Upper Midwest Environmental Sciences Center
Home/ Overview/ Science Programs/ Data Library/ Products and Publications/States/ Rivers/Teachers and Students/ Links/ Contact/ Search |
Estimated trends in LTRMP data are adjusted for year and selection probabilities effects. Years (generally equivalent to sampling events) are treated as primary sampling units (i.e., as random) while selection probabilities are estimated following procedures in Estimating Means and Standard Errors from LTRMP Survey Data. Where adjustment has nontrivial influence on the estimated standard error of a trend, estimates relying on poststratification (e.g., trends in upper Pool 4) are adjusted for poststratification. Models of binomial and multinomial outcomes assume means vary linearly on the log odds scale, while models of count outcomes assume means vary linearly on the log scale; data are not transformed prior to trend estimation.
Temporal correlation is estimated for those variables that may reasonably be presumed to be correlated at the annual scale (i.e., for annual means of fish, macroinvertebrate, and vegetation outcomes but not for annual means of water constituents). If an estimate of first-order temporal correlation among annual means,, is significantly different from zero (using the Durbin-Watson statistic; Myers 1990) at P < 0.10, then the standard error associated with the trend estimate is divided by. Myers, R. H. 1990. Classical and modern regression with applications. Second edition. PWS-Kent, Boston. Contact: Further information about estimating trends in LTRMP data may be obtained from Brian Gray, LTRMP statistician, Upper Midwest Environmental Sciences Center, La Crosse, Wisconsin, at brgray@usgs.gov. |