Search tips
Search criteria 


Logo of hsresearchLink to Publisher's site
Health Serv Res. 2002 April; 37(2): 397–415.
PMCID: PMC1430364

Estimating Paid and Unpaid Hours of Personal Assistance Services in Activities of Daily Living Provided to Adults Living at Home



To estimate the total hours of paid and unpaid personal assistance of daily living provided to adults living at home in the United States using nationally representative household survey data.

Data Sources

The Disability Followback Survey of the National Health Interview Survey on Disability (NHIS-D) conducted from 1994 to 1997.

Data Collection/Extraction Methods

Data were obtained on persons receiving help with up to 5 ADLs and 10 IADLs, for up to 4 helpers, including the activities they helped with, whether the helper was paid or not, and the number of hours of help provided in the two weeks prior to the survey. The sample consists of 8,471 household-resident adults ages 18 and older receiving help with personal assistance. About 22 percent of the sample has missing data on hours, which we impute by multiple regression models using demographic, ADL, and IADL variables.


We estimate that 13.2 million noninstitutionalized adults receive an average of 31.4 hours per week of personal assistance in ADLs and IADLs per week, with 3.2 million people receiving an average of 17.6 hours of paid help and 11.7 million receiving an average of 30.7 hours of unpaid help. More persons ages 18–64 received help than those ages 65 and older (6.9 versus 6.2 million), but working-age recipients had fewer hours (27.4 versus 35.9) per week, due in part to less severe levels of disability.


Personal assistance provided to adults with disabilities amounts to 21.5 billion hours of help per year, with an economic value in 1996 approaching 200billion.Only16percentofthistotalispaid,representing32 billion in home health services spent annually. This study, the first to estimate hours of assistance for both working-age and older adults, documents that older persons are more likely to receive paid personal assistance, while working-age people rely to a greater extent on unpaid help. This study begins to articulate the division of labor in the provision of personal assistance. Estimates of paid and unpaid hours of help by number of ADLs should inform policy concerning eligibility boundaries in long term care.

Keywords: Personal assistance services, long term care, hours of help, paid help, informal help, elderly, working-age, imputation, NHIS-D

Personal assistance services (PAS)1 with activities of daily living (ADLs) and instrumental ADLs (IADLs) are increasingly recognized as important for enabling many millions of individuals with disabilities of all ages to live in their own homes. The steady aging of the population and a strong preference for home-based services among both working-age and older persons with disabilities have likely increased the demand for PAS. However,despite the enormity of PAS as a health and welfare concern,the precise number of persons who receive such services and the hours of PAS provided to individuals living at home is not known. Existing studies are piecemeal and seldom include the full range of informal (unpaid) and formal (paid) help. Moreover, existing data pertain only to the older population, overlooking several million working-age people with disabilities who receive help in ADLs and IADLs (Kennedy and LaPlante 1997). This situation presents an obstacle to estimating the need and cost of PAS in the United States (U.S. General Accounting Office 1999).

Estimates of PAS hours vary by study, which have invariably excluded working-age recipients. An analysis of the 1996 National Survey of Caregivers, a small (n = 1,509) telephone survey, estimated that there were approximately 22-million U.S. households (23 percent of total households) providing an average of 17.9 hours per caregiver of informal help to individuals at least 50 years old (National Alliance for Caregiving 1997). Because most individuals have multiple caregivers, the average hours per recipient is considerably higher than this estimate.

Arno, Levine, and Memmott (1999) were interested in estimating the total hours of informal care giving nationally. They used data from two older national surveys (the 1986 Survey of Income and Program Participation and the 1987 and 1988 National Survey of Families and Households) yielding the number of helpers providing personal assistance, which they extrapolated to 1997 assuming that the ratio of helpers to the total population remained constant. The estimate of the number of helpers was multiplied by the estimated 17.9 hours from the 1996 National Survey of Caregivers to yield an estimated 22–26 billion hours of informal help.

A 1984–1985 study of helpers of older individuals (ages 70 and older) living in their own homes in Massachusetts found that they received an average of 34.6 hours per week of informal help (Crawford, Tennstedt, and McKinlay 1995). Boaz (1996) used the 1989 National Long Term Care Survey and the companion National Informal Caregiver Survey, and estimated that 36 hours of unpaid care giving were provided to older persons with ADL limitations on a weekly basis in 1989. Liu, Manton, and Aragon (2000), using 1994 data from the National Long Term Care Survey, estimated an average of 21.6 total hours per week of paid and unpaid help for older persons with any ADL or IADL assistance needs, including 12.6 hours of paid help per week.

In this study, we provide a statistically precise estimate of the total hours of paid and unpaid assistance provided to all adults living at home, whether or not they are older. We also measure differences in hours by age. We used data from the Disability Followback Survey (DFS), Phase II of the National Health Interview Survey on Disability (NHIS-D). Conducted between 1994 and 1997, the DFS provides the most recent national survey data covering all adults. Another aim of the study was to investigate what factors predict hours of help.



The NHIS is a nationally representative survey of U.S. households conducted annually by the Census Bureau for the National Center for Health Statistics. Re-spondents to the l994 and l995 NHIS also took part in two supplemental surveys, known collectively as the NHIS on Disability (NHIS-D). Phase I of the NHIS-D was designed to gather detailed data on the U.S. population with disabilities living in the community. Administered at the same time as the NHIS core, Phase I screened 126,704 adults for any indication of disability, using an extensive set of criteria, including functional limitation, specific disabling diagnoses, perception of disability, and use of disability-related services. Individuals with disabilities were later reinterviewed in Phase II for the DFS; this sample of 25,805 adults was used for estimating hours of help. The core response rate was 94 percent, and the Phase I and Phase II response rates were 92 and 89 percent, respectively. The sample design was changed in 1995 to include more Primary Sampling Units and oversampling for Hispanics. Sampling weights were used in all estimation procedures. The Phase II weight incorporates the final core survey weight with additional corrections for nonresponse in each wave.

Survey Questions

In Phase II of the NHIS-D, adults with disabilities were asked if they had difficulty with any of seven ADLs and eight IADLs (we moved walking and getting outside from the ADL to the IADL list) and whether they received help or needed help from another person, whether hands on or supervisory. In addition, if they received such help, they were asked who helped with these activities. For each of up to four helpers, the respondent was asked which of the 15 activities the helper helped with, the relationship of the helper, whether the helper was paid and the sources for any payments, and the number of days and the average hours per day that each helper provided assistance during the past 2 weeks. We constructed a variable for the total hours of help received for each respondent by multiplying the number of days that help was received by the average hours per day and summing this over all helpers.

Imputation for Missing Data

We observed that 22 percent of those who said they were helped in the past 2weeks did not specify either the number of days that they got help or the number of hours per day that they got help. Missing data on hours per day were more frequent than on the number of days per week.

The missing data on hours of help were nonrandom, which could produce less efficient estimates as well as biased estimates of the population parameters under study. Moreover, to produce an estimate of the total hours of help in the adult population, the missing data had to be imputed. Following Crawford, Tennstedt, and McKinlay (1995), we considered four imputation methods to handle the missing data. These methods are complete-case analysis, mean imputation, single regression model imputation, and multiple regression model imputation.

In complete-case analysis, records with missing data are ignored. In mean imputation, the average value of hours of help for nonmissing respondents is the imputed estimate. In single-regression model imputation, the predicted value from a linear regression model is the imputed estimate. However, both of these procedures treat imputed values as if they are estimated with certainty and have no error. The last technique, multiple-regression model imputation, introduces a degree of uncertainty for each imputed estimate by adding a random component to the regression parameters (coefficients and regression variance) based on their estimated variances and replicating the imputation up to m times. Each missing value is replaced with the average of m acceptable values, representing a distribution of possibilities based on the variability of the complete cases. We employed m = 10, within the range that Rubin (1987) suggested.

We estimated three linear regression models for three outcome variables covering the 2 weeks prior to the survey interview: (1) hours provided by all paid helpers, (2) hours provided by a primary unpaid helper, and (3) hours provided by secondary unpaid helpers. The models can be expressed as follows:

equation image

where Y= log (hours) for paid helper(s), primary unpaid helper, secondary unpaid helper(s); X = vector of explanatory predisposing, enabling, and need variables; and ei=error term.

The specification of explanatory variables is provided in Table 1 along with their means and standard deviations and that of the dependent variables. The predictive variables for the models were categorized into predisposing, enabling, and need factors (Andersen 1995) and included the following: The predisposing factors were age, female, non-White race and education; the enabling factors included living alone, income, Medicare coverage, Medicaid coverage, living in an urban area, the number of paid helpers, and the number of unpaid helpers. The need factors were 5 ADLs (bathing, dressing, transferring, toileting, and eating) and 10 IADLs (such as grocery shopping, walking, meal preparation, light housekeeping) specified as (0,1) dummy variables. These variables indicate, for each of the three sources of help, whether the respondent was helped in that activity. A significant effect means that help with that particular activity increased or decreased total hours of help received by that class of helper. The model measures the independent effects of these activities that are generally correlated. These explanatory variables were selected based on previous studies of formal and informal long-term care.

Table 1
Descriptive Statistics on Hours of Personal Assistance and for Explanatory Variables Used to Predict Hours Among Populations with Any Helper(s), Paid Helper(s), a Primary Unpaid Helper, and Secondary Unpaid Helper(s)

The imputation procedure is described as follows:

1. Estimate a linear regression model using complete data cases only. Let   ttacca and

equation image

be the estimated regression coefficient vector and the estimated regression variance in complete case analysis (cca).

2. Draw a χnp2 random variable k*, and let

equation image

where n is the number of complete data cases, and p is the number of regression coefficients.

3. Add a number randomly drawn from N(0,1),zj*, to the jth regression coefficient,

equation image
equation image

where scca(j) is the standard error of the jth regression coefficient from step1.

4. A new imputed value for missing data is computed as

equation image

where zi* is a random number drawn from N(0,1), and n1 is the number of missing data.

5. Repeat steps 2 through 4 m times. The final imputed value

equation image

for missing data Yi is computed as

equation image

As the sample distributions of paid hours and unpaid hours were skewed toward high values, we fitted the regression models on the logarithm (natural) of hours. Missing values imputed on the logarithmictransformed scale were retransformed to the original scale using a nonparametric method in order to achieve unbiased and consistent quantities on the original scale (Duan 1983). The final results presented in this article are based on the multiple regression model imputation. Empirically, the multiple regression model imputation provided a statistically more satisfactory set of results, with the outcome variable being more normally distributed than the other methods. Further details of the multiple regression model imputation are given in Rubin (1987) and in Crawford, Tennstedt, and McKinlay (1995).


We discovered that the hours for the primary unpaid helper (the first-listed unpaid helper if more than one was mentioned) had to be estimated separately from the hours for the secondary unpaid helpers because the predictors were different. Moreover, we found that different factors were associated with paid versus unpaid help. Approximately 17 percent of paid hours were missing, as were 27 percent of unpaid hours provided by the primary helper, and 21 percent of unpaid hours provided by the secondary helpers. These three models individually performed better than a model that predicted total hours. To facilitate comparison across the three outcome variables, we included the same predictors in all three models, deleting those variables that were not significant in any. Medicaid coverage was included as a complement to Medicare coverage, despite lack of significance. The complete set of ADL and IADL variables was included for comparison, although transferring, toileting, and managing money were not significant in any model.

Variance Estimation

The NHIS-D employed a complex (i.e., nonrandom) sample design that involved multiple stages of selection with a high degree of stratification and clustering. Because of this sampling design, individual responses are correlated (Lee, Forthover, and Lorimor 1989). We used SUDAAN software to adjust for this nonrandomness for both point and regression estimates using Taylor linearization.


The sample included 8,471 individuals representing a total population of 13.2-million persons who needed or received help with ADLs/IADLs in the 2 weeks prior to being interviewed (Table 1).2 As expected, most people received informal or unpaid help. Help from one or more paid helpers was received by 3.2-million persons (23.9 percent); 11.7-million persons received help from a primary unpaid helper (88.5 percent), which tended to be a spouse or often a parent, and 4.0 million (30.1 percent) received additional help from one or more secondary unpaid helpers, often children and friends. Of all recipients of help, 65 percent were female, and 17 percent were non-White. The average age was 60.6 years—6.9-million working-age persons (18–64 years) and 6.2-million older persons (65 years and over). The average education was just under completion of the eleventh grade. Almost a quarter (24 percent) lived alone. Sixty-two percent had Medicare. Twenty-six percent had Medicaid, and 73 percent lived in urban areas. On average, recipients had 1.6 helpers, 0.3 paid and 1.3 unpaid. Those who received paid help were 69.5 years old—almost 10 years older than those who received unpaid help—and they were more likely to live alone, have higher incomes, to be on Medicare, to be helped with ADLs, and to be helped with housework. People who got help from one or more secondary unpaid helpers were less likely to be helped with bathing and dressing than those who got help from a primary unpaid helper but were about as likely to be helped with other ADLs and IADLs.

In order to impute missing hours, we estimated linear prediction models for the three sources of hours of help. Table 2 shows the estimated multiple regression imputation models for the log of paid hours of help and hours provided by unpaid primary and secondary helpers. The model for paid hours accounted for 44.3 percent of the variation, adjusted for the number of predictor variables. Being non-White, living in an urban area, and the number of paid caregivers were positive predictors of paid hours of care, whereas the number of unpaid helpers was a significant negative predictor. Thus, the greater the number of informal helpers a person has, the lower the hours that are provided by paid helpers—the more paid helpers, the more paid hours. With respect to the need factors, these models predict the number of hours that each individual received based on the activities that the particular class of helper(s) actually helped with. The coefficients represent the independent effects of each activity. Bathing was the only ADL factor that was a significant predictor of paid hours, in the positive direction. Six IADLs were significant positive predictors of paid hours, with the strongest factor being help in preparing meals. One IADL—receiving heavy housework—was a negative predictor of paid hours of care. Heavy housework is the most common activity in which people get help, and it represents the group with the lowest level of disability.

Table 2
Estimated Weighted Regression Models for the Log Hours of Personal Assistance Services Among Populations with Paid Helper(s), a Primary Unpaid Helper, and Secondary Unpaid Helper(s)

The model for hours provided by the primary unpaid helper was different from the paid helper model. Non-White persons had higher unpaid hours, as they did for paid hours, and the number of unpaid caregivers had a negative association with hours received from the primary unpaid helper. Thus, the greater the number of informal helpers that a person has, the lower the hours that are provided by the primary informal helper. Unlike paid hours, the number of paid helpers was not associated with informal hours. As with paid hours, age had no impact. However, being female, having higher levels of education, living alone, and having a higher monthly income were negative predictors of unpaid hours, and Medicare was a positive predictor, although none of these variables were significant predictors of paid hours. As with paid hours, getting help with bathing was significant. Dressing and eating had no effect on paid hours but were associated with greater unpaid hours. Walking and getting outside were two IADLs that increased unpaid hours, but not paid hours. As with paid hours, preparing meals had the largest effect on unpaid hours. This model explained 51 percent of the variance.

The model for secondary unpaid hours was different from the model for primary unpaid helper hours. Being non-White was more strongly significant. Older persons received fewer hours from secondary unpaid helpers. This is the only model in which age had any effect on hours. People with higher education and those who lived alone received fewer secondary hours, whereas the number of unpaid helpers was a positive factor. Bathing was a significant positive factor, but unlike for primary helper hours, dressing and eating were not significant. Preparing meals and walking were the strongest IADL factors, followed by light housework, managing medication, and getting outside. The model for unpaid secondary hours explained slightly less variance than the other two models (40 percent). In both unpaid models, living alone had a very large impact with an absolute effect as large as that of help with meal preparation, the strongest of the need (ADL/IADL) predictors. Walking and getting outside increased hours of unpaid help, especially among secondary unpaid helpers.

The final results of the imputation on the estimated total hours, which is the sum of the hours over the three populations, are shown in Table 3. The mean hours per person were 30.9 for the known cases and 31.4 for the multiple regression model imputation method. The greatest difference was for paid hours. The multiple regression model imputation method produced an estimate of 17.6 hours compared with 19.1 for the known cases. This resulted because, of those who received paid services, the missing data were more likely among those with less disability, which the regression model adjusts for.

Table 3
Estimated Means and Standard Deviations of Hours of Assistance in ADLs/IADLs for Respondents with Known Hours and Including Imputed Hours

Table 4 shows the total paid and unpaid hours, including the imputed data, for all persons receiving help. Those who received paid help got 17.6 hours of paid help, and those who received unpaid help got 30.7 hours of unpaid help per week. Thus, although the minority of recipients got paid help, they also got lower hours of paid help than unpaid help.

Table 4
Estimated Paid, Unpaid, and Total Hours of Personal Assistance Services Provided to Adults Living at Home

Those who get help with any ADLs average 57.0 hours of help per week. The 8.3-million people who do not receive help with ADLs but received help with any of the IADLs received just 16.3 hours per week. The total hours of help increased from those needing help in bathing receiving an average of 63.6 hours per week to those needing help in eating receiving 106.7 hours. Unpaid hours increase more with the number of ADLs and IADLs a person gets help with than do paid hours. Of persons getting help with five ADLs, those who get unpaid help get 108.7 unpaid hours, and those who get paid help get 50.8 paid hours.

More than 9-million individuals received help with heavy housework, but these individuals received the fewest hours (30.5 hours per week) compared with the other IADLs.

The fewest number of persons (1.3 million) received help with using the telephone, but they received an average of 87.2 total hours of help per week, the highest number of hours of the IADLs.


This is the first study to estimate the total hours of PAS provided to adults living in the community. Based on a large national survey covering all adults 18 years and older, we found that 13.2-million adults received help in ADL/IADLs, averaging 31.4 hours per week. This amounts to 21.5 billion hours of PAS help per year, of which 13.4 percent is paid and the remainder is unpaid. If paid hours are valued at the wage rate of the average home health worker at $11 per hour in 1996, the market value of home health services would be roughly $32 billion a year. This estimate is identical to another independent estimate of national home health expenditures (Levit et al. 1997). This study estimated the total hours of informal help at 18.7-billion hours, somewhat less than the 22- to 26-billion estimated by Arno et al., who relied on the average number of hours for family caregivers taken from one survey and multiplied by the number of helpers from two other surveys, projected into the future. The advantage of this study is that we estimate this number directly from a large national survey. The market value of informal PAS is approximately $166 billion if the 18.7-billion hours are likewise valued at $9 per hour, which is above the minimum wage but less than the wage of a home health worker.

Moreover, we can disaggregate these estimates by age. Although working-age (18–64 years old) PAS recipients outnumber older recipients (6.9 vs. 6.2 million), they receive fewer hours of help (27.4 vs. 35.9 hours per recipient per week). This results in a total of 9.9-billion hours provided to working-age people with disabilities compared with 11.6 billion provided to the older population. Although the hours of paid help are the same for working-age and older recipients, the majority (70 percent) of recipients of paid help are older. Thus, of the $32 billion in paid help provided annually, $9.3 billion is provided to working-age people with disabilities and $23.7 billion to older people with disabilities.

Although it is beyond the scope of this article to assess fully the potential disparities in PAS by age, we note that the mean number of ADL/IADLs was somewhat higher for older recipients (mean of 4.67 vs. 3.81). Because hours of care increase steadily with the number of ADL/IADL dependencies, a crude adjustment of hours for differences in ADL/IADL dependencies can be made by simply dividing the mean hours by the mean number of ADL/IADLs. This reveals that working-age recipients received 7.2 hours per ADL/IADL compared with 7.7 hours for older recipients. Alternatively, if we adjust by the mean number of ADLs in the two age groups (0.8 for 18–64, 1.05 for 65+), we obtain 35.3 and 34.2 hours per ADL limitation. Thus, the lower mean hours among working-age PAS recipients is partly due to having fewer ADL/IADL dependencies. In any case, this study clearly shows that health and welfare policy must comprehend that long-term care is, at this point in time, much more than a problem of just the older population.

A recent study by Liu, Manton, and Aragon (2000), based on the National Long Term Care Survey, estimated that 4.0-million older persons (65 years and older) received a total of 25.4 hours3 of help per week, including 12.6 hours of paid help, in 1994. We obtain both a higher number of older recipients (6.2 million) and higher average hours per recipient (35.9 hours total, 17.9 hours of paid help). One difference is that the National Long Term Care Survey used a 1-week reference period, whereas the NHIS-D used a 2-week reference period, which is likely to capture a greater number of recipients because helping is to some degree episodic. Also, data on recipients and hours were obtained in Phase II of the NHIS-D, which occurred from 7 months to up to 26 months (median of 13.6 months) after individuals were first screened in on the basis of any level of disability. The number of recipients getting help with ADL/IADLs was higher in Phase II than in Phase I. It is possible that individuals previously screened would have become more disabled as time progressed, which would tend to increase the number of recipients and possibly the average hours of help. However, we found that the number of individuals was higher in Phase II, regardless of the length of time that expired between interviews. It is possible that the DFS questionnaire and procedures identified more recipients and hours. Although these differences in the older population are important, they require additional analyses to isolate the reasons why. Nevertheless, the findings we report are within the range identified in the literature.

We found that the need factors (ADLs and IADLs) were the most important predictors of hours of help. In all the models examining paid and unpaid hours of help, bathing was the most important ADL measure in predicting hours, controlling for other ADLs and IADLs for which help was received. Controlling for disability and other factors, age was of little importance. However, living alone had a substantial negative impact on unpaid hours, even adjusted for the number of helpers, and rivaled the effects of the strongest need factors. This suggests that informal providers fail to provide as many hours of help to people who live alone, independent of their disability needs. All of the models explained a high degree of the variation in hours, from 40 to 51 percent.

This study offers some evidence that informal help substitutes for formal help. The greater the number of informal helpers that are available, the lower the hours that are provided by paid helpers. The number of paid helpers had no impact on informal hours, however. The number of unpaid caregivers also reduced the amount of care provided by the first helper. The presence of additional informal caregivers, often children, reduces the hours provided by the primary caregiver, which is often the recipient's spouse.

We found that the predictors of paid and unpaid hours are different. Paid hours were somewhat less predictable by the variables tested than were unpaid hours. Paid hours are often constrained by budget caps, which are not measurable. Unpaid hours were predicted by a greater number of ADL/IADL activities, particularly help with mobility, and whether recipients lived alone. As we know, informal hours depend particularly on the availability of family and friends who appear to be more likely to spend time helping with many activities that paid helpers are less likely to assist with, such as walking and getting outside. This study begins to articulate the division of labor in the provision of personal assistance.

Overall, the study demonstrated that the paid and unpaid hours increase rather dramatically as the number of ADLs and IADLs increase, as would be expected. Having more precise estimates of total paid and unpaid hours of care in the United States for different levels of ADLs and IADLs should provide some guidelines for planners and policy makers who are attempting to address the needs of different types of disabled populations. The findings also showed that the amount of paid care is dramatically less than the unpaid care provided by helpers, representing only 13.4 percent of the total hours of PAS. This understanding may be useful for policy makers who would like to expand the amount of paid care provided to individuals with ADL and IADL limitations. Future studies could use these estimates to model the costs of expanding paid coverage for different levels of care, disability groups, and income groups.


1We prefer the term personal assistance over long-term care. The term care is often equated with help for those who are sick or frail. Many people, particularly those who are not older, who need assistance with ADLs and IADLs, are neither sick nor frail.

2A weighted 2-million people said that they got help with these activities, but not in the past 2 weeks. They are not included in this analysis.

3The estimate reported by Liu, Manton, and Aragon (2000) is 21.6 hours, but the denominator includes 0.7-million older persons who did not get help during the survey reference week. Because our study excluded people who did not get help in the NHIS-D reference period, we recomputed Liu et al.'s mean hours excluding this group.

This research was funded by the National Institute on Disability and Rehabilitation Research (H133B980045). This article is the responsibility of the authors and does not necessarily reflect the opinions of the funding agency.


  • Andersen RM. “Revisiting the Behavioral Model and Access to Medical Care: Does It Matter?” Journal of Health and Social Behavior. 1995;36:1–10. [PubMed]
  • Arno PS, Levine C, Memmott MM. “The Economic Value of Informal Caregiving.” Health Affairs. 1999;18(2):182–8. [PubMed]
  • Boaz RF. “Full-Time Employment and Informal Caregiving in the l980s.” Medical Care. 1996;34(6):524–36. [PubMed]
  • Crawford SL, Tennstedt SL, McKinlay JB. “A Comparison of Analytic Methods for Non-Random Missingness of Outcome Data.” Journal of Clinical Epidemiology. 1995;48(2):209–19. [PubMed]
  • Duan N. “A Nonparametric Retransformation Method.” Journal of the American Statistical Association. 1983;78:605–10.
  • Kennedy J, LaPlante M. A Profile of Adults Needs Assistance with Activities of Daily Living Disability Statistics Report, 11. Washington DC: National Institute on Disability and Rehabilitation Research; 1997.
  • Lee ES, Forthover RN, Lorimor RJ. Analyzing Complex Survey Data. Newbury Park: Sage; 1989.
  • Levit KR, Lazenby HC, Braden BR. “National Health Expenditures, 1996.” Health Care Financing Review. 1997;19(1):161–200. [PubMed]
  • Liu K, Manton KG, Aragon C. “Changes in Home Care Use by Disabled Elderly Persons: 1982–1994.” Journal of Gerontology: Social Sciences. 2000;55B(4):S245–53. [PubMed]
  • National Alliance for Caregiving and the American Association of Retired Persons. Family Caregiving in the U.S.: Findings from a National Survey. Bethesda MD: Author; 1997.
  • Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York: John Wiley and Sons; 1987.
  • U.S. General Accounting Office. Adults with Severe Disabilities: Federal and State Approaches for Personal Care and Other Services GAO/HEHS-99-101. Washington DC: Author; 1999.

Articles from Health Services Research are provided here courtesy of Health Research & Educational Trust