Accounting for pseudoreplication is not possible when the source of nonindependence is unknown

Repeated observations of the same individuals or other units, which can lead to clustered observations, are common in animal behaviour research, and mixed models are commonly employed to model and account for such clustering in the data and avoid pseudoreplication. However, in some cases, while the data might comprise repeated samples from the same individuals, the precise identity of the individuals from which samples originated is unknown. In a recent paper Garamszegi (2016, Animal Behaviour, 120, 223–234) suggested an approach to account for pseudoreplication which is based on repeatedly assigning random subject identities to the samples and then analysing the data using a mixed model or averaged values for each randomly assigned identity. Here we tested this approach using a simulation study. We found that the approach suggested by Garamszegi leads to clearly inflated type I error rates that were essentially the same as those obtained from a naïve linear model simply ignoring individual identity and that only a model based on the correct subject identities roughly produced the nominal type I error rate. We conclude that, currently, there is no method available that allows pseudoreplication to be controlled when subject identities are unknown.