Division of Financial Responsibility within Mixed-Gender Couples

Abstract

This paper studies the dynamics of financial responsibility division within mixed-gender couples. Analysis is based on individuals’ self-assessments of their own contribution to four household activities collected in the Survey of Consumer Payment Choice. A series of logistic regressions link reported roles from 3728 households to respondent gender and six household characteristics, representing aggregate and relative attributes with respect to age, education, and income. A second, longitudinally-based analysis relates reported contribution levels in subsequent survey years to changes in household income dynamics. For bill payments, the data are consistent with a bargaining model in which relative income rankings, more so than other household variables, relate to responsibility shares. For decisions about saving and investments and decisions on other financial matters, in addition to income rank, there is also some evidence that greater relative educational attainment coincides with greater responsibility shares. For household shopping, however, tendencies in household role assignment seem predominantly driven by gender considerations. Females across all household types consistently do more of the shopping, and females are much more likely to increase their contribution, even when they become the primary earner.

Appendix A

In general, I adopt weakly informative priors, meant to guide estimates away from unrealistic values rather than serve as an influential source of information. Characteristic level effects that achieve the desired monotonicity condition in (5) are parameterized as follows:

$$\begin{aligned} \beta ^{(c)}_1&\sim \textrm{Normal}(0,1)\\ \gamma ^c&\sim \textrm{Normal}(0,1)\\ \tau ^c_i&\sim \textrm{Normal}^+(0,1), \text {for}\ i=2,\ldots ,\ell _c \\ \beta ^{(c)}_i&= \beta ^{(c)}_{i-1}+\tau ^c_i\times \gamma ^c, \text {for}\ i=2,\ldots ,\ell _c, \end{aligned}$$

where $\textrm{Normal}^+(0,1)$ refers to a normal distribution with standard deviation 1 restricted to non-negative values. The parameter priors are chosen with regard to the size of the jump from one characteristic level to another, so that such a change, given by $\tau _i\times \gamma$, is less than one-half about 78% of the time and less than one about 94% of the time, nontrivial changes on the logistic scale. Estimation with a more conservative prior on jump size, $\tau _c^i\sim \textrm{Normal}^+(0,0.5)$ was also used and yielded virtually identical 90% posterior intervals.

Priors for the cubic effects relating to changes in income dynamics are given $\lambda _0 \sim \textrm{Normal}(0,1)$ and $\lambda _1,\lambda _2,\lambda _3 \sim \textrm{Normal}(0,.5)$.