Assessment of degeree of OVB with psacalc command - Negative value for delta (relative degree of selection)?

[Abby Connor]

Hi everyone,

I am using the psacalc command and the strategy proposed by Emily Oster (2016) on unobservable selection and coefficient stability in order to evaluate the degree of omitted variable bias in my OLS regressions.

I use the command

psacalc delta variable_of_interest

to get a value for delta. If I understand correctly, a value of delta=3 implies that the unobservables would have to be three times as important as the observables (i.e. included controls) to render the effect of the variable of interest 0, i.e. to get a zero treatment effect.

In some (rare) cases, I get a negative value of delta and I have problems interpreting this negative result.

Can anyone help me with the interpretation? Or is this result because I am implementing the command wrong in any way?

Thanks and kind regards,
Abby

[Arthur Morris]

One of the package's authors has answered this question here: Unobservable Selection(Oster 2016) Using psacalc