Search tips
Search criteria 


Logo of gerontLink to Publisher's site
Gerontologist. 2011 October; 51(5): 597–609.
Published online 2011 April 15. doi:  10.1093/geront/gnr025
PMCID: PMC3179664

Stayers, Leavers, and Switchers Among Certified Nursing Assistants in Nursing Homes: A Longitudinal Investigation of Turnover Intent, Staff Retention, and Turnover

Jules Rosen, MD,*,1,2 Emily M. Stiehl, BA,corresponding author2 Vikas Mittal, PhD,corresponding author3,4 and Carrie R. Leana, PhDcorresponding author2


Purpose: Studies of certified nursing assistant (CNA) turnover in nursing homes are typically cross-sectional and include full-time and part-time workers. We conducted a longitudinal study to evaluate the job factors and work attitudes associated with just full-time staying or leaving. For those who did not stay, we assessed reasons for leaving and satisfaction following job transition. Design and Methods: A random sample of CNAs identified through the Pennsylvania Department of Health's CNA registry, working≥30 hr weekly in a nursing facility was surveyed by telephone at baseline and 1 year later. Results: Of the 620 responding to both surveys, 532 (85.8%) remained (stayers), 52 (8.4%) switched to another facility (switchers), and 36 (5.8%) left the industry (leavers). At baseline, switchers reported higher turnover intentions and fewer benefits compared with stayers and left for new opportunities. Leavers had lower job satisfaction and emotional well-being and left for health reasons. Turnover intentions were predicted by low job satisfaction and low emotional well-being. Actual turnover was predicted only by turnover intentions and by the absence of health insurance. Pay was not a predictor of turnover intent or turnover. Implications: There are two distinct groups of CNAs contributing to turnover. Attitudinal factors, such as job satisfaction and emotional well-being, are mediated via turnover intentions to effect actual turnover. Even accounting for methodological differences, this turnover rate is lower than previous studies, which use alternative methods and include part-time workers. This study should help nursing home administrators better understand the work-related factors associated with staff turnover.

Keywords: Nursing homes, Workforce Issues, Organizational & Institutional issues, Long-term Care, Caregiving—Formal

The recruitment and retention of direct care workers (DCWs) in the U.S. long-term care industry have been the subject of numerous studies over several decades (Castle, Engberg, Anderson, & Men, 2007; Lagnado, 2008; Span, 2009). The Institute of Medicine Report, Retooling for an Aging America, identifies an unstable direct care workforce as a primary challenge that must be addressed to care for the growing number of aging baby boomers in the Unites States (Committee on the Future Health Care Workforce for Older Americans, I. o. M., 2008). Staff turnover not only increases the financial burden of caring for elders (Seavey, 2004), but it also interferes with quality of care (Castle, Engberg, & Men, 2007) and quality improvement initiatives (Rosen et al., 2005). Earlier studies that focused on certified nursing assistants (CNAs) in nursing homes have linked a number of constructs, including job satisfaction, with intent to leave the job (Castle, Degenholtz, & Rosen, 2006; Parsons, Simmons, Penn, & Furlough, 2003) and actual quitting (Castle, Engberg, Anderson, et al., 2007; Kiyak, Namazi, & Kahana, 1997) or have evaluated the impact of organizational factors on administrator-reported staff turnover (Banaszak-Holl & Hines, 1996; Brannon, Zinn, Mor, & Davis, 2002; Brennan & Moos, 1990). In this article, we examine the impact of employee work attitudes and job factors on intentions to leave the job and on actual turnover.

CNA turnover rates reported in the literature are high: A scholarly review by Castle (2006) reported annual CNA turnover rates between 23% and 346%. However, such findings must be interpreted with caution in light of the limitations of the research on CNA turnover. Virtually, all previous studies use cross-sectional data provided by nursing home administrators who employ inconsistent and variable methods to determine turnover. Furthermore, these studies fail to differentiate between part-time and full-time employees, even though prior research suggests that part-time employees are likely to be very different than full-time employees in terms of job satisfaction and job behaviors (Eberhardt & Shani, 1984; Feldman, 1990; Werbel, 1985). Finally, nearly all report turnover rates rather than retention rates, which are not reciprocal (Castle, 2006).

Public policy discussions of high turnover rates, the increasing costs of care due to turnover, and discrepancies reported in the academic literature regarding the stability of this workforce warrant a fresh approach to this problem. Rather than rely on cross-sectional data or self-reports by facility managers, we conducted a longitudinal study of CNAs using telephone surveys completed at baseline and then 12 months later to determine the actual behavior among CNAs in terms of staying with or leaving their jobs. Unlike earlier studies, participants in this study were randomly selected from a statewide database of CNAs rather than by their association with participating facilities and were followed longitudinally. Also differing from previous studies, our respondents were working at least 30 hr per week in a single facility, limiting the number of part-time workers participating.

This study addresses the following regarding CNAs employed in nursing homes: (a) the annual rate of turnover and of retention, (b) factors at baseline that differ between workers who stayed at their original job (stayers), switched to another facility (switchers), or left the eldercare workforce entirely (leavers) 12 months later, (c) cross-sectional differences between these groups at follow-up, (d) longitudinal differences at baseline and follow-up within each group, and (e) a retrospective assessment of reasons for leaving. We expect that attitudinal and job-related factors at baseline will differ among stayers, switchers, and leavers. As shown in Figure 1, we hypothesize that differences in these job-related attitudes and factors will manifest as differences in participants’ intentions to stay at their job or turnover. In the subsequent time period, turnover intentions can result in the action of switching or leaving (i.e., actual turnover). For the purpose of this study, we are primarily interested in job-related attitude factors (i.e., job satisfaction, emotional distress, supervisor respect, and coworker respect) and job factors (i.e., paid sick and vacation days, health insurance, and promotion opportunities), both of which have the potential of being modified through administrative actions (Price & Mueller, 1981; Werbel, 1985).

Figure 1.
A conceptual model of turnover among certified nursing aides.


We conducted a longitudinal study of DCWs employed in long-term care settings in Pennsylvania (PA). This study was approved by the Institutional Review Board of the University of Pittsburgh, and respondents were compensated for their time following the completion of each survey. To gain insight on issues that affect the decision to stay with or to leave a DCW job, we initially conducted focus groups and telephone interviews with 55 DCWs, classified as “frequent leavers” or “persistent stayers” using self-report data about whether they had changed jobs within the past three years. The results of these focus groups have been previously described and were used to develop the survey used in the current study (Mittal, Rosen, & Leana, 2009). More specifically, we identified factors such as supervisor and coworker respect, emotional distress, and specific job-related benefits as factors that differed among frequent leavers and persistent stayers. Building on those findings and previous research, we developed a survey.

We followed the Dillman method in developing our survey design and implementation methodology (Dillman, 1978; Dillman, Smyth, & Christian, 2009). A key aspect of this survey design is the explicit recognition that increasing the survey length is negatively correlated with the quality of data due to respondent fatigue and boredom. Therefore, rather than being exhaustive, we focused on factors that were based on prior research, our focus group findings, and also deemed managerially actionable. Trained professional interviewers administered two telephone interviews, each approximately 45-min long at Time 1 (T1; March, 2008) and one year later at Time 2 (T2).

The T1 survey asked respondents to report on (a) demographic information; (b) on-the-job factors, including pay, paid sick and vacation days, health insurance, and opportunities for promotion; (c) attitudinal factors, including relationships with coworkers and supervisors, job satisfaction, and emotional distress; and finally (d) turnover intentions. The T2 survey, administered 12 months later, assessed many of these same factors but also inquired about whether the respondent had changed jobs and, if so, how important various reasons were in their decision to leave their T1 job. Job satisfaction was a single item measured with a 4-point Likert scale (1 = not at all satisfied, 4 = very satisfied). Emotional distress was measured by three items with a 5-point Likert scale. Supervisor respect and coworker respect were both measured on a 4-point Likert scale (1 = not at all, 4 = a lot). Turnover intentions were measured with three items on a 4-point Likert scale (1 = not at all likely 4 = very likely). The specific items comprising the key constructs and coefficient alpha are shown in Table 1. All scales have satisfactory (.70 or higher) reliability coefficients (Cronbach's alpha), a measure of internal consistency (Nunnally, 1978). Job benefits (e.g., paid sick and vacation days, health insurance, and opportunities for promotion) were measured as dichotomous variables (1 = has access to it, 0 = does not). Respondents reported whether their organization offered each benefit to them.

Table 1.
Items and Reliability for Each Construct

To be eligible for the study, the respondent had to be employed as a CNA in a single nursing facility in Pennsylvania for at least 30 hr per week and had to be sufficiently proficient in English to complete the survey. The CNA registry in Pennsylvania is maintained by the PA Department of Health, and it is mandated that all CNAs working in nursing facilities update the registry with any new information. Of the 2,366 CNAs randomly selected from the CNA registry for initial screening between February and August, 2008, 1,360 (57.5%) were eligible to participate. Ineligible respondents were not currently working or worked less than 30 hr per week (38%), did not reside in PA (3%), or did not speak English (0.1%). From the set of eligible CNA respondents, 121 (8.9%) refused and 425 (31.3%) were working in sites other than nursing homes. Thus, 814 (59.8%) completed the T1 survey.

All 814 respondents from T1 were recontacted by telephone after 12 months to participate in the T2 survey. We obtained a total of 620 responses at T2 (76.1% response rate). Each respondent received two follow-up calls before they could be classified as “missing.” Respondents were assigned to three primary categories based on their reported employment at T2: (a) “stayers” work in the same job for the same organization, (b) “switchers” continue to work at least 30 hr per week as a CNA but for a different organization than at T1, and (c) “leavers” now work in a non-DCW job or have left the workforce entirely. Failure to complete the T2 survey was due to the inability to coordinate the time for the interview due to (a) logistical issues such as travel out of the area, scheduling, or illness (69.4%), (b) refusal (14.6%), and (c) inability to locate the respondent (15.9%).

We conducted three sets of analyses to understand why CNAs stay or do not stay in their jobs. First, we conducted bivariate analyses to cross-sectionally compare constructs between the three groups. We used t test for means and z-scores for proportions as needed. Second, we conducted bivariate analysis to longitudinally compare any shifts in means or proportions within each subgroup (e.g., differences among stayers from T1 to T2). The within-group means were tested using same-sample t tests, and the differences for proportions from T1 to T2 were tested using McNemar's test (McNemar, 1947).

Third, we conducted a set of multivariate analyses. We first examined whether various work attitudes and job factors predict a person's intentions to leave their job at T1. As reported in Table 4, we estimated a series of hierarchical regression models, sequentially adding blocks of variables: demographics, job factors, and attitudes measured at T1 to predict turnover intentions. As described earlier, we relied on our focus group findings and prior literature to include factors and attitudes that occur at work and that may be under a manager's control (Price & Mueller, 1981). We include the demographic variables as controls.

Table 4.
Predictors of Turnover Intentions at T1

Next, we tested whether these factors affect actual turnover behaviors. We assessed whether CNA's turnover behavior was associated with turnover intentions and other blocks of variables: demographics, job factors, and attitudes. Specifically, in addition to turnover intentions, we included age, race, gender, tenure, paid sick days, paid vacation days, health insurance, opportunities for promotion, hourly pay, and work attitudes (e.g., job satisfaction, emotional distress and supervisor and coworker respect) in the model. These models are described in detail later.

To test whether turnover intentions mediate the relationships between the significant variables in Table 4 (e.g., job satisfaction and emotional distress) and actual turnover, we employed a nonparametric bootstrapping procedure (Preacher & Hayes, 2004, 2008). This procedure draws 5,000 new random samples from our initial sample, with replacement, which it uses to develop 5,000 estimates of the mediating effect, from the independent variable to actual turnover through turnover intentions. From these estimates, the procedure constructs a 95% confidence interval (CI). Mediation exists when the CI does not include zero (i.e., we can reject the null hypothesis that the effect of mediation is zero at p < .05). Bootstrapping does not require the data to be normally distributed nor does it require a significant direct relationship to exist between the dependent and independent variables in order to test for mediation (Hayes & Matthes, 2009).


Of the 620 nursing facility workers who completed both surveys, 532 (85.8%) reported they had not changed jobs over the past year. Of the remaining 88 respondents (14.2%) who were no longer at their T1 job, 52 (59%) were switchers and 36 (41%) were leavers. Of the 36 leavers, 25 had left the workforce (i.e., were no longer employed), leaving only 11 leavers to answer questions about work at T2.

The descriptive results are shown in Table 2. This table contains both cross-sectional comparisons among the three groups and longitudinal comparisons within each group (e.g., stayers at T1 vs. stayers at T2). Columns A, B, and C provide cross-sectional comparisons of the three groups at T1. Columns E, F, and G compare the three groups at T2. Column D provides the T1 data from “missing” CNAs, who completed the initial survey but did not complete the T2 survey. Significant cross-sectional differences (p < .05) between the three groups are indicated with superscripts (a, b, c, and d for T1 and e, f, and g for T2). For instance, the mean stayer age is significantly different from the average ages of switchers, leavers, or missing data at T1. This is denoted by superscripts b, c, and d. Next, within each group, we made longitudinal within-group comparisons across T1 and T2. Within each group, significant longitudinal differences are superscripted with a “t” in columns E, F, or G. For example, among stayers, hourly pay changed significantly (p < .01) from T1 to T2, increasing from $13.03 at T1 to $13.56 at T2. This is denoted by the superscript t** in column E.

Table 2.
Demographics, Job Factors, and Key Attitudinal Constructs Between Groups of Certified Nursing Assistants at T1 and T2

The upper panel of Table 2 provides demographic information about the respondents. Stayers are older than switchers and leavers with mean ages of 47.9, 41.7, and 42.2 years, respectively, and stayers report longer job tenure (11.28 years) than switchers (6.83 years) and leavers (7.83 years). Approximately 95% of each group is female, and 76% was White, but there are no discernable patterns in terms of gender and race among the three groups.

Comparing Groups at T1

In Table 2 (columns A, B, and C), stayers report better access to health insurance and paid time off than do switchers at T1. This pattern persists into T2. In contrast, the results of pay or paid sick or vacation days for leavers are not significantly different from those of stayers or switchers and tend to lie between the two. At T1, promotion opportunities were similar for all three groups.

In terms of work attitudes, Table 2 shows both raw means and covariate-adjusted means for each subgroup, the latter controls for covariates including age, tenure, gender, and race. The covariate-adjusted means are included to compare the three groups while controlling for variations in sample composition. Stayers and switchers are mostly similar, except with regard to their turnover intentions or “intent to leave,” where stayers were significantly less likely to report an intention to leave their current jobs (1.41 vs. 1.79, p < .05). Leavers, on the other hand, reported higher turnover intentions, greater emotional distress, less job satisfaction, and lower supervisor respect than stayers and differed from switchers in terms of emotional distress and job satisfaction.

Comparing Groups at T2

At T2, a pattern of job factor differences between stayers and switchers appears (Table 2, columns E, F, and G). Switchers are more likely to report greater promotion opportunities than stayers at T2 (37.5% vs. 23.0%), even though they reported having numerically fewer promotion opportunities at T1 (29.4% vs. 38.3%). Although there are no differences between groups in terms of pay at T1, switchers receive significantly lower wages than stayers at T2 ($12.25 and $13.56, respectively). Switchers also experienced lower emotional distress than stayers at T2. Leavers reported lower coworker respect than stayers or switchers at T2 (2.91 vs. 3.47 vs. 3.51). Interestingly, at T2, stayers still had lower turnover intentions than switchers or leavers.

Longitudinal Comparison of Job Factors and Attitudes Within Groups Over Time (T1 to T2)

As shown in Table 2, stayers report significantly fewer promotion opportunities but higher wages at T2 than at T1. Switchers and leavers show numerical reductions in wages, paid time off, and health benefits at T2, although these differences did not reach significance. Emotional distress decreased from T1 to T2 for each group, with the greatest decrease occurring among switchers. In addition, switchers reported a trend toward greater job satisfaction and access to fewer paid sick days. Leavers reported increased job satisfaction and greater supervisor respect at T2.

Comparing “Missing” CNAs Who Only Completed T1

It could be argued that CNAs completing only T1 surveys and not T2 (“missing”) may be more likely to leave their job. To assess this possibility, we compared the CNAs missing at T2 with other subgroups. As seen in column D of Table 2, the “missing” group is different from the other groups, although there is an inconsistent pattern of differences compared with each of the other groups on demographic and job factors. For example, in terms of demographics, missing CNAs are most similar to leavers and are significantly younger than stayers (43.73 vs. 47.94 years, p < .05), with lower tenure (8.31 vs. 11.28 years, p < .05) and less hourly pay ($12.57 vs. $13.03, p < .05). However, missing CNAs are most like stayers and switchers in terms of race, with a significantly smaller proportion of Whites among the missing group than among the leavers (71.9% vs. 85.7%, p < .05).

As for job factors, missing CNAs are most similar to stayers and are more likely to have health insurance than switchers are (78.5% vs. 57.7%, p < .05). Missing CNAs also report more promotion opportunities than leavers (41.3% vs. 22.2%, p < .05).

In terms of work attitudes, when adjusting for demographic factors, the missing CNAs have significantly lower turnover intentions than switchers or leavers (1.45 vs. 1.69 or 1.93, p < .05). Missing CNAs also have significantly greater levels of job satisfaction (3.31 vs. 2.85, p < .05) and supervisor respect (3.15 vs. 2.83, p < .05) than do leavers. Thus, in terms of attitudes, the missing CNAs seems to lie somewhere between stayers and leavers and may be substantively closer to switchers in many ways except in their turnover intentions and access to health care.

Reasons for Leaving T1 Job

At T2, respondents who had left their T1 DCW job (i.e., switchers and leavers) were asked to rate the importance of several reasons (1 = not at all important to 4 = very important) contributing to their decision to leave. In Table 3, we compare switchers and leavers who rated each factor's importance as either low importance (1 = not at all important, 2 = a little important) or high importance (3 = somewhat important, 4 = very important). We find that switchers and leavers differ in terms of the high importance they attribute to physical health problems (21.1% vs. 65.6%, p < .05) and the pursuit of other opportunities (87.2% vs. 63.3%, p < .05). These results suggest that switchers are more likely to leave their T1 jobs in order to pursue other opportunities than leavers. In contrast, leavers are more likely to report physical health problems as their primary reason for leaving. Pay (44.7% vs. 46.7%), demanding work (51.4% vs. 43.3%), and problems with supervisors (35.1% vs. 26.7%) were not significantly different between the two groups in terms of high importance. These factors are also less important than pursuit of better opportunities to both switchers and leavers when explaining why they left their T1 jobs.

Table 3.
Reasons for Leaving the Job Held During T1 (Reasons Provided at T2)

Multivariate Analysis (Predicting Turnover Intent)

In the full regression model, Table 4, Model 3, turnover intentions are associated positively with emotional distress (p < .01) and negatively with job satisfaction (p < .0001). They are also associated, to a lesser degree, with paid sick days (p < .10) and paid vacation days (p < .10). Turnover intentions diminish as the age of the workers in our sample increases (p < .05) and are higher for workers with lower tenure (p < .01) and for non-White workers (p < .0001). Notably, pay is not a predictor of turnover intentions. As a robustness check, we found that inclusion of the “missing” CNAs (either as stayers or as nonstayers) did not alter the results.

Multivariate Analysis (Predicting Turnover)

In Table 5, we report a longitudinal analysis to determine factors that predict turnover. Specifically, measures obtained at T1 were used to predict actual turnover at T2. In this model, the dependent measure is dichotomous, with values 1 (stayers) and 0 (nonstayers or switchers and leavers). The model predicts the likelihood of not staying, that is, actual turnover. Table 5 contains logistic regression coefficients, their p values, odds ratios, and 95% CIs for the odds ratios for each of the constructs. The estimated coefficient can be used to calculate the probability of actual turnover for given levels of turnover intentions. Based on Table 2 and the estimate from the full model in Table 5, this probability of turning over (i.e., switching or leaving) is 65% for a person who responded “1” on the turnover intent scale (i.e., low turnover intentions), 77% for a rating of “2” on the turnover intent scale, 86% for a rating of “3” on the turnover intent scale, and 92% for a rating of “4” on the turnover intent scale. The probability of actual turnover is strongly associated with an increase in turnover intent (p = .001). Finally, in Model 4, we find that turnover intentions have a significant positive relationship with actual turnover after controlling for demographics, job factors, and job attitudes (Table 5). Receiving health insurance is also significantly associated with turnover (p < .05), though it was not associated with turnover intentions.

Table 5.
Logistic Regression Predicting Actual Turnover Between T1 and T2, Comparing Stayers and Nonstayers

As a robustness check, we conducted a multinomial logit analysis (not included) using three levels to represent stayers, switchers, and leavers. The analysis produced results similar to those in Table 5 such that turnover intentions had a significant effect on actual turnover (chi-square = 12.23, p = .0022), most notably through its effect on switchers versus stayers. Beyond this result, the multinomial logit provided no additional statistically significant results, likely due to the small size of the switcher and leaver samples. Because our goal is to determine which factors influence whether employees stay or do not stay in their jobs and the two samples are so small, we decided to combine switchers and leavers into one group for the analysis in Table 5. We ran two additional logistic regression models (not included) that combine the “missing” CNAs with (a) stayers and (b) nonstayers and found that the results remained largely unchanged. Again, in both analyses, turnover intentions remained significant (b = 0.66, p < .0001 vs. b = 0.287, p < .05), although the effect was dampened in the second analysis where the missing data were added to the nonstayers. The significance of health insurance (b = 0.284, p < .10 vs. b = 0.31, p < .05) and promotion opportunities (b = 0.246, p < .10 vs. b = 0.279, p < .10) increased slightly when the missing data were added to the stayers, whereas that of age (b = −0.238, p < .10 vs. b = −0.229, p < .01) and paid vacation days (b = 0.246, p < .10 vs. b = 0.416, p < .05) increased when the missing data were added to the nonstayers. These results are consistent with the differences in Table 2 between the missing respondents and the stayers, switchers, and leavers.

Testing for Mediation

In Tables 4 and and5,5, there are several variables that predict turnover intentions, which subsequently predicts actual turnover. Using a nonparametric bootstrapping procedure, we find that turnover intentions mediate the relationship between job satisfaction and actual turnover (95% CI: −0.082 to −0.018) and emotional distress and actual turnover (95% CI: 0.016–0.056), controlling for age, tenure, race, and gender. To our earlier point, the effects of health insurance are not mediated by turnover intentions (95% CI: −0.43 to 0.003).


Although there have been multiple publications regarding nursing home turnover, this study utilizes a different methodological approach to provide new insights into this vexing problem. Specifically, this is the first study to assess nursing home CNAs longitudinally to determine the rates of retention (stayers) and turnover (switchers/leavers) to assess reasons for leaving and to evaluate the outcomes for workers 12 months later who either remained on their job or transitioned. The participating CNAs were randomly identified from a registry rather than based on their association with specific nursing homes, and they all worked 30 hr or more per week in a single facility. This is also the first study to differentiate two groups of CNAs who transitioned from their baseline jobs to be classified as either “switchers” or “leavers.”

Switchers are similar to stayers at baseline in terms of attitudinal factors, including job satisfaction and emotional distress. They differ from stayers primarily in terms of job factors, specifically sick leave, paid vacations, and health insurance. The most frequently cited reason for leaving was to pursue other opportunities. After switching to another facility, they experienced greater emotional well-being and more opportunities for promotion compared with stayers. It appears that the absence of benefits may not be a driving force in terms of switching as the switchers at T2 are no more likely to have benefits than they were at baseline. However, having benefits may promote retention for workers as shown by the multivariate analysis. Clearly, retention and turnover and factors affecting them are not simply reciprocals of each other (Mittal et al., 2009). Dissatisfaction with pay is not a significant predictor of turnover intentions or actual turnover nor is its importance as a self-reported reason for switching or leaving very high. Of note, switching jobs actually resulted in a numerically lower wage at T2.

Leavers were similar to stayers in terms of job factors, including paid leave and health insurance but differed from them in terms of key attitudinal factors, reporting greater emotional distress, lower job satisfaction, and less supervisor respect at baseline. Notably, despite the small sample of leavers (N = 36), we find some statistically significant differences between leavers and switchers, suggesting that physical health and the pursuit of other opportunities may be important in a person's decision to simply change jobs or to leave the industry entirely. Future research is needed to further compare switchers and leavers using larger sample sizes.

From a longitudinal perspective, workers transitioning from their baseline job into a new job appear to achieve positive outcomes with lower emotional stress and greater job satisfaction after switching jobs, whereas job satisfaction was unchanged for stayers. Thus, although turnover is undesirable from the nursing home industry's perspective, it appears to benefit the individual workers. In fact, switchers changing jobs to create better opportunities or more favorable schedules appear to gain some measure of success as a result of the transition. Leavers, on the other hand, appear to be less inclined to this work or may be physically less able to continue DCW work and, by leaving the industry, also find greater satisfaction.

Our results also shed light on the underlying process through which job factors and job attitudes may lead to turnover. Jointly, the final models in Tables 4 and and55 paint a relatively complex picture of actual turnover. They suggest that that age, tenure (years in job), paid sick days, and paid vacation days affect turnover intentions but do not directly affect actual turnover. Similarly, job satisfaction and emotional distress affect turnover intentions but do not directly affect actual turnover. Non-White workers also have higher turnover intentions, a finding that requires further exploration. The effect of these variables on actual turnover is indirect and fully mediated by turnover intentions. Thus, rather than directly causing a person to leave their job, these variables affect behavior through intentions. In contrast, we find that health insurance does not affect turnover intentions, but it directly affects actual turnover. Although it is possible that the absence of health insurance “permits” CNAs to switch jobs or leave the workforce more easily, it is not clear that the presence of health insurance would prevent turnover as these workers may have been insured through a spouse or partner. This finding will require further exploration.

An unexpected finding is that hourly pay does not affect turnover intentions or actual turnover. Though counterintuitive, this finding makes sense from a transaction-cost perspective (Williamson, 1981). According to this theory, instrumental benefits (e.g., pay and rewards) may be offset by noninstrumental and indirect costs associated with the status quo. Our data show that the range of hourly pay is quite restricted with 95% earning less than $16.05 per hour, whereas the range for the remaining 5% is $16.09–$31.00. This suggests that switching from one job to another may lead to only modest, if any, increases in actual pay and may actually result in a slight loss in pay due to loss of tenure. Moreover, there may be other benefits such as supervisor and coworker respect, promotion opportunities, and paid vacation days that may reduce turnover intentions despite the possibility of higher pay in alternative jobs. Although CNAs do not cite pay as a primary reason for leaving, it is unclear whether a substantial increase in pay (e.g., several dollars per hour) would affect turnover. Clearly, more research is needed to fully ascertain the trade-offs between higher pay and other factors affecting turnover.

The rate of turnover of 14.2% (±2.8%) in our study is considerably lower than previously reported in the literature as reviewed by Castle (2006). This finding may underrepresent the actual turnover because almost 25% of the CNAs did not participate in the T2 survey. Although the intent to leave for the “missing” CNAs was similar to stayers, unforeseen events, such as illness, relocation, and other reasons, might increase the turnover rate for this group. Note, however, the inclusion of this group in the analysis did not change the association of predictors of turnover with intentions or actual turnover. Still, 14% should be considered the lower bound, with the upper bound estimate of turnover at 35% ([88 leavers + 194 non-responders]/814; ±3.8%). This range is still considerably lower than that of annual turnover rates in more than 300 facilities across five states (81%–119%) reported by Castle.

There are several factors that may contribute to the relatively lower rate of turnover in the current report. Of the 25 reports of DCW turnover in nursing homes reviewed by Castle (2006), only 11 specified the method used to calculate turnover, with considerable methodological variability between studies. Differing from most previous reports, in our study, the rates of turnover and retention are reciprocals. In a recent study of more than 1,000 facility administrators’ self-reports, the number of CNAs retained for over 12 months was 62.7% (meaning 37.3% of positions were vacated), whereas the annualized turnover rate was 75.2% (Ong, Ricklkes, Matthias, & Benjamin, 2002). This apparent discrepancy is due to rapid turnover of new recruits requiring a single position to be filled by multiple CNAs during a 12-month period. When reporting retention, nursing home CNAs appear to be a considerably more stable workforce than when describing turnover. Even so, the results of our study still reflect a substantially greater retention rate that needs to be explained. Most likely, the restriction of casual and most part-time workers from our sample accounts for this discrepancy.

Virtually, all previous reports of nursing home staff turnover describe full- and part-time CNAs as “full-time equivalents.” In other words, they do not separate full-time and part-time employees despite many studies showing they are different (Eberhardt & Shani, 1984; Feldman, 1990; Werbel, 1985). In other industries, such as education (Stuit & Smith, 2010) and sales (Werbel, 1985), higher turnover among part-time employees as compared with full-time employees has been documented. In a study of turnover addressing full- and part-time staff separately in more than 350 facilities, administrators reported the overall turnover rate as 67% with the rate of turnover for part-time CNAs at 103% (Castle, 2006). Another study of CNA self-reports confirms a significantly higher rate of leaving for part-time workers (Castle, Engberg, Anderson, et al., 2007). In our presurvey screening, 30.8% of the CNAs in PA working in nursing home were ineligible for the current study because they were working less than 30 hr per week. Similarly, a study of CNAs in California reports that 1/3 of the workers are part-time or temporary (Ong et al., 2002). According to the Current Population Survey of March 2010, 61.3% of the CNAs in nursing facilities work year round and full- time in the United States. Collectively, these data suggest that the nursing home workforce comprised a substantial proportion of part-time workers who have lower retention rates than the CNAs participating in this study. More importantly, the different turnover rate—that is, lower turnover in our sample—may be due to the inclusion of workers working 30 or more hours. Theoretically, this suggests the need for studying part-time and full-time CNA workers as distinct groups rather than aggregating them as previous studies tend to do.

This may represent an opportunity for nursing home administrators to consider changes that could alter CNA retention rates. Addressing low job satisfaction and providing health insurance are likely to reduce the intent to leave and actual leaving. If part-time employees contribute disproportionately to the rates of turnover among nursing home staffs, modifying hiring practices to employ fewer part-time and more full-time workers could add stability to this industry. Additional studies with larger samples are needed to better understand the rates and the reasons for part-time CNA employment and the contribution of this segment of the workforce to turnover and retention rates.

Many of the limitations of this study represent choices flowing from resource restrictions, though they also represent critical research opportunities. These include the restriction on including part-time CNAs in the study (although deliberate), limiting recruitment only to CNAs working in Pennsylvania, not following the workers for more than 12 months, and not conducting additional surveys at 6-month intervals. Similarly, several factors were not included in the survey to keep the survey duration manageable (Dillman, 1978). Factors such as “calling,” household income, urban versus rural setting, and many others should be included in future research. We focused primarily on factors that could be addressed by a nursing home administration, such as job satisfaction, health insurance, and opportunities for promotion. Furthermore, our focus was not on examining facility-based differences. Thus, we did not collect any facility-level data. To reliably obtain such data, a different study design where nursing home information is obtained through secondary sources and merged with CNA responses is needed. Though resource intensive, such a design could provide additional insights about the interplay of nursing home characteristics and job factors. It should also be noted, the last six months of data collection coincided with a major downturn in the U.S. economy, and regional unemployment rates have been shown to have a significant impact on CNA job stability (Donoghue, 2010). Finally, the lower than expected rate of turnover limits the understanding of all possible reasons for leaving, especially among the “leavers.” We hope that despite these limitations, our results will provide grounds for future research in this area.


Funding was provided by the Jewish Healthcare Foundation of Pittsburgh, the Pennsylvania Department of Labor and Industry, the Heinz Endowments, the University of Pittsburgh Research Council, and the National Institute of Nursing Research (R01 NR009573).


  • Banaszak-Holl J, Hines MA. Factors associated with nursing home staff turnover. The Gerontologist. 1996;36:512–517. [PubMed]
  • Brannon D, Zinn JS, Mor V, Davis J. An exploration of job, organizational, and environmental factors associated with high and low nursing assistant turnover. The Gerontologist. 2002;42:159–168. [PubMed]
  • Brennan PL, Moos RH. Physical design, social climate, and staff turnover in skilled nursing facilities. Journal of Long Term Care Administration. 1990;18:22–27. [PubMed]
  • Castle NG. Measuring staff turnover in nursing homes. The Gerontologist. 2006;46:210–219. [PubMed]
  • Castle N, Degenholtz H, Rosen J. Determinants of staff job satisfaction of caregivers in two nursing homes in Pennsylvania. BMC Health Services Research. 2006;6:60. [PMC free article] [PubMed]
  • Castle NG, Engberg J, Anderson R, Men A. Job satisfaction of nurse aides in nursing homes: Intent to leave and turnover. The Gerontologist. 2007;47:193–204. [PubMed]
  • Castle NG, Engberg J, Men A. Nursing home staff turnover: Impact on nursing home compare quality measures. The Gerontologist. 2007;47:650–661. [PubMed]
  • Committee on the Future Health Care Workforce for Older Americans, Institute of Medicine. Retooling for an aging America: Building the health care workforce. Washington, DC: National Academies Press; 2008. (Report No. ISBN: 0-309-11588-4. )
  • Dillman DA. Mail and telephone surveys: The total design method. New York: Wiley-Interscience; 1978.
  • Dillman DA, Smyth JD, Christian ML. Internet, mail, and mixed-mode surveys: The tailored design method. 3rd ed. Hoboken, NJ: John Wiley; 2009.
  • Donoghue C. Nursing home staff turnover and retention: An analysis of national level data. Journal of Applied Gerontology. 2010;29:89–106.
  • Eberhardt BJ, Shani AB. The effects of full-time versus part-time employment status on attitudes toward specific organizational characteristics and overall job satisfaction. Academy of Management Journal. 1984;27:893–900. [PubMed]
  • Feldman DC. Reconceptualizing the nature and consequences of part-time work. Academy of Management Review. 1990;15:103–112. [PubMed]
  • Hayes AF, Matthes J. Computational procedures for probing interactions in OLS and logistic regression: SPSS and SAS implementations 2. Behavior Research Methods. 2009;41:924–936. [PubMed]
  • Kiyak HA, Namazi KH, Kahana EF. Job commitment and turnover among women working in facilities serving older persons. Research on Aging. 1997;19:223–246.
  • Lagnado L. Rising challenger takes on health care system. The Wall Street Journal. 2008, June 24 Retrieved from
  • McNemar Q. Note on the sampling error of the difference between correlated proportions or percentages. Psychometrika. 1947;12:153–157. [PubMed]
  • Mittal V, Rosen J, Leana C. A dual-driver model of retention and turnover in the direct care workforce. The Gerontologist. 2009;49:623–634. [PubMed]
  • Nunnally JC. Psychometric theory. 2nd ed. New York: McGraw-Hill; 1978.
  • Ong PM, Ricklkes J, Matthias R, Benjamin AE. California caregivers: Final labor market analysis. Los Angeles, CA: UCLA School of Public Policy and Social Research; 2002.
  • Parsons SK, Simmons WP, Penn K, Furlough M. Determinants of satisfaction and turnover among nursing assistants. The results of a statewide survey. Journal of Gerontological Nursing. 2003;29:51–58. [PubMed]
  • Preacher KJ, Hayes AF. SPSS and SAS procedures for estimating indirect effects in simple mediation models. Behavior Research Methods, Instruments, & Computers. 2004;36:717–731. [PubMed]
  • Preacher KJ, Hayes AF. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods. 2008;40:879–891. [PubMed]
  • Price JL, Mueller CW. A causal model of turnover for nurses. Academy of Management Journal. 1981;24:543–565. [PubMed]
  • Rosen J, Mittal V, Degenholtz H, Castle N, Mulsant B, Rhee YJ, et al. Organizational change and quality improvement in nursing homes: Approaching success. Journal for Healthcare Quality. 2005;27:6–14. 21, 44. [PubMed]
  • Seavey D. The cost of frontline turnover in long-term care. Washington, DC: American Association of Homes and Services for the Aging; 2004.
  • Span P. The nursing home as battle zone. The New York Times. 2009, October 7 Retrieved from
  • Stuit D, Smith TM. Teacher turnover in charter schools. National Center on School Choice; 2010. Retrieved from
  • Werbel JD. The impact of primary life involvements on turnover: A comparison of part-time and full-time employees. Journal of Organizational Behavior. 1985;6:251–258.
  • Williamson OE. The economics of organization: The transaction cost approach. American Journal of Sociology. 1981;87:548–577.

Articles from The Gerontologist are provided here courtesy of Oxford University Press