You are here:
Statistical Power for Trend Detection Under Alternative Panel Designs for Surveys over Time
Citation:
Olsen, Tony, Tom Kincaid, AND Steve Paulsen. Statistical Power for Trend Detection Under Alternative Panel Designs for Surveys over Time. 11th National Monitoring Conference, Denver, Colorado, March 25 - 29, 2019.
Impact/Purpose:
Monitoring programs imply that sampling is conducted over time, typically years, to determine if a trend is present. For probability-based survey designs, the program must determine whether the same sites should always be sampled or if it is better to sample new sites (all or some) in each year sampling occurs. The answer to the question depends on the statistical power that the "panel design" has for detecting trends. We provide information that monitoring program managers can use to help them make decisions on the panel design for their program.
Description:
The National Aquatic Resource Surveys (NARS), and many other monitoring programs that use probability-based survey designs, have an objective to investigate trends over time. Questions that immediately arise are what is meant by trend and what is the best survey design over time to detect that trend. Our focus is not on trends at individual sites but trend trajectories of population estimates from surveys and their summary as linear trends over time. While it is common to monitor the same sites over time, many alternative panel designs are available where sets of sites are monitored on different schedules. A panel of sites is a collection of sites that are all sampled at the same points in time. Common panel designs are the always revisit, serially alternating, rotating panel and split panel designs, although many others exist. The panel design that has good power to detect a linear change in the trend trajectory depends on the relative contributions of variation across sites, across year, site by year interaction, local variation and variability of linear trends for individual sites. Based on estimates of these sources of variation from the NARS, we investigate power for a collection of panel designs that have been use or considered for use by monitoring programs. We also discuss how the choice of a panel design involves other considerations related to implementation and impact on non-response.