> We are interested in estimating the importance of exposure to adult neighbors in a particular occupation, given by β. The challenge is that we lack data on νi, and we also likely lack data on relevant local amenities. Both of these unobserved factors are plausibly correlated with the presence of neighbors in a given occupation, creating a potential source of bias.

> To illustrate the problem, consider children growing up in coal-mining towns. These children are more likely than the average child in the full-count census to live next door to coal miners, simply because coal mining is the dominant local occupation. They are also more likely to become coal miners themselves. However, this relationship may be driven as much by proximity to coal mines as by direct exposure to adult coal miners living nearby. As another example, wealthy households—where many adults hold high-status occupations such as doctors or lawyers—tend to sort into the same high-prestige neighborhoods. As a result, children from high-income families are more likely to live near adults in these occupations and, in turn, to enter similar professions. This pattern could arise even if true causal exposure effects were zero, since high-income parents possess the resources and social capital to steer their children toward high-paying occupations. These examples illustrate the well-known reflection problem articulated by Manski (1993).

> We argue that comparing individuals at extremely fine geographic scales can resolve this reflection problem. Individuals residing on the same census manuscript sheet—on average encompassing only 7.6 households—are exposed to the same local amenities within that microgeographic area. At the same time, households cannot choose the occupations of their immediate next-door neighbors. Consequently, conditional on residing on the same manuscript sheet, all households are equally likely to live immediately next door to a neighbor in any given occupation.