Conway-paper_for_Oct08 by SabeerAli1


									                     NO COUNTRY FOR OLD MEN (OR WOMEN) --

Karen Smith Conway                                   Jonathan C. Rork
Department of Economics                              Department of Economics
University of New Hampshire                          Andrew Young School of Policy Studies
15 College Road                                      and Gerontology Institute
Durham NH 03824                                      Georgia State University
                                                     PO Box 3992
                                                     Atlanta, GA 30302                           

                                                  October 2008
                      Preliminary Draft – please do not quote without permission

For presentation at the 2008 National Tax Association Meetings and Southern Economic
Association Meetings.

This project was supported by Grant Number 5R03AG028479-02 from the National Institutes of Health. Its
contents are solely the responsibility of the authors and do not necessarily represent the official views of NIH. We
thank Josh Stillwagon for his fine research assistance. We are also indebted to Jon Bakija, who generously made his
data on state income tax preferences available to us under the conditions of this grant, and to Edward Coffield, who
helped us organize the tax information in a coherent manner.
1. Introduction

Earlier this year, the state of Missouri enacted a new income tax deduction for elderly taxpayers

who receive social security benefits or other retirement income; Georgia enacted a similar

deduction in 2006 and Iowa is currently phasing out its tax on social security benefits. These

states are far from isolated cases. The number and type of income tax breaks offered to elderly

taxpayers by states has grown steadily over the last forty years (Zahn and Gold 1985, Mackey

and Carter 1994, Conway and Rork 2008b,c). During this same time period, a large majority of

states have also reduced or eliminated their so-called EIG or „death‟ taxes -- taxes on estates,

inheritances and gifts (hence, EIG) – taxes which are borne disproportionately by the elderly

(Conway and Rork 2004).

       While „fairness‟ concerns are often given as one rationale for such policies, an attempt to

recruit the elderly or a fear that high taxes will drive them away are also offered as reasons. For

instance in the case of Missouri, the proponents argued that taxpayers have „already paid taxes

on their contributions‟ to social security and that dramatically increasing numbers are having to

pay taxes on such benefits, including those suffering from kidney failure. However, they also

emphasized the role such taxes may play in location decisions:

       “There is active recruitment by some states to lure retirees to locate in other
       states. A state‟s tax burden is one of the top three reasons to move to a state.”
       (, accessed

Opponents, which included AARP, argued strongly against the equity argument, noting

that single (married) households with incomes less than $25,000 ($32,000) would see no

benefit from the tax break and that such benefits would benefit the wealthy. The concern

that taxes drive away the elderly was left unaddressed. Similar policy debates have taken

or are taking place across the country.1

         Are state tax policies a major consideration in elderly location decisions?

Certainly, the information about tax policies appears readily available and widely

dispersed. The AARP publishes regularly a handbook that summarizes each state‟s

major tax policies and their implications for retirees (State Handbook of Economic,

Demographic and Fiscal Indicators).2 Federal employees are briefed on the differences

in state tax treatment in their retirement planning seminars.3               Numerous web sites, such

  A quick internet search leads to several politicians making these arguments. Among these are Dave Nabity, a
former candidate for Nebraska governor, whose 2004 campaign web site asserts “Unfortunately, one of our most
cherished treasures – our retirees – are leaving in favor of more tax-friendly states,”, accessed 9/25/08). The legislation introduced by
Governor Perdue that reduced taxes on retirees was justified similarly in the flagship Georgia newspaper: “‟This tax
cut will allow seniors to better cover the costs of prescription drugs and health care, or spend more time with their
families,‟ Perdue said in a statement. „It will help attract retirees to our state and make our economy even stronger.‟
Perdue has sought to cut or eliminate retirement income taxes to help the (sic) Georgia compete with states like
Florida, which doesn‟t have an income tax,” (
blogs/ajc/georgia/entries/2007/01/29/bill_eliminates.html, accessed 9/25/08). The Wisconsin Taxpayer‟s Alliance
(WISTAX) made a similar argument that weather isn‟t the only reason retirees are leaving Wisconsin: “However,
WISTAX researchers note that taxes may also have played a role, particularly for the more „well-to-do.‟ While
Florida and Arizona were the preferred destinations, about 50% more retirees went to Florida, a state with no
income tax. In fact, the study found that the top nine states gaining the most high-income Wisconsinites all had
income and property taxes 25% to 73% lower than Wisconsin‟s,”
(, accessed 9/25/08). States that continue to tax retirees face
a great deal pressure, as represented by an article in Gray Times (an online resource for Minnesota seniors)
criticizing Minnesota‟s tax policy and publicizing the recent reductions in other states: “According to a recent article
in Where to Retire magazine, if taxation is your only factor in choosing a place to live, you can probably do better
than Minnesota. While there are a few breaks for Minnesota retirees in the tax code, the state doesn't bend over
backwards to keep or attract retirees through incentives to the extent some others do… While Minnesota treats
retirees similarly to other taxpayers and has an income tax, many other states are aggressively providing incentives
to retain their retirees or draw them from other states. For example, Missouri is phasing in a full exemption for
pensions from its income tax. Georgia is also increasing the amount of retirement income a taxpayer can exclude on
income tax returns. Iowa will phase in a full exemption for Social Security by 2014,”, accessed 9/25/08). This entire set of quotations
was found in less than 20 minutes by simply searching on „state taxes on retirees,‟ sometimes also including a
particular state‟s name.
 The 2008 edition (reporting data for 2006) is the most current edition at this writing. At least five earlier editions
of this handbook exist as well, one for 1996, 1998, 2000, 2003 and 2006.
  For example, federal employees in retirement planning seminars are given a workbook that suggests moving to a
lower cost location (both in terms of housing costs and taxes) as a strategy for stretching retirement income and
as, and investment reports, such as “Retiree Tax Heaven (and Hell)”

(published August 2008 in Kiplingers), offer up to date reports and advice on the tax

consequences of location. Despite widespread changes in state tax policies towards the

elderly and improved information about such policies, however, the overall interstate

migration rate of the elderly is remarkably constant at around 4-5% and the patterns of

movement are very persistent (Conway and Rork 2008a). However, there is some

evidence that elderly migration is becoming less concentrated (Longino and Bradley

2003, Conway and Rork 2008a) and that the elderly may have become more mobile

relative to other age groups (Wolf and Longino 2005).

        Most past econometric research suggests state tax policies do affect elderly

migration, albeit in often modest or counterintuitive ways (see, for example, Duncombe,

et al 2001, 2003; Conway and Houtenville 1998, 2001, 2003; and Bakija and Slemrod

2004). However, most suffer from at least two shortcomings. Foremost, the vast

majority are cross-sectional in nature.4 The persistence of elderly migration patterns that

is readily apparent from census data (Flynn et al 1985, Longino and Bradley 2002,

Conway and Rork 2008a) combined with the likelihood of important state-specific

unobservables (such as cultural and natural amenities) make the probability of a spurious

correlation between tax policy and migration likely. For example, two states that have

historically enjoyed a disproportionately large influx of the elderly are Florida and

Nevada; neither state has an income tax. Is the lack of an income tax really the reason?

includes a worksheet to calculate costs at multiple locations. See Part 3 and worksheet 8 in Guidebook to Help Late
Savers Prepare for Retirement, Workbook 1, National Endowment for Financial Education, 2003.
 Two notable exceptions, discussed in greater detail shortly, are Bakija and Slemrod (2004), and Conway and Rork
(2006). The latter study explicitly addresses the perils of using only cross-sectional data in studying migration
         If tax policies are an important consideration to the elderly in their location

decisions, then their migration behaviors should change in response to policy changes as

well – i.e., the effects should remain in panel analyses that control for unobservable state

characteristics. The one existing panel analysis of elderly migration data, however, found

no statistically significant evidence that EIG taxes (or any other state taxes) affect elderly

migration (Conway and Rork 2006).

        The second shortcoming is that most focus only on income tax policies in general

(such as the average burden or an estimated tax rate) rather than also considering income

tax preferences specifically targeting the elderly – such as the policies enacted by

Missouri, Georgia and Iowa.5 Given that these policies have changed a great deal and are

specifically yet fairly universally targeted at the elderly, they seem a natural policy tool

for recruiting the elderly – and have been justified on that basis, as noted above. And,

unlike EIG taxes that only affect the very wealthiest individuals – and only after their

death -- elderly income tax preferences affect a much wider range of individuals and do

so every year. Such preferences therefore seem far more likely to affect elderly

migration than EIG taxes, which have received greater attention in the literature.

        Income tax preferences also provide a crisp way of isolating possible effects of

income tax policy on elderly migration. Such preferences seem less likely to capture the

effects of other elements of the state‟s government than general income tax burdens do.

For example, states with a heavy reliance on income taxes may be different in many ways

from states with smaller income taxes, such as a greater desire for progressivity or a

  Conway and Houtenville (1998, 2001, 2003) are notable exceptions in that they include in their models the amount
of pension income exempted in each state in addition to the usual income tax variables of estimated marginal tax
rate and income tax liability. However, other preferences such as extra deductions, exemptions or favorable
treatment of social security benefits are excluded. Moreover, these studies are all cross-sectional.
larger public sector, or the monopoly power a state may have over high skill workers who

benefit from urban agglomeration effects (Bakija and Slemrod 2004). Elderly tax

preferences are likely less correlated with such unobservables, especially over time and

once the reliance on income taxes more generally is controlled for. They also present an

opportunity to search for spurious correlation by estimating whether such preferences

have an effect on the location decisions of the young or the low income elderly (who pay

little or no income taxes even without the preferences). While elderly and non-elderly

migration behaviors are certainly linked (for example, the elderly may move to be closer

to their adult children), one expects elderly tax preferences to have a stronger effect on

elderly migrants than non-elderly ones, ceteris paribus. Moreover, high income elderly

migrants should be affected the most strongly.

          The research presented here addresses both shortcomings. Using state-to-state migration

flows from four different census years (spanning 1965-70 to 1995-2000) and information on

state fiscal policies – especially those targeting the elderly – as well as other state characteristics,

we investigate whether elderly location decisions respond to changes in state tax policies. By

estimating panel migration flow models that allow for unobservable state- and even flow-specific

effects for both elderly and non-elderly, we isolate the unique, differential effect that state taxes

have on elderly migration. In contrast to the assertions made by policy makers and advocates,

our research to date finds no consistent evidence that elderly migration is affected by state tax


2. Background

Our analysis combines the wisdom from and contributes to two strands of research – (1)

empirical studies that estimate the role that fiscal policies and other factors play in elderly
migration behavior, and (2) mostly descriptive studies that document the different types of state

income tax preferences and their effects on estimated tax burdens, and that discuss how and

perhaps why such preferences have changed over time.

A. Past Elderly Migration Research

Patterns, motives and determinants of elderly migration have been studied extensively in other

disciplines, especially gerontology (e.g., see Walters 2002 for a survey). The effect of state

fiscal policy on elderly location decisions has received a small, but steadily growing, amount of

attention from economists. These studies have taken one of two empirical approaches. The first

approach uses individual level data and discrete choice models to estimate the effects of location

characteristics on the decision to move. Examples include Dresher (1994) and Duncombe et al

(2001), (2003), two studies that use a conditional logit framework and compare the

characteristics of a random sample of locations with the one actually chosen.6 Such an approach

has the advantage of being able to control for individual characteristics, but also has significant

limitations. The sample sizes tend to be small (e.g., Dresher‟s analysis using the PSID contained

91 elderly movers) and the time frame considered is short (e.g., Dresher 1994 considered moves

during 1983-88; Duncombe et al 2001, 2003 considered 1995-2000). Furthermore, the empirical

approach makes it very difficult, if not impossible, to control for unobservable state

characteristics. To our knowledge, no study has attempted to control for these unobservables.

         The second approach uses aggregate data to measure patterns of migration. Some

studies, such as Cebula (1990), Conway and Houtenville (1998), and Gale and Heath (2000), use

inter-state migration rates – i.e., how many elderly individuals are moving into (in-migration) or

 A similar, but slightly different, approach is taken by Farnham and Sevak (2006). They use longitudinal data to
estimate the effects of moving on the local fiscal bundle facing the household. Their emphasis is primarily on local
policy and therefore is not as relevant for our research.
out of (out-migration) each state. (Net in-migration is in-migration minus out-migration.)

Others, such as Voss et al (1988) and Conway and Houtenville (2001, 2003), use state-to-state

migration flow data (e.g. how many individuals moved from Alabama to Arkansas), which is

clearly richer data because it includes information on both the origin and destination state. These

studies, too, have focused on a narrow time period, typically the 5 year period spanned by

questions on mobility from a single US census.7 These studies have frequently been plagued by

the „same sign‟ problem, in which a characteristic affects migration decisions in a logically

inconsistent manner. Using migration rate data, this problem manifests itself when coefficients

are the same sign in both in-migration and out-migration equations; with migration flow data, it

shows up as origin and destination coefficients having the same sign. For example, EIG taxes

are frequently found to have a negative effect on both in-migration and out-migration (Conway

and Rork 2006) and, in analyses using flow data, have negative coefficients for both the origin

and destination (Voss et al 1988, Conway and Houtenville 2001).

         Both types of studies – individual level and aggregate – tend to find that state taxes exert

a negative effect on elderly migration. However, a recent study suggests that these findings do

not stand up when one allows for state unobservables. Conway and Rork (2006) employ

migration rate data from four different censuses and find that the negative effect of EIG taxes

frequently found in cross-sectional studies completely disappears when panel data and

techniques are used. The effects of other state policies are similarly affected. Moreover, their

approach, which combines panel analysis with a difference-in-difference approach that uses the

non-elderly as a pseudo control group, largely eliminates the „same sign‟ problem.

  Migration is inferred in census data by comparing the individual‟s current residence to the residence s/he reports
living in 5 years prior; it therefore measures movement during the 5 years prior to the census year (e.g., 1995-2000
migration is measured in the 2000 census).
         The only other panel study of which we are aware is Bakija and Slemrod (2004), who use

federal estate tax return data, tallied by wealth class, state and year, as a proxy for the location

decisions of the rich (and likely elderly). The authors develop and then use sophisticated state

income and EIG tax calculators to create the combined federal-state tax burdens facing a

representative taxpayer from each wealth class, compared to the burden faced if there were no

state income/incremental EIG tax.8 In contrast to Conway and Rork (2006), they find that state

EIG and income taxes discourage the rich from locating in the state.9 The differing conclusions

between the two studies make sense. Conway and Rork (2006) study the aggregate movement of

all elderly, the vast majority of whom will pay no EIG taxes upon death and may enjoy

significant advantages from income taxation rather than other forms. Bakija and Slemrod

(2004) focus on the very top (typically far less than 10%) of the income distribution, a group for

whom EIG and income tax considerations are likely to be far more important. Such a group is

also much more likely to own more than one home and thus have greater flexibility in choosing a

legal residence.

         Our reading of current policy debates is that states are not just attempting to recruit the

very rich elderly; rather, they view the middle class elderly as a possible growth industry, a

group that stimulates local demand for goods and services while placing little strain on local

public services (e.g., see Wisconsin Policy Research Institute Report 2006 as well as footnote 1).

This group is not likely to face state or federal EIG taxes, especially in recent years.10 Income

 These tax calculators are remarkably detailed, but not currently publicly available. See Bakija (2008) for further
discussion of the income tax calculator.
  Note that their framework is similar to that of net in-migration studies in that there is no possibility of a „same
sign‟ problem. In essence, the federal estates observed in a given state in a given year reflect the net migration of
rich individuals in the past.
  See Bakija and Slemrod 2004 and Conway and Rork 2004, 2006 for discussion of how these taxes have changed
over the past 40-50 years and how the proportion of the income distribution facing these taxes has diminished.
taxes are likely to be a much more important consideration; however, the absence or minimal

burden of an income tax may not, by itself, be attractive to the elderly. States still must fund

their programs and it could be that alternative funding mechanisms, such as fees and other taxes,

are actually more burdensome to the elderly. For this reason, studies typically attempt to control

for other types of state and local taxes and, sometimes, the expenditure side as well. Still, it is

difficult to argue that one can successfully control for all of the other characteristics of a state‟s

public sector and overall environment that are likely associated with the income tax burden it


B. A Primer on Income Tax Preferences for the Elderly

        Income tax preferences for the elderly – such as those recently enacted by Missouri,

Georgia and several other states – present a unique opportunity to examine whether the middle

income and upper middle income elderly respond to tax incentives. Unlike the overall income

tax burden, the value of targeted income tax preferences should be unambiguously attractive to

the elderly. While such preferences certainly have revenue consequences for the state (e.g.,

Bernstein 2004), they seem less likely to be associated with other unobserved aspects of the

public sector and their consequences are much smaller than less narrow tax provisions (such as

top marginal tax rate). A heavy reliance on income taxes could be either attractive or not to an

elderly taxpayer depending on her economic status and the revenue alternatives, but an extra

exemption given based on age should be unambiguously desirable. Such preferences also tend to

be easily visible and publicized; for instance, they are reported prominently in the AARP

Handbook. As such, they may send a strong signal of a state‟s desirable treatment of the elderly

in general. Yet, these preferences have been mostly ignored in past elderly migration research. 11

         Perhaps states have recognized this, as they seem increasingly likely to offer such

preferences -- in stark contrast to the federal government which has reduced its preferential

treatment of elderly taxpayers since the early 1980‟s (Conway and Rork 2008b). As discussed in

greater detail elsewhere, these income tax preferences typically take three different forms.12 The

first is an extra deduction, exemption or tax credit (henceforth called deduction) given solely on

the basis of age (typically over age 65). The federal government and nearly all states with broad-

based income taxes grant such deductions.13 The Tax Reform Act of 1986 substantially reduced

the relative size of the federal deduction, and most states followed suit.

         The second is the favorable tax treatment of social security benefits. Until the 1983

federal legislation that began taxing a portion of Social Security benefits for high income

households, such benefits had not been subject to federal or state taxation. Beginning in 1983,

up to half of social security benefits were subject to federal income taxation for high income

   Rather, most studies include an overall measure of the income tax burden, such as an effective average tax rate or
tax liability. Two exceptions are Conway and Houtenville (2001, 2003), but only the exemption for pension income
is considered and, as noted above, their analyses are limited to one cross-section (the 1990 census).
  See Zahn and Gold (1985) and Nelson (1983) for discussion of early income tax preferences for the elderly. More
recent work includes Penner (2000), Edwards and Wallace (2004) and Manzi et al (2005), as well as Conway and
Rork (2008b,c) who are the first to examine the changes in federal and state income tax preferences over time.
Conway and Rork (2008b) focus on recent policy; they report these preferences in isolated years and estimate the
differential tax burdens facing representative elderly vs. non-elderly households from 1977-2002 using CPS data and
the NBER‟s TAXSIM calculator. Conway and Rork (2008c) trace the historical beginnings of both federal and state
income tax preferences and find that while such preferences began in the 1930‟s, they did not become widespread
until the mid-1950‟s. Furthermore and important for this research, they find that state income tax preferences have
changed substantially throughout their history and in largely unsystematic ways.
   We limit our discussion to the 40 states, plus the District of Columbia, that had broad-based income tax systems
for most of the period under study. We include states that enacted income tax systems during the period (e.g.,
Illinois, Pennsylvania, Rhode Island, etc.) but exclude those with a very narrow base (i.e., New Hampshire and
households, and 13 states followed suit. 14,15 In 1993, an additional set of (higher) thresholds

were imposed past which up to 85% of social security benefits could be taxed. Unlike the extra

deduction, this appears to be a policy in flux at the state level. The number of states that tax

social security benefits grew to a high of 18 states in the early 1990‟s (and the most recent year

relevant for our empirical analysis), but has been declining since. According to the AARP 2008

Handbook, 15 states continue to tax social security benefits in 2006; however, two of those

(Colorado and Connecticut) offer more favorable treatment than the federal government and

another three (Iowa, Missouri and Wisconsin) are phasing out the taxation of these benefits. The

debate in Missouri clearly highlights the role of tax incentives in retiree location decisions; yet,

no formal analysis has been performed to see if elderly tax preferences actually influence

migration behavior.

         The third and perhaps most far-reaching preference is an exemption for private pension

income. (Most public pensions are at least partially exempt from both federal and state income

taxation.) The federal government has always taxed private pension income, but beginning in

the 1950‟s states began enacting exemptions for retirement income, especially pension income

(Conway and Rork 2008c). The number and size of these exemptions has grown dramatically in

recent years. In 1964, the first year we consider in our empirical analysis, two states exempted

   The income thresholds are $25,000 for single households and $32,000 for married ones; however, income refers
to combined income, which is all forms of taxable income plus one-half of social security benefits received. See
Page and Conway (2007) for a detailed discussion of the federal and state income tax policies regarding taxation of
social security benefits, as well as the likely behavioral effects and the prevalence of the tax. One state that did tax
social security benefits prior to the 1983 law was Mississippi, but it has fully exempted benefits since 1980 (Conway
and Rork 2008c, Table 2). In our subsequent empirical analyses, we investigate the robustness of our results to our
treatment of Mississippi during this early period and find it makes no qualitative difference.
  Three states, Nebraska, Rhode Island and Vermont, have income tax systems based on the federal income tax and
so automatically began taxing benefits as well. Ten other states -- Colorado, Iowa, Kansas, Minnesota, Missouri,
Montana, North Dakota, Oklahoma, Utah and Wisconsin – began taxing social security benefits in the same or
similar manner soon afterwards.
pension income – Delaware and Hawaii, the latter of which exempted all pension income.16 By

1994, the last year of policy we consider in our analysis, the list had grown to 24 states and the

number of states exempting all pension income had grown to five (Alabama, Illinois, Mississippi

and Pennsylvania joined Hawaii).17 Both the parameters of these exemptions and the number of

states granting exemptions continue to change. The list of states granting exemptions has grown

to 28 in 2006, with Iowa, Kentucky, Maine, Missouri, New Mexico and Oklahoma joining the

list (AARP 2008 Handbook).

         State income tax preferences for the elderly have therefore displayed wide cross-sectional

and temporal variation over the last 40-50 years. Yet, no elderly migration study has carefully

studied whether such preferences are effective policy tools in recruiting or retaining the elderly.

We contribute to both past research and current policy debates by using elderly migration data

and information on state income tax preferences, as well as other state characteristics such as

EIG and other taxes as well as certain types of expenditures, over a forty year period to better

investigate the relationship between tax breaks for the elderly and their location decisions.

3. Empirical Strategy and Data

         Our migration data comes from census data in which a person is counted as a migrant if

s/he lives in a different state (county) at the time of the Census than s/he reports living in five

years prior. This method is that typically used in measuring migration, but has the limitation of

underestimating movement – e.g., individuals who move away and then return during the five

  Hawaii only fully exempts pension income from employer contributions; pension income that includes employee
contributions are partially taxable.
  Alabama only exempts defined benefit pension plans. Two more states, Vermont (1971-92) and West Virginia
(1973-80), enacted and then subsequently repealed pension exemptions during this period. In addition, the amount
and other details of these exemptions changed for a large number of states during this period. This policy therefore
displays ample temporal and cross-sectional variation during our study period. See Conway and Rork (2008c) for a
more detailed listing of these policy changes.
year period are not counted. Age is measured at the time of the census and thus individuals can

be up to five years younger when the move actually took place.

         Our main source of migration data is the County-to-County Migration Flows calculated

by the Census for 1970, 1980, 1990, and 2000. These flows report the number of individuals of

a certain age category (typically, ages 5 and over) who moved between each set of counties

during the 5 years prior to the census. We then aggregate these data by state (eliminating

intrastate movers), and refer to these as (state-to-state) Census flows. These flows are publicly

available in all four census years for ages 5 and over, but only since 1980 for ages 65 and over.

We therefore requisitioned a special tabulation from the Census for the elderly in 1970. By

subtracting elderly flows (ages 65+) from total flows (ages 5+), we obtain non-elderly flows

(ages 5-64). We also perform supplementary analyses using migration information for high

income and low income elderly using data from the Integrated Public Use Microdata Series

(IPUMS). In this way, we can refine our analyses to those elderly most (least) likely to be

affected by the tax incentives. However, the IPUMS is based on a much smaller sample and is

thus likely to contain more sampling error and a greater preponderance of zero flows.18 For this

reason, we use the Census tabulations in our main analyses.

   Past studies of elderly migration patterns – especially those focusing on changes over time -- have mostly utilized
the Public Use Micro-Samples (PUMS) for various census years. The IPUMS is the result of a collaborative effort
of the Minnesota Population Center to bring each Census year‟s PUMS into one place and improve the uniformity of
coding across the years to ease comparisons (Ruggles et al 2004). The IPUMS is therefore based on the PUMS and
is directly comparable to past analyses using the PUMS. The (I)PUMS data has the advantage of being reported at
the individual-level and contains a wide range of characteristics, including age, income, education, etc. It therefore
provides researchers more flexibility in constructing migration flows (based on different age groupings or retirement
status, for instance), in addition to allowing researchers to study the characteristics of migrants and thus possible
determinants of migration behaviors (e.g., see Walters 2002). However, it has the significant drawback of being
based on a substantially smaller proportion of the population (5% at best) whereas the census tabulations are based
on an approximately 1 in 6 sample. Conway and Rork (2008a) demonstrate that the increased sampling error that
results from using the smaller PUMS data is magnified when one investigates changes in flows over time. As a
result, elderly migration flows appear to have changed substantially more (and erroneously so) if one uses the
(I)PUMS data rather than the census tabulations. We therefore choose to use the census tabulations in our main
analyses, which are limited by the way the data is tabulated by census – i.e., all individuals ages 65+ and all
         With the migration flow data, we estimate the following general gravity model of

migration flows,

(1)      ln M g ijt    Dij  d t   o ln Popit   d ln Pop jt   o X it   d X jt   ijt ,

where M g ijt denotes the number of individuals in group g (= elderly, non-elderly, high income

elderly, low income elderly), that moved from state i to state j during census time period t (=

1965-70, 1975-80, 1985-90, and 1995-2000). This specification follows the typical gravity

model in which the natural log of the migration flow is modeled as a function of the log of the

populations at the origin ( ln Popit ) and at the destination ( ln Pop jt ), plus the log of the distance

between the two (captured in equation 1 by Dij). The superscripts o and d denote the parameters

for the characteristics of the origin (state i) and destination (state j).

         To our knowledge, we are the first to estimate a panel elderly migration flow model,

which allows us to add a time period index t, or compare it to that estimated for the non-elderly

or different income groups of elderly (thus the group index, g).19 To allow for general shifts

over time, we include a full set of census year dummy variables, denoted by dt, in all of the

models we estimate. Observed characteristics of the origin and destination states are included as

well and are denoted by X. The „same sign‟ problem discussed above manifests itself when the

                                                                      ^ o      ^ d
coefficients on X are estimated with the same sign – e.g.,  and  > 0. Such a result suggests

that a variable such as cost of living encourages both out-migration (   0 ), as expected, as

individuals age 5+. However, as discussed shortly, our results here are surprisingly robust to the elderly sample
used – full census, full IPUMS, high income IPUMS and low income IPUMS.
  Lin (1999) is an important exception as he estimates a flow model including 3 waves of PUMS data. However,
his analysis includes no state policy variables and is instead more focused on whether migration responses to time-
invariant variables such as climate and distance have changed. See also Frees (1992) for a general discussion of
how one can estimate panel flow models.
                             ^ d
well as in-migration (   0 ), which is counter-intuitive. Past migration flow studies have been

plagued with this problem. One byproduct of our research is to see if panel data analysis can

help eliminate this counter-intuitive result in flow models as Conway and Rork (2006) found that

it did in models using in-migration and out-migration rates. We also explore whether their

finding that EIG taxes appear to discourage elderly migration only in cross-sectional analyses –

i.e., they fail to matter when panel analysis is used – remains in an elderly migration flow model.

As discussed in greater detail in section 3B, we take as our starting point the models estimated by

Conway and Rork (2006) in terms of the variables included. The availability of flow data,

however, requires that we estimate a gravity model specification instead and the availability of

panel flow data allows us to estimate several variants of it.

A. Alternative Econometric Specifications

           Adding a time dimension allows us to treat distance, as well as unobservable state

characteristics, in a variety of ways. Specifically, we begin by estimating the standard gravity

model -- Dij in equation (1) is simply the natural log of the distance „as the crow flies‟ between

the geographic centers of states i and j.20 This specification completely ignores the possibility of

unobserved state characteristics and can also be estimated with a single cross section of panel

migration flows. Therefore, we not only estimate this model on the entire pooled sample (1970-

2000 censuses), but also on each census separately. Estimating the model separately for each

census allows us to compare our results to those found by past researchers who use a single

census (i.e., Voss et al 1988, Conway and Houtenville 2001, 2003) and also to see if the

relationships appear fairly stable over time.

     Distance is frequently expressed in log form in gravity models to allow for a diminishing effect.
        Next, we add origin and destination state fixed effects to the model; i.e., Dij now

includes distance plus J-1 dummy variables for each destination state and I-1 dummy variables

for each origin state. We refer to this as our panel gravity model. As always, any state

characteristic that does not change over time must be dropped from the model. The origin and

destination state fixed effects capture any time-invariant, unobserved state characteristic that may

affect migration decisions, such as climate, culture and natural amenities. Moreover, the effects

of any characteristic or policy (δ) are identified only by those states that experienced a change in

that characteristic or policy.

        Our last and least restrictive specification replaces the distance variable and

origin/destination fixed effects with a full set of flow-specific dummy variables. Every ij flow

combination is assigned its own dummy variable. State-to-state migration flows have been

found to be very stable over time, and this specification models that persistence explicitly. Note

that the flow-specific dummy variables subsume the distance variable. More generally, the flow-

specific fixed effects capture the effects of any time-invariant origin or destination state

characteristic as well as any variable associated with the flow, such as information networks,

vacation patterns, etc. We refer to this as our flow-specific model.

        Even our flow-specific model requires that we control for (changes in) state population,

as the population distribution has shifted over time. A recurring question in the elderly migration

literature is which population to use – the total population or the elderly (non-elderly)

population? While elderly population is probably more closely associated with elderly migration

flows, it has the disadvantage of more likely being endogenous – it reflects the effects of past (or

anticipated) policies towards the elderly. In addition, using the age-specific population requires

that we estimate different models for the elderly and non-elderly and thus complicates our

comparisons. It also ignores the strong linkages in migration across age groups, where elderly

parents move to be closer to their adult children, for example. We therefore use the total

population in our main specifications and explore the effects that using the age-specific

population instead has on our results.

       Another econometric issue to grapple with is the preponderance of zero flows. Certain

flow combinations (e.g., Wyoming to Vermont) are rare occurrences in the census and thus are

frequently zero. The typical approach in past migration studies is to drop zero flows from the

analysis. In our case, however, this approach leads to an unbalanced panel. It also leads to

differences in the sample size and composition across our different samples. We therefore

explore the robustness of our results to an alternative approach, which is to set those zero flows

to 1.0 (zero when logged) and include them in the estimation.

       A final problem is that of possible endogeneity. As has been discussed extensively in

policy research, state policies do not occur randomly but rather arise from a particular political

process and policy environment. A concern in migration research is that elderly migrants may

affect state fiscal policy as they become part of the electorate. (They could also affect other

characteristics such as cost of living and crime.) We address this problem in the typical manner

by including the state characteristics, policies, etc. in place the year prior to the migration period.

Because our migration data refers to 1965-70, 1975-80, 1985-90 and 1995-2000, whenever

possible we include state characteristics and policies for 1964, 1974, 1984 and 1994,

respectively. We note the few instances when we are unable to find data for all of those years in

the next section which discusses the variables we include.

B. Variables Included

        The state variables, X, that we include in our baseline model follow that of Conway and

Rork (2006). In this way, we can see if the results they find for panel elderly migration rate

models and EIG taxes carry over when one has access to richer migration flow data. We then

augment their specification by including detailed information on elderly state income tax

preferences. In general, emphasis was given to those variables that could be gathered in a

comparable way over the entire time period. Because our other variables follow directly from

Conway and Rork (2006) and have been discussed in detail there, we only describe them briefly.

Variable definitions and sources are reported in Table 1.

        Our explanatory variables fall into four main categories – cost of living (COL), amenities,

government expenditures and tax policies. Housing prices are frequently used as a measure of

COL; however, no state-specific housing indices are available prior to 1975 (Bakija and Slemrod

2004). We therefore use the median house value, calculated using the IPUMS from each census

year, as our proxy for cost of living in each state.21 For amenities, we include heating degree

days (to capture climate, which is dropped from the panel analyses), the crime rate and the

percentage of the population that is elderly. We also include the average manufacturing wage

and the unemployment rate; these are key labor market variables for younger migrants whereas

they may further capture COL for elderly migrants. Government expenditure variables are

measured as state and local expenditures per capita and are broken into several categories – 1)

health and hospitals, 2) education, 3) welfare programs, and 4) all other.

  This is the same measure used by Conway and Rork (2006); such is the case for all variables discussed here
unless otherwise noted. In future work, we may investigate using the measure used by Bakija and Slemrod (2004),
which combines the quality-adjusted Office of Federal Housing Enterprise Oversight (OFHEO) housing indices that
became available beginning in 1975 with regional-specific indices of new single family homes in the years prior to
that. Most of their data refers to years since 1975 and so the richness of the OFHEO index is especially beneficial.
Because half of our sample refers to 1974 or earlier, the value of the OFHEO data is more dubious.
           In addition to income tax preferences for the elderly, we must also control for other forms

of state taxation. Again, we borrow from Conway and Rork (2006) and use the measures that

they emphasize. For EIG taxes, we use two alternative, complementary measures: 1) a

dichotomous dummy variable for whether the state has an incremental EIG tax or not (i.e., a state

that imposes only a „pick-up‟ or „soak-up‟ tax is coded with a zero), and 2) the effective average

state EIG tax rate on a $1 million (in constant 1996 dollars) bequest divided equally between two adult

children, reported by Bakija and Slemrod (2004, Table 2).22 These rates are reported every 5 years from

1965 to 2000; we therefore use 1965, 1975, 1985 and 1995. Our results are very robust to the EIG

measure we include. For consistency, then, when our income tax preferences variables are measured in a

dichotomous form, we report the results from models in which EIG taxes are measured similarly (the first

measure above). When our income tax preference measures are continuous and thus reflect their

magnitude, we instead report results from models that use the effective EIG tax rate (the second measure).

The rest of the tax system is captured in the identical way as the primary specification in Conway and

Rork (2006); we use average tax rates for the personal income tax and sales tax, and revenue shares for

property and all other taxes.

           Finally, as in Conway and Rork (2006), we explore the robustness of results to the control

variables that are included. We estimate several variants of the model that exclude sets of variables,

including one model that excludes all but the key tax variables. Our main results are quite robust to the

variables included, and so we emphasize the results from the broadest model.23

C. Measuring State Income Tax Preferences for the Elderly

           Because income tax preferences are our primary variables of interest and there are no

previous migration studies from which to draw, we explore several alternative measures. These

     We thank Jon Bakija for giving us permission to use this data.
  The only strongly affected coefficients are those referring to the EIG taxes. In the more parsimonious models, the
coefficients go from negative at both the origin and destination to positive at both.
measures vary along two dimensions – discrete versus continuous, and individual

components/features versus aggregate. The four measures and associated model specifications

we consider are:

1) Dummy variables for each of the three components. In this model, we simply include three

dummy variables for whether the state had each type of preference (deduction, social security

exemption, and private pension exemption) in the year prior to the migration period. Although

probably the simplest and most straightforward, this approach fails to capture the magnitude of

the preferences. In this model, we emphasize the results from specifications that control for EIG

taxes in a similar fashion, using only the dichotomous variable for whether a state has an

incremental EIG tax or not. In the rest of the specifications that attempt to capture magnitude --

2) through 4) below -- we emphasize those that include the effective EIG tax rate from Bakija

and Slemrod (2004) instead.

2) The maximum tax benefit associated with each component. This measure is calculated by

multiplying the amount of the deduction or exemption by the maximum marginal income tax rate

a household could face in the state. In this way, we are capturing the maximum possible tax

benefit experienced.24 In order to create this measure for social security exemptions, we use the

maximum social security benefits that a household could receive.25 For states that completely

exempt social security benefits, we multiply the maximum benefits by the marginal tax rate, as

that is the maximum value of the exemption. For states that followed the 1983 federal

government‟s policy of taxing social security benefits, we multiplied half of the maximum

benefits (the maximum amount that could be subject to tax) by the marginal tax rate. The

   For states that grant a tax credit instead of a deduction or exemption, the dollar amount of the tax credit is used.
We are grateful to Jon Bakija for providing us with the information regarding the specific tax preferences (under the
terms of our grant). The maximum marginal income tax rates are from the World Tax Database.
  The source for these benefits is Table 2.A28 from the Social Security Administration‟s Annual Statistical
Supplement, 2006. We also thank Alberta Presberry who helped us interpret and best use this data.
measure is calculated in a similar fashion for states that followed the 1993 federal government

policy‟s of taxing up to 85% of benefits (i.e., 15% of benefits are multiplied by the marginal tax

rate). In this way, all three measures are calculated in the same manner and each captures the

maximum possible value of the tax preference.26

3) Aggregated maximum tax benefit. Although we have a large number of state-to-state flows

(and thus observations), we still have only forty or so states with income taxes. It therefore may

be asking too much of the data to estimate separately the effect of each tax preference. 27 We

therefore create an alternative measure, which is the simple sum of the three components

described in 2). This measure is interpreted as the maximum income tax benefit granted the

elderly and does not distinguish between the different policy instruments.

4) TAXSIM-estimated tax benefit. The above three measures, while straightforward, likely miss

many complexities of the state income tax systems as well as overstate the actual tax benefits

from these preferences. As a complementary measure, we therefore use the estimated tax

benefit of being a representative high income elderly household calculated in Conway and Rork

(2008b). Summarizing briefly, the authors use data from the Current Population Survey (CPS) to

create a representative „high income‟ elderly household (defined as the top quartile) and a

comparable non-elderly household (who has the same level of income). These two households

therefore vary only by age and the sources of their income (e.g., the non-elderly household has

few social security benefits and much higher earnings). The state income tax burden facing each

   Note that the social security exemption presents a challenge in this regard. Because 50% (85%) is the maximum
benefits that can be taxed, our measure is not actually the maximum possible value. Rather, the maximum possible
tax benefit is 100% of the benefits even for those states who tax social security benefits for high income households.
However, using 100% of benefits as the maximum value obviously defeats the purpose of identifying states that
treat social security benefits more generously than others. An alternative is to instead use the maximum tax one
could pay on social security benefits. Such a measure has two serious drawbacks. First, as a maximum „tax,‟ its
coefficient would have an opposite interpretation from the others. Second and more seriously, such a measure could
not be used to aggregate all of the preferences as we do in our third model specification.
  This is also why we estimate several more parsimonious versions of our model including one that only contains
income tax variables.
type of household is calculated via TAXSIM. The tax benefit of being elderly is the difference

between the estimated state tax liability of the non-elderly household and the elderly

household.28 This approach also provides an alternative measure of the average income tax

burden, and so in these specifications we replace the pseudo average income tax rate with the

actual average income tax rate facing a non-elderly household (i.e., the estimated tax liability

divided by household income).

           While the TAXSIM measure better captures the subtleties of the state and federal income

tax code and may better reflect the tax benefits experience by a „typical‟ high income elderly

household, it has a critical drawback. TAXSIM (and the CPS) is only available since 1977. We

therefore must omit the 1970 data from any analysis using this measure. Moreover, if we adhere

to our practice of using the year prior to the migration period, we lose the 1980 data as well. We

take two approaches to this problem. First, we estimate models for the 1980-2000 data using the

midpoint average (e.g., 1977-78 average for 1975-80 migration) for the TAXSIM variables.

Second, we estimate models for 1990 and 2000 following our standard practice of using the year

prior to the migration period (1984 and 1994, respectively). This issue highlights that our results

may differ between 3) and 4) above either because of the difference in time periods or because of

the different way that the policy variables line up with the time period. To explore the

importance of these differences, we also estimate 3) using only 1980-2000 and 1990-2000 data.

Fortunately, our findings are robust to these exercises.29

           We therefore estimate seven different models in all – one each for 1) and 2), three for 3)

and two for 4). For the sake of brevity, we focus primarily on the models at the two extremes –

the first specification that includes dummy variables for each component and the fourth

  See Conway and Rork (2008a), in particular Tables 1 and 2, for further discussion of how these measures are
constructed and Feenberg and Coutts (1993) for discussion of the TAXSIM estimator.
     All exercises discussed but not reported are available upon request.
specification that uses the TAXSIM measure for 1980-2000. (However, unique findings from

the other models are discussed as well.) This provides an almost completely complementary set

of analyses – different time periods (but not so short as to diminish the richness of the data),

different EIG tax measures, magnitude versus discrete and components versus aggregate. It also

uses tax preference information from two completely different sources.

4. Empirical Results

A. The importance of panel data redux and the role of income tax preferences

         Our first goal is to see whether the findings of Conway and Rork (2006) stand up when

one uses migration flow data. We also investigate whether one might erroneously conclude that

income tax preferences matter if only cross-sectional analyses are used – as the authors found

was true for EIG taxes. We therefore begin by estimating the same specification as in Conway

and Rork (2006, Tables 4, 5, A1 and A2) except that migration flow data is used. We first

simply pool the cross sections (our standard gravity model described above), then add origin and

destination state fixed effects (panel gravity model) and finally include flow-specific fixed


         Results from these three different models are reported in the top panel of Table 2. It is

immediately apparent that the basic findings of Conway and Rork (2006) carry over to migration

flow data. In the standard gravity model on the pooled cross-sections (the first two columns), the

estimated EIG tax coefficients are again strongly statistically significant – at both the destination

(as expected) and the origin. These effects, however, are completely eliminated by the inclusion

  Results from the individual cross-sections – i.e., estimating each census year separately -- are reasonably similar
to those obtained from the pooled analysis and so we emphasize the pooled model results for brevity.
of either origin/destination or flow-specific fixed effects. Other tax and expenditure variables are

affected similarly. Also as in Conway and Rork (2006), we find that panel methods remove the

persistent „same-sign‟ problem of crime that has been found repeatedly in the literature. Once

panel methods are used, crime has the intuitive effect of driving out the elderly (   0 ) and

                                           ^ d
discouraging them from entering (   0 ). It is only fair to note, however, that panel methods

do not eliminate all of the same-sign results; cost-of-living (median house value) and per capita

income both continue to suffer from it.31 In the case of cost-of-living, we can at least take

comfort in the fact that the destination effect (the intuitive result) has the greater magnitude.

         The bottom panel of Table 2 adds the dummy variables for the different elderly income

tax preferences states may offer (model 1 above). The results for the rest of the variables are

largely unaffected by their inclusion. Thus, the insignificant effect of EIG taxes on elderly

migration cannot be attributed to a failure to include other aspects of the income tax system that

appeal to the wealthy elderly. In general, the results for these other variables, EIG taxes in

particular, are remarkably robust in the multitude of models we estimate including different

measures of state income tax preferences using different samples.

         Income tax preferences are similar to the other state policies in that they are statistically

significant in models that ignore state unobservables (the first two columns). Unlike EIG taxes,

however, their estimated effects are much more likely to be counter-intuitive and to persist even

after origin/destination or flow-specific effects are included. In the pooled cross-section

   Population and, to a lesser extent, percent elderly are expected to exert the same forces at the origin and
destination. Population at both destination and origin are expected to positively affect migration flows. Percent
elderly population is likely capturing the higher elderly population in these states as well and is partly adjusting for
our use of total population. However, to the extent that it is also capturing an amenity of the state – which could be
either good or bad to the elderly – one would expect its sign to differ. Many of the other coefficients also have the
same sign, but both are rarely if ever statistically significant.
analysis, deductions and social security exemptions both appear to drive away elderly migrants –

a counter-intuitive finding. Pension exemptions appear to retain the elderly, but also discourage

in-migrants with an even greater magnitude. Many of these results are eliminated by the

inclusion of fixed effects, but the remaining statistically significant effects are universally

counter-intuitive. Specifically, fully exempting social security benefits appears to drive away the

elderly and discourage in-migration. Offering an exemption for private pension income also

drives the elderly away from the state. This first set of results, then, completely refutes the

notion that state income tax preferences for the elderly are an effective vehicle for recruiting

and/or retaining a state‟s elderly population. Next, we turn to whether these results are robust to

alternative measures of income tax preferences and more refined samples of the elderly.

B. Alternative measures of income tax preferences

        Table 2 also shows that including flow-specific rather than origin/destination state fixed

effects has little effect on the results and leads to more statistically significant coefficients, if

anything. This finding carries over to the multitude of models we estimate. We therefore

emphasize the results from the flow-specific fixed effects model in our alternative models. Our

first exercise explores the effects of using alternative measures of the income tax preferences.

Table 3 reports the key coefficients from five of the seven specifications discussed above. The

first two models include the different types of preferences (first as dummy variables – the model

reported in Table 2 -- then as maximum tax values). Models 3 and 4 aggregate the separate

preferences into the maximum total amount of tax benefits. Model 3 uses the full sample period

(1970-2000) while model 4 limits the analysis to 1980-2000 in order to be more comparable with

Model 5, which by using the TAXSIM measure must be limited to this time frame.

           EIG taxes and the general income tax burden (average income tax rate) continue to have

no statistically significant effect on migration. Moreover, to the extent that they approach even

marginal statistical significance, they are of a counter-intuitive sign (either discouraging out-

migration or encouraging in-migration).

           The estimated effects of income tax preferences do vary depending upon the measure(s)

used, but in general lend little support to the notion that such policies are attractive to potential

elderly migrants. When we use the total tax value of the different components (columns 3 and

4), the counter-intuitive effect of exempting social security benefits remain while the one for

pension income exemptions are eliminated. The value of a deduction is the first instance --and

ultimately the only instance -- of income tax preferences having the expected effect on

migration. A large deduction discourages out-migration and encourages in-migration, although

the latter is not statistically significant. Of the three components, this is actually the one that we

least expected to have a strong effect, given the relative small size of the deductions, credits,

etc.32 Summing the values of the different types of preferences (Models 3 and 4, columns 5-8)

finds no statistically significant of preferential tax treatment.

           The last two columns (Model 5) use a completely different source of information on

elderly tax preferences (simulated tax burdens and benefits from TAXSIM). Yet, these results

once again refute the notion that tax preferences attract/retain elderly migrants. To the contrary,

they suggest that tax benefits drive away the elderly and discourage in-migration.

           These results are remarkably robust to a wide set of sensitivity checks. We re-estimated

these models alternately omitting observations with no income taxes, changing our treatment of

Mississippi (the lone state that taxed social security benefits prior to 1983), aggregating

     Most tax credits are $100 or less and deductions/exemptions are less than $2000.
expenditures into total per capita, limiting the number of amenity variables that are included

(omitting the unemployment rate, per capita income and percent elderly), substituting flow fixed

effects with destination and origin fixed effects, including observations with zero flows, and

estimating a model that only includes EIG and income tax variables. The results are all

qualitatively the same.

C. Rich man, poor man

        Perhaps our results are due to a form of aggregation bias. As the gerontology literature

points out, the elderly move for a variety of reasons (e.g., Litwak and Longino 1987). By

looking at the migration patterns of all elderly individuals, perhaps real effects on the mobile,

high income elderly are obscured. To explore this further, we use the IPUMS from 1970, 1980,

1990 and 2000 to construct state-to-state migration flows of the high income elderly, defined as

individuals whose household income places them in the top quartile of the elderly income

distribution. We construct migration flows for the low income elderly (lowest quartile) in a

similar fashion. The low income elderly make an interesting control group as they would pay

little if any income taxes even without elderly tax preferences.33

        The smaller sample sizes of the IPUMS raise two concerns.34 First, Conway and Rork

(2008a) show that even with the enormous sample sizes of the IPUMS and the census, the

relatively smaller size of the IPUMS leads to much greater variation in migration flows over time

than in the larger census. This larger variation has the potential to make location characteristics

look more important than they are, simply due to possible spurious correlation. We tackle this

 See Conway and Rork (2008c, Table 2), which reports the estimated tax liabilities of the low income non-elderly
and elderly over several years.
  We use the 1% sample for 1970, the largest available that includes migration information, and the 5% samples for
the other three.
problem by first estimating the models in Table 3 using the migration flows for all elderly

individuals using the IPUMS. The results are nearly identical to those that use census flows.

        The second concern is that there are many more instances of zero flows, which likely

makes our treatment of them more important. We therefore re-estimate the models including the

zero flow observations and setting them to 1.0 (such that the log is zero).

        Before estimating the models for these different income groups, however, we first

explore how different their migration patterns are. Table 4 reports the top 10 state-to-state flows

in each year for the total elderly census sample, then the high income and low income IPUMS

samples. The table also reports the number of zero flows, the migration rate and the geographic

concentration of migration (measured as the percentage of all elderly migrants accounted for by

the top 10 flows) for each sample/year.

        Looking across the table (across samples), the similarity of the top 10 flows is clearly

evident. The elderly are consistently leaving cold, northeastern or north central states in favor of

warmer climates, especially Florida. Differences across the samples are as one would expect.

The smaller IPUMS samples lead to many more instances of zero flows, especially in the 1970

sample that is only based on 1%.35 The high income elderly have a higher rate of migration than

the low income, although both are small (approximately 4 percent). Perhaps the greatest

difference is how less concentrated the migration flows of the low income elderly are. This

finding is consistent with Litwak and Longino (1987) and others who find that the assistance

  This difference presents yet another argument for re-estimating our models excluding 1970. Still another benefit
of doing so is to explore the influence of late adopters of the income tax, such as Illinois and Michigan, most of
whom enacted income taxes in between 1964 and 1974 and so show up as changing dramatically only when 1970 is
motive for moving is more prevalent among less economically advantaged elderly migrants; one

would expect such moves to be less geographically concentrated.

       Looking down the table (changes over time) reveals the same patterns that others have

described (Flynn et al 1985, Bradley and Longino 2003, and Conway and Rork 2008a). Elderly

migration rates are very stable during this period, ranging from 3.2% to 4.9% across all

samples/years and displaying no consistent trend over time. Likewise, the same flows tend to

appear in the top 10, although some key differences are evident. Florida‟s dominance as a

destination is diminishing while Nevada‟s is growing. California has gone from a major receiver

to a major sender. More generally, migration flows have become much less geographically

concentrated over time, especially for the high income elderly.

       Next, we turn to whether location characteristics, in particular state tax policies towards

the upper income elderly, have differing effects on these two groups. Given that low income

elderly pay little income taxes and are very unlikely to pay any EIG taxes upon death, we would

not expect these policies to have an impact on their migration behavior. It is important to note

that we use the same policy variables for both groups – the estimated effects for a high income

household – which should be largely irrelevant for low income individuals. We therefore view

the low income sample as a „straw man‟ or „control‟ group against which to compare our results

for the high income elderly.

       Tables 5 and 6 repeat the exercises summarized in Table 3 estimated with the high

income and low income IPUMS samples, respectively. Immediately apparent is how similar the

results are across the two income samples, as well as the census flows, and how once again any

statistically significant results for EIG taxes and income tax preferences tend to be counter-

intuitive. As before, social security exemptions and the TAXSIM-calculated elderly tax benefit
increase out-migration. The one and only supportive finding from the main analyses remains as

well; the maximum value of the deduction tends to discourage out-migration. Moreover, this

effect is only statistically significant for the high income elderly. The one difference we see

from this analysis versus the census analysis is that EIG taxes are now occasionally significant –

but still of the „wrong‟ sign (they discourage out-migration).

       Finally, we explore the sensitivity of these findings to our treatment of zero flows.

Appendices 1 and 2 summarize the results from the two models at the extremes, the one with

dummy variables for each component for 1970-2000 and the one using the TAXSIM measures

for 1980-2000, both with and without zero flows included. The first two columns therefore

correspond to the first two columns of Tables 5 and 6, and the fifth and sixth columns correspond

to the last two. The intervening columns report the estimates for these models when zero flow

observations are included. While it is evident that including zero flows alters the results, it does

not change the finding that almost all statistically significant results are of the wrong (counter-

intuitive) sign. Furthermore, although not reported in the appendices, including zero flows

eliminates our only finding in support of tax incentives‟ effectiveness in recruiting or retaining

the elderly; the statistically significant effect of the maximum value of elderly deductions is

completely eliminated.

5. Concluding Remarks

       Our research investigates whether little investigated yet much debated elderly tax

incentives, such as exemptions for retirement income, have an effect on elderly migration

behavior. To our knowledge, we are the first to explore different measures of these incentives

and include them as possible factors influencing elderly interstate migration. Unlike past
research that studies the effect of state policies on elderly migration, ours employs migration

flow data from four different censuses so that persistent flow patterns can be explicitly modeled

via panel methods. We subject our analyses to a wide range of sensitivity checks. Numerous

alternative measures of tax incentives are used which come from two very different sources and

approaches (TAXSIM tax liability estimates versus actual policy parameters). We estimate the

models using different time periods, including different sets of control variables and treating

differently states with no income taxes during part or all of the sample period. We also explore

the impact of different panel methods, such as including origin and destination, or flow-specific

fixed effects. Perhaps most importantly, we extend our analyses to the elderly most likely to

react to these incentives – the high income elderly – and compare our findings to those obtained

from the least likely (low income). Our findings overwhelmingly suggest that these incentives

have no credible effect on elderly migration. The few results that are statistically significant are

almost always of the wrong (counter-intuitive) sign and appear for the low income elderly as

well. Moreover, these findings carry over to other state policies, including much-debated estate,

inheritance and gift (EIG) taxes.

       However, policy debates across the country, as well as the recent actions by states to

increase tax breaks for their elderly citizens and eliminate EIG taxes, suggest that many believe

otherwise. Our findings are therefore likely a surprise and, as such, should be subjected to even

further scrutiny. Still, they are very much in line with recent research that reveals how stable

elderly migration flows have been over time. If elderly migration patterns are so persistent in the

face of dramatic changes in state EIG taxes and elderly income tax preferences, then we should

not expect to find that such policies have had much effect.36 Our results, combined with the

consistently low rate of elderly interstate migration, should therefore give pause to those who

believe that offering state tax breaks is an effective way to attract and retain the elderly.

  Conway and Rork (2008a) find that more than 93% of the total variation in elderly migration flows during this
period is explained by persistent flow patterns (as measured by flow-specific fixed effects). Thus, at most 7 % of
elderly interstate migration has been affected by changes in state policies. For a longer time period and analyses that
extend to other age groups, see also Wolf and Longino (2005).

Bakija, Jon, “Documentation for a Comprehensive Historical U.S. Federal and State
         Income Tax Calculator Program,” revised January 2008, available at, accessed

Bakija, Jon and Joel Slemrod, “Do the Rich Flee from High State Taxes? Evidence from Federal
        Estate Tax Returns,” NBER Working Paper No. 10645, July 2004.

Bernstein, David. 2004. Demographic change and the future of state income taxes. Working Paper,
        George Mason University.

Bradley, Don E. and Charles Longino, “A First Look at Retirement Migration,” The Gerontologist, 43,
        2003, pp. 904-907.

Conway, K.S. and A.J. Houtenville, “Out with the Old, In with the Old: A Closer Look at
       Younger versus Older Elderly Migration,” Social Science Quarterly, June 2003, 84(2),
       pp. 309-28.

Conway, K.S. and A.J. Houtenville, “Elderly Migration Flows and State Government Policy –
      Evidence from the 1990 Census Migration Flows,” National Tax Journal, March 2001,
      Vol. LIV, No. 1, pp. 103-23.

Conway, K.S. and A.J. Houtenville, “Do the Elderly 'Vote with their Feet'?” Public Choice,
      December 1998, 97, p. 663-85.

Conway, K. S. and J. C. Rork. (2004). Diagnosis murder: The death of state “death” taxes.
       Economic Inquiry, 42(4), 537-59.

Conway, K. S. and J. C. Rork. “State „Death‟ Taxes and Elderly Migration–the Chicken or the
       Egg?” National Tax Journal, March 2006, Vol. LIX, No. 1, pp. 97-128.

Conway, K.S. and J.C. Rork. (2008a). 40 years of „going with the flow‟ – how much has elderly
      migration changed? Working paper.

Conway, K. S. and J. C. Rork. (2008b). Income tax preferences for the elderly. Public Finance
       Review, 36(5), 523-562.

Conway, K.S. and J.C. Rork. (2008c). The evolution of state income tax preferences for the
       elderly – intelligent design? Working paper.

Edwards, Barbara and Sally Wallace. 2004. State income tax treatment of the elderly. Public Budgeting
      and Finance, 24(2): 1-20.

Farnham, Martin and Purvi Sevak. (2006). State fiscal institutions and empty-nest migration: are
       Tiebout voters hobbled? Journal of Public Economics, 90, 407-27.

Feenberg, Daniel and Elisabeth Coutts. 1993. An introduction to the TAXSIM model. Journal of
       Policy Analysis and Management, 12(1): 189-94.

Flynn, C. B., C. F. Longino Jr., R. F. Wiseman, and J. C. Biggar. (1985). The redistribution of America‟s
        older population: Major national migration patterns for three Census decades, 1960-80. The
        Gerontologist, 25(3), 292-96.

Frees, E. W. (1992). Forecasting state-to-state migration rates. Journal of Business and Economic
        Statistics, 10, 153-67.

Gale, Lewis R. and Will Carrington Heath, “Elderly Internal Migration in the United States Revisited,”
        Public Finance Review, 28(2), March 2000, pp. 153-70.

Haas, W. H. III, D. E. Bradley, C. F. Longino Jr., E. P. Stoller, and W. J. Serow. (2006). In retirement
       migration, who counts? A methodological question with economic policy implications. The
       Gerontologist, 46(6), 815-20.

Lin, G. (1999). Assessing changes in interstate migration patterns of the United States elderly
        population, 1965-90. International Journal of Population Geography, 5, 411-24.

Litwak, E. and C. F. Longino. (1987). Migration patterns among the elderly: A developmental
        perspective,” The Gerontologist, 27(3), 266-72.

Longino, Charles, “Going Home: Aged Return Migration in the United States 1965-70,” Journal of
       Gerontology, 34(5), 1979, pp. 736-45.

Longino, C. F. and D. E. Bradley. (2003). A first look at retirement migration trends in 2000. The
       Gerontologist, 43(6), 904-07.

Manzi, Nina, Joel Michael and Paul Wilson. 2005. State income tax revenues in 2002 and 2030: The
       impact of the retirement of the baby boom. National Tax Association Proceedings of the 98th
       Annual Conference of Taxation 2005. 118-130.

Nelson, Gary M. 1983. Tax expenditures for the elderly. The Gerontologist 23(5): 471-78.

Penner, Rudolph G. 2000. Tax benefits for the elderly. The Retirement Project, Urban
        Institute, Occasional Paper Number 5.

Page, T. F. and K. S. Conway. (2007). The labor supply effects of taxing Social Security benefits:
        Theory, evidence and implications for other Social Security policies. Working Paper, University
        of New Hampshire.

Rappaport, J. (2007). Moving to nice weather. Regional Science and Urban Economics, 37, 375-98.

Rogers, A. (1990). Return migration to region of birth among retirement-age persons in the United
        States. Journal of Gerontology, 45(3), S128-34.

Rogers, A. and J. Raymer. (1999). The regional demographics of the elderly foreign-born and native-
        born populations in the United States since 1950. Research on Aging, 21(1), 3-35.

Ruggles, S., M. Sobek, T. Alexander, C. A. Fitch, R. Goeken, P. K. Hall, M. King, and C.

Ronnander. (2004). Integrated Public Use Microdata Series: Version 3.0. Minneapolis, MN: Minnesota
      Population Center.

Sergeant, J. F., D. J. Ekerdt, and R. Chapin. (2008). Measurement of late-life residential relocation:
       Why are rates for such a manifest event so varied? Journal of Gerontology: Social Sciences,
       63B(2), s92-s98.

Serow, W. J. (1978). Return Migration of the Elderly in the USA: 1955-60 and 1965-70. Journal of
       Gerontology, 33(2), 288-95.

Serow, W. J. (1987). Determinants of interstate migration: Differences between elderly and nonelderly
       movers. Journal of Gerontology, 42(1), 95-100.

Serow, W. J. and D. A. Charity. (1988). Return migration of the elderly in the United States. Research
       on Aging, 10(2), 155-68.

Walters, William H. (2002). Later Life Migration in the United States: A Review of Recent Research.
        Journal of Planning Literature, 17(1), 37-66.

Wisconsin Policy Research Institute Report. (2006). “ Retirement Migration in Wisconsin – the Impact of
       State Taxes,” Report, 19(8).

Wilmoth, J. M. and C. F. Longino. (2006). Demographic trends that will shape U.S. policy in the
      twenty-first century. Research on Aging, 28(3), 269-88.

Wolf, D. A. and C. F. Longino. (2005). Our „increasingly mobile society‟? The curious persistence of a
       false belief. The Gerontologist, 45(1), 5-11.

Zahn, M. A. and S. D. Gold. (1985). State Tax Policy and Senior Citizens – A Legislator’s Guide.
       Denver CO: National Conference of State Legislatures.

Table 1 -- Sources of Variables

VARIABLE NAME                                                          SOURCE

1. Migration Variables
migration flows of all elderly                                         Census County to County Migration Files
migration flows of high-income (top quartile) elderly                  Authors' calculations from IPUMS
migration flows of low-income (bottom quartile) elderly                Authors' calculations from IPUMS

2. EIG Tax Measures
incremental EIG tax (YES = 1)                                          Authors' calculations
Bakija & Slemrod (2004) tax rate                                       Bakija and Slemrod (2004)

3. Income Tax Measures
average income tax rate                                                State Government Finances
high income non-elderly average income tax rate                        TAXSIM
exemption of additional income tax deductions or credits for elderly   Bakija (2008)
exemption of social security from taxation                             Bakija (2008)
exemption of pension income from taxation                              Bakija (2008)
high-income elderly tax bonus                                          TAXSIM

4. Other Tax Variables
sales tax rate                                                         World Tax Data Base
property tax share                                                     State Government Finances
other per capita tax share                                             State Government Finances

5. Amenities and Cost of Living
heating days                                                           Statistical Abstract of the United States
median house values (cost of living)                                   Calculated from PUMS
crime rate                                                             Statistical Abstract of the United States
average mfg wage                                                       Statistical Abstract of the United States
state unemployment rate                                                Statistical Abstract of the United States
% population age 65 and over                                           Statistical Abstract of the United States
per capita income                                                      Statistical Abstract of the United States

6. Per Capita Government Expenditures
health and hospitals                                                   State Government Finances
education expenditures                                                 State Government Finances
welfare expenditures                                                   State Government Finances
all other expenditures                                                 State Government Finances

Table 2 -- Regression Results with Alternative Panel Methods
                                                        EIG TAXES ONLY
                                     origin   destination    origin  destination                                     origin      destination
incremental EIG tax (1 = YES)       -0.296 *** -0.468 *** 0.021        -0.015                                        0.017         -0.008
                                   [-11.03]     [-17.62]     [0.62]    [-0.43]                                       [0.68]        [-0.32]
additional credits or deductions
for elderly (1 = YES)

social security income exemption
(1 = YES)

pension income exemption
(1 = YES)

average income tax rate                         -13.35 *** -11.628 ***             2.913             1.675           2.074         1.514
                                                [-9.30]     [-8.07]                [1.07]            [0.61]          [1.06]        [0.77]
property tax share                               0.176      0.429 **                0.036             0.364           0.052         0.518 ***
                                                 [0.84]    [-2.05]                  [0.13]            [1.38]          [0.27]        [2.71]
sales tax rate                                   0.029 *** -0.044 ***              -0.018            -0.012          -0.011        -0.014
                                                [-3.53]    [-5.28]                 [-1.42]           [-0.92]         [-1.20]       [-1.58]
other tax share                                 -0.966 *** -0.546 **               -0.602             0.242          -0.620 **      0.255
                                                [-4.31]    [-2.43]                 [-1.62]            [0.65]         [-2.35]        [0.97]
heating days                                   9.0E-06           -9.5E-05    ***
                                                 [1.01]           [-10.58]
state median house value                       3.7E-06     ***    4.5E-06    *** -1.8E-06    ** -2.5E-06 *** -1.7E-06 ** -2.6E-06 ***
                                                 [4.10]             [4.97]        [-1.99]        [-2.85]       [-2.70]     [-4.10]
state crime rate                               3.0E-04     ***    4.0E-04    *** 3.8E-05     ** -2.1E-04      4.2E-05 *** -2.9E-05 **
                                               [25.30]             [34.35]         [2.09]        [-1.13]        [3.27]     [-2.25]
per capita income                               -0.024     **      -0.078    *** 0.038       *** 0.032 ** 0.033 *** 0.037 ***
                                                [-2.49]            [-8.17]         [2.71]         [2.30]        [3.25]      [3.64]
average state mfg wage                           0.011              0.021    ** 0.036        ** 0.022           0.040 *** 0.023 *
                                                 [1.11]             [2.09]         [2.05]         [1.24]        [3.21]      [1.84]
state unemployment rate                         -0.040     ***     -0.070    *** -0.010           0.002        -0.012 *     0.001
                                                [-4.65]            [-8.15]        [-1.02]         [0.17]       [-1.73]      [0.21]
% population aged 65 or more                     8.643     ***      7.016    *** 4.899       *** 6.670 *** 5.463 *** 6.316 ***
                                               [12.79]             [10.39]         [3.31]         [4.50]        [5.29]      [6.11]
expenditures on health & hospitals              -0.923     *** -1.361        ***  0.486 *** 0.171                     0.465 *** 0.182
                                                [-7.15]        [-10.56]           [3.02]     [1.07]                   [4.02]     [1.58]
education expenditures                           0.475     *** 0.540         *** -0.044      0.062                   -0.039      0.042
                                                 [6.09]          [6.96]          [-0.44]     [0.62]                  [-0.55]     [0.58]
welfare expenditures                            -0.215     *** -0.463        *** 0.061       0.057                    0.058      0.046
                                                [-3.04]         [-6.46]           [0.91]     [0.84]                   [1.21]     [0.92]
other expenditures                               0.315     *** 0.230         *** -0.005     -0.071                    0.000     -0.084 **
                                                 [7.19]          [5.25]          [-0.10]    [-1.44]                   [0.01]    [-2.37]
logged state population                          0.810 *** 0.652 *** 1.369 ***                       0.922     ***    1.324 *** 0.886 ***
                                                [49.68]    [39.73]   [13.78]                         [9.30]          [18.64]    [12.50]
logged distance between states                  -1.279 ***
year fixed effects                               YES                  YES                                            YES
origin and destination fixed effects              NO                  YES                                            NO
flow-specific fixed effects                       NO                   NO                                            YES
NOTE: t-statistics reported in brackets. ***,**, * indicated significant at 1%, 5% and 10% levels.
Table 2 (continued) -- Regression Results with Alternative Panel Methods
                                            INCOME TAX PREFERENCES INCLUDED
                                     origin    destination  origin   destination   origin                                        destination
incremental EIG tax (1 = YES)       -0.307 *** -0.460 *** 0.004        -0.007    -3.5E-05                                          1.1E-04
                                  [--11.47]     [-17.39]    [0.13]     [-0.19]     [0.00]                                           [0.00]
additional credits or deductions     0.060 ** 0.041         0.004      -0.011     -0.011                                           -0.020
for eldelry (1 = YES)                [2.19]       [1.52]    [0.13]     [-0.36]    [-0.53]                                          [-0.93]

social security income exemption                 0.197      ***    0.205       ***   0.050          -0.048          0.041    *     -0.061 ***
(1 = YES)                                        [6.17]            [6.39]            [1.49]         [-1.45]         [1.73]         [-2.58]

pension income exemption                         -0.075 *** -0.266 ***               0.079    **    -0.007          0.079    ***    0.003
(1 = YES)                                        [-2.86]    [-10.15]                 [2.50]         [-0.21]         [3.50]          [0.14]

average income tax rate                         -13.336 *** -9.609 ***               1.915           1.613          1.276           1.369
                                                 [-8.42]    [-6.06]                  [0.67]          [0.56]         [0.62]          [0.67]
property tax share                                0.085     -0.635 *** -0.036                        0.378          -0.032          0.527 ***
                                                  [0.41]    [-3.07]    [-0.13]                       [1.42]         [-0.17]         [2.74]
sales tax rate                                   -0.026 *** -0.051 *** -0.013                       -0.012          -0.005         -0.014
                                                 [-3.10]    [-6.05]    [-0.99]                      [-0.95]         [-0.59]        [-1.54]
other tax share                                  -0.984 *** -0.782 *** -0.414                        0.174          -0.443 *        0.187
                                                 [-4.32]    [-3.43]    [-1.10]                       [0.46]         [-1.66]         [0.70]
heating days                                    1.3E-05           -8.5E-05     ***
                                                  [1.44]            [-9.56]
state median house value                        3.5E-06     ***    3.4E-06     *** -1.5E-06      -2.5E-06 *** -1.4E-06 ** -2.5E-06 ***
                                                  [3.90]             [3.72]         [-1.62]       [-2.78]       [-2.15]     [-3.88]
state crime rate                                3.1E-04     ***    4.3E-04     *** 3.6E-05    * -2.1E-05       3.9E-05 *** -3.0E-05 **
                                                [25.61]            [35.17]           [1.93]       [-1.17]        [2.96]     [-2.32]
per capita income                                -0.032     ***     -0.083     *** 0.035      ** 0.034 ** 0.029 *** 0.038 ***
                                                 [-3.37]            [-8.62]          [2.46]        [2.40]        [2.86]      [3.75]
average state mfg wage                            0.014              0.029     *** 0.034      *    0.022         0.039 *** 0.023 *
                                                  [1.40]             [2.93]          [1.94]        [1.24]        [3.12]      [1.82]
state unemployment rate                          -0.050     ***     -0.077     *** -0.010          0.003        -0.012 *     0.003
                                                 [-5.71]            [-8.77]         [-1.03]        [0.32]       [-1.69]      [0.49]
% population aged 65 or more                      8.833     ***      5.948     *** 4.954      *** 6.872 *** 5.507 *** 6.687 ***
                                                [12.59]              [8.51]          [3.33]        [4.61]        [5.16]      [6.25]
expenditures on health & hospitals               -0.938     *** -1.369         ***  0.398 **         0.203           0.386 *** 0.206 *
                                                 [-7.29]        [-10.65]            [2.43]           [1.24]          [3.28]     [1.76]
education expenditures                            0.515     *** 0.530          *** -0.021            0.059          -0.018      0.045
                                                  [6.61]          [6.84]           [-0.21]           [0.60]         [-0.25]     [0.63]
welfare expenditures                             -0.170     ** -0.392          *** 0.046             0.053           0.042      0.035
                                                 [-2.41]         [-5.49]            [0.68]           [0.77]          [0.86]     [0.70]
other expenditures                                0.348     *** 0.264          *** -0.024           -0.078          -0.023     -0.097 ***
                                                  [7.84]          [5.97]           [-0.48]          [-1.54]         [-0.64]    [-2.68]
logged state population                           0.815 *** 0.672 *** 1.349 ***                      0.937    ***    1.308 *** 0.902 ***
                                                 [50.17]    [41.10]    [13.51]                       [9.40]         [18.34]    [12.68]
logged distance between states                   -1.297 ***            -1.517 ***
                                                [-84.43]             [-117.50]
year fixed effects                                YES                   YES                                         YES
origin and destination fixed effects               NO                   YES                                         NO
flow-specific fixed effects                        NO                    NO                                         YES
t-statistics reported in brackets. ***,**, * indicate significant at 1%, 5% and 10% levels respectively.
Table 3 -- Key Coefficients from Regressions Using Alternative Measures of Income Tax Preferences
                                   origin     destination   origin    destination origin     destination                                      origin     destination    origin   destination
                                      (from table 2)         (component amt)         total amounts                                          total amt (1980-2000)      TAXSIM (1980-2000)
EIG Taxation
    incremental EIG tax (YES=1) -3.5E-05       1.1E-04
                                   [0.00]        [0.00]
    Bakija and Slemrod tax rate                             0.001       0.001     0.004        -0.001                                        -0.012        -0.001      -0.011       -0.002
                                                            [0.15]      [0.21]    [0.56]       [-0.19]                                       [-1.52]       [-0.11]     [-1.44]      [-0.20]
Average Income Tax Rate
    taxes/income                   1.276         1.369      1.339       3.045     2.188         1.475                                        2.631         0.944
                                   [0.62]        [0.67]     [0.67]      [1.52]    [1.13]        [0.76]                                       [1.05]        [0.37]
    non-elderly average tax                                                                                                                                            -2.380       1.947
    rate via TAXSIM                                                                                                                                                    [-1.50]      [1.21]
Additional Credits/Deductions
for Elderly
    are any offered (1=YES)       -0.011        -0.020
                                  [-0.53]       [-0.93]
    value of credit/deduction                             -2.3E-04 ** 1.2E-04
                                                           [-2.14]      [1.05]
Social Security Exemption          0.041 * -0.061 ***
    is it exempt? (1=YES)          [1.73]       [-2.58]

    value of exemption                                                       2.4E-05        -6.0E-05 ***
                                                                              [1.27]         [-3.13]
Pension Income Exemption
  is it exempt? (1=YES)                       0.079     *** 0.003
                                              [3.50]        [0.14]
    value of exemption                                                       3.6E-06         1.7E-06
                                                                              [0.46]          [0.22]
Total Value of Elderly Tax
   sum of (3), (4), & (5)                                                                                     4.6E-08         -8.7E-07      -2.6E-06      -9.6E-07
                                                                                                               [0.05]          [-1.05]       [-1.32]       [-0.52]
Elderly Tax Savings
   TAXSIM calculation                                                                                                                                                  6.1E-05 *** -4.7E-05
                                                                                                                                                                        [2.01]      [-1.54]
Flow-Specific fixed effects and year fixed effects are included.
Variables from Table 2 listed under other taxes, cost of living, per capita expenditures and gravity model are included but not reported.
t-statistics reported in brackets. ***,**, * indicate significant at 1%, 5% and 10% levels respectively.
TABLE 4 -- Top 10 Elderly Migration State-to-State Flows, by Source and Income
Rank              Census                 IPUMS Highest 25% Income                IPUMS Lowest 25% Income
                  Origin   Destination         Origin   Destination              Origin   Destination
   1              NY       FL                  NY       FL                       NY       FL
   2              MI       FL                  IL       FL                       MI       FL
   3              IL       FL                  MI       FL                       IL       FL
   4              NY       NJ                  NJ       FL                       PA       FL
   5              OH       FL                  OH       FL                       OH       FL
   6              NJ       FL                  NY       NJ                       NJ       FL
   7              PA       FL                  PA       FL                       NY       NJ
   8              IL       CA                  IL       CA                       IL       CA
   9              NY       CA                  MA       FL                       IN       FL
  10              PA       NJ                  IN       FL                       MA       FL
# zero flows                    299                           1712                              1696
migration rate              0.0368                          0.0434                            0.0320
top 10 as % of all flows       30.3                            30.1                              21.1
   1               NY      FL                  NY       FL                       NY       FL
   2               NJ      FL                  NJ       FL                       NJ       FL
   3               OH      FL                  OH       FL                       OH       FL
   4               IL      FL                  IL       FL                       PA       FL
   5               PA      FL                  PA       FL                       NY       NJ
   6               MI      FL                  MI       FL                       MI       FL
   7               NY      NJ                  NY       NJ                       IL       FL
   8               NY      CA                  MA       FL                       MA       FL
   9               MA      FL                  NY       CA                       NY       CA
  10               CA      OR                  IL       CA                       CA       OR
# zero flows                    272                            1223                              1175
migration rate              0.0447                           0.0480                            0.0403
top 10 as % of all flows       29.7                             29.3                              17.8
   1               NY      FL                  NY       FL                       NY       FL
   2               NJ      FL                  NJ       FL                       NJ       FL
   3               CA      AZ                  MI       FL                       MI       FL
   4               MI      FL                  PA       FL                       OH       FL
   5               NY      NJ                  OH       FL                       CA       OR
   6               CA      OR                  IL       FL                       NY       NJ
   7               OH      FL                  MA       FL                       PA       FL
   8               CA      NV                  CT       FL                       MA       FL
   9               PA      FL                  CA       AZ                       CA       AZ
  10               IL      FL                  NY       NJ                       IL       FL
# zero flows                    139                              803                               834
migration rate              0.0437                           0.0492                            0.0379
top 10 as % of all flows       24.4                             22.8                              16.8
   1               NY      FL                  NY       FL                       NY       FL
   2               NJ      FL                  NJ       FL                       NY       NJ
   3               NY      NJ                  MI       FL                       NJ       FL
   4               OH      FL                  PA       FL                       MI       FL
   5               MI      FL                  CA       AZ                       PA       FL
   6               CA      AZ                  OH       FL                       CA       AZ
   7               PA      FL                  IL       FL                       CA       NV
   8               CA      NV                  MA       FL                       FL       GA
   9               MA      FL                  NY       NJ                       OH       FL
  10               IL      FL                  CA       NV                       FL       NY
# zero flows                     88                              694                               710
migration rate              0.0431                           0.0465                            0.0380
top 10 as % of all flows       21.7                             17.3                              14.2
Table 5 -- Key Coefficients from Regressions Using Alternative Measures of Income Tax Preferences for High-Income IPUMS Sample
                                   origin     destination       origin   destination   origin     destination   origin     destination   origin   destination
                                      (from table 2)             (component amt)          total amounts       total amt (1980-2000)    TAXSIM (1980-2000)
EIG Taxation
    incremental EIG tax (YES=1)   -0.168 *** -0.061 ***
                                  [-3.90]       [-1.40]
    Bakija and Slemrod tax rate                                -0.010      -0.010     -0.006        -0.008     -0.011        -0.020 *   -0.003      -0.018
                                                               [-0.84]     [-0.82]    [-0.46]       [-0.65]    [-0.80]       [-1.66]    [-0.27]     [-1.36]
Average Income Tax Rate
    taxes/income                   7.841 ** 2.024               2.305      -1.633      4.615        -0.744      5.527
                                   [2.05]      [0.52]           [0.63]     [-0.43]     [1.27]       [-0.20]     [1.30]
    non-elderly TAXSIM burden                                                                                                           -6.652 *** -1.310
                                                                                                                                        [-2.60]     [-0.50]
Additional Credits/Deductions
for Elderly
    are any offered (1=YES)        0.013        -0.018
                                   [0.35]       [-0.46]
    value of credit/deduction                                -5.5E-04 ** -2.1E-04
                                                               [-2.41]     [-0.89]
Social Security Exemption                                ***
    is it exempt? (1=YES)          0.208 *** 0.017
                                   [4.80]      [0.39]
    value of exemption                                        1.0E-04 *** 6.0E-05 *
                                                                [3.10]      [1.86]
Pension Income Exemption
    is it exempt? (1=YES)         -0.001        -0.065
                                  [-0.03]       [-1.56]
    value of exemption                                       -1.1E-06     -1.8E-05
                                                               [-0.07]     [-1.28]
Total Value of Elderly Tax
    sum of (3), (4), & (5)                                                           -2.5E-07      -1.8E-06   7.9E-06 ** 4.9E-06 *
                                                                                      [-0.14]       [-1.07]     [2.41]        [1.74]
Elderly Tax Savings
    TAXSIM calculation                                                                                                                 6.6E-05     -5.0E-06
                                                                                                                                         [1.34]     [-0.10]
Flow-Specific fixed effects and year fixed effects are included.
Variables from Table 2 listed under other taxes, cost of living, per capita expenditures and gravity model are included but not reported.
t-statistics reported in brackets. ***,**, * indicate significant at 1%, 5% and 10% levels respectively.
Table 6 -- Key Coefficients from Regressions Using Alternative Measures of Income Tax Preferences for Low-Income IPUMS Sample
                                   origin     destination    origin   destination   origin     destination   origin     destination   origin  destination
                                      (from table 2)          (component amt)          total amounts       total amt (1980-2000)    TAXSIM (1980-2000)
EIG Taxation
    incremental EIG tax (YES=1)   -0.126 *** -0.014
                                  [-2.87]       [-0.32]
    Bakija and Slemrod tax rate                             -0.010        0.016    -0.008         0.014     -0.013         0.015     -0.011      0.018
                                                            [-0.77]       [1.22]   [-0.68]        [1.12]    [-1.01]        [1.08]    [-0.82]     [1.35]
Average Income Tax Rate
    taxes/income                   0.663         2.506      -3.523        0.900    -1.289         2.101     -5.748        -2.000
                                   [0.17]      [0.64]       [-0.92]       [0.23]   [-0.35]        [0.56]    [-1.28]       [-0.45]
    non-elderly TAXSIM burden                                                                                                        -6.644 ** -4.498
                                                                                                                                     [-2.47]    [-1.59]
Additional Credits/Deductions
for Elderly
    are any offered (1=YES)       -0.005         0.041
                                  [-0.14]        [1.05]
    value of credit/deduction                             -1.9E-04      3.6E-04
                                                            [-0.88]       [1.52]
Social Security Exemption
    is it exempt? (1=YES)          0.213 *** 0.095 **
                                   [4.94]      [2.16]
    value of exemption                                     7.8E-05 *** 6.4E-05 *
                                                             [2.32]       [1.92]
Pension Income Exemption
    is it exempt? (1=YES)          0.055        -0.009
                                   [1.30]       [-0.21]
    value of exemption                                     1.3E-05     -1.9E-05
                                                             [0.78]      [-1.35]
Total Value of Elderly Tax
    sum of (3), (4), & (5)                                                        3.5E-06       -1.5E-06   1.8E-06       2.2E-06
                                                                                    [1.60]       [-0.93]     [0.57]        [0.72]
Elderly Tax Savings
    TAXSIM calculation                                                                                                              9.1E-05 * 6.8E-05
                                                                                                                                      [1.78]     [1.32]
Flow-Specific fixed effects and year fixed effects are included.
Variables from Table 2 listed under other taxes, cost of living, per capita expenditures and gravity model are included but not reported.
t-statistics reported in brackets. ***,**, * indicate significant at 1%, 5% and 10% levels respectively.

Appendix 1 -- Impact of Zero Flows on Regressions Using High-Income IPUMS Sample

                                                          dummy variable model (1970-2000)                                                   TAXSIM model (1980-2000)
                                              origin        destination             origin          destination             origin          destination       origin      destination

EIG Taxation
   incremental EIG tax (YES=1)               -0.168 ***       -0.061               -0.118 *          -0.054
                                             [-3.90]          [-1.40]              [-1.68]           [-0.76]
    Bakija and Slemrod tax rate                                                                                            -0.003            -0.018           0.005        -0.011
                                                                                                                           [-0.27]           [-1.36]          [0.25]       [-0.61]

Average Income Tax Rate
   taxes/income                               7.841 **        2.024                 7.510            0.7714
                                              [2.05]          [0.52]                [1.34]            [0.14]
    non-elderly TAXSIM burden                                                                                              -6.652 *** -1.310                  -4.114        1.337
                                                                                                                           [-2.60]    [-0.50]                 [-1.04]       [0.34]

Additional Credits/Deductions
for Elderly
    are any offered (1=YES)                   0.013           -0.018               -0.038             0.019
                                              [0.35]          [-0.46]              [-0.67]            [0.33]

Social Security Exemption
   is it exempt? (1=YES)                      0.208 ***       0.017                 0.019            -0.128 *
                                              [4.80]          [0.39]                [0.28]           [-1.94]

Pension Income Exemption
   is it exempt? (1=YES)                     -0.001           -0.065                0.169     ***     0.099
                                             [-0.03]          [-1.56]               [2.71]            [1.59]

Elderly Tax Savings
   TAXSIM calculation                                                                                                     6.6E-05           -5.0E-06          1.4E-04 *   -6.4E-05
                                                                                                                           [1.34]            [-0.10]          [1.84]       [-0.86]
Include zero flows                              no                                   yes                                     no                                 yes

Flow-Specific fixed effects and year fixed effects are included.
Variables from Table 2 listed under other taxes, cost of living, per capita expenditures and gravity model are included but not reported.
t-statistics reported in brackets. ***,**, * indicate significant at 1%, 5% and 10% levels respectively.

Appendix 2 -- Impact of Zero Flows on Regressions Using Low-Income IPUMS Sample

                                                           dummy variable model (1970-2000)                                                 TAXSIM model (1980-2000)
                                              origin       destination              origin        destination            origin        destination           origin    destination

EIG Taxation
   incremental EIG tax (YES=1)                -0.126 *** -0.014                     0.011           -0.007
                                              [-2.87]    [-0.32]                    [0.17]          [-0.12]
    Bakija and Slemrod tax rate                                                                                          -0.011             0.018           0.0004      0.0004
                                                                                                                         [-0.82]            [1.35]           [0.02]      [0.02]

Average Income Tax Rate
   taxes/income                               0.663           2.506                11.223 **        -3.357
                                              [0.17]          [0.64]                [2.20]          [-0.66]
    non-elderly TAXSIM burden                                                                                            -6.644 *** -4.498                  -1.958      -2.433
                                                                                                                         [-2.47]    [-1.59]                 [-0.50]     [-0.62]

Additional Credits/Deductions
for Elderly
    are any offered (1=YES)                   -0.005          0.041                 0.049            0.106      **
                                              [-0.14]         [1.05]                [0.94]           [2.03]

Social Security Exemption
   is it exempt? (1=YES)                      0.213 *** 0.095 **                    0.103     *     -0.060
                                              [4.94]    [2.16]                      [1.71]          [-0.99]

Pension Income Exemption
   is it exempt? (1=YES)                      0.055          -0.009                 0.064            0.050
                                              [1.30]         [-0.21]                [1.13]           [0.88]

Elderly Tax Savings
   TAXSIM calculation                                                                                                   0.0001 *        0.0001              0.0002 *** 0.0001
                                                                                                                         [1.78]          [1.32]              [2.62]     [1.43]
Include zero flows                              no                                   yes                                   no                                 yes

Flow-Specific fixed effects and year fixed effects are included.
Variables from Table 2 listed under other taxes, cost of living, per capita expenditures and gravity model are included but not reported.
t-statistics reported in brackets. ***,**, * indicate significant at 1%, 5% and 10% levels respectively.


To top