Geoscience Reference
In-Depth Information
Table 7.1
Monte Carlo Simulation Procedure Used to Estimate the Power of
Population-Monitoring Programs to Detect Trends
Step
Procedure
1. Basic structure of the monitoring program is defined (i.e., number of plots
surveyed, survey frequency, and a series of survey years).
2. Deterministic linear trends are projected from the initial abundance index on
each plot over the series of survey years.
3.
Sample counts are generated at each survey occasion across all plots and for each
trend. Sample counts are random deviates drawn from a normal distribution
(truncated at 0) with mean equal to the deterministic projection on a particular
monitoring occasion and with a variance approximated by the standard
deviation in initial abundance (constant variances over time).
4.
The slope of a least-squares regression of sample abundances versus survey
occasion is determined for each plot and each trend.
5.
The mean and variance for slope estimates are calculated across plots for each
trend.
6.
Whether the mean slope estimate is statistically different from zero for each
trend is determined.
7.
Steps 1 through 6 are repeated many times, whereupon the proportion of
repetitions in which the mean slope estimate was different from zero is
determined. The resulting proportion represents the power estimate, which
ranges from 0 (low power) to 1 (high power) and indicates how often the survey
program correctly detected an ongoing trend.
more, these estimates can be integrated with power analyses to provide general
guidance on sampling protocols that animal ecologists can use to design robust
monitoring programs for local populations.
To this end, count series of local animal and plant populations that ex-
tended more than 5 years were obtained by examining 25 major ecology jour-
nals published from 1940 to the present (nonwoody plants are also presented
here because animal ecologists often must monitor plant populations in the
course of their animal studies). Variability of each count series thus obtained
was estimated by dividing the standard deviation of the counts by the mean
count to determine the coefficient of variation (CV). To remove trends in the
counts (which might have inflated variance estimates), the standard deviation
was determined from the standardized residuals of a linear regression of counts
against time. Furthermore, because the variability of a time series is related in
part to its length (Warner et al. 1995), a 5-year moving CV (similar in concept
to a moving average) was calculated for each count series. (However, most
