Causality method for logit model

[Adeola Kolade]

Hello all,

I will appreciate if someone can advise on method to use for causality when my outcome and independent variables are binary and I don't have an instrument which means using instrumental variable is out of the options. Thanks in advance.

[Richard Williams]

I would use logistic regression. If you think there are some variables you should have but don't then you can offer caveats about model misspecification.

[Adeola Kolade]

Many thanks.

[Jeff Wooldridge]

Tell us more about your control variables. Also, you have a binary treatment? The doubly robust estimator available in teffects can be very useful. Specify a logit response model and logit treatment model in ipwra.

[Adeola Kolade]

Thanks for your reply. This is the logit model for the relationship between childhood obesity (binary treatment) and educational attainment in adulthood (binary outcome)

educational attainment = β₀ + β₁child obesity + β₂sex + β₃

childhood

social class + β₄childhood cognitive skills+ β₅

childhood

school absence + β₆fathers education+ β₇smoking + e_i

I will appreciate your reply sir.

[Jeff Wooldridge]

The following command estimates a logit model for child obesity using the controls. Then it estimates separate logits for educational attainment for the two values of child obesity, using weights from the first estimation. The estimator has a "double robustness" property: only one the outcome models or treatment model need to be correctly specified for consistent estimation of the ATE. This is in my 2007 Journal of Econometrics paper, "Inverse Probability Weighted Estimation for General Missing Data Problems."

teffects ipwra (education_attainment x1 x2 ... xk, logit) (child_obesity x1 x2 ... xk, logit)

[Adeola Kolade]

Thanks for your reply. I tried running it but it gives me the note shown below (please how do I go about this?)

treatment 1 has 17 propensity scores less than 1.00e-07
treatment overlap assumption has been violated; use option osample() to identify the overlap violators.