The short answer: I apply UWLS via {fixest} and then avg_predictions(). Outcomes are CATEs in the form of log risk ratios. Regressors are study variables. Looks something like the code attached. (wlsmra() is a custom function that builds on fixest::feols().)

For those curious, the first applicant, on average, receives 35% (95%CI = [−42%, −26%]) fewer positive responses, while the latter applicant receives 18% (95%CI = [−27%, −7%]) fewer positive responses compared to a plausible control group in that market.

I can virtually "place" profiles in the labour market, such as a Middle Eastern jobseeker looking for a Manager role in Europe over the past 3 years, and compare their hiring penalty with that of a Black applicant seeking a sales role in the US during the same period, all else equal.

Quite funny conclusion given Boren’s “intelligence is what the test tests” (sic) take on intelligence in 1923, implicitly referring to the common underlying g factor. Appears these language model intelligence tests behave similarly.

I expected exactly what you expected. Not sure whether I used this in any of my code before, but seems like normal behaviour to me.

Read the full article here: wol.iza.org/articles/hir.... This is joint work with Stijn Baert and Brecht Neyt.

We also plead that policymaking supports research that
- evaluates under-researched types of discrimination;
- provides broader insights into discrimination mechanisms;
- pilot tests the efficacy of policy interventions.

- ethnic discrimination varies by origin and is highest for MENA candidates;
- the past decades have seen little reduction in hiring discrimination;
- research increasingly focuses on identifying the mechanisms that may explain discrimination, but no conclusive view has emerged.

Other insights are:
- hiring discrimination based on gender appears mainly in occupations dominated by a specific gender;
- age discrimination in hiring is substantial and under-researched in many countries;
- age discrimination is lower in the US than in Europe;

This insight is incompatible with policies focusing exclusively on one or a few vulnerable groups. Broad diversity policies, prioritising targeting those most disadvantaged, are needed.

Our key insight is that many groups face hiring penalties. These penalties are highest for people with a disability, less physically attractive people, older people, and people of a minority race, ethnicity or national origin, respectively.

I've learned not to overinterpret them (by eye-balling). You should be best off using wviechtb.github.io/metafor/refe..., conducting sensitivity analyses with wviechtb.github.io/metafor/refe..., or immediately correcting for various forms of publication bias.

We are also looking for a postdoc (90% research, 10% teaching). Application deadline approaching fast (21 January 2025).

Actually didn't know of this package yet. Just tested out the functions. Seems extremely useful.

I’ve been recently experiencing issues with modelsummary() not displaying tables in the Viewer pane, too. Must be RStudio then.

Thanks, Lukas!

👋🏻