ArticlePDF Available

Causes of Caregiver Turnover and the Potential Effectiveness of Wage Subsidies for Solving the Long-Term Care Workforce 'Crisis'

Authors:

Abstract and Figures

Detailed data on private providers of long-term community-based residential services for persons with developmental disabilities permit investigation of the causes of frontline worker turnover. The endogeneity of turnover with compensation variables is accounted for in the estimation using instrumental variables. Turnover is determined by resident characteristics, frontline-worker compensation, and establishment characteristics. The share of higher-need residents and agency size predict higher turnover, while compensation and non-profit status are associated with lower turnover. Our findings indicate that public policies to reduce turnover through compensation subsidization can be effective. Our preferred estimates suggest an approximate one-quarter increase in total compensation would cut turnover by one-third.
Content may be subject to copyright.
The B.E. Journal of Economic
Analysis & Policy
Contributions
Volume 10, Issue 1 2010 Article 5
Causes of Caregiver Turnover and the Potential
Effectiveness of Wage Subsidies for Solving
the Long-Term Care Workforce ‘Crisis’
Elizabeth T. PowersNicholas J. Powers
University of Illinois at Urbana-Champaign, epowers@illinois.edu
njpowers@yahoo.com
Recommended Citation
Elizabeth T. Powers and Nicholas J. Powers (2010) “Causes of Caregiver Turnover and the Po-
tential Effectiveness of Wage Subsidies for Solving the Long-Term Care Workforce ‘Crisis’,The
B.E. Journal of Economic Analysis & Policy: Vol. 10: Iss. 1 (Contributions), Article 5.
Available at: http://www.bepress.com/bejeap/vol10/iss1/art5
Copyright c
2010 The Berkeley Electronic Press. All rights reserved.
Causes of Caregiver Turnover and the Potential
Effectiveness of Wage Subsidies for Solving
the Long-Term Care Workforce ‘Crisis’
Elizabeth T. Powers and Nicholas J. Powers
Abstract
Detailed data on private providers of long-term community-based residential services for
persons with developmental disabilities permit investigation of the causes of frontline worker
turnover. The endogeneity of turnover with compensation variables is accounted for in the esti-
mation using instrumental variables. Turnover is determined by resident characteristics, frontline-
worker compensation, and establishment characteristics. The share of higher-need residents and
agency size predict higher turnover, while compensation and non-profit status are associated with
lower turnover. Our findings indicate that public policies to reduce turnover through compensa-
tion subsidization can be effective. Our preferred estimates suggest an approximate one-quarter
increase in total compensation would cut turnover by one-third.
KEYWORDS: turnover, long-term care, Medicaid, wages
Veronica Alaimo and Sarah Jackson provided excellent research assistance. This work benefitted
from the comments of Economics Department seminar participants at the University of Kentucky,
Indiana University-Purdue University at Indianapolis, and the University of Illinois at Urbana-
Champaign. The authors appreciate the helpful comments of the editor and an anonymous referee.
INTRODUCTION
This paper examines the determinants of frontline worker turnover in the long-
term care industry. In particular, we study the causes of high turnover rates of
Direct Service Providers (DSPs) at small, community-based residential facilities
(‘group homes’) for individuals with developmental disabilities (DDs). In
addition to detailed information about the group home work environment, we
possess important information about their operators’ organization and
compensation practices.
Our paper makes several contributions. First, in contrast to past studies of
other sectors, we identify multiple factors strongly predictive of turnover,
including employee compensation, resident characteristics, and agency
characteristics. In particular, our findings on compensation support the
proposition that public subsidies could solve the ‘caregiver crisis.’ Second, we
also introduce a novel instrumental variables (IV) strategy for identifying the
exogenous effect of compensation on turnover. Third, using original data
collected from agencies operating group homes for individuals with DDs
throughout the state of Illinois, we estimate the determinants of turnover, treating
compensation as an endogenous variable. We find that this consideration is quite
important. Endogeneity-corrected estimates of the effect of compensation on
turnover are up to 65 percent greater than their ordinary least squares (OLS)
counterparts.
The DD sector is understudied considering its relative importance in the
long-term care industry. Public spending for DD services in the U.S. is large in
absolute terms ($34.64 billion in 2002, according to Rizzolo et al., 2004).1
Although more money is spent on the elderly, spending on individuals with DDs
is large relative to total long-term care spending, comprising 20 percent of all
long-term care spending and 40 percent of Medicaid spending on long-term care
in 2002 (Congressional Research Service, 2006). Further, our findings are likely
relevant for the fastest growing segment of long-term care—home- and
community-based care—because the DSP job parallels that of other home- and
community-based care workers. Both primarily provide custodial care, assisting
residents with the ‘activities of daily living,’ and both secondarily provide
developmental services, including helping residents maintain and restore physical
and mental skills and assisting residents with social activities. Because home-
and community-based care is both cheaper than and preferred by caregiving
recipients to traditional nursing home care (Congressional Research Service,
1 $22.60 billion was dedicated to community living arrangements, $7.66 billion to institutions, and
$4.39 billion to individual and family support services that enable persons with DDs to live on
their own or with their families. These public expenditures for long-term care omit public
spending on ‘traditional’ health care for persons with DDs.
1
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
2006), it is expected to continue to grow dramatically in the face of the financial
pressures created by the increased demand for living assistance due to population
aging.
TURNOVER IN THE CAREGIVER WORKFORCE
Frontline workers in the caregiving fields are often low-skilled and low-paid, and
turnover rates are among the highest of comparable industries in the United States
(Stephens and Riley, 2005). Direct-care worker turnover—measured as annual
total separations as a percent of annual average employment—is estimated
nationally at 50 percent (U.S. Department of Health and Human Services, 2006).
Turnover is particularly problematic when the worker is the ‘point of service’
(i.e., there is no intermediation between client and worker). Turnover has been
reported to lower the quality of long-term care (see Seavey, 2004, on long-term
care in general and Institute for the Future of Aging Services, 2007, on the frail
and disabled elderly), including DD services (U.S. Department of Health and
Human Services, 2006). In particular, Larson et al. (2004) find that frontline
worker turnover in the DD sector severely curtails residents’ opportunities, while
Test et al. (2003) report that it decreases resident safety, creates personal
problems including sadness, and generally diminishes the quality of life by
disrupting resident life and routines.
In addition to its detrimental impact on the quality of long-term care,
turnover plays a crucial role in the much-discussed U.S. caregiver shortage
‘crisis’ (Stone with Wiener, 2001, provides an overview). Rather than continually
recruiting new workers to fill open positions through increased vocational
training, immigration, or other measures, the demand for frontline caregivers in
the DD field over the next decade could be met by a less than one-third reduction
in the turnover rate corresponding to reduced attrition from the field (U.S.
Department of Health and Human Services, 2006).
High turnover also poses an obstacle to government meeting its statutory
mandates to persons with disabilities. The American with Disabilities Act of
1990, its interpretation by the U.S. Supreme Court in Olmstead v. Zimring, and
the Developmental Disabilities Assistance Act of 2000 recognize and reaffirm a
national commitment to offer institutionalized individuals a meaningful choice to
reside in the community, including the necessary support to do so. Workforce
issues have been central to the judiciary’s interpretation of these federal laws,
finding that scarce services inhibit the movement from institution to community
with “reasonable promptness.”2 High turnover outside of state-operated
2 For example, a federal district court ordered the state of Arizona to provide a wage to frontline
workers sufficient to ensure an adequate supply of the care services to which the plaintiffs were
entitled (Ball et al. v. Biedess et al., U.S. District Court of Arizona, August 12, 2004).
2
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
institutions likely contributes to guardian resistance to de-institutionalization
(Davis et al., 2000).
PRIOR RESEARCH
Other fields have produced a plethora of studies on frontline worker turnover, but
the questions asked are often fundamentally different from those asked here, and
few take an economic perspective. Much of this work emphasizes worker socio-
demographic factors (age, sex, race, marital status, and education) and working
conditions (type of ownership, size, workload, workplace social supports, and
internal job ladders; see, e.g., Wai Chi Tai et al., 1998, for a review of the nursing
literature) but often ignores basic factors of greatest interest to economists and
policymakers, chiefly compensation. Exceptionally, Mitchell and Braddock
(1994) estimate direct-care worker turnover for a sample of agencies providing
DD services. They find negative correlations between turnover and a variety of
compensation variables, but their univariate empirical analysis is a severe
shortcoming.
Studies of worker turnover in the economics literature are either highly
aggregated or focus on sectors that have little applicability to the long-term
caregiver workforce. When data are pooled across diverse industries, estimated
average effects are not particularly informative about specific types of workers.
An exception is Howes (2005), who examines the results of a near doubling of
wages and the introduction of health and dental care benefit plans for California
Bay-Area providers of home- and personal-care services over a 5-year period.
She estimates that a $1.00 per hour wage increase from $8.85 to $9.95 reduces the
annual turnover rate for new workers by 12 percentage points (from 0.32 to 0.20).
She also finds large reductions in turnover associated with the introduction of
health and dental benefits. Her findings suggest that turnover is quite responsive
to compensation. There are differences between many frontline caregivers and
workers in California’s homecare system, however. In addition to the work
taking place in the resident’s private home (true of all homecare workers), about
half of California homecare workers are family members of the resident, the
resident is the employer, and most of the workers are independent agents.
Perhaps most important, in this special setting, wages are set administratively at
the county level for all workers, regardless of experience or training, Therefore,
Howes (2005) need not consider the potential endogeneity of compensation with
turnover.
Research on worker turnover in childcare, nursing, and teaching is
relevant to turnover in long-term care to the extent that workers in these
occupations also provide custodial care and developmental services, to varying
degrees. For a sample of U.S. childcare centers, Powell et al. (1994) find a small
3
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
negative wage effect on turnover but fail to find consistent or significant effects of
other factors (e.g., center size, non-profit status, and child-staff ratios). For a
sample of Canadian childcare centers, Cleveland and Hyatt (2002) find that higher
wages and pension benefits, specialized childcare credentials, prior childcare
experience, and non-unionization all reduce turnover, as measured by the
worker’s reported intention to quit. Wages also have a statistically significant, but
small, negative effect on registered nursing turnover (see, e.g., Ahlburg and
Mahoney, 1996; Schumacher, 1997), although this wage effect is perhaps 50
percent larger holding split-shift work status constant (Holmas, 2002). Hanushek
et al. (2004) find that for Texas public schools, a higher salary reduces the
propensity to quit, but other school-level factors (large numbers of academically
disadvantaged, African-American, or Hispanic students) are much more
important. Falch and Strøm (2005), studying the population of public schools in
Norway, where there is uniform pay, confirm the importance of student
characteristics; teachers are more likely to leave schools with high shares of
minority and special-needs students.
Reasons for the small wage effect on turnover found in the related
occupations of childcare, nursing, and teaching vary. Given occupation-specific
training requirements (particularly in nursing and teaching), workers in these
fields may be less mobile and therefore less responsive to wages in the short run.
Findings from these studies may also be biased due to the unaddressed
endogeneity of compensation with turnover. Institutional features such as
collective bargaining (particularly in teaching) may also constrain differential
compensation. Findings for the more highly skilled nursing and teaching
professions may also reflect an increased importance of nonpecuniary job
characteristics at higher income levels.
The article proceeds as follows. After providing background on the DD
community residential services industry in the second section ‘Developmental
Disabilities Services,’ we discuss the sample and variables in the ‘Data and
Variable Construction’ section. In the section ‘Specifications, Estimation
Strategies, and First-Stage Results,’ we outline our empirical strategy and discuss
the first-stage findings from several instrumental variables approaches. Our main
empirical findings are then presented in the ‘Findings’ section, followed by a
discussion of their implications in the ‘Discussions and Conclusions’ section.
DEVELOPMENTAL DISABILITIES SERVICES
About 4.5 million Americans are developmentally disabled (Rizzolo et al., 2004),
experiencing a physical or mental impairment with onset prior to age 22 that has
altered or substantially inhibited their capacity to care for themselves or live on
their own. DDs are severe, chronic, and likely to continue indefinitely. Examples
4
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
of specific conditions that may result in DDs are mental retardation, cerebral
palsy, autism, and epilepsy.
Pursuant to a mandate from the Illinois General Assembly, we collected
data on group home operations from the non-state agencies that provide DD
services in Illinois.3 Individuals with DDs may receive services in residential or
nonresidential settings. Residential services provide both a place to live and
personal supports for daily living. Nonresidential services include developmental
activities and therapies to maintain and improve functioning, vocational and
employment support programs, and respite care or in-home personal assistance.
The foremost policy for maintaining adults with DDs in the community is
the group home (few individuals with DDs receive assistance in their private
homes). The Community Integrated Living Arrangement (CILA), a Medicaid
program, is the chief group home arrangement in Illinois, supporting persons with
DDs in residences of up to 8 persons. The state heavily regulates the CILA
industry, mandating minimum levels of many inputs, from staffing ratios to
square footage per resident. The state also sets per diem reimbursement rates for
CILA residents. Opportunities for further increasing per resident revenue are
severely limited by state-set service caps (CILAs are not for the medically fragile,
and allowances for nursing and other specialized care are strictly restricted).4
While agencies can freely enter and exit the CILA market, state policies ration
demand by capping the number of CILA residents at 7,200 and through explicit
program expenditure caps. Agencies may not recruit CILA residents; the state
distributes residents to agencies through another set of private agencies acting on
its behalf.
The DSP is the dominant frontline staff position at CILAs, accounting for
83 percent of all staff hours. The next-most-common position, house manager
(also a low-skilled position), accounts for another 9 percent of worker hours
(Powers et al., 2006). The state requires that DSPs possess a high school degree
or GED; formal training in human services is not required. The state allows a
maximum amount of reimbursable training for newly hired staff; agencies are
reimbursed for up to 40 hours of classroom training and up to 80 hours of on-the-
job training per new worker (Illinois Department of Human Services, 2002).5
Unlike other settings (e.g., California’s personal care/homecare system), agencies
have flexibility in wage-setting. There is no statutory obligation on the agencies
3 While the state of Illinois provides services directly to a small number of people through its
state-operated developmental centers, most DD services are provided through a state-funded
network of private and occasionally county-operated agencies.
4 In the latter respect, CILAs might be considered most similar to “assisted living.”
5 Upon completion, an inexperienced worker is certified as a DSP. The state also recommended 3
days of training annually for every incumbent DSP at the time the survey was conducted but does
not fund this additional training.
5
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
in this sample to pay a particular wage. Nor are agencies’ total payrolls
predetermined by the state; 28 percent of revenues are from non-state sources
(Powers et al., 2006).
The CILA-DSP job is physically and emotionally demanding. Residents
require assistance to varying degrees with the activities of daily living (e.g.,
toileting, bathing, dressing, hygiene, and feeding). Since Medicaid policy
requires nearly all CILA residents to attend a developmental day program, the
need for assistance is greatest in the early morning, late afternoon, and evening.
DSPs are also needed on weekends to help with shopping, cooking, housekeeping,
and social outings. Nonstandard work schedules pose hardships for many
workers.
While we lack detailed information on the characteristics of the DSPs in
our sample, based on discussions with agency chief executive officers (CEOs),
they appear similar to others in the low-skilled caregiving workforce:
disproportionately low-educated, female, and members of racial minorities
(mostly African-American, in Illinois).
DATA AND VARIABLE CONSTRUCTION
A state-provided list of all 24-hour-staffed CILA sites in Illinois, identified by
agency operator and street address, formed the basis for a sample of specific
CILA sites. Agencies were solicited repeatedly by telephone and mail to
complete a multiple-module web-based questionnaire on their services, workers,
residents, revenues, and costs. To keep the survey manageable for respondents,
each CILA-operating agency was asked to provide information on a maximum of
5 randomly selected CILA sites.6 Despite the five-CILA limit, information about
most agencies’ CILA sites is fairly complete. Fifty-four percent of the responding
agencies reported on all of their CILAs (i.e., 54 percent operated 5 or fewer
CILAs), while 77 percent provided information on at least one-half.
The final data consist of 61 agencies and 200 associated CILA sites.7
Based on the few variables available for the population, the responding CILA
sites appear representative. Fifty-six percent of the responding sites are located in
the densely populated Chicago metropolitan area, as are 55.5 percent in the
6 To be perfectly clear, the sample contains only information on the CILA sites that were pre-
selected by the authors. In no case did the agency determine the sites it would report on, and
therefore there are no sample-selection problems arising from this source. In a few cases, selected
CILAs were found to have ceased operating. In this case, we asked the agency for an up-to-date
list of its sites and randomly re-selected sites for sampling.
7 The response rate of CILA-operating agencies was 33.3 percent (61 of 183 agencies operating
CILAs responded). The response rate of CILA sites was 33.2 percent (200 of the 603 CILA sites
responded). Surveys were completed by the agencies’ chief financial officers. The survey and
instructions are available upon request.
6
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
population. Thirty-seven percent of the responding sites are located in Cook
county (the core of the Chicago MSA), compared with 41 percent of the
population.
VARIABLES AND DESCRIPTIVE STATISTICS
Information was collected about the entire agency, about the agency’s aggregated
CILA operations, and about individual CILA sites. For purposes of estimating
worker turnover (available at the agency level), explanators that are CILA
characteristics are constructed by averaging the individual site-level
characteristics within agency. Due to the random sample design of CILA-site
data collection, these variables are unbiased estimates of the average
characteristics of an agency’s CILAs.
Table 1 presents the descriptive statistics of the agency-level sample used
for the turnover analysis. Turnover is measured as the share of the agency’s
employees on June 30, 2004 hired during state fiscal year 2004. Note that this is
not CILA-DSP-specific turnover. For now, we assume that turnover at CILA-
operating agencies is dominated by turnover of CILA-DSPs. Evidence presented
below indicates that this assumption is well supported by the data. Although
sources of turnover (quits, fires, layoffs, and retirements) are unavailable, our
discussions with agency officials indicate that new-worker quits are
overwhelmingly the prime driver of turnover.8
Table 1 indicates that, on average, 26.7 percent of workers employed on
June 30, 2004 were with an agency fewer than 12 months. This understates the
turnover rate for CILA DSPs because more-skilled workers with longer tenure are
included in the total and because this measure does not account for the same
position turning over multiple times during a year.9
There is considerable variation in average CILA resident characteristics
(see Table 1). A majority are males (54.1 percent) and 38 percent are older than
age 49 (21.4 percent of all residents are males older than age49).10 In addition to
8 Discussions with agency executives indicate that the typical pattern is for new hires in the DSP
position to quit quickly after finding the job is not to their liking. Firings are rare and this is a
stable industry without layoffs.
9 See U.S. Department of Health and Human Services (2006) and Stone with Wiener (2001) for
surveys of the literature and expert consensus estimates. Turnover rates comparable to ours are
provided by Larson and Lakin (1999), who determine that 35 percent of Minnesota frontline group
home workers are on the job for less than a year in the mid 1990s, and Howes (2005), who finds a
32 percent turnover rate for homecare workers in the late 1990s-early 2000s, which falls to 27
percent after eliminating resident-originated service terminations.
10 Residents are relatively old because many adults with DDs live at home until their parents
become infirm or die. While age 49 is not the standard cutoff for ‘elder’ in the general population,
many residents with DDs experience accelerated aging due to Down’s syndrome.
7
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
Table 1: Descriptive statistics
Turnover
Turnover Share of workers hired within the last year 0.267
(0.126)
Resident characteristics
Share male residents Site-average share of residents male 0.541
(0.234)
Share elder residents Site-average share of residents over age 49 0.380
(0.204)
Share residents elder
male
Site-average share of residents male and over age 49 0.214
(0.148)
Share nonambulatory
residents
Site-average share of residents nonambulatory 0.102
(0.137)
Share dually-
diagnosed residents
Site-average share of residents with developmental disabilities
and mental illnesses
0.357
(0.302)
Share blind/deaf
residents
Site-average share of residents blind, deaf, or both 0.058
(0.081)
ICAP score Site-average Inventory for Client and Agency Planning score 47.501
(9.069)
Log of ICAP score Natural logarithm of ICAP score 3.842
(0.201)
Job characteristics
Average wage Site-average hourly pretax wage for DSPs 9.970
(1.646)
Log of average wage Natural logarithm of DSP Average wage 2.286
(0.168)
Entry wage Site-average entry-level hourly pretax wage for DSPs 8.590
(1.089)
Log of entry wage Natural logarithm of DSP Entry wage 2.143
(0.125)
Wage schedule Site-average ratio of Experienced wage for DSPs to Entry
wage for DSPs
1.179
(0.176)
Log of wage schedule Natural logarithm of DSP Wage schedule 0.156
(0.131)
Health insurance Share of sites offering DSPs a health insurance plan 0.854
(0.342)
Retirement plan Share of sites offering DSPs a retirement plan 0.636
(0.473)
8
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
Table 1 (continued): Descriptive statistics
Job characteristics (continued)
Total compensation Site-average hourly pretax wage and fringe benefits for DSPs 11.065
(1.979)
Log of total
compensation
Natural logarithm of DSP Total compensation 2.387
(0.185)
Work load CILA residents per full-time equivalent DSP 1.172
(0.482)
Log of work load Natural logarithm of DSP Work load 0.095
(0.348)
Union Share of sites with any unionized workers 0.180
(0.363)
County
unemployment rate
County unemployment rate 0.065
(0.009)
Log of county
unemployment rate
Natural logarithm of County unemployment rate -2.741
(0.138)
Home health aide
(HHA) wage
Median hourly wage for Home health aide (HHA) by
Economic Development Region (EDR)
8.717
(0.311)
Log of home health
aide (HHA) wage
Natural logarithm of Home health aide (HHA) wage 2.165
(0.036)
Nursing aide (NA)
wage
Median hourly wage for Nursing aide (NA) by Economic
Development Region (EDR)
9.661
(0.662)
Log of nursing aide
(NA) wage
Natural logarithm of Nursing aide (NA) wage 2.266
(0.073)
Local average wage Site-average hourly pretax wage for DSPs at neighboring,
non-relative CILA sites
10.165
(1.000)
Log of local average
wage
Natural logarithm of DSP Local average wage 2.314
(0.101)
Agency characteristics
For-profit Binary variable = 1 if the agency is for-profit 0.180
(0.388)
Private nonprofit Binary variable = 1 if the agency is private nonprofit 0.705
(0.460)
Public nonprofit Binary variable = 1 if the agency is public nonprofit 0.115
(0.321)
9
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
Table 1 (continued): Descriptive statistics
Agency characteristics (continued)
Size Total employees 198.499
(164.510)
Log of size Natural logarithm of Size 4.836
(1.121)
Any MH services Binary variable = 1 if the agency provides any mental health
(MH) services
0.311
(0.467)
Number of agencies 61
Notes: Means reported with standard errors in parentheses beneath.
having a DD, 35.7 percent are diagnosed with mental illness (termed ‘dually
diagnosed’), 10.2 percent are nonambulatory, and 5.8 percent are blind and/or
deaf (‘blind/deaf’). These incidences are substantially higher than in the
similarly-aged general population. Residents’ Inventory for Client and Agency
Planning (ICAP) score, ranging in value from 0 to 100, is intended to indicate the
intensity of required supervision as determined by both the frequency and severity
of maladaptive behavior (although agency CEOs dismissed its usefulness in
conversation). The CILA residents of the average agency have an average ICAP
score of 47.5.
The mean of site-average CILA-DSP (nominal) hourly earnings is $9.97
circa 2004. The average ‘entry’ wage offered to an inexperienced worker is
$8.59. The ‘experience’ wage is that paid to a worker with 5 years’ experience.
The average ratio of experience-to-entry wage is 1.18, and the ratio of the average
wage to the entry wage is 1.16.
Most DSPs are offered fringe benefits; 91.5 percent of all CILA sites offer
a health insurance plan and 67.5 percent offer a retirement plan (figures not
reported in tables). Most agencies that offer DSPs health insurance make an
employer contribution for health insurance. Only 7 of the 183 CILA sites offering
DSPs health insurance do not. All agencies offering retirement plans contribute
to them. On average, employer contributions to health insurance and retirement
benefits constitute 9.4 percent of total DSP compensation (not reported in table).
The resident-to-DSP ratio, an indicator of the DSP work load, averages
1.17 CILA residents per full-time-equivalent (FTE) DSP (Table 1). The average
agency unionization rate (i.e., the average of agencies’ share of CILA sites with
any union workers) is 18 percent (Table 1). Twenty-four percent of the CILA
sites are unionized (not reported in a table), indicating that smaller agencies are
less likely to be unionized.
Forty-three agencies (70.5 percent of agencies), representing 160 CILA
sites (80 percent of CILA sites), are private non-profit, while 18 percent of
agencies (10.5 percent of CILA sites) are for-profit, and 11.5 percent of agencies
10
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
(9.5 percent of CILA sites) are county-operated (public non-profit). The average
agency has a workforce of nearly 200. Most agencies provide diverse services;
75.4 percent provide nonresidential DD services (typically CILA-complementary
day programs; figures not reported in Table) and 31.1 percent provide mental
health services (Table 1). Nearly 40 percent of all state payments made to these
agencies are reimbursements for the provision of CILA services (not reported in
table). We use variation in this measure of ‘CILA intensity’ (i.e., the dominance
of CILA operations, as measured in relative revenue from that activity) to explore
the robustness of our estimates below.
Additional variables were either merged from government sources
according to geographic location of the agency’s headquarters or constructed from
the sample. The average county unemployment rate (obtained from the Bureau of
Labor Statistics, BLS) is 6.5 percent (Table 1). As discussed below, we use two
wage measures—one for home health aide (HHA) and another for nursing aide
(NA)—in our assessment of the validity of other instruments for total
compensation. HHAs provide routine personal care in the resident’s home. NAs
include certified nursing assistants and hospital aides who provide routine
personal care under the direction of nursing staff. The median hourly wages for
these two similar occupations are obtained from the Illinois Department of
Employment Security by Economic Development Region (EDR), a broad
designation of Illinois counties into 10 geographic areas. The averages of these
wages are $8.72 and $9.66, respectively. Finally, as described in detail below, we
use address information to construct a ‘local wage’ for each CILA site, defined as
the average DSP wage at nearby competitors’ CILA sites. The average value of
this wage is $10.17.
SPECIFICATIONS, ESTIMATION STRATEGIES, AND FIRST-STAGE
RESULTS
In this section, we motivate and present the empirical specifications for employee
turnover. Potential problems of endogeneity and simultaneity with regard to
turnover and employee compensation are outlined. Instrumental variable (IV)
strategies to address these issues are explained, and empirical findings related to
the first stage, including tests of endogeneity as well as validity and weakness of
the instruments, are presented. Although the sample of agencies is small, the
preferred estimates account for the small sample size biases associated with two-
stage least squares (2SLS).
There is an extensive theoretical literature on job search and its
implications for turnover. The major determinants of turnover are the worker’s
appropriately discounted stream of perceived pecuniary and nonpecuniary
benefits and costs flowing from the current position relative to the expected
11
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
streams from alternative jobs, considering also the costs of search and of learning
about the current employer through experience. Key results are that wages
increase with tenure, quit rates fall with tenure, and firms may reduce turnover by
back-loading the wage schedule (Rogerson et al., 2005).
We specify turnover as a function of four types of variables influencing
these flows: compensation, the immediate CILA work environment, agency
personnel practices, and employer-employee relations. We lack information on
the personal characteristics of the DSP workers that may influence turnover. To
the extent that these characteristics are systematically correlated with included
variables, estimates are biased by this omission. The IV approach, however, helps
eliminate biases from this source with regard to the key explanator, total
compensation.
A higher wage reduces the benefits of job search. Fringe benefits
additionally contribute importantly to binding employees to employers, both
through ‘job lock’ (especially in the low-wage sector, where alternative jobs may
not offer fringe benefits) and because workers with a preference for job stability
may also have a strong taste for fringe benefits.
Higher total compensation is predicted to reduce turnover. There are
several potential ways to express compensation, and in principle it would be
desirable to break it down into its wage and fringe benefit components. We do
not present findings based on this approach. As a practical matter, promising
candidate instruments for the fringe benefit component (analogues of the ‘local’
average wage approach we develop below) have little explanatory power for
fringe benefits. When fringe benefits are treated as endogenous, the findings with
respect to the wage are stable and similar in magnitude to the findings for total
compensation presented below. Most importantly, reliable estimation and
diagnostic tests that allow for possible weakness in the instruments have been
developed for the case of a single endogenous explanator only; adding a second
troublesome explanator leaves one with no reliable inferential tools (Murray,
2006a, 2006b). In our judgment, given the current state of knowledge, it is
preferable to obtain reliable estimates for the total compensation case, as opposed
to entering a no-man’s-land of inference by expanding the specification.
The wage schedule (ratio of experience to entry-level wage) is included in
the turnover specification because back-loaded pay may attract more stable
workers to the agency and increase retention generally. The unemployment rate
controls for alternative job opportunities. Agencies located in counties with low
unemployment rates are expected to experience more turnover.11
11 The unemployment rate poses two potential limitations. First, to the extent that the county
unemployment rate does not adequately capture all of the influence of outside job opportunities
and such remaining influences are correlated with the instruments for total compensation, the
instrumented variables’ effect incorporates the influence of unmeasured outside job opportunities.
12
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
Resident characteristics principally control for the intensity of care and
corresponding worker effort required. Turnover is expected to be higher when
more residents are male, non-ambulatory, and dually-diagnosed with mental
illness. Male residents are on average stronger and heavier. Caring for non-
ambulatory residents is physically arduous and lifting poses significant injury
risk. Dually-diagnosed residents may be less cooperative, more unpredictable in
their behavior, and experience more frequent stressful crises. Turnover is
expected to be lower when residents are older or the average resident ICAP score
is higher, because caring for relatively inactive residents or residents with more
predictable behaviors is easier (aging residents who develop serious medical
problems are relocated from CILAs), and these residents may provide more self-
care. The share of blind/deaf residents may be associated with lower turnover,
because the DSPs who care for these residents possess unusual skills (e.g., sign
language) that make their retention a priority. We also include the work load
(resident-to-worker ratio), which is expected to increase turnover.
Agency personnel practices and worker-management relations may also
affect turnover. Economies of scale in recruiting should make higher turnover
rates more tolerable for larger agencies. Larger agencies may be more impersonal
and bureaucratic, and workers who cannot effectively voice their concerns are
more likely to address them through exit. On the other hand, larger agencies may
be more able to offer internal career ladders that aid retention.
We expect lower turnover at non-profit than for-profit agencies if non-
profits place greater priority on both resident and worker well-being, both of
which are presumably reduced by turnover. A sense of shared mission with a
non-profit employer may also engender worker loyalty. Unionization may reduce
turnover if unions and management work together to effectively voice and address
workplace concerns. In the general population, unionized workers tend to
experience longer tenure, ceteris paribus (Freeman, 2005). On the other hand, the
union itself may constitute a network for informally or formally sharing
information about alternative job opportunities, increasing turnover. Finally, we
include a variable indicating whether the agency offers mental health (MH)
services. Recall that our turnover variable is agency-wide. MH workers are more
highly skilled, specialized in their training, and better compensated, and we wish
to control for the depressing effect their presence may have on the overall
turnover rate.12
Second, if the market is tighter for an agency than the county unemployment rate suggests, they
may pay a higher wage but still face greater turnover. In this case, our estimated wage effect is
conservative.
12 Note that the DSP workers we study provide specific DD services at CILA sites. There should
be no systematic variation in DSP-CILA worker characteristics with respect to activities in
13
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
Total compensation is a potentially troublesome explanator for several
reasons. First, agencies likely determine total compensation and turnover
simultaneously. Unobservable turnover-reducing shocks may result in the agency
choosing lower total compensation, generating a positive correlation between total
compensation and turnover. If so, the ordinary least squares (OLS) coefficient of
total compensation understates its effect on turnover. Second, omitted variables
may be correlated with total compensation. Workers’ unobserved characteristics
may make them both high-quality and long-tenure. A positive correlation
between unobserved worker characteristics and total compensation implies that
the OLS coefficient for total compensation overstates its turnover-reducing effect.
Given that the primary compensation variable is based on the average wage and
there is no measure of tenure, higher compensation may simply reflect a longer-
tenure workforce with its associated lower turnover. The direction of the net bias
to OLS estimates resulting from these two concerns is ambiguous. While random
measurement error in total compensation would bias towards understatement, this
is not a concern; measures of hours and compensation come directly from payroll
records and should be quite accurate.
IV strategies are employed to address the potential biases outlined above.
Two alternative instruments for total compensation are proposed—the agency’s
entry-level DSP wage and the ‘local’ average DSP wage. In both cases, the home
health aide (HHA) and nursing aide (NA) wages for the agency’s EDR are used to
implement Anderson-Rubin overidentification tests. We describe the entry-level
and local-average wage IV approaches and the first-stage results in turn.
INSTRUMENTING AVERAGE TOTAL COMPENSATION WITH THE ENTRY-LEVEL WAGE
Rationales for using the entry-level wage as an IV are that it is based on the
prevailing wage for low-skilled workers in the agency’s local labor market and
population attributes that affect worker quality secularly (e.g., the quality of local
high school education); that it is offered to inexperienced new hires prior to the
agency obtaining idiosyncratic information about worker quality through
monitoring; and that it is unrelated to the tenure distribution of the current
workforce. Under these assumptions, the entry-level wage is a valid instrument
for total compensation. It is likely highly correlated with an agency’s total
compensation while actual compensation (and not entry-level wage) claims the
direct influence on turnover.
Table 2 presents the first-stage findings. Column 1 reports the coefficient
of the (natural log of the) entry wage when it is regressed on (the natural log of)
average total compensation alone (a finding of possible interest to some readers),
unrelated service areas (like mental health) that their agencies may offer. In short, we do not
expect DSP-CILA-worker variables to be collinear with the MH service variable.
14
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
while column 2 presents the first-stage regression using entry-level wage as the
instrument for total compensation.13 The coefficient for entry-level wage is
positive, as expected. Entry-level wage also largely overcomes the ‘weak’
instrument problem that biases 2SLS estimates of the coefficient and its standard
error (Murray, 2006a, 2006b). The first-stage F-statistic (F(1,52) = 12.19)
indicates that the true significance level of the hypothesis test based on 2SLS is
below 15 percent when the nominal level is 5 percent (Stock and Yogo, 2002).14
Nevertheless, some might still reasonably argue that this instrument is potentially
weak, and this possibility is accounted for in the calculation of preferred estimates
below.
While compelling to us, some may reasonably doubt whether the entry-
level wage is indeed a valid instrument. Additionally, the entry wage could also
conceivably reflect agencies’ unobserved characteristics. If it is possible to
differentiate workers prior to hire by offering a higher entry wage, then it is a
relevant explanator of turnover. We are skeptical of this argument, however,
given our conversations with agency CEOs about their entry-level hiring strategy,
which is to cast a wide net, knowing that most new hires have no experience in
long-term care and will not find the DSP job to their liking. Given this situation,
it seems futile to offer a higher wage to inexperienced applicants in hopes of
making a better match.
To address the validity concern further, we employ a version of the
Anderson-Rubin over-identification restrictions test that permits an assessment of
the validity of a subset of instruments whose validity is in doubt, conditional on
other instruments whose validity is not in doubt (Murray, 2006b). Because HHA
and NA wages are aggregated over wide geographic areas and because entry-level
DSPs select from a broad menu of alternative low-wage jobs spanning assorted
industries, there is a strong case that HHA and NA wages do not belong in the
turnover equation. DSP wages nevertheless should be correlated with HHA and
NA wages due to the similar skills required and shared industrial sector.
The Anderson-Rubin overidentification test statistic for HHA and NA
wages is X2(1) = 0.003 (p-value = 0.9583), supporting the notion that HHA and
NA wages are not relevant explanators in the turnover equation. HHA and NA
wages are ‘weak’ instruments (the first-stage F-statistic is F(2,51) = 2.19). This is
likely due to geographic aggregation over large areas (up to 19 counties), which
breaks down the correlation with particular agencies’ wages. Unfortunately,
moredisaggregated wage data are not available. They are nevertheless useful for
assessing whether the entry-level wage, a stronger potential instrument whose
13 Throughout the paper, small-sample corrections are made to all test statistics.
14 The Stock-Yogo test for the reduction of OLS bias achieved by 2SLS cannot be calculated
because this test requires at least three instruments when there is one endogenous variable (Stock
and Yogo, 2002).
15
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
Table 2: First- stage regressions for Turnover equation (Instrumenting Log of total
compensation)
Variables (1) (2) (3) (4)
Log of entry-level wage 0.595****
(0.176)
0.584****
(0.167)
Log of local average wage
0.536***
(0.227)
0.660****
(0.205)
Covariates
Resident characteristics
Share elder residents 0.252****
(0.101)
0.211***
(0.101)
Share non-ambulatory
residents
0.041
(0.158)
0.086
(0.164)
Share dually- diagnosed
residents
-0.071
(0.067)
-0.093
(0.068)
Share blind/deaf residents -0.303
(0.249)
-0.295
(0.252)
Other job characteristics
Log of county unemployment
rate
-0.114
(0.147)
-0.250*
(0.152)
Agency characteristics
Private nonprofit 0.080*
(0.054)
0.054
(0.054)
Log of size 0.042**
(0.023)
0.054***
(0.022)
F-test, instrumental variables F(1,59) =
11.40
p-value =
0.001
F(1,52) =
12.19
p-value =
0.001
F(1,59) =
5.54
p-value =
0.021
F(1,52) =
10.28
p-value =
0.002
F-test, all controls -- F(8,52) =
5.43
p-value =
0.000
-- F(8,52) =
5.08
p-value =
0.000
Notes: Entries are estimated coefficients with standard errors in parentheses beneath.
(*,**,***,****) indicate statistical significance based on a one-tail t-test exceeding the (90, 95,
97.5 and 99) percent levels of confidence. All test statistics and confidence intervals are adjusted
for sample size.
16
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
validity is in doubt, is indeed exogenous, because the Anderson-Rubin
overidentification test is robust to weak but valid instruments (Murray, 2006b).
The test statistic of X2(1) = 1.770 (p-value = 0.183) supports the notion that the
entry-level wage is not a relevant explanator in the turnover equation.15
INSTRUMENTING AVERAGE TOTAL COMPENSATION WITH THE LOCAL AVERAGE DSP WAGE
The alternative instrument for total compensation is the local average DSP wage.
This variable is constructed for each CILA site by averaging the average DSP
wage at the five closest CILA sites operated by the agency’s competitors within a
30-mile radius.16 The local average wage is exogenous with respect to an agency
but should be correlated with that agency’s total compensation due to overlapping
labor markets. In particular, its exogeneity with respect to the agency solves the
simultaneity problem and any problems due to unobserved agency attributes.
Since it is also exogenous with respect to worker characteristics, it is also
uncorrelated with worker tenure and heterogeneous worker characteristics at the
agency in question.
In the first-stage regression the coefficient of the (natural logarithm of)
local average wage in the specification for the (natural logarithm of) total
compensation is positive, as expected (see column (4) of Table 2; as before,
column (3), immediately to the left, contains the coefficient from a specification
with no additional explanators). Local average wage largely overcomes the
‘weak’ instrument problem. The first-stage F-statistic is F(1,52)= 10.28,
indicating that the true significance level of hypothesis tests based on 2SLS is
below 15 percent when the nominal level is 5 percent (Stock and Yogo, 2002).
Nevertheless, some might still reasonably argue that this instrument is potentially
weak, and this possibility is accounted for in the calculation of preferred estimates
below. The Anderson-Rubin test of whether the local average wage is valid
conditional on the validity of the HHA and NA wages is X2(1) = 0.717 (p-value =
0.397), supporting the notion that the local average wage is not a relevant
explanator in the turnover equation.17
15 The Anderson-Rubin test of the validity of a subset of instruments is performed using ‘ivreg2’, a
downloadable add-on command for Stata.
16 For the vast majority of the CILA sites (192 sites), the local average wage is based on five non-
relative, neighboring CILA sites located within 24.9 miles. For more isolated sites, fewer than 5
competitors’ sites were selected. The 30 mile radius was chosen under the assumption that on
average a worker would readily commute up to about 30 minutes from her current workplace.
17 Two other instruments for total compensation were found to be quite weak. Local entry-level
wage and local total compensation were constructed in the same manner as local average wage.
The first-stage F-statistic is F(1,52) = 2.53 for local entry-level wage and F(1,52) = 7.24 for local
total compensation. Local total compensation is not a strong instrument for total compensation
because marginal variation in total compensation is largely driven by its wage component. While
17
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
Before proceeding, we note that the findings for the various instruments
are as one might expect based on their variation in the data. The coefficient of
variation (CV) for the logarithm of DSP total compensation is 0.078. The CVs
for the wage instruments are uniformly lower because idiosyncratic agency
components are averaged out, leaving only ‘systematic’ variation. The CVs are
0.058 and 0.044 for the logarithms of entry wage and local average wage,
respectively, our two most promising instruments for total compensation. In
contrast, the CVs for the logarithms of home health aide wage and nursing aide
wage are substantially less at only 0.017 and 0.032, respectively. These two later
wage measures are, not surprisingly given their relative lack of variation, found to
be weak instruments for total compensation. They nevertheless are valid
instruments and are useful for assessing the validity of the entry wage and local
average wage, our two alternative wage instruments, both of which are strong
instruments, but whose validity may be in doubt.
FINDINGS
All specifications are estimated with OLS, two-stage least squares, or Fuller’s
estimator, as appropriate, using a semi-log form, where controls that are not either
binary or shares are logged. In fact, the findings differ little across linear,
semilog, and log-log specifications.
DETERMINANTS OF TURNOVER
Table 3 presents estimates of the agency turnover rate as a function of resident,
job, and agency characteristics. Column (1) presents findings from the complete
specification discussed above. The second column presents findings for the
parsimonious specification, dropping variables whose estimated effects differ
insignificantly from zero at a 20 percent or higher confidence level in a one-sided
t-test. The coefficients of selected resident characteristics (sex and ICAP), some
job characteristics (wage schedule, work load, union), and agency characteristics
(public non-profit and MH service provider binaries) are insignificant. A
comparison of columns (1) and (2) shows that the estimated coefficients for the
parsimonious specification are reasonably insensitive with respect to dropping the
other variables from the specification.18
there is little cross-site variation in fringe benefit compensation within an agency, there is cross-
site variation in wages in response to local labor market conditions. The second-stage findings
using these two ‘weak’ instruments are qualitatively similar to those reported below.
18 In particular, the ‘union’ effect is still insignificant when the union variable is maintained while
other insignificant variables are dropped. Because only 13 of the 61 agencies had any unionized
DSPs, it is quite possible that the incidence of unions in our sample is too low to identify a
potential independent union effect.
18
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
Table 3: OLS and 2SLS estimates of Turnover equation (Instrumenting Log of total
compensation)
OLS
estimates
OLS
estimates
2SLS
estimates
(Instrument is
Log of entry-
level wage)
2SLS
estimates
(Instrument is
Log of local
average
wage)
Variables (1) (2) (3) (4)
Resident characteristics
Share male residents -0.048
(0.087)
Share elder residents -0.108
(0.120)
-0.137***
(0.066)
-0.110*
(0.075)
-0.126**
(0.074)
Share elder male residents -0.044
(0.189)
Share nonambulatory
residents
0.467****
(0.114)
0.506****
(0.102)
0.498****
(0.106)
0.503****
(0.103)
Share dually-diagnosed
residents
0.083**
(0.049)
0.088**
(0.044)
0.074*
(0.049)
0.083**
(0.048)
Share blind/deaf residents -0.434***
(0.193)
-0.372***
(0.165)
-0.416***
(0.177)
-0.390***
(0.174)
Log of ICAP score -0.013
(0.083)
Job characteristics
Log of total compensation -0.247****
(0.094)
-0.237***
(0.082)
-0.391***
(0.194)
-0.302*
(0.203)
Log of wage schedule -0.040
(0.109)
Log of work load -0.016
(0.047)
Union -0.015
(0.047)
Log of county
unemployment rate
-0.186**
(0.110)
-0.173**
(0.097)
-0.196**
(0.103)
-0.183**
(0.101)
Agency characteristics
Private nonprofit -0.099**
(0.054)
-0.068**
(0.035)
-0.060*
(0.037)
-0.065**
(0.036)
19
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
Table 3 (continued): OLS and 2SLS estimates of Turnover equation (Instrumenting Log
of total compensation)
Agency characteristics (continued)
Public nonprofit -0.056
(0.063)
Log of size 0.040**
(0.020)
0.029**
(0.015)
0.038**
(0.019)
0.033**
(0.019)
Any MH services -0.026
(0.032)
Number of observations 61 61 61 61
R2 0.525 0.487 0.452 0.481
Adjusted R2 0.352 0.408 0.368 0.401
F-test F(16,44) =
3.04
p-value =
0.001
F(8,52) =
6.17
p-value =
0.000
F(8,52) =
5.31
p-value =
0.000
F(8,52) =
5.35
p-value =
0.000
Notes: Entries are estimated coefficients with standard errors in parentheses beneath.
(*,**,***,****) indicates statistical significance based on a one-tail t-test exceeding the (90, 95,
97.5 and 99) percent levels of confidence. All test statistics and confidence intervals are adjusted
for sample size.
Column (3) presents the findings when the entry-level wage instruments
total compensation, and column (4) presents the findings when the local DSP
average wage is the instrument. Higher total compensation always reduces
turnover, as hypothesized, and the magnitudes of these estimated effects are
substantial. A 10 percent increase in total compensation (a $1.11 increase at the
mean) reduces turnover from 3.02 to 3.91 percentage points, depending on the
instrument, representing a 27 to 65 percent larger effect than the OLS estimate
reported in column (2). The fact that both IV coefficient estimates exceed their
OLS counterparts suggests that simultaneity of total compensation with turnover
is the major source of bias.
When instruments are weak, confidence intervals are understated, and
2SLS point estimates can be severely biased as well (Murray, 2006a, 2006b).
While the first-stage findings for both IVs presented in the previous section
indicate that they are reasonably strong, we nevertheless explore the robustness of
the findings to any potential weakness. Moreira’s (2003) two-sided conditional
likelihood ratio (CLR) approach corrects for biased confidence intervals.
Following Murray’s (2006a, 2006b) recommendation for choosing a point
20
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
estimate for the coefficient of an endogenous explanator when instruments are
potentially weak, we compute point estimates using two of Fuller’s estimators
(with parameter values a = 1 and a = 4).19
Using entry-level wage as the instrument for total compensation, the CLR
test yields a 95-percent confidence interval of [-0.9571, -0.0046] with a p-value
for the null hypothesis of no effect of total compensation on turnover of 0.0304.
The Fuller estimates are -0.376 when a = 1 and -0.344 when a = 4 (the most
conservative endogeneity-corrected estimate). Both are reasonably close to the
2SLS point estimate of -0.391. These findings strongly suggest that the effect of
total compensation on turnover is negative and considerably larger—45 to 65
percent larger in absolute magnitude—taking account of endogeneity.
When local average wage is the instrument for total compensation, the
CLR test yields a 95-percent confidence interval of [-0.8645, 0.1632] and a p-
value for the null hypothesis of no effect of total compensation on turnover of
0.1252. The 2SLS estimate of the effect of total compensation on turnover is
-0.302; Fuller estimates are -0.295 (a = 1) and -0.280 (a = 4). While the local
average wage-instrumented findings are weaker, they again suggest a
conservative estimate of the effect of total compensation on turnover (-0.280) that
is quite large in absolute magnitude, or 18 percent larger in absolute size, taking
endeogneity into account.
The evidence collectively indicates that the simultaneity of total
compensation and turnover is a source of bias in the OLS estimates. The
estimated magnitude of this bias depends on the instrument set and the estimator.
In particular, the Fuller (a = 1) estimates are from 24 to 59 percent larger than
their OLS counterparts; the Fuller (a = 4) estimates are from 18 to 45 percent
larger than their OLS counterparts; and the 2SLS estimates are from 27 to 65
percent larger than their OLS counterparts.
The unemployment rate has the expected effect. A 10 percentage-point
increase in the unemployment rate reduces turnover by 1.73 to 1.96 percentage
points (columns (1) – (4)). Resident characteristics have the hypothesized effects,
and these effects are of substantial magnitude. A 10 percentage-point increase in
the share of nonambulatory residents increases turnover by 4.67 to 5.06
percentage points; a 10 percentage-point increase in the share of blind/deaf
residents reduces turnover by 3.72 to 4.34 percentage points; a 10 percentage-
point increase in the share of older residents reduces turnover by 1.08 to 1.37
percentage points; and a 10 percentage-point increase in the share of dually-
diagnosed residents increases turnover by 0.74 to 0.88 percentage points.
Agency characteristics affect turnover as expected. Private non-profit
agencies experience 6.0 to 9.9 percentage points less (or about one third less)
19 In the context of weak instruments, when a = 1 Fuller’s estimator is approximately unbiased.
When a = 4, Fuller’s estimator is biased, but its mean square error is less than when a = 1.
21
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
turnover than for-profits. Turnover is higher at larger agencies; a 10 percent
increase in agency size (a 20 worker increase at the mean) is associated with a
0.29 to 0.40 percentage-point increase in turnover.
ROBUSTNESS OF THE TURNOVER ESTIMATES
While the turnover measure is agency-wide, the right-hand-side variables, except
for agency characteristics, are specific to CILA operations. Since there is
variation in the extent to which agencies’ activities are dominated by CILAs, we
explore whether the turnover estimates, particularly with regard to the CILA-
DSP-specific variables, strengthen as we increasingly focus on agencies most
dedicated to CILAs. We re-estimate the basic OLS turnover equation, dropping
the 16, 25, and 33 percent of the agencies in our sample with the smallest CILA-
reimbursement revenue shares. Findings are consistent with CILA operations
dominating agency turnover. All estimated coefficients are robust with respect to
these sample alterations and the effects of many—notably resident characteristics
and total compensation—increase in absolute magnitude. The overall fit also
strengthens (the adjusted R-squared rises from 0.39 to 0.44).20
Several other pieces of evidence support our assumption that agency-wide
turnover is governed by CILA-DSP turnover. CILA-operating agencies
experience turnover that is 50 percent higher than that of agencies without CILA
operations. Further, when we estimate turnover equations—in each case
replacing CILA-specific variables with other program-area-specific variables
(MH and DD nonresidential programs)—there are no significant predictors of
turnover. Similarly, when we pool information across all program areas (MH,
DD nonresidential, and CILA) to create program-area-weighted-average
explanatory variables, no statistically significant relationships are found. Thus
assorted evidence collectively and strongly supports the notion that DD-agency
turnover is chiefly a CILA-DSP staffing problem.
DISCUSSION AND CONCLUSIONS
We have identified multiple factors influencing turnover of frontline caregivers.
Resident characteristics indicating a higher caregiving burden are associated with
increased turnover. Thus, the incidence of turnover is unevenly distributed, and
residents with the greatest caregiving needs (those who are younger,
nonambulatory, and dually diagnosed with a mental illness) face systematically
higher turnover and potentially lower-quality care. Turnover is higher at larger
agencies, consistent with personnel cost efficiencies or exit theory, or both.
Agencies paying higher total compensation to frontline caregivers experience less
20 Findings available upon request.
22
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
turnover. Depending on the chosen IV estimate, an increase of from 23.9 to 30.5
percent in total compensation (or an increase of $2.65 to $3.38 per hour at the
sample mean of $11.065 per hour) cuts turnover by one-third.21 Other findings of
interest are the effects of DSP unionization and agencies’ nonprofit status.
Unionization does not appear to reduce turnover directly. Nonprofit agencies
have lower turnover, consistent with the notion that they are more concerned with
both quality care and worker welfare.
The finding of an important influence of compensation on turnover in this
research stands in stark contrast to much prior work. Research on the childcare
and teaching sectors has produced surprisingly little evidence in favor of a
significant effect of compensation, while research on the nursing field has tended
to find only small wage effects. Industry differences, problems of accurately
measuring compensation, and the lack of endogeneity correction (which we find
increases the estimated effect of compensation) likely explain this difference.
While the childcare literature has been largely unsuccessful in identifying many
significant influences on turnover at all, our work is similar to prior research on
teaching in finding that ‘client’ characteristics (students, in the case of teaching,
and group home residents, as studied here) do have an important influence on
worker turnover. Given the intense and potentially long-term nature of the
relationships between workers and ‘clients’ in both settings, it strikes us as quite
plausible that ‘client’ characteristics have a great influence on the psychic costs
and benefits in the work environment in these industries.
Our work informs the debate over the adequacy of pay in the caregiving
sector by quantifying the relationship of turnover to compensation in addition to
other factors. Our estimates are comparable in magnitude to those found by
Howes (2005). At a nominal wage of $8.85 in 2001, her estimates imply that a $1
(or 12.4 percent) increase in the wage cuts turnover of new workers by 17.6
percent.
Agencies can manipulate job characteristics to affect voluntary worker
turnover. The prudent agency balances the associated costs against the benefits of
reducing turnover (higher productivity, reduced recruiting costs, reduced training
costs, and other costs). Since agencies in the sample tolerate fairly high turnover
rates, these benefits are presumably insufficient to cause them to lower turnover.22
Nevertheless, it may be socially, but not privately, efficient to do so. In other
21 These estimated effects are from the coefficient estimates of -0.376 and -0.295 obtained
applying Fuller’s (a = 1) estimates to models in which total compensation is instrumented with
entry wage and local average wage, respectively.
22 In an early article, Leonard (1987) concludes that the 200 high-technology plants he studies
have an economically justifiable level of turnover. However, this conclusion stems from a finding
of a very small compensation effect on turnover which necessitates enormous (and expensive)
wage increases to induce noticeable turnover declines.
23
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
work (Powers and Powers, 2009), we present estimates of potential cost savings
from cutting turnover by one-third for the sample of agencies analyzed here.23
Four parties potentially benefit from this wage increase—DSP workers, residents
(through quality improvements), agencies (through cost reductions), and the state
(through cost reductions on subsidized items).24 Findings from the same data (see
Powers and Powers, 2009, for a detailed presentation) suggest that,
conservatively, an increase in total compensation sufficient to cut turnover by
one-third would be socially justified without appeal to worker welfare if
residents’ valuation of increased service quality (emanating from reduced worker
turnover) is from 9 to 12 percent of the annual payment made to agencies by the
state for their care (i.e., the state’s declared cost of providing food, shelter, and
supervision, after assumed allowances for client payments, or $39,678 per person;
Powers et al., 2006).25 Our perception is that the upper bound of 12 percent on
resident valuation is reasonably small.26
Findings with regard to the other model variables indicate that policies to
encourage non-profit entry and smaller agencies could also reduce turnover.
Some of the characteristics associated with increased turnover are arguably
encouraged by peculiar features of Illinois’ DD policy (see Powers et al., 2006,
for a detailed discussion). The state’s reimbursement system shifts financial risk
onto providers, increasing incentives to pool risk over large resident bases with
diverse characteristics. The state’s reimbursement schedule also creates
incentives for agencies to offer a variety of compliment services in order to
exploit opportunities for cross-subsidization. Both factors encourage large
agencies, which we have found to be associated with greater turnover. State
policies to reimburse entry-level worker training underwrite one of the potential
major costs of turnover to agencies (see Powers and Powers, 2009, for a detailed
analysis of this issue). The findings presented here also suggest that
reimbursement differentials for extra-special-needs (dually-diagnosed and non-
ambulatory) residents are inadequate to compensate DSP workers for their
additional effort in caring for these individuals. An additional implication of our
23 The choice of a specific level of turnover reduction for these calculations is somewhat arbitrary.
We choose one-third because that is the level of reduction in turnover determined by U.S.
Department of Health and Human Services (2006) to support an adequate caregiving workforce to
the DD sector, which might be perceived as a reasonable policy goal.
24 Efforts to raise direct care workers’ compensation is in the spirit of the ‘Living Wage’
movement (e.g., Adams and Neumark, 2005), which targets government contractors employing
low-wage workers.
25 It is important to point out that the typical client has little income aside from Supplemental
Security Income payments that are already dedicated to living expenses.
26 Alternative estimates for a range of assumptions yield similar magnitudes (see Powers and
Powers, 2009, for details).
24
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
findings is that the secular aging of the population may serve to reduce turnover at
group homes for individuals with DD.
Several specific public policies are promising for reducing turnover. A
federal Medicaid wage pass-through policy applied to the DD sector is one. This
option permits states to increase Medicaid reimbursements to agencies above
current per diems if the increase is dedicated to DSP compensation (see U.S.
Department of Health and Human Services and Institute for the Future of Aging
Services, 2002, for discussion). Enhancing resident choice in the long-term care
system likely would spur agencies to compete on quality, creating incentives for
agencies to reduce their turnover. While Illinois’ current policy of subsidizing
entry-level training implicitly encourages turnover, we do not advocate for its
elimination. Agencies would tend to under-train DSP workers in its absence
because they cannot internalize the benefits of their training to other employers.
Future work to understand low-wage worker turnover, its role in the long-
term care ‘crisis,’ and its implications generally is well served by the availability
of position-specific data at the establishment level for other types of
establishments in the long-term care sector. The present study, while limited to
long-term community-based care for persons with developmental disabilities,
sheds new light on this important topic.
REFERENCES
Adams, Scott, and David Neumark, 2005. “The Effects of Living Wage Laws:
Evidence from the Failed and Derailed Living Wage Campaigns,” Journal of
Urban Economics 58(2, September): 177-202.
Ahlburg, Dennis A., and Christine Brown Mahoney, 1996. “The Effect of Wages
on the Retention of Nurses,” The Canadian Journal of Economics 29(Part 1,
April): S126-S129.
Ball et al. v. Biedess et al., U.S. District Court of Arizona, August 12, 2004.
Cleveland, Gordon H., and Douglas E. Hyatt, 2002. “Child Care Workers’
Wages: New Evidence on Returns to Education, Experience, Job Tenure and
Auspice,” Journal of Population Economics 15: 575-97.
Congressional Research Service, 2006. Long-Term Care: Trends in Public and
Private Spending, CRS Report RL33357, The Library of Congress, April 11.
25
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
Davis, DeWayne, Wendy Fox-Grage, and Shelly Gehshan, 2000. De-
Institutionalization of Persons with Developmental Disabilities: A Technical
Assistance Report for Legislators. National Conference of State Legislators
Item #6683. (http://www.ncsl.org/programs/health/Forum/pub6683.htm).
Falch, Torberg, and Bjarne Strøm, 2005. “Teacher Turnover and Non-Pecuniary
Factors,” Economics of Education Review 24: 611-31.
Freeman, Richard B., 2005. “What Do Unions Do? The 2004 M-Brane
Stringtwister Edition,” Journal of Labor Research 26(4, Fall): 641-68.
Hanushek, Eric A., John F. Kain, and Steven G. Rivkin, 2004. “Why Public
Schools Lose Teachers,” The Journal of Human Resources 39(2): 326-54.
Holmas, Tor Helge, 2002. “Keeping Nurses at Work: A Duration Analysis,”
Health Economics 11: 493-503.
Howes, Candace, 2005. “Living Wages and Retention of Homecare Workers in
San Francisco,” Industrial Relations 44(1, January): 139-63.
Illinois Department of Human Services, 2002. House Resolution 582 Report to
the Illinois General Assembly: A Review of the Community-Integrated Living
Arrangement (CILA) Individual Rate Determination Model: Reimbursement
Rates for Community Agencies Providing Residential Supports to Persons
with Developmental Disabilities Living in Community Group Home Settings
in Illinois.
Institute for the Future of Aging Services, 2007. The Long-Term Care Workforce:
Can the Crisis be Fixed? Problems, Causes and Options. Report to the
National Commission for Quality Long-Term Care. Washington, D.C.
Larson, Sheryl A., Amy S. Hewitt, and K. Charlie Lakin, 2004. “Multiperspective
Analysis of Workforce Challenges and Their Effects on Consumer and
Family Quality of Life,” American Journal on Mental Retardation 109(6,
November): 481-500.
______________ and K. Charlie Lakin, 1999. “Longitudinal Study of
Recruitment and Retention in Small Community Homes Supporting Persons
With Developmental Disabilities,” Mental Retardation 37(4, August): 267-
80.
26
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
Leonard, Jonathan S., 1987. “Carrots and Sticks: Pay, Supervision, and
Turnover,” Journal of Labor Economics 5(4, October): S136-S152.
Mitchell, Dale, and David Braddock, 1994. “Compensation and Turnover of
Direct-Care Staff in Developmental Disabilities Residential Facilities in the
United States. II: Turnover,” Mental Retardation 32(1, February): 34-42.
Moreira, Marcelo J., 2003. “A Conditional Likelihood Test for Structural
Models,” Econometrica 71(4, July): 1027-48.
Murray, Michael P., 2006a. “Avoiding Invalid Instruments and Coping with
Weak Instruments,” Journal of Economics Perspectives 20(4, Fall): 111-32.
_____________, 2006b. “The Bad, the Weak, and the Ugly: Avoiding the Pitfalls
of Instrumental Variables Estimation.” Social Science Research Network
Working Paper No. 843185.
Olmstead v. Zimring, 527 U.S. 581 (1999).
Powell, Irene, Mark Montgomery, and James Cosgrove, 1994. “Compensation
Structure and Establishment Quit and Fire Rates,” Industrial Relations 33(2,
April): 229-48.
Powers, Elizabeth T., and Nicholas J. Powers, 2009. “Who Bears the Costs of
Turnover? Consequences of Frontline Worker Turnover in Community-
Based Residential Care.” Working Paper.
Powers, Elizabeth T., Nicholas J. Powers, and David Merriman, 2006. The
Adequacy of State Payments to Community-Based Agencies for Services
Provided to Illinois Residents with Mental Illness and/or Developmental
Disabilities: Final Report to the Illinois General Assembly Requesters
Pursuant to Public Act 93-842.
Rizzolo, Mary C., Richard Hemp, David Braddock, and Amy Pomeranz-Essley,
2004. The State of the States in Developmental Disabilities. American
Association on Mental Retardation.
Rogerson, Richard, Robert Shimer, and Randall Wright, 2005. “Search-Theoretic
Models of the Labor Market: A Survey,” Journal of Economic Literature
43(4, December): 959-88.
27
Powers and Powers: Causes of Caregiver Turnover
Published by The Berkeley Electronic Press, 2010
Schumacher, Edward J., 1997. “Relative Wages and Exit Behavior among
Registered Nurses.” Journal of Labor Research 18(4, Fall): 581-92.
Seavey, Dorie, 2004. The Cost of Frontline Turnover in Long-Term Care. A
Better Jobs Better Care Practice and Policy Report. Better Jobs Better Care.
Stephens, Brady, and Kim Riley, 2005. Developing Annual Estimates of Hires
and Separations, Office of Survey Methods Research, Bureau of Labor
Statistics, Washington, D.C., August. http://www.bls.gov/osmr/pdf/st050250.pdf.
Stock, James H. and Motohiro Yogo, 2002. “Testing for Weak Instruments in
Linear IV Regression.” NBER Technical Working Paper 284. Cambridge,
MA: National Bureau of Economic Research. (Revised January 2004 version
posted at http://ksghome.harvard.edu/~jstock/ams/websupp/rfa_7.pdf)
Stone, Robyn I., with Joshua M. Wiener, 2001. Who Will Care for Us?
Addressing the Long-Term Care Workforce Crisis. Washington, D.C.: The
Urban Institute. http://www.urban.org/url.cfm?ID=310304.
Test, David W., Claudia Flowers, Amy Hewitt, and Jill Solow, 2003. “Statewide
Study of the Direct Support Staff Workforce,” Mental Retardation 41(4,
August): 276-85.
U.S. Department of Health and Human Services, 2006. The Supply of Direct
Support Professionals Serving Individuals with Intellectual Disabilities and
Other Developmental Disabilities: Report to Congress. Office of the
Assistant Secretary for Planning and Evaluation, Office of Disability, Aging
and Long-Term Care Policy.
U.S. Department of Health and Human Services and Institute for the Future of
Aging Services, 2002. State Wage Pass-Through Legislation: An Analysis.
Workforce Issues No. 1. Washington D.C.: U.S. Department of Health and
Human Services and Institute for the Future of Aging Services.
Wai Chi Tai, Teresa, Sherry I. Bame, and Chester D. Robinson, 1998. “Review of
Nursing Turnover Research, 1977-1996,” Social Science and Medicine
47(12): 1905-24.
28
The B.E. Journal of Economic Analysis & Policy, Vol. 10 [2010], Iss. 1 (Contributions), Art. 5
http://www.bepress.com/bejeap/vol10/iss1/art5
... Past research has indicated that wage increases are effective at reducing PSW turnover (Dawson, 2007;Howes, 2005; Office of the Governor of Wyoming, 2005;Powers & Powers, 2010;Smith & Baughman, 2007). Increased investment in HCC PSW wages has the potential to help stabilize the HCC workforce and enable the growth of HCC capacity (Jabola-Carolus, Luce, & Milkman, 2021). ...
... Multiple studies have quantified the impact of real-world wage increases on PSW retention. To estimate the impact of wage parity for Ontario HCC PSWs, retention rates from these real-world studies can be scaled to the 26 per cent ($4.98/hr) increase in wages required to reach wage parity (Dawson, 2007;Howes, 2005; Office of the Governor of Wyoming, 2005;Powers & Powers, 2010;Smith & Baughman, 2007). For this 26 per cent wage increase, predicted gains in retention are between 5 per cent (Smith & Baughman, 2007) and 33 per cent (Powers & Powers, 2010), with an average retention rate of 21 per cent across five studies (Dawson, 2007;Howes, 2005; Office of the Governor of Wyoming, 2005;Powers & Powers, 2010;Smith & Baughman, 2007). ...
... To estimate the impact of wage parity for Ontario HCC PSWs, retention rates from these real-world studies can be scaled to the 26 per cent ($4.98/hr) increase in wages required to reach wage parity (Dawson, 2007;Howes, 2005; Office of the Governor of Wyoming, 2005;Powers & Powers, 2010;Smith & Baughman, 2007). For this 26 per cent wage increase, predicted gains in retention are between 5 per cent (Smith & Baughman, 2007) and 33 per cent (Powers & Powers, 2010), with an average retention rate of 21 per cent across five studies (Dawson, 2007;Howes, 2005; Office of the Governor of Wyoming, 2005;Powers & Powers, 2010;Smith & Baughman, 2007). ...
Article
Full-text available
Most older adults prefer to age in place, which for many will require home and community care (HCC) support. Unfortunately, HCC capacity is insufficient to meet demand due in part to low wages, particularly for personal support workers (PSWs) who provide the majority of paid care. Using Ontario as a case study, this paper estimates the cost and capacity impacts of implementing wage parity between PSWs employed in HCC and institutional long-term care (ILTC). Specifically, we consider the cost of increased HCC PSW wages versus expected savings from avoiding unnecessary ILTC placement for those accommodated by HCC capacity growth. The expected increase in HCC PSW retention would create HCC capacity for approximately 160,000 people, reduce annual health system costs by approximately $7 billion, and provide an 88 per cent return on investment. Updating wage structures to reduce turnover and enable HCC capacity growth is a cost-efficient option for expanding health system capacity.
... That staff development and support would help to improve their collective understanding and increase the nursing staff's sense of honor, thus improving their care. Powers et al. (2010) [47] found that senior caregivers were dissatisfied with the level of support they received from managers and society in the nursing facility, which made them not develop a sense of honor for their work and led to their high turnover rate. Therefore, a sense of honor for the caregivers' work is also essential. ...
... That staff development and support would help to improve their collective understanding and increase the nursing staff's sense of honor, thus improving their care. Powers et al. (2010) [47] found that senior caregivers were dissatisfied with the level of support they received from managers and society in the nursing facility, which made them not develop a sense of honor for their work and led to their high turnover rate. Therefore, a sense of honor for the caregivers' work is also essential. ...
... Nursing workers care for the senior population with various physical or psychological illnesses in senior care institutions, so they could have this sense of good service and love for their work to serve the senior population well. This conclusion is different from some conclusions of others [47,48], who believe that the sense of honor is also a vital component indicator. Zhao and Liu (2021) [49] believe that nurses are mostly senior women with low psychological resistance, insufficient physical strength, and poor learning and acceptance of new things. ...
Article
Full-text available
As the problem of the aging population becomes more and more serious, building an intelligent senior care service model and optimizing the senior care service industry become key to the development of the senior care service industry. The key to developing intelligent senior care services is to improve the overall senior care personnel quality and construct a competency model of intelligent institutional senior care professionals. This study used literature research and interviews to establish 31 relevant institutional senior care professional talent competency elements. We proposed six research propositions, prepared questionnaires for empirical analysis, and took caregivers of senior care institutions implementing intelligent management in some cities in Hebei Province, China as samples. This study established and validated 28 competency quality index models of senior care professionals in intelligent institutions in four dimensions: nursing knowledge, professional ability, personal quality, and professional attitude through exploratory factor analysis and confirmatory factor analysis. Based on the index system, this study suggests three aspects: improving the talent recruitment and selection mechanism, talent training and development mechanism, and assessment and incentive mechanism. The traditional talent competency model only focuses on fundamental aspects, such as competence. This study comprehensively establishes an evaluation model from four aspects, providing theoretical and practical significance for selecting and developing talents in intelligent institutions.
... It is important to note that unobserved heterogeneity has not been accounted for in the above studies, but has been found to bias wage effects towards zero in studies on other sectors (Manning, 2003). One of the few studies that took into account the endogeneity of wages in the LTC context is Powers & Powers (2010). Using data from 61 community-based care sites in Illinois and instrumental variable methods to identify the effect of wages on employer-level turnover rates, they found a negative effect and a significant downward bias when not controlling for unobserved factors. ...
... Not controlling for unobserved heterogeneity can lead to either upward or downward biased results. As mentioned earlier, omitted variable bias has been found to bias wage effects on turnover or job separations towards zero (Baughman & Smith, 2012;Manning, 2003;Powers & Powers, 2010). ...
Article
Full-text available
Background and Objectives Most job leavers in the long-term care (LTC) sector in England do not leave the sector, but rather move to other LTC employers. Nevertheless, the high ‘churn’ can have a negative impact on continuity and quality of care, care providers’ recruitment and training costs, and the remaining staff workload and motivation. This study aimed to provide quantitative evidence on the drivers of direct care workers’ job separation in England, with a focus on job quality. Research Design and Methods We used yearly data (2016-2019) from the Adult Social Care Workforce Data Set (ASC-WDS), the leading source of LTC workforce data in England and including information on both LTC workers and employers. The analysis considered panel data econometric methods that accounted for unobserved heterogeneity at worker and employer level. Results After controlling for observed individual, organizational, and local market characteristics as well as unobserved worker and employer heterogeneity, we found that, everything else being equal, wages and employment conditions (i.e., full-time contracts and contracts with guaranteed working hours) significantly reduce job separation. For example, a 10% wage increase from the sample mean would reduce the job separation rate by about 3 percentage points. This wage effect was more than halved (i.e., downward biased) when not accounting for unobserved effects. Discussion and Implications The persistent high staff turnover in LTC in England highlights the need for finding practical solutions faced by care providers and policymakers. Our findings showed that improving pay and employment conditions can be the way forward while methodologically stressed the importance of accounting for unobserved variable bias.
... Achieving wage parity between HCC and ILTC would require a funding increase of $6.23/hour for PSWs in HCC, including a 25% allowance for benefits. Wage parity would improve the relative attractiveness of working in HCC, create muchneeded stability in the sector and enable increased capacity to provide HCC (Powers and Powers 2010;Zagrodney 2022). From an ethical and pay equity standpoint, it is concerning that the historically unrecognized and poorly valued HCC PSW workforce includes a higher proportion of women of older age and from visible minority groups, compared to other better-compensated sectors (Laporte et al. 2020). ...
Article
The home and community care (HCC) sector is in a health human resource crisis. Particularly concerning is the shortage of personal support workers (PSWs) who provide the majority of HCC. This paper outlines a strategy to mitigate the HCC PSW shortage by applying appropriate funding to HCC and focusing on equal pay between HCC and institutional long-term care facilities' PSWs. Using publicly available data, our calculations estimate substantial government cost-savings from investing in HCC PSWs to increase HCC capacity. Beyond the economic evidence, how such investments would benefit those seeking care are also highlighted.
... Medicaid wage pass-through policies) found they improved staff wages (Baughman & Smith, 2010) and employment (Feng et al., 2010;Foster & Lee, 2015;Wu et al., 2021). Moreover, higher wages have been shown to improve LTC staff retention (Baughman & Smith, 2012;Gleason & Miller, 2021;Kennedy et al., 2021;Powers & Powers, 2010;Vadean & Saloniki, 2023;Wiener, 2009). ...
Article
Full-text available
Background and objectives: In many countries, a large proportion of long-term care staff are paid at, or near, minimum wage, leading to concerns of negative effects on care outcomes. This study analysed the effect of staff wages on care home quality ratings in England. Research design and methods: A national staffing database of long-term care providers was matched with local area information on needs and supply to construct a three-year panel (2016-2018) of English care home observations. Using multiple imputation methods to address missing data provided a dataset of 12,055 observations of 5,556 care facilities (both residential and nursing homes). We analysed the effect of facility-level average hourly wage of care staff on national regulator quality ratings. A measure of impact of exogenous changes in national minimum wage on care facilities was used as an instrument for wage. Results: We find that wages positively impact on care home quality ratings. Other things equal, a 10% increase in average hourly wage of direct care workers would lead to a 7.1% increase in the likelihood that a care home will have a high quality rating. The wage effect on quality was significant when controlling for staff skill mix, i.e., share of registered nurses in nursing home staff. Discussion and implications: This study provides important evidence on the positive impact that staff pay can have on the quality of long-term care. Our finding has important implications about appropriate levels of pay and the funding of long-term care.
... Reported predictors of DSW turnover are low wages, a lack of benefits such as paid annual leave, high client support needs, a lack of opportunities for promotion, and inexperienced supervisors overseeing their practice (Dowler et al., 2011;Gaventa, 2008;Lakin et al., 2003;Larson, Hewitt, & Lakin, 2004;Larson & Lakin, 1999;Misra, Walls, & Orslene, 2010;Powers & Powers, 2010). ...
Article
Direct support professionals (DSPs) provide a range of supports in a variety of settings to people with intellectual and developmental disabilities (IDD) who count on these supports to live, work, and contribute in their communities. Despite this, high annual DSP turnover rates are problematic. DSP turnover is disruptive to people who receive supports as the lack of stable, reliable supports can negatively impact their important day-to-day outcomes (e.g., safety, community participation, and choice). Turnover also comes at a cost to provider organization in the hiring and training of new employees. To retain DSPs, organizations offer incentives (e.g., bonuses, retirement plans, health insurance). This study utilized National Core Indicators® (NCI®) Staff Stability Survey 2018 data to examine the relationships between wages, different types of incentives, including benefits (e.g., paid time off, access to health insurance, disability insurance, wage bonuses, health incentives programs, etc.) to annual turnover in participating states in the United States. Results indicated that incentives were not positively associated with DSP retention. Staff wages were the most notable factor associated with differences in DSP retention rates, along with the state in which the organization was located as well as organization vacancy rates.
Article
Full-text available
Background: :In 2014, Ontario increased its "minimum wage" for personal support workers (PSWs) in publicly funded home care. Objective: The objective of this article is to determine the short-term results of this policy for home care PSWs' wages, hours and job stability. Methods: This study uses descriptive graphs and ordinary least squares and unconditional quantile regressions, using PSWs across Canada as comparison groups. Results: Pre-policy nominal wages for Ontario home care PSWs stagnated, whereas real wages declined. The policy increased home care PSWs' wages without noticeably affecting hours or job stability. However, wages were already increasing for low-wage home care workers in the rest of Canada. Conclusions: Ontario exercises monopsony power in the home care market and, before the wage increase, kept nominal wages stable compared to rising real and nominal wages in the rest of Canada. This PSW-specific wage increase did not represent a drastic change relative to market conditions.
Article
Full-text available
This study records the impact on workforce retention of the nearly doubling of wages for homecare workers in San Francisco County over a 52-month period. Using descriptive statistics and logistic regression analysis the author finds that the annual retention rate of new providers rose from 39 percent to 74 percent following significant wage and benefit increases and that a 1increaseinthewageratefrom1 increase in the wage rate from 8 an hour - the national average wage for homecare - would increase retention by 17 percentage points. The author also shows that adding health insurance increases the retention rate by 21 percentage points.
Article
Full-text available
Many school districts experience difficulties attracting and retaining teachers, and the impending retirement of a substantial fraction of public school teachers raises the specter of severe shortages in some public schools. Schools in urban areas serving economically disadvantaged and minority students appear particularly vulnerable. This paper investigates those factors that affect the probabilities that teachers switch schools or exit the public schools entirely. The results indicate that teacher mobility is much more strongly related to characteristics of the students, particularly race and achievement, than to salary, although salary exerts a modest impact once compensating differentials are taken into account.
Article
Despite the claims of efficiency wage theory, there is surprisingly little direct evidence that firms that pay higher wages have lower turnover rates than otherwise similar firms. This study uses data on 205 child care establishments to examine the influence of wages, fringe benefits, and the dispersion of wages within a skill class on establishment quit and fire rates. We separately examine one high-skill group (teachers) and one lowskill group (teacher aides). While we find that wages (alone) significantly reduce establishment quit and fire rates, the effect seems too small to be consistent with the efficiency wage hypothesis.
Article
Instrumental variables estimation can, in principle, avoid biases that ordinary least squares estimation suffers when explanatory variables are correlated with the disturbances. Finding appropriate instruments is a challenge. This paper uses seven recently published empirical papers to illustrate exemplary practices in IV estimation. Nine strategies for avoiding bad instruments (those correlated with the disturbances), as well as recently developed best practices for coping with weak instruments (those little correlated with the troublesome explanatory variable), are summarized and illustrated. The ugly interpretive perils posed by heterogeneity in agents' behavioral responses to a troublesome explanator are also described and illustrated. All procedures recommended in the paper can be implemented using existing commands (some of them quite newly constructed) for one or more standard econometric packages.
Article
Living wage campaigns have succeeded in about 100 jurisdictions in the United States but have also been unsuccessful in numerous cities. These unsuccessful campaigns provide a better control group or counterfactual for estimating the effects of living wage laws than the broader set of all cities without a law, and also permit the separate estimation of the effects of living wage laws and living wage campaigns. We find that living wage laws raise wages of low-wage workers but reduce employment among the least-skilled, especially when the laws cover business assistance recipients or are accompanied by similar laws in nearby cities.
Article
This paper studies teacher mobility using matched employee–employer panel data from Norwegian primary and lower secondary schools. The Norwegian institutional setup with completely centralized wage setting for teachers is ideal to analyze the effect of non-pecuniary job attributes on quit decisions. We find that teachers tend to leave schools with high share of minority students and high share of students with special needs. In addition, the composition of teachers and the school size affect the propensity to quit. These results are robust across different econometric specifications and sub-samples.