The objective of this study was to identify phenotypic characteristics of dogs predictive of adoption after being received into a shelter. Individual dog records for 2017 were requested from shelters in five states that received municipal funding and utilized electronic record keeping methods. Records from 17 shelters were merged into a dataset of 19,514 potentially adoptable dogs. A simple random sample of 4500 dogs was used for modelling. Variables describing coat length, estimated adult size, and skull type were imputed from breed phenotype. A Cox proportional hazard model with a random effect of shelter was developed for the outcome of adoption using manual forward variable selection. Significance for model inclusion was set at alpha = 0.05. Dogs from shelters in the North were more likely to be adopted than dogs from shelters in the South (hazard ratio (HR) = 3.13, 95% C.I. 1.27–7.67), as were dogs from Western shelters versus those from Southern shelters (HR = 3.81, 95% C.I. 1.43–10.14). The effect of estimated adult size, skull type, and age group on adoption were each modified by time in the shelter (p < 0.001). The results of this study indicate that what dogs look like is predictive of their hazard for adoption from shelters, but the effect of some characteristics on hazard for adoption depend on time in the shelter. Further, this study demonstrates that adopters prefer a certain phenotype of shelter dog including those that are puppies, small sized and not brachycephalic, when accounting for time in the sheltering environment.

The objective of this study was to identify phenotypic characteristics predicting the outcome of euthanasia for dogs entering shelters compared to live release. Individual dog records for 2017 were requested from shelters in five states (Mississippi, Pennsylvania, Michigan, Colorado, and Oklahoma) receiving municipal funding and using electronic records. Duplicate dogs were removed and records from 17 shelters were merged into a dataset of 25,047 unique dogs with variables of breed, gender, coat color, size, age, region, and time in shelter. Only data from dogs with the potential to be adopted (n = 19,514) were analyzed. From these data, a simple random sample of 6200 dogs was used for modelling. Variables describing coat length, estimated adult size, and skull type were imputed from the breed description. A Cox proportional hazard model with a random effect of shelter was developed for the outcome of euthanasia using manual forward variable selection and significance for variable retention at alpha = 0.05. A size by geographic region interaction was associated with the hazard of euthanasia (p = 0.0204). Additionally, age group and skull type were both associated with euthanasia compared to live release (p < 0.0001). The results of this study indicate that phenotypic characteristics of dogs are predictive of their hazard for euthanasia in shelters.