METHODS

The socio-economic panel study (SOEP) is a longitudinal survey of the German population that began in 1984 and was enlarged to include the re-united East Germany in 1990.²⁷ This report uses the 16th wave from 1999 because, in this year, multiple questions concerning the home and neighbourhood environments were included in the survey.

Sampling was performed through a regionally clustered, multistage design. Within a private household, every member aged 16 and older was asked to participate. People were followed up even if they moved to another location and started a new household (though not outside the country), and all variables used in this study represent the most current household. In addition to the personal interviews, a single household member was asked to complete an additional survey about general household and neighbourhood items. Continuity of the sample between waves and response rates within waves have remained high.²⁷ All people who were interviewed in 1999 and who had the health outcome variable were included (n = 14 055).

Outcome measure

Self rated health, which has been shown to be a reliable indicator of overall health status²⁸ and predictive of mortality in Germany,²⁹ was assessed with a single item answered on a five point Likert scale. Responses were dichotomised with “not so good” and “bad” indicating poor health.

Demographic features

Age was divided into four groups as based on the sample distribution, and sex was controlled for in the study. As foreigners living in Germany were oversampled in the survey and because their housing situation has been shown to be different from Geman born persons,³⁰ immigrant status was included.

Socioeconomic variables

Home ownership was divided into owner and renters (including those who rent privately and from state subsidised plans). Education was assessed through years of schooling needed to attain their current degree, and these years were classified into three groups based on the educational system. A measure for relative household poverty was constructed as follows: firstly, the total monthly household income (in DM, after tax deduction and national insurance contributions) was divided by the number of household members (with the first household member having a weight of 1.0 and each additional member with a weight of 0.8) and, secondly, relative poverty was defined as those households having an income less than 50% of the sample’s mean.

Household features

To assess the physical household environment, household representatives were asked: “How would you characterise the condition of the house in which you live?” Households were dichotomised into homes in need of renovation (partial to complete) compared with those not in need of renovation. Household crowding was defined by the number of persons in the household divided by the number of rooms. Households were then dichotomised into a low crowding group (two thirds of the households) compared with a high crowding.

Housing type was defined as either a single or double family house (either detached or as part of a row) compared with a building with three or more apartments. A dummy variable was also created if the person lived alone or with other people. This is used here as a crude measure of social ties within the home.³¹

Neighbourhood indicators

Neighbourhood indicators were selected in order to describe the physical and social environment. As living in a rural or urban environment may intersect with these findings, distance to nearest city (in km) was asked. According to the distribution in our sample, distance from city centre was classified as less than compared with greater than 25 km. Each household representative was asked the extent to which they were affected by noise and air pollution (graded on a five point Likert scale). As these two items were highly correlated (0.714, p<0.000), they were summed and then dichotomised into a high pollution (one third of households) compared with low. With regard to the social neighbourhood environment, the household representative was asked how close was their relationship to their neighbours (answered on a five point Likert scale ranging from “very close” to “almost no contact”). Being distant with neighbours was defined as “not close” and “almost no contact”.

Whether the respondent lived in the former East or West Germany was also included as a control variable in the study as housing conditions between the two regions have been shown to vary.³²

Statistical analysis

To explore how health as well as home ownership varied between the different variables, cross tabulations with χ² statistics were used.

Because bivariate and single level logistic analyses do not account for the fact that people within households will be more alike than people from different households, the analyses are likely to underestimate standard errors. Therefore, multilevel statistical methods were used with individuals nested within households. In the following equation, the outcome Π_ij for the ith individual in the jth household is explained by the fixed intercept B0 and the m fixed effects or predictor variables:

The individual error term is represented by e which in binary analyses is confined to 1, and u represents the level 2 (household) error term.^33,34 Analyses were conducted using MLwiN version 1.1³⁵ with second order penalised quasilikelihood (PQL) used to estimate coefficients and standard errors.³⁶ As self rated health may be conceptualised as a continuum, a “latent variable approach” is used to estimate the household variance partition coefficient.³⁷ This coefficient is used to show the relative importance of the household level variation.

RESULTS

The nested structure of the dataset shows 14 055 people living in 7381 homes. Between one and six people reside in a home, and in 23.4% of homes, people live alone. The sample is almost equally divided between men (48.3%) and women (51.7%). The age distribution is as follows: 31.4% are less than 35 years old, 30.3% are between 35 and 49 years old, 23.6% are between 50 and 64, and 14.7% are 65 or older. People who were foreign born comprise 15.9% of the sample. With years of schooling, 18.8% having nine or fewer years, 51.5% having between 9 and 12 years, and 29.7% having 12 or more. Of the sample, 16.3% (2293 people) describe themselves as being in “not so good” or “bad” health.

Table 1 shows the distributions of the household and neighbourhood indicators. Of the participants, 44.0% live in their own homes and 11.1% live in relative poverty. About half (48.3%) live in buildings with three or more apartments.

Bivariate analyses by home ownership (shown in table 2) show the clustering and distribution across the study population. In relation to the socioeconomic indicators, people who live in poverty and those who have less than or equal to nine years of schooling have higher percentages who rent their homes. Those foreign born have a higher percentage who rent their homes than people who are German born. Younger people, women, people living alone, and those who live in apartment buildings, near cities, have crowded homes and homes in need of renovation, report high pollution, and distant contact with neighbours all have a higher percentage who rent their homes than the respective reference groups.

Turning to the mulitivariate analyses, the null model shows significant variation at the household level with respect to self rated health (see table 3). The variance partition coefficient is reduced to 25.6% with the inclusion of demographic features, and further reduced to 24.0% with the addition of housing tenure, education, and poverty.

In model 2, which includes demographic information, it is found that age, sex, and immigrant status are all related to the likelihood of reporting “not so good” or “poor” self rated health. In model 3, every variable is tested individually in a model containing demographic information. Household crowding, years spent in your own home, and whether a person resided in the East or the West were all non-significant. People who rented homes had an odds ratio of 1.48 (CI 1.31 to 1.68). In the model that included years of education, immigrant status was no longer significant. Socioeconomic indicators were added simultaneously to an equation containing demographic information in model 4. The effect of housing tenure is only slightly reduced (OR 1.46, CI 1.29 to 1.66). To further test this association, relative poverty was replaced with household equivalent income categorised into quartiles and then tertiles (not shown). Home ownership remained a significant factor.

In model 5, remaining household and neighbourhood variables were added. Only those variables which reached significance in model 3 were included. Here the type of house, whether a person lives alone, and the distance to the city centre were not significantly associated with the health outcome. Housing tenure did remain significant and again decreased with these additional factors (OR 1.23, CI 1.04 to 1.46). Renovation and relative poverty were the significant household factors. Perceptions of pollution and relationships with neighbours were the two neighbourhood factors that remained significant in the final model.

DISCUSSION

The study supports the argument that after controlling for multiple measures of socioeconomic position, housing tenure is associated with self rated health in Germany. This contrasts with results from two previous studies, which found that home ownership was not significantly related to self rated health, depression, or functional limitations in Germany; however, these studies had smaller samples restricted to elderly populations.^38,39 This study further found that the association between housing tenure and health may be mediated by elements of the physical and social environments in the home and surrounding community.

Though home ownership is related to higher levels of education and lower levels of poverty, controlling for education and poverty in multilevel analyses did not eliminate housing tenure’s association with health. Furthermore, home ownership in this study is unlikely to just reflect occupation status. In a subgroup analysis of 7150 full time employed people (not shown), it was found that by including occupational prestige (with low status defined as people in the sample with the lowest third on the Treiman scale)⁴⁰ along with education and income, renting your own home remained significantly associated with poor self rated health, though the association was further decreased (OR 1.35, CI 1.13 to 1.60). Thus, home ownership is different from many commonly used measures of socioeconomic position with respect to self rated health.

Moreover, controlling for income, education, and occupation may underestimate the relations between home ownership and health because it does not consider indirect effects, that is, how home ownership may operate through these factors.⁴¹ Home ownership represents a comparatively stable investment that may generate income through appreciation, tax benefits, or other forms of government subsidies.^24,26 Home owners may live in areas with better school systems or employment opportunities.⁴² Though these indirect effects seem to be small in the model that simultaneously controlled for all three factors, the life stage of the respondent needs to be taken into account. Access to schools, for example, has a very different meaning for someone who is school age compared with an older person. This study also did not control for household assets, savings, insecure home ownership,⁴³ or unemployment,⁴⁴ all of which may be important in further elucidating the relation between housing tenure, socioeconomic position, and health.

Though housing tenure may be related to health through similar mechanisms proposed for socioeconomic position (such as engaging in unhealthy behaviours, low levels of social support, and low access to medical care⁴⁵), this study directs us to features of the home and neighbourhood environment. Psychosocial features may be one link between housing tenure and health. Home has been postulated to be a site of attachment, familiarity, and identity,^46,47 and it has also been argued to be a site of ontological security where control can be exercised.⁴⁸ Greater attachment and an increased sense of security may accompany owning your own home. Research that has instrumentalised these factors and tested their associations with respect to housing tenure and to health have been rare.^15,22,49

Physical housing conditions may also help to explain the association between housing tenure and health. Worse physical features of the home—need for renovation and crowding—were both related to home renting. The need for renovation was further associated with worse self rated health. The need for household renovation is a broad category that may include such aspects as mould and dampness. As it does not adequately delineate among the various factors, more explicit questions and quantitative measures are important in future studies. Furthermore, renters may have less control over renovations and, thus, may have responded to the survey question differently. Crowding was not associated with self rated health, which may reflect the fact that diseases are less likely to be of infectious origin in contemporary Western societies. Crowding was defined relative to the sample distribution, and it may be the case that either a subjective sense of crowding or an absolute “over-crowded” level may be more important for health.

Mechanisms that link home ownership and health must also situate the home in the surrounding neighbourhood. Rented homes tend to be clustered together, creating areas of lower socioeconomic status. Neighbourhoods have been postulated to affect health through the physical, social, and service environment.^13,14,41 In terms of the physical environment, living in a rented home was more common closer to cities and in areas where pollution was high. Pollution was associated with self rated health and decreased the magnitude of the association between home ownership and health. This is consistent with past studies that report air and noise pollution are more likely to be found among poorer areas and that they negatively affect health in Germany.^20,21

The social environment of the neighbourhood may also mediate the effects of housing tenure. This study found that home owners were more likely to report that they were close with their neighbours and this was related to better self rated health. One reason for this may be that home owners in the sample tended to reside in their homes for longer periods of time (analyses not shown) and they may feel more committed to the neighbourhood. These may allow for community networks and social ties to develop^42,50 and have a beneficial effect on health. This study supports previous evidence from Germany that found that individual level indicators of social capital (reciprocity and civic trust) were correlated with housing satisfaction among the elderly population in Germany and that these were significantly associated with self rated health (submitted data). Though the service environment (such as access to healthy groceries and recreational facilities) was not included in our study, it may also mediate the relation between home ownership and health.

The use of multilevel methods in this study strengthens the validity of the results. Because people from the same household are not assumed to be independent in multilevel modelling, it allowed for the inclusion of the cases within a household while still obtaining unbiased coefficients and standard errors.³³ Multilevel statistics also provide a framework in which housing variables may be defined at the appropriate level. It showed that household level variation (42.6% in the null model) is high, thus highlighting the importance of investigating and developing interventions at the household level. Residual variation (24.1% in the final model) indicates that unmeasured factors at the household level, including a better assessment of the psychosocial and physical environments, require further elaboration.

With few people per household, it is difficult to use multilevel models to specify a very extensive random portion, and thus all variables were entered as fixed effects. As this was one of the first studies examining home ownership and health in Germany, interaction terms were not explored. Future studies should examine these interaction terms to better investigate how the associations are changed by sex, age, race, immigration status, whether the person resides in East or West Germany, and income inequality in the neighbourhood.^13,40,51 The fact that women were more likely to report worse health, even after controlling for sociodemographic indicators is consistent with previous studies and in need of further investigation.⁵²

Additional limitations need to be discussed. Firstly, because multilevel statistics were used, each person was given the same weight, and the results are therefore not nationally representative. Secondly, as this study is cross sectional, it is subject to the problem of common method variance, and no causal inference can be drawn as to the association between housing tenures and self rated health. Furthermore, the health measure was obtained through self report rather than objective measures or events and other potential health confounders, including mental illness, were not included. Depression, for example, may bias not only the reported health status but also perceptions of home and the local environment. The fact that self reported health was highly correlated with being legally disabled (0.353, p<0.000) and depression (0.513, p<0.000) lends further credence to its importance as a indicator of overall health. These limitations call for a prospective study design that uses objective measures of health.

Thirdly, as household and neighbourhood level variables were assessed by a single household representative, they do not account for varying opinions that may exist between members of the same household. In addition, the head of household may have felt pressured to make the home and neighbourhood seem better (so as to appear as a good provider). Fourthly, obtaining objective measures of the home and environment, on such features as dampness, pollution, and local amenities are also important in future studies.

Advantages of this study include its use of a large scale, population based sample to explore how home ownership is related to self rated health, controlling for multiple socioeconomic confounders and investigating numerous potential mediators. The use of multilevel methods further allows for the correct specifications and estimations. Rather than simply implementing policies to encourage home ownership, further exploration is needed to better understand how the physical and psychosocial environments in the home and the surrounding neighbourhood may affect health.