For a binary explanatory variable, explain why the cumulative logit model with proportional odds structure is unlikely to fit well if, for an underlying latent response, the two groups have similar location but very different dispersion.