| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Center for Policy Research, Syracuse University, New York.
2 Center for Policy Analysis and Management, University of Connecticut, Storrs.
Address correspondence to William Duncombe, Center for Policy Research, 426 Eggers Hall, Syracuse University, Syracuse, NY 13244-1090. E-mail: duncombe{at}maxwell.syr.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Methods. We place decisions regarding whether to move and where to locate in a single discrete-choice framework. We estimate an individual-level location-choice model by using a combination of place-characteristics data and Census county-to-county migration data for the period 1985–1990.
Results. We find that levels of tax burdens and public services can affect location decisions. Of the fiscal variables, income taxes have the largest relative effects. However, other factors, including climate, economic conditions, and population characteristics, appear to play much larger roles in migration and location decisions.
Discussion. Although income tax reductions might influence the location decisions of recent retirees, whether or not the economic benefits associated with these population gains outweigh the accompanying revenue losses is an important area for further research. A more cost-effective strategy may be for states to focus on marketing their amenities rather than using fiscal policy to recruit retirees.
Population aging in the United States has made elderly migration more significant in absolute terms, contributing to heightened interest in the migration and residential location decisions of elderly people. Some areas have become elder "magnets," attracting large numbers of recent retirees (Frey, Liaw, & Lin, 1999
). Competing claims have been made about whether these older migrants drive a need for additional government services (Bryant & El-Attar, 1984; Longino & Biggar, 1981) or enhance a local government's economy and tax base without imposing heavy demands on local services (Stallman, Deller, & Shields 1999; Longino & Crown, 1989; Woods, Miller, Voth, Song, & Jones, 1997). Some states have developed strategies for attracting retirees (Eckl, 1986; Reeder, 1998; Stallmann & Siegel, 1995) analogous to economic development programs aimed at attracting high-tech or manufacturing firms. Besides marketing their state's amenities, governments have planned, enacted, or encouraged a variety of tax and fee incentives to lure retirees.
Although the residential location decisions of retirees may have major impacts on the ability of local governments to raise revenue and provide services, there has been relatively little research on whether state and local governments can, in fact, influence older persons' migration decisions. Do taxes and public services influence retirees' residential decisions, or do amenity and place characteristics dominate such choices? We investigate whether fiscal factors influence the migration decisions of retirees, and we study the relative importance of fiscal and other factors in these decisions. We incorporate a broad range of amenity and cost factors found in the migration literature, and we examine the determinants of retiree migration by using county-to-county migration data for the period 1985–1990.
This article makes several contributions to research on retirement migration. We use a discrete-choice framework in which individuals choose from a large (3,068) set of potential locations. In our model, an individual's chances of moving to any particular location depend on the characteristics of all potential locations. We exploit this feature of the model by computing the expected change in each location's population in response to changes in its characteristics. The framework represents the decisions of nonmovers as well as those of movers. We also estimate differences by racial group and educational level in the effects of state and local attributes on migration and location decisions. Finally, although our model is formulated in terms of individual location choices, our use of grouped data permits us to use a large sample of origin counties, producing very precise estimates of tax, expenditure, and amenity effects.
Analytic Framework
Litwak and Longino (1987) presented a three-stage classification of retiree mobility, consisting of amenity-driven location decisions in early retirement, a subsequent move prompted by actual or pending health problems or the death of a spouse, and, possibly, ultimate relocation to an institution for full-time assistance. In this study we focus on the first stage of this process. States are likely to focus their recruitment campaigns on the "young old" who have higher incomes and lower service demands, and who are more apt to be influenced by state policies because several alternative destinations may provide similar amenities (Haas & Serow, 1993). Local amenities are thought to play a large role in first-stage migration relative to the proximity of family or friends. In fact, the first move often has the result that elders live farther away from their children (Clark & Wolf, 1992), which is important given that the principal data source for this study, the decennial census migration files, does not include information on the location of kin.
Much of the existing research on elder migration uses state- or county-level data on aggregate migration flows. Models typically include characteristics of the origin, destination, or both places, but they do not consider characteristics of nonselected alternatives including not moving at all (Assadian, 1995; Clark & Hunter, 1992; Conway & Houtenville, 2001; Kallan, 1993). Although these types of migration models provide information on migration patterns and their correlates, they provide an incomplete view of the residential location decisions of retirees. First, decisions of nonmovers are not considered, yet over 80% of 65- to 74-year-olds make no moves. Second, when retirees consider potential retirement destinations, the characteristics of many potential locations are relevant to their decision. Finally, many past studies focus exclusively on state-to-state migration patterns, but this obscures the significant within-state heterogeneity that exists, particularly in large states.
A more general approach is to model residential location decisions rather than migration. During the early stages of retirement, an individual faces decisions regarding whether and where to move. Like Pampel, Levin, Louviere, Meyer, and Rushton (1984), we view both decisions as part of a single process in which the characteristics of the present location are compared with various alternatives over an extended period (Haas & Serow, 1993). Because we observe only the outcomes of this process, we treat residential location as a single-choice problem. We develop a discrete-choice model of where individuals choose to live during the early years of retirement. We consider the decisions of nonmovers as well as movers, and we take account of the characteristics of all possible locations.
Past research on elderly migration has identified several correlates of location choices, which can be organized broadly into amenity and cost factors (Walters, 1994, 2002). Amenities include physical amenities, cultural amenities, and access to services. Physical amenities, such as a pleasant climate and proximity to recreational opportunities (e.g., seacoast), are often found to be important determinants of migration (Clark & Hunter, 1992; Graves, 1979). Crime rates have often been included as a negative amenity but have produced inconsistent results. Given the high demand of many elders for health services, studies have often included measures of health services, such as the numbers of physicians, hospital beds, and nursing home beds (Walters, 2002). Only a few migration studies have included measures of public services, typically in the form of category-specific per capita spending. Higher public welfare spending has generally been found to detract from a destination, whereas results for education spending and public safety are inconsistent (Clark & Hunter, 1992; Conway & Houtenville, 2001; Dresher, 1994).
Moving costs include both fixed and variable components. Distance has been included in most migration studies to reflect variable costs of moving, and it has consistently been shown to be negatively related to migration. Fixed moving costs, which include not only resource costs but also the psychic costs associated with leaving friends, family, and familiar surroundings, have not typically not been included in past migration research. High living costs, all else equal, should also make a location less attractive. Cost of living variations are closely linked to housing costs (McMahon, 1991) and taxes. Although we would expect higher tax burdens to reduce net migration, holding service levels constant, the impact of taxes is complicated by the fact that the elderly population may be more or less affected by various taxes (Mackey & Carter, 1995a, 1995b). Most studies find that higher per capita property, estate, and inheritance taxes make a place less attractive, but results for income taxes and sales taxes are quite varied (Assadian, 1995; Clark & Hunter, 1992; Conway & Houtenville, 2001; Dresher, 1994; Voss, Gunderson, & Manchin, 1996).
|
METHODS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Formally, we assume that potential migrant i's evaluation of the net benefits of living in location k, assessed from the perspective of the current location j, can be represented as
|
|
), Djk represents the distance between locations j and k (which equals 0 when 
), and eijk is a random factor. Given the definitions of Mjk and Djk, we see that 
0 and 1 represent the fixed and the variable costs, respectively, of moving. Following the conventions of discrete-choice theory, we assume that each decision maker chooses the location offering the greatest net benefits. A move to location k will occur only if 
.
We adopt the conventional assumption that all the elements of eijk for persons and location choices are independently extreme-value distributed. This implies, in turn, a standard logistic discrete-choice model for the probability of locating in each of 3,068 counties in the year 1990, given the 1985 county of residence:
|
|
. Thus we are able to estimate an individual-level location-choice model using group-level data by using weighted multinomial logit with the counts (the njk) serving as weights. The discrete-choice estimation software requires a row of data for each choice (i.e., the 3068 counties) facing individuals in each origin county. Physical constraints prevented us from using the full 3068 x 3068 data file. Instead, we defined 767 size strata by sorting counties by total population, grouping them into successive groups of four, and randomly selecting one county from each stratum. This strategy ensures that approximately 25% of the population is used in the estimation. Hypothesis tests demonstrate that the sample is not statistically different (at the.01 level) from the full population of counties for all the variables used in the analysis (a table is available from the authors upon request).
Interpreting coefficients in discrete-choice models is difficult, particularly in the presence of very large choice sets such as ours. Furthermore, the probability that an individual will move from one county to any other individual county over a 5-year period is extremely small. In order to develop a set of summary indices that permit us to determine the relative sizes of responses to changes in different county attributes, we expressed each county's expected 1990 population in terms of the model parameters, that is,
|
|
Data and Measures
Migration data
The decennial Census asks those enumerated to indicate their current place of residence as well as that of 5 years ago. These data are attractive because they identify origin and destination of migrants at the county level, include both migrants and nonmigrants, and provide a complete enumeration of survivors at the end of the 5-year period. However, these retrospective data may miss intermediate and return moves within the 5-year period, they fail to record moves made by persons who fail to survive to the data-collection date, and the values of possible correlates of moves are recorded at the end of, rather than at the beginning of, the time interval. Despite these shortcomings, we used the county-to-county (CTC) migration flow data from the 1990 census, because the data cover the entire population in 1990. The county-level data are disaggregated enough to eliminate much of the heterogeneity in place characteristics in a given location, yet most relevant location attributes are available at the county level. We confined our analysis to the 48 contiguous states plus the District of Columbia (treated as the equivalent of a county), but we discarded two counties from Texas (Loving and King Counties) that were found to be substantial outliers with respect to fiscal variables. These restrictions produced a total of 3,068 counties for our analysis.
The CTC data include cross-tabulations of migration with some demographic characteristics, permitting separate analyses of migration for each such group. To capture primarily first-stage migration, we analyze the migratory flows of persons aged 60–69 in 1985 (65–74 in 1990), estimating separate models for the six combinations of two racial categories (White and non-White) and three educational groups (not high school graduates, high school graduates, and college graduates). Education captures differences in tastes and information, and it also serves as a proxy for income level. Because these characteristics are fixed for most elders, postmigration measurement is not an issue. With respect to individual characteristics, Newbold (1996) and Chevan (1995) found that more-educated elders are significantly more likely to move. Kallan (1993) found that income is directly related to the likelihood of moving.
Place characteristics
Drawing from the growing literature on elderly migration summarized in Walters (2002), we assembled an extensive set of county attributes from sources such as the Census of Retail Trade, Census of Service Industries, Census of Governments, Consumer Expenditure Survey, Significant Features of Fiscal Federalism, and the County and City Data Book. Health system characteristics and population centroids came from the Area Resource File (ARF). Land features such as coastline and recreational areas were coded by hand with a standard road atlas. The names, sources, years, and descriptive statistics of all location-specific variables used in our analysis are presented in Table 1. Variables representing fiscal factors pertain to 1987, because the Census of Governments data source is only available for 1982 and 1987; the data for 1987 are likely to more accurately represent the fiscal environment facing potential migrants during 1985–1990.
|
There are two basic kinds of death taxes: estate taxes imposed on decedents' estates prior to disbursement to beneficiaries, and inheritance taxes assessed on beneficiaries in their state of residence. In general, elderly people can most directly control their estate tax burden because it applies to their state of residence. Each state's tax structure is different and interacts with the federal estate tax. In order to ensure variability in the death tax measures, we calculated effective estate and inheritance tax burdens by using the national average size of estates on which any tax was paid, $822,000. State and local income tax burdens on the elderly population can vary across states because of different tax rates and adjustments to the tax base, including treatment of pension income, personal exemptions for elderly people, and other special provisions (Mackey & Carter, 1995a). Without detailed information on household characteristics, it is not possible to estimate an effective tax rate for a particular household. Instead, we calculate the combined state and local effective rate (tax divided by income) for elderly filers with taxable incomes of $30,000. The same estate size and income level is used for all counties to calculate the tax variables to facilitate comparisons of comparable households across counties.
Our sales tax variable reflects both state and county sales tax rates. To proxy the average sales tax rate facing the elderly population, we applied these rates to the taxable portion of an average consumption bundle for a householder between 65 and 74 years of age. City tax rates were included if they were different than the county rate, in cases in which the county and city share common boundaries. Property tax rates relative to the market value of the property are not available at a national level. Instead, we used data from the Census of Governments on property tax revenues and divided this by the number of households in a county to obtain the average property tax per household. Measures of property tax relief for senior citizens were included in earlier versions of the model, but they were not generally found to be statistically significant.
Ideally, public services should be measured by the quantity and quality of the services provided in a given county. Unfortunately, measures of public service outputs are rarely available. Spending per household provides a rough proxy for the level of public services. For public safety and education, we used per household spending by all local governments in the county. For public welfare, public housing, and recreation, we combined per household county spending with average state spending per household. These variables capture the categories of expenditures most frequently mentioned as being of interest to elderly people.
We included a range of variables to capture desirable amenities for the elderly population. We used three measures of climate: (a) the average number of clear days, which captures both sunlight and precipitation; (b) the number of heating degree days, which measures how cold a place is; and (c) the relative humidity in July, used as a measure of comfort. These data come from the National Oceanic and Atmospheric Administration (1987) and pertain to each county's nearest weather station. To measure physical amenities, we have included dummy variables representing whether the county abuts the coast or has one or more recreational lakes. To capture the possible existence of urban services attractive to elderly people, we have included measures of urbanization (population density and the percentage of the population living in urban areas). The tendency of migrants to seek locations with similar demographic groups is measured by the percentage of the population aged 65 to 74 years old, and the percentage of population that is non-White.
To represent the health care services available in a county, we utilize the extensive data on physicians and care facilities available in the ARF. Specifically, we include measures of the number of medical specialists per household, the number of nursing home beds per capita, and the percentage of a comprehensive list of hospital services that are offered in the county.
As housing represents one of the principal differences in cost of living across counties, we include median house value as a rough proxy for housing prices (McMahon, 1991). Additional factors that might influence an area's attractiveness include indicators of various social problems such as the housing unit vacancy rate, the unemployment rate, the percentage of female-headed households, teen births per capita, and the crime rate. To capture the most visible and threatening crimes facing potential migrants, we use the violent crime rate; we allow for a possible nonlinear relationship by including the square of this variable. Because housing vacancy rates and unemployment rates might be negatively related to housing prices, the relationship between these variables and the attractiveness of a destination is ambiguous.
To capture the fixed costs of moving, we have included a dummy variable [Mjk in Equation (1)] indicating that the origin and potential destination counties are different. The variable costs of moving are reflected in the distance between origin and destination counties. We computed the distance, in miles, between the population centroids of all pairs of counties. To allow for a nonlinear relationship between distance and cost, we used the natural log of distance in the model (Newbold, 1996). All models also include dummy variables indicating 8 (of 9) census regions, although we do not report the coefficients of those variables.
|
RESULTS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
|
How does education (and therefore income) affect responses to tax burdens? We expect that more-educated individuals would be more repelled by all taxes except sales taxes, because they are expected to bear a higher burden of these taxes. For Whites, the expected patterns emerge only with respect to inheritance and sales taxes. The coefficient on income taxes for college-educated people is half of that for those that are not college graduates. The differences in the coefficient across education groups for non-Whites fit the expected pattern for income and inheritance taxes. Except for estate taxes, there are no statistically significant differences between non-White high school graduates and those without high school diplomas.
Turning to public services, we would expect that services heavily used or valued by elderly people should attract them to a location. Commonly mentioned in this category are public safety, recreation, and health services (to be discussed with amenities). Public education or public welfare programs might, however, fall in the category of services not used or valued by many of the elderly group (Button, 1992). Our results provide support for the view that some public services affect elderly migration decisions. As expected, there is a positive relationship between the level of public safety expenditures per household and the choice of location. Surprisingly, spending on recreation services is not valued by elders. It may be that the mix of recreation services provided by most local governments focuses on youth rather than senior citizens. Higher spending on public welfare appears to deter migration to a location among Whites and non-Whites in every education group. Surprisingly, for those in the one group that might potentially benefit from public housing, being in a lower income household (as proxied by education) is not positively associated with migration into a county. In contrast, the housing coefficients for high school graduates and college graduates are positive for Whites and are mixed for non-Whites.
Finally, the coefficients on public education expenditures and school enrollment provide evidence that White elders generally avoid locations with large school-age populations and high education spending. There are no significant differences in the negative education spending coefficients between education groups, and the coefficient on school enrollment actually becomes more negative with higher education. For non-Whites the impact of education on location appears to be less important, as four out of the six coefficients are not statistically significant from zero. The two significant coefficients are negative. The results for Whites, in particular, are consistent with recent publicity about elders voting against school budget increases in Florida and Arizona (Keller, 2000).
Older individuals are uniformly attracted to warmer climates. Non-Whites are attracted generally to sunnier climates with lower humidity, whereas the opposite appears to be the case for college-educated White seniors. Recreational lakes are viewed as a positive amenity in all groups, but, surprisingly, only White college-educated elders appear to be attracted to counties on the seacoast. Non-Whites in general, and Whites without college degrees, are attracted to urban counties. Less densely populated urban areas, such as suburban counties, appear attractive to college-educated Whites. Consistent with findings from previous research, retirees are generally attracted to locations with a higher percentage of older households, whereas Whites appear to avoid areas with relatively large non-White populations.
Among the health service variables, elders appear to be attracted to destinations with a broader range of hospital services, and the importance of these services generally goes up with education. Greater per capita availability of nursing home beds is negatively related to location decisions. It is likely that recent retirees are not considering nursing home facilities in making their migration decisions, but it is not clear why such facilities would be a negative amenity.
The uncertainty associated with moving to a new location imposes psychological costs on the potential movers. To minimize this uncertainty, people may look for cues about locations with negative amenities factors, such as rates crime, unemployment, teenage pregnancy, housing vacancy, and single-parent households. A complex story emerges with regard to these variables, whose effects vary by race and educational attainment. Places with a high unemployment rate are less attractive to those in most racial and education groups, but another indicator of economic vitality, the housing vacancy rate, is positively related to in-migration for Whites (when housing prices are held constant). A larger share of female-headed households is a strong deterrent to Whites, but for non-Whites this is positively associated with location in a county. To capture a potential nonlinearity in the relationship between crime and migration to a location, we include the violent crime rate and its square in the equation. For all groups, the crime has an inverted U shape with a high inflexion point, indicating crime becomes a deterrent only for the highest crime rates. It is possible that the results for crime are partially capturing omitted urban variables. Another location-specific disamenity is high living costs. We assume that housing values serve as a rough proxy for living costs, and we find that higher housing costs do not have a large impact on location decisions. Housing prices may be picking up both cost of living differences and differences in positive amenities across locations.
Moving imposes tangible and intangible costs on the migrating household. Consistent with past research, we find that higher variable costs of moving, reflected in distance, lower the probability of a move to a particular destination. The effect of distance goes down with education, which is consistent with previous research findings that higher income and more-educated elders are more likely to make interstate moves. The opposite pattern emerges for the fixed costs of moving, which rise with education. Higher income elders may have stronger roots in a community, and thus have more difficulty, all else equal, making the decision to move.
Relative Importance of Fiscal Factors in Location Decisions
The findings from our model of retiree location decisions indicate that taxes and expenditures have statistically significant effects on the county locations chosen by persons of retirement age for most education and race groups. The signs on these variables generally fit our expectations—taxes discourage migration, and expenditures have a mixed effect depending on the service. Of more relevance for policy is the practical significance of our results. How large a decrease in taxes is required to attract more retirees to a county? To assess this we have calculated elasticities indicating the percent change in total retirement-age population in a county that are due to 1% changes in selected independent variables. Table 4 reports both the average elasticity over all counties and selected points in the distribution of responses.
|
|
DISCUSSION |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
The objective of this article has been to add to the growing research on elderly migration by focusing directly on the influence of state and local fiscal policies. We have made several contributions to the existing literature. Our model places the choices of whether and where to move into an integrated framework that allows estimates of the impact of policy changes in one location on the elderly migration flows to all other locations. We estimated this model by using data on a broad array of amenity, cost, and fiscal variables that may affect the location decisions of people of retirement age. Instead of focusing solely on interstate migration, we provide a more disaggregated view of migration decisions by examining county-to-county migration flows. Migration was further decomposed by race and educational attainment.
Our findings regarding fiscal variables generally conform to expectations. Higher tax rates, particularly for income tax, are associated with less in-migration. Public safety spending is generally viewed positively, and welfare and education spending are viewed negatively by people in this age group. White retirees are also generally deterred from locations with a high relative public school enrollment.
Although state and local fiscal policies do appear to influence location decisions, except for income taxes they are secondary in the migration decisions of most households. A 1% increase in income tax rates may reduce a population by up to 1% in some locations. Significant changes in other taxes and public service spending are associated with relatively small changes in population. In contrast, location characteristics well outside the realm of government intervention (such as climate) or only slightly subject to policy influence (such as housing costs and population characteristics) are much more influential in migration and location decisions.
What are the implications of our findings for state policy makers contemplating strategies for attracting elder migrants? Changes in tax burdens and service levels will generally have small effects on migration, except for possibly income tax. Thus, if a fiscal tool is to be used to recruit footloose elders, income tax appears to be best choice. However, very large income tax reductions would be required to generate significant increases in elder populations within most states, even if they were targeted to only the elderly population. A key area for future research is to examine the fiscal benefits and costs of large age-based income tax reductions. A lower cost alternative might be for states to focus on marketing their amenities to potential migrants, rather than using fiscal policy to recruit retirees.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Received for publication September 27, 2002. Accepted for publication February 4, 2003.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  K. S. Conway and J. C. Rork Income Tax Preferences for the Elderly Public Finance Review, September 1, 2008; 36(5): 523 - 562. [Abstract] [PDF]
|
|
|
|
 |
|
 J. D. Fisher, D. S. Johnson, J. T. Marchand, T. M. Smeeding, and B. B. Torrey No Place Like Home: Older Adults and Their Housing J. Gerontol. B. Psychol. Sci. Soc. Sci., March 1, 2007; 62(2): S120 - S128. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
