Estimating Nest Success Rates & the Number of Nests Initiated by Radiomarked Mallards Carl James Schwarz cschwarz@stat.sfu.ca Estimating Nest Success Rates & the Number of Nests Initiated Typical Sampling Protocol: - search for viable nests - follow detected nests until success or failure - DANGER: length biased sampling - Estimators to account for this bias: * Mayfield (1961, 1975) method - assumes that daily survival is constant over nesting period Farnsworth et al (2000) – heterogeneity -> bias Used Kaplan-Meier with staggered entry to allow variation - measures survival from first encounter rather than initiation Pollock and Cornelius (1988), Bromaghin and McDonald (1993) model encounter probabilities. Estimating Nest Success Rates & the Number of Nests Initiated Observed number/density is an underestimate because not all nests detected Miller and Johnson (1978), Johnson and Shaffer (1990) divided observed successful nests/prob of sucess N̂ MJ xJ Ŝ1Ŝ2 L ŜJ - only uses successful nests and hence loss of information Bromaghin and McDonald (1993a) provided a more general approach where the sampling design that locates nests is also modeled. Estimating Nest Success Rates & the Number of Nests Initiated Our Study: - 1993–2000, 27 study sites, Western Canada - in 1997, 135 female mallards captured, radio tagged - radiomarked females located 2–3 times daily - when a nest found, initiation date determined. - nest followed until success or failure Estimating Nest Success Rates & the Number of Nests Initiated Statistics: xi Number of nests seen for the first time at age i. ri Number of nests that survived from age i to age i + 1. Ri Number of nests at risk at age i. Usually, Ri = ri–1 + xi, but censoring possible Age (days) Ri ri 1 16 12 2 26 19 3 43 40 4 64 60 + … 21 more days of data …. Total 1,868 1,752 xi 19 17 25 24 175 Estimating Nest Success Rates & the Number of Nests Initiated Parameters: N Si pi i Total number of nests initiated by the radiomarked sample. Probability a nest survived from age i to i + 1, given the nest survived to age i. Probability that a nest was newly found at age i, given the nest has survived to age i and has not been previously found. Probability a nest was newly found at age i. i 1 1 p1 ; 2 1 p1 S1 p2 ; i pi S j 1 p j . j 1 Estimating Nest Success Rates & the Number of Nests Initiated Likelihood model – 2 components (1) How nests are found? N J N xi xi J J L1 i 1 i x1 ,..., xJ , N xi i 1 i 1 . i 1 (2) How nests survive: Ri ri Ri ri L2 Si 1 Si . i 1 ri J 2J + 1 parms to fit to 2J sufficient statistics – one restriction pJ = 1 Estimating Nest Success Rates & the Number of Nests Initiated Key Assumptions - radiomarked birds must be representative of the overall population. - radiomarking must not affect subsequent nesting behavior. - newly found nests must be aged correctly. - detection and survival probabilities will depend only on age and not on time. For example, a nest of age i discovered on 15 May must have the same probability of subsequent survival as a nest of age i discovered on 15 June. - radiomarked females cannot leave the study site and nest where they are undetectable by telemetry surveys. - the act of finding a nest must be independent of subsequent survival. Estimating Nest Success Rates & the Number of Nests Initiated No closed forms except for estimator of N which takes the form: N̂ MLE x• ̂ • 1 ̂ nests • , where J x• xi i 1 J is the total number of nests observed and J i 1 i 1 j 1 ̂ • ̂ i p̂i Ŝ j 1 p̂ j i 1 This is a “Horvitz-Thompson” like estimator Estimating Nest Success Rates & the Number of Nests Initiated Summary of model fitting and parameter estimates Upper and N̂ lower 95% logL K AICc AICc Modela (SE) CI for N 232.0 {N p3t S} 211 260 –132.6 5 275.2 0.0 (12.0) 243.4 {N p3t S2t} 216 282 –131.6 6 275.2 0.0 (16.4) 242.6 {N p S} 217 Inf –145.8 3 297.6 22.4 (15.9) 232.7 {N pt S} 212 261 –123.1 26 298.5 23.4 (12.2) Model 237.1 214 272 averaged (15.3) 1.91 Nests/female 1.73 2.19 (0.12) “Mayfield” 236 178 320 w 0.505 0.495 0.000 0.000 Estimating Nest Success Rates & the Number of Nests Initiated Model estimates from two “best” models from previous table Estimated Estimated detection daily survival rates (SE) rates (SE) 1–3 4–6 ≥7 1–10 ≥11 Modela days days days days days 0.10 0.22 0.24 0.94 0.94 {N p3t S} (.013) (.025) (.041) (.005) (.005) 0.10 0.21 0.24 0.93 0.94 {N p3t S2t} (.013) (.025) (.041) (.01) (.006) Estimating Nest Success Rates & the Number of Nests Initiated Summary: - Mayfield methods can be extended to account for non-constant survival - Mayfield methods can be extended to model nest discovery CRUCIAL – some nests must be discovered at AGE 1! - Mayfield-related methods do not use entire information present Power analysis shows that can reduce number of nests by 6x using our method! - Goodness of fit passed for this dataset - because of HT form, robust to heterogenous survival - because of HT form, biased by heterogeneous discovery - We are relatively confident that assumptions are nearly satisfied