Including independent variables in propensity score matching

Including dependent variables in propensity score matching

I am doing a comparison of economic delevopment in two groups of companies. One group has had a tax audit. I have used a logistic regression to create a propensity score for beeing audited. I have then matched an equal number of non-audited companies with close score matches. The variables included in the logistic regression are related to the probability of beeing audited. But for some dependent variables (wage expences, reported income), there are differences in the two groups even prior to the audit taking place, even after the score matching

My natural instinct would be to use my dependent variables in the logistic regression model, so that the two groups would also be equal prior to beeing audited. That way, if the groups have different levels of income and wage expences after the audit, and all else beeing equal, the difference could be an effect of beeing audited.

But from what I understand, the variables used in making a propensity score should be related to the probability of beeing in the treatment group, and not related to values on the dependent variables. And I am here talking about the values prior to the intervention.

Should I do a new score matching that assures that both groups are equal with respect to the dependent variable prior to intervention? Or is this the wrong approach?
Last edited: