Brookings/ICF Long Term Care Financing Model: Model Assumptions

ATTACHMENTS: Memos Related to Data Analysis for the Brookings/ICF Long Term Care Financing Model

TABLE OF CONTENTS

Memo 1 (2/14/89): 1988 Social Security Trustee's and Bureau of the Census Population Projections

TABLE 1. Comparison of Key Assumptions for Population Projections from the Bureau of the Census and the Office of the Actuary

TABLE 2. Comparison of Total Population Projections from the Bureau of the Census and the Office of the Actuary

TABLE 3. Comparison of Elderly Population Projections from the Bureau of the Census and the Office of the Actuary

TABLE 4. Comparison of Elderly Population Projections by Age and Sex in 2020 from the Bureau of the Census and the Office of the Actuary

TABLE 5. Ratio of Elderly Population Projections by Age and Sex in 2020

Memo 2 (6/6/89): SIPP Data on Support for Adults Living in Nursing Homes

TABLE 1. Supported Adults Living in Nursing Homes, by Relationship to Provider, 1985

TABLE 2. Supported Adults Living in Nursing Homes, by Relationship to Provider, 1985

Memo 3 (7/14/89): Status Report on Analysis of SIPP Data on Assets of the Elderly

EXHIBIT 1. Total Assets of the Elderly as Reported in Wave 7 of the 1984 Panel of the Survey of Income and Program Participation, 1985

EXHIBIT 2. Financial Assets of the Elderly as Reported in Wave 7 of the 1984 Panel of the Survey of Income and Program Participation, 1985

EXHIBIT 3. Change in the Financial Asset Stock of the Elderly from Late 1984 to Late 1985 as Reported in the Survey of Income and Program Participation

EXHIBIT 4. Home Equity of the Elderly as Reported in Wave 7 of the 1984 Panel of the Survey of Income and Program Participation, 1985

EXHIBIT 5. Change in Home Equity of the Elderly from Late 1984 to Late 1985 as Reported in the Survey of Income and Program Participation

TABLE 1. Percentage of Elderly Families with Increases in Assets, 1984-1985, by Type and Level of Asset

TABLE 2. Demographic Characteristics of Elderly Home Sellers and Home Buyers

TABLE 3. Home Equity and Financial Assets of the Elderly as Reported in Wave 7 of the 1984 Panel of the Survey of Income and Program Participation, 1985

Memo 4 (7/14/89): Profile of the SIPP Elderly Who Responded in 1984 but not 1985

EXHIBIT 1. Comparison of Dropouts and Non-Dropouts Wave 4 to Wave 7 of the 1984 Panel of the Survey of Income and Program Participation

Memo 5 (8/11/89): Update on Savings Rate of Elderly Families, 1984-1985

EXHIBIT 1. Savings Rates of the Elderly 1984-1985 by Selected Demographic Characteristics

EXHIBIT 2. Savings Rates of the Elderly 1984-1985 by Selected Demographic Characteristics

Memo 6 (10/18/89): Induced Demand

REPORT: Induced Demand in the Brookings/ICF Long-Term Care Model

EXHIBIT 1. Summary of Selected Studies

TABLE A-1. Scanlon's Regression Results

TABLE A-2. Chiswick's Regression Results

TABLE A-3. Nyman's Regression Results

Memo 7 (10/19/89): Disability and Income

REPORT: Is Income a Significant Predictor of Disability Among the Elderly?

TABLE 1. Annual Disability Transition Probability Matrices for the Noninstitutionalized Elderly

TABLE 2. Brookings/ICF Long Term Care Financing Model Disability Prevalence Rates for the Noninstitutionalized

TABLE 3. Disability Prevalence Rates for the Noninstitutionalized from SIPP

TABLE 4. Tabulations for Persons Age 65-67

TABLE 5. Relative Likelihood Having One or More Severe ADL Limitations

TABLE 6. Regression Results

TABLE 7. Correlation Matrix

Memo 8 (1/9/90): Income and Asset Distribution of Elderly Families

TABLE 1. Percent Distribution of Elderly Families by Amount of Financial Assets for Various Demographic Characteristics, 1984

TABLE 2. Percent Distribution of Elderly Families by Amount of Home Equity for Various Demographic Characteristics, 1984

TABLE 3. Percent Distribution of Elderly Families by Financial Asset Level and Income as Percent of the Poverty Level

Memo 9 (1/10/90): Additional Information on the Income and Asset Distribution of Elderly Families

TABLE 1. Percent Distribution of Disabled Elderly Families by Amount of Financial Assets for Various Demographic Characteristics, 1984

TABLE 2. Percent Distribution of Non-Disabled Elderly Families by Amount of Financial Assets for Various Demographic Characteristics, 1984

TABLE 3. Percent Distribution of Disabled Elderly Families by Amount of Home Equity for Various Demographic Characteristics, 1984

TABLE 4. Percent Distribution of Non-Disabled Elderly Families by Amount of Home Equity for Various Demographic Characteristics, 1984

Memo 10 (1/12/90): Additional Information on the Income and Asset Distribution of Elderly Families

TABLE 1. Average Monthly Income and Distribution of Elderly Families by Monthly Income for the Disabled and Non-Disabled, 1984

Memo 11 (7/18/90): Life Insurance Values Held by the Elderly

TABLE 1. Percent of Elderly Persons by the Face Value of Life Insurance Held in 1984

Memo 12 (4/9/91): Table Specs for Distribution of Assets and Income

TABLE 1. Distribution of Income and Assets for Elderly Households

TABLE 2. Distribution of Income and Assets for Elderly Households

Memo 13 (4/25/91): Living Arrangement and Disability

TABLE 1. Disabled Elderly Persons by Alternate Definitions of Disability

Memo 14 (5/16/91): Medigap Analysis Results Using the 1984 SIPP/CES Match File and the 1989 CES

EXHIBIT 1. Age and Income Characteristics of the Non-Institutionalized Elderly with Medicare and Private Health Insurance: 1984

EXHIBIT 2. Poverty and Income of Non-Institutionalized Elderly by Health Care Coverage: 1984

EXHIBIT 3. Comparison of Medicare Supplemental Coverage and Premiums Among Elderly Households to Determine the Percent of Households with Retiree Health Benefits

EXHIBIT 4. Average Annual Out-of-Pocket Expenses for Single Elderly Households by the Amount Spent on Private Insurance Premiums and Earnings Status: 1989

TABLE 1: Non-Institutionalized Elderly Health Care Coverage Characteristics: 1984

TABLE 2: Characteristics of the Non-Institutionalized Elderly with Medicare and Private Health Insurance Coverage: 1984

TABLE 3: Elderly Married Couples and Elderly Single Households Paying Both Medicare and Medicare Supplemental Premiums: 1989

TABLE 4: Average Annual Expenditures for Elderly Married Couples and Elderly Single Households, by Age and Earning Status: 1989

TABLE 5: Average Annual Expenditures as a Percent of Income After Taxes, for Elderly Married Couples and Elderly Single Households, by Age and Earning Status: 1989

TABLE 6: Average Annual Expenditures as a Percent of Income After Taxes, for Elderly Married Couples and Elderly Single Households, by Income and Earning Status: 1989

TABLE 7: Average Annual Health Expenditures, for Elderly Married Couples and Elderly Single Households, by Age and Earning Status: 1989

TABLE 8: Average Health Expenditures as a Percent of Income After Taxes, for Elderly Married Couples and Elderly Single Households, by Age and Earning Status: 1989

TABLE 9: Annual Health Expenditures as a Percent of Income After Taxes, for Elderly Married Couples and Elderly Single Households, by Income and Earning Status: 1989

TABLE 10: Average Annual Out-of-Pocket Expenses for Elderly Households by Amount Spent on Private Insurance Premiums, by the Number of Members in Consumer Unit and by Marital Status: 1989

MEMORANDUM

DATE: February 14, 1989
TO: John Drabek
FROM: Dave Kennell, Lisa Alecxih
SUBJECT: 1988 Social Security Trustee's and Bureau of the Census Population Projections

This memo describes differences in the population projections from the Social Security Area Population Projections: 1988 and the Bureau of the Census Projections of the Population of the United States, by Age, Sex, and Race: 1988 to 2080. It also discusses the impact these differences could have for the model.

When we compared the two reports we found that there are differences in the geographic area covered in the two projections. The projections prepared by the Bureau of the Census are generally for only the United States and the armed forces overseas. Those prepared by the Office of the Actuary include Puerto Rico, Guam, American Samoa, the Virgin Islands, and other United States citizens living abroad.

We also compared the three key assumptions for population projections in the two reports: fertility, mortality and immigration. Table 1 shows that there are differences for all three key areas. The Bureau of the Census assumes lower fertility and net immigration rates than the Office of the Actuary. The death rate per 1,000 persons in the Census report is lower than the Office of the Actuary report in the short-run and higher by 2020.

The net effect of the different assumptions on the projections of the population is that the total population from the Office of the Actuary projections is 2.7 percent higher in 1987 and 4.3 percent higher in 2020 than the Census projections (see Table 2). The 1987 Social Security area population baseline is higher because it covers a larger geographic area. The difference increases in the future because the Office of the Actuary uses higher birth and immigration rate assumptions and lower long-range death rate assumptions.

With respect to the elderly population, Table 3 shows that the Office of the Actuary and the Census projections of the number of elderly are very similar, and both indicate that the percentage of elderly in the population will increase in the future. The only major difference is that the Bureau of the Census projections for the percentage of elderly in the population are higher initially and increase at a faster rate than the Office of the Actuary projections. This more rapid increase in the percentage of elderly in the population is most likely due to effect of the Census' lower fertility rate assumption over time because the rate of increase of the elderly as a percent of the total population between 1987 to 2000 is about the same in both reports -- 7.4 percent for the Census and 7.5 percent for the Office of the Actuary.

Overall, the differences in the Bureau of the Census and the Office of the Actuary's population projections would have very little effect on the model's projections of long-term care utilization or expenditures. Long-term care utilization and expenditures in the model are driven by assumptions based on the demographic characteristics of the elderly population. Table 4 shows that there is very little difference in the two projections of the number of elderly in 2020 by age and sex. Table 5 presents the ratio of the number of elderly persons by age and sex projected by the Office of the Actuary over the number projected by the Bureau of the Census. Most of the cells are within one percentage point of each other, and none of the remaining cells have a difference greater than five percent. Differences in the percentage of the entire population that is elderly described earlier would not effect model results because long-term care utilization is not simulated for the non-elderly.

If you have any questions or comments, please give me a call.

MEMORANDUM

DATE: June 6, 1989
TO: John Drabek
FROM: Dave Kennell, Kathy Chaurette
SUBJECT: SIPP Data on Support for Adults Living in Nursing Homes

In 1985, SIPP data indicated that 166,601 adults lived in nursing homes and received financial support from their family. Approximately 25 percent received support from a spouse, 48 percent received support from a child, 9 percent received support from a parent, and 18 percent received support from some other relative.

On average, spouses provided $8,153 per year to their spouse living in a nursing home, while nonspouses provided an average of $818 to their relative. However, the family relationship of nonspouses substantially influenced the amount of support they provided. For example, children gave an average of $528 per year, whereas other relatives (excluding parents) gave $1,658 per year.

Among spouses providing support, 82 percent gave at least $2,500 per year. Of nonspouses providing support, 72 percent gave under $1,000 per year; 82 percent of children provided under $1,000, and 15 percent of other relatives (excluding parents) provided under $1,000. Among persons providing at least $3,500 per year, spouses gave an average of $12,674, while nonspouses gave an average of $6,000.

These estimates are from the Topical Module of the Survey of Income and Program Participation 1984 Panel Wave 5. The amount of support provided is the amount reported for the twelve months preceding the survey month. The Wave 5 survey was conducted from January through April of 1985.

In this survey, only 30 respondents indicated that they provided support to an adult living in a nursing home. One person reported supporting two nursing home residents. Only 9 respondents reported supporting a spouse, and 22 reported supporting some other relative. The small sample size indicates that the estimates should be used with caution.

MEMORANDUM

DATE: July 14, 1989
TO: John Drabek
FROM: Dave Kennell, Kathy Chaurette
SUBJECT: Status Report on Analysis of SIPP Data on Assets of the Elderly

I. OVERVIEW

We have conducted a preliminary analysis of the savings behavior of the elderly to determine whether the elderly accumulate and decumulate assets over time. We hope to use the results of this analysis to modify the assumptions in the Brookings/ICF Long Term Care Financing Model on the savings behavior of the elderly. The base case of the model assumes that, except for long term care expenses, no savings or dissavings is done by the elderly.

We are using longitudinal data from SIPP for 1984 and 1985 to study the saving/dissaving patterns of the elderly by selected demographic and economic characteristics. Although by using SIPP we are able to analyze assets at two points in time for the same group of elderly, the data were collected only 12 months apart. Consequently, we are restricted to a study of asset accumulation/decumulation over a twelve month period beginning in late 1984.

II. METHODOLOGY

The 1984 Panel of SIPP was asked questions on assets twice, once in late 1984, and exactly twelve months later in 1985. We extracted the records of all elderly persons and couples from both the 1984 assets module and the 1985 assets module of SIPP. We then matched the 1984 records to the 1985 records, using the unique identifier provided in the SIPP data, so that we could analyze changes in asset stocks over time. We were unable to use 1,496 (31 percent) of the 1984 records because they did not have data for 1985, leaving 3,353 records with data for both 1984 and 1985. However, only 343 (9 percent) of the total 1985 records were not used because of missing data for 1984.

We used the following definitions in our preliminary analyses:

• For married couples:
  • income and assets are pooled;
  • the age of the oldest partner is used to categorize the couple by age;
  • if either partner in a married couple is non-white then the couple is considered to be non-white;
  • the education level of the partner with the highest education level is used;
  • if at least one partner in a married couple was disabled, then the couple is classified as disabled;

  • a person is classified as disabled if she reported having difficulty walking upstairs, trouble getting around outside the house alone, or needed help with personal needs such as dressing;

• In the calculations, we used the 1985 SIPP weight to weight the sample observations to the population

• When calculating the change in financial assets, if the person or couple did not report financial assets in 1984, the percentage change could not be calculated and the person or couple is placed in a separate category called "no financial assets in 1984";

• When calculating the change in home equity; a) if the person or couple reported home equity in 1984 but did not in 1985, the person or couple is placed in a category called "home sellers", b) if the person or couple did not report home equity in 1984 but did in 1985, the person or couple is placed in a category called “home buyers”.¹

III. FINDINGS

Total Assets in 1985

In 1985, SIPP data (shown in Exhibit 1) indicate that 9 percent of the elderly did not own any assets including both financial assets and home equity. Another 16 percent of the elderly owned assets worth less than $10,000, 9 percent had assets worth $10,000 - $24,999, 15 percent had assets worth $25,000 - $49,999, 22 percent owned assets worth $50,000 - $99,999, and 29 percent had assets worth at least $100,000. Among the elderly who reported assets, the average amount reported was $102,315. This relatively high average indicates that the elderly in the highest asset range hold substantial amounts of assets.

The amount of assets reported varied considerably by age, disability status, marital status, education level, race, and income. Low income, low education level, non-white race, presence of a disability, increased age, and single marital status are associated with lower assets. For example, 33 percent of the disabled elderly had under $10,000 in assets, while 19 percent of the non-disabled elderly owned under $10,000 in assets. Also, 45 percent of non-whites, but only 22 percent of whites, owned under $10,000 in assets in 1985.

Surprisingly, although single men reported a higher average amount of assets than single women, 36 percent of the single men owned under $10,000 in assets, whereas 32 percent of single women had under $10,000 in assets. Education level is also an important indicator of total assets. Only 9 percent of the elderly who had attended college owned under $10,000 in asset; 22 percent of those who did not attend beyond high school, and 40 percent of the elderly who did not attend high school owned assets worth under $10,000 in 1985.

Financial Assets in 1985

In 1985, SIPP data (shown in Exhibit 2) indicate that 12 percent of the elderly had no financial assets, another 29 percent had under $10,000 in financial assets, 15 percent owned financial assets worth $10,000 - $24,999, 14 percent held financial assets worth $25,000 - $49,999, 15 percent owned $50,000 - $99,999 in financial assets, and 14 percent owned financial assets worth at least $100,000. Among the elderly who reported having financial assets, the average amount reported was $64,182.

The average amount reported varied considerably by marital status, age, race, education level, disability status, and income. Overall, whites, married couples, and the younger elderly reported more financial assets than non-whites, singles, and the older elderly. Furthermore, higher education level was associated with higher financial assets. Only 3 percent of the elderly who did not attend high school had $100,000 or more of financial assets; while 12 percent of those who had attended high school, and 33 percent of those who had attended college had $100,000 or more of financial assets.

Disability and income are also important indicators of asset level. For example, while 33 percent of the non-disabled elderly had under $10,000 in financial assets, 53 percent of the disabled elderly had financial assets under $10,000. Additionally, 81 percent of the elderly who reported under $500 monthly income had less than $10,000 in financial assets, 50 percent of elderly with monthly income between $500 and $999 had less than $10,000 in financial assets; 22 percent with monthly income between $1,000 and $1,499, 13 percent with monthly income between $1,500 and $1,999, and under 10 percent of the elderly with at least $2,000 monthly income, owned less than $10,000 in financial assets in 1985.

Changes in Financial Assets

We also analyzed the change in financial asset stock over a one year period from late 1984 to late 1985 using matched SUP data² (see Exhibit 3). This analysis indicates that between 1984 and 1985 there was no predominate pattern of saving/dissaving among the elderly.³ For example, 18 percent of the elderly reported at least a 75 percent increase in the value of their financial assets over the year, while 13 percent reported a reduction in financial assets of at least 75 percent. The remaining elderly were split between saving and dissaving with 15 percent having an increase of between 10 and 75 percent, another 10 percent increased their financial assets by 0 to 10 percent, 7 percent dissaved by less than 10 percent, and 27 percent dissaved by 10 to 75 percent. Overall, 43 percent of the elderly increased or at least maintained their financial assets over the year, and 47 percent dissaved. The remaining 9 percent reported owning no financial assets in 1984.⁴

The saving/dissaving pattern varied by race, education level, marital status, and income. Surprisingly, there was little variation by age. For example, about 43 percent of the elderly age 65 to 69 (in 1984) saved over the year, 45 percent of those age 70 to 74, 44 percent of the elderly age 75 to 79, 42 percent of those age 80 to 84, and 41 percent of those 85 or over increased the value of their financial assets over the year.

Grouping the elderly by disability status, marital status, and amount of asset holdings as of 1984 reveals more distinct patterns of saving/dissaving. For example, among the disabled elderly, 40 percent reported an increase in financial assets over the year, while 46 percent of the non- disabled elderly reported an increase in financial assets over the year. Among married couples, 47 percent saved, compared to 41 percent of single females and 39 percent of single males.

Additionally, the elderly with lower levels of financial assets were more likely to save and less likely to dissave than were the elderly with higher levels of financial assets. As shown in Table 1, over 50 percent of families with financial assets under $25,000 in 1984 saved while only 35 percent of families with $150,000 or more in financial assets saved.

Home Equity in 1985

In 1985, SIPP data (shown in Exhibit 4) indicate that 40 percent of the elderly had no home equity, another 8 percent had under $25,000 in home equity, 19 percent had home equity worth $25,000 - $49,999, 23 percent had $50,000 - $99,999 in home equity, and 10 percent had home equity worth at least $100,000. Among the elderly who reported having home equity, the average amount reported was $61,706.

The average amount reported varied substantially by age, race, education level, marital status, disability status, and income. In general, whites, married couples, the more highly educated, and the younger elderly reported more home equity than non-whites, singles, the less educated, and the older elderly. For example, 35 percent of whites, but only 15 percent of non-whites, had home equity worth at least $50,000. Also, 57 percent of the elderly who had attended college reported home equity of at least $50,000; while 34 percent of those who did not attend beyond high school, and 16 percent of those who did not attend high school had $50,000 or more in home equity.

Disability and income are both important determinants of the value of home equity. For example, while 24 percent of the disabled elderly had home equity worth at least $50,000, 40 percent of the non-disabled elderly had home equity worth at least $50,000. Additionally, 11 percent of the elderly who reported under $500 monthly income reported home equity worth at least $50,000; 23 percent with monthly income between $500 and $999, 39 percent with monthly income between $1,000 and $1,499, 52 percent with monthly income between $1,500 and $1,999, 55 percent with monthly income between $2,000 and $2,499, 66 percent with monthly income between $2,500 and $2,999, and 76 percent with at least $3,000 monthly income had home equity worth at least $50,000 in 1985.

Changes in Home Equity

We analyzed the change in home equity over the year 1984 - 1985 for elderly families who reported home equity in either 1984 or 1985 (see Exhibit 5). overall, about 35 percent of the elderly homeowners reported a decrease in home equity over the year, and 65 percent reported an increase in home equity. Among elderly homeowners, 4 percent sold their home between 1984 and 1985, and about 1 percent bought a home between 1984 and 1985.

Changes in home equity for families who owned a home in 1984 varied by income. For example, 55 percent of the elderly who reported under $500 monthly income reported an increase in home equity, while between 62 and 66 percent of the elderly who reported $500 - $2,499 monthly income, and 73 percent of the elderly who reported at least $3,000 monthly income reported an increase in home equity from 1984 to 1985.

Changes in home equity also varied by disability status, age, education level, marital status, and value of the equity. As Exhibit 5 shows, 60 percent of the disabled elderly, and 68 percent of the non-disabled elderly reported an increase in home equity (including home buyers) over the year. The change in home equity varied only slightly by age. For example, 68 percent of the elderly age 65 to 69 reported an increase in home equity; 65 percent of the elderly age 70 to 74, 63 percent of those age 75 to 84, and 64 percent of those age 85 or more reported an increase in home equity over the year. Additionally, Table 1 shows that the percentage of families with increases in home equity declined considerably as home equity increased.

Home Sales

About 4 percent of the elderly sold their home between 1984 and 1985 (see Exhibit 5). The disabled, low-income, single, less educated, and older were more likely to sell their home than were other elderly. For example, 6 percent of the disabled elderly were home sellers, while 3 percent of the non-disabled elderly were home sellers. Additionally, 6 percent of the elderly who did not attend high school sold their home; 3.5 percent of those who did not attend beyond high school, and 3 percent of the elderly who had attended college sold their home over the year.

Single men were more likely to sell their home than were single women and married couples. About 7 percent of the single male homeowners sold their home over the year; whereas 4 percent of single women, and 3 percent of married couples sold their home over the year. Income is also an important indicator of home sales. For example, 7 percent of elderly homeowners with under $500 monthly income sold their home over the year; 5 percent with $500 $999, 4 percent with $1,000 - $1,499, 3 percent with $1,500 - $1,999, 2 percent with $2,000 - $2,499, and less than 1 percent of the elderly with at least $2,500 monthly income sold were home sellers.

Among home sellers (see Table 2), 57 percent were disabled compared to 43 percent who were not disabled. Furthermore, 90 percent were white, 81 percent did not attend beyond high school, 65 percent reported under $1,000 monthly income, 17 percent were single men, 37 percent were single women, and 45 percent were married couples.

Home Purchases

As shown in Exhibit 5, about 1 percent of the elderly bought a home between 1984 and 1985. Interestingly, the disabled, low-income, and age 85 and over were more likely to buy a home than were other elderly. Home buying varied only slightly by marital status, education level, and race.

About 2 percent of the disabled elderly bought a home between 1984 and 1985, while 1 percent of the non-disabled elderly bought a home over the year. Additionally, 3 percent of the elderly with under $500 monthly income purchased a home, while not more than 2 percent of the elderly with at least $500 monthly income bought a home during the year. Furthermore, while the percentage of elderly who purchased a home declined from 2 percent to .5 percent as age increased from 65 to 84, 5 percent of the elderly age 85 and over were home buyers.

The distribution of elderly home buyers by demographic characteristics is very similar to that found among elderly home sellers. This indicates that certain groups of the elderly have a more dynamic housing pattern than other groups of elderly. Among elderly home buyers (see Table 2), 58 percent were disabled, 57 percent reported under $1,000 monthly income, 82 percent did not attend beyond high school, 89 percent were white, 43 percent were single women, 20 percent were single men, and 38 percent were married couples.

Financial Assets and Home Equity Interaction

As shown in Table 3, the value of financial assets is closely related to the value of home equity. In general, as the value of financial assets increases, the value of home equity increases. Overall, 67 percent of the elderly families who did not have financial assets also did not have home equity in 1985.

Of the elderly families who reported financial assets in 1985, 52 percent of those with under $10,000 in financial assets had no home equity, and another 33 percent had under $50,000 in home equity; 40 percent of the elderly with $10,000$24,999 in financial assets had no home equity, and 53 percent had $10,000 - $99,999 in home equity; 30 percent of those with $25,000 - $49,999 in financial assets had no home equity, and another 60 percent had $25,000 - $149,999 in home equity; 30 percent of the elderly with $50,000 - $99,999 in financial assets had no home equity, and 63 percent had home equity worth $25,000 - $149,999; 18 percent of those with $100,000 - $149,999 in financial assets had no home equity, and another 70 percent had $25,000 - $149,999 in home equity; and 12 percent of those with at least $150,000 in financial assets had no home equity, and another 75 percent had home equity worth at least $50,000.

IV. NEXT STEPS

We are currently identifying factors that appear to influence the saving/dissaving behavior of the elderly, such as, age, income, and disability status controlling for other factors. This is in preparation for the next step of formulating a simple regression model that will be useful for estimating the saving/dissaving behavior of the elderly from cross-sectional data.

We are also planning to calculate the savings rate of elderly families by specific demographic and economic characteristics. We have prepared the data for analyses that control for imputation of asset amounts.

NOTES

1. If total financial assets in 1984 or 1985 were valued at less than zero, the family was excluded from Exhibit 3. Similarly, if home equity in 1984 or 1985 was valued at less than zero, the family was excluded from Exhibit 5.

2. The percentage change in financial assets over the year was calculated using the following formula for each family that had total financial assets in 1984 that were valued at greater than zero, and total financial assets in 1985 that were not negative: ((assets in 1985 - assets in 1984) / assets in 1984) * 100.

3. This result may be affected by amputations of asset data performed by the Bureau of the Census on the SIPP data. We will explore this problem and control for amputations in our regression analysis.

4. Another 1 percent of the families reported negative assets in 1985 and were excluded from Exhibit 3.

MEMORANDUM

DATE: July 14, 1989
TO: John Drabek
FROM: Dave Kennell, Kathy Chaurette
SUBJECT: Profile of the SIPP Elderly Who Responded in 1984 but not 1985

A total of 1,166 (representing 5.4 million in 1984) elderly families were assigned a weight in Wave 4 but not Wave 7 of the 1984 Panel of the Survey of Income and Program Participation. An analysis of these families (see Exhibit 1) indicates that the dropouts were more likely to be disabled, older, single, report monthly income under $500, report no home equity, and report under $10,000 in financial assets than were non-dropouts. Education level and race were not important indicators of the likelihood of participation in Wave 7 of the survey.

Disability and age appear to be the most important determinants of participation in both years. About 47 percent of the dropouts were disabled compared to 42 percent of the non-dropouts. Furthermore, 48 percent of the dropouts were age 75 and over, whereas 41 precent of the non- dropouts were age 75 and over. Additionally, 62 percent of the dropouts and 58 percent of the non-dropouts were single.

Financial status is also an important indicator of participation in both years. About 29 percent of the dropouts reported monthly income under $500 compared to 26 percent of non-dropouts. Similarly. 42 percent of dropouts reported no home equity, and 39 percent of non-dropouts reported no home equity in 1984; 42 percent of dropouts reported under $10,000 in financial assets, and 40 percent of non-dropouts reported under $10,000 in financial assets in 1984.

MEMORANDUM

DATE: August 11, 1989
TO: John Drabek
FROM: Dave Kennell, Kathy Chaurette
SUBJECT: Update on Savings Rate of Elderly Families, 1984-1985

I. METHODOLOGY

We calculated the savings rates for 1984 to 1985 for elderly families in four ways (see Exhibit 1): 1) using all families with positive financial assets in 1984 (method 1); 2) using all families with positive financial assets in 1984 after setting specific asset items to zero in both years if the item was imputed in both years, or to the non-imputed year's amount if the item was imputed in only one year (method 2); 3) using all families with positive financial assets in 1984 after setting specific asset items to zero in both years if the item was imputed in either year (method 3); and 4) excluding families that had any asset item imputed in either year (method 4). The definition of savings rate used for this analysis was the sum of financial assets in 1985 minus financial assets in 1984 for a group, divided by total financial assets in 1984 for the same group.

We also grouped the elderly into three savings categories using each of the four methods described above (see Exhibit 2). Category 1 includes elderly whose savings rate was over 10 percent from 1984 to 1985 (savers), category 2 includes those whose savings rate did not exceed 10 percent and who did not dissave by more than 10 percent (no change), and category 3 includes those elderly who dissaved by more than 10 percent from 1984 to 1985. A 10 percent band was selected based on our earlier analysis of the change in assets stock over the year which indicated that saving/dissaving among the elderly is not tightly concentrated around zero. This is probably partly explained by the difficulty of consistently valuing assets from year to year, especially property holdings. A small discrepancy from year to year in the value of an asset represents a relatively high savings/dissavings rate for those elderly who own a small amount of assets.

II. FINDINGS

The savings rate varied considerably depending upon the treatment of imputed assets. The savings rate for all families with positive assets in 1984, without adjusting for imputed assets, was 3.0 percent (see Exhibit 1). After adjusting for imputed assets by method 2, 47 (1.5%) families were exluded because they no longer had positive assets in 1984, and the savings rate dropped to 1.4 percent. Alternatively, adjusting for imputed assets by method 3 excluded 79 (2.6%) families that no longer had positive assets in 1984, and the savings rate dropped to 1.5 percent. Finally, after adjusting for imputed assets by method 4, 1,252 (41%) families were excluded, and the savings rate dropped to -4.3 percent.

The savings rate varied substantially by disability status, income, age, marital status, education level, race, and assets stock. For example, without adjusting for imputed assets, the savings rate for the disabled was 8.6 percent, while the savings rate for elderly families that were not disabled was 1 percent. Furthermore, single females had a savings rate of 6.4 percent, compared to -4.5 percent for single males, and 3.0 percent for married couples. After adjusting for imputed assets by method 2, the savings rate for the disabled elderly was 2.8 percent, while the savings rate for other elderly remained at about 1 percent. Additionally, single females had a savings rate of 4.3 percent, compared to -3.5 percent for single males, and 1.2 percent for married couples.

Families with lower income and assets stock tended to have a higher savings rate both before and after adjusting for imputed assets. For example, without adjusting for imputed assets, elderly families with under $500 monthly income had a savings rate of 19.7 percent; the savings rate was 5.6 percent for elderly with $500 to $999 monthly income, 5.8 percent for those with $1,000 to $1,499 monthly income, -2.1 percent for those with $1,500 to $1,999, 2.8 percent for elderly with $2,000 to $2,499, 9.7 percent for those with $2,500 to $2,999, and -1.1 percent for elderly with at least $3,000 monthly income. Similarly, the elderly with $1 to $9,999 in total assets (financial plus home) had a savings rate of 101.8 percent; the savings rate was 25.4 percent for those with $10,000 to $24,999, 37.7 percent for elderly with $25,000 to $49,999, 21.9 percent for those with $50,000 to $99,999, 6.7 percent for elderly with $100,000 to $149,999, and -5.2 percent for those with at least $150,000 in total assets in 1984. The higher savings rates for the lower income elderly and those with smaller assets stock probably reflects the fact that a dollar saved by a family with small asset holdings represents a relatively larger savings than a dollar saved by a family with large asset holdings.

Grouping the elderly into the three savings categories of savers, no change, and dissavers, resulted in slightly different distributions of elderly depending upon which of the four methods was used to handle amputations (see Exhibit 2). When no adjustments were made for imputed assets, 37 percent of the elderly were categorized as savers, 19 percent were categorized as no change, and 44 percent were dissavers. After adjusting for imputed assets using method 2, 35 percent of the elderly were categorized as savers, 24 percent as no change, and 42 percent as dissavers. Similarly, after adjusting for imputed assets using method 3, 35 percent of the elderly were categorized asd savers, 21 percent as no change, and 43 percent as dissavers. Finally, after adjusting for imputed assets using method 4, 33 percent of the elderly were categorized as savers, 25 percent as no change, and 42 percent as dissavers.

The distribution of the elderly across the savings categories differs slightly by disability status, age, marital status, education level, and race. In general, disabled, older, single, female, less educated, and white elderly were less likely to save and more likely to dissave than were other elderly. This is true regardless of the treatment of imputed assets. Additionally, the elderly with lower monthly income were considerably less likely to be savers than were the elderly with higher monthly income. For example, without adjusting for imputed assets, 32 percent of the elderly with $0 - $499 in monthly income were savers; 37 percent of those with $500 - $999, 38 percent of the the elderly with $1,000 - $2,499, 41 percent of those with $2,500 - $2,999, and 40 percent of the elderly with at least $3,000 monthly income were savers. Furthermore, the lower income elderly were substantially more likely to be dissavers than were higher income elderly. The distribution of the elderly by monthly income and savings category is similar regardless of the treatment of imputed assets.

The distribution of the elderly across savings categories also varies considerably by level of financial assets and total assets (financial plus home). Overall, elderly with smaller assets were more likely to be savers and less likely to be dissavers than were elderly with larger levels of assets. This is true for each of the four methods of handling imputed assets.

III. NEXT STEPS

We are continuing our analysis of the relationship between savings and dissavings and demographic characteristics of elderly families. We will attempt to estimate some regressions to control for the key variables which appear to affect the savings rate.

MEMORANDUM

DATE: October 18, 1989
TO: John Drabek
FROM: Dave Kennell
SUBJECT: Induced Demand

Enclosed is a draft paper written by Teresa Fama on induced demand and how it might be incorporated into the model.

After you have had a chance to review this paper, we can set up a time to discuss its implications for the model.

Please feel free to call Teresa at 842-8979 if you have questions about the paper.

INDUCED DEMAND IN THE BROOKINGS/ICF LONG-TERM CARE MODEL

Submitted to: DHHS/ASPE/SSP
By: Lewin/ICF
October 17, 1989 (Draft)

I. OVERVIEW

The Brookings/ICF Long-Term Care Financing Model is designed to simulate the impact of changes in long-term care programs on the sources and levels of long-term care financing. A key factor in simulating the impact of a new program is the expected level of increase in consumer demand due to the reduced cost of services to consumers in a new program (called induced demand). This reduced cost may occur because of a reduction in required consumer cost sharing.

Canada's health care system provides evidence of the effects of relaxing financial eligibility for long-term care or expanding benefits. Canada's provinces instituted a long-term care program in the 1970s offering community-based care, case management services, and nursing home care on the basis of functional impairment (i.e., no cost sharing). A study of the Canadian system by Kane and Kane (1985) revealed that the availability of community services did not reduce the supply of nursing home beds. Of equal importance, the number of homemaker hours in one province studied, British Columbia, increased almost four times during the three year period after implementation of the expanded benefits for long-term care. This increase, however, levelled off after this time period.

In the Brookings/ICF Long-Term Care Financing Model, induced demand is modelled through user-specified assumptions about the expected increase in utilization due to a new public or private financing program. The model is structured so that the effect of a given level of induced demand can be modelled by simulating additional nursing home admissions or home health care users.¹ The induced demand assumption in the model is structured in this manner because of the lack of data concerning the effects of induced demand on nursing home and home care utilization.

In order to examine the potential effects of induced demand in the United States, this paper reviews the literature relating to induced demand, including John Nyman's analyses of excess demand for nursing home care, information gathered in the Channeling studies to determine the increase in home care use, and results from the Rand health insurance study. A summary of the studies' results is found in Exhibit 1.

Based on this review, we propose a methodology for incorporating or estimating induced demand under differing scenarios. We also explore the feasibility of incorporating an elasticity of demand equation into the Brookings/ICF Long-Term Care Model.

II. INDUCED DEMAND FOR NURSING HOME CARE

Three studies of the demand for nursing home care have been done by Scanlon (1980), Chiswick (1976), and Nyman (1989). All three studies used different methodologies; however, the range of price elasticities was narrow compared to the wide range found in the demand for health insurance literature. Detailed results and model specifications for these three studies are included in the Appendix A.

Scanlon (1980) found a segmented market characterized by private-pay patients experiencing an unlimited supply of nursing home beds and public-pay patients (or Medicaid patients) demanding the beds that remain after the private patients' demands are met. Based on this theory of the nursing home market, Scanlon's private demand equation is a true demand function. That is, it does not make sense to analyze total utilization as the dependent variable because it is a product of the demand for nursing home care and the probability of acquiring an empty bed. Because private-pay patients face an unlimited bed supply, the probability of acquiring an empty bed equals one. Therefore, total utilization equals the demand for nursing home care, which also is the private demand for nursing home care.

Scanlon used aggregate state data from 43 states for the years 1969 and 1973. He estimated two equations, one for total utilization and one for private demand. The dependent variable for the former equation was all nursing home residents per population over age 65; the dependent variable for the latter equation was private pay residents in all homes per population over age 65. (See Table A-1 in Appendix A).

Using both ordinary least squares (OLS) and two-stage least squares (2SLS), Scanlon found a highly significant negative coefficient for the private price variable. The price variable coefficient was insignificant in the total utilization model. A -1.12 coefficient for the 2SLS equation implies that as price increases 10 percent, demand for nursing home care decreases 11 percent. The OLS elasticity estimate is similar, -0.90, which implies that a 10 percent increase in price results in a 9 percent decrease in demand for care. The 2SLS equation was preferred by Scanlon because of the price-quantity simultaneity of the demand equation. As Scanlon points out, the quantity demanded is affected by the price and the price is affected by the anticipated quantity demanded (p.35).

The other significant variables in the equations are percent urban (-) and per capita income (+).² The income elasticities for the OLS and 2SLS equations were large, 2.83 and 2.27, respectively. Based on the negative elasticities for price and the positive elasticities for income, Scanlon concludes that nursing home care is a normal good.

Chiswick (1976) used data from all 50 states for the year 1967. The dependent variable for his equation was the number of nursing home residents in the SMSA (Standard Metropolitan Statistical Area) per thousand per population over age 65 in the SMSA. (See Table A-2 in Appendix A). Chiswick found the significant independent variables to be price (-), income (+), the percent of adult married women in the labor force (+), the percent of those aged 65 and over who are female (-), the percent of those over age 65 who are age 85+ (+), and the death rate for the aged as an index of poor health (+). The relevant elasticities, price and income, are -2.3 and 0.9, respectively. The price elasticity implies that as price increases 10 percent quantity demanded decreases 23 percent. The income elasticity implies that as income increases 10 percent quantity demanded increases 9 percent. Again, Chiswick's findings suggest that nursing home care is a normal good.

The most recent analysis of the demand for nursing home care was done by John Nyman (forthcoming in the Journal of Health Economics). He, like Scanlon, looked at the private demand function for two reasons. First, because the states (and federal government) pay for Medicaid patients' care, these patients are not likely to respond to price. Second, due to a limited nursing home bed supply, private patients are preferred to Medicaid patients because of the higher per them rate that private patients pay. Nyman states that homes compete for private pay patients' business; therefore, these patients have an “unconstrained choice among homes (Nyman 1985 and 1988).”

Nyman presents some explanations for why his analysis differs from Chiswick's and Scanlon's. First, Chiswick's analysis does not differentiate between private and Medicaid patients. Second, Chiswick and Scanlon use geographic areas as the unit of observation. Therefore, quantity and price variables must be aggregated across firms. Nyman uses the firm (nursing home) as the unit of observation because he does not think that nursing home care is homogeneous across homes with respect to quantity and quality. Also, it is not clear that average price accurately represents a price level. As Nyman points out, private patients face a range of prices. Third, Chiswick used average revenue as his price variable. Nyman notes that because revenue comes from both private and Medicaid patients, it is not a good measure of private price.

Nyman presents results from a number of different models. For the full sample of 314 nursing homes, he estimates two equations. The dependent variables are the number of patients in the home (equation A) and the number of private admissions during the year (equation B). (See Table A-3 in Appendix A). Two-stage least squares was used in order to address the simultaneity of price and quantity demanded, on the one hand, and quality and quantity, on the other. Each dependent variable was regressed on a number of independent variables. The significant variables (at the 5 percent level) for equation A are SNF private per them price (-), per capita income in the county elderly population in the county (+), percentage of married women in the county who work (+), number of nursing homes in the county (-), percentage of patients who are mentally ill or mentally retarded (-), average length of stay of patients in the home (-), and number of beds in the home (+). The significant variables for equation B are the deaths per population in the county (+), percentage of patients who are mentally ill or retarded (-), average length of stay of patients in the home (-), and number of beds in the home (+).

The price elasticity of demand for equation A is -1.7 and highly significant. The price elasticity for equation B is -0.3 and insignificant. The income elasticity in equation A is 1.2; equation B has an income elasticity of 0.86 and is significant only at the 8 percent level. Because of the insignificance of the price variable in equation B, Nyman concluded that some homes have a number of false private admissions, i.e., admissions that are private for a very short period of time and result in becoming Medicaid dependent. He, therefore, eliminated homes with private admissions greater than or equal to the number of private patients and reran the regressions on the resulting sample of 156 homes. The price elasticities for the two equations are -2.3 for equation A and -2.6 for equation B; both were highly significant. The income elasticities for equations A and B are 1.5 and 1.9, respectively.

Nyman's results are similar to Chiswick's, even though he used a different unit of observation. The range of elasticity estimates from these three studies is -1.1 to -2.6, with Scanlon's results being the lowest and Nyman's results for his subsample of 156 homes being the highest. Regardless of which elasticity is "correct", even the lowest range implies that private consumers of nursing home care are very responsive to price changes.

III. INDUCED DEMAND FOR HOME HEALTH CARE

We have found no estimates of the price elasticity of demand for home health care; however, several studies have analyzed the relationship between the amount of formal home care received and variables such as disability level, living arrangements, and income (excluding price).

The Channeling demonstration was designed to determine if formal home care services provided in the community would reduce nursing home care. Two models were designed to study this relationship. The first model, the basic case management model, provided limited funds to purchase community services, and case management was intended to coordinate services under the existing system. The second model, the financial control model, expanded service coverage, provided more funds for services, and gave case managers the authority to determine the amount, duration, and scope of services paid for by these increased funds.

The results from the demonstration showed that formal home care services increased significantly in both models; however, there was a larger treatment/control group difference in the financial control model compared to the basic case management model (Corson, Grannemann, and Holden, 1988). Corson, et al. believe that the effects were stronger in the financial control model because of the greater provision of funds for in-home services compared to the basic case management model. Utilization of formal home care services did not increase because more individuals remained in the community; rather, it increased because those using community services increased their use of these services.

The Channeling demonstration researchers found only a slight reduction in informal care in both models (Christianson, 1988). The treatment/control group difference was significant for the financial control model results only. A comparison of the effects at six months showed that a 5 percentage point increase in the percent receiving formal services was associated with a 1 percentage point decrease in the percent receiving informal care (under the financial control model). Moreover, the reduction in informal care was due to a decrease in non-family caregivers, not primary caregivers.

Soldo (1985) found that the receipt of formal home care services depended on ADL dependency and the client's living arrangements. She did not include a price variable in her model, but did include an income variable, measured as the annual cash income of the family in the preceding calendar year. The effect of income on receipt of formal services was insignificant. Using logistic regression analysis, Soldo found that the probability of receiving formal services increased as needs (i.e. ADL dependency) increased. In addition, at any given level of need, the probability of receiving formal services is greatest for those living with non-relatives, followed in descending order by women living alone, with other relatives, and with spouses. These conclusions do not provide an estimate of the price-elasticity of demand for formal home care services; however, the evidence suggests that elasticities may be very sensitive to level of need. Soldo points out that, "Reductions in price may not alter the service utilization patterns of the extremely dependent elderly, but as the price to the consumer declines, the demand could increase from those with lesser care needs (p.301)." These results suggest that any price elasticity or demand equation incorporated into the model should take disability level into account.

IV. INDUCED DEMAND AND THE RAND HEALTH INSURANCE EXPERIMENT

A review of the literature for estimates of the price elasticity of demand for medical care reveals that estimates vary by a factor of 10 or more (Manning, et al, 1987). To narrow this uncertainty about how demand responds to insurance-induced changes in price, the federal government sponsored a health insurance experiment in 1974 (the Rand Experiment). The price elasticity estimates from the Rand experiment probably are not indicative of those for home health or nursing home care because the Rand experiment's sample did not include anyone 62 years of age and older at the time of enrollment. The Rand experiment is reviewed, however, because it provides a framework for determining the variables to be used in an elasticity of demand equation.

The independent variables used in the Rand analysis were type of insurance plan, age, sex, race, family income, health status, family size, and site. There were five types of insurance plans: a free plan, with no out-of-pocket costs, 25 percent coinsurance rate plan, 50 percent coinsurance rate plan, 95 percent coinsurance rate plan, and a 95 percent coinsurance rate plan for out-patient services and free inpatient care. A four-equation model was used to estimate medical expenditure per person enrolled in the experiment. The first equation is a probit equation for the probability of receiving medical services (inpatient or outpatient). The second equation also is a probit equation for the conditional probability that an enrollee will have at least one inpatient stay, given that the enrollee has some medical use. The third equation is a linear regression for the logarithm of total annual medical expenses of enrollees using out-patient services only. The fourth equation is like the third. but the medical expenses are for users of any inpatient services.

Their results indicate that the likelihood of any medical service use increases as the coinsurance rate decreases. For example, the likelihood of medical service use under the 95 percent coinsurance plan is 68 percent compared to 87 percent under the free plan. Moreover, medical expenditures are 46 percent higher in the free plan compared to the 95 percent coinsurance plan. An analysis by subgroups such as income and health status, revealed that the probability of any medical service use increases with income, with larger increases for the coinsurance plans than for the free plan. However, the probability of inpatient use decreases as income increases. Surprisingly, even though health status was a strong predictor of expenditure level, there was no differential response to health insurance coverage based on health status.

The authors' estimate of price elasticity of demand for medical care is much smaller than the estimates for nursing home care that Scanlon, Nyman, and Chiswick found. In order to compare the results from the Rand experiment with results from the literature, Manning, et al, used an average coinsurance rate method and two other methods to determine a price elasticity of demand. Estimates from the Rand analysis were similar to values in the lower range of the nonexperimental literature (Newhouse, 1981); depending on the method they used to estimate the price elasticity, the estimates were in the -0.1 to -0.2 range. Therefore, a 10 percent increase in price of health care would result in a 1 to 2 percent decrease in demand for medical care.

VI. IMPLICATIONS FOR THE MODEL

There are at least five issues we must address in modelling induced demand. The first issue is how we would measure the reduction in price from a public or private insurance program. In the model, if the insurance plan is strictly a copayment or a fixed reduction in the per diem rate, then the price reduction is simple to determine -- it is 1 minus the copayment in the former case or the ratio of the old per them rate to the new per them rate in the latter case. However, typical insurance plans do not operate that simply. Most insurance plans have a deductible and a fixed indemnity. In this more complex case, further research is required in order to devise a method for determining the reduction in price.

The second issue is determining if a distinction should be made between the induced demand effect under a public insurance plan and the effect under a private insurance plan. A price elasticity of -1.2 is meaningless if the number of beds available cannot meet the increased demand of 12 percent (as price decreases 10 percent). If a private insurance plan is offered, a price reduction most likely would generate a large number of new private pay admissions due to the fact that for every nursing home resident there are two similar persons in the community (Doty, Liu, and Weiner, 1985). Because nursing homes prefer private pay patients, it seems likely that the increased demand would be met. Rather than trying to determine if the induced demand effect would be different under a public or private insurance plan, incorporating assumptions about the supply of beds into the model may be more fruitful. Currently, the model assumes that the supply of nursing home beds will increase with the rate of growth of the elderly population. However, in reality this is not the case. Research into the rate of growth of the supply of nursing home beds and the factors that affect that growth could be used to generate assumptions for the model.

Another important issue is developing assumptions about participation rates by insurance plan (i.e. the base from which to draw the increased use). Clearly, in any public plan the full effects of a liberalization of Medicare or Medicaid benefits would be experienced in a short time period; however, the full effects for a private insurance plan would be incremental as evidenced by the low participation rates of current long-term care insurance policies. The model's current assumptions about who will purchase private long-term care insurance seem adequate in determining the participation rates of a private plan.³ The probabilities that are used currently to determine who will use services among the policy holders can be multiplied by the induced demand effect; the non-service users would be subjected to this resulting probability. For example, if the probability of using services in the current model is 10 percent and the induced demand effect is 15 percent, then the policy holders who did not use services would be subjected to a probability of using services equal to 1.5 percent.

A fourth issue is how to measure the effect of induced demand on the length of use of those currently using the service. In particular, will a reduction in price cause persons in nursing homes or in home care to increase their length of stay or number of visits? Because additional home care and nursing home days must be considered medically necessary under Medicare and Medicaid, an increased length of stay or number of visits is uncertain. Therefore, in the short-run our focus will be on the effect of induced demand on the increase in the number of new users.

A fifth issue relates to the degree to which we should interpret the nursing home price elasticities found in the Nyman, Scanlon, and Chiswick studies. There are several issues to address. First, data from the Scanlon and Chiswick studies are old and may not reflect current behavior. Data from Nyman's study are limited because they are from one state, Wisconsin. Second, these studies are cross-sectional; therefore, as Newhouse points out (1981) the variety of price elasticity estimates may reflect true differences in specification in the populations sampled or may reflect differences in specification (p.89). A carefully designed experiment can test more directly whether price affects demand. The only disadvantage of an experiment or demonstration with an experimental design is the lack of methods to test whether behavior in an experimental situation differs from behavior in a true operational program. While the Rand experiment did not measure price elasticities for long-term care, the price elasticities found were much smaller than the elasticities found in the studies reviewed in this paper. Third, the Channelling demonstration found that informal caregiving did not reduce significantly during the demonstration. While this finding is viewed with caution due to the temporary nature of the demonstration, other studies support this finding (Greene, 1983 and Moscovice, et al, 1988).⁴ Fourth, according to the 1982 National Long-Term Care Survey, the majority of disabled elderly want to stay out of nursing homes. These four issues indicate that nursing home care may not be as sensitive to price as the three studies have suggested.

These issues lead us to conclude that there are two options we could undertake. The first option is to choose a price elasticity estimate from the range of price estimates and multiply it by (1 - copayment) or by the ratio of the old per them nursing home rate to the new per them rate to calculate the induced demand effect. This parameter would be included in the model and could vary by type of insurance plan. The price elasticity to be used could be the midpoint of the range of estimates found in the literature or a sensitivity analysis could be done with a low, medium, and high range to determine how much the results would vary based on the elasticity estimate used.

Because of the concerns raised above, a second option is to incorporate an elasticity of demand equation into the model. Before we attempt this approach, we would require a data source in order to estimate the elasticity of demand equation. The most appropriate data source to date is the National Medical Expenditures Survey (NMES) which contains accurate and complete expenditure information for individuals in the community and in institutions as well as demographic data such as age, sex, race, income, and living arrangements. The NMES data are recent (1987) and consist of a large nationally representative sample of elderly. A drawback to using this file is that only one year of data are available; therefore, longitudinal analyses are not possible.

NOTES

1. The model assumes these new admissions are based on the same pattern of nursing home admissions or formal home care use reflected in the entry probabilities (for nursing homes) and annual start probabilities (for formal home care) calculated using data from the 1982-1984 National Long-Term CareSurvey and the 1985 National Nursing Home Survey. Specifically, the entry probabilities are multiplied by the assumed percent increase in admissions, and those persons who had not entered a nursing home or received home care services as a result of the base case (non-induced demand) probabilities would be subjected to the additional probabilities.

2. Direction of effect is in parentheses.

3. We have proposed that group long-term care insurance participation start at zero percent of workers with pension coverage in 1989 and increase to 20 to 30 percent of workers in 2009. Growth rates for individual participation are under consideration.

4. Greene found a large substitution effect between informal and formal care with respect to community-based care. However, he points out that this effect may be due informal caregivers concentrating their efforts in fewer areas (a "specialization effect") rather than decrease their total effort. Moscovice, et al, found no empirical evidence for the hypothesis that formal care substitutes for informal care. The authors, like Greene, state that they do not know the extent to which informal caregivers specialize; therefore, variation in formal services might not be expected to influence strongly the quantity of informal support.

REFERENCES

Chiswick, B.R., 1976. "The Demand for Nursing Home Care: An Analysis of the Substitution between Institutional and Non-Institutional Care." Journal of Human Resources 11:295-316.

Christianson, J.B. , 1988. "The Effect of Channeling on Informal Caregiving." Health Services Research 23:99-117.

Corson, W., Grannemann, T., and Holden, N., 1988. "Formal Community Services under Channeling." Health Services Research 23:83-98.

Doty, P., Liu, K., and Wiener, J., 1985. "An Overview of Long-Term Care." Health Care Financing Review 6:69-78.

Greene, V.L., 1983. "Substitution Between Formally and Informally Provided Care for the Impaired Elderly in the Community." Medical Care 21:609-619.

Kane, R.A. and Kane, R.L., 1985. "The Feasibility of Universal Long-Term Care Benefits: Ideas from Canada." The New England Journal of Medicine 312:1357-1364.

Manning, W.G. et al., 1987. "Health Insurance and the Demand for Medical Care: Evidence from a Randomized Experiment." The American Economic Review 77:251-277.

Moscovice, I., Davidson, G., and McCaffrey, D., 1988. "Substitution of Formal and Informal Care for the Community-Based Elderly." Medical Care 26:971-981.

Newhouse, J.P., 1981. "The Demand for Medical Care Services: A Retrospect and Prospect." in J. van der Gaag and M. Perlman, eds., Health, Economics, and Health Economics. North- Holland Publishing Company, pp.85-100.

Nyman, J.A., 1985. "Prospective and 'Cost-Plus' Medicaid Reimbursement, Excess Medicaid Demand, and the Quality of Nursing Home Care." Journal of Health Economics 4:237-259.

Nyman, J.A., 1988. "Excess Demand, the Percentage of Medicaid Patients, and the Quality of Nursing Home Care." Journal of Human Resources 23:76-92.

Nyman, J.A., in press. "The Private Demand for Nursing Home Care." Journal of Health Economics.

Scanlon, W.J., 1980. “A Theory of the Nursing Home Market.” Inquiry 17:25-41.

Soldo, B.J., 1985. "In-Home Services for the Dependent Elderly." Research on Aging 7:281- 304.

APPENDIX A

MEMORANDUM

DATE: October 19, 1989
TO: John Drabek
FROM: Dave Kennell
SUBJECT: Disability and Income

Enclosed is a draft paper written by Polly Ericksen and Christie Provost on the relationship between income and disability.

After you have had a chance to review this paper, we can set up a time to discuss its implications for the model.

Please feel free to call me or Lisa at 842-8927 if you have questions about the paper.

IS INCOME A SIGNIFICANT PREDICTOR OF DISABILITY AMONG THE ELDERLY?

Submitted to: DHHS/ASPE/SSP
By: Polly Ericksen and Christie Provost, Lewin/ICF
October 17, 1989

INTRODUCTION

In the revised version of the Brookings/ICF Long Term Care Financing Model, disability status for the elderly is simulated in several steps. In the first year that an individual turns 65, he or she may be assigned a disability level.^{1, 2} At the start of each subsequent simulation year, the model simulates the disability status of every individual. During the year, different events occur which affect the number of disabled elderly persons:

• some persons become disabled;
• some disabled persons have increased disability;
• some disabled persons die;
• some disabled persons have decreased disability or recover from their disability; and
• some disabled persons age 64 turn age 65.

In the Brookings/ICF simulations, disabled individuals age 65 and over are defined as those who are unable to conduct at least one instrumental activity of daily living (doing heavy work, doing light work, preparing meals, shopping for groceries or other personal items, getting around inside, walking outside, managing money, and using the telephone) or unable to conduct at least any one of five activities of daily living (eating, bathing, dressing, toileting, and getting in and out of bed).³ In the model, an individual may be assigned one of four disability levels: 1) a deficiency in one or more instrumental activities of daily living (IADL only); 2) a deficiency in one activity of daily living (IADL); 3) a deficiency in two or more activities of daily living (2+ ADLs); or 4) no disability.

The model notes intra-year changes in disability status for persons who are admitted to or discharged from a nursing home or who start to use noninstitutional services. For all other persons in the model, changes in disability status are simulated at the start of each simulation year using the matrices in Table 1. The model takes these changes in disability status into account and simulates additional persons to become disabled to adjust for remissions from disability and death. The model uses the prevalence rates in Table 2 to ensure that there is a correct proportion of disabled persons in the community and simulates disability for additional persons, if necessary, to match the prevalence rates.

The prevalence rates vary by level of disability, age, and marital status and are assumed to hold constant over time.⁴ Analysts have wondered whether these rates also should vary by income level and other variables. There is evidence from previous research that people with lower incomes are more likely to be disabled, and, conversely, that people with disabilities have lower incomes.

Most of the research has examined the relationship between health status and retirement income. Much of the evidence suggests that poor health or functional limitations lead to low retirement income -- partly because persons with poor health work less than average over their lifetime and partly because disabled workers retire earlier. The causality between occupational status and health status is unclear, although poor health and low income can both increase with age. For examples of work in this area, see Burtless (1985), Kingson (1982), and Chapman, et al. (1986). Their work has relied on data bases such as the Retirement History Survey, the National Longitudinal Survey, and the New Beneficiary Survey.

There has been less research on the relationship between disability and income after retirement, as elderly people move into their seventies and eighties. Garber (1987) found no evidence of a decline in wealth with age or with an increase in the number of functional limitations, but he used only the 1982 National Long Term Care Survey (NLTCS) which is a one year cross- section and does not analyze persons' wealth over time. More recently, Weissert, et al., using several data sources, found that age, sex, race, and the percent of elderly in poverty in a geographic area were strong predictors of disability for elderly persons.

Disability and income are not directly linked in the model. However, 60 percent of Disability Insurance (DI) recipients at age 62 are simulated to continue to be disabled at 65. Since these people are defined to have work disabilities, in keeping with the eligibility requirements for DI benefits, their earnings and pensions are generally lower. In addition, the model does not permit disabled persons to be simulated to work. For all other elderly persons, disability and income are not directly related.

PURPOSE AND METHODOLOGY

The purpose of this paper is twofold: 1) to investigate whether income is a significant predictor of disability among the elderly; and 2) to investigate whether the model should simulate disability at age 65 as a function of income and/or other variables in addition to marital status and age (and previous receipt of DI benefits, for some persons).

We used data from Wave 3 of the 1984 panel of the Survey of Income and Program Participation (SIPP.) to conduct this analysis. Wave 3 is appropriate for this type of analysis because, in addition to the detailed questions on demographic characteristics and income from all sources included in the core questionnaire, Wave 3 also includes a topical module on work and disability. The SUP sample includes about 6,000 persons over age 65, which is more than adequate as a base for reliable estimates for elderly persons. For this analysis, we prepared an extract of Wave 3 which included age, sex, marital status, years of schooling, race, total income, pension income, employment earnings, income from assets, and a series of questions on disability status. (See Appendix 1 for the disability questions asked in SIPP.)

We made several further modifications to this data base. For most of the analysis we focussed only on persons age 65-67, because we felt SUP was most appropriate to investigate the prevalence rates of disability in the model, not the transition matrices.⁵ In the model, after 1979 the prevalence rates primarily are used to assign disability at age 65. About 1,200 persons age 65-67 were included in the Wave 3 sample; this is a large enough group to ensure the reliability of our estimates.

Because disability is an individual characteristic, we decided that our analysis should be done at the person level, not the family level. In order to do this we had to redefine the total income variable so that individuals who were married could each be assigned a portion of jointly received sources of income.⁶ In our analysis, we used two different variables: family income and individual income. For single people, individual and family income were the same. For married people, family income was the sum of both spouses, total income, and individual income was the sum of an individual's pension and earnings plus half of all jointly received income -- social security, SSI and asset income.

Finally, we developed a definition of disability status using the Wave 3 questionnaire that was comparable to the disability definitions used in the model. In our extract we included questions about functional limitations: requiring an aid or crutch to get around, walking a quarter of a mile, walking up stairs, carrying a 10 lb. object, getting around inside, getting around outside, and getting in and out of bed; and two areas of assistance, household management and personal care. Individuals could be measured as either having difficulty with these functional activities or as being unable to perform or needing the help of another person to perform these activities. We decided to base our proxy disability definition on the more severe degree of disability -- being unable to perform the above functions (called a severe functional limitation) -- and to classify individuals into four groups: those with no severe functional limitations, those with one severe functional limitation, those with two severe functional limitations, and those with three or more severe functional limitations. See Table 3 for the disability prevalence rates by age and marital status.

To analyze the relationship between disability and income, we used simple cross-tabulations. We also estimated a regression model, which is described below.

FINDINGS FROM CROSS TABULATIONS

Before estimating a regression equation for persons age 65-67, we looked at simple cross-tabulations of disability and several demographic characteristics: race, sex, marital status, education level, whether or not the individual had income from earnings, family income and personal income. We chose these variables because they are all currently available in the model (even though we do not use them all to model disability) and because we believed that these demographic variables also were related to income level.⁷

The tabulations for our sample are presented in Table 4. The variables we examined seem to have some degree of a relationship with disability level. In several instances, the trends for persons with no limitations are opposite the trends for persons with limitations. Also, for several variables the difference between persons with one limitation and persons with two limitations is greater than the difference between persons with two limitations and persons with three or more limitations. This finding indicates there may be certain factors with less influence as the severity of limitation increases.

The majority of persons with no limitations are married. Differences in the proportion of persons by marital status by disability level exist between persons with one limitation and persons with two limitations, but not for persons with three or more limitations.

Education has a strong influence on level of disability: 76.6 percent of persons with 13 or more years of schooling have no limitations versus 51.1 percent of persons with 0-8 years of schooling. Of the persons who have one or more limitations there are twice as many with only 0-8 years of schooling than there are with 13 or more years of schooling.

The differences by race also vary by disability level. A majority of persons with no limitations are white. Equal proportions of whites and nonwhites have one limitation. Non- whites are more than twice as likely to have two or more limitations than are whites. There is little variance in the proportion by sex.

Finally, we examined the relationship between disability and three different measures of income: family income, individual income, and whether the person received income from employment. For both family and individual income, the trends are consistent for all disability levels. Persons with $10,000 or more are much less likely to have one or more limitations than persons with less than $10,000. Over 70 percent of the persons with family or individual income over $10,000 had no limitations. The results for employment earnings are similar. Of persons age 65-67 with earnings, 20 percent are disabled, while 40 percent of those without earnings are disabled. Persons with no earnings are nearly three times as likely to have three or more limitations than people with earnings.

We have computed the relative likelihood of having one or more severe functional limitations as it varies by our different dichotomous variables in Table 5. For example, 50.4 percent of non-whites have one or more limitations, compared to 33.2 percent of whites. Thus, non-whites are 1.52 times as likely as whites to have one or more limitations. Variables related to income more than double the relative likelihood of disability. Table 5 also indicates that the relative likelihood of disability for persons with 0-8 years of schooling is 1.67 times the likelihood of persons with 13+ years of schooling. The relative likelihood of disability for persons with $0-$5,000 in individual income is 2.27 times the likelihood of persons with $10,000 or more.

REGRESSION RESULTS

The cross-tabulations proved helpful in formulating a regression model. Our basic equation regressed a categorical variable for being disabled on the demographic and income variables presented in Table 4.⁸ We used dummy variables for age 65, age 66, male, white, 0-8 years of schooling, 9-12 years of schooling, married, and receives employment earnings. The income variables were continuous; either total family income or the difference between individual income and earnings was used. The difference between individual income and earnings was used to estimate the relationship between disability and income after retirement. We used this variable to test the hypothesis that a history of disability may reduce individual retirement income. A separate model using continuous variables for both earnings and unearned income produced similar estimates for the income variables. However, we decided to test if the receipt of earnings influenced disability status and used the difference between the individual income and earnings. For the dependent variable we used 1+ severe limitations. The small proportions of disabled people with two or more limitations (18.1 percent of the cases) and three or more limitations (7.1 percent of the cases) in our sample made it infeasible to estimate accurately the influence of income on disability by disability levels.

We estimated two multivariate regressions, one with individual income and one with family income, using two techniques: an ordinary least squares (OLS) model and a logistic model. We used OLS because the linear function produces easily understandable coefficients. We supplemented this with a logistic model because a logit function constrains the fitted value of the dependent variable to fall between the 0 and 1 limits for probabilities, whereas a linear model permits estimated probabilities to exceed 1 or be less than 0. The logistic model assumes that the probability of the outcome of interest occurring is of the form:

P = e^xB/(l+e^xB)

where P = the probability of disability conditional on XB;
     XB = B_o + B₁x₁ + ... + B_kx_k;
     B_o = an estimated intercept for the model;
     and B₁,....,B_k are the estimated regression coefficients corresponding to the variables x₁,...x_k.

Results for all equations are presented in Table 6.

Because we only looked at 65-67 year olds, age was not a significant variable in our models. In a model that included individuals 65+, however, age had a significant effect on disability status. Sex also was not significant in our models. Although these variables were not significant for either equation, the regression results indicated that several of the other variables had a significant effect on disability status. All variables that were significant in the OLS model were significant in the logistic model.

A negative effect on disability as a result of marital status was significant at the 95 percent confidence interval in equation 2, using the individual income variable, but was not significant in equation 1, using the family income variable. This result may have occurred because the definition of individual income is based on marital status. The correlation coefficient between married and family income was .309. Income, therefore, may have absorbed some of the effect of marital status.

Having 9-12 years of schooling relative to 13+ years of schooling was not significant, but having 0-8 years of schooling relative to 13+ years of schooling was significant at the 95 percent confidence interval for both equations. The positive coefficient for 0-8 years of schooling indicated that less education was positively correlated with the likelihood of being disabled.

Race was significant in both equations. The negative coefficient for white indicated that white people were less likely to be disabled than non-whites.

Income was also significant in each equation. This was true for all three income-related variables, although the variable for the difference between individual income and earnings had a lower t- statistic. The negative coefficients for the three income variables indicated that the lower one's income, the higher the probability of being disabled.

We constructed a correlation matrix in order to see whether any of the variables were particularly highly correlated with one another, as this might have biased the regression results. The matrix in Table 7 indicates that none of the variables in the regression models were very highly correlated with one another.

IMPLICATIONS OF FINDINGS

These results suggest that there is a strong relationship between disability and income for persons age 65-67. The coefficients for all income-related variables were consistently negative and significant at the 95 percent confidence interval, implying elderly persons age 65-67 with lower income are more likely to be disabled. To illustrate the relationship we selected extreme cases of high disability risk and low disability risk and fluctuated individual and family income to determine the impact they have on the probability of being disabled. Using the logit regression estimates, the probabilities of disability with and without earnings and with different income levels were determined for the following cases:

The likelihood probabilities of disability substantially decreased with a $10,000 increase in income. Increases in both individual income and family income decreased the probability of disability in each case, with income having a greater influence on the high risk case. The receipt of earnings also has a substantial impact. In both cases the likelihood probability of disability decreased by at least .127. Again, the high risk case is more influenced by the receipt of earnings than the low risk case.

The marginal probabilities, provided by the OLS estimates, highlight the magnitude of influence the income variables had on disability status (see Table 6). For every $10,000 increase in family income and individual income, the probability of disability decreased by 3.5 percent and 6.3 percent respectively. A change in earnings status decreased the likelihood probability of disability by 15.1 percent for the model including family income and 19.6 percent for the model including individual income.

IMPLICATIONS FOR THE MODEL

The long-term care financing model currently captures the effect of earnings on disability status because disabled persons are not allowed to work. However, the model currently does not include individual or family income in determining disability status. Since both income variables prove to have significant influence on the likelihood of disability, the model's simulations of disability may be improved if these variables were included in the determination of disability status.

These results lead us to revisit the question of whether or not income is a significant predictor of nursing home care and home health care utilization. Previous analyses found that income was not a good predictor of long-term care use; however, these analyses only included the disabled population. The lack of variation in the disabled population's incomes could explain the statistically insignificant income variable in the model. This leads us to the tentative conclusion that there is a stronger relationship between disability and income than income and long- term care service use. We wish to discuss further with the Department the need for more research into this area and the possibility of assigning disability status based on income.

NOTES

1. This occurs in every year except 1979, when the model simulation starts. In this year, a disability status is simulated for all persons age 65 and over, based on the prevalence rates in Table 1.

2. 60 percent of individuals receiving Disability Insurance program benefits at age 62 are simulated to continue to be disabled when they reach age 65.

3. In the 1982-1984 NLTC Survey, disability was defined as the inability to conduct any of the Activities of Daily Living or Instrumental Activities of Daily Living due to a health condition which had or would endure for 90 days or more. The measure used in the model is based upon the control card in the NLTCS and therefore includes persons who require active human assistance, stand-by assistance, or assistance from special equipment.

4. These rates were calculated using data from the 1982-84 National Long Term Care Survey.

5. To investigate the transition matrices one needs a longitudinal data base.

6. In SIPP, jointly received income sources were assigned to the record of the spouse who was interviewed first.

7. Later in this section we present a correlation matrix for all of these variables.

8. Family income and individual income were used as substitutes for one another in separate equations.

BIBLIOGRAPHY

Burtless, Gary, "Occupational Effects on the Health and Work Capacity of Older Men." In Work, Health and Income Among the Elderly, edited by Gary Burtless, 103-150. Washington, D.C.: The Brookings Institution, 1987.

Chapman, Stewartt, Mitchell P. LaPlante, and Gail Wilensky. "Life Expectancy and Health Status of the Aged." Social Security Bulletin, Vol.49, No.10 (October 1986):24-48.

Garber, Alan M. "Long Term Care, Wealth, and Health of the Disabled Elderly Living in the Community." NBER Working Paper No.2328. Cambridge, MA: National Bureau of Economic Research, July 1987.

"Increasing the Social Security Retirement Age: Older Workers in Physically Demanding Occupations or Ill Health." Social Security Bulletin, Vol.49, No.10 (October 1986):5-23.

Kingson, Eric R. "The Health of Very Early Retirees." Social Security Bulletin, Vol.45, No.9 (September 1982):3-9.

Weissert, William G., et.al. Small Area Estimation of Dependency Final Report. Under ASPE Grant No.87ASPE181A. Photocopy. [http://aspe.hhs.gov/daltcp/reports/smareaes.htm]