 10,000 registered charter vessels in the United States. In Alaska, 1,400 charter vessels landed over 60 percent of the reported recreational catch of halibut and lingcod in recent years, with this percentage reaching over 70 percent in southeast Alaska.2 In the Gulf of Mexico, charter vessels land an average of 70 percent of the recreational red snapper catch (35 percent of the total directed catch), and as a result, a charter vessel moratorium program is being implemented to limit the potential catch from this sector (Gulf of Mexico Fishery Management Council, 2004a). Due to the large potential contribution of this sector to total removals, it is important that it be monitored accurately. Several years ago, it was recognized that the MRFSS was not effective for assessing the for-hire sector, and consequently, there are now alternative surveys in place in most states for collecting data from the for-hire sector. The most important of these are the For-Hire Survey and the Party Charter Survey (see Appendix B). Both of these surveys are designed to ascertain fishing effort and CPUE data, just as the original MRFSS aims to do. However, the major change is that effort is determined from boat directory telephone lists instead of the RDD frame. Use of these list frames is much more efficient than use of the RDD frame. This allows for a greater sample size specific to this sector. In addition, the potential for bias is eliminated since fishing effort for both 2 Personal communication, Allen Bingham, Alaska Department of Fish and Game, Sportfish Division, Anchorage.

OCR for page 57
REMOVAL ESTIMATION 73 local and nonlocal anglers can be estimated directly from the charter companies. There is no need to adjust for effort by out-of-frame anglers. The current surveys are capable of monitoring the for-hire sector better than what was achieved through the MRFSS. However, design issues associated with these surveys still exist. The estimation of CPUE still relies on intercept sampling at points of landing; therefore, they are still subject to the problems discussed in the previous chapter about interviewer choice. In fact, intercept issues for this sector may be an even bigger problem since cluster effects arise from multiple anglers partic- ipating in the same fishing experience. These effects can be significant and must be accounted for in the estimation for this fishing mode. Another difficulty in surveying the for-hire sector is that operations range from very small to very large, with some being transient. License frames for this sector are likely to suffer from some incomplete coverage, especially for the small or transient operations. An alternative to the current sampling surveys is the use of mandatory logbooks or diaries of all the fishing effort and catch on for- hire boats, as a condition of the vessel's license. The captain would be responsible for filling out the logbooks as fishing progressed each day, and he or she would be required to turn the logbooks in on a timely basis as a condition for continued licensing. Having the license of the vessel tied to the logbook requirement would be the mechanism to achieve a complete list frame for this sector. Therefore, a census of this sector theoretically is possible because the population of charter and head boats is defined more easily than that of the total angler population. Also, the captains and crew generally have a greater knowledge of the local fish species and could provide more reliable catch data, including species identification and the location of catch. The question of whether to use a survey or census for the for-hire sector is not a new one. In 2001, the Recreational Technical Committee (RTC) of the Atlantic Coast Cooperative Statistics Program (ACCSP) undertook a one-year assessment of three programs designed to measure the fishing activity of the for-hire sector of the South Carolina marine fishery (Ditton et al., 2002). The purpose was to provide information for determining the best and most acceptable method of collecting data from the for-hire sector that could be adopted as a standard by ACCSP. They reviewed (1) the MRFSS; (2) the mandatory South Carolina Charter Logbook Survey, combined with the NMFS Headboat Logbook Survey; and (3) the NMFS Vessel Directory Telephone Survey (VDTS), com- bined with the MRFSS intercept component (with augmented sampl- ing)--the precursor design for the For-Hire Survey.

OCR for page 57
74 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS RTC noted marked improvements with VDTS and the South Carolina logbook methods when compared to the MRFSS (Ditton et al., 2002). RTC found that the advantages of the logbook program were that it had the most credibility of the three methods with the public, it had the best timeliness of data availability, and it had the most complete sampling frame and coverage. It also found that the logbook program sampled 99 percent of the for-hire vessels in South Carolina and was successful because it was mandatory, enforceable (with measurable enforcement actions), and financially sustainable. The disadvantages of the South Carolina Charter Logbook Survey were the possibility of an incomplete sampling frame because of potential rogue vessels, underreporting on vessels, and lack of biological sampling dockside. RTC provisionally recommended the VDTS program over the logbook program because of implementation issues it anticipated for a coastwide program--primarily lack of funding and commitment of agencies to enforcement and validation. At least some of the potential problems identified for the South Carolina Charter Logbook Survey could be addressed through a logbooklicense linkage, as described above. Although RTC was concerned that there could be implementation issues for logbook programs, this will be true with any fundamental change in sampling protocol. Given the magnitude of the for-hire sector in some regions and the potential scale of fishery removals for this sector, the committee finds compelling arguments for the use of mandatory logbooks as the source of catch and effort data for the for-hire sector. Furthermore, ACCSP found that recreational and charter fishing constituents along the Atlantic coast have a strong desire to participate more actively in data collection (Loftus et al., 1999), and the committee heard similar comments during public testimony.3 Not only can mandatory reporting from the for-hire sector increase the public's acceptance of the credibility of recreational catch statistics, it may help to facilitate "ownership" of these data by the for-hire sector. If the data they are supplying are a component of the final estimation, there may be fewer criticisms of these final estimates. The use of logbook data is particularly important for fisheries in which fishery-independent surveys are conducted infrequently or not at all because these data will be an essential component of stock assessment calculations. The committee recognizes that logbooks should not be re- 3The testimony from a small number of individuals on this topic may not repre- sent the whole fishing community; however, this testimony originated from the largest national organization of charter boat operators.

OCR for page 57
REMOVAL ESTIMATION 75 quired by more than one level of government (state, regional, federal, and international), and agencies must be coordinated to avoid the burden of duplicate reporting. The committee sees the state as the appropriate level of implementation for this requirement, with adherence to national reporting standards and program coordination at the national level. Validation of data acquired through any source is an issue of concern, and it would be no less so for a mandatory logbook program. The data collected through logbook programs will be reliable only if there are strict verification and enforcement components of the program. Since the information obtained from the logbooks is owner supplied, there is the need for verification for both CPUE and effort. Effort and kept catch could be checked by dockside inspection of angler parties and their catch. However, accurate and timely logbook submission as a condition of license is important. While the normal process of validation through creel surveys and random sampling of individual clients on the vessels could still be used, there would be direct and effective accountability because of the legal requirement for the logbooks, as well as the economic incentive associated with continued licensing of the charter operation. Also, the logbook program will serve as a participation record for any more detailed allocation discussions (e.g., the use of individual quotas for charter vessels, which is being contemplated in some jurisdictions). Finally, a mandatory logbook program provides a comparison vehicle for data acquired independently via offsite, random, individual angler-based or panel-based surveys. A for-hire logbook program represents a significant step in mon- itoring of this sector, but it will not solve all problems of monitoring. For example, accurate accounting and verification of catch-and-release activity will be addressed only partially through such a program. Alternative verification of catch and release via observers or electronic monitoring may be required. However, the committee views the for-hire sector as a business enterprise--the business being the connection of people and fishing opportunity. Therefore, this sector should be subject to a greater level of reporting than independent anglers, as a corollary of conducting business based on a public resource. Such a program will require additional resources to maintain a logbook-based data infrastructure. However, the substantial benefits of the program, as recognized in previous reviews and some existing programs, argue for its adoption and the commitment of resources to its implementation. Also, there are significant design issues associated with stratification by size of charter operation and geographic locality. These design issues are further addressed in Chapter 6.

OCR for page 57
76 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS ANALYSIS AND ESTIMATION TECHNIQUES Below is a detailed overview of analysis and estimation issues re- lated to the MRFSS that could revolutionize the way the survey is analyzed, especially at smaller spatial scales. It is deliberately presented at a higher technical level than some of the other sections because of the complexities involved. Generally, the analysis issues are focused on the use of auxiliary information to increase precision and the special prob- lems with estimation of subpopulations. In virtually all surveys, esti- mates are required not only for the population as a whole but for various subpopulations, called domains. For human populations, domains may be demographic groups (e.g., age, race, sex), occupational groups, or geo- graphic groups. For natural resource inventories, domains are typically geographic (e.g., county, state, state waters, federal waters) or ecological subdivisions (e.g., ecoregion, watershed). Geographic subpopulations are called areas. In fisheries, domains could be geographic areas or temporal periods. Often domains are not sampling strata so the sample size within domains is not pre-allocated but is determined randomly from the sampling. Three useful classes of domains are large domains, medium domains, and small domains, based on the sample sizes attained in those domains. Large Domains and Direct Estimation Large domains are likely to be sampling strata (i.e., predefined subpopulations that are sampled independently using predetermined sample allocations), but even if they are not, they are large enough to have a high probability of a large sample size. This large sample size ensures that standard design-based survey estimation procedures yield estimators of adequate precision. These standard estimators are called direct estimators because they use data only from the study units in the domain and time period of interest. These estimators have good design properties, and they are typically unbiased (or asymptotically unbiased), asymptotically normal, and allow for statistically consistent variance estimation and valid confidence intervals. All of these good statistical properties are justified by the randomization used in the probability sampling design and do not depend on the validity of any statistical model. This is the approach usually used in the current MRFSS analyses.

OCR for page 57
REMOVAL ESTIMATION 77 Medium Domains and Survey Regression Estimation Direct estimation is not reliable if the sample size is too small. In medium domains, the sample size is moderate but not extremely small. For such domains, if auxiliary information is available at both the population level and the sample level, it is often possible to construct a survey regression estimator (e.g., Cochran, 1977) with greater precision than that of the simple direct estimator. Such an estimator fits a global regression model to all of the survey data and predicts the responses for unsampled population elements using the fitted model. Survey regression estimators may be either model-based or model-assisted. Model-based survey regression estimators estimate the total for a domain by adding the responses for the sampled elements to the predicted responses for the unsampled elements. Such estimators are highly efficient if the model is right but can be biased and even inconsistent if the model is wrong. On the other hand, a model-assisted survey regression estimator predicts all elements using the fitted model and adds them up over the domain of interest. Since this prediction may be biased if the model is not specified correctly, the model-assisted estimator adds on a design-bias adjustment computed as the weighted difference between the observed and predicted responses over the domain. If the model is right, the estimator is highly efficient. The key result is that whether or not the model is right, the model-assisted estimator retains the good design properties of a direct estimator (i.e., it is asymptotically unbiased, asymptotically normal, and allows for consistent variance estimation and valid confidence intervals) (Srndal et al., 1992). The type of survey regression estimator depends on the types of available auxiliary information. With categorical covariates only, the survey regression estimator is a post-stratified estimator. With a single continuous covariate, the survey regression estimator could be a ratio estimator, classical regression estimator, or even a kernel or spline- based nonparametric survey regression estimator (e.g., Breidt and Opsomer, 2000). Generalizations to multiple covariates are also possible. To ensure the quality and timeliness of any of these survey regres- sion estimators, all covariates that enter the regression must be of high quality and must be readily available in a timely manner. Definitions of the covariates and protocols for their measurement should change as little as possible over time. Missing covariate information should be minimal. Indeed, all of the quality standards applicable to responses in the original survey are applicable to the covariates as well.

OCR for page 57
78 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS In the context of fisheries surveys, possible covariates for effort could include business-related covariates (e.g., bait sales, boat rentals) and weather-related covariates (e.g., precipitation, temperature, wave height). The business-related covariates could be difficult to obtain and use in an ongoing survey. Establishments vary in size and in the resources they devote to maintaining accounting records. Quality can vary considerably from establishment to establishment and from year to year. Definitions would need to be standardized, and cycles of data compilations would need to be synchronized. Thus, tracking down and compiling sales or rental data could be as difficult as conducting the original survey. The weather-related covariates, on the other hand, are readily available in a timely and consistent manner from a centralized source--the National Oceanic and Atmospheric Administration itself. Use of these weather data should involve minimal additional cost. Small Domains and Small Area Estimation The final domain classification is the small domain, called a small area in a geographic context (Ghosh and Rao, 1994; Rao, 2003). Here, direct estimators or model-assisted survey regression estimators are not sufficiently precise for the inferential problems of interest. Typically, the random sample size in a small domain or area is small and may be zero in some cases. There is no hope for direct estimation with such small sample sizes so small domain estimation problems lead to indirect estimators. Unlike direct estimators, indirect estimators use data from outside the domain or time period of interest to "borrow strength" across time or space, and the validity of these indirect methods depends on the correctness of the model specification. Perhaps the simplest small area estimator is the synthetic estimator in which all elements in a domain are predicted from a fitted global model relating the response variable to the covariates. The model borrows strength from the entire sample in the fitting of the regression model, which typically has common coefficients for all domains in the pop- ulation. The synthetic estimator can be computed for a given small do- main even if there are no samples in that domain and usually has very low variance since it is fitted on the basis of the entire sample, but it may have large bias if the model is incorrectly specified. A composite estimator attempts to trade off the low bias but high variability of a direct estimator with the high bias but low variability of a synthetic estimator by computing a combination of the two estimators.

OCR for page 57
REMOVAL ESTIMATION 79 The weights in this composite estimator can be chosen in an ad hoc way, such as by making the weight on the direct estimator larger if the sample size in that domain gets larger. The weights also can be chosen on the basis of a formal statistical model. The standard approach to formal composite estimation is to choose the composite weights as functions of the parameters from a fitted model. Two classes of models appear in the literature, depending on the type of available auxiliary information. Element-level models require auxiliary information for every sampled element (e.g., Battese et al., 1988), while area-level models require auxiliary information only for each small area. In either case, the small area model is hierarchical. In area-level models, much of the complexity of the survey design is averaged out, and nonnormality in responses tends to average out as well. Here, the focus is on the area-level model. Assuming that auxiliary information is available for each small area, the model describes the distribution of the direct estimates given the true domain parameters, and the distribution of the true domain parameters given the covariates. Usually, the direct estimate is modeled as truth + sampling error where the sampling error has a mean of zero and known variance. The true domain parameters are modeled with a global regression function of the covariates, plus domain-specific deviations from the global model. The domain-specific deviations are random effects that may have some correlation structure, such as temporal correlation structure in a time- indirect context or spatial correlation structure in a domain-indirect geographic context. The small domain model has two ways to borrow strength: globally through the regression fitted to all the data and locally through the temporally or spatially correlated random effects. Temporal correlation structure can be described with a state space model, special cases of which can include autoregressive moving average models (e.g., Brockwell and Davis, 1991). State-level unemployment estimates from the Current Population Survey, for example, combine a regression mod- el, a basic structural model for stochastic trend and seasonality, and an autoregressive moving average model for the correlated sampling errors (Tiller, 1992). Spatial correlation structure in an area-level model can be described with a lattice model, such as a conditional autoregression model (Cressie, 1993). Small area estimation models are fitted using standard statistical procedures, such as through estimation of variancecovariance parameters by restricted maximum likelihood or other methods followed by joint estimation and prediction of the fixed and random effects in the

OCR for page 57
80 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS model. This approach is known as empirical best linear unbiased prediction and has relatively straightforward computation that may be implemented using standard statistical software (e.g., the PROC MIXED function in SAS software, the lme function in S-Plus software), but these methods do not account fully for uncertainty since they treat the estimated variancecovariance parameters as known. Hierarchical Bayesian analysis is also possible. Prior distributions for all unknown parameters (including variancecovariance parameters) are assigned, and then numerical techniques, such as Markov Chain Monte Carlo, are used to compute posterior distributions of the unknown parameters given the direct estimates. Computation is more complex, but now these methods are routinely taught to statisticians in graduate school and routinely implemented in many government agencies that employ statisticians. Use of These Techniques in Angler Surveys In the context of angler surveys, use of auxiliary variables and small area estimation techniques might be applied to the effort estimates, the CPUE estimates, or the final catch estimates, perhaps after some transformation. Suitable auxiliary information for effort modeling may include weather-related covariates; suitable auxiliary information for CPUE may or may not be available. Identification of suitable covariates and specification of an appropriate regression model or models would be a critical part of a small area analysis. Even without suitable covariates, estimation of both effort and CPUE might be assisted by temporal, spatial, and multivariate correlation. The data are collected in temporal waves, and wave-to-wave or year-to-year correlation might be helpful in predicting current wave values. Also, the data are spatially explicit, so borrowing information from similar, nearby areas might help to improve predictions. Finally, the data are multivariate (catch by species), and the correlation structure among the different species might help in predicting individual species components. To conclude, the current estimation methodology used in the MRFSS is primarily direct estimation for large domains. Auxiliary information and survey regression estimation methods enter in minor ways, such as in some simple ratio adjustments and temporal pooling of estimators. It appears that with relatively modest additional resources, the MRFSS could add more formal survey regression methods, extending the inferential scale to medium-sized domains. Small area estimation would require a much greater investment of resources. This estimation method-

OCR for page 57
REMOVAL ESTIMATION 81 ology would require stronger assumptions, more sophisticated model specification (both in the regression model and in the covariance struc- ture), more detailed diagnostics, and heavier computations. However, the potential pay-off is enormous in that it extends the inferential scale to finer spatial resolutions, which seems to be what managers currently require. These recommendations will require a rethinking of the program management of angler surveys (see Chapter 6 on program management and support). CONCLUSIONS AND RECOMMENDATIONS The committee concludes that the current methods used in the MRFSS for sampling the universe of anglers and for determining their catch and effort are inadequate. Sampling of each group of anglers (i.e., private, guided, head boat, and charter boat) presents chal- lenges that can differ across the groups. Two complementary methods of sampling angler catch and effort are used in the MRFSS. One is onsite (i.e., intercepting anglers while they are fishing or at their access [landing] points). The other is offsite, which includes a variety of sampl- ing techniques for contacting anglers after they have completed their trips. Both onsite and offsite methods suffer from weaknesses that may lead to biases in catch and effort estimation. This necessitates major changes in both the design and analysis procedures. A comprehensive, universal sampling frame with national coverage should be established. The most effective ways to achieve this are through a national registry of all saltwater anglers or through new or existing state saltwater license programs that would allow no exemptions and that would provide appropriate contact and information from anglers fishing in all marine waters, both state and federal. Any gaps in such a program (e.g., a lack of registration in a particular region or mode, ex- emptions of various classes of anglers) would compromise the use of the sampling frame and, hence, the quality of the survey program. Future telephone surveys should be based on the above universal sampling frame. Dual-frame procedures should be used wherever possible to reduce sample bias. For example, if a state has an incomplete list frame based on licenses, the use of an additional sampling frame of the state's residents (e.g., RDD) would reduce the bias. The existence of a universal frame described above would make this approach unnecessary for offsite sampling, provided there are no exemptions. Complemented surveys

OCR for page 57
82 REVIEW OF RECREATIONAL FISHERIES SURVEY METHODS should be used more widely in regional surveys where reliable esti- mates are required for management of a small suite of very im- portant species at small regional scales. Panel surveys, which contact individual anglers repeatedly over time, should be considered in recreational fishing surveys to gather angler trend data and to improve the efficiency of data collection. This is especially true for the telephone portion of the MRFSS. Internet surveys should be considered for their potential use in recreational fishing surveys, especially in panel surveys, as a way for anglers to submit information. They could be used in the for-hire surveys, in private angler surveys like the MRFSS, or in social and eco- nomic surveys. In most cases, charter boat, head boat, and other for-hire recreational fishing operations should be required to maintain log- books of fish landed and kept, as well as fish caught and released. Providing the information should be mandatory for continued operation in this sector, and all the information should be verifiable and made available to the survey program in a timely manner. Onboard observers could be used on a sample of vessels to verify logbook information. A sample survey may be more appropriate in fisheries where the for-hire sector is a small component of the catch or where verification and enforcement are particularly problematic. The reported release alive of captured fish (catch and release) is increasingly common in many marine recreational fisheries. Although released fish suffer lower mortality than retained fish (the mortality of retained fish is, of course, 100 percent), there still is some mortality, and in some cases, it can exceed 50 percent. The survey fails to provide a valid and reliable method of adequately accounting for fish caught and not brought to the dock (including fish released alive or dead, as well as fish caught for bait or given away before reaching the dock). This shortcoming affects estimates of catch and total removals. Current analysis procedures used in estimation for the MRFSS do not exploit the current knowledge of finite population sampling theory. The current estimates are particularly deficient when applied to small areas because they do not use information in adjoining areas or time periods, nor do they consider relationships between species that occur together. Therefore, they are of lower precision than would be possible if this information were used. Improvements in these estimates would be of great use to managers who need to make quick decisions concerning spatial areas that are smaller than typical in the early years of the MRFSS.