Dear All, A laud to numhess: <hansl> include a_eff.gfn # not uploaded version, since it requires 2017d open transport.gdt logit auto 0 dtime bl = $model exey(&bl) #including "logit" into exey() took 1 minute of coding!!! <hansl> output: logit elasticities on Pr(auto = 1): coefficient std. error z p-value ------------------------------------------------------ dtime -0.0373775 0.0205797 -1.816 0.0693 * <stata> use http://www.principlesofeconometrics.com/poe4/data/stata/transport.dta, clear logit auto dtime margins, eyex(dtime) atmeans ------------------------------------------------------------------------------ | Delta-method | ey/ex Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- dtime | -.0373775 .0205797 -1.82 0.069 -.077713 .002958 ------------------------------------------------------------------------------ The same is with probit. The timing is even better than for mfx() which uses analytical formulas but for b[k]*x + b[k+1]*x^2 Oleh