scale of logit model

Hi,

I am trying to simulate some samples for a logit model, and get confused
about the scale of the error term. It is said the variance of the error term
is prescaled to pi^2/6 with gumbel distribution scale parameter equal to
one. So if I have a set of "true" utility parameters B, and corresponding
attributes X, utility function V = BX (U = BX + epsilon), for binary
discrete choice, how do I generate epsilon (location and scale) regarding to
the scale of B and X, or are they really irrelevant?

Kind regards,


.