The Hot Topics thread

Causal Inference Book Club
1

This is a thread to share papers that pop up in the mainstream media. if nothing newsworthy happens, I guess we could also post 90s-mall-goth memes.

Hot-topic-logo-of-DOOM%21
Hot-topic-logo-of-DOOM%21850×350 23.9 KB

“Eclipse of Rent-Sharing: The Effects of Managers’ Business Education on Wages and Labor Share in the US and Denmark”

Pluralistic: 06 Jun 2022 – Pluralistic: Daily links from Cory Doctorow - Cory Doctorow’s take:

The paper makes a somewhat nuanced and technically complex argument, but let me paraphrase its conclusion: Going to business-school makes you the kind of person who cuts wages in bad times and refuses to increase wages in good times. When companies are run by MBAs, their workers’ wages decline.

I used to teach intro to stats for MBA students, so on the one hand, I’d like to believe this isn’t true, but on the other hand, thinking about some of the students, totally plausible.

From the paper, the model is

y_it = γB_it + X'_itβ_t + λ_i + δ_t + ε_it

which they describe as

where Bit is an indicator variable for whether the manager at firm i in year t has a business degree.
In addition, Xit denotes a vector of covariates, λi summarizes the firm fixed effects, δt corresponds to time effects, and εit is an error term. The coefficient of interest is γ, which is the effect of business managers on firm and worker outcomes. In our event studies, we allow the effects to vary by event time, and in some of the specifications we allow these effects to vary by worker skill or wage percentile.
We use a number of different strategies to estimate equation (1). Our first and most central strategy is a series of event studies, focusing on firms that transition from being run by non-business managers to being run by business managers. These event studies enable us to confirm that firms switching to business managers are not on differential trends before the events and provide a transparent way of estimating and displaying our results.

So they have a panel data set, firm by year, and “management has an mba” is the effect of interest.

Throughout, we follow Borusyak et al. (2021) and use an “imputation” estimator to compute the
event-study estimates. This estimator ensures consistency in the presence of two-way fixed effects and avoids issues of spurious identification and negative weights on some observations (de Chaisemartin and D’Haultfœuille, 2020). In practice, this estimator is constructed in three steps. First, the unit fixed effects, time fixed effects, and the coefficients of other control variables are estimated from regressions using untreated observations only.
Second, the treatment effect for each treated observation is computed from the first-step regression as the difference between the actual outcome and the potential untreated outcome. Finally, using the second-step estimated effects, we compute the average treatment effect on the treated.

and

In the US worker-level regressions, because the total number of workers is very large (over 100
million), we adopt a matching procedure for implementing worker-level regressions

Anyways I’m going to spend a little more time with this study. I haven’t found the section that clearly defines the outcome Y yet.

BTW @RavinKumar can you add the MathJax plugin? Discourse Math - plugin - Discourse Meta

1 Like
2

Enabled mathjax. Thanks for creating this topic, also what a conclusion MBA school trains you to cut workers salaries. I wonder what your students think of that

\overbrace{y= ax+b}^{SomeFancy Math}
3

Gender Pay Gap across Cultures

Today’s paper claims that 50% of the gender pay gap in executives is explained by “cultural factors.”

We employ a cross-country sample to examine whether cultural differences help explain gender
compensation variations across corporate executives. The results show that the cultural
differences, which are embedded in societies from long prior to the compensation decisions,
provide significant explanatory power to the observed gender gap in executive compensation.
Using an Oaxaca-Blinder decomposition with variables that have previously been shown to be
significant determinants of executive compensation, we find that adding cultural measures to the
model increases the explanatory power from 44% to 95% of the gender compensation gap.

The decomp is here: Blinder–Oaxaca decomposition - Wikipedia

** Good example of some algebra that exploits the fact that residuals are constrained to zero.

The paper itself stacks a factor analysis into a log-linear model, which seems potentially yikes to me - why do that and not a full SEM?

1 Like
4

They use data from Social Security + Criminal Justice, regression discontinuity design with a break point at age 18 when the benefits are re-evaluated with adult criteria.

5

This is a new study that is trending on twitter. It says “friendship between rich and poor can reduce poverty!”

Complete article here :

Reading the mixtape book, I can reason that the population of poor people who do have a friendship with a rich person is not representative of the total population below poverty line.

Let me know your thoughts.

6

Lol they only studied Facebook friends. Are those even actually real friends anymore?

I know people that are required to “friend” their (higher income) boss, but not a lot of poverty reduction going on there, yknow.

But at least they shared some data - https://www.socialcapital.org/ - idk what I’d do with it though

7

So I did what everyone does and looked up myself, heres a puzzle:

I went to washburn university. 91st percentile for low friending bias based on parent income (+3.7% bias). But switch the income measure to current income (around age 30 IIRC) and it drops to 27th percentile (-2.4% bias). Can we interpret this as the biasing happens after graduating? Or more realistically, maybe there’s an unobserved “unfriending” process where the rich people move away from KS and homophilic ties survive…

Going to ask my old stats prof how they would interpret it…

2 Likes
8

Saw this today, a diff-on-diff on school masking: https://www.medrxiv.org/content/10.1101/2022.08.09.22278385v1

9

https://siepr.stanford.edu/publications/working-paper/do-pre-registration-and-pre-analysis-plans-reduce-p-hacking-and

Do Pre-Registration and Pre-analysis Plans Reduce p-Hacking and Publication Bias?

Randomized controlled trials (RCTs) are increasingly prominent in economics, with pre-registration and pre-analysis plans (PAPs) promoted as important in ensuring the credibility of findings. We investigate whether these tools reduce the extent of p-hacking and publication bias by collecting and studying the universe of test statistics, 15,992 in total, from RCTs published in 15 leading economics journals from 2018 through 2021. In our primary analysis, we find no meaningful difference in the distribution of test statistics from pre-registered studies, compared to their non-pre-registered counterparts. However, pre-registered studies that have a complete PAP are significantly less p-hacked. These results point to the importance of PAPs, rather than pre-registration in itself, in ensuring credibility.

10

This time from /behavioral/ econ. BTW I still hate dynamite plunger plots.

11

I had seen this paper a short time ago. It also deals with the treatment of propensity scores.

12

https://onlinelibrary.wiley.com/doi/abs/10.1111/kykl.12316

We use new individual-level data from MasterChef, a television show in the United States in order to objectively capture situations of fear of failure. We codify situations in which the contestants are on the verge of being eliminated from competition and situations where they explicitly express fear of failing. These new data have the distinct advantage of being purely objective. We cover ten seasons, from 2010 to 2020 and include nearly 200 observations to study the role of fear of failure on performance. Using ordinary least squares, we show that extreme fear of failure is associated with an increase of two to four positions in the final placement of the cooking competition. This positive link between fear of failure and performance tends to contradict the conventional wisdom in both psychology and behavioral economics that such a link tends to be negative. Our findings are robust to broad changes in specification.

13

and zvi on car seats:

1 Like
14

https://rss.onlinelibrary.wiley.com/doi/10.1111/1740-9713.01670

Good, easy article from the ASA student magazine:

Based on our results, one could conclude that dogs are associated with a 14% lower risk of death. While this is not wrong per se, it is an incomplete result, which could lead to unnecessary further investigations. The logical next step would be to focus research on dogs to determine what part of them decreases the risk of death in Covid-19 patients. However, we must first consider if this result was the consequence of confounding or selection bias.

The takeaway is pretty interesting bc it’s a really good example how MNAR can bias your results and how causal thinking can help tease out the issue (even if it doesn’t address the missingness problem itself).

I worry about this issue every time I see a left join.

15

For a psych viewpoint (as opposed to econ).

16

https://osf.io/preprints/socarxiv/9qj4f

Instrumental variable (IV) analysis assumes the instrument only affects the dependent variable via its relationship with the independent variable. Other possible causal routes from the IV to the dependent variable are exclusion-restriction violations and invalidate the instrument. Weather has been widely used as an instrumental variable in social science to predict many different variables. The use of weather to instrument different independent variables represents strong prima facie evidence of exclusion violations for all studies using weather IVs. A review of 288 studies reveals 192 variables previously linked to weather: all representing potential exclusion violations. Using sensitivity analysis, I show that the magnitude of many of these violations is sufficient to overturn numerous existing IV results. I conclude with practical steps to systematically review existing literature to identify possible exclusion violations when using IV designs.

Also don’t trust the weather

2 Likes
17

Saw an example of how not to use instrumental variables on twitter:

2 Likes
18

More COVID RDDs:

1 Like
19

Just saw this. Twitter is such a cesspool. Thought I would chime in with asking what exactly is ‘ancestry-adjusted UV radiation’, other than begging the question? Seems Singapore has high IQ and of course high UV exposure. But, just so happens many Singaporeans are from elsewhere… phew… so their racist preconception survives! What is the model/mechanism for this effect? Does my 3 generations of European history in northern Australia (high sun exposure, lower ozone layer) mean I am a doofus? How many generations does it take? Or is ‘high UV’ just a proxy for something else?
</end of rant>

1 Like
20

The whole thing is just another example of a long line of hereditary based pseudo-science, starting from phrenology to whatever this IQ research claims to be. Still crazy to me that people still try to push this crap in 2022. But yeah, ‘ancestry-adjusted UV radiation’ seems like a barely passible way for them to be racist without actually saying it out loud.

1 Like