Look at the changes, not at the levels (Part II)

I have a new paper out that looks at the causal impact of caring for an elderly or disabled person. A pretty large literature has suggested that carers suffer large penalties in employment, wages and happiness. But the problem with this is that almost all the previous studies have used cross-sectional variation. When you look at panel data (following the same people over time), the employment impacts are negative but much smaller, and the other effects are close to zero.

Substantively, this paper is about caregiving. But the methodological issue is probably more important. For policy purposes, cross-sectional comparisons risk being biased by ‘unobservable’ factors. Panel data analysis helps take care of this, at least to the extent that those unobservables don’t vary within the same person over time.*

As I’ve mentioned recently, the same issue arises across countries. If you compare outcomes at a single point in time, you can get quite a different answer than if you track a panel of countries over time.

Anyhow, here’s the abstract of the carers paper.

Informal Care and Labor Market Participation
Understanding the effect of informal care for an elderly or disabled person on labor market outcomes is important for developing policies targeted towards caregivers. However, because of omitted variables bias, simple cross-sectional relationships may provide a misleading picture of the causal impact of informal care provision on labor force status. To address this, I use panel data for the period 2001−2007, which make it possible to track the same individuals over time, and observe how their outcomes alter as their care arrangements change. While caregiving does appear to have a modest negative impact on labor force participation, this impact is only one-quarter to one-sixth as large in the panel as in the cross-section. Taking account of individual heterogeneity, the impact of caregiving on other labor force outcomes (and on life satisfaction) seems to be small or non-existent. Large estimated effects from cross-sectional regressions are most likely driven by individual heterogeneity. One possible interpretation of this result is that the impact of caregiving on labor market outcomes and life satisfaction takes several years to manifest itself. Another is that the causal effect of caregiving on labor force outcomes and life satisfaction is quite small.

* Of course, a downside of the panel data approach is that the outcome measure must also vary within the same individual over time. So while I’d warrant that the recent “TV kills” study suffers from unobservables bias, we can hardly use panel data to solve the problem.

1 thought on “Look at the changes, not at the levels (Part II)”

  1. Very timely as usual Andrew! I have been looking at this area in the context of demand for respite and family support (ie does increasing the level of respite increase participation in the workplace). This looks fascinating.


Comments are closed.