iebaltab: t-test column when using fixed effects, covariates

[de-barros]

Hi ietoolkit team, @kbjarkefur, @luizaandrade

I find the iebaltab t-test column confusing when fixed effects are used. Currently, the mean difference is displayed, and this is what you may expect from the "(1)-(2)" header. However, this is not what you'd expect from the t-test header, and from the table note (on fixed effects). Rather, I'd expect to see the regression coefficient (preferably with an option to show standard errors, too).

For my work, I'm currently only requiring fixed effects; but probably, the same would also make sense for the -covariates()- option.

Would you add this? It would be great to just make two columns optional, when fixed effects are used: One for the regression coefficient and / or one for the mean difference. You may also get rid of the t-test header.

[luizaandrade]

Thank you for your feedback, @de-barros! What is currently being displayed is the difference between the unconditional mean. The t-test used to determine if there is a significant difference between groups, on the other hand, comes from a regression with fixed effects and/or covariates (if specified).

We chose to display the difference exactly because displaying the regression coefficient could create confusion among non-technical audiences when fixed effects and covariates are included. But we could certainly also add an option to display the regression coefficient and standard errors in another column.

Two more related comments:

1. We're also planning to add an option for custom names for the t-test column in LaTeX output -- see #187.
2. Currently, there are a few options to determine what is being displayed in the last column:

• pttest shows the P-value of the t-test instead of the difference between unconditional means
• nottest supresses the t-test column
• normdiff includes an extra column with pairwise normalized differences
• feqtest includes an extra column with an F-test of joint orthogonality across all treatment arms

[luizaandrade]

I'm adding this issue to iebaltab re-write. I think the only decision left to make is whether we should display significance start in both the new column and the t-test column. That would be redundant when they are both used at the same time.

It's easy to remove the stars using starsnoadd, but not to choose in which column they should be displayed.

[aram92]

Hello @luizaandrade,

In the sense of the discussion above, I am wondering if there is a possibility to display regression coefficients and p-values (below in brackets instead of standard errors) in the last column, when using fixed-effects?

Thanks,

Aram