Date Published: May 30, 2019
Publisher: Public Library of Science
Author(s): Lina Hedman, David Manley, Maarten van Ham, Petri Böckerman.
Previous research has reported evidence of intergenerational transmissions of neighbourhood status and social and economic outcomes later in life. Research also shows neighbourhood effects on adult incomes of both childhood and adult neighbourhood experiences. However, these estimates of neighbourhood effects may be biased because confounding factors originating from the childhood family context. It is likely that part of the neighbourhood effects observed for adults, are actually lingering effects of the family in which someone grew up. This study uses a sibling design to disentangle family and neighbourhood effects on income, with contextual sibling pairs used as a control group. The sibling design helps us to separate the effects of childhood family and neighbourhood context from adult neighbourhood experiences. Using data from Swedish population registers, including the full Swedish population, we show that the neighbourhood effect on income from both childhood and adult neighbourhood experiences, is biased upwards by the influence of the childhood family context. Ultimately, we conclude that there is a neighbourhood effect on income from adult neighbourhood experiences, but that the childhood neighbourhood effect is actually a childhood family context effect. We find that there is a long lasting effect of the family context on income later in life, and that this effect is strong regardless the individual neighbourhood pathway later in life.
There is an emerging body of literature that highlights the importance of taking into account the neighbourhood in which an individual grew up as a means to understand their later in life trajectories. Empirical evidence suggests that there is a correlation between the neighbourhood types experienced during childhood and the neighbourhoods where one lives in adulthood [1,2,3,4,5,6]. Other studies show that the neighbourhood environment experienced during childhood has a causal and long-lasting influence on adulthood socio-economic status outcomes, such as income [7,8,9,10,11,12]. The size of the effects of the childhood neighbourhood on individual outcomes is unclear, both in absolute and relative terms. When reviewing literature on neighbourhood effects on children’s outcomes, Ginther and colleagues  found effects which varied from substantial to almost non-existent. They argued that one explanation for these disparities could be that models of neighbourhood effects often do not control for family characteristics, which can result in biased outcomes. There are some examples of studies which argue that neighbourhood effects are very small or even non-existent when taking the family context into account (for instance see ). It is notable that family effects on children’s outcomes are generally found to be substantially higher than neighbourhood effects .
Neighbourhood effects arise due to critical spatial context exposures that affect individual life opportunities through a set of transmission mechanisms . Although the residential neighbourhood does not represent the full range of exposures that an individual experiences , it acts like an access point through which many other contextual spaces are accessed. Hence, geographic variation in the local spatial opportunity structure  not only concerns the neighbourhood but also higher geographic levels in which the neighbourhood is situated within (for example, school attachment areas, city districts, the municipality etc.). There is a vast literature analysing how neighbourhood exposures affect individual life opportunities. This literature includes outcomes such as indicators of socio-economic status, school performance, health, cognitive abilities, behaviours etc., and encompasses studies from different countries and cities, using different methodological approached and data sets, as well as varying neighbourhood definitions. Most of these studies find evidence of neighbourhood effects (there are however also examples of studies finding no effects at all; see [19, 20, 21]). Studies have also found neighbourhood effects to vary by individual characteristics [22, 23], spatial scale [24, 25] and length of exposure to certain neighbourhood types [23, 26, 27].
Using a sibling design has been argued to be a promising approach to separate neighbourhood context effects from family context effects, although such a design is not used very often . Within pairs of genetically related individuals, who also share a similar family background (siblings), many of the unmeasured influences on individual outcomes can be controlled for. If siblings are sufficiently close in age, they will have experienced a similar household environment, and it can be assumed that they have also been exposed to the same family norms, values and attitudes. They will also have similar childhood neighbourhood experiences, at least in terms of their residential locations. Any sibling correlation can thus be assumed to represent a joint effect of shared family and community characteristics . Sibling correlations in income are generally found to be about 0.45 for the U.S and 0.25 for the Scandinavian countries . This means that, in the U.S., almost half of the inequality in earnings can be attributable to siblings’ shared background.
This paper is part of a project funded by the European Research Council (ERC). As part of the granting procedure, the project proposal was evaluated by both the Delft University of Technology institutional ethics committee, and the ERC ethics committee. Both committees approved the project. The data used for this study are derived from GeoSweden, a register based longitudinal individual level micro-database owned by the Institute for Housing and Urban Research, Uppsala University. The GeoSweden database is not based on a sample, but it contains the entire Swedish population tracked from 1990 to 2010. The database is constructed from a set of different annual administrative registers including, demographic, geographic, socio-economic and real estate data for each individual living in Sweden each year. For each person in the dataset it is possible to identify their parents and through them also their siblings. Although the data used cannot be publicly shared, we have made our Stata code available through protocols.io to enhance the reproducibility of our research (http://dx.doi.org/10.17504/protocols.io.z6af9ae).
Table 2 shows the results from our ‘individual models’–the models estimating the effect of adult neighbourhood exposure at the level of individuals on income for work, separately for male and female same-sex siblings. Models 2a and 2c includes only characteristics that are not influenced by parents/childhood neighbourhood (age and country of birth). In Models 2b and 2d we add time-varying variables that are known to affect income from work (family composition, education level, employment status, tenure), but which are also highly likely to be influenced by childhood family context and childhood neighbourhood exposure.
This paper set out to better understand the effects of childhood neighbourhood context, and adulthood neighbourhood experiences on individual income from work later in life. The paper started with the idea that estimation of these neighbourhood effects is likely to be affected by the influence of the childhood family context. The childhood family sorts children into certain childhood neighbourhoods, affects adult neighbourhood careers, but also affects later in life income from work. Separating these different effects is a major challenge in neighbourhood effects research, because any childhood family effect might bias estimates of independent causal effects on income of childhood and adult neighbourhood experiences.
The results suggest that there is an adulthood neighbourhood effect on income from work, net of the effect of the childhood neighbourhood and childhood family context effects. The results also suggest that any effects on later in life income from the childhood neighbourhood context are in fact childhood family context effects. That is not to say that the childhood neighbourhood is not important at all, but likely that the childhood neighbourhood effect is the result of non-random selection of families into neighbourhoods based on family characteristics. Our analyses show that individuals with a sibling who does well in terms of their (adult) neighbourhood pathway (in other words has a low cumulative exposure to low-income neighbourhoods), have a higher predicted income from work compared to individuals with a sibling with a high exposure to low-income neighbourhoods. We interpret this as a family context effect. Those with siblings in low income neighbourhoods are assumed to come from a less resourceful or advantageous family (either in terms of finances, time investments or other unobservable but important traits such as genetics), whereas individuals whose siblings live in better neighbourhoods are assumed to benefit from a more positive family background. Our overall conclusion, therefore, is that the childhood family context has a lasting effect on adult income, even when taking both childhood and adult neighbourhood path into account. Part of what appeared to be a neighbourhood effect was in fact a lasting ‘family effect’. For the wider research literature, it is clear that, when possible, models of neighbourhood effects should control for the childhood family context to avoid bias in estimates.
A possible limitation of our study is the construction of the contextual sibling pairs. Because of pragmatic and conceptual restrictions we have used a relatively simple way to construct a control group of contextual siblings. Although we had access to full population data, imposing more restrictions on the contextual siblings would reduce the size of the control group further. A larger control group could be constructed in countries with larger populations, or by using multiple cohorts within the data. A further limitation is that the real sibling pairs differ in ways we cannot observe in the data. To reduce these possible differences, a dataset of real (preferably identical) twins could be used, but that requires a dataset with a large number of twins, requiring at least a birth cohort study or preferably a twin study. In these cases we would likely be able to acquire genetic information as well allowing further control of currently unobservable factors. However, using our design, we got the most out of the register data at our disposal.