Sources of bias in observations of Murre provisioning behavior

ABSTRACT Although many studies involve observations of parent birds feeding their young, few investigators have attempted to quantify possible sources of bias associated with such observations. To address this issue, we observed the provisioning behavior of Thick‐billed Murres (Uria lomvia) at a breeding colony on Coats Island, Nunavut, Canada from 2004 to 2007. We also attached electronic recorders that indicated every return to the colony and, for some prey items, allowed us to determine whether they were correctly identified based on the profile of the dive preceding delivery. We recorded when the bird arrived without a fish and, because our electronic recorders did not impact feeding rates, we were able to convert the number of arrivals into feeding rates. Of 3744 arrivals observed by the recorders, 13% of arrivals were missed by the observers at the colony, 13% were unidentified and 5% were misidentified. The proportion of missed feedings (AIC weight = 0.46) and unidentified prey (AIC weight = 0.48) increased with decreasing light level, whereas the proportion of misidentified prey increased with increasing light level (AIC weight = 0.30). Distance from the observation blind to the breeding site also influenced the proportion of missed feedings. Unidentified and misidentified feeds occurred more often when the breeding site was partially obscured. Unidentified prey also varied with observer (P= 0.0005), although this did not correlate with observer experience. Fish lengths recorded by observers viewing photographs of fish subsequently collected, and therefore of known length, were consistently about one centimeter lower than actual fish length (R²= 0.71). After correcting for missed feeds and misidentification of fish lengths, we obtained robust estimates of chick energy intake rates (accuracy ± 2%). We concluded that light level was the largest source of bias in feeding watches, but that other factors, such as breeding site location, needed to be considered when analyzing data collected from these watches.