research

We study how the release of ChatGPT affected posted U.S. labor demand, using 368 million Lightcast job postings (2016–2025) and a usage-anchored measure of LLM exposure. While we find a negative and causal effect on the volume of postings in AI-exposed occupations, we find no post-release effect on their seniority composition. After November 2022, posting volume in top-quartile-exposed occupations contracts by 9 log points (approximately 8.6 percent) relative to less-exposed occupations within the same metropolitan market and month. The estimate survives an extensive battery of robustness checks and is concentrated in measured LLM usage rather than AI capability. In exposed occupations, junior postings fall considerably more than senior postings after the release, but the divergence predates ChatGPT: it opens in 2021–22, during the COVID recovery, and shows no break at the release. The differential that a simple pre/post comparison would attribute to ChatGPT is absorbed by predetermined remote-work exposure. Our results caution against reading exposure-based entry-level declines as evidence of AI-driven seniority-biased technological change.

generative AI; ChatGPT; labor demand; seniority; job postings; difference-in-differences.

J23, J24, O33, M51.

Why can corruption attract dishonest workers into public service while deterring honest ones? We introduce self-control into occupational choice. Low-integrity agents enter for corruption rents, high-integrity agents remain untempted, and intermediate agents resist at a utility cost. This creates an interior deficit in the entry frontier, unlike the selection patterns generated by the reputation and Roy benchmarks considered here. The mechanism distinguishes rent prevention from delayed sanctions. Holding the commitment payoff fixed, prevention weakly increases entry and reduces the mass and share of corrupt entrants relative to sanctions, because temptation underweights delayed punishment. Under severe temptation, prevention yields a double dividend in the uniform benchmark: less corruption and more public employment. In the linear model, the geometry generates distinct tipping thresholds for public employment and average integrity. Both converge to 1 + √2 as rents vanish under regular joint type distributions. For sufficiently small positive rents, the integrity threshold lies above the employment threshold, creating a region where employment contracts while integrity deteriorates. Only the employment threshold remains exactly 1 + √2 in the independent-uniform benchmark. The two-kink frontier also identifies temptation parameters under stated support conditions and commensurable cardinal measures of integrity and outside opportunities, jointly observed with realized entry. Existing experiments provide descriptive motivation but do not implement this structural test.

occupational selection; self-control and temptation; corruption; reputation and self-image; public sector labor markets; institutional design.

D73, J45, D90, H83.

This paper studies whether exposure to geolocated crime news shifts individual voting behavior in Italian elections. Using a panel of non-relocating Bologna voters observed across multiple election rounds, we exploit within-individual variation in exposure to nearby crime news during the pre-election month, holding constant individual and election fixed effects. Aggregate crime-news exposure yields weak and imprecise associations. Disaggregating by the offender’s perceived nationality reveals a sharp asymmetry. At national elections, exposure to immigrant-attributed crime news near a voter’s residence is associated with a within-individual decline of roughly 2 percentage points in support for the Five Star Movement—the largest party without a clear immigration stance during the sample period—and with a comparable rise in abstention; crime news attributed to Italian citizens carries no comparable association. The response is specific to crime content: comparable local non-crime news produces no analogous shift. These asymmetric responses, absent in aggregate measures, are more consistent with a nationality cue in crime reporting that activates an immigration-and-security frame than with a uniform reaction to crime salience. The evidence points to the attribution of crime in local news, rather than its mere occurrence, as the margin that matters for electoral behavior. A timing falsification supports the campaign-window reading but is sensitive to the offender-nationality classifier; we accordingly read the estimates as associations, with a cross-source replication under way to settle the causal interpretation.

crime; immigration; elections; news media; individual voting behavior.

D72, D83, K42, L82.

Using administrative turnout records from Bologna, Italy (2004–2013), linked to annual population registers, we examine how childbirth changes the mother–father turnout gap. Linking both parents to the same child, we find that the gap worsens for the mother by 5.7 percentage points in the first month after a birth and attenuates over the first year. In a separate comparison, mothers delivering in the week before the polling period are 16 points less likely to vote than those delivering in the week after. The maternal turnout shortfall relative to the pre-maternity benchmark diminishes over time, but the evidence does not establish full recovery. At two to five years after the first identified birth, estimates are −0.9 points in the full sample and +0.3 in a high-confidence sample. The main full-sample specification excludes losses exceeding approximately 2.4 points at the 95 percent confidence level; this bound is sensitive to the reference elections used.

voter turnout; motherhood; childbirth; gender; political participation; household production.

D72, J13, J16, J22.

Does proximity to locally provided public goods raise the political participation of the households they are meant to serve? We study Bologna, Italy, with administrative turnout records for all registered voters at four elections (2004–2013), linked to annual population registers, income, exact dates of birth, and a purpose-built geocoded census of childcare facilities. Our baseline follows parents with one co-resident child in the annual register as the child ages across service regimes, so that the relevant distance switches within person, and estimates separate distance slopes for local and national elections. The municipal-facility distance slope in local elections is statistically indistinguishable from zero, with a symmetric equivalence bound of 0.47 percentage points per 100 meters, or 1.25 points across the interquartile range of distance. The local–national contrast that the accountability hypothesis restricts is estimated less precisely: its bound is 0.79 points per 100 meters, and moderate effects on it remain compatible with the evidence. Proximity does not measurably moderate the parenthood penalty, no gender interaction survives multiplicity adjustment, and a switching cohort and six additional local amenities and services yield the same result. An earlier marginal contrast among parents with verified address continuity is not robust to specification and is reported as exploratory. The estimand is proximity: the estimates bound the mobilizing effect of living near a facility and leave the effect of obtaining a place in it unmeasured.

voter turnout; public goods; childcare; gender; local elections.

D72, H41, J13, J16.