Average Marginal Effects, margins, dydx(*) command in Stata

[Joseph Malcontento]

Hello! I'm currently doing a probit analysis on marital infidelity and a host of explanatory variables ranging from dummy variable to continuous variables. After running the probit regression I obtained the average marginals effects with the following command: margins, dydx(*), since coefficients from the initial probit regression cannot be interpreted as the effect of my explanatory variable on the probability of Y occuring (yes?).

Given these average marginal effects, how do I interpret these coefficients properly? Take note that the significant explanatory variable are a mix of continuous and dummy variables. A proper phrasing template would be nice.

Furthermore, do these average marginal effects, as seen in their coefficients give me the increase/decrease in the likelihood of Y occurring?

[Clyde Schechter]

As is so often the case, your photo attachment is unreadable (at least on my computer). Such attachments are discouraged. (If you must attach an image, .png files work best.) The best way to post Stata output so that it can be read by others is not with a photo, but by copying from the Results Window or your log file to your clipboard and then pasting into a code block on this forum. See FAQ #12 7th paragraph for instructions on setting up a code block.

You are correct that probit regression coefficients cannot be interpreted as effects of explanatory variables on the probability of the outcome.

In general, the average marginal effects of the type you estimated with your command represent the expected difference in outcome probability associated with a 1-unit increase in the predictor variable, adjusted to the sample distributions of all the variables in your model. If the marginal effect is positive, it signifies that the probability increases, if negative it signifies that the probability decreases, with a 1 unit increase in the predictor.

The interpretations for indicator (dummy) and continuous variables differ slightly. For indicator variables, -margins- actually calculates the predicted values at predictor values that differ by 1 and takes their difference. For continuous variables, -margins- estimates the first partial derivative of the probability with respect to the predictor. This distinction applies only if your original probit regression used factor variable notation to distinguish the discrete variables from the continuous ones. If you did not use factor variable notation, then -margins- has no way to know which type of variable it is dealing with, so it defaults to treating all of them as continuous.

Note that average marginal effects depend strongly on the actual distributions of all the model variables in your particular sample. As such, the findings may not generalize to other situations. Depending on your research goals, that may or may not be a problem for you.

[Richard Williams]

Joseph, at least on my machine, your photo is totally unviewable. Please copy and paste your results using code tags. See pt. 12 of the FAQ for information on how to do this and post questions effectively.

To get you started on interpretation, you may wish to look at

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

[Richard Williams]

Clyde posted while I was writing. I pretty much agree with him. I'll add that I am not very fond of marginal effects for continuous variables, as I find them harder to interpret than is the case with categorical variables. I find graphing helps with continuous variables. Patrick Royston's mcp command is documented in

http://www.stata-journal.com/article...article=gr0056

[sandy emad]

hello, I am also doing a probit regression and have the margins, dydx (*) results for all my covariates as well as the individual margins i calculated using margins, predict (pr) over (variable name).

I understand that in interpreting the individual margins, it's the marginal effects at unique values of my variable used, but for the dydx i am having some trouble understanding what it means.
especially since the coefficients from dydx dont line up with the margins at unique values.
most of my variables are categorical and dummies.

can someone help me understand what the dydx margins coeeficients mean? or at least how they differ from the other margins predict command?

i am very grateful.

[Richard Williams]

Sandy, I would again suggest checking out the sources mentioned above. Or else see

https://www3.nd.edu/~rwilliam/stats3/Margins01.pdf

https://www3.nd.edu/~rwilliam/stats3/Margins03.pdf

If you still have Qs, post some of the output that is confusing you. Use code tags. See pt #12 of the FAQ.

especially since the coefficients from dydx dont line up with the margins at unique values.

It would help see what you mean.