Exporting Average Marginal Effects from Logistic Regression in Stata

Desired Outcome

Obtain the average marginal effects (AME) for each predictor in a logistic regression model and export those estimates to an Excel workbook for further reporting or sharing.

Prerequisites

• Stata version 13 or later (the dydx(*) syntax for margins was introduced in version 13).
• Write permission to the directory where you intend to save the Stata dataset and the Excel file.
• A dataset loaded in memory with a binary dependent variable and at least one independent variable (continuous or categorical).

Procedure

1. Fit the logistic regression model. Use logit (or logistic) with factor notation for any categorical predictors. Example using the built‑in auto dataset:

   sysuse auto, clear
   logit foreign i.foreign mpg weight

   Here foreign is the dependent variable (0 = Domestic, 1 = Foreign) and mpg and weight are continuous predictors.
2. Compute the average marginal effects. Run margins with the dydx(*) option to obtain the AME for every predictor. Store the results in a temporary Stata dataset:

   margins, dydx(*) save(margins_eff)

   After this command, Stata creates a dataset named margins_eff in memory that contains the estimates in r(b) and the variance‑covariance matrix in r(V).
3. Export the results to Excel. Use export excel to write the dataset to a workbook. Specify that the first row should contain variable names:

   export excel using margins_eff.xlsx, first(variable names) replace

   The file margins_eff.xlsx will appear in the current working directory.

Expected Checks

• After margins, inspect the output table to ensure that standard errors are present and not excessively large (which could indicate convergence problems).
• Run return list to verify that the numbers shown in the table match the contents of r(b) and that the squared standard errors match the diagonal of r(V):

  return list

• Optionally, perform a manual check for a continuous predictor (e.g., mpg) by computing the difference in predicted probabilities at x and x+1 using predict and comparing it to the AME reported by margins. The two values should agree within rounding error.
• Assess model adequacy before trusting the margins: estat gof or linktest can be run after the logistic regression to detect major misspecification.

Recovery Options

• If the margins command fails because of missing factor notation consistency, re‑estimate the model using the exact same factor specification (e.g., i.foreign) and then rerun margins.
• If the Excel export fails due to lack of write permission, choose a different directory where you have rights, or run Stata with elevated privileges.
• If the AME estimates appear unstable, repeat the procedure with robust or bootstrap standard errors: logit foreign i.foreign mpg weight, vce(bootstrap) followed by margins, dydx(*).