We have observed some unexpected behaviour in RooStats::AsymptoticCalculator (in particular: MakeAsimovData) while performing a statistical combination of two analyses. We found that MakeAsimovData gives different results depending on the order in which the likelihoods are multiplied. It appears to only set the observables in the first likelihood of the product, while ignoring the observables of subsequent likelihoods. This leads to different behaviour, depending on the order in which the product of likelihoods is taken. The correct behaviour is restored when a raw product of the Poisson terms and Gaussian constraints of each individual analysis is taken as the likelihood. In the attached file is a minimal example in PyRoot to illustrate this behaviour.