Citation:

Blocksom, Karen A, Eric W Fox, AND A. Olsen. Adjusting for non-response in National Aquatic Resource Surveys. Annual meeting of the Pacific Northwest Chapter of the Society for Freshwater Science, La Conner, WA, November 07 - 09, 2017.

Impact/Purpose:

The National Rivers and Streams Assessment (NRSA) uses a probability-based survey design to estimate condition of the streams and rivers across the 48 contiguous states of the US. Of the 2580 sites that were evaluated and considered within the scope of the NRSA 2008-2009 survey, 656 could not be sampled due to inaccessibility, landowner denial, or other reasons. These sites represented almost 21% of the total stream length of ~2.43 million km included in the survey, but the current approach is to report only on the stream length represented by sampled sites. In order to better account for these unsampled sites in reporting, Karen Blocksom, Eric Fox, and Tony Olsen of EPA-WED applied an approach to adjusting the weights of sampled sites for the NRSA 2008-2009 survey. Their approach was to group sites into similar classes and then redistribute the weights of unsampled sites among sampled sites within those classes. They used a predictive model of macroinvertebrate condition to look for bias in predicted values by different categories, and determined that aggregated ecoregions and ownership type (private/public) were important grouping categories. Adjusted weights were then used in estimating stream length in good, fair, and poor condition for several condition indicators measured in the survey, including macroinvertebrates, fish, physical habitat, and water chemistry. After applying this approach, they compared the results to the original estimates of condition. With the revised estimates based on adjusted weights, the full stream length included in the survey design is represented in the estimates of condition based on sampled sites. Although the estimated stream length increased across all indicators and all condition classes, there were only minimal changes to the estimates of percent stream length. This presentation focused on the results from the Xeric and Western Mountains aggregated ecoregions to reflect the location and interests of the meeting attendees in the Pacific Northwest. This Abstract contributes to SSWR 3.01A.

Description:

The National Rivers and Streams Assessment (NRSA) is a national-scale probabilistic survey. As part of the survey design, each site is associated with a weight reflecting the number of stream kilometers in the target population represented by that site. During each round of the survey, many sites are evaluated but not sampled, with some subset of these sites considered part of the target population (streams and rivers within the 48 contiguous states with flowing water during the survey index period). These unsampled target sites are considered unit non-responses, and occur for various reasons, including landowner denial and inaccessibility of sites. The current approach to addressing the issue of non-responses is to report on only the sampled population, which leaves a potentially large proportion of stream kilometers excluded from the population estimates of condition. For this study, we employed an approach that involves categorizing sites into classes and assuming non-responses occur at random within those classes. Then the weights of sampled sites are adjusted to account for non-response sites within classes. Population-level estimates of condition are then calculated using the adjusted weights and the condition data from sampled sites. For this presentation, we used data from the NRSA 2008-9 survey to determine an appropriate classification of sites, performed this weight adjustment, and compared the results to the original estimates based only on the sampled population, focusing on the Western Mountains and Xeric ecoregions. The estimated stream kilometers increased dramatically across condition classes with the adjusted weights, as expected, but we observed only very small changes in percentage estimates, regardless of indicator or condition class.

Record Details: