A neighbourhood effect is commonly defined as an impact on people’s socio-economic outcomes that can be attributed to differences in the neighbourhood context. Neighbourhood effects have received increasing attention in the last decade from social researchers. However, many recent reviews of the empirical literature on neighbourhood effects reach the conclusion that no consensus has been reached on whether neighbourhood effects exist, or on which neighbourhood contexts produce which outcomes, which causal relationships are at work, or which groups are affected most. The empirical identification of neighbourhood effects is impeded both by theoretical and methodological challenges and, more importantly, lack of data.
Neighbourhood effects studies aim to explain characteristics or behaviours of individuals through the neighbourhood context and assume that the populations of small geographical entities are highly homogeneous and that the areas’ characteristics can therefore be used to explain individual-level outcomes. Moreover, it is assumed that spatial proximity and social similarity leads people to become even more similar to each other.
However, once we allow the individual’s behaviour to be influenced by the context we have to acknowledge that the individual does at the same time influence the context. By choosing the neighbourhoods in which they live, people also actively seek the contexts that will influence them. It is analytically and methodologically difficult to disentangle these effects. A further problem in the identification of neighbourhood effects is that of defining the relevant contexts, both in terms of the boundaries of the spatial units and in terms of the characteristics of the areas. Which contexts matter for which outcomes? Are ‘neighbours’ just the people in the adjacent homes, or everyone in the same town? Can neighbourhood effects meaningfully be established when we consider that people are becoming increasingly mobile and that they are confronted with a number of different local and social contexts? Even if the neighbourhood effects research could provide sufficiently good answers to these questions, it would still be difficult to find the kind of data needed to investigate the matter empirically.
In the German case, it has been the lack of data that has impeded neighbourhood effects studies. It is extremely difficult and costly for individual researchers to get data describing people’s neighbourhoods. In contrast to most countries, Germany does not collect the population census data traditionally used by neighbourhood effects researchers to establish the neighbourhood context of respondents to large-scale nationally representative surveys. In Germany, the last census was collected in 1987 (East Germany: 1981). Data protection laws furthermore prohibit the disclosure of most geographically referenced data that might otherwise be available. As a result, there are only a handful of case studies for selected German cities that investigate the possible operation of neighbourhood effects (Hamburg: Alisch and Dangschat (1993), Cologne: Friedrichs (1998), Berlin: Häußermann and Kapphan (1999)).
Most of the empirical evidence on neighbourhood effects is anecdotal and does not allow causal inference. The methodological design of the studies neither allowed researchers to look at a control group (for instance, poor people living in an affluent neighbourhood) nor to follow people over time as they move to other neighbourhoods or as their neighbourhoods undergo changes.
The cornerstone for representative and longitudinal neighbourhood effects studies in Germany was laid when it became possible to match data of the German Socio-economic Panel Study (SOEP) with data at the zip-code level. Geographically referenced data at this scale have been obtained from a number of Statistical Offices of German cities. In addition, the data pools of micro-marketers offer a wide-range of qualitative neighbourhood indicators. Following strict data protection measures, select waves of SOEP have been matched on the basis of the study’s sensitive address records with a selection of these indicators. The point at which this was undertaken was the starting point for the research presented in this doctoral thesis.
This study starts with an extensive review of theoretical and empirical work on neighbourhood effects (Chapter 2). It re-examines first theoretical and then empirical research from the disciplines of economics and sociology. There have been a number of recent reviews of neighbourhood effects studies (for instance Dietz 2002; Sampson, Morenoff et al. 2002; Durlauf 2003) and we may argue that there is no need to provide yet another one. However, the economics and sociology literature proved to overlap only to a minor extent and the studies offer fairly distinct insights into neighbourhood effects. One contribution of this thesis is to bring these schools of thought together.
Comparing sociological and economic theories of neighbourhood effects, we find that the former have a tendency to focus on just parts of the broader picture. We look at the three most prominent models proposed in sociology – ‘contagion models’, ‘collective socialisation models’, and ‘institutional models’ – which are drawn upon to explain negative effects of living in neighbourhoods with a higher population of socio-economically disadvantaged people. A fourth model, the relative deprivation model, on the other hand, is used to explain positive effects of living in neighbourhoods with poor aggregate characteristics. All four models, however, could potentially also explain the opposite direction of effects. The strength of the economic theories – ‘interactions-based model’ and ‘models with local and global interactions’ in particular – is that they identify the conditions under which disadvantaged and advantaged neighbourhoods may or may not have negative effects on individual-level outcomes.
The review of empirical studies, which follows the discussion of neighbourhood effects theories, shows that most empirical studies do not set out to test any of the propositions made by theorists. We re-examine sociological and economic studies that draw on a wide range of methodologies, including ethnographical, experimental and quantitative studies in both fields. European studies are taken into account where possible given the fact that recent reviews of neighbourhood effects studies have tended to overlook the European research.
This thesis adds to the literature on neighbourhood effects by testing empirically whether people’s life satisfaction depends on their relative income position in the neighbourhood (Chapter 3). From the perspective of neighbourhood effects research this is an empirical test of relative deprivation theory, which posits that people are unhappier the better off their neighbours are. It has already been shown that people’s own income and others’ incomes matter for life satisfaction (e.g., Clark and Oswald 1996; Ferrer-i-Carbonell 2005). This makes a proof of the existence of the effect per se redundant, allowing us to focus on the identification of the particular context effect of neighbours’ income on life satisfaction. From the perspective of happiness research, the research presented in Chapter 3 is a test of whether or not the so-called ‘relative income’ hypothesis also holds when the reference group concerned is one’s neighbours.
We use a unique dataset for these analyses – the 1994 and 1999 waves of the SOEP matched with neighbourhood indicators at the German zip-code level (roughly 4,000 households). This way we know for every respondent of the SOEP how well off they are and also how well off the people in their local environment are. The richness of the dataset allows us to control extensively for other characteristics of individuals, their families and their neighbours and to formulate more sophisticated hypotheses about possible routes for the comparison effect to operate. In addition, the longitudinal structure of both our neighbourhood context dataset and the SOEP allows us to control for unobserved heterogeneity at the neighbourhood and at the individual level.
The empirical results presented in Chapter 3 suggest that there are sizeable neighbourhood effects on people’s life satisfaction: People living in small communities, self-owned property and in proximity to facilities that serve recreational purposes are happier than others. However, people in Germany are not unhappier the more income their neighbours have, as relative deprivation theory predicts, but – if anything - in fact happier. We find positive effects of neighbourhood income on happiness in all cross-sectional models and this is robust to a number of robustness tests, including adding in more controls for neighbourhood quality, changing the outcome variable, and interacting neighbourhood income with indicators that proxy the extent to which individuals may be assumed to interact with their neighbours. The impact of neighbours’ income on happiness is not highly statistically significant. In fact, it is only statistically significant in 1999 and borderline statistically significant in most models. But the neighbourhood income effect does not turn negative. The only negative effect of neighbourhood income on happiness we identify is for individuals with young children in the household. For these it appears to be a struggle to “keep up with the Schmidts” .
Do we have to conclude from these findings that the predictions of the relative deprivation theory are wrong? Or are there other reasons that may explain why we do not identify a negative comparison effect?
It is a common critique against neighbourhood effects research that it uses statistical constructs of neighbourhood that are too big and too crudely delineated to satisfy any more sophisticated definition of a neighbourhood, i.e., as a physical and social space. The scale at which we operationalise ‘neighbourhood’ is the smallest geographical entity that has ever been taken to establish the local context in German research. However, the German zip-code areas may be too large to detect the particular social comparison effect that we try to identify.
During the course of this research, more micro-geographical data at different scales of neighbourhood became available, facilitating empirical analyses that would not have been possible at the start of this project. In addition to indicators of neighbourhood income at the zip-code level, indicators that apply the same income definition became available at two much more local scales, i.e., the market-cell level and the street-section level. This allows us to investigate systematically in Chapter 4 whether the social comparison effect operates at smaller scales of neighbourhood and also to test whether critiques that have been brought up against the use of large, statistically defined neighbourhood units are confirmed by the empirical evidence. We look at whether larger neighbourhood units confound heterogeneity of the population in the neighbourhoods, and re-estimate models used in Chapter 3 including neighbourhood incomes at three different scales.
The majority of the results are the same whichever neighbourhood scale the reference income is measured at, and at the smaller geographic scales the effects are statistically significant. People in Germany are happier the more income their neighbours have, and we only find negative effects for individuals with young children in the household. For this group of the population we find that the neighbourhood income effect is more adverse when measured at the zip-code level than at the more immediate neighbourhood scales. We then address the critique that neighbourhood effects research operates with too poor a conceptualisation of neighbourhood by testing whether our neighbourhood effects are the same when we restrict the sample to individuals for whom we may assume that our neighbourhood units are not merely physical but also social spaces. The neighbourhood income effects remain positive but, for this smaller sample, they turn statistically insignificant.
It is possible that neighbourhood income effects operate at multiple scales simultaneously. At the larger geographical scale, the prosperity in the neighbourhood may proxy better employment prospects, which may translate into greater happiness, while the comparison effect may only operate at the very local level. We show that not controlling for these area effects at larger scales biases the coefficients on our more locally defined neighbourhood characteristics. We find that the income in the market cell (corresponding to the average income of approx. 400 households around a person’s home) has the biggest influence on happiness, and is positive. At this scales of neighbourhood, the size of the coefficient is least affected by omission of neighbourhood characteristics at the other scales of neighbourhood. In these models, a number of neighbourhood income effects turn negative. However, the negative effects are not statistically significant.
Perhaps close neighbours do not matter after all. This would be in line with the empirical result that we do not identify a negative comparison effect of the closest neighbours, while there are sizeable effects of higher prosperity in wider areas. The neighbourhood used to be considered the place where people work, where family and relatives live, and where friends are found. With increasing modernisation, however, people have become less dependent on their immediate local environment. They venture out into other places. Jobs, shops and places for recreation are outside peoples’ neighbourhoods. Access to public transportation and telecommunications make it possible to maintain contacts to like-minded people irrespective of where they live. Increasingly, when people perceive their prospects in the area as poor, they move away, even from family and friends. Greater mobility means that people are influenced not just by their neighbourhood but also by other contexts. This undermines the neighbourhood context hypothesis. The close neighbourhood may have lost its importance.
In Chapter 5 we investigate empirically whether it may indeed be true that greater mobility is associated with close neighbours becoming less important. We focus on three different types of mobility that have been proposed to negatively impact on neighbourliness, namely residential mobility, access to modes of public and private transport, and changes in the availability of modern communications technologies. Our proxy for how much neighbours matter is the SOEP respondents’ accounts of how often they visit with their neighbours. In comparison, we look at the strength of people’s ties with their families and the sensitivity thereof to changes in mobility.
Residential mobility, so the argument goes, increases the relative costs of investing in and maintaining links to people in the neighbourhood. Conversely, residentially mobile people have been suggested to maintain closer links to their family and relatives, i.e., the people who have been permanent companions and to whom strong emotional ties exist. Greater physical mobility means that people can more easily cover distances to reach people and places outside the neighbourhood. Since many people do not have family members living nearby, access to train services or their own car makes it possible to visit family members. Yet Internet use – dubbed virtual mobility – takes time that could be used to do other things. This in particular may have changed people’s inclination to interact with others face-to-face, not only with neighbours but also with family (or friends). It is a multifaceted tool which offers many types of activities and has an addictive potential. Its use has become less expensive over time as prices for personal computers have dropped and more Internet providers have entered the market. The Internet may be the community of the future as local communities lose their significance.
The kind of data needed to empirically investigate a complex matter like this is not readily available in any single wave of SOEP (or in fact in any other social survey). Consecutive waves of the survey were pooled and matched with micro-marketing data, ensuring that people did not change neighbourhoods between the surveys – otherwise neighbourhood characteristics, accounts of social interactions with neighbours and mobility portfolios would not refer to the same place.
The empirical analyses show increases in all forms of mobility in the decade from 1994 to 2004. Virtual mobility has entered into people’s lives at an immense pace, while residential mobility and physical mobility have increased more gradually. Parallel to this the incidence and frequency of social interactions with neighbours has declined, and we show that this is negatively associated with residential mobility, access to transport and Internet use. In contrast, the incidence and frequency of visits with family has remained constant over time. We show that family visits are not associated with changes in access to modes of transport, and that the correlations with residential and virtual mobility are in opposite directions. Greater residential mobility is correlated with more family visits and virtual mobility appears to weaken family ties as well as neighbourhood ties.
Our multivariate analyses for 1999 and 2004 confirm these results. We test for a number of alternative explanations why mobility may be associated with declines in visits with neighbours. For instance, we test whether people’s decisions to visit neighbours or family may be driven by personal preferences rather than mobility portfolios by restricting the sample to individuals who we may assume not to have chosen their neighbourhood based on personal preferences (young people living with their parents), by interacting mobility effects with personal characteristics that suggest a particular inclination toward mobility and that may be correlated with people’s sociability, and also by controlling for unobserved heterogeneity. The effects of residential and virtual mobility are robust to these alternative explanations and we do not find effects of physical mobility. While residential mobility appears to have a trade-off effect on visiting with neighbours to the benefit of visiting with family, Internet use undermines sociability with neighbours and family. The negative effect of Internet use on people’s sociability in the real world may not continue, however, since most people are already using the Internet and the magnitude of the effect was much smaller in 2004 than in 1999. Overall, our results show that social contacts with neighbours are more volatile than family ties to changes in people’s mobility. With further increases in mobility, close neighbours may become less significant but even in a mobile society such as that of Germany in 2004, the incidence of visits with neighbours is sizeable and not close to zero – in contrast to the frequent assertion in the literature that the neighbourhood does not matter.
The neighbourhood does matter – the research presented in this thesis documents this using the German example. It may not matter in the way that neighbourhood effects researchers expected and some effects prove to be difficult to identify even when very local and very timely data are used (as is the case here). Studying neighbourhoods is a very inspiring subject and the research presented in this thesis cannot necessarily cover all aspects. Chapter 6 summarises the main conclusions of this research and points out avenues that may be taken in the future to further understand how neighbourhoods affect people’s lives.