| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
Correspondence: Address correspondence to Bita A. Kash, MBA, Department of Health Policy and Management, Texas A&M University Health Science Center, School of Rural Public Health, TAMU 1266, College Station, TX 77843. E-mail: bakash{at}srph.tamhsc.edu
 |
Abstract |
|---|
 TOP Abstract MethodologyResultsDiscussionReferences |
|---|
Key Words: Nursing homes • Nurse staffing • Staff turnover
Current minimum staffing requirements specified in state and federal laws that govern nursing homes have not been able to ensure acceptable quality of care. In fact, the vast majority of nursing homes do not have sufficient nursing staff (Harrington, Kovner, et al., 2000; Harrington, Mullan, & Carrillo, 2004; Walshe & Harrington, 2002). Potential reasons for this might be that (a) staffing requirements have been neither uniformly implemented by facilities nor enforced by regulatory agencies, and (b) written staffing standards alone may be insufficient to influence staffing levels, because other policies, such as reimbursement rates, and facility-specific features, such as resident case mix, ownership, and turnover, may affect staffing levels (Mueller et al., 2006). In addition, existing nursing home regulations fail to address the issue of staff turnover, despite its known association with quality of care. Because public policy has yet to achieve adequate staffing levels and lower turnover rates, it is important to explore methods other than staffing standards to address these problems.
Previous research on nursing home staffing has not fully accounted for the apparent reciprocal relationship between staffing levels and turnover. The assumption traditionally made is that low staffing levels will result in overburdened staff and poor quality, leading to increased staff turnover, which in turn increases vacancies (Harrington & Swan, 2003). However, the relationship may be more complex. Indeed, our approach assumes that staff turnover affects staffing levels rather than only the reverse. Further, we believe that it is possible to identify organizational factors that affect turnover but not staffing levels, allowing for a more appropriate staffing and turnover modeling approach.
Two factors led us to believe that staff turnover affects staffing levels. First, as several studies and reports have demonstrated, there is a shortage of individuals willing to work in nursing, and most health care providers, from hospitals to nursing homes, are grappling with these shortages (IOM, 2004; Seago, Spetz, Alvarado, Keane, & Grumbach, 2006). This factor alone may account for the effect of staff turnover on staffing levels, as new vacancies become increasingly difficult to fill with new hires (Staw, 1980). Second, however, high staff turnover also may be due to specific conditions that make a facility an unattractive place to work, such as poor management and staff mix. These factors may not directly affect staffing levels, but, when turnover occurs, they make it difficult to recruit new staff to fill vacancies and thus negatively affect staffing levels over time (Bowers, Esmond, & Jacobson, 2003; Castle & Engberg, 2006). This type of relationship is often called endogeneity. Endogeneity occurs when, as a result of omitted variables, an independent variable (staff turnover) is correlated with the error term in the staffing prediction model (Wooldridge, 2003). The presence of endogenous variables can lead to biased results; in our case, the underlying causes of low staffing levels and high turnover may be misidentified.
Clarifying the relationship between staffing levels and turnover, as well as understanding the factors associated with each, is critical to improving nursing home quality. In this study we attempt to clarify the underlying relationship by including staff turnover (an endogenous variable) in models that predict staffing levels. This involves the use of instrumental variables, that is, identifying predictors of staff turnover that are not associated with staffing levels (Wooldridge, 2003). Thus, in this study we attempted to build upon the research of previous studies (e.g., Harrington & Swan, 2003) on staffing and turnover by correcting for the endogenous relationship through the use of instrumental variables for staff turnover in two-stage least squares (2SLS) models that predict staffing levels. We used results from previous studies on staff turnover to identify facility-specific characteristics (instrumental variables) that are associated with staff turnover rates but not related to staffing levels. In addition to this methodological approach, we examined direct care staffing levels by focusing on three staff categories: registered nurses (RNs), licensed vocational nurses (LVNs), and certified nursing assistants (CNAs).
Literature Review
Many studies have examined staffing levels in U.S. nursing homes, but only a limited number have attempted to explain staffing levels as a function of staff turnover. Most studies have used staffing as a predictor variable in models of nursing home quality. We found only five published articles (from 1990 to 2005) that used a measure of staffing level as the dependent variable (Cohen & Spector, 1996; Grabowski, 2001; Harrington & Swan, 2003; Konetzka, Yi, Norton, & Kilpatrick, 2004; Zinn, 1994).
A 1994 study of RN staffing aggregated facility information to the county level, which helped with the identification of market factors related to staffing levels but lacked focus on facility-level factors affecting staffing and turnover (Zinn, 1994). The focus of the most recent studies has been on the effect of reimbursement systems and level on quality in nursing homes. Cohen and Spector (1996) found that reimbursement level was associated with higher staffing levels, which was associated with better quality of care. Using a more recent dataset and alternative methodology, Grabowski (2001) confirmed these findings and concluded that a retrospective-based reimbursement system was associated with a higher average number of RNs than a prospective-based system. Another recent study considered the effect of policy variables related to Medicare-payment changes on staffing levels (Konetzka et al., 2004). That study of skilled nursing facilities concluded that Medicare's Prospective Payment System had a negative effect on professional staffing (RNs and LVNs). None of these studies explored the effect of staff turnover on staffing levels.
The study of California nursing homes by Harrington and Swan (2003) did consider staff turnover as a predictor of staffing levels; it examined the apparently reciprocal relationship but did not offer instrumental variables for staff turnover when examining this relationship. The analysis of staffing levels as a function of staff turnover may produce biased results if one does not identify instrumental variables for staff turnover first. Further, this study found that the Medicaid-reimbursement level and the proportion of residents whose care is paid for by Medicare or Medicaid versus private insurance are significant predictors of facility decisions about hiring and retention of direct care staff (Harrington & Swan, 2003).
These prior studies examining the staffing of nursing homes have stressed the concept of resource dependency in the process of decision making by facility operators about staffing levels. The argument is that decisions about staff intensity and configuration are often influenced by the level of available resources to the nursing homes, and that additional research on a broader and more specific array of organizational characteristics that affect staffing levels and turnover is necessary to fully understand these decisions (Harrington & Swan, 2003; Konetzka et al., 2004).
The literature currently supports the idea that staff turnover has an adverse effect on a variety of quality measures in nursing homes (Burgio, Fisher, Fairchild, Scilley, & Hardin, 2004; Castle, 2001; Castle & Engberg, 2005; Zimmerman, Gruber-Baldini, Hebel, Sloane, & Magaziner, 2002). Additional consequences of high turnover in nursing homes have been lower standards of care, increased workload for the remaining staff, and higher costs for the facility (Caudill & Patrick, 1991; Knapp & Missiakoulis, 1983; Staw, 1980). Despite these findings, there has been insufficient attention to the relationship between staff turnover and staffing levels. We attempted to identify organizational characteristics (instrumental variables), beyond those affecting both staffing levels and staff turnover, that influence staff-turnover rates only. These instrumental variables, we believe, are facility factors that are realized by direct care staff after the initial hiring period.
Development of Hypotheses
Experts have recommended that nursing homes dedicate financial resources to the support of nurse training and skill improvement in order to ensure quality of care and patient safety (IOM, 2004). We expect these recommended measures, such as staff training and improved management practices, to also reduce staff turnover rates in nursing homes. Studies of nursing home staff turnover have identified specific factors associated with turnover rates. These include staff benefits, in-house CNA training, and management continuity, and these are potentially important organizational factors that are useful in the development of retention strategies (Bowers et al., 2003; Castle, 2005). On the basis of these selected results from prior research on factors affecting staff turnover, we were able to identify predictors of turnover (instrumental variables) before we evaluated the effects of turnover on staffing levels. We expected to find a significant relationship between staff turnover and staffing levels for all three staff types.
|
Methodology |
|---|
|
TOPAbstractMethodologyResultsDiscussionReferences |
|---|
Data Sources
We drew our population of nursing homes from the 2002 Texas Nursing Facility Medicaid Cost Report, which included 1,017 facilities. We dropped 3 facilities because they had extremely low occupancy rates related to a relatively short period of operation; this reduced the sample size to 1,014. This study does not include hospital-based facilities, because these are not included in the Texas Nursing Facility Medicaid Cost Report and would constitute a different population of residents and staff. Because the cost report was corrected and audited by the Texas Health and Human Services Commission (THHSC), the data did not include any omitted variables or observations; we calculated facility-level measures, such as occupancy rates and reimbursement rates, by following commission instructions.
We extracted county-level market factors from the 2003 Area Resource File, which combines 2000 Census data with the most recent data from the Bureau of Labor Statistics and other health-care-specific data sources. Because county codes from the two data sources did not correspond, we manually recoded all Area Resource File county codes before we merged the two data sources.
Dependent Variables
The first dependent variables of interest were direct care staffing levels. Direct care staff included RNs, LVNs, and all CNAs (including medication and respiratory aides). We measured staffing levels for each staff type by using the commonly used hours per resident day measure of staffing levels. This measure of staffing takes into account both staff hours and resident days, which captures the amount of direct care provided to each resident per day.
The second dependent variable of interest was staff-turnover rate. We measured staff turnover by dividing the number of employees who are no longer employed (total number of W2 forms filed minus the number of employees at the end of the reporting period) by the number of employees at the end of the reporting period for each category of direct care staff. This calculation is close to the formula recommended by the Bureau of Labor Statistics, which is defined as "the number of total separations for the year divided by the average employment level for the year" (Department of Labor, 2005).
Independent Variables
Organizational Factors Affecting Staffing and Turnover
Facility-level variables included profit status and chain membership; number of licensed beds; occupancy rate; level of resources (Medicare, Medicaid, and private-pay resident day percentages as well as Medicaid-reimbursement rates); and hourly wages for the three direct care staff categories. We included the facility's average case-mix index (CMI) to control for the level of the residents' needs for staff assistance, supervision, and monitoring. The CMI is a composite measure of resident acuity at the facility level, based on the average Texas Index of Level of Effort, a case-mix classification system similar to the Resource Utilization Groups used for Medicaid-reimbursement purposes in other states and in the Medicare program (Fries et al., 1994).
Demographic and Labor Market Factors That Affect Staffing and Turnover
Following the example of Harrington and Swan (2003), we included covariates in the prediction models. These include demographic variables, such as the proportion of individuals in the population who are aged 85 years and older, the proportion of racial or ethnic populations, and per capita personal income. We also included labor market variables such as the percentage of women in the labor force; county unemployment rates; the proportion of RNs, LVNs, and CNAs in the population; and female unemployment rates for women in our staffing and turnover models. Using the market share of facility beds in the county, we measured level of market concentration with the Herfindahl index, which is a capacity-based market-concentration measure. We also included the urban influence code, which rates level of urban influence at the county level on a scale from 1 (most urban) to 9 (least urban). Many of these demographic and market variables have been reported to be significant in prior studies of nursing home quality and staffing (Cohen & Spector, 1996; Harrington & Swan, 2003; Zinn, 1994). We expected staffing levels and turnover rates to be more affected by organizational factors and less affected by market factors.
Instrumental Variables That Affect Turnover
The facility-level characteristics that we used as predictors of staff turnover were staff training expense ratio (total resident-care-staff training expense/net resident revenues), direct care staff benefit expense ratio (direct-care-staff employee-benefits expense/net resident revenues), professional staff ratio (RN and LVN hours/CNA hours), contracted staff ratio (contracted direct-care-staff hours/employed direct-care-staff hours), administrative expense ratio (total administrative and central office expenses/net resident revenue), RN turnover rates, and in-house CNA training (a dummy variable). We included RN turnover as a potential negative predictor of LVN and CNA turnover on the basis of recent research findings that linked administrator (management) turnover to direct care staff turnover (Castle, 2005). Research has also shown that in-house CNA training may have a negative effect on retention, and therefore we examined this variable as a possible instrumental variable for CNA turnover (Brannon, Zinn, Mor, & Davis, 2002).
Analysis
The variables of interest in this study were staff turnover rates for RNs, LVNs, and CNAs. We modeled staffing levels for the three nurse types by using a set of organizational characteristics including the respective nurse-type wages and a set of demographic and labor market variables (which also included the respective nurse-type populations). We included respective staff turnover rates as the variable of interest in both OLS and 2SLS models for each of the three nurse-type staffing models. We performed formal tests of endogeneity and concluded that staff turnover was indeed endogenous in all three staffing-level prediction models. We followed the commonly recommended residual analysis steps in testing for endogeneity (Wooldridge, 2003).
We addressed staff-turnover endogeneity by applying 2SLS models, using groups of instrumental variables associated with staff turnover but not staffing levels (Wooldridge, 2003). We included instrumental variables as predictors of staff turnover in the first-stage models, but not in the second-stage staffing-level regressions. We first started with all potential instrumental variables in the turnover models and tested for the significance of groups of instrumental variables by using series of F tests. Next, we evaluated the significance of staff turnover as a predictor of staffing levels for all three staff types by using OLS and 2SLS. In order to answer the question of how important staff turnover is as a predictor of staffing levels when compared with other significant factors that affect staffing, we calculated fully standardized beta coefficients from the OLS results. The standardized beta coefficient is a useful measure of the relative impact of each independent variable on staffing levels, because it eliminates the units of measurement (metrics) and just reports effect size in terms of standard deviations (Long & Freese, 2003). Finally, we ruled out CMI endogeneity by performing formal statistical tests for endogeneity, and we treated CMI as an exogenous variable in the OLS and 2SLS models. We tested for CMI endogeneity by following the same procedures used to test for turnover endogeneity (Wooldridge).
|
Results |
|---|
|
TOPAbstractMethodologyResultsDiscussionReferences |
|---|
|
Staff Turnover
On the basis of the first-stage regression results presented in Table 2, our attempt to explain staff turnover was very successful for LVNs, somewhat successful for CNAs, and not successful for RNs. Instrumental variables actually used in the first-stage turnover regressions are those with coefficient estimates in Table 2.
|
The LVN turnover model, in contrast, was useful in explaining LVN-turnover variation, as we can see from the adjusted R2 value of 22%. RN turnover was a highly significant predictor of LVN turnover, confirming results from previous studies on management-turnover effects (Castle, 2005). We found that the ratio of professional staff to nonprofessional staff and the ratio of contracted to employed staff were both significant predictors of LVN turnover. A higher professional staff ratio reduced LVN turnover, whereas a higher contracted staff ratio (agency staff) mix increased LVN-turnover rates. For-profit nursing homes were associated with higher LVN turnover—a consistent pattern across all staff types.
We were also successful in explaining CNA-turnover rates by using a selected group of instrumental variables. The significant predictor of CNA turnover was the administrative expense ratio, which had a negative association with CNA turnover. This result confirms previous research results linking better management practices and capacity with reduced CNA-turnover rates (Banaszak-Holl & Hines, 1996; Castle, 2005). We did not observe the expected negative correlation between staff training expense and CNA turnover. For-profit facilities and higher proportion of Medicare resident days were associated with higher CNA-turnover rates. One important observation is that higher CNA wages did indeed reduce CNA turnover, a relationship that is unique to CNAs only.
Staffing Levels
Results from both OLS and 2SLS models (second-stage results) for RN, LVN, and CNA staffing intensity are presented in Table 3. We were able to identify instrumental variables for LVN and CNA turnover, but not for RN turnover. Therefore, we recommend the use of OLS models for RN staffing levels.
|
Higher occupancy rates had a negative effect on RN and LVN staffing and a significantly positive effect on CNA staffing levels. Nursing home size had a negative effect on RN staffing levels and a positive effect on LVN and CNA staffing; this is possibly an adverse effect of the minimum standards on RN staffing imposed by current regulations. Reimbursement rate was a significant and positive predictor of staffing levels, confirming resource dependency of staffing decisions (Harrington & Swan, 2003). As we expected, the percentage of Medicaid days had a significant negative effect on staffing levels, whereas the percentage of Medicare days had a significant positive effect on both RN and LVN staffing. Significant results related to demographic and market factors were detected for RNs and LVNs only. There seemed to be a positive relationship between the county's per capita income and the nursing home's ability to hire RNs. Higher proportions of women in the labor force and more LVNs in the county population had a significant positive association with LVN staffing levels. This might indicate that LVN staffing levels are more sensitive to labor supply factors than are the other direct care staff categories. As we can see from the OLS and 2SLS regression results, significant factors affecting staffing levels are consistent across models and parameter estimates are very similar.
Finally, the analysis of staff turnover as a predictor of staffing levels revealed mixed results, depending on staff type. RN turnover was associated with RN staffing ratios, but this relationship was only significant in the preferred OLS model. Therefore, we find support for Hypothesis 1 and conclude that RN turnover might indeed be a significant predictor of RN staffing levels. LVN turnover was not associated with LVN staffing levels. This result was consistent across OLS and 2SLS models. Therefore, we could not support Hypothesis 2 and concluded that there is no significant association between LVN turnover and LVN staffing levels. Results from OLS and 2SLS regressions confirmed that CNA turnover is indeed a significant predictor of CNA staffing levels in nursing homes. Therefore, we find support for Hypothesis 3 and conclude that CNA turnover and CNA staffing levels are related and the relationship is significant even after we correct for the endogeneity of CNA turnover and control for all covariates.
Relative Impact of Staff Turnover on Staffing Levels
We examined the relative impact of staff turnover on staffing levels compared with other significant predictors of staffing levels by calculating fully standardized beta coefficients (Long & Freese, 2003). We present and compare standardized coefficients from the OLS regression for all three staff types in Table 4.
|
LVN staffing levels were most affected by occupancy rates (negative association), LVN wages (negative association), and reimbursement rates (positive association). The next most important predictors of LVN staffing were for-profit ownership and percentage of Medicaid days (both had negative association with LVN staffing levels). Number six and seven in terms of highest impact were the supply of LVNs and the proportion of women in the labor force. LVN turnover was not a significant predictor of LVN staffing levels.
In the case of CNAs, we see that the predictor variable with the highest relative impact on CNA staffing was reimbursement rate (positive association), followed by ownership type, facility size (positive association), and proportion of Medicaid days (negative association). The next two variables with the highest relative impact were occupancy rate (positive association) and CNA turnover (negative association). We see that CNA turnover ranked sixth when it is compared with other significant predictors of CNA staffing levels, which are mostly related to facility resources and capacity.
|
Discussion |
|---|
|
TOPAbstractMethodologyResultsDiscussionReferences |
|---|
Our results confirmed that the most significant predictors of staffing levels and staff turnover were organizational characteristics, making staffing intensity less dependent on market factors and more sensitive to ownership status and facility resources. LVN turnover was not associated with LVN staffing levels, although we were very successful in identifying instrumental variables for LVN turnover. RN turnover rates were an important predictor of LVN turnover. We also found that LVN staffing intensity is indeed affected by important market factors, such as the proportion of female workers in the labor force and LVN labor supply. The insight this study has added to our understanding of nursing home staffing and turnover is that management (RN) turnover is a significant predictor of LVN turnover, and that turnover does not necessarily affect staffing levels when staffing levels are highly sensitive to market factors, as in the case of LVNs.
This analysis of Texas nursing homes reveals a significant relationship between CNA wages and CNA turnover rates, and a negative correlation of wages and staffing levels in general. Although wages seem not to be effective recruitment incentives in a market dominated by for-profit nursing homes, they do reduce turnover rates for CNAs significantly. Therefore, higher wage rates can be a disincentive to hire more staff at the facility level and an incentive to continue working at a nursing home for CNAs. This result would seem to lend some support to a recent CMS (2001) report suggesting that a $2 per hour pay increase would reduce high CNA turnover.
CNA turnover was also affected by administrative expenses (a measure of management capacity). Our results show that higher administrative expenses, including central office expenses related to multifacility administration, were associated with lower CNA turnover rates. These results may suggest that better management, in the form of qualified administrators and higher management capacity, as well as higher wages would help with CNA retention. This finding supports research showing the possible spillover effects in the nursing home setting coming from top management (Castle, 2001, 2005). Moreover, as Castle (2005) asserts, improving top management issues in nursing homes also represents another tool available to reduce staff turnover.
Policy Implications
Most attempts to achieve "good" staffing levels have focused on specifying minimum standards and factors that may affect facility hiring decisions, such as reimbursement rates. However, our findings suggest that more research is needed to understand the dynamics of turnover and that attempts to achieve and maintain adequate staffing levels in nursing homes should include policies specifically aimed at improving retention rates.
Nursing home staffing and turnover, according to these Texas facilities, were not always related as we had expected. RN turnover was associated with RN staffing, and CNA turnover was associated with CNA staffing; while LVN turnover was not significantly related to LVN staffing levels. Therefore, it is important for policy incentives to affect both staffing levels and turnover rates, since both have proven to be associated with quality of care in nursing homes and are not always associated with one another.
In order to improve CNA retention, incentives should be directed toward an increase in CNA wages and the development of management capacity and better management practices in nursing homes. Policy incentives at the facility level should also focus on increasing the number of RNs and LVNs compared with CNAs and reducing the reliance on contracted staff in order to improve LVN retention. LVN turnover rates were also highly sensitive to administrative factors, such as management continuity measured by RN turnover rates. Therefore, policy initiatives that involve management capacity building in nursing homes could improve both LVN and CNA retention.
At the market level, policy could be directed toward improving the ability of nursing homes to hire more LVNs. LVN staffing levels were affected by market factors, such as supply of licensed nurses in the county population and percentage of female workers in the labor force. This might explain the relatively high vacancy rates for LVNs compared with other staff types, based on a recent survey of nursing homes (Decker et al., 2003). The observed market dependency of LVN staffing is potentially amendable to policy interventions such as better public access to licensed nursing programs and promotional activities in high schools and community colleges to encourage licensed nursing careers in nursing homes.
Limitations
Texas has a large number of nursing homes compared with other states with a high percentage of for-profit facilities. Texas also has a well-established Medicaid-cost-report process, allowing for a thorough examination of expense categories, staffing levels, and turnover rates by staff type. Looking at quality-enforcement activity compared with other states, we see that Texas falls in the lower third quartile (Harrington et al., 2004), and therefore it is expected to be less affected by enforcement and more by organizational and market factors when it comes to staffing levels. Nevertheless, we cannot make the case that our findings are nationally representative.
Other limitations include the lack of information about the kind of training offered to staff. We assumed that all training expenses were strictly for long-term-care-specific training of CNAs. Further, we attempted to measure "management capacity" by using an administrative expense ratio. Ideally, higher administrative expenses should be associated with higher management capacity, but this relationship cannot always be assumed. In order to detect the true effects of management capacity on staff turnover, organizational-level data on administrator qualifications, educational attainment, experience, and management styles would be necessary.
Conclusion
Prior research has stressed the importance of understanding the factors associated with staff levels and turnover in nursing homes. This study of Texas nursing homes provides a detailed analysis of the relationship between staffing levels and staff turnover. Our findings show that staff turnover is not always associated with staffing levels. Therefore, policy initiatives should be directed toward improving staff levels as well as retention. Results from this study offer new information for policy affecting RN, LVN, and CNA recruitment and retention. On the basis of our analysis, staffing levels have a strong association with reimbursement rates and ownership type. Better management capacity and practices combined with higher CNA wages can help improve CNA retention. Increasing the population of licensed nurses can improve the ability of nursing homes to hire more LVNs.
|
Footnotes |
|---|
1 Department of Health Policy and Management, Texas A&M University, College Station.
2 Department of Health Policy and Management, University of Pittsburgh, PA.
3 Department of Economics, Texas A&M University, College Station.
Decision Editor: Linda S. Noelker, PhD
Received for publication December 6, 2005. Accepted for publication May 5, 2006.
|
References |
|---|
|
TOPAbstractMethodologyResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  N. G. Castle Nursing Home Caregiver Staffing Levels and Quality of Care: A Literature Review Journal of Applied Gerontology, August 1, 2008; 27(4): 375 - 405. [Abstract] [PDF]
|
|
|
|
 |
|
 C. E. Bishop, D. B. Weinberg, W. Leutz, A. Dossa, S. G. Pfefferle, and R. M. Zincavage Nursing Assistants' Job Commitment: Effect of Nursing Home Organizational Factors and Impact on Resident Well-Being Gerontologist, July 1, 2008; 48(suppl_1): 36 - 45. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. C. Stearns and L. P. D'Arcy Staying the Course: Facility and Profession Retention Among Nursing Assistants in Nursing Homes J. Gerontol. B. Psychol. Sci. Soc. Sci., May 1, 2008; 63(3): S113 - S121. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
