Jean Joohyun Oh

Weekly Research Digest

Weekly collation of research in entrepreneurship, genAI & work, and orgs & inequality. Modeled after Kevin Bryan's Useful Knowledge.

Issue No. 004 · Nov 4, 2026 · 33 papers · 12 highlights ← No. 003 No. 005 →

Highlights··

Management journals 5

001

Looking the Part? How Candidates’ Race and Attire Shape Employers’ Hiring Decisions in a Low-Wage Labor Market

Gig-staffing platform data and preregistered experiment; professional attire yields smaller hiring returns for Black candidates, mediated by perceived professionalism.

AbstractRacial disparities in employment persist in the United States, particularly in low-wage labor markets in which employers frequently cite self-presentation, especially attire, as a basis for hiring decisions. Yet, whether improving professional attire yields equal returns across racial groups remains largely unexamined. Drawing on status characteristics theory, we argue that professional attire functions as an ambiguous proxy for unobservable traits such as motivation and reliability, making it susceptible to racialized interpretations. We predict that professional attire will yield weaker hiring returns for Black candidates than for their White counterparts and that perceived professionalism mediates this gap. We test these arguments using data from a gig-staffing platform comprising over 1 million job applications with profile photos, supplemented by a preregistered experiment that manipulates attire within candidates, using two photos of each candidate: one in professional attire, one in casual. The results converge across field and experimental data. Professional attire improves hiring prospects overall but yields substantially smaller returns for Black candidates. Moderated mediation tests confirm that perceived professionalism drives this disparity, such that race discounts how attire translates into perceived professionalism and how perceived professionalism translates into hiring. The racial gap attenuates in customer-facing roles in which appearance is task relevant. These findings challenge supply-side accounts of hiring inequality and reveal how a demand-side interpretation of equivalent signals perpetuates racial disadvantages.
Racial disparities in employment persist in the United States, particularly in low-wage labor markets in which employers frequently cite self-presentation, especially attire, as a basis for hiring decisions. Yet, whether improving professional attire yields equal returns across racial groups remains largely unexamined. Drawing on status characteristics theory, we argue that professional attire functions as an ambiguous proxy for unobservable traits such as motivation and reliability, making it susceptible to racialized interpretations. We predict that professional attire will yield weaker hiring returns for Black candidates than for their White counterparts and that perceived professionalism mediates this gap. We test these arguments using data from a gig-staffing platform comprising over 1 million job applications with profile photos, supplemented by a preregistered experiment that manipulates attire within candidates, using two photos of each candidate: one in professional attire, one in casual. The results converge across field and experimental data. Professional attire improves hiring prospects overall but yields substantially smaller returns for Black candidates. Moderated mediation tests confirm that perceived professionalism drives this disparity, such that race discounts how attire translates into perceived professionalism and how perceived professionalism translates into hiring. The racial gap attenuates in customer-facing roles in which appearance is task relevant. These findings challenge supply-side accounts of hiring inequality and reveal how a demand-side interpretation of equivalent signals perpetuates racial disadvantages.

Abstract via crossref

002

Crossing the Design–Use Divide: How Organizations Select Innovations to Adopt

Ethnography of entrepreneurs piloting AI-based innovations with adopting organizations; committee selection processes shape innovation design and implementation.

AbstractPrior scholarship on technology and organizations has focused either on how innovations are designed or how they are used, often assuming that successful selection stems from successful design. But this assumption has limited our understanding of patterns of resistance and advocacy that can influence how and whether innovations reach organizations in the first place. To explore these patterns, this study asks, how does selection unfold within organizations, and what are the consequences of this process for the design and use of innovations? Drawing on a two-year ethnographic study, I examine how entrepreneurs piloted their AI-based innovations with potential adopting organizations. Rather than testing for learning or generalizability, organizations set up committees that engaged in process manipulation by narrowing the scope of pilots, controlling conversations, assigning pilot responsibility, and using goal claiming to align pilot outcomes with committees’ majority preferences. Committees with more adversarial members designed pilots that tested advanced aspects of entrepreneurs’ innovations, hoping for failure. Committees with a majority that championed the innovations tested basic aspects, ensuring success. While pilots run by champions mostly succeeded, these innovations faced more resistance later, during implementation, than did innovations piloted successfully by adversaries. This study shows that innovation is endogenous to selection, and reveals that champions can temper innovation, while adversaries can enable it.
Prior scholarship on technology and organizations has focused either on how innovations are designed or how they are used, often assuming that successful selection stems from successful design. But this assumption has limited our understanding of patterns of resistance and advocacy that can influence how and whether innovations reach organizations in the first place. To explore these patterns, this study asks, how does selection unfold within organizations, and what are the consequences of this process for the design and use of innovations? Drawing on a two-year ethnographic study, I examine how entrepreneurs piloted their AI-based innovations with potential adopting organizations. Rather than testing for learning or generalizability, organizations set up committees that engaged in process manipulation by narrowing the scope of pilots, controlling conversations, assigning pilot responsibility, and using goal claiming to align pilot outcomes with committees’ majority preferences. Committees with more adversarial members designed pilots that tested advanced aspects of entrepreneurs’ innovations, hoping for failure. Committees with a majority that championed the innovations tested basic aspects, ensuring success. While pilots run by champions mostly succeeded, these innovations faced more resistance later, during implementation, than did innovations piloted successfully by adversaries. This study shows that innovation is endogenous to selection, and reveals that champions can temper innovation, while adversaries can enable it.

Abstract via crossref

003

Navigating the Family Support Dilemma in Female Entrepreneurship: An Embeddedness Perspective

Interview study of female entrepreneurs in Ethiopia; gendered family support constrains venture growth.

AbstractDrawing on family embeddedness and gender performativity, we examine how family support is enacted and governed through gendered norms, and how female entrepreneurs navigate these dynamics. Based on indepth interviews with female entrepreneurs and family members in Ethiopia, we introduce the family support dilemma: a process in which support enabling female entrepreneurship simultaneously generates explicit or internalized obligations that constrain sustained engagement and growth over time. Our findings show that family support is embedded in gendered social structures, functioning as a site of norm enforcement where support is granted or withdrawn based on women’s enactment of gender role expectations.
Drawing on family embeddedness and gender performativity, we examine how family support is enacted and governed through gendered norms, and how female entrepreneurs navigate these dynamics. Based on indepth interviews with female entrepreneurs and family members in Ethiopia, we introduce the family support dilemma: a process in which support enabling female entrepreneurship simultaneously generates explicit or internalized obligations that constrain sustained engagement and growth over time. Our findings show that family support is embedded in gendered social structures, functioning as a site of norm enforcement where support is granted or withdrawn based on women’s enactment of gender role expectations.

Abstract via crossref

004

The intelligence loop and value creation in the age of AI agents

Conceptual theory of AI agents' intelligence loop, value creation, and venture implications including one-person companies; speaks to AI and work.

AbstractResearch Summary Artificial intelligence (AI) is moving from generating outputs to executing work. An agent that executes work autonomously produces something generative AI does not: operational experience that can raise its future intelligence. Prior research on AI emphasizes value creation through demand‐side feedback, notably data network effects, in which user data improve prediction and reinforce adoption. We develop the supply‐side counterpart, the intelligence loop: delegated usage generates operational experience, agentic capability converts it into future intelligence, and the resulting value maintains or expands usage. The loop cannot operate until autonomous execution is viable, a threshold we call minimum viable intelligence (MVI). Crossing MVI does not guarantee rising intelligence. Comparable deployments of AI agents can diverge as their loops decay, sustain, or accelerate. Three implications follow. Competitive advantage may depend more on the loop than on installed base. An above‐MVI agent supplies scarce expert skill as a non‐rival, replicable resource, so productive capacity can grow faster than expert headcount, with the one‐person company as a limiting case. And a venture with adequate demand, capital, and a capable team can still fail because its loop never functions or later deteriorates. Managerial Summary Artificial intelligence (AI) agents can execute work, and the experience generated through execution can raise their future intelligence, a feedback process we call the intelligence loop. An agent crosses a workflow's minimum viable intelligence (MVI) when it executes reliably without step‐by‐step direction, autonomous execution creates at least as much net value as the best alternative after oversight and correction costs, and governance arrangements keep residual risk acceptable and accountability clear. Investment belongs where one of these conditions keeps a valuable workflow below MVI. Deployment can begin with narrow, feedback‐rich workflows near MVI that ideally lead to harder workflows, since base model upgrades can match gains on simple work. Even above MVI, the loop is not automatic: it stalls when feedback is not timely and diagnostic, experience yields little intelligence gain, gains add little user value, or value does not maintain or expand usage. Human involvement shifts from executing work to specifying goals, evaluating outputs, governing execution, and supplying tacit knowledge. When the loop sustains or accelerates, productive capacity can grow faster than expert headcount. With capable base models widely available, advantage depends less on access to AI than on building and governing the loop, controlling proprietary operational experience, and converting it into intelligence more effectively than rivals. A venture with adequate demand, capital, and a capable team can fail because its loop never functions or later deteriorates. Cutting evaluation or oversight to scale faster may increase output in the short run but weaken the loop even while revenue remains healthy.
Research Summary Artificial intelligence (AI) is moving from generating outputs to executing work. An agent that executes work autonomously produces something generative AI does not: operational experience that can raise its future intelligence. Prior research on AI emphasizes value creation through demand‐side feedback, notably data network effects, in which user data improve prediction and reinforce adoption. We develop the supply‐side counterpart, the intelligence loop: delegated usage generates operational experience, agentic capability converts it into future intelligence, and the resulting value maintains or expands usage. The loop cannot operate until autonomous execution is viable, a threshold we call minimum viable intelligence (MVI). Crossing MVI does not guarantee rising intelligence. Comparable deployments of AI agents can diverge as their loops decay, sustain, or accelerate. Three implications follow. Competitive advantage may depend more on the loop than on installed base. An above‐MVI agent supplies scarce expert skill as a non‐rival, replicable resource, so productive capacity can grow faster than expert headcount, with the one‐person company as a limiting case. And a venture with adequate demand, capital, and a capable team can still fail because its loop never functions or later deteriorates. Managerial Summary Artificial intelligence (AI) agents can execute work, and the experience generated through execution can raise their future intelligence, a feedback process we call the intelligence loop. An agent crosses a workflow's minimum viable intelligence (MVI) when it executes reliably without step‐by‐step direction, autonomous execution creates at least as much net value as the best alternative after oversight and correction costs, and governance arrangements keep residual risk acceptable and accountability clear. Investment belongs where one of these conditions keeps a valuable workflow below MVI. Deployment can begin with narrow, feedback‐rich workflows near MVI that ideally lead to harder workflows, since base model upgrades can match gains on simple work. Even above MVI, the loop is not automatic: it stalls when feedback is not timely and diagnostic, experience yields little intelligence gain, gains add little user value, or value does not maintain or expand usage. Human involvement shifts from executing work to specifying goals, evaluating outputs, governing execution, and supplying tacit knowledge. When the loop sustains or accelerates, productive capacity can grow faster than expert headcount. With capable base models widely available, advantage depends less on access to AI than on building and governing the loop, controlling proprietary operational experience, and converting it into intelligence more effectively than rivals. A venture with adequate demand, capital, and a capable team can fail because its loop never functions or later deteriorates. Cutting evaluation or oversight to scale faster may increase output in the short run but weaken the loop even while revenue remains healthy.

Abstract via crossref

005

Should I stay or should I go: intentions to freelance across 16 countries

Cross-country survey on intentions to leave salaried jobs for freelance self-employment; vulnerable workers show higher intentions.

AbstractA substantial literature explores workers’ transitions into entrepreneurial business ownership, while freelance, on-demand self-employment, a distinct and increasingly prevalent mode of labour market participation, remains comparatively understudied. Using an original large survey conducted in 16 countries, we investigate the determinants of workers’ intentions to leave their main salaried job to engage in freelance, on-demand self-employment. We find that socio-demographic characteristics, job conditions, and behavioural traits are systematically associated with these intentions, though several patterns diverge from those documented in the entrepreneurship literature. A detailed heterogeneity analysis substantiates variation across occupations, task types, and individual characteristics. Workers in more vulnerable positions exhibit higher intentions to transition into freelance, on-demand self-employment, a form of work often associated with lower protection and higher income risk.
A substantial literature explores workers’ transitions into entrepreneurial business ownership, while freelance, on-demand self-employment, a distinct and increasingly prevalent mode of labour market participation, remains comparatively understudied. Using an original large survey conducted in 16 countries, we investigate the determinants of workers’ intentions to leave their main salaried job to engage in freelance, on-demand self-employment. We find that socio-demographic characteristics, job conditions, and behavioural traits are systematically associated with these intentions, though several patterns diverge from those documented in the entrepreneurship literature. A detailed heterogeneity analysis substantiates variation across occupations, task types, and individual characteristics. Workers in more vulnerable positions exhibit higher intentions to transition into freelance, on-demand self-employment, a form of work often associated with lower protection and higher income risk.

Abstract via crossref

Sociology journals 2

006

Who Benefits from Citizenship? Varying Effects of Naturalization Decisions in the Early-Twentieth-Century United States

Linked historical census data on naturalization returns for immigrants' occupational status; larger gains for disadvantaged immigrants.

AbstractCitizenship acquisition is widely viewed as a mechanism for improving economic outcomes, yet little is known about whether its benefits are distributed evenly. This article examines the varied returns to naturalization in the early-twentieth-century United States, assessing whether naturalization uniformly benefits all immigrants or is especially impactful for those already positioned to profit. Using naturalization records from New York City’s Southern District in the early 1900s, we construct a novel historical dataset for 1,947 immigrants who declared their intent to naturalize. By linking these records to U.S. census data from 1920, 1930, and 1940, we are able to observe immigrants’ characteristics at the point when they make naturalization decisions and to track their outcomes over time. Our findings reveal that immigrants who completed the naturalization process attained significantly higher occupational status than those who initiated but did not complete naturalization. In addition, citizenship is most transformative for those facing the greatest disadvantages. Additional analyses corroborate these findings in linked full-count census data for the entire United States. Citizenship confers occupational advantages not only through formal rights but as a resource that enables claims-making in the labor market, with greater returns to less advantaged immigrants.
Citizenship acquisition is widely viewed as a mechanism for improving economic outcomes, yet little is known about whether its benefits are distributed evenly. This article examines the varied returns to naturalization in the early-twentieth-century United States, assessing whether naturalization uniformly benefits all immigrants or is especially impactful for those already positioned to profit. Using naturalization records from New York City’s Southern District in the early 1900s, we construct a novel historical dataset for 1,947 immigrants who declared their intent to naturalize. By linking these records to U.S. census data from 1920, 1930, and 1940, we are able to observe immigrants’ characteristics at the point when they make naturalization decisions and to track their outcomes over time. Our findings reveal that immigrants who completed the naturalization process attained significantly higher occupational status than those who initiated but did not complete naturalization. In addition, citizenship is most transformative for those facing the greatest disadvantages. Additional analyses corroborate these findings in linked full-count census data for the entire United States. Citizenship confers occupational advantages not only through formal rights but as a resource that enables claims-making in the labor market, with greater returns to less advantaged immigrants.

Abstract via crossref

007

Workplace authority in China: gender, parenthood, and work sectors

Panel fixed-effects study of parenthood and gender gaps in workplace authority across Chinese state and private sectors.

AbstractWorkplace authority is a key individual and organizational resource to which women typically have more limited access than men. Yet, less is known about how parenthood shapes this gender disparity and how the relationship varies across employment regimes. Drawing on nationally representative, longitudinal data from the China Family Panel Studies (2010–2022), I examine within-person changes in women’s and men’s access to supervisory authority following parenthood, with particular attention to variation across the state and private sectors. Fixed-effects results show that, overall, women’s likelihood of holding authority declines after becoming mothers, especially after having two or more children, whereas men experience little change. These dynamics differ by work sector. A considerable motherhood penalty is concentrated in the private sector, whereas a fatherhood premium emerges only in the state sector. Parenthood therefore widens the gender gap in authority in both sectors through distinct patterns. These findings suggest that employment regimes condition the gendered meanings of parenthood and how they translate into access to authority. In China, the state sector’s formal protections may buffer motherhood-related disadvantages, whereas its male-dominated hierarchy and pronatalist agendas may sustain fatherhood-related rewards.
Workplace authority is a key individual and organizational resource to which women typically have more limited access than men. Yet, less is known about how parenthood shapes this gender disparity and how the relationship varies across employment regimes. Drawing on nationally representative, longitudinal data from the China Family Panel Studies (2010–2022), I examine within-person changes in women’s and men’s access to supervisory authority following parenthood, with particular attention to variation across the state and private sectors. Fixed-effects results show that, overall, women’s likelihood of holding authority declines after becoming mothers, especially after having two or more children, whereas men experience little change. These dynamics differ by work sector. A considerable motherhood penalty is concentrated in the private sector, whereas a fatherhood premium emerges only in the state sector. Parenthood therefore widens the gender gap in authority in both sectors through distinct patterns. These findings suggest that employment regimes condition the gendered meanings of parenthood and how they translate into access to authority. In China, the state sector’s formal protections may buffer motherhood-related disadvantages, whereas its male-dominated hierarchy and pronatalist agendas may sustain fatherhood-related rewards.

Abstract via crossref

General science journals 2

008

Decredentialed jobs and upward mobility among non-college workers in the USA

Links job postings to administrative earnings; employers dropping degree requirements and upward mobility of non-college workers, with race and gender composition.

AbstractProminent employers are increasingly dropping college degree requirements when hiring. Does this provide upward mobility for non-college workers? Matching job postings to 446,500 job starts in US administrative data from 2011 to 2021, we show that employers reduced the share of postings requiring a degree by more than 10%. Non-college workers who move to decredentialed jobs earn 12 log points more than matched workers moving to never-credentialed jobs (β = 0.124; 95% confidence interval, (0.042, 0.206)), around US$4,600 per year—a premium that endures for at least five years. These hires are more likely to be Black, Hispanic and female than the workers they replace. Employers also benefit, as removing degree requirements reduces labour costs. Yet most employers that drop requirements do not hire non-college workers (non-college hiring rises 1.2 percentage points; 95% confidence interval, (0.5, 2.0)). Decredentialing is a promising tool for social mobility, but only when employers follow through. Linking job postings to US Census earnings records from 2011 to 2021, Lyttelton et al. find that non-college workers hired into jobs that dropped degree requirements earned around US$4,600 more per year, yet employers rarely hired non-college workers after decredentialing.
Prominent employers are increasingly dropping college degree requirements when hiring. Does this provide upward mobility for non-college workers? Matching job postings to 446,500 job starts in US administrative data from 2011 to 2021, we show that employers reduced the share of postings requiring a degree by more than 10%. Non-college workers who move to decredentialed jobs earn 12 log points more than matched workers moving to never-credentialed jobs (β = 0.124; 95% confidence interval, (0.042, 0.206)), around US$4,600 per year—a premium that endures for at least five years. These hires are more likely to be Black, Hispanic and female than the workers they replace. Employers also benefit, as removing degree requirements reduces labour costs. Yet most employers that drop requirements do not hire non-college workers (non-college hiring rises 1.2 percentage points; 95% confidence interval, (0.5, 2.0)). Decredentialing is a promising tool for social mobility, but only when employers follow through. Linking job postings to US Census earnings records from 2011 to 2021, Lyttelton et al. find that non-college workers hired into jobs that dropped degree requirements earned around US$4,600 more per year, yet employers rarely hired non-college workers after decredentialing.

Abstract via openalex

009

Unpacking transatlantic income inequality

Harmonized data on disposable and market income inequality and redistribution across 17 countries, 1995–2021.

AbstractThis paper uses newly harmonized survey and register data on incomes and labor market outcomes as the basis for consistent comparisons of inequalities across 17 North American and European countries over the period 1995–2021. We examine trends in disposable income inequality in the working-age population, defined as persons aged 25 to 60, as well as its component parts of market income inequality and redistribution through government tax and transfer systems. We adjust incomes to allow for changes in education, population aging, and household structure in each country. We test for convergence across countries in demographically-adjusted disposable income inequality, along with market income inequality and redistribution. There is strong evidence of convergence over time in disposable income inequality across the 17 countries, despite a divergence in the dispersion of market income inequality driven by very heterogeneous trends for hours of work and the number of workers in households across countries. Consistent with this we find a strong convergence across countries in the dispersion of our measure of redistribution. There are reductions over time in inequality-reducing redistribution in the Nordic countries and increases in some other European and North American countries, with the net effect being more similarity across countries, a lower average level of redistribution, and a higher average level of disposable income inequality. These trends are being driven by the bottom 95% of the income distribution. Our results are robust to a variety of alternative specifications including the measure of inequality, trimming of data outliers, and sample period.
This paper uses newly harmonized survey and register data on incomes and labor market outcomes as the basis for consistent comparisons of inequalities across 17 North American and European countries over the period 1995–2021. We examine trends in disposable income inequality in the working-age population, defined as persons aged 25 to 60, as well as its component parts of market income inequality and redistribution through government tax and transfer systems. We adjust incomes to allow for changes in education, population aging, and household structure in each country. We test for convergence across countries in demographically-adjusted disposable income inequality, along with market income inequality and redistribution. There is strong evidence of convergence over time in disposable income inequality across the 17 countries, despite a divergence in the dispersion of market income inequality driven by very heterogeneous trends for hours of work and the number of workers in households across countries. Consistent with this we find a strong convergence across countries in the dispersion of our measure of redistribution. There are reductions over time in inequality-reducing redistribution in the Nordic countries and increases in some other European and North American countries, with the net effect being more similarity across countries, a lower average level of redistribution, and a higher average level of disposable income inequality. These trends are being driven by the bottom 95% of the income distribution. Our results are robust to a variety of alternative specifications including the measure of inequality, trimming of data outliers, and sample period.

Abstract via crossref

Working papers 6

010

Breaking into Supply Chains : Experimental Evidence on Procurement Training for Women-Owned SMEs

Randomized trial of procurement training for women-owned SMEs in Colombia; effects on bidding and contract wins.

AbstractWomen-owned-or-led small and medium-sized enterprises (WSMEs) are underrepresented in public and corporate supply chains worldwide. There is growing interest in promoting SMEs’ and WSMEs’ integration into value chains, but limited causal evidence whether interventions to support their participation in procurement are effective. This paper conducts a phased randomized controlled trial to estimate the impacts of corporate procurement training for WSMEs in Colombia. The program increases procurement knowledge and readiness, resulting in a ten percentage point increase in registrations to corporate portals, and an eight percentage point increase in the likelihood of submitting a bid. Impacts on winning bids are concentrated in WSMEs with prior procurement knowledge or experience, whose probability of winning a contract increases by eight to nine percentage points. These impacts are approximately twenty percent of the control means. More broadly, the paper illustrates how complier-weighted treatment effects can be used to evaluate programs delivered through a funnel-shaped participation process, common in private sector development interventions.
Women-owned-or-led small and medium-sized enterprises (WSMEs) are underrepresented in public and corporate supply chains worldwide. There is growing interest in promoting SMEs’ and WSMEs’ integration into value chains, but limited causal evidence whether interventions to support their participation in procurement are effective. This paper conducts a phased randomized controlled trial to estimate the impacts of corporate procurement training for WSMEs in Colombia. The program increases procurement knowledge and readiness, resulting in a ten percentage point increase in registrations to corporate portals, and an eight percentage point increase in the likelihood of submitting a bid. Impacts on winning bids are concentrated in WSMEs with prior procurement knowledge or experience, whose probability of winning a contract increases by eight to nine percentage points. These impacts are approximately twenty percent of the control means. More broadly, the paper illustrates how complier-weighted treatment effects can be used to evaluate programs delivered through a funnel-shaped participation process, common in private sector development interventions.

Abstract via repec

011

Start-up Financing, Entry and Innovation

Equilibrium model of venture capital market estimating start-up shutdown distortions and US-UK differences in funding and innovation.

AbstractI develop and estimate an equilibrium model of the VC market to quantify distortions and explain cross-country differences in VC activity. The model maps directly to observable start-up funding histories and outcomes, facilitating estimation. For US start-ups first funded in 2005–2015, my estimates suggest that 40% shut down despite having positive continuation value; with continued funding, half would achieve a successful exit. In the UK, the model attributes the gap in activity to the US to financing conditions and acquisition opportunities, and counterfactuals suggest there may be gains to retargeting existing government support towards late-stage start-ups.
I develop and estimate an equilibrium model of the VC market to quantify distortions and explain cross-country differences in VC activity. The model maps directly to observable start-up funding histories and outcomes, facilitating estimation. For US start-ups first funded in 2005–2015, my estimates suggest that 40% shut down despite having positive continuation value; with continued funding, half would achieve a successful exit. In the UK, the model attributes the gap in activity to the US to financing conditions and acquisition opportunities, and counterfactuals suggest there may be gains to retargeting existing government support towards late-stage start-ups.

Abstract via repec

012

Supporting Small-Scale Micro-Entrepreneurship in South Africa’s Townships

Qualitative study of township micro-entrepreneurs in South Africa: necessity entry, networks, registration, government support, success factors.

AbstractDrawing on qualitative data from 11 key informants and 75 small-scale firm owners in South Africa’s townships, this paper identifies five factors shaping micro-entrepreneurial success. First, micro-entrepreneurship often begins as employment of last resort but evolves into a deliberate choice as owners value its autonomy and independence. Second, social and communal networks expand access to shared resources, information, and markets. Third, complex processes and limited benefits discourage firm registration, even though entrepreneurs feel that it boosts customer trust and patronage. Fourth, access to government support is low, hindered by perceived program irrelevance and stringent eligibility criteria. Fifth, success depends less on finance than on resilience, resourcefulness, and mindset. These findings underscore the need for holistic programs offering business training, socio-emotional skills, mentorship, and collective business strategies tailored to the township economy.
Drawing on qualitative data from 11 key informants and 75 small-scale firm owners in South Africa’s townships, this paper identifies five factors shaping micro-entrepreneurial success. First, micro-entrepreneurship often begins as employment of last resort but evolves into a deliberate choice as owners value its autonomy and independence. Second, social and communal networks expand access to shared resources, information, and markets. Third, complex processes and limited benefits discourage firm registration, even though entrepreneurs feel that it boosts customer trust and patronage. Fourth, access to government support is low, hindered by perceived program irrelevance and stringent eligibility criteria. Fifth, success depends less on finance than on resilience, resourcefulness, and mindset. These findings underscore the need for holistic programs offering business training, socio-emotional skills, mentorship, and collective business strategies tailored to the township economy.

Abstract via repec

013

Affirmative Action, Faculty Productivity and Caste Interactions: Evidence from Engineering Colleges in India

Compares productivity of caste-reservation and general faculty in Indian engineering colleges; affirmative action in hiring.

AbstractAffirmative action programs are often criticized because of concerns that they result in lower worker productivity and efficiency losses.We study the relative productivity of workers benefiting from an aggressive affirmative action policy in a setting where hiring constraints are mandated upon employers and are typically likely to bind.In India, colleges are required to reserve approximately 50 percent of faculty hires for individuals from disadvantaged caste and social class groups.We collect and analyze data from a nationally representative sample of 50 engineering and technology colleges in India, some of which randomly assign students to classrooms.We find that reservation category faculty have lower levels of education, lower professorial ranks and fewer years of experience in academia than general category faculty who are not hired through reservations.Yet, even with lower qualifications, we do not detect meaningful differences in instructional quality between reservation category faculty and general category faculty across a wide range of measures that include course grades, follow-on course grades, standardized test scores, dropout, attendance, graduate school plans, and graduation.In fact, we find that, at least for immediate effects on course grades, students taught by reservation category faculty perform slightly better than students taught by general category faculty.We also estimate "teacher-like-me" effects of reservation category faculty on the relative course performance and longer-term outcomes of reservation category students and cannot reject the hypothesis of null effects.Furthermore, even in the face of potential discrimination and resentment against faculty hiring quotas, general category students perform slightly better in classrooms taught by reservation category faculty than general category faculty.Finally, estimates of differences between reservation category and general category faculty in several measures of research productivity and administrative service are small in magnitude and not statistically different from zero.The findings have implications for the heated debates over affirmative action programs found in many countries around the world and in India which is now the largest country in the world.
Affirmative action programs are often criticized because of concerns that they result in lower worker productivity and efficiency losses.We study the relative productivity of workers benefiting from an aggressive affirmative action policy in a setting where hiring constraints are mandated upon employers and are typically likely to bind.In India, colleges are required to reserve approximately 50 percent of faculty hires for individuals from disadvantaged caste and social class groups.We collect and analyze data from a nationally representative sample of 50 engineering and technology colleges in India, some of which randomly assign students to classrooms.We find that reservation category faculty have lower levels of education, lower professorial ranks and fewer years of experience in academia than general category faculty who are not hired through reservations.Yet, even with lower qualifications, we do not detect meaningful differences in instructional quality between reservation category faculty and general category faculty across a wide range of measures that include course grades, follow-on course grades, standardized test scores, dropout, attendance, graduate school plans, and graduation.In fact, we find that, at least for immediate effects on course grades, students taught by reservation category faculty perform slightly better than students taught by general category faculty.We also estimate "teacher-like-me" effects of reservation category faculty on the relative course performance and longer-term outcomes of reservation category students and cannot reject the hypothesis of null effects.Furthermore, even in the face of potential discrimination and resentment against faculty hiring quotas, general category students perform slightly better in classrooms taught by reservation category faculty than general category faculty.Finally, estimates of differences between reservation category and general category faculty in several measures of research productivity and administrative service are small in magnitude and not statistically different from zero.The findings have implications for the heated debates over affirmative action programs found in many countries around the world and in India which is now the largest country in the world.

Abstract via openalex

014

Income Inequality and Intergenerational Mobility in Russia: Trends, Sources, and Measurement

Review of income inequality trends and intergenerational mobility in Russia, with family background associations in education, occupation, income.

AbstractThis chapter documents trends in income and inequality in Russia from the collapse of the Soviet Union through 2025, drawing on official statistics, the existing literature, and household survey microdata.It also reviews evidence on intergenerational mobility.While there is a consensus that inequality dramatically increased during the 1990s, its post-2000 trajectory is contested.Initial household survey studies usually found a decline, whereas official statistics reported a comparatively stable Gini coefficient.Our analysis of Russia's major household surveys, harmonized using common measurement conventions, aligns with the survey-based literature, indicating that the Gini coefficient generally fell from approximately 0.45-0.48 in the early 2000s to 0.29-0.38 in the early 2020s.We further find that most inequality arises within rather than between regions, urban and rural areas, and education and occupation groups.Household labormarket attachment, region, and household composition account for the largest shares of the explained variance in log income.The large literature on inequality is complemented by a small literature on intergenerational mobility.Here, the various studies on intergenerational dynamics in education, occupation, and income across post-Soviet cohorts conclude that, even against the background of dramatic structural changes, family background remains strongly associated with children's outcomes.
This chapter documents trends in income and inequality in Russia from the collapse of the Soviet Union through 2025, drawing on official statistics, the existing literature, and household survey microdata.It also reviews evidence on intergenerational mobility.While there is a consensus that inequality dramatically increased during the 1990s, its post-2000 trajectory is contested.Initial household survey studies usually found a decline, whereas official statistics reported a comparatively stable Gini coefficient.Our analysis of Russia's major household surveys, harmonized using common measurement conventions, aligns with the survey-based literature, indicating that the Gini coefficient generally fell from approximately 0.45-0.48 in the early 2000s to 0.29-0.38 in the early 2020s.We further find that most inequality arises within rather than between regions, urban and rural areas, and education and occupation groups.Household labormarket attachment, region, and household composition account for the largest shares of the explained variance in log income.The large literature on inequality is complemented by a small literature on intergenerational mobility.Here, the various studies on intergenerational dynamics in education, occupation, and income across post-Soviet cohorts conclude that, even against the background of dramatic structural changes, family background remains strongly associated with children's outcomes.

Abstract via openalex

015

Public R&D and the Automation of Manufacturing: Evidence from the Space Race

NASA R&D shaped automation-intensive patents and lowered labor share in exposed industries; direction of innovation and labor effects.

AbstractThis paper examines how public research and development shaped the direction of technological change during the Cold War-era Space Race and its consequences for US manufacturing.While prior research has documented the effects of public R&D on innovation and economic growth, less is known about whether mission-oriented public investment directed innovation toward automation and how this affected labor markets.We exploit variation in exposure to NASA's Space Race investments across industries and locations to estimate the effects of public R&D on innovation and manufacturing outcomes from 1947 to 1992.Combining newly developed measures of automation in full-text patent data with historical evidence on government involvement in patenting, we show that NASA-sponsored patents are systematically more automation-intensive than non-NASA patents within the same technological classes.Industries and locations more exposed to Space Race activity experienced persistent declines in labor share, consistent with the diffusion of labor-saving technologies developed for space applications.In contrast, we find little evidence of systematic changes in skill composition.These findings show that mission-oriented public R&D can shape the direction of technological change and have lasting distributional consequences.
This paper examines how public research and development shaped the direction of technological change during the Cold War-era Space Race and its consequences for US manufacturing.While prior research has documented the effects of public R&D on innovation and economic growth, less is known about whether mission-oriented public investment directed innovation toward automation and how this affected labor markets.We exploit variation in exposure to NASA's Space Race investments across industries and locations to estimate the effects of public R&D on innovation and manufacturing outcomes from 1947 to 1992.Combining newly developed measures of automation in full-text patent data with historical evidence on government involvement in patenting, we show that NASA-sponsored patents are systematically more automation-intensive than non-NASA patents within the same technological classes.Industries and locations more exposed to Space Race activity experienced persistent declines in labor share, consistent with the diffusion of labor-saving technologies developed for space applications.In contrast, we find little evidence of systematic changes in skill composition.These findings show that mission-oriented public R&D can shape the direction of technological change and have lasting distributional consequences.

Abstract via openalex

Preprints 16

016

Learning about Markets from Models: Evidence from an AI Hiring Platform

Firm-specific ML hiring models on an AI platform for low-skill workers; retention predictions and employer differences in valuing traits.

AbstractMachine learning models are increasingly used to not only inform, but make decisions. Sometimes platforms add value by pooling data across contexts to improve models' decision making---but when the optimal decision differs by context, pooling may not increase accuracy. We argue that a platform storing starkly distinct models across contexts nevertheless captures a model comparison externality. Together, these models reflect how decision makers respond differently to the same data in the current equilibrium. Policymakers can use this differentiation to prioritize interventions; firms can learn how they are already differentiated from competitors to focus future investments. Our evidence comes from an AI hiring platform used by large employers of low-skill workers. The platform uses a standardized psychometric assessment to create firm-specific models estimating each application's retention likelihood at that firm. Each firm's model is highly predictive for itself but no better than chance at other firms. We use these models to create a firm-application matrix: each firm scores each application, not only the ones it receives. An application's improved retention likelihood if submitted to a different firm is highly heterogeneous based on traits. Further, these gains are due to two distinct mechanisms. Finally, firms value traits differently even in this low skill market: they disagree about whether important traits improve or worsen retention at the margin. These insights are only possible when a platform can access both the full application pool and each firm-specific model. We explain the value of these insights to firms, policymakers, and platforms.
Machine learning models are increasingly used to not only inform, but make decisions. Sometimes platforms add value by pooling data across contexts to improve models' decision making---but when the optimal decision differs by context, pooling may not increase accuracy. We argue that a platform storing starkly distinct models across contexts nevertheless captures a model comparison externality. Together, these models reflect how decision makers respond differently to the same data in the current equilibrium. Policymakers can use this differentiation to prioritize interventions; firms can learn how they are already differentiated from competitors to focus future investments. Our evidence comes from an AI hiring platform used by large employers of low-skill workers. The platform uses a standardized psychometric assessment to create firm-specific models estimating each application's retention likelihood at that firm. Each firm's model is highly predictive for itself but no better than chance at other firms. We use these models to create a firm-application matrix: each firm scores each application, not only the ones it receives. An application's improved retention likelihood if submitted to a different firm is highly heterogeneous based on traits. Further, these gains are due to two distinct mechanisms. Finally, firms value traits differently even in this low skill market: they disagree about whether important traits improve or worsen retention at the margin. These insights are only possible when a platform can access both the full application pool and each firm-specific model. We explain the value of these insights to firms, policymakers, and platforms.

Abstract via arxiv

017

What is Work For? Labor, Politics, and Meaning in the Age of AI

Cross-national survey of work motivations and basic income substitution, framed around AI labor transition.

AbstractThis paper investigates the role of work in providing meaning, purpose, community, and other non-monetary benefits. Using an original cross-national survey of 8,234 working-age adults across the UK, US, Denmark, and India, I study the factors that motivate people in their work, with an eye toward the future AI economic transformation. I find that 81% of respondents report being satisfied with their job, and a majority across all four countries are strongly motivated by the non-monetary components of their employment. Using an original "Work Values" Index, I show that one third of working respondents are motivated intrinsically by their job--valuing their role for its own sake. Nearly one quarter are driven by social and community-related aspects of their employment, 17% by social status, and just over one quarter by the transactional financial benefits of their job. Although a majority of people would leave their role if offered their current salary as a guaranteed income, one fifth of workers would stay even if offered up to 120% of their salary. Intrinsically motivated workers were disproportionately likely to refuse a basic income at any substitution level. Groups attached to their job for non-monetary reasons were also more likely to pursue volunteering or work-like activities if they received guaranteed income substitution but could no longer perform their role. Crucially, these non-monetary work motivations are evenly distributed across the income distribution and between white- and blue-collar workers. I argue that these findings have implications for managing the AI transition, understanding attitudes to AI, and making sense of policy preferences. Guaranteeing a basic income may indeed be necessary for managing any future AI labor shock, yet it could ultimately fail to compensate workers for any loss of the non-monetary benefits they derive from work.
This paper investigates the role of work in providing meaning, purpose, community, and other non-monetary benefits. Using an original cross-national survey of 8,234 working-age adults across the UK, US, Denmark, and India, I study the factors that motivate people in their work, with an eye toward the future AI economic transformation. I find that 81% of respondents report being satisfied with their job, and a majority across all four countries are strongly motivated by the non-monetary components of their employment. Using an original "Work Values" Index, I show that one third of working respondents are motivated intrinsically by their job--valuing their role for its own sake. Nearly one quarter are driven by social and community-related aspects of their employment, 17% by social status, and just over one quarter by the transactional financial benefits of their job. Although a majority of people would leave their role if offered their current salary as a guaranteed income, one fifth of workers would stay even if offered up to 120% of their salary. Intrinsically motivated workers were disproportionately likely to refuse a basic income at any substitution level. Groups attached to their job for non-monetary reasons were also more likely to pursue volunteering or work-like activities if they received guaranteed income substitution but could no longer perform their role. Crucially, these non-monetary work motivations are evenly distributed across the income distribution and between white- and blue-collar workers. I argue that these findings have implications for managing the AI transition, understanding attitudes to AI, and making sense of policy preferences. Guaranteeing a basic income may indeed be necessary for managing any future AI labor shock, yet it could ultimately fail to compensate workers for any loss of the non-monetary benefits they derive from work.

Abstract via arxiv

018

Algorithmic Alertness and Imprinted Exits: Artificial Intelligence, Founder Characteristics, Opportunity Recognition, and Exit in South America

Conceptual essay on AI-enabled opportunity recognition, founder characteristics and exit in South American ventures, with propositions.

AbstractArtificial intelligence (AI) is reshaping how new ventures are conceived, built, and ultimately exited, yet most theorizing about these shifts has been developed in North American and European settings where capital markets are deep and institutions comparatively stable. This essay develops a conceptual framework that links founder characteristics, AI-enabled opportunity recognition, and entrepreneurial exit in South America, a region marked by institutional voids, concentrated family ownership, rapid consumer adoption of AI, and thin, contracting equity markets. Drawing on upper echelons and imprinting theories, the cognitive literature on opportunity recognition, and an emerging view of AI as an entrepreneurial actor rather than a passive tool, we argue that AI does not so much substitute for founder characteristics as re-weight them. Algorithmic search lowers the cost of identifying opportunities but raises the premium on founder judgment, entrepreneurial drive, and political skill, which govern whether AI-surfaced opportunities can be legitimized and exploited where formal institutions are weak. At the exit stage, founder imprints formed under the region's recurrent macroeconomic crises shape not only whether founders sell or list, but whether they later reverse a public listing altogether, a pattern currently visible in the wave of delistings on Brazil's B3. The essay advances six propositions and outlines an agenda for research on AI-augmented entrepreneurship in emerging economies.
Artificial intelligence (AI) is reshaping how new ventures are conceived, built, and ultimately exited, yet most theorizing about these shifts has been developed in North American and European settings where capital markets are deep and institutions comparatively stable. This essay develops a conceptual framework that links founder characteristics, AI-enabled opportunity recognition, and entrepreneurial exit in South America, a region marked by institutional voids, concentrated family ownership, rapid consumer adoption of AI, and thin, contracting equity markets. Drawing on upper echelons and imprinting theories, the cognitive literature on opportunity recognition, and an emerging view of AI as an entrepreneurial actor rather than a passive tool, we argue that AI does not so much substitute for founder characteristics as re-weight them. Algorithmic search lowers the cost of identifying opportunities but raises the premium on founder judgment, entrepreneurial drive, and political skill, which govern whether AI-surfaced opportunities can be legitimized and exploited where formal institutions are weak. At the exit stage, founder imprints formed under the region's recurrent macroeconomic crises shape not only whether founders sell or list, but whether they later reverse a public listing altogether, a pattern currently visible in the wave of delistings on Brazil's B3. The essay advances six propositions and outlines an agenda for research on AI-augmented entrepreneurship in emerging economies.

Abstract via osf

019

ChatGPT Access and Writing Skill: A Bayesian Reanalysis of an Online Randomized Experiment

Bayesian reanalysis of the online experiment on ChatGPT access in professional writing; effects on skill differences, enjoyment, workplace use.

AbstractGenerative writing tools are entering professional work, and their effects differ across people. I reanalyze a published online randomized experiment in which college-educated professionals in six writing-intensive occupations completed two incentivized writing tasks, and a randomly assigned half were given access to ChatGPT before the second task (Noy & Zhang, 2023a). I ask whether access weakens the link between first-task and second-task writing grades, whether that weakening is stronger for writing than for content, whether access narrows differences between people more for originality than for writing quality, and how access relates to felt effectiveness and enjoyment on the task. Grades and ratings are standardized; the analysis plan, fixed beforehand, retests hypotheses already tested on these data. Bayesian regression models with weakly informative priors are fit to the linear and convex incentive conditions. Each estimate is a posterior median with a 95% equal-tailed credible interval and the posterior probability of the predicted direction, P(direction), with no decision cutoffs. Under access, second-task writing grades depended less on first-task grades, −0.159 [−0.317, −0.002], with P(direction) = .976; the data did not distinguish content from writing interactions, or originality from writing spread ratios. Access was associated with higher enjoyment (0.44 [0.29, 0.60]); felt effectiveness was 0.12 [−0.06, 0.29], P(direction) = .905. After a disclosed code correction, two-month workplace use of ChatGPT, the secondary outcome, was 13.4 [3.0, 23.5] percentage points higher under access. For organizations, results suggest one structured encounter was associated with higher enjoyment and followed by more use at work.
Generative writing tools are entering professional work, and their effects differ across people. I reanalyze a published online randomized experiment in which college-educated professionals in six writing-intensive occupations completed two incentivized writing tasks, and a randomly assigned half were given access to ChatGPT before the second task (Noy & Zhang, 2023a). I ask whether access weakens the link between first-task and second-task writing grades, whether that weakening is stronger for writing than for content, whether access narrows differences between people more for originality than for writing quality, and how access relates to felt effectiveness and enjoyment on the task. Grades and ratings are standardized; the analysis plan, fixed beforehand, retests hypotheses already tested on these data. Bayesian regression models with weakly informative priors are fit to the linear and convex incentive conditions. Each estimate is a posterior median with a 95% equal-tailed credible interval and the posterior probability of the predicted direction, P(direction), with no decision cutoffs. Under access, second-task writing grades depended less on first-task grades, −0.159 [−0.317, −0.002], with P(direction) = .976; the data did not distinguish content from writing interactions, or originality from writing spread ratios. Access was associated with higher enjoyment (0.44 [0.29, 0.60]); felt effectiveness was 0.12 [−0.06, 0.29], P(direction) = .905. After a disclosed code correction, two-month workplace use of ChatGPT, the secondary outcome, was 13.4 [3.0, 23.5] percentage points higher under access. For organizations, results suggest one structured encounter was associated with higher enjoyment and followed by more use at work.

Abstract via osf

020

Le Transducteur. La fabrique ordinale de l'inégalité réelle. Une étude de cas par simulation dans les sociétés racialisées

Agent-based simulation of how ordinal sorting criteria in organizations reproduce group access gaps in racialized societies.

AbstractThe Transducer: The Ordinal Making of Real Inequality. A Simulation Case Study in Racialized Societies. Abstract How does inequality reproduce itself when no rule commands it? This article forges the concept of the Transducer, the ordinal sorting infrastructure that captures the fold, the sedimented history of past inequalities, within the apparently neutral criteria of organizations, and converts it into durable exclusion without intention, doctrine or bias. An agent-based simulation establishes its logical sufficiency: two properties, sorting the weakest and excluding to the point of renunciation, generate a widening access gap between groups, invisible to audits controlling the file and carried mostly by those who stop applying. Ordinal sorting thus hollows out real equality at its very core, access to resources, as rule compliance coexists with widening gaps. The field of application is racialized societies, where documentation is richest, and the text extracts three falsifiable signatures: the divergence between the gap at the counter and the gap over the population, the share of renunciation in the total gap, and the share of the gap retained by the record. Its counterpart, the Detransducer, points to ex ante vigilance over access distributions. Computational package archived on Zenodo, fixed seeds, exact bit-level reproduction. Résumé Comment l'inégalité se reproduit-elle alors qu'aucune règle ne la commande ? Cet article forge le concept de Transducteur, l'infrastructure de tri ordinal qui capture le pli, l'histoire sédimentée des inégalités passées, dans les critères apparemment neutres des organisations, et le convertit en exclusion durable sans intention, doctrine ni biais. Une simulation multi-agents en établit la suffisance logique : deux propriétés, trier les plus faibles et exclure jusqu'au renoncement, engendrent un écart d'accès croissant entre les groupes, invisible pour l'audit qui contrôle le dossier, porté pour l'essentiel par les personnes qui cessent de demander. Le tri ordinal creuse ainsi l'égalité réelle en son cœur même, l'accès aux ressources, puisque la conformité des règles coexiste avec l'élargissement des écarts. Le champ d'application désigne les sociétés racialisées, là où la documentation est la plus riche, et le texte en extrait trois signatures falsifiables, la divergence entre l'écart au guichet et l'écart sur la population, la part du renoncement dans l'écart total, la part de l'écart que le registre conserve. Son envers, le Détransducteur, désigne une vigilance ex ante sur la distribution des accès. Paquet computationnel archivé sur Zenodo, graines fixées, reproduction exacte au bit près.
The Transducer: The Ordinal Making of Real Inequality. A Simulation Case Study in Racialized Societies. Abstract How does inequality reproduce itself when no rule commands it? This article forges the concept of the Transducer, the ordinal sorting infrastructure that captures the fold, the sedimented history of past inequalities, within the apparently neutral criteria of organizations, and converts it into durable exclusion without intention, doctrine or bias. An agent-based simulation establishes its logical sufficiency: two properties, sorting the weakest and excluding to the point of renunciation, generate a widening access gap between groups, invisible to audits controlling the file and carried mostly by those who stop applying. Ordinal sorting thus hollows out real equality at its very core, access to resources, as rule compliance coexists with widening gaps. The field of application is racialized societies, where documentation is richest, and the text extracts three falsifiable signatures: the divergence between the gap at the counter and the gap over the population, the share of renunciation in the total gap, and the share of the gap retained by the record. Its counterpart, the Detransducer, points to ex ante vigilance over access distributions. Computational package archived on Zenodo, fixed seeds, exact bit-level reproduction. Résumé Comment l'inégalité se reproduit-elle alors qu'aucune règle ne la commande ? Cet article forge le concept de Transducteur, l'infrastructure de tri ordinal qui capture le pli, l'histoire sédimentée des inégalités passées, dans les critères apparemment neutres des organisations, et le convertit en exclusion durable sans intention, doctrine ni biais. Une simulation multi-agents en établit la suffisance logique : deux propriétés, trier les plus faibles et exclure jusqu'au renoncement, engendrent un écart d'accès croissant entre les groupes, invisible pour l'audit qui contrôle le dossier, porté pour l'essentiel par les personnes qui cessent de demander. Le tri ordinal creuse ainsi l'égalité réelle en son cœur même, l'accès aux ressources, puisque la conformité des règles coexiste avec l'élargissement des écarts. Le champ d'application désigne les sociétés racialisées, là où la documentation est la plus riche, et le texte en extrait trois signatures falsifiables, la divergence entre l'écart au guichet et l'écart sur la population, la part du renoncement dans l'écart total, la part de l'écart que le registre conserve. Son envers, le Détransducteur, désigne une vigilance ex ante sur la distribution des accès. Paquet computationnel archivé sur Zenodo, graines fixées, reproduction exacte au bit près.

Abstract via osf

021

Where Work Is Moving Occupations, Academic Programs and Places in the 2035 AI Workforce Scenario: A Roadmap for Higher Education

Task-level model of AI adoption estimating occupation, program and regional labor-demand shifts by 2035, with higher-education implications.

AbstractThe McKinsey Global Institute (MGI) estimates that, by 2035, automation could reduce U.S. labor demand by about a fifth of current work hours, while broader growth creates more jobs than automation removes. About 11 million workers may need to change occupations in the next decade. Colleges and universities plan programs years ahead, so they need to know what this scenario means for the programs they offer and the places they serve. We translate the MGI scenario into estimates for 830 detailed occupations, 1,432 academic program families at four award levels, and 530 metropolitan and nonmetropolitan areas. A task-level model of AI adoption, built from language-model exposure, observed AI success and review time, the cost of errors and governance, distributes MGI's totals across occupations. We also report the model's independent estimate and explain where and why it differs from MGI. Almost no occupation disappears: one falls by more than half and 62 fall by 20–50%. About 9.5 million jobs are in occupations that decline, almost half of them in office and administrative support. Program risk sits below the bachelor's degree. About 8% of certificate and 7% of associate completions are in programs whose graduates' occupations shrink by 10% or more, while 48% of certificates and 32% of associate degrees are in programs whose fields grow by 10% or more. At the bachelor's and graduate levels demand is broadly stable, but AI is expected to adopt or augment more than half of graduates' work hours, so the task is curriculum reinvention rather than closure. The need to change jobs is greatest in rural, agricultural and lower-wage areas; the wages at stake are greatest in high-wage hubs. We set out a roadmap for investment, reinvention, consolidation and review.
The McKinsey Global Institute (MGI) estimates that, by 2035, automation could reduce U.S. labor demand by about a fifth of current work hours, while broader growth creates more jobs than automation removes. About 11 million workers may need to change occupations in the next decade. Colleges and universities plan programs years ahead, so they need to know what this scenario means for the programs they offer and the places they serve. We translate the MGI scenario into estimates for 830 detailed occupations, 1,432 academic program families at four award levels, and 530 metropolitan and nonmetropolitan areas. A task-level model of AI adoption, built from language-model exposure, observed AI success and review time, the cost of errors and governance, distributes MGI's totals across occupations. We also report the model's independent estimate and explain where and why it differs from MGI. Almost no occupation disappears: one falls by more than half and 62 fall by 20–50%. About 9.5 million jobs are in occupations that decline, almost half of them in office and administrative support. Program risk sits below the bachelor's degree. About 8% of certificate and 7% of associate completions are in programs whose graduates' occupations shrink by 10% or more, while 48% of certificates and 32% of associate degrees are in programs whose fields grow by 10% or more. At the bachelor's and graduate levels demand is broadly stable, but AI is expected to adopt or augment more than half of graduates' work hours, so the task is curriculum reinvention rather than closure. The need to change jobs is greatest in rural, agricultural and lower-wage areas; the wages at stake are greatest in high-wage hubs. We set out a roadmap for investment, reinvention, consolidation and review.

Abstract via osf

022

Do Some Schools Mitigate and Others Exacerbate Inequality? Between-School Variation in the Income-Achievement Association

Norwegian register data on between-school variation in the parental income-achievement link; socioeconomic stratification through schooling.

AbstractWhether schools mitigate or exacerbate social inequality has been debated for decades. Much of the evidence concerns the average effect of schooling relative to its absence, while less is known about whether schools differ in how strongly parental income predicts achievement. Using population-wide register data on 759,037 students in 2,699 Norwegian primary schools, we show that the income-achievement association in 8th grade varies considerably across schools. In some schools, the association is more than twice as large as in others, and the variation persists when we adjust for family background and 5th-grade test scores. It is largely unrelated to the socioeconomic composition of schools, but the association is somewhat weaker in schools with higher value-added, which contribute more to students’ learning. If this variation reflects what schools do, schooling can be compensatory on average while some schools do much less than others to weaken this association. Schools then differ in how much parental income matters for their students’ achievement, not only in how much their students learn.
Whether schools mitigate or exacerbate social inequality has been debated for decades. Much of the evidence concerns the average effect of schooling relative to its absence, while less is known about whether schools differ in how strongly parental income predicts achievement. Using population-wide register data on 759,037 students in 2,699 Norwegian primary schools, we show that the income-achievement association in 8th grade varies considerably across schools. In some schools, the association is more than twice as large as in others, and the variation persists when we adjust for family background and 5th-grade test scores. It is largely unrelated to the socioeconomic composition of schools, but the association is somewhat weaker in schools with higher value-added, which contribute more to students’ learning. If this variation reflects what schools do, schooling can be compensatory on average while some schools do much less than others to weaken this association. Schools then differ in how much parental income matters for their students’ achievement, not only in how much their students learn.

Abstract via osf

023

How earnings and public transfers shape the educational gradient in partnered mothers’ poverty trajectories in Austria

Austrian register event-study of mothers' earnings losses and poverty after childbirth, by education; gender and class inequality.

AbstractThe transition to parenthood is a major source of gender inequality in economic vulnerability. This paper examines how mothers’ own earnings, public transfers, and partners’ income jointly shape poverty risks after first birth and how the importance of different sources of income varies by education. Using Austrian administrative register data covering more than 34,000 first-time mothers and their partners, we follow couples from two years before to six years after childbirth using an event-study design. We compare individual and household poverty measures and decompose disposable income into earnings and public transfers to assess the role of public and private resources. We find that although mothers experience substantial and persistent earnings losses after childbirth, increases in economic vulnerability are considerably smaller at the household than at the individual level and are partly cushioned by public transfers, reflecting protection through partners’ earnings and public transfers, especially family-related ones. Educational differences are pronounced: higher-educated mothers experience larger earnings losses but remain comparatively protected, both through higher partner earnings and more effective public transfers, whereas lower-educated mothers face substantially higher and more persistent poverty risks despite smaller earnings losses. Our findings show that the economic consequences of childbirth cannot be inferred from earnings losses alone and highlight the importance of both public and private resources in shaping inequality after parenthood.
The transition to parenthood is a major source of gender inequality in economic vulnerability. This paper examines how mothers’ own earnings, public transfers, and partners’ income jointly shape poverty risks after first birth and how the importance of different sources of income varies by education. Using Austrian administrative register data covering more than 34,000 first-time mothers and their partners, we follow couples from two years before to six years after childbirth using an event-study design. We compare individual and household poverty measures and decompose disposable income into earnings and public transfers to assess the role of public and private resources. We find that although mothers experience substantial and persistent earnings losses after childbirth, increases in economic vulnerability are considerably smaller at the household than at the individual level and are partly cushioned by public transfers, reflecting protection through partners’ earnings and public transfers, especially family-related ones. Educational differences are pronounced: higher-educated mothers experience larger earnings losses but remain comparatively protected, both through higher partner earnings and more effective public transfers, whereas lower-educated mothers face substantially higher and more persistent poverty risks despite smaller earnings losses. Our findings show that the economic consequences of childbirth cannot be inferred from earnings losses alone and highlight the importance of both public and private resources in shaping inequality after parenthood.

Abstract via osf

024

Mapping Women’s Contributions to Artificial Intelligence Using Large-Scale Text Analysis

Text analysis of 500,000 AI papers showing women's underrepresentation, distinct topics and lower citations; gender inequality in AI research.

AbstractParticipation in AI research shapes which type of AI is built, whose societal needs it addresses, and how it is used in society. Despite increasing attention to gender diversity in AI, we lack systematic evidence on the research topics, methods used, and collaboration structures associated with female AI researchers. This paper addresses this significant gap by developing the first large-scale, validated mapping of women’s contributions to both generative and non-generative AI research across data science and the main social science disciplines (economics, political science, and sociology). Using a sample of over 500,000 papers published between 2023 and 2026, we document three key findings. First, women are under-represented among academic AI authors across all four disciplines and recent years, and least represented in data science. Second, conditional on participating, women’s AI research is thematically distinctive: it is more applied, social and purpose-driven, concentrated in health, education and fairness, and, in data science, oriented towards using rather than building AI. Third, this distinctive work is heard less: conditional on working on the same AI topic women and men are cited equally, but the topics women disproportionately work on are themselves cited less, so women’s ideas receive systematically fewer citations. This voice gap is therefore a topic-composition penalty rather than a same-topic penalty, and it reduces efficiency by giving a distinctive, society-oriented research agenda too little voice in how AI develops.
Participation in AI research shapes which type of AI is built, whose societal needs it addresses, and how it is used in society. Despite increasing attention to gender diversity in AI, we lack systematic evidence on the research topics, methods used, and collaboration structures associated with female AI researchers. This paper addresses this significant gap by developing the first large-scale, validated mapping of women’s contributions to both generative and non-generative AI research across data science and the main social science disciplines (economics, political science, and sociology). Using a sample of over 500,000 papers published between 2023 and 2026, we document three key findings. First, women are under-represented among academic AI authors across all four disciplines and recent years, and least represented in data science. Second, conditional on participating, women’s AI research is thematically distinctive: it is more applied, social and purpose-driven, concentrated in health, education and fairness, and, in data science, oriented towards using rather than building AI. Third, this distinctive work is heard less: conditional on working on the same AI topic women and men are cited equally, but the topics women disproportionately work on are themselves cited less, so women’s ideas receive systematically fewer citations. This voice gap is therefore a topic-composition penalty rather than a same-topic penalty, and it reduces efficiency by giving a distinctive, society-oriented research agenda too little voice in how AI develops.

Abstract via osf

025

Same Job, Different Status? The Social Significance of Network Status Deviation Across 28 Societies

Introduces network status deviation relative to occupational peers across 28 societies; stratification and status beyond occupation.

AbstractSocial status is a key dimension of stratification. Since Weber, social honor belongs to persons and groups and is expressed partly through the company they keep. Yet occupational status scales assign the same status to everyone in the same job. We introduce Network Status Deviation (NSD): the distance between the weighted mean status of a person’s contacts and that of their occupational peers’ contacts, capturing whether a person keeps better, worse, or similar company than their peers. Leveraging ISSP 2017 position-generator data across 28 societies, we derive this measure from original commensality, ISEI, and SIOPS scores. Net of class, education, and income, NSD and occupational status are jointly associated with subjective social position, health and well-being, social trust, and socio-political participation and orientations, with magnitudes comparable to class and education. To illuminate the social significance of status, the company people keep is as important as their occupation.
Social status is a key dimension of stratification. Since Weber, social honor belongs to persons and groups and is expressed partly through the company they keep. Yet occupational status scales assign the same status to everyone in the same job. We introduce Network Status Deviation (NSD): the distance between the weighted mean status of a person’s contacts and that of their occupational peers’ contacts, capturing whether a person keeps better, worse, or similar company than their peers. Leveraging ISSP 2017 position-generator data across 28 societies, we derive this measure from original commensality, ISEI, and SIOPS scores. Net of class, education, and income, NSD and occupational status are jointly associated with subjective social position, health and well-being, social trust, and socio-political participation and orientations, with magnitudes comparable to class and education. To illuminate the social significance of status, the company people keep is as important as their occupation.

Abstract via osf

026

Schools and Learning Inequality: A Theoretical Model and Quasi-Experimental Evidence

Formal model and Danish register quasi-experiment on school quality and socioeconomic learning gaps; stratification and reproduction.

AbstractDo schools reduce or amplify learning gaps by socioeconomic background? I develop a formal model that clarifies the estimands implicit in the literature and synthesizes its seemingly contradictory findings by distinguishing between exposure, or how much schooling children receive, and quality, or which schools they attend. The model shows that schools can compensate for inequality, by displacing more unequal learning outside school, while simultaneously reproducing it, both positionally, through the sorting of children into schools of unequal quality, and dispositionally, through differences in how children convert schooling into learning. Empirically, I test three of the model's predictions using Danish register data and variation from changes in school district boundaries over time. I find that the effects of school quality exceed those of exposure, are concentrated among socioeconomically disadvantaged children, and are larger where sorting into schools is stronger. Equalizing school quality can therefore be an effective means of reducing learning inequalities.
Do schools reduce or amplify learning gaps by socioeconomic background? I develop a formal model that clarifies the estimands implicit in the literature and synthesizes its seemingly contradictory findings by distinguishing between exposure, or how much schooling children receive, and quality, or which schools they attend. The model shows that schools can compensate for inequality, by displacing more unequal learning outside school, while simultaneously reproducing it, both positionally, through the sorting of children into schools of unequal quality, and dispositionally, through differences in how children convert schooling into learning. Empirically, I test three of the model's predictions using Danish register data and variation from changes in school district boundaries over time. I find that the effects of school quality exceed those of exposure, are concentrated among socioeconomically disadvantaged children, and are larger where sorting into schools is stronger. Equalizing school quality can therefore be an effective means of reducing learning inequalities.

Abstract via osf

027

VOICE, POWER, AND AGENTIC AI: WOMEN’S STRATEGIC AGENCY IN IDENTITY-DIVERSE DECISION ENVIRONMENTS IN THE GLOBAL SOUTH

Survey of 376 Pakistani women on enterprise agentic AI use, leadership, fairness and psychological empowerment in workplaces.

AbstractGenerative and agentic artificial intelligence are reshaping strategic decision-making, but their empowering effects depend on how participation, leadership, technological capability, hierarchy, and algorithmic legitimacy interact. Drawing on Sociotechnical Systems Theory and Psychological Empowerment Theory, this study tests a moderated-mediation model using survey data from 376 women in decision-involved roles in Pakistan and PLS-SEM. Women’s Engagement in Decision-Making, Inclusive Leadership, and Generative AI Capability positively predict Perceived Enterprise Use of Agentic AI Capabilities (PEAIC), which in turn increases Psychological Empowerment and mediates all three relationships. Power Distance Orientation weakens the three antecedent–PEAIC links, whereas Perceived Algorithmic Fairness strengthens the PEAIC–empowerment relationship. The findings show that human-centric AI transformation depends not only on technical capability but also on voice, authority, and perceived fairness in identity-diverse Global South workplaces.
Generative and agentic artificial intelligence are reshaping strategic decision-making, but their empowering effects depend on how participation, leadership, technological capability, hierarchy, and algorithmic legitimacy interact. Drawing on Sociotechnical Systems Theory and Psychological Empowerment Theory, this study tests a moderated-mediation model using survey data from 376 women in decision-involved roles in Pakistan and PLS-SEM. Women’s Engagement in Decision-Making, Inclusive Leadership, and Generative AI Capability positively predict Perceived Enterprise Use of Agentic AI Capabilities (PEAIC), which in turn increases Psychological Empowerment and mediates all three relationships. Power Distance Orientation weakens the three antecedent–PEAIC links, whereas Perceived Algorithmic Fairness strengthens the PEAIC–empowerment relationship. The findings show that human-centric AI transformation depends not only on technical capability but also on voice, authority, and perceived fairness in identity-diverse Global South workplaces.

Abstract via osf

028

AI evaluators favor job applications written by certain AI models, independent of candidate quality

Audit of LLM-written cover letters screened by LLM evaluators; model-choice bias and fairness in AI-assisted hiring.

AbstractJob applicants increasingly use large language models (LLMs) to write cover letters, while employers deploy LLMs to screen them. In this all-AI setting, we ask whether the \emph{choice of model} affects hiring outcomes independently of candidate qualifications. We introduce a multi-model auditing framework pairing eleven LLM writer agents with nine API-hosted evaluator agents across ten job descriptions and fifty candidates per job. Four findings emerge. First, we identify a generative preference hierarchy: a consistent global ranking of writer models by evaluator acceptance, in which locally hosted models are systematically disadvantaged regardless of their license, compounded by self-preference bias where evaluators favour their own outputs. Second, evaluator models differ substantially in how much bias they introduce based on applicants' tool choice, offering actionable guidance for fairer screening design. Third, this hierarchy has measurable competitive consequences: in the most adverse pairing, approximately two (1.8) out of twenty-five qualified candidates are filtered out purely due to model mismatch. Fourth, a common mechanism explains the patterns above: writer preference and self-preference trace to a combination of how much each writer leans on specific stylistic cues and how strongly evaluators, often unanimously, reward or penalise those same cues. A further check finds little evidence that an unmeasured "writer fingerprint" remains once these same features are accounted for. We discuss implications for algorithmic fairness in employment, regulation of AI-assisted screening, and the design of auditing frameworks that can be reapplied as the model landscape evolves.
Job applicants increasingly use large language models (LLMs) to write cover letters, while employers deploy LLMs to screen them. In this all-AI setting, we ask whether the \emph{choice of model} affects hiring outcomes independently of candidate qualifications. We introduce a multi-model auditing framework pairing eleven LLM writer agents with nine API-hosted evaluator agents across ten job descriptions and fifty candidates per job. Four findings emerge. First, we identify a generative preference hierarchy: a consistent global ranking of writer models by evaluator acceptance, in which locally hosted models are systematically disadvantaged regardless of their license, compounded by self-preference bias where evaluators favour their own outputs. Second, evaluator models differ substantially in how much bias they introduce based on applicants' tool choice, offering actionable guidance for fairer screening design. Third, this hierarchy has measurable competitive consequences: in the most adverse pairing, approximately two (1.8) out of twenty-five qualified candidates are filtered out purely due to model mismatch. Fourth, a common mechanism explains the patterns above: writer preference and self-preference trace to a combination of how much each writer leans on specific stylistic cues and how strongly evaluators, often unanimously, reward or penalise those same cues. A further check finds little evidence that an unmeasured "writer fingerprint" remains once these same features are accounted for. We discuss implications for algorithmic fairness in employment, regulation of AI-assisted screening, and the design of auditing frameworks that can be reapplied as the model landscape evolves.

Abstract via arxiv

029

When AI performs, what remains? An integrative review of generative AI and cognitive capital in individuals and organizations

Integrative review of generative AI in knowledge work, proposing a cognitive capital framework with mechanisms for skill retention and organizational learning.

AbstractGenerative artificial intelligence is rapidly entering knowledge work, and its short-term productivity benefits are increasingly documented. Less is known about what happens, over time, to the knowledge, skills and judgment that remain in people and organizations. This integrative narrative review addresses that question through the construct of cognitive capital: the accumulated, renewable and depletable stock of knowledge, cognitive and metacognitive capabilities, and practices that enables an individual or an organization to learn, interpret information, exercise judgment, solve problems and adapt. We distinguish individual and organizational cognitive capital from the capability of the human-AI system, so that gains in joint performance and losses in human capability can be observed at the same time. Drawing on experimental, field and conceptual work, we propose four mechanisms through which interaction with AI affects cognitive capital: substitution of practice, scaffolding and feedback, metacognitive calibration, and organizational aggregation. Seven principles summarize the current evidence, including that assisted performance differs from retained competence, that the same technology can support or displace learning depending on design and use, and that individual gains can coexist with reduced collective diversity. For each mechanism we derive testable propositions, and we discuss five design levers for organizations as proposals to be validated rather than established practice. We conclude that cognitive capital offers a multilevel frame for governing AI adoption and outline a research agenda centered on longitudinal, multilevel measurement.
Generative artificial intelligence is rapidly entering knowledge work, and its short-term productivity benefits are increasingly documented. Less is known about what happens, over time, to the knowledge, skills and judgment that remain in people and organizations. This integrative narrative review addresses that question through the construct of cognitive capital: the accumulated, renewable and depletable stock of knowledge, cognitive and metacognitive capabilities, and practices that enables an individual or an organization to learn, interpret information, exercise judgment, solve problems and adapt. We distinguish individual and organizational cognitive capital from the capability of the human-AI system, so that gains in joint performance and losses in human capability can be observed at the same time. Drawing on experimental, field and conceptual work, we propose four mechanisms through which interaction with AI affects cognitive capital: substitution of practice, scaffolding and feedback, metacognitive calibration, and organizational aggregation. Seven principles summarize the current evidence, including that assisted performance differs from retained competence, that the same technology can support or displace learning depending on design and use, and that individual gains can coexist with reduced collective diversity. For each mechanism we derive testable propositions, and we discuss five design levers for organizations as proposals to be validated rather than established practice. We conclude that cognitive capital offers a multilevel frame for governing AI adoption and outline a research agenda centered on longitudinal, multilevel measurement.

Abstract via arxiv

030

We Ran a Research Sprint With AI. The AI Worked. The Research Didn't

Pilot research sprint with teams using generative AI; consumption bottleneck and AI-induced silos in collaborative work.

AbstractGenerative AI tools for research are widely deployed, yet individual productivity gains do not guarantee better team research. We argue that how AI use is structured matters alongside access. Personal AI tools may amplify two failures under deadline pressure: a consumption bottleneck, in which generation outpaces team-visible evaluation, and AI-induced research silos, in which private human-AI loops displace shared decision-making. We motivate this position with a mechanism-generating pilot: a 100-minute research sprint with twelve researchers in three independently operating teams, within a seventeen-person session. Both patterns recurred across teams and evidence sources. Of 33 logged AI interactions, one was accepted outright and 27 lacked a recorded disposition; missing records alone do not establish absent evaluation. A parallel autonomous pipeline produced a complete draft without in-session human evaluation, illustrating production capacity rather than comparative quality. We propose Human-AI-Human collaboration: an interaction topology in which AI connects researchers' contexts and supports inspectable collective evaluation and decision-making. The pilot does not isolate AI-specific effects from time pressure or workflow practices. We close with seven modular, falsifiable design hypotheses, prioritizing shared context and team-visible evaluation checkpoints for controlled follow-up.
Generative AI tools for research are widely deployed, yet individual productivity gains do not guarantee better team research. We argue that how AI use is structured matters alongside access. Personal AI tools may amplify two failures under deadline pressure: a consumption bottleneck, in which generation outpaces team-visible evaluation, and AI-induced research silos, in which private human-AI loops displace shared decision-making. We motivate this position with a mechanism-generating pilot: a 100-minute research sprint with twelve researchers in three independently operating teams, within a seventeen-person session. Both patterns recurred across teams and evidence sources. Of 33 logged AI interactions, one was accepted outright and 27 lacked a recorded disposition; missing records alone do not establish absent evaluation. A parallel autonomous pipeline produced a complete draft without in-session human evaluation, illustrating production capacity rather than comparative quality. We propose Human-AI-Human collaboration: an interaction topology in which AI connects researchers' contexts and supports inspectable collective evaluation and decision-making. The pilot does not isolate AI-specific effects from time pressure or workflow practices. We close with seven modular, falsifiable design hypotheses, prioritizing shared context and team-visible evaluation checkpoints for controlled follow-up.

Abstract via arxiv

031

When Disclosure Is Compulsory: Evaluating Ride-Hailing Rebalancing Policies with Self-Interested LLM Driver Agents

Simulation of LLM driver agents under EU platform work disclosure rules; effects on driver income and platform profit.

AbstractThe EU Platform Work Directive, to be transposed by December 2026, requires ride-hailing platforms to state in plain language the main parameters of their automated decisions and how a worker's behaviour affects them. Such rules are exactly what a large language model (LLM) can read, so drivers will be able to let one decide for them. We compare two main families of rebalancing policy once their rules are public: a reactive policy (R) that sends idle drivers to cover requests just lost, and an anticipatory policy (A) that plans moves on a demand forecast. Each driver is an LLM agent maximizing its own net income, given no rules (N) or the rules at the directive's level of detail (L). We test two LLMs, Claude Haiku 5.5 and Jev, which understand the rules equally well, in a three-zone model and in Manhattan. With the rules disclosed, drivers follow the platform more closely: in Manhattan, the share of relocation offers they decline falls from about four in five to one in four under both policies. Yet the effects differ by policy. For the platform, disclosure costs R 3.7% of its profit in Manhattan, as R now pays for moves drivers used to make for free, while A's profit is unchanged; the two policies, which tie on profit under full compliance, end 6.1% apart. For drivers, disclosure raises income in the small model under both policies and LLMs; for users, it raises the service rate there by 10 to 18 percentage points. In Manhattan, self-interested drivers also make A a win-win-win against full compliance, raising driver income, platform profit and service rate by 10 to 12% each. Rebalancing policies that perform equally well under full compliance can thus show opposite results under mandatory disclosure. The study offers a framework to evaluate such policies under the new requirement and is, to our knowledge, the first to do so with LLM agents.
The EU Platform Work Directive, to be transposed by December 2026, requires ride-hailing platforms to state in plain language the main parameters of their automated decisions and how a worker's behaviour affects them. Such rules are exactly what a large language model (LLM) can read, so drivers will be able to let one decide for them. We compare two main families of rebalancing policy once their rules are public: a reactive policy (R) that sends idle drivers to cover requests just lost, and an anticipatory policy (A) that plans moves on a demand forecast. Each driver is an LLM agent maximizing its own net income, given no rules (N) or the rules at the directive's level of detail (L). We test two LLMs, Claude Haiku 5.5 and Jev, which understand the rules equally well, in a three-zone model and in Manhattan. With the rules disclosed, drivers follow the platform more closely: in Manhattan, the share of relocation offers they decline falls from about four in five to one in four under both policies. Yet the effects differ by policy. For the platform, disclosure costs R 3.7% of its profit in Manhattan, as R now pays for moves drivers used to make for free, while A's profit is unchanged; the two policies, which tie on profit under full compliance, end 6.1% apart. For drivers, disclosure raises income in the small model under both policies and LLMs; for users, it raises the service rate there by 10 to 18 percentage points. In Manhattan, self-interested drivers also make A a win-win-win against full compliance, raising driver income, platform profit and service rate by 10 to 12% each. Rebalancing policies that perform equally well under full compliance can thus show opposite results under mandatory disclosure. The study offers a framework to evaluate such policies under the new requirement and is, to our knowledge, the first to do so with LLM agents.

Abstract via arxiv

SSRN & Scholar alerts 2

032

Disentangling the roots of success: How the CEO’s social-class origin affects firm growth

German CEO sample linking childhood economic and cultural resources of social-class origin to firm growth.

AbstractManagement scholars have recently started to disentangle how the social-class background of top executives influences firms’ strategic outcomes. Using a large sample of German CEOs, we explore the role that economic and cultural resources available during childhood play in fostering firm growth, both individually and in combination. Our analyses reveal that the economic and cultural resources marking a CEO’s social-class origin relate to firm growth in complex ways. While both are positively associated with firm growth, their relative importance differs across growth measures, and their interaction is negatively related to sales and total asset growth. Contour plot analyses attribute lower growth rates particularly to firms led by CEOs who grew up with either low or high combinations of economic and cultural resources. Additional analyses show that these relationships differ within the CEO group depending on CEOs’ own educational attainment, pointing to mobility-related heterogeneity in how class origin translates into firm growth. Besides highlighting the relevance of imprinting, our exploratory study emphasizes the importance of disentangling social subclasses within the CEO group and their impact on different growth types.
Management scholars have recently started to disentangle how the social-class background of top executives influences firms’ strategic outcomes. Using a large sample of German CEOs, we explore the role that economic and cultural resources available during childhood play in fostering firm growth, both individually and in combination. Our analyses reveal that the economic and cultural resources marking a CEO’s social-class origin relate to firm growth in complex ways. While both are positively associated with firm growth, their relative importance differs across growth measures, and their interaction is negatively related to sales and total asset growth. Contour plot analyses attribute lower growth rates particularly to firms led by CEOs who grew up with either low or high combinations of economic and cultural resources. Additional analyses show that these relationships differ within the CEO group depending on CEOs’ own educational attainment, pointing to mobility-related heterogeneity in how class origin translates into firm growth. Besides highlighting the relevance of imprinting, our exploratory study emphasizes the importance of disentangling social subclasses within the CEO group and their impact on different growth types.

Abstract via openalex

033

Explaining Social Class Identity Over Time: The Increasing Role of Asset Ownership

Australian survey series showing asset ownership increasingly shapes subjective social class identity.

AbstractThe contemporary social sciences are seeing a resurgence of interest in social class, prompted by widening economic inequalities. While much research associated with this resurgence remains true to the core maxims of class analysis, it has yet to fully grapple with contemporary transformations in class structure. In advanced economies such as Australia, the combined effects of wage stagnation and asset price inflation, supported by pro-asset owner policy regimes, have made asset ownership increasingly central to life chances. In this article, we explore whether the rise in asset-based inequalities is effecting a change in subjective social class identities. We ask: who identifies as working or middle class? And, have the determinants of social class identification changed over time? To investigate these questions, we analyse the series of Australian Election Study surveys fielded from 1987 to 2022, focusing on the period since 2001. By examining change in the determinants of social class identity over time, we show that homeownership has become an increasingly important factor shaping identification with the middle class, while renters are becoming more likely to identify as working class. While asset ownership has become increasingly important for class identities, the impact of more traditional indicators of class, including education and supervisory roles at work, has weakened over time. This article therefore demonstrates how asset ownership is transforming social class identities.
The contemporary social sciences are seeing a resurgence of interest in social class, prompted by widening economic inequalities. While much research associated with this resurgence remains true to the core maxims of class analysis, it has yet to fully grapple with contemporary transformations in class structure. In advanced economies such as Australia, the combined effects of wage stagnation and asset price inflation, supported by pro-asset owner policy regimes, have made asset ownership increasingly central to life chances. In this article, we explore whether the rise in asset-based inequalities is effecting a change in subjective social class identities. We ask: who identifies as working or middle class? And, have the determinants of social class identification changed over time? To investigate these questions, we analyse the series of Australian Election Study surveys fielded from 1987 to 2022, focusing on the period since 2001. By examining change in the determinants of social class identity over time, we show that homeownership has become an increasingly important factor shaping identification with the middle class, while renters are becoming more likely to identify as working class. While asset ownership has become increasingly important for class identities, the impact of more traditional indicators of class, including education and supervisory roles at work, has weakened over time. This article therefore demonstrates how asset ownership is transforming social class identities.

Abstract via crossref

Built Nov 18, 2026 · 848 papers seen, 33 in scope