Northern Prairie Wildlife Research Center
The variance of an estimated DSR is inversely proportional to the number of exposure days (Johnson 1979), and (for certain species) differences in numbers of exposure days among some habitat classes, study areas, and years greatly influenced the precision of our DSR estimates. We used a linear model (PROC GLM, SAS Institute, Inc. 1989) fitted by the method of weighted least squares, with weights equal to the number of exposure days (Snedecor and Cochran 1980) to overcome imbalance due to small numbers of exposure days in some categories. The initial model included effects for area-year, half-area-year within area-year, habitat, species, and interactions between area-year and species. We tested for significant (P < 0.05 throughout, unless otherwise noted) effects by extending the method of Johnson (1990) to multiple effects. We subsequently removed interactions between area-year and species because they were not significant and fitted a reduced model involving the remaining effects (Appendix C). Analysis indicated all remaining effects were significant (P < 0.05) except species (P = 0.16). We chose to leave species effects in the model because early-nesting species have lower nest success than late-nesting species (Klett et al. 1988) and we wished to retain this option in our analysis. Our assumption that habitat effects were similar among area-years and species precluded comparisons of nest success by habitat on individual study areas.