What is the default at() of the margins command?

[Ecce]

If I run something like

logit y x1 x2
margins, dydx(*) atmeans

I can interpret it as the marginal effect evaluated at the the average for all variables.

But what is the default if I don't specify any at() command? I checked the values and they are not the same as atmeans and I couldn't find any mentioning of the default in the help file.

[Marcos Almeida]

I gather the default is the overall margin, as stated in the manual.

[Ecce]

Thank you Marcos, could you please clarify what is meant by overall margins? Does it mean the margin at each value in the dataset which are then averaged out among themselves? Also, could you kindly provide a link to the manual stating this?

[Richard Williams]

If you use -atmeans-, you get MEMs, Marginal Effects at the Means. If you don't use -atmeans- you are basically using -asobserved-, which gives you AMEs, Average Marginal Effects. The differences between the two (for categorical variables, anyway) are described in

http://www3.nd.edu/~rwilliam/xsoc73994/Margins01.pdf

For continuous variables, see

http://www3.nd.edu/~rwilliam/xsoc73994/Margins02.pdf

http://www3.nd.edu/~rwilliam/xsoc73994/Margins03.pdf

[Marcos Almeida]

Hello "Ecce",

The references mentioned by Richard Williams are benchmark presentations of this matter. They are a "must read", so to speak.

That said, just to please you with the link of the manual, here it goes: (http://www.stata.com/manuals14/r.pdf)

Below, an excerpt, specifically related to your query:

Overall predictive margin, the average predicted probability of y = 1, after logit y a##b x1 x2

. margins

To end, I kindly recommend to re-register with full name, as recommended in the FAQ. You may do this just by clicking on the "contact us" button below. Thanks.

[Richard Williams]

Well, I don't know if the handouts are that good, but thanks for the kind words. ;-)

One thing to keep straight, that often confuses people. Margins can estimate both adjusted predictions and marginal effects. Ecce's original Q pertained to marginal effects, whereas Marcos comment (the margins command without dydx) pertained to adjusted predictions.

[Sasha Volodarsky]

Dear Richard and Marcos

Can you please clarify (if you can) what STATA does if you ask for predicted values with and without "atmeans" option. I see in my results that there are small differences.
I thought that "atmeans" is a default but apparently not.