predictnl and factor variables

[Dario Maimone Ansaldo Patti]

Dear All,

suppose I have the following:

logit Y X1 X2 X3 i.COUNTRY i.YEAR, cluster(S003)

After the estimation I need to calculate a value using the following:

predictnl abc=1/(1+exp(-(_b[X1]*X1+_b[i.COUNTRY]*I.COUNTRY+_b[i.YEAR]*i.YEAR)))

So i need to exclude some of the regressors. I have many countries and many years, so I tried to use factor variables, hoping that predictnl could recognize it. Apparently, this is not the case. So is there any way to use factor variable within predictnl? I need to calculate several of those values above and it is quite annoying to write down all the coefficients/variables associated to the country and year dummies. On the side, using factor notation, I need to check every time which country and year has been dropped to avoid collinearity.

Thanks in advance for the help.

[Andrew Musau]

As long as you know the non-linear transformation required, you do not need to type any long expressions. Contrast this to the predict syntax where you never type in an expression as the linear transformation is unique.

webuse lbw
probit low lwt smoke ptl ht
predictnl phat = normal(_b[_cons] +_b[ht]*ht +_b[ptl]*ptl +_b[smoke]*smoke+ _b[lwt]*lwt), se(phat_se)
predictnl phat2 = normal(xb(low)), se(phat_se2)
assert phat==phat2
assert phat_se==phat_se2

Note that you can difference the sum of the excluded regressors multiplied by their coefficients. Example:

webuse lbw
probit low lwt smoke ptl ht
predictnl phat3 = normal(_b[_cons] +_b[ht]*ht + _b[lwt]*lwt)
predictnl phat4 = normal(xb(low)-_b[ptl]*ptl -_b[smoke]*smoke)
assert phat3==phat4

[Dario Maimone Ansaldo Patti]

Andrew Musau Thanks a lot. That saved me a lot of time.