Gender Differences in Caregiver Stressors, Social Resources, and Health: An Updated Meta-Analysis

For the first question — how large are the gender gaps among caregivers — the design is sound: a large, systematically assembled pool of studies, random-effects pooling that allows true effects to vary, and inclusion of 23 unpublished studies and 7 book chapters so publication bias could be tested. The design fails for the headline causal question. Establishing that caregiving widens the gender gap requires comparing caregivers with non-caregivers, and the paper makes no such comparison; at least six included studies had non-caregiver control groups that went unused for this purpose. The comparison is two literatures placed side by side. The adjustment analysis compounds this: the raw gender gaps come from all studies, but the adjusted gaps come from a regression on the subset reporting enough variables. For physical health, the gap within that subset is already about 0.10 before any adjustment, versus the 0.17 reported from the full pool; adjustment moves it only to 0.08. Other unaddressed confounds include the entanglement of gender with kinship (in 136 of 229 studies "female" pools wives and daughters, "male" pools husbands and sons), selection of men into the caregiver role, and shared datasets counted as separate samples.

The pooled gender differences are computed correctly, confidence intervals are internally consistent for the main outcomes, and the conversions to variance explained and to the binomial effect size display check out. The authors dropped about 20% of eligible studies for not reporting a usable gender split — studies disproportionately likely to have found nothing — and ran no formal publication-bias test. Non-independence is unhandled: the unit is "sample" rather than dataset, and several publications draw on the same underlying data. The authors ran 84 moderator tests without correction; fourteen reached conventional significance where about four would by chance, and moderator regressions fit seven predictors to as few as 28 samples. The caregiver-versus-population comparison uses reference values that are inconsistent within the paper (the population depression figure is attributed to two different sources; the population physical-health figure appears as two different values, one outside its own stated interval). The adjusted depression confidence interval is a fixed-effects single-sample interval, narrower than the raw random-effects interval despite using 38% of the participants, and is then tested for overlap against a random-effects interval from another paper. A null result for years in the caregiving role is flagged as highly significant. Several table entries have intervals not centered on their point estimates.

Every variable is a caregiver's self-report, so a "gender difference in depression" is a gender difference in how many symptoms men and women write down. Women disclose symptoms more readily than men across health surveys (Verbrugge 1985, which the paper cites). The stressor measures have the same source: the care recipient's behavior problems, the strongest predictor of depression in the adjustment model, are rated by the same caregiver whose depression is the outcome. A low mood that makes the patient seem more difficult would produce the observed correlation without any causal path from stressor to health. "Practical significance" is a 1%-of-variance convention from Cohen 1992, not a demonstration that anything differs in practice, and the variance figure for burden is 2.8% in one table and 1.7% in another. The claim that caregiving widens the gap compares populations that differ in age (the physical-health reference covers people 60 and over; roughly half the caregivers here are younger), era, and instruments. The mediation claim has three links, none observed within any study; the measured gender-to-stressor link is tiny (around 0.1–0.2 standard deviations, with zero difference in care-recipient impairment or years in the role).

The descriptive reasoning is honest: the authors expected women to face more care-recipient impairment and less social support, found neither, and said so. Adjusting for stressors reduces the depression gap from about 0.34 to 0.29 standard deviations — perhaps a quarter of the gender effect — yet the conclusion says caregiving widens the gap "primarily" because women face more stressors. The discussion credits "lower levels of social resources among female caregivers" with part of the explanation, but no support measure differed by gender, and adding support variables to the depression model changed the gender coefficient by zero. The depression comparison against the population benchmark shares an endpoint under the paper's own non-overlap rule, yet the paper calls it "rather unequivocal." The abstract closes by saying the results support stress-and-coping theory even though most of that theory's predictions failed. A hedge in the limitations ("must be interpreted with caution") does not reach the five numbered conclusions. The cohort conclusion rests on 4 of 24 tests, with the shortfall explained by gender roles being "slow to change," which would also explain zero hits. The first conclusion lists "quality of the relationship with the care recipient" as a practically important gender difference; that variable was never analyzed.

The included studies are listed and coded by variable in the reference list, funding is disclosed, and a limitations section is candid about heterogeneity and the multiple-testing risk. Nothing else needed to reproduce the analysis is available: no study-level dataset, no coding manual, no inter-coder agreement, no count of screened or excluded records beyond "approximately 20%," no count of trimmed outliers, no software or code. Two of the 229 "studies" are raw data files the authors analyzed themselves, with no description of how. No registration existed for meta-analyses in 2006, so its absence is typical of the era. The limitations section omits the points that bear most on the headline claims — the cross-literature comparison, the sample shrinkage between raw and adjusted estimates, the self-sourced benchmark — and several cross-table inconsistencies cannot be resolved from the paper alone.

Framing citations are used in the standard way and prior reviews are characterized fairly. The critical citations are the two external benchmarks for general-population gender gaps, and these are handled loosely: the population depression gap is attributed to Davis 1999 on one page and to Pinquart & Sörensen 2001 on the next, and the cited Davis paper is a meta-analysis of life events, not depression. Two of the three benchmarks are the authors' own earlier work. Nolen-Hoeksema's reviews, the standard reference for the general-population depression gap, are not cited. The paper states duplicate results from the same dataset were omitted, but Gould 2003 and Hyer 2003 share a title and author group and both contribute burden estimates; Beach 2000 and Bookwala & Schulz 2000 both draw on the Caregiver Health Effects Study. One attribution error: women's "more extensive formal support system" is credited to Okun & Keith 1998, a study of informal exchanges with family and friends.

The only funding is a career-development grant from the U.S. National Institute on Aging; there is no commercial stake in whether gender differences are large or small, and the paper reports several results that run against theories the authors cite. The benchmark against which the headline "caregiving-specific" excess is measured comes, for two of three outcomes, from the authors' own prior meta-analysis. The theoretical frame they endorse is their own field's, and at least one of their primary studies is in the pool. None of this is improper, but the conclusions that favor the authors' framework are exactly the ones with the weakest support, and nothing in the design — no preregistration, no independent analyst — guards against that.