2026-10-03 Weekly Good Reads: The Minority Being Stronger May Simply Be Due to a Higher Threshold

Title: 2026-10-03 Weekly Good Reads: The Minority Is Stronger, Maybe Just Because the Bar Is Higher

Body:

Why I picked it

Recent WeChat forwards keep circling back to a few connected questions: What value does management actually create? How do ordinary people without connections fight for opportunities? Why is the same thing worth completely different amounts to different people? And once an opportunity appears, can an individual actually hold onto it? They all assume a premise that's easy to overlook: Are the "capable people" we see really filtered out from the same starting line, through the same threshold?

Nava Ashraf, Oriana Bandiera, Virginia Minni, and Víctor Quintas-Martínez's September 2026 working paper, "Gender Gaps Across the Spectrum of Development: Local Talent and Firm Productivity," puts this question inside a consumer goods multinational spanning 101 countries. The authors observed that in countries and age groups with lower female labor force participation, fewer women enter the company, yet those who do perform better on pay, promotions, and manager ratings. The paper's explanation isn't "women are naturally stronger," nor "diversity automatically boosts productivity." It's selection effect: when a group has to cross a higher threshold to enter the labor market, the people we end up seeing have already been through stricter screening.

I picked it because it adds an organizational-level correction to "value varies by person." A person's résumé, position, and performance don't just reflect individual ability—they also reflect what candidate pool they came out of and what thresholds filtered them along the way. More importantly, this paper doesn't stop at a correct but vague "eliminate bias." It puts what firms can do, what a single firm can't do, and how social thresholds and corporate pay systems constrain each other into the same model. Its conclusions are illuminating and easily abused, which is why it deserves a slow read.

Core arguments of the paper

The paper uses administrative records from a single multinational company. The sample covers 2015 to 2019, 101 countries, 100,819 local full-time white-collar employees, totaling 303,759 employee-year observations. Employees mainly work in sales, engineering, marketing, HR, R&D, and management, typically requiring a university degree. The authors can see job levels, roles, tenure, annual performance ratings from managers, and fixed and variable income from the payroll system.

The authors then group employees by gender, country, and four decade-based age brackets, matching them to the male-female labor force participation rate in their country when they entered the labor market around age twenty. The core metric is the female labor force participation rate relative to men's. The point is to use the same company's global personnel system to hold firm-side differences as constant as possible, then observe how labor market thresholds in different places and generations change the gender composition the company ultimately sees.

The first set of results is descriptive. The lower the local female-to-male labor force participation rate, the lower the female-to-male ratio inside the company; but the women who stay are more likely to appear at the high-pay end, in management, and among faster promoters. The gender pay gap also varies with participation rates: in environments with the lowest female participation, women's average pay may actually exceed that of men with equivalent experience, tenure, and job function; as female participation rises, this reverse gap gradually disappears and approaches the more common industry pattern of women earning about 10% less.

The paper explains this as positive selection. The higher the cost of entering the labor market, the more likely only those with sufficiently high expected returns and strong ability or motivation cross the threshold. The authors use the pay distribution to estimate a latent "productivity" variable. Results show that female employees inside the company average 0.10 standard deviations higher than men; in countries where female labor force participation is below the median, this gap is about 0.18 standard deviations, and in high-participation countries about 0.07. This estimate also predicts manager ratings and first-year pay growth for new hires that weren't directly used in the model, but it remains a latent variable calibrated by the model, not a direct output like production line volume or sales.

The second set of results is counterfactuals from the structural model. The authors ask: If firms know that the female candidate pool has been more strongly screened, can adjusting only the pay structure attract more high-ability women and raise average productivity? Among the 114 country-age group cells that can be solved, the model's "productivity-maximizing" solution would raise the female-to-male employee ratio from 0.71 to 0.92 and the average productivity metric from 0.31 to 0.47 standard deviations, roughly a 50% increase over baseline.

But the cost is equally striking. This solution would lower men's fixed pay, raise women's fixed pay, and make income rise more steeply with latent productivity; the female-minus-male pay gap in the model would go from negative 0.10 log points to positive 0.63, a difference of 73 log points. The authors themselves emphasize that such a gap might be unacceptable both socially and within the organization. It's not a "firms should do this" recommendation, but an extreme optimal solution illustrating: when the external labor market sets different thresholds for two groups, a single firm trying to correct this through wages alone converts external inequality into internal inequality.

The authors then add a constraint: "equal ability uses the same pay rule." In the 88 solvable cells, the female-to-male employee ratio rises from 0.71 to 0.78, and the productivity metric from 0.39 to 0.46, capturing about 60% of the available gain from the unconstrained case. Average pay still wouldn't be fully equal, because the model holds that the two candidate pools have different average latent productivity. In other words, "same rule," "same average outcome," and "eliminating external thresholds" are three different things.

The most critical counterfactual isn't about who the firm pays more. It's assuming men and women face the same external options and entry thresholds. At that point, the firm re-optimizing pay can simultaneously achieve near-equal employee numbers, zero average pay gap, and higher productivity than the status quo. The paper thus draws a clear boundary: firms can improve screening, pay, and promotion, but the household division of labor, social norms, and labor market entry barriers that cause candidate pool differences are not something any single firm can solve alone.

Points worth questioning

First, this is not a randomized experiment. The relationships between female labor force participation, company employee composition, pay, and promotions may also be simultaneously affected by education quality, industry structure, economic cycles, legal enforcement, family policy, and the company's local business footprint. The authors use the same company, different age groups within countries, and various controls to reduce confounding, but they cannot treat "lower labor force participation causes higher individual ability" as a directly proven causal chain.

Second, "101 countries" sounds like broad coverage, but there's only one company. The sample is concentrated in local full-time white-collar workers at a large multinational consumer goods company, and most roles require a university degree. It can't represent small businesses, manufacturing front lines, the public sector, or gig workers, nor can it directly yield hiring or promotion rules for any specific Chinese company. Global coverage increases environmental variation but doesn't automatically solve firm representativeness.

Third, the productivity in the paper is mainly a latent variable calibrated from the pay distribution. Pay records come from the payroll system and are precisely measured; but "wages have no measurement error" doesn't equal "wages perfectly equal productivity." Manager ratings, promotions, and wages can all be affected by organizational politics, position scarcity, negotiation ability, and existing bias. The correlation between the model and additional performance metrics increases credibility but doesn't turn latent ability into a directly observable fact.

Fourth, the headline 50% productivity gain and 73 log point pay change are highly dependent on model assumptions. The counterfactual treats this multinational's pay rules as sufficient to shift the entire labor participation decision, holds employee numbers and total payroll fixed, doesn't let the firm change recruiting investment, training, mentoring, or work flexibility, and doesn't consider attracting already-employed women from other employers. Of the 260 structural sample cells, the full optimal solution can only be solved in 114; adding the same-pay-rule constraint leaves only 88. These are stress tests to help understand mechanisms, not directly replicable budget plans.

Fifth, group average differences cannot become identity scores for individuals. The paper concludes by proposing: when two candidates have identical observable conditions, someone from an underrepresented group may have better unobservable characteristics. This idea can remind managers to pay attention to "the distance traveled," but if gender itself is treated as a proxy for ability, it can also create new discrimination and ignore that specific industries, times, and candidate pools have already changed. The correct use is to examine screening thresholds and fill in individual evidence, not replace judgment about individuals with a group average.

Sixth, the narrowing of the average ability gap among women inside the company as female participation rises doesn't mean social progress brought "declining talent quality." A more accurate explanation is that the candidate pool expanded: previously only the strongest small batch could enter, now more ordinary people can work too. The male candidate pool already includes a full distribution from ordinary to excellent. The goal of equality isn't to keep women forever as an elite sample screened out by high thresholds, but to give them the right to also be ordinary workers.

Takeaways connected to recent interests

This paper adds a counterintuitive angle to "how do people without connections build a future." A high threshold can indeed train those who remain to be more resilient, and can make organizations see higher average performance from survivors; but this isn't a cultivation mechanism worth celebrating. Those blocked at the door leave no data, and the extra costs survivors pay aren't automatically compensated. You can't infer that the threshold itself is reasonable just because a few success stories are strong.

For individuals, "the distance traveled" is real information, but you can't expect managers to automatically understand it. People without connections, crossing industries, or coming from disadvantaged environments need to articulate both hidden thresholds and verifiable results: what was accomplished under what resource constraints, which abilities are reproducible, which were just a one-time survival. Telling only hardship gets dismissed as emotion; telling only results makes others assume the starting line was the same. What's truly useful is connecting constraints, actions, evidence, and outcomes into a checkable chain.

For managers, what's most worth auditing isn't how many women are on the final list, but what happens at each layer of the funnel: who saw the position, who dared to apply, who was blocked by résumé thresholds, who got interviews, who received high-risk assignments, whose achievements were recorded, who exited before promotion. Looking only at those who already entered the organization makes it easy to mistake multi-screened survivors for the entire group, and to mistake candidate pool problems for individual employee problems.

This also responds to "is management building castles on sand." Good management isn't picking the people who look most like strong performers from the existing list, but recognizing how the current list was shaped by institutions, then designing roles, development, and evaluation evidence to be closer to real ability. The paper doesn't prove that any specific management system necessarily works, but it reminds managers: you think you're evaluating talent, but often you're actually evaluating the shadow talent leaves after passing through old thresholds.

For AI tools and personal projects, this selection effect exists too. The success stories we see often come from people willing to invest extra time, handle failure, and document processes; you can't directly extrapolate these survivors' results to all users. When evaluating a tool, ask how the candidate pool formed, where the failures went, and whether average effects are inflated by selection. Change the scenario, and this question isn't fundamentally different from the labor participation thresholds in the paper.

How to read it

Start with the abstract and the first seven pages of the introduction, grasping just three layers: the fact is that women perform higher when relatively scarce in the same company; the explanation is that higher entry thresholds cause positive selection; the policy implication is that internal firm rules and external social thresholds must be separated. Don't immediately simplify the results into "women are better than men."

Step two, look at the data section. Remember 2015 to 2019, 100,819 local full-time white-collar workers, 101 countries, and that the company is just one. Then look at the matching method for female labor force participation: the authors use the country and era when employees entered the labor market around age twenty, not precise records of each person's job search choices that year. This design has explanatory power but also leaves clear proxy variable boundaries.

Step three, go directly to Table 5 and the assumptions before the counterfactuals. Write the 50% productivity gain, 73 log point pay change, and 114 solvable cells together. Any interpretation that cites only the first two numbers without mentioning the sample shrinkage and the strong assumption that wages can shift labor participation is passing off model arithmetic as real-world effects.

Finally, read the conclusion and do a transfer exercise: pick a hiring, promotion, education, or product case you're familiar with, and separately list "who had a chance to enter the candidate pool," "who passed subsequent screening," and "what the final metric actually measured." If these three layers are mixed together, the ability, value, and success rates we see have likely all been reshaped by thresholds.

Original introduction: Becker Friedman Institute: Gender Gaps Across the Spectrum of Development

Paper PDF: BFI Working Paper No. 2026-124