43 



2.2.2 



DESIGN AND STATISTICAL CONCERNS FOR MONITORING 



Dr. Roger H. Green 



Department of Zoology 



University of Western Ontario 



London, Ontario 



CANADA, N6A 5B7 



INTRODUCTION 



The emphasis of this presentation is on concerns for long-term monitoring and impact detection. There are three 

 themes: hypotheses and response variables, basic questions of study design, and problems in study design related 

 to natural variation. 



HYPOTHESES AND RESPONSE VARIABLE 



In designing any environmental study there should be a logical sequence of: purpose = > question = > 

 hypotheses = > model = > study design statistical analysis and tests of hypotheses (Green 1984). Hypotheses 

 are usually hierarchical, especially when there is natural variation. For example, the null hypothesis might be 

 "natural variation only" and the alternate hypothesis "natural variation plus impact of development." Criteria for 

 a good response variable include (1) relevance to impact effects and sensitivity of response, (2) intrinsic economic 

 or aesthetic value, (3) low natural temporal and spatial variation, and (4) estimation quickly, precisely, and at 

 low cost (Green 1987). Carney (1987) discusses appropriate response variables for studies of impacts related 

 to more offshore petroleum exploitation. Bayne et al. (1988) provide examples of diverse response variables 

 ranging from biochemical and cellular to community levels, within one marine pollution study. Further discussion 

 of options and constraints in choice of response variables for environmental impact and monitoring studies is 

 found in Green (1989). We should always be open to the potential of new response variables, such as isotope 

 ratios in biological tissues or community level indices, in addition to well-established and accepted response 

 variables (Green 1987, 1989). 



BASIC QUESTIONS OF STUDY DESIGN 



Emphasis is on three related questions. First, how many samples should one take? Some replication (n) at each 

 location (1) and time (t) is always desirable, so the total number of samples would be equal to nit, with n usually 

 3 or more. Replication should be sufficient to provide at least 10 error degrees of freedom in any statistical test. 

 Second, what number of samples is needed to perform tests of hypotheses with a given power to detect real 

 impact effects of a given magnitude? A pilot study can provide information about natural variation so that the 

 error variance can be estimated, and then the necessary number of samples can easily be calculated given the 

 error variance, the effect to be detected, and the desired Type I and II error levels (the Type I level usually = 

 0.05 and the Type II level = 1 - power) (Green (1979, 1989, in press, and references cited therein). Third, what 

 is the appropriate level of variation in the study design for calculating the error variance to be used in tests of 

 hypotheses about impact? We commonly use the variation among replicate field samples (at a given time and 

 place) for this purpose, but in many instances this is inappropriate (Green 1984). Hurlbert (1984) has referred 

 to this as "pseudoreplication". Often it is most appropriate to use re-sampled sites as replicates rather than re- 

 randomized samples within sites, and the error term for tests is often calculated from the interaction between 

 among-sites variation and among-times variation at given sites, i.e. variation in time trends among sites (Green 

 1989). 



STUDY DESIGN RELATED TO NATURAL VARIATION 



Obviously it is necessary to have knowledge of the extent and nature of natural variation in any particular 

 situation, which requires a pilot study, or previous research on that system by other workers, or a long-term 

 baseline study. Some natural variation can be "stratified out" by an appropriate study design (Green 1984), for 

 example among-site variation or within-year temporal variation (e.g., seasonal, diel, tidal). However some within- 

 year temporal variation can not be stratified out (e.g., seasonal patterns vary from year to year), and year-to- 



