I mildly disagree that our opinions should not be averaged. A commentary on Carpentras and Quayle (2023)

The article “The psychometric house-of-mirrors: the effect of measurement distortions on agent-based models’ predictions” claims that in general ordinal scales are biased measures of latent constructs and parametrisation of Agent-Based Models (ABMs) after parametric estimates fit on measurements on ordinal scales do not generally converge into coherent inferences. I argue that the assumptions of their results do not generalise and I use to argument to claim that the parametric fit of ordinal scales can be relaxed under less restrictive methodological assumptions, conversely that the sample of respondents is unbiased and that items are semantically well-designed. My logic is the following: i. differently by the authors, I notice that ordinal scales differ substantially by rankings in their probabilistic structure; ii. it follows that, without assuming strong ontological axioms, Uniformity is not ideal for ordinal scales, hence the simulation of the authors holds only a partial validity; iii. I conclude that well-designed surveys are relatively immune from the kind of distortions modeled by authors because with a good survey design idiosyncratic distortions annihilate into regular noise. In the Conclusions I suggest adopting an approach for the parametrisation of Agent-Based Models based on mixture models.