characteristics of study sites may change in longer-term studies, and thus matching may be unreliable.

When data are lacking before an impact, the control-impact design may be used. This type of study differs from the BACI design only in the lack of pre-impact data. As in the BACI design, if a significant difference is attributed to the impact of a perturbation the assumption is that nothing else could cause a change of that magnitude (Manly 2001). Before-after designs can be used when data from a control area cannot be obtained. A change immediately following an impact is assumed to be a result of the impact and not from some other cause. In the absence of data from control areas, the attribution of cause may be difficult to support, unless the impact is large and easily attributable to the cause. For example, a decline in bird abundance following the construction of a wind-energy facility might be attributed to the facility by finding large numbers of bird carcasses killed by turbines. In the absence of strong corroborative evidence, attributing the change in abundance to the wind-energy plant may be difficult to defend.

The impact-gradient design may be used for quantifying impacts in relatively small assessment areas with homogeneous environments (Anderson et al. 1999; Manly 2001). With this design, an effect is assumed if it appears to be reduced as the distance increases from the source of the impact (Manly 2001). The most important assumption made when using the impact-gradient design is that the environment is homogeneous. Homogeneity is relatively uncommon in the environment and the analysis of data resulting from this study design should take spatial correlation into account (Manly 2001). For example, wind turbines are typically placed on the windiest sites available in a wind-resource area, such as ridge tops. Thus, moderating environmental conditions as a function of distance from the turbines may create subtle differences in the characteristics of the sites that could mask impacts.

Morrison et al. (2001) suggested improving observational studies by using several general approaches to study design that can increase precision without requiring increased replication. Their suggestions include:

  • Vary sampling effort (or apply treatments) within homogenous groups of experimental units (blocking).

  • Measure non-treatment factors (co-variates) and use analysis of covariance when analyzing the response to a treatment to consider the added influence of variables having a measurable influence on the dependent variable.

  • Refine experimental techniques, including greater sampling precision within experimental units (Cochran and Cox 1957; Cox 1958).

Mensurative studies involve making measurements of uncontrolled



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