Over the past several decades, we have witnessed a long historical debate about work disincentives in public housing programs, and the federal government has begun to design policies, assisting recipients to find employment and become self-sufficient. The Moving to Work (MTW) is a HUD demonstration legislated by Sect. 204 of the Omnibus Consolidated Recessions and Appropriations Act (OCRAA) of 1996. It allows greater flexibilities and wavers in the use of their administrative funds to participating public housing agencies (PHAs) to avoid excessive regulatory burdens, which may impede the effectiveness of programs in provisions of the 1937 Housing Act (Abravanel et al., 2004; Levy et al., 2020). The total funding provided to MTW agencies has been gradually increased over the past 2 decades, and, in 2017, 39 MTW agencies received nearly $4.4 billion, which accounted for 17% of all HUD funding to PHAs. Additionally, MTW agencies served approximately 12% of all PHA-assisted households in 2016, which reflected an increase from 8% in 2008 (Galvez, Gourevitch, et al., 2020). The demonstration has three main objectives: (i) reduce costs and increase cost-effectiveness in federal expenditures, (ii) promote employment and economic self-sufficiency, and (iii) increase housing choices for low-income families. To achieve the aforementioned statutory objectives, participating PHAs have implemented unique locally tailored programs, including—but not limited to—rent policy, neighborhood choice and mobility, housing development, and work requirements (see, e.g., Galvez, Teles, et al., 2020; Levy et al., 2020; Riccio, 2020).

Particularly concerning work disincentive policies, nine out of 39 MTW agencies have developed work requirement policies as of 2015, although policy rules, targeted population, and supportive services varied across agencies, they have targeted work-able recipients to make the transition to a life of self-sufficiency without relying on public assistance (Levy et al., 2019). Rohe et al. (2016) evaluated the effect of work requirements across five public housing sites in the Charlotte Housing Authority (CHA) and found that public housing residents who were subject to work requirements showed a substantial increase in employment but did not increase average hours worked. Lee and McNamara (2018) also showed that work-able recipients (both public housing and housing choice voucher holders) who were subject to work requirements in the Housing Authority of Champaign County (HACC) increased average earnings by $2,283 between 1 year prior to and 1 year following work requirement enforcement, compared to similar recipients in a nearby housing authority. Similarly, Levy et al. (2019) found that 94% of the recipients who were subject to work requirements in the Chicago Housing Authority were in compliance with the requirements, and their annual income increased by 9.9% over 7 years. Additional MTW case studies and annual reports demonstrate the economic benefits of work requirements.Footnote 1 However, only a few or no studies have investigated the relationship between work requirement enforcement and mental health of recipients served in MTW agencies.

In this study, we examine the effect of work requirements on two measures of mental health (depression and hopefulness) for heads of households (HHs) that were subject to work requirements in the HACC. As of January 2013, the HACC imposed work requirements to all working-aged (18–54) and non-disabled recipients and supported their transition to work with case management services. Hence, to understand changes in mental health of the impacted HHs during the early phase of work requirements, we invited all work-requirement-eligible HHs to complete the survey in 2012 as a benchmark for evaluation. We then re-asked these HHs to complete a follow-up survey in 2014. We also surveyed HHs that met the eligibility for HACC’s work requirement policy but were assisted by a nearby conventional housing authority as a control group of this study.

We employ a difference-in-differences (DD) method to estimate the average treatment effects of work requirements on changes in measures of mental health that the recipients had experienced during the analysis period. The benefit of the methodology enables us to control for unobserved time-invariant heterogeneity in the effect estimation. However, the DD estimator requires the parallel trend assumption, indicating that, in the absence of program intervention, the difference between the “experiment” and “control” group is constant over time (Angrist & Pischke, 2008). However, our control recipients, served by another nearby housing authority, would have different socioeconomic characteristics as well as different economic opportunities and neighborhood attributes than those of the recipients in HACC. When the parallel trends are not presumed to be achieved, one widely used approach that can be applied is propensity score matching (PSM) to construct a statistical counterfactual that shares approximately similar likelihoods (propensity scores) of being in the treatment group. The usefulness of matching relies on the conditional independence assumption (CIA) because it supports DD’s parallel condition by making potential outcomes to be independent of the treatment. Our rich set of matching covariates, collected from the survey and administration data, can help to create more convincing statistical counterfactuals. Our main results show that, during the early phase of work requirements, the impacted (working-aged and non-disabled) HHs were predicted to increase depression scores and decrease hopefulness scores than changes in mental health measures of the matched control groups over the same period. We find that the estimates of DD with PSM tend to be larger than the DD estimates, indicating that the differences in observed characteristics between the treatment and control group induce some downward bias in work requirements effect estimates.


As of 2015, 9 MTW agencies (Atlanta, Champaign, Charlotte, Chicago, Delaware, Lawrence-Douglas, Lexington, Louisville, and San Bernardino) have adopted work requirements, mandating either some or all work-able recipients to work between 15 and 37.5 hours per week (Levey et al., 2019). Nearly all work requirement agencies have also offered an alternative way to fulfill recipients’ work requirement obligation through attending vocational training or formal education programs (general educational development or GED and postsecondary education) (Webb et al., 2016). Moreover, all these agencies have provided case management services, of which three agencies (Champaign, Lexington, and Lawrence-Douglas) have mandated participation in the services along with work requirements (Levey et al., 2018). Several work requirement agencies have also set time limits for assistance. Although the duration of time limits vary by the housing authority and targeted population, subsidized households must leave the housing authority once they reach the time limit. Non-compliant households may face the loss of housing subsidy and/or eviction.

As of January 2013, the HACC enforced work requirements for all working-aged (18–54) and non-disabled household members to work a certain number of hours weekly or attend job training or education institutions. The HHs were first required to develop a self-sufficiency plan that identified goals and objectives for each adult household member in the first recertification. Recertification was repeated annually, and households with no income were first invited to set up the plan because they were the program’s main target for a transition to work. The HACC required that at least one adult member of the MTW household must work a minimum of 20 hours weekly or enroll in an education institution by the second recertification. In the fourth recertification, one adult member must work a minimum of 20 hours weekly; and all other adult members must be employed a minimum of 20 hours weekly or attend education institutions. In the sixth recertification, all adult members must be working a minimum of 20 hours a week or attending education institutions. Furthermore, along with the work requirements, the HACC enforced participation in case management services. Each household was assigned to a case manager who assisted in advising and assessing career planning and employment barriers. Moreover, the case manager provided referrals to local social service partners such as public health district, childcare resource centers, education and job training services, financial budgeting, and transportation services. Recipients who comply with work requirements are exempt from case management participation. Additionally, the HACC announced that maintaining the non-compliance status results in a loss of housing assistance; however, there were hardship exemptions for households with extenuating circumstances such as illness or loss of employment through no fault of the household member.

Conceptual Model

We utilize the difference-in-differences (DD) method to estimate the effect of work requirements on mental health of work-requirement-eligible HHs. Specifically, we compare the change in two measures (depression and hopefulness) of mental health of the targeted (work-able) HHs in HACC to those of the HHs in the control housing authority over the same period. Let \({Y}_{ijt}\) be a mental health index value of recipient i in housing authority j and period t; \({D}_{jt}\) be a treatment dummy (1 for HACC and 0 for control housing authority); and \({T}_{t}\) is a time dummy (1 for follow-up year and 0 for baseline year). We then use the ordinary least squires (OLS) regression in which \({y}_{ijt}\) is regressed on the treatment dummy \({D}_{jt}\), time dummy \({T}_{t}\), and interaction of the treatment and time dummies \({D}_{jt \bullet }{T}_{t}\), written as:

$${Y}_{ijt}={\beta }_{0}+{ \beta }_{1} {D}_{j}+{ \beta }_{2} {T}_{t}+{\beta }_{3} \left({D}_{j}\bullet {T}_{t}\right) + {\varepsilon }_{ijt},$$

where \({\beta }_{0}\) is designed to capture unobserved time-invariant heterogeneity and \({\varepsilon }_{ijt}\) is an idiosyncratic error term. \({\beta }_{3}\) identifies the program effect, estimated by,

$$\begin{aligned}DD= & \left\{E\left[Y_{ijt}\vert D_j=1,T=1\right]-E\left[Y_{ijt}\vert D_j=1,T=0\right]\right\} \\ & -\left\{\mathrm E\left[{\mathrm Y}_{\mathrm{ijt}}\vert{\mathrm D}_{\mathrm j}=0,\mathrm T=1\right]-\mathrm E\left[{\mathrm Y}_{\mathrm{ijt}}\vert{\mathrm D}_{\mathrm j}=0,\mathrm T=0\right]\right\}=\beta_3\\\end{aligned}$$

The DD estimate will be biased if the parallel trend assumption is violated. This condition assumes that, in the absence of program intervention, the differences in individual, PHA, and regional characteristics between the treatment and control group is constant over time (Angrist & Pischke, 2008). However, the MTW demonstration was not designed as a random-assignment experiment (Lee & McNamara, 2018). Specifically, it was a PHA’s own decision to apply for the demonstration, and the US Congress may authorize specific PHAs to become part of the MTW demonstration, or it may allow the US Department of Housing and Urban Development (HUD) to select PHAs through its criteria. Cadik and Nogic (2010) stated that “the scoring and selection process was not straightforward” (p.13). They reported that several highest scored PHAs failed to join the MTW demonstration due to the limited program scope and lack of innovative activity proposals. On the other hand, some PHAs that did not belong to the highest scoring group were recommended to participate in the demonstration. Furthermore, some selected PHAs withdrew from the demonstration since the initially requested deregulation was not secured in negotiated agreements. These factors would raise the selection problem: some unobserved characteristics correlated with agency selection processes might also affect recipients’ mental health.

Since our control recipients were served by another nearby housing authority, and therefore they would have different socioeconomic characteristics  as well as face different economic opportunities and neighborhood attributes than those of the recipients in HACC. When the parallel trends are not presumed to be achieved, one widely used approach that can be applied is propensity score matching (PSM) to construct a statistical counterfactual that shares approximately similar likelihoods (propensity scores) of being in the treatment group (Caliendo & Kopeinig, 2008; Rosenbaum & Rubin, 1985). To reduce selection bias, we control for a set of individual and family characteristics. The individual characteristics include head’s age, gender, race, marital status, educational attainment, employment, criminal history, presence of driver’s license, computer skills, residential mobility, and English-speaking skills. The household characteristics include household size, having a child, living with an adult member, other sources of income from adult members other than the head, and recent residential mobility. The rich set of variables obtained from the conducted surveys and administrative data allows us to create a more convincing statistical sample of the control group by using PSM. More specifically on the estimation procedures, we first estimate the propensity score, which indicates the probability of being in the treatment group, to match treatment and control recipients given matching covariates in the baseline year. We then use a weighted least squares regression, in which control observations are weighted by pre-estimated propensity scores, to calculate the DD estimates.


Study Sample

We administered a family self-sufficiency survey to all working-age and able-bodied HHs in the treatment and control housing authorities, and recipients voluntarily decided to participate in the study. At baseline (year 2012), 1,224 HHs in the treatment housing authority and 761 HHs in the control housing authority met the eligibility for HACC’s work requirements, and about 94.2% and 82.1% of them were invited, respectively.Footnote 2 By the end of the baseline year, 148 HHs (12.1%) completed the survey in the treatment group, and 79 HHs (10.4%) completed the survey in the control group.

Selection of Control Housing Authority

We considerably chose the control housing authority, based on the following criteria: (i) a housing authority has not been a member of the MTW demonstration; (ii) a housing authority was located in the same state since different state rules and policies might influence unobserved household and neighborhood characteristics; (iii) a housing authority did not serve a large city or its suburban areas due to heterogeneity inherent in different regional characteristics; and (iv) a housing authority was in the close distance and shared the similar neighborhood characteristics and economic opportunities. According to the criteria, we chose a housing authority that served a small metropolitan statistical area (MSA) and its suburban areas, located southwest of Champaign County. Under the consent of the control housing authority, we surveyed the HHs that satisfied eligibility of work requirements (Lee & McNamara, 2018).

Appendix 1 presents descriptive statistics for the selected economic, demographic, and housing characteristics of each MSA with the treatment and control housing authorities. The results show that the total number of individuals and households in MSA where the HACC was located was about twice as large as the one with the control housing authority, while the population density was similar in both MSAs. In addition, the change in area's unemployment rates between 2012 and 2014 was quite close to each other. Moreover, both MSAs showed similar median annual income and household-unit ratio. The poverty rates at the household-level in MSA with HACC tended to be higher, but the family-level poverty rates and families below the severe poverty level were similar in both MSAs.

Measuring Mental Health

This study uses two widely used measures of mental health.Footnote 3 We use a multi-item scale known as the Center for Epidemiologic Studies Depression (CES-D) to measure the level of depressive symptomatology (Radloff, 1977). The CES-D scale measure is based on response to 20 items, including four components—depressive affect, positive affect, somatic/retarded activity, and interpersonal relations. Respondents indicate how often within the last week they experienced symptoms of depression on a 4-point scale. In general, the 20 items are summed to create an overall CES-D scale score ranging from 0 to 60, with a higher score indicating more depressive symptoms. Also, the entire CES-D scale is considered missing if more than four items are missing. However, this method could deliver wrong information if a respondent answers different numbers of items over time. Therefore, in order to have a precise measurement, we average these 20 items to construct the CES-D scale score.

We also use the Adult Trait Hope (ATH) scale, which is designed to measure the level of hopefulness based on a combination of believing there is a way to achieve personal goals (agency) and being motivated to take action to pursue those goals (pathways) (Snyder et al., 1991). The scale consists of 12 items with a 4-point scale. The two subscales—agency and pathways—use four items each. The agency subscale score is derived by summing items 2, 9, 10, and 12; and the pathway subscale score is derived by adding items 1, 4, 6, and 8. Moreover, four additional items are used as distracters (items 3, 5, 7, and 11). These two subscales are summed to create an overall hope score. Higher overall hope scores indicate that individuals are more hopeful and motivated to achieve their goals and more capable of designing means to achieve their goals. Similar to the CES-D scale, we average the 8 single-item scales (agency and pathway), allowing up to 1 missing item.


Table 1 presents descriptive statistics for mental health measures at baseline and follow-up years. The data show that the treatment group’s CES-D scale scores increased, while the control group’s CES-D scale scores decreased over the same time. We also observe that the treatment group’s ATH scale scores decreased, while the control group’s ATH scale scores increased over time.

Table 1 Descriptive Statistics for Mental Health Index

Table 2 reports the description of control covariates and balancing test results (two-tailed t-statistics) to test a statistical difference in observables between the treatment and control groups in the baseline year. If the control group is well established, we would expect that none of the balancing test coefficients differs from zero. However, the results show that the treatment group tended to have a higher proportion of non-white heads and a lower proportion of HHs with a criminal history and high school diploma (or GED) than those of the control group. These results suggest a need to adjust group differences in the baseline year to ensure that the groups we were comparing shared similar characteristics in the baseline year.

Table 2 Baseline Balancing Test

We use nearest neighbor (NN) matching with a caliper of 0.25 standard deviation of the estimated propensity scores (Rosenbaum & Rubin, 1985). Caliper matching is one form of imposing a common support condition, thereby enhancing matching quality by avoiding bad matches from the estimation (Caliendo & Kopeinig, 2008). Table 3 presents several indicators of matching quality. Specifically, after matching, the pseudo R2 shows that the proposed model only explains 1–2% variation of the treatment condition. The likelihood ratio (LR) test leads us to accept the null hypothesis, indicating that the joint significance of all matching covariates is not statistically significant. Additionally, the mean and median standardized bias decrease by half or more after matching. Figure 1 shows the density distribution of the estimated propensity scores for the treatment and control groups before and after matching. The more the two distributions are similar (overlap), the larger common support they have.

Table 3 The Effect of Work Requirements On Mental Health
Fig. 1
figure 1

Density distribution of propensity score

Table 3 presents regression results obtained from the DD and DD with PSM. The DD estimates show that the treatment group increased the CES-D scale scores by 0.14, three times greater than the control group’s changes (–0.047). The treatment group also decreased the ATH scale scores by 0.2 or 1.5 times less than its change (0.13) of the control group. We find that the estimates of DD with PSM tend to be larger than the DD estimates, indicating that the differences in some socioeconomic variables between the treatment and control group induce a downward bias in work requirements effect estimates, leading to a smaller impact size. Additionally, including controls in regression makes only a minimal change in the size of effect estimates.


Causes and Consequences of Mental Distress Among Recipients Who Were Subject to Work Requirements

Our study results show that, during the early phase of work requirements, impacted recipients (who were subject to work requirements) experienced increased depression scale scores and decreased hopefulness scale scores compared to changes in those of the matched control group. These results implicitly suggest that the adverse effects of work requirements on mental health, such as challenges and hardships to fulfill work requirements, might overwhelm its beneficial effects, such as having a job or increased earnings.

Specifically, according to Lee and McNamara (2018), work-requirement-eligible HHs’ probability of working increased substantially in HACC during this period. Our data also confirmed that an additional 13% of the study sample reported working in the follow-up survey, while the control group’s employment rates remained similar. Thus, considering previous literature, demonstrating a positive relationship between labor market outcomes and mental health, the HHs who complied with work requirements might achieve some degrees of satisfaction linked to an improvement of mental health (e.g., Clark & Oswald, 1994; Ettners, 1996; Ettner et al., 1997; McKee-Ryan et al., 2005; Murphy & Athanasou, 1999; Winkelmann & Winkelmann, 1995). As a group, these studies showed that the benefits of employment to mental health include a sense of financial security, enhanced social interaction and self-esteem, and other quality of life improvements.

In contrast, the recipients might also suffer from mental distress related to being in dead-end jobs, following coercive rules and activities, and fear of sanctions as a result of non-compliance with work requirements. First, considering that the recipients tended to have poor job skills and limited work experience, the jobs they could find paid low wages and often did not last. A number of studies showed that low wages and uncertainty about the length of the contract were associated with work dissatisfaction, which in turn worsens individuals’ mental distress and problematic parenting (Conger et al., 1992; Perry-Jenkins et al., 2000; Ross & Mirowsky, 1995; Yeung et al., 2002). Additionally, research on welfare reform showed that during the early phase of the implementation of welfare reform, the recipients with unstable and low-wage jobs were more likely to experience greater financial hardships and family disruption—some families were even worse off than they were on welfare—than those with stable and higher-wage jobs (Scott et al., 2004). Other studies also concluded that low-wage jobs—the recipients were employable—might not be sufficient to generate health improvement (Raver, 2003; Wilde et al., 2014).

Second, previous studies that explored women’s role overload and conflict argued that, although the direction of causation is unclear, the combination of work and child-rearing could put employed single mothers under greater mental stress than comparable married mothers. That is, potential positive effects of employment and fulfillment of work requirements might be counterbalanced by the strain that employment added to their child care responsibilities (Ali & Avison, 1997; Hsueh & Yoshikawa, 2007). Considering that nearly 9 in 10 sampled recipients in our study were single mothers with dependents, overall lower levels of mental health measures for recipients who were subject to work requirements might be attributed to the aforementioned sources of mental stress from juggling motherhood and employment. Furthermore, mandating individuals to enter the labor market and increase their number of hours worked if they were otherwise unwilling to do so would be highly stressful (Chase-Lansdale et al., 2011).

Third, beyond individual barriers such as the lack of education and work experience, the recipients might face structural barriers, including limited availability of child care providers and transportation, which would restrict their (regional) job search boundaries and chances to find and keep better quality jobs (Danziger et al., 2000; Edin & Lein, 1997; Raver, 2003; Teitler et al., 2004). A majority of the recipients was more likely to experience multiple employment barriers, and the number and severity of these barriers were strong indicators of failure to fulfilling work requirements and benefit sanctions. Contrary to the role of sanctions that encourage recipients to work and become self-sufficient, sanctions can be also viewed as a punishment for those who did not comply with the regulations. Hence, experiencing imposition of work requirements and sanctions, especially for those who were at risk of not meeting work requirements, could worsen mental health.

Policy Recommendations

A majority of policies aimed at enhancing economic independence for publicly assisted individuals has put an emphasis on addressing individual barriers (e.g., education and work experience) and structural barriers (e.g., access to jobs paying a living wage and the availability and costs of child care and transportation). However, a growing body of research also suggests that mental health problems appear to be a critical barrier to self-sufficiency, resulting in lower rates of employment and reduced number of hours worked weekly (Ettner et al., 1997; McKee-Ryan et al., 2005; Murphy & Athanasou, 1999).

The research on welfare reform shows that one in five families who left welfare between 1997 and 1999 returned to welfare by 1999, of which higher than 20% of families exiting welfare returned within the first year after exit. Moreover, members of the return-to-welfare group were much more likely to be unprepared for employment (39%) and suffer from physical/mental health problems (28%) than those who left welfare for work (Loprest, 2002). Additional evidence demonstrating mental health’s role on welfare-to-work is exclusively documented in Jayakody and Stauffer (2000) and Jayakody et al. (2000).

Our findings of decreased mental health measures for recipients who were subject to work requirements might provide some valuable insights about how and where to invest federal funds, gained from MTW flexibilities, to achieve recipients’ self-sufficiency. First, if work requirements impaired recipients’ mental health, procedures to achieve the self-sufficiency statutory objective would be to (i) increase the housing authority’s outlays on offering careful assessment and screening to identify those with mental health problems and other serious employment barriers who may be at risk of not fulfilling work requirements or being sanctioned; (ii) integrate case management services with mental health counseling or care services through a partnership with local community organizations; (iii) allow flexibilities in sanctions policies to prevent recipients at potential risk of eviction due to their mental problems from leaving the program without a stable financial settlement, and (iv) alleviate recipients’ mental problem factors by reinforcing provision of or referrals to community social services (e.g., child care services) and by supporting to search stable jobs that pay living wages.

Study Limitations and Future Research

One-fourth of the MTW agencies have applied the “stick” approaches in the form of work requirements and time limits, which mandate work-able recipients to find employment and restrict the length of time they may receive assistance (Webb et al., 2016). On the other hand, some agencies have chosen to offer “carrot” in the form of escrow saving account to facilitate work-able recipients’ transition to employment. For example, CHA deposits between $10 and $50 monthly, depending on the income band in which recipients belong to, into escrow saving accounts established for those with income between $12,500 and $35,000. Delaware Housing Authority deposits rent beyond a ceiling rent into a savings account based on the income level. King County and Minneapolis housing authorities established similar, but locally driven, escrow accounts (Webb et al., 2016).

However, despite efforts to develop and practice innovative policies, the comparison of the effectiveness between carrot and stick approaches has not been fully answered yet because many MTW agencies implemented multiple policies simultaneously, making it extremely difficult to disentangle the effect of one policy from another. Additionally, each agency sets its own policies in terms of definitions of work-able individuals and self-sufficiency, specifics of work requirements, the duration of time limits, and voluntary (or mandatory) nature of participation.Footnote 4 Therefore, to better understand the impact of such experimented policies, evaluation plans should be carefully designed to avoid confounding effects among policies that may obscure the effects being tested and lead to erroneous interpretation. Since both financial carrot and stick approaches are designed to incentivize targeted individuals to improve employment, but differently, the direction of their effects on labor market outcomes may be in the same direction, although the effect size may vary. However, the direction of their effects on mental health may not be in the same direction since carrot approaches do not have a punitive nature of work requirements, and financial incentives may even positively impact recipients’ mental health.

Additionally, conducting operational research will expand our current understanding of (i) why MTW agencies chose a certain set of self-sufficiency policies over the others, (ii) whether those selected policies were successful or not, (iii) what lessons and challenges they have earned, and (iv) how these lessons and challenges influenced the agencies to plan their future self-sufficiency strategies. Furthermore, the imposition of work requirement may invoke behavioral change for current and potentially assisted households by applying to other (non-MTW) housing authorities without work requirement to secure housing assistance regardless of work effort. This may change the composition of assisted households in housing authorities with work requirement that assisted households and households in their waitlists tend to have better employment skills and mental health to meet the work conditions than households previously served. Answers to these questions will provide a more complete empirical portrait of the challenges and success of MTW agencies’ heterogeneous self-sufficiency policies. Lastly, MTW agencies’ privilege as a housing policy lab is to experiment with various innovative approaches replicating and localizing important dimensions of existing financial incentive programs such as the Earned Income Tax Credit (EITC) and the Child Tax Credit.

This study has limitations. Our study is based on 12.1% and 10.4% of the entire treatment and control group populations, as well as recipients who were willing to participate in the study completed the survey. That is, some groups of people might be over-represented or under-represented in the sample; hence, findings based on this subset of the population would not be generalized to all HHs who were subject to work requirements. However, conversely, this means that about 10% of the sample HHs who were subject to work requirements might have a relatively higher risk of mental health illness during the early phase of work requirement enforcement. Nevertheless, our evidence can provide meaningful insights of the implementation of work requirements and possible consequences into other MTW and non-MTW housing authorities that serve a similar subset of the population as the study sample.