In 2018, Medicaid spent $92b on home and community based services (HCBS), which is more than the US spent on the Supplemental Nutrition Assistance Program (“food stamps”), Temporary Assistance to Needy Families, the Earned Income Tax Credit, and numerous other social programs (Blom, 2019). This spending provides help to over 4 million disabled Americans annually (O’Malley Watts et al., 2020), a number that is sure to grow as the US population continues to age. HCBS funds largely support individuals who need assistance with day-to-day tasks like bathing, dressing, and preparing food, but prefer home-based care to living in a nursing home. There is concern that the quality of care provided by a large group of HCBS caregivers, personal care aides (“caregivers” hereafter), is sub-optimal since the vast majority receive no formal training related to the services they provide (LeBlanc et al., 2001, Spetz et al., 2019).1

In recent years, several policies have sought to promote caregiver training. The Personal and Home Care Aide State Training Demonstration Program, part of the Affordable Care Act, supported efforts to develop training for caregivers and led to free, online training programs in some states. President Biden’s Build Back Better agenda also called for investment in training programs to help caregivers deliver higher-quality care (H.R.5376). Despite the policy push to increase caregiver training, we know very little about the demand for training and consequently, how effective attempts to increase training will be.

To take first steps in this direction, we partnered with a home care agency to design, conduct, and evaluate a randomized controlled trial (RCT) that offers home-based caregivers the opportunity to complete an online training program. The majority of caregivers in our sample were once informal caregivers who are now being paid for their services through state HCBS waiver programs. Our partner agency handles the logistical tasks that the Medicaid recipient would otherwise have to complete as an employer.2 As part of our experiment, a random subset of participants are offered a $50 incentive, equivalent to nearly four-times the average worker’s hourly rate, to complete the one hour training. This incentive effectively varies the price of the training and allows us to estimate a demand curve for the training. This demand curve is useful because it quantifies information on caregivers’ willingness to pay for training and provides a first step towards analyses of policies that encourage caregiver training. We find that among individuals not offered any financial incentive, 13 percent completed training. The $50 incentive increased the completion rate by nearly 67 percent, or nine percentage points.

All of the caregivers in the experiment were contacted via email. In an effort to gauge the fraction that read the message and were therefore aware of the opportunity, we also offered a random subset of participants $10 for simply replying to the email. Based on the replies we received and training completion rates, approximately 37 percent of caregivers were aware of the training opportunity. Among those who were aware of the training, we estimate that between 35 and 51 percent of caregivers not receiving a financial incentive completed training, while 59 to 65 percent of incentivized caregivers completed. These results suggest that for training programs like the one used in our experiment, a policy that makes training freely available, and potentially offers a financial incentive for completion, can result in a majority of caregivers taking up the training without great expense. However, if the policy aims to have 100 percent participation, a large financial incentive is required. As always, these types of extrapolations rely on the validity of our estimates for other populations.

We conclude our analysis by using the results from our experiment and the Marginal Value of Public Funds (MVPF) framework (Finkelstein and Hendren, 2020) to evaluate the welfare implications of a policy that offers caregivers a free online training program.3 Such a policy may be justified in this setting by an externality that could prevent efficient levels of training: Many HCBS recipients are insured through Medicaid, meaning at least part of the benefit of training—higher quality care, which results in lower health care costs—is captured by the government insurer and not the caregiver or care recipient.4 We draw two main conclusions from our analysis. First, the policy is likely to increase welfare per dollar spent more than increasing caregiver wages in Medicaid, the primary payer for home-based caregiving. Key to this finding is the fact that offering online training is relatively inexpensive (the cost-equivalent increase in Medicaid caregiver wages is less than two pennies per hour) and our experiment suggests that many caregivers will choose to complete training even without a financial incentive. Second, if training caregivers reduces healthcare expenditures even by a small amount, then the policy is likely to cost less to finance than the healthcare expenditures saved. As with any welfare analysis, these conclusions require a number of strong assumptions, which we highlight and discuss below.

Our work contributes to literatures in economics, health policy, and medicine studying various aspects of the long-term care market including insurance (e.g., Brown and Finkelstein, 2011, Mommaerts, 2020), hospital incentives (e.g., Einav et al., 2018), the supply of caregivers (e.g., Spetz et al., 2015, Spetz et al., 2019), and trade-offs associated with informal care provision (e.g., Van Houtven et al., 2013, Skira, 2015, Barczyk and Kredler, 2018, Mommaerts and Truskinovsky, 2020). Although skills training has been shown to improve knowledge, skills, and patient outcomes for physicians and nurses (Davis et al., 1995, Marinopoulos et al., 2007, Khatony et al., 2009), the strand of this literature focusing on caregiver training is quite limited. Cooper et al. (2017) conduct a meta-analysis of papers that evaluate the efficacy of training programs for paid home caregivers (broadly defined) and find just six quantitative papers, only one of which they deem high quality.5 More recently, Van Houtven et al. (2019) conducted an RCT that provided training to informal, family caregivers of military veterans and found that training increased experienced quality of care. The authors also examined veterans’ subsequent use of health care along a number of dimensions, but lacked the statistical power to reject the null for fairly large point estimates. For example, veterans whose family caregivers received training were approximately 10 percent less likely to have an emergency department visit or a hospitalization, but these effects were not statistically different from zero. Our RCT complements this past work by providing estimates of the demand for training in the context of a newly available, relatively short online training module.

We also contribute to a strand of literature on the supply of and demand for skills training (see Leuven and Oosterbeek, 1997 for a review). Demand estimation in our paper is most similar to Hidalgo et al. (2014), which randomized vouchers for training programs to a set of low-skilled individuals in the Netherlands. The authors found that lowering the price of training by approximately €21 per hour ($30 in real 2021 U.S. dollars) increased training completion by 44 percent.6 We find broadly similar results, a $50 price reduction increased training completion by 67 percent. Despite the large differences in contexts and the content of the training, this similarity suggests that our results may be more broadly informative than otherwise expected.

The remainder of this paper is organized as follows: Section 2 provides background information on caregiving in the US. Section 3 describes our partner organization and the experiment. Section 4 discusses our data and empirical strategy. We present our findings in Section 5 and discuss the implied demand curve in Section 6. Our welfare analysis can be found in Section 7. Finally, we discuss our findings and summarize the limitations of our analysis in Section 8.