Postestimation prediction question

[Gautam Sethi]

I'm running the following regression: Y = β0 + β1Χ1 + β2Χ2 + .. +.βnXn + ε.

What's the easiest way to calculate Y_hat = b0 + b1Χ1 + b2Χ2_bar + .. +.bnXn_bar

where the bs are estimates of βs, and Xi_bar is the mean of Xi?

I did it by storing the means of Xis as scalar and going the longhand way but my code looks kludgy. I was wondering if there is a simpler, cleaner way to do it.

[Ken Chui]

That'd be just one number, correct? -margins, atmean- after the regression command may be able to get what you need:

sysuse auto, clear

* Storing the means for checking later:
foreach x in weight mpg headroom length{
    quietly sum `x'
    scalar m`x' = r(mean)
}

* Here is the regression
reg price weight mpg headroom length

* The "long way":
display _b[_cons] + _b[weight]*`=mweight' + _b[mpg]*`=mmpg' + _b[headroom]*`=mheadroom' + _b[length]*`=mlength'

* Shortcut to get the same answer:
margins, atmean

[Gautam Sethi]

Dear Ken:
Thank you for your response. I should have been more explicit. The output is a vector: note the expression "b0 + b1Χ1 + b2Χ2_bar + .. +.bnXn_bar" in which I'm using the observed values of X1 but the means of X2 through Xn.

I'm essentially following the long way you suggested. My code looks like:

display _b[_cons] + _b[weight]*weight + _b[mpg]*`=mmpg' + _b[headroom]*`=mheadroom' + _b[length]*`=mlength'

I was hoping the margins command would have more options to do what I need it to do.

[Ken Chui]

I see. Sorry about the misunderstanding. In that case I think your way may be more direct. Perhaps change that into a -gen- command so that the results will be added to your data set.

[Richard Williams]

The help for margins says

at((means) _all (asobserved) x2) is a convenient way to set all covariates except x2 to the mean.

Is that what you want?

Having said that, in an OLS regression I don't think any of this much matters -- I think you get the mean of Y whatever method you use.