Ian Lundberg

PhD Candidate, Sociology and Social Policy, Princeton

I study topics in stratification and inequality. I strive to produce substantive findings that are conceptually precise and which rely on credible assumptions. These principles often lead me to computational and machine learning methods and the development of new approaches.

My CV contains links to all papers and replication files. Scroll down for an overview of my research targeting estimands that are predictive, descriptive, and causal.

Estimands are the starting point for methodological choices

In a vision of social science research laid out with Rebecca Johnson and Brandon Stewart, I argue that social scientists should set the research goal in precise terms that do not involve regression coefficients. This frees us to state the goal of greatest interest, even if our methods to achieve that goal involve a model that is only a rough approximation.

Predictive estimands call for the direct application of data science

It is well known that social science models do not predict well. But is this just for lack of trying?

A new way of doing science. We collaborated with hundreds of social scientists and data scientists in a research design optimized for prediction. Teams trained predictive models on a standard social science dataset. We evaluated them on a holdout set locked away until the end.

We learned new things. The best predictive performance observed holds new weight because (1) it was evaluated on holdout data and (2) it represents the best out of many diverse attempts.

Descriptive estimands show where policies are failing

Housing eviction is more common than you think

Demographers often summarize events per person-year. By this metric, eviction is rare: only 2-3 % of households per year.

A new goal. But often we care whether someone ever experiences an event over a longer period, such as at any point in childhood. It only takes one eviction to upend a child's life.

The goal matters. More than 1 in 4 children born in a large U.S. city from 1998 to 2000 experienced eviction by age 15.

Causal estimands prescribe policy solutions

Public housing protects families from eviction

Public housing provides tenants with reduced rent as well as an internal grievance procedure to resolve conflicts with the housing authority. Does public housing reduce eviction?

A new goal. An explicitly causal estimand clarifies precise assumptions for observational data point to a policy solution.

The goal matters. In our target population, public housing reduces eviction from 11 percent to 3 percent. It is difficult to argue that this large difference arises from confounding alone: a causal effect is more plausible.

Seemingly descriptive quantities often involve a hypothetical intervention

Demographers frequently study socioeconomic gaps over demographic groups conditional on policy-amenable variables.

A new goal. Drawing on work in epidemiology, one can formalize the goal as a gap-closing estimand: the expected gap if a random individual was sampled from each demographic group and assigned to a single treatment value.

The goal matters. The goal guides adaptation of double machine learning this demographic goal.

Interpretation of complex empirical quantities often requires sharpened theory

Cousins' incomes are sometimes similar. This does not imply a direct grandparent effect.

The cousins' incomes are sometimes remarkably similar. This might suggest something about how family background constrains life chances.

A new goal. What do we really mean by the "influence" of family background? We can formalize our theoretical model mathematically.

The goal matters. Several plausible theoretical models could generate any given set of sibling and cousin correlations.