STATUS OF THE "MONITOR" WEB SITE
July 15, 2003
Patuxent has temporarily disabled its websites related to power analysis and program MONITOR, until we have had the opportunity to re-craft guidance on use of power analysis and estimation of sample size requirements for monitoring programs. We hope to engage a broad spectrum of statisticians in a dialogue on this topic. Our decision reflects our growing awareness (and wariness) of the complexity of power analysis in general, and also acknowledges some specific limitations of MONITOR for many real-world applications. Our chief, immediate concern is that many users of MONITOR may be unaware of these limitations and may be using the program inappropriately.
Below are comments from one of our statisticians on some of the aspects of MONITOR that users should be cognizant of:
"There are numerous issues with how Program Monitor calculates statistical power and sample size. One issue concerns the default option whereby the user assumes independence of plots or sites from one time period to the next. If you are randomly sampling new sites or plots each time period, then it is correct to assume independence (assuming that finite population correction factor is not an issue, which depends on how many plots or sites you are sampling, relative to the total population size of potential plots or sites). If you are sampling the same plots or sites repeatedly over time, however, then the default option in Program Monitor is unlikely to give a correct calculation of statistical power or sample size. If plots or sites are positively autocorrelated over time, as is usually the case in biological surveys, then Program Monitor will underestimate sample size, or conversely, it will overestimate the statistical power. The correct sample size estimate is likely to be greater, and depending upon the amount of autocorrelation, the correct sample size could be vastly greater to achieve a stated power objective.
A more fundamental issue concerns the null model one chooses for the trend in population growth. Program Monitor assumes a relatively simple linear trend in population growth, but this is a controversial issue, because there are potentially an infinite number of models one could use. If pilot data are available, then it may be possible to estimate autocorrelation and try to make some choices concerning the type of model to use as the null model for a power calculation, but regardless of how you decide to proceed, it would be a good idea to consult a statistician to determine an approach that fits your needs and data. No matter what additional flexibility is built into the modeling, however, it will always be possible to posit the existence of further structure which if overlooked will produce misleading results. For a pertinent discussion of some of these issues, please see Elzinga et al. (1998). Although this reference deals specifically with plant populations, the fundamental statistical issues are similar whether you are sampling plant or animal populations.
Literature Citation: Elzinga, C.L., D.W. Salzer, and J.W. Willoughby. 1998. Measuring and monitoring plant populations. BLM Technical Reference 1730-1, Denver, CO. 477 pages."
If you find that program MONITOR meets the needs of your situation and you are unable to wait until we update our site, please contact Sam Droege (sam_droege@usgs.gov 301-497-5840), who can email you a windows version of the program.