Research

Publications and Working Papers

Receiving vs. believing (mis)information from friends

Experimental evidence from India.

Social Networks

Belief Formation

Working paper

How much do people (additionally) believe a news story because it’s been shared by a friend? How much should they believe it for that reason? To answer these questions we conduct a series of lab experiments in India with ~800 pairs of real-life friends. Using a custom platform, we collect detailed information on how individuals (sharers) decide which stories to share, and how their friends (receivers) update their beliefs in response. We find receivers over-interpret sharing as a sign of a story’s veracity, while discounting other reasons/motivations for sharing. As a result, receivers’ trust in shared stories increases irrespective of the sharers’ belief in them, with false stories accruing the greatest additional trust. To identify mechanisms, we measure how receivers update if, instead of observing their friend’s decisions they (i) learn their friend’s beliefs, and (ii) receive computer-generated clues of known accuracy (instead of subjective signals from their friend). We find several mechanisms contribute to receivers’ biased inference: receivers overestimate how well sharers’ beliefs predict a story’s veracity; miscalculate the relevance of those beliefs to sharing decisions; and exhibit base-rate neglect, updating the most on stories they least believed originally.

The American Dream in the Great Depression

Absolute Income Mobility in the United States, 1915-1940

Inequality

Income mobility

Working paper

We use two newly digitized historical datasets to estimate rates of absolute income mobility for the cohort of sons born in the early 1910s in the United States. We find that the mobility for this cohort was 30-50 percentage points lower than for the 1940 cohort, a finding robust to alternative income imputations, inflation indices and age adjustments. The rates are similar to men born in the United States in the 1980s, suggesting that today’s young adults face mobility prospects comparable to those of adults entering the labor market during the Great Depression. But while low mobility rates for men born in the 1910s were due mainly to slow economic growth during the Depression, those of men coming of age today are due mostly to rising income inequality.

James Feigenbaum, Maximillian Hell, Robert Manduca, Jimmy Narang

The Fading American Dream

Trends in absolute income mobility since 1940

Inequality

Income mobility

We estimated rates of absolute income mobility — the fraction of children who earn more than their parents — by combining data from U.S. Census and Current Population Survey cross sections with panel data from de-identified tax records. We found that rates of absolute mobility have fallen from approximately 90% for children born in 1940 to 50% for children born in the 1980s. Increasing Gross Domestic Product (GDP) growth rates alone cannot restore absolute mobility to the rates experienced by children born in the 1940s. However, distributing current GDP growth more equally across income groups as in the 1940 birth cohort would reverse more than 70% of the decline in mobility. These results imply that reviving the “American dream” of high rates of absolute mobility would require economic growth that is shared more broadly across the income distribution.

Raj Chetty, Nathaniel Hendren, David Grusky, Maximillian Hell, Robert Manduca, Jimmy Narang

10.1126/science.aal4617

In Progress

Priors vs. Desires: what happens when biases interact

Belief Formation

In Progress

A long literature in psychology and economics documents that people update their beliefs in a biased manner upon receiving news that agrees with their priors (confirmation bias) or their desires (motivated updating / desirability bias). But what happens when these biases (i) align or (ii) conflict with each other? By combining data from existing studies with new experiments, I find that confirmation bias is first-order, and mediates the extent to which desirability bias occurs. Subjects’ beliefs show almost no movement upon receiving news that confirms their priors, irrespective of its favorability to their ego or politics. However, subjects update their beliefs considerably upon receiving disconfirming (surprising) news, with larger updates on favorable news.

Trends in the Usage of Large Language Models in the United States

Evidence from UAS data

Artificial Intelligence

LLM

Inequality

In Progress

Using a nationally representative panel (N ~ 15,000), we conducted two waves of surveys to track awareness, usage, and perceptions of large language models (LLMs) in the United States. Our data capture longitudinal usage not just by demographics and occupations but also by personal traits and preferences. We find that by Spring 2024, 24% of U.S. residents had used LLMs, and another 48% were aware of them, compared to 18% and 50.5%, respectively, in Fall 2023. LLM Usage was significantly higher among young men, Whites and Asians, those with income above $100,000, college education, analytical occupations, Democrats, and individuals above median in cognitive ability, internet skills, and openness to new experiences, with gaps relatively stable over time. Reasons for LLM usage shifted within users: work- and school-related applications increased, while curiosity- and entertainment-driven usage declined. Those using LLMs for work were 11 percentage points more likely to engage frequently (daily or weekly) and remain frequent users. These findings provide a dynamic view of LLM adoption across socio-demographic groups, highlighting persistent disparities and emerging inequalities in AI engagement.

Marco Angrisani, Maria Cassanova, Nathanael Fast, Jimmy Narang, Juliana Schroeder