Northern Prairie Wildlife Research Center
We reviewed published and unpublished studies of nest success from parkland and grassland regions in 3 provinces (Alta., Manit., Sask.) and 2 states (N.D. and S.D.; Fig. 1) where long-term data were available for 5 species of upland-nesting ducks: blue-winged teal, gadwall, mallard, northern shoveler, and northern pintail. We included only nonisland sites where it was reported no organized predator control was conducted. (Beauchamp et al. 1996 analyzed nest success on islands and at sites with predator control.)
|Fig. 1. Locations of study sites used in analyses of nest success of upland-nesting ducks in the Prairie Pothole Region of North America during 1935-92.|
To ensure comparable point estimates of nest success, in space and time, we excluded studies from our analysis if data had been combined for >1 study site or for ≥2 years at 1 site. These 2 criteria resulted respectively in the exclusion of only 7 and 9 nest success estimates. Further, we did not include estimates of nest success based on <10 nests. We used 143 point estimates of nest success from 37 sources; the earliest study was conducted in 1935 and the most recent in 1992 (Appendix A).
We created 2 datasets, 1 with estimates of nest success for each species (unpooled) at each site in each year (n = 232, Appendix A), and another with nest success pooled across species at each site in each year (n = 143). To test for differences in nest success among species, we analyzed unpooled data. However, multiple estimates of nest success (i.e., from >1 species) at a given site in a given year are not likely independent because all species would be subjected to similar environmental conditions. For this reason, we used pooled data in all other analyses to reduce effects of nonindependence. The pooled data is also valuable in that more study sites are represented (67 compared with 49 in the unpooled dataset), because some authors did not report nest success separately for each species.
One problem in undertaking a temporal analysis is that incomparable estimators of nest success have been used over time. The "apparent" estimator used in older studies is almost always biased high (Mayfield 1961, 1975; Miller and Johnson 1978), but the contemporary Mayfield method more accurately estimates "true" nest success (Mayfield 1961, 1975). Eleven studies (1977-90) that we used in our analyses reported Mayfield estimates (Appendix A). When only apparent nest success was reported, we converted it to Green's (1989) "Mayfield-equivalent" (Appendix A), thereby enabling us to examine long-term variation in nest success. Johnson (1991) reported no directional bias for Green's transformation provided the probability of finding nests did not vary with nesting stage. This equal chance criterion might not be met if a study area was searched frequently and thoroughly. In this situation, nest success would be underestimated because Green's transformation would overcorrect for nest exposure (Johnson 1991). Because early studies lacked the efficiency of the cable-chain drag (Higgins et al. 1969) for nest searching, these sites were likely covered less systematically than those searched later and would not, therefore, violate the assumption of Green's transformation. Thus we considered it reasonable to compare Green-transformed estimates of apparent nest success from earlier studies with Mayfield estimates from later studies.
We treated sites as random samples in the analyses and assumed that all estimates of nest success were equivalent. The problem with the assumption of equivalence is that most nest abandonments were counted as nest failures in the older studies, whereas many recent studies did not use abandoned nests to estimate nest success if the abandonments were thought to have been caused by investigators.
We used conserved soil moisture (CSM) indices, available from other broad-scale studies conducted in the Canadian part of the Prairie Pothole Region (Bethke and Nudds 1993), to estimate yearly and regional variation in climatic conditions. (Similar data for the U.S. portion of our study area were not available.) Conserved soil moisture is a weighted mean of total precipitation in the 21 months preceding 1 May in any given year (Williams and Robertson 1965, Boyd 1981). More weight is given to precipitation in fall and winter because rainfall during the summer growing season does not contribute as much to persistent soil moisture (Boyd 1981). We estimated CSM for each of the 31 Canadian study sites using precipitation data from the nearest weather station(s) (Mon. Rec., Atmos. Environ. Serv., Environ. Can., Ottawa, Ont.) for each year that nest success data were available. For study areas >50 km from the nearest weather station, we used mean CSM of the closest 2-3 stations.
We conducted linear regression and analysis of covariance (ANCOVA) using the General Linear Model (GLM) procedure of SAS (SAS Inst. Inc. 1985, Freund et al. 1986). Frequency distributions of the residuals from the linear models departed from normality; a log10 transformation best normalized the data and was used in our analyses.
We conducted preliminary analyses to compare the effects of using different combinations of unweighted data, weighting by the number of nests, including studies with 10 to 20 nests, and truncating the dataset at various years (i.e., excluding the 1930s, excluding the 1940s, etc.). We found that our conclusions were robust and that the statistical parameter estimates differed only slightly. Such uniformity was not surprising, because of the large sample size and the large amount of inherent variation in the data. Further, it is difficult to determine a weighting scheme a priori, or to justify one a posteriori. We chose not to weight the data, to increase our sample size by including studies with ≥10 nests, and to include all years for which we could find data.
To test whether nest success declined over time, we regressed pooled nest-success estimates against year. To determine whether precipitation explained additional variation in nest success (after accounting for yr), we conducted a separate regression of nest success on year using only data from Canadian sites for which we had estimates of CSM. We then regressed residuals against CSM. We conducted a full ANCOVA to detect interactions among year, species, and region. If there were none, we conducted separate ANCOVAs to test for differences in nest success among species (using unpooled data) and between regions (using pooled data) (Freund et al. 1986:202-203). Before testing for differences (in intercepts) between species or regions using ANCOVA, we tested the assumption of homogeneity of slopes among groups (Freund et al. 1986:200-205).
To examine differences among species, we conducted multiple comparisons of least-squares means (Freund et al. 1986), adjusted for year effect, using a Bonferroni adjusted α = 0.01 per comparison to ensure an overall error rate of <0.05.