Abstract
This study applies a multilevel design to study determinants of poverty in the Netherlands. Years of measurement, households, neighborhoods, and municipalities are treated as separate levels to answer the general questions of (1) the relative importance of these levels in predicting the likelihood that a household is poor, (2) which variables measured at the different levels explain differences in poverty risks between households, and (3) to what extent differences between neighborhoods and between municipalities arise from compositional or contextual factors. The random coefficient models, estimated using the Dutch Social and Economic Panel Survey 1984-2002 of Statistics Netherlands, show that the household level is most influential, but that the neighborhood and municipality do independently contribute to the explanation of poverty. Households with a head of 25 to 34 years, who is lower educated, who has a low status job, or who does not work, run a relatively high risk to be poor. So do couples with young children, single households, and single parent families. For a large part, differences in the risk of poverty between neighborhoods and municipalities are based on compositional differences with regard to these household specific factors. ‘Pure’ neighborhood effects nevertheless show up: the lower the socio-economic status of a neighborhood, and the higher the percentage of non-employed, the higher becomes the risk of neighborhood residents to be poor. The results also show that when the general unemployment rate is high, the likelihood of poverty is high as well.
