There are many causes of nest failure in birds. Although partitioning the risk of nest failure among causes has long been of interest to ornithologists, application of formal methods for estimating competing risks has received little attention in the literature. We describe how evidence collected at nests can be formally incorporated into likelihood functions for competing risks to control classification uncertainty. We briefly review estimators for an idealized case in which all fates are known with certainty, and then we introduce new estimators for four cases in which evidence is used. In the evidence-based models, we consider several distinct types of evidence, including videographic evidence, static ecological evidence, and ecological evidence that decays over time. For each of the four cases using evidence, we compare the asymptotic sampling variance to the ideal case in which fates are always known. We also develop closed-form expressions for expected bias when assumptions underlying the estimators are not met. In all cases, use of evidence results in larger sampling variances for the failure probabilities than when fates are known with certainty. Typically, the magnitude of increase in sampling variance depends on the inverse of the evidence probabilities. We also show that disjoint evidence does not reduce sampling variance and should be ignored. Finally, we show that violations of underlying assumptions cause bias, though the bias may be tolerable in some cases.