Wealth plays an important role in social stratification but the results that can be obtained when analyzing wealth as a predictor variable depend on modeling decisions. Although wealth consists of multiple components it is often operationalized as net worth. Moreover, wealth effects are likely non-linear, but the functional form is often unknown. To overcome these problems, we propose to 1) split up net worth into gross wealth and debt and evaluate their joint effect and 2) use non-parametric Generalized Additive Models. We show in a simulation study that this approach describes systematic wealth differences in more detail and overfits less to random variation in the data than standard approaches. We then apply the approach to re-analyze wealth gaps in educational attainment in the US. We find that the operationalization of wealth as net worth results in a misclassification of which children have the best and the worst educational prospects. Not negative net worth is associated with the worst educational prospects but only the combination of low gross wealth and low debt. The most advantaged group are not only children with high net worth but all children with high gross wealth independent of the households’ amount of debt.

Close

财富在社会分层中发挥着重要作用，但将财富作为预测变量进行分析时可以获得的结果取决于建模决策。尽管财富由多个组成部分组成，但它通常以净资产的形式进行操作。此外，财富效应可能是非线性的，但其函数形式往往是未知的。为了克服这些问题，我们建议 1）将净资产分解为总财富和债务，并评估它们的联合效应；2）使用非参数广义加性模型。我们在模拟研究中表明，与标准方法相比，这种方法更详细地描述了系统性财富差异，并且对数据随机变化的过度拟合更少。然后，我们应用该方法重新分析美国教育程度方面的贫富差距。我们发现，将财富作为净值进行操作会导致对哪些孩子的教育前景最好和最差的分类错误。负净资产并不与最差的教育前景相关，而只与低总财富和低债务的组合相关。最有利的群体不仅是高净资产的儿童，而且是所有拥有高总财富的儿童，与家庭债务数额无关。

关闭