Models to analyse discrete choice data that account for heterogeneity in error variance (scale) across respondents are increasingly common, e.g. heteroscedastic conditional logit or scale adjusted latent class models. In this paper we do not question the need to allow for scale heterogeneity. Rather, we examine the interpretation of results from these models. We provide five empirical examples using discrete choice experiments, analysed using conditional logit, heteroscedastic conditional logit, or scale adjusted latent class models. We show that analysts may incorrectly conclude that preferences are consistent across respondents even if they are not, or that classes of respondents may have (in)significant preferences for some or all attributes of the experiment, when they do not. We recommend that future studies employing scale heterogeneity models explicitly state scale factors for all samples, choice contexts, and/or latent scale classes, and report rescaled preference parameters for each of these groups.