Dear Statalist,

Currently using the -cmclogit- command. I've been estimating marginal effects (dydx) and semielasticities (dyex) using the respective options with -margins- command. When comparing these values with the ones calculated via contrast in predicted probabilities (of one-unit increment), the values don't match for dyex but match (almost) for marginal effects.

After running cmclogit choice price, basealternative(bn) casevars(fwg33 age), the marginal effects/semielasticities are computed, as well as the contrast between the respective increments, that is, dydx (one-unit increment) and dyex (1% increment).

Results for dydx and one-unit increment change. As shown, they are very similar, almost equal.

Array

However, when comparing for dyex and 1% increment in price (manually), the difference (eg. -0.0706/-0.00070) is that the first is 100 times higher than the latter.

Array

Am I doing something wrong? Doesn't the dyex represent a 1% increase, as in "following a 1% increase in price, the predicted probability of choosing BN decreases by 0.0706"? Maybe the increment is not "0.01*price", as I've computed, but then, what is it?

Probably both are correct, I must be reading them wrong. Does anyone have a clue?

Thanks in advance!