t-test VS propensity score matching

[Angeliki Skoura]

Dear all,

I would like to ask a question about the statistical significance of the values of the two methods (T-test & propensity score matching).
May the value of difference of means resulting from the t test not be statistically significant but the propensity score matching value be statistically significant using the same variables or the two methods are not related to each other?

I am thankful for any response.

Best,
Angeliki

[Felix Bittmann]

A t-test tells you whether the mean of two groups is different. Here, no adjustment happens. That is, if there are confounders present (which is usually the case if no experiment was conducted), you cannot interpret the result as causal. For example, if you compare males and females, you cannot conclude that the difference is due to gender. PSM, just as related methods like OLS regressions, attempt to estimate a causal effect. If you are able to account for all confounders by selecting controls, you might be able to state that the group is the cause of the difference. In summary, a t-test does not use control variables but PSM does.

[Angeliki Skoura]

Thank you @Felix Bittmann!

[Jared Greathouse]

To be a little more concrete, PSM is only valid when the unconfoundedness assumption holds. That is, do you have enough covariates such that there's no major imbalance between the average outcomes treated units and untreated ones in the pre-intervention period?