Predicted probabilities in logit regression with control variables set to mean or mode

[Emil Alnor]

I have made a lot of logistic regression with one binary independent variable on a lot of different binary dependent variables. The models includes both categorical and continous control variables. The interpretation of the coefficients are not intuitive, so I want to show the predicted probabilities of the dependented variables when my independent variable is 0 and 1, repsectively, with the control variables set to mode or mean (for categorical and continous control variables, respectively).

In the auto data, this is what I try to do:

sysuse auto, clear
recode price (0/5000 = 1 "Cheap") (5001/999999 = 0 "Expensive"), gen(cheap)
logit foreign i.cheap i.rep length
margins cheap, at(rep=mode length=mean)

Can anyone help with specifying the ", at"-option correctly? What I am trying to do is calcalute the predicted probabilities for 'foreign' for cheap and expensive cars when rep is at its mode and length is at its mean. Since in my real data, I have many models and varibles, manually inserting the values of the modes and means would be very time consuming.

[Felix Bittmann]

You need to compute these statistics of interest beforehand:

sysuse auto, clear
ssc install mmodes, replace
mmodes rep
sum length

recode price (0/5000 = 1 "Cheap") (5001/999999 = 0 "Expensive"), gen(cheap)
logit foreign i.cheap i.rep length
margins cheap, at(rep=3 length=187.93)

[Emil Alnor]

Felix Bittmann thanks for your suggestion, but, as I wrote in my post, since in my real data, I have many models and varibles, manually inserting the values of the modes and means would be very time consuming. Therefore, computing the statistic beforehand is not a viable option.

[Andrew Musau]

There is no option to include modes automatically. For one, modes may not be unique - so you need some criterion to select them. The option -atmeans- will automatically include means, so you can do something like

margins cheap, at(rep=3) atmeans

where 3 is the mode of rep78.

I have many models and varibles, manually inserting the values of the modes and means would be very time consuming.

You can write a program that will compute the modes of a specified list of variables and store them in a local that you feed straight to margins, so that your command becomes

margins cheap, at(`modes') atmeans

I do not have time to volunteer code at the moment.

[Emil Alnor]

There is no option to include modes automatically

Thanks for clarifying that, got to do some programming then.