Association for Public Policy Analysis & Management

Both research and our everyday experience confirm that children growing up in different (types of) neighborhoods fare differently on a number of important outcomes later in their life. William J. Wilson e.g. forcefully demonstrated this in his influential book on The truly disadvantaged: The inner city, the underclass and public policy, first published 40 years ago. A spatially uneven distribution of outcomes may be considered problematic from a social justice perspective, and consequently for social policy. Part of this uneven spatial distribution is due to sorting; different people with different initial endowments of life chances reside in different locations. However, one cannot rule out the possibility that characteristics of the childhood neighborhood exercise a causal impact upon outcomes in later life.

We study attachment to education and labor force in early adulthood, i.e. at the age of 20-22. More concisely, we analyze the probability of three different positions: that an individual is not in employment or training (NET), has completed high school and that (s)he is enrolled in higher education. The basic question is how childhood neighborhood characteristics affect the probability of each of these three outcomes. Our empirical analysis is based on a longitudinal register data set with annual information on incomes, educational achievements and family composition, covering all individuals born in Norway from 1986-1994, i.e., almost 500,000 individual observations. A nice feature of using register data is that there is no sample attrition (in excess of mortality and emigration); this is crucial when studying the interdependency between phenomena far removed in time. Furthermore, the quality of the administrative registers has been proven to be high.

In order to identify the causal effect of childhood neighborhood experiences on different outcomes one would ideally want to observe the same individual simultaneously in different neighborhood contexts and compare outcomes. This is the hypothetical counterfactual or potential outcomes approach, which is obviously impossible to achieve. We will pursue an approximate strategy that utilizes the fact that siblings share unobserved (and observed) family characteristics, and the fact that families sort themselves across neighborhoods according to these characteristics. Hence, we estimate sibling fixed effect linear probability models where the core explanatories are neighborhood context experienced during childhood.

To put it simply, our sibling fixed effect approach test whether differences in outcomes between similar individuals can be attributed to differences in childhood neighborhood qualities. The context variables used will be neighborhood income levels and variation therein, prevalence of unemployment and reliance on welfare transfers, neighborhood share of adults with high and low educations levels, and the composition of the neighborhood housing stock. Siblings residing in the same family will experience the same neighborhood context at the same chronological time, but at different ages. Consequently, our results will be driven by variations between different-age siblings, i.e. non-twins. Between-siblings variation in neighborhood qualities will in part stem from variation over time in qualities of each particular neighborhood and in part, it will arise from mobility of families.