Total state in-migration rates and public policy in the United States: a comparative analysis of the Great Recession and the pre- and post-Great Recession years

Abstract

This study adopts state-level data to investigate empirically the Tiebout hypothesis (as extended by Tullock) of ‘voting with one’s feet’ for the period referred to in the United States as the ‘Great Recession’ (2007–09). As compared with previous studies, more recent data are used and estimates are provided for three time periods: the ‘Great Recession’ (from 1 July 2007 to 30 June 2009), the pre-Great Recession period (1 July 2004 to 30 June 2006) and the post-Great Recession period (1 July 2009 to 30 June 2011). This analysis also differs from most previous literature by including a separate cost-of-living variable and a variable measuring effective state personal income tax rates. After allowing for various economic factors and quality of life/climate variables, migrants (consumer-voters) over the 2007–09 period appear to prefer states with lower effective state personal income tax rates and higher levels of ‘fiscal surplus’, defined in this study for each state as the total outlay per pupil on primary and secondary public education minus the per capita property tax level. The three empirical estimates all demonstrate that the Tiebout/Tullock hypothesis was operational not only during but also both before and after the Great Recession since for all three time periods migrants (consumer-voters) manifested a preference for lower effective state personal income tax rates and higher levels of fiscal surplus.

Keywords:

• total in-migration
• effective state income tax rates
• fiscal surplus

Introduction

Determinants of human migration, especially within the United States, but elsewhere as well, have been and continue to be a topic of research interest (Ashby, 2007; Carrington, Detragiache, & Vishwanath, 1996; Chi & Voss, 2005; Conway & Houtenville, 2001, 2003; Francis, 2007; Fu & Gabriel, 2013; Landry, Bin, Hindsley, Whitehead, & Wilson, 2007; Nechyba, 2000; Partridge & Rickman, 2006; Percy, Hawkins, & Maier, 1995; Peters, 2012; Plantinga, Detang-Dessendre, Hunt, & Piguet, 2013; Renas, 1983; Rhode & Strumpf, 2003; Saltz, 1998; Vedder, Gallaway, Graves, & Sexton, 1986). The nature of issues considered within the context of internal migration determinants is extremely diverse; indeed, it has (naturally) become increasingly diverse over time as a consequence of increasingly more complex research by multiple researchers and in a variety of disciplines over time. One of the areas receiving the greatest attention involves the so-called Tiebout (1956) hypothesis (Banzhaf & Walsh, 2008; Cebula, 1978, 1990; Cebula & Alexander, 2006; Renas, 1980; Rhode & Strumpf, 2003), also sometimes referred to as the Tiebout (1956)/Tullock (1971) hypothesis.

According to Tiebout (1956, p. 418), ‘the consumer-voter may be viewed as picking that community which best satisfies his preferences for public goods […] the consumer-voter moves to that community whose local government best satisfies his set of preferences.’ As Tullock (1971, p. 917) further observes, this hypothesis can effectively be extended such that it holds that, ceteris paribus, the ‘individual deciding where to live will take into account the private effects upon himself of the bundle of government services and taxes […]’. Thus, Tullock more explicitly than Tiebout emphasizes that the consumer-voter evaluates both the government goods and services and the tax burden at the relevant locations of choice.

Most studies on the impact of public policy on migration in the United States have effectively involved migration prior to the Great Recession, which is identified by the National Bureau of Economic Research (NBER) as the period between December 2007 and the end of June 2009. A modest number of recently published studies have addressed the economic response of regions to a major recession (Connaughton & Madsen, 2012; Hassink, 2010; Martin, 2011; Walden, 2012).

Hassink (2010) studies the linkage of regional economic resilience to regional economic adaptability under major recessionary circumstances, whereasMartin (2011) further develops the idea of regional economic resilience and investigates its usefulness, especially when combined with the notion of hysteresis, in understanding the response of regional economies to a major recessionary shock or (simply major recessions) in the UK.

Connaughton & Madsen (2012) identify US states that suffered the greatest job losses and those that experienced the least job losses during the Great Recession. They find large differences between states in terms of the negative consequences (job losses and higher unemployment rates) from the Great Recession. Connaughton & Madsen (2012) also find that the industrial composition within each state contributed to its economic resilience and its adaptability to the shock of the Great Recession and hence to the magnitude of job losses and to the extent that unemployment rates were elevated. Walden (2012) finds preliminary evidence that in a large percentage of cases during Great Recession in the United States, migrants who moved into states with lower unemployment rates were themselves unemployed; hence, they were more strongly motivated by the need to find employment and secure an income than they were to seek lower state and local tax burdens and higher public school outlays per se. Thus, Walden raises the idea that the motivations underlying state in-migration during the Great Recession were rather different from those prior to the Great Recession, and that the linkage between migration and public policy may have been weakened by the economic experience of the Great Recession. Although an interesting idea, the ad hoc model and empirical evidence regarding this proposition in his single ordinary least squares (OLS) estimate are limited, leaving the issue open for further investigation.

Interestingly, the Tiebout–Tullock hypothesis has not been empirically studied for the United States for the period of the Great Recession and the years immediately prior to and following the Great Recession. Accordingly, the present study seeks to help fill this void by empirically investigating this hypothesis using contemporary state-level data for the United States. Conducting this study also provides an opportunity to investigate empirically the preliminary hypothesis by Walden as summarized above.

The present empirical study first addresses the determinants of the state-level total in-migration rate during the Great Recession (2007–09). During this period it is estimated that approximately 14 136 246 persons in the United States moved their principal residence from one state to another. ‘The total state in-migration rate’ for each time period studied is defined in this study as the ratio of the total number of domestic migrants moving into each given state from all the remaining 49 states during a given study period relative to the beginning-of-period total population in each state. This study seeks to shed light on whether fiscal factors such as outlays per pupil on public primary and secondary education, per capita property tax liabilities, and state personal income tax rates influenced consumer-voters’ mobility decisions during this Great Recession study period. In addition, this study seeks to demonstrate whether these same factors, i.e. outlays per pupil on public primary and secondary education, per capita property tax burdens, and state personal income tax rates, influenced state total in-migration rates in the pre-Great Recession (2004–06) and post-Great Recession years (2009–11).¹ Thus, three sets of empirical findings are provided, each with the same explanatory variables, although reflecting values for different years, depending upon the study period. Lagged values of explanatory variables were used in all estimates to address issues of potential endogeneity.

Numerous previous studies have empirically addressed determinants of internal migration within the United States. A number of these emphasize the migration impact not only of economic and fiscal factors but also of non-economic factors, including ‘quality-of-life’ factors, especially climate (Cebula & Alexander, 2006; Cebula & Belton, 1994; Clark & Hunter, 1992; Conway & Houtenville, 1998, 2001, 2003; Davies, Greenwood, & Li, 2001; Gale & Heath, 2000; Milligan, 2000; Renas, 1978, 1980, 1983; Saltz, 1998; Vedder, 1976). As demonstrated in these studies, the omission of non-economic factors, especially climate-reflecting variables, from an empirical migration analysis constitutes an omitted-variable problem that generally compromises the integrity of that analysis. As a consequence, this empirical study will include not only fiscal factors and economic factors but also purely quality-of-life factors.

The basic migration decision framework

This section follows the economic path of most migration studies in the United States since the contributions by Sjaastad (1962) and Riew (1973), among others, and treats the consumer-voter’s overall migration decision as an investment decision such that the decision to migrate from area i to area j requires that his/her expected net discounted present value of migration from area i to area j, DPVij, be both (1) positive and (2) the maximum net discounted present value that can be expected from moving from area i to any other known and plausible alternative area.

Following in principle the models given in Tiebout (1956), Sjaastad (1962), Tullock (1971), Riew (1973), Vedder (1976), Cebula (1978), Renas (1983), Vedder et al. (1986), and Cebula & Alexander (2006), among others, DPVij in this study consists of three broad sets of considerations:

• Economic conditions in those areas.

• Fiscal factors in those areas.

• Environmental characteristics of the areas.

According to this investment framework, it follows that migration will flow from area i to area j only if:(1)

where z represents all the plausible known alternative locations to area i. Given the focus in this study on state migration, area j is actually state j. Clearly, although the framework is an investment model, either a carefully crafted constrained-utility maximization model or a carefully constructed cost–benefit model (Cebula, 1978) could have been adopted and used to introduce the same categories of variables.

The total state in-migration rate for the Great Recession, MIGj, is the total number of domestic in-migrants to state j over the period 1 July 2007–30 June 2009, expressed as a percentage of the year 2007 population in state j. This specification in percentage terms allows comparisons of each state’s total in-migration rate with those of the other states. The only available data on state total in-migration in the United States follow this 1 July–30 June time frame pattern; thus, for roughly the first five months of the Great Recession study period, the economy was not officially/technically in recession, although it was materially slowing down prior to December 2007 (Council of Economic Advisors, 2009, table B-2).

Three variables are used to measure the economic conditions in state j. The first of these purely economic variables is the per capita personal income in state j, PCPERSINCj. Conventional wisdom suggests that the higher the per capita personal income in a state, the higher the expected income in that state and hence the more appealing it would be to move to that state. Thus, MIGj, the total domestic in-migration rate to state j, is hypothesized to be an increasing function of PCPERSINCj, ceteris paribus. This variable constitutes an average measure of expected income/wage prospects in state j if one were actually to have a job in that state.

The second purely economic variable included in the model is the percentage unemployment rate in state j, URj, which is the average unemployment rate of the civilian labour force over the course of the year in state j. The higher the unemployment rate in a state, the less promising are the job prospects in the state and hence the less appealing that state is as a place of residence (Cebula, 1979; Riew, 1973; Sommers & Suits, 1973; Walden, 2012). Clearly, then, the total domestic in-migration rate to state j is expected to be a decreasing function of URj, ceteris paribus.

The third purely economic variable is COSTj, the overall cost of living in state j for the average four-person family, expressed as an index, with COSTj = 100.00 being the theoretical mean value of this indexed variable. In the absence of money illusion, the total domestic in-migration rate to state j is hypothesized to be a decreasing function of COSTj, ceteris paribus, because the higher the cost of living, the lower one’s real income and the lower one’s standard of living. The adoption of a variable such as COSTj in migration studies is becoming more common (Cebula, 1979; Cebula & Alexander, 2006; Conway & Houtenville, 1998, 2001, 2003; Gale & Heath, 2000; Renas, 1978, 1980, 1983).

Three variables are adopted to measure fiscal factors, although two of them are eventually merged into one to avoid multicollinearity. The first fiscal factor variable considered in this study is AVSTINCTRj, the average effective percentage state personal income tax rate in state j. This variable is the average state personal income tax paid by residents of state j, expressed as a percentage of the average family income in state j. Such a variable has usually been overlooked in studies of a Tiebout-type framework, although it has been considered on a limited basis, i.e. by a few studies (Cebula, 1990; Cebula & Alexander, 2006; Conway & Houtenville, 2001). It is hypothesized here that the total domestic in-migration rate to state j is a deceasing function of AVSTINCTRj, ceteris paribus (Tullock, 1971), because a higher effective state personal income tax implies a lower disposable income. Since nine states do not have a state income tax, for these states the value of AVSTINCTRj = 0. The nine states in question are Alaska, Florida, Nevada, New Hampshire, South Dakota, Tennessee, Texas, Washington and Wyoming.

The variable PCPROPTXj is defined as the total local (city plus county) per capita nominal property tax liability on residential property in state j. Such a variable has often been considered in studies of a Tiebout-type framework (Barsby & Cox, 1975; Cebula & Alexander, 2006; Conway & Houtenville, 1998, 2001; Gale & Heath, 2000; Greene, 1977; Liu, 1977; Pack, 1973; Renas, 1980; Rhode & Strumpf, 2003). It is expected that the total domestic in-migration rate to state j is a decreasing function of PCPROPTXj, ceteris paribus, because a higher property tax burden implies a lower disposable income. The principal outlay financed by property taxes is of course public primary and secondary education. Continuing, the variable ANNUALPERPUPj is the nominal outlay in state j per pupil on primary and secondary public education from all sources, federal, state and local. The variable ANNUALPERPUPj is be distinguished from the very commonly adopted variable per capita public education outlays found in many previous studies (Cebula, 1979; Conway & Houtenville, 1998, 2001; Gale & Heath, 2000; Greene, 1977; Hinze, 1977; Pack, 1973; Renas, 1980; Rhode & Strumpf, 2003). Indeed, it is stressed in this study, as hypothesized in a prior study by Cebula & Alexander (2006), that higher per pupil outlays on public education are a more direct measure of a commitment to better quality education than higher per capita outlays on public education. In any event, it is hypothesized here that, ceteris paribus, the total domestic in-migration rate to state j is an increasing function of ANNUALPERPUPj because a greater outlay on public education per pupil arguably implies a greater commitment to the pursuit of a better quality public education system in the state, as well as a reduced inclination or perceived need to choose private education (with its accompanying pecuniary costs, especially tuition) over public education.

Not surprisingly, the variables PCPROPTXj and ANNUALPERPUPj are highly correlated (r = 0.63). Accordingly, a ‘fiscal surplus’ variable labelled FISCSURPj is constructed, which is the value of ANNUALPERPUPj minus the value of PCPROPTXj. The variable FISCSURPj is a measure of the average excess of the value received from per pupil outlays on primary and secondary public education in state j minus the per capita property tax associated with paying most of those outlays in state j. According to Tiebout (1956) and Tullock (1971), as well as Buchanan (1950), the consumer-voter seeks to maximize fiscal surplus. Thus, it is hypothesized in this study that the total domestic in-migration rate to state j is an increasing function of FISCSURPj, ceteris paribus.

Finally, there are three quality-of-life variables included in the model. Two are climatic in nature. The first of these climatic variables is JANTEMPj, defined here as the mean January temperature in state j (1971–2000), as a measure of warmer climatic conditions. As in so many migration studies (Cebula & Alexander, 2006; Cebula & Belton, 1994; Clark & Hunter, 1992; Conway & Houtenville, 1998, 2001, 2003; Davies et al., 2001; Gale & Heath, 2000; Milligan, 2000; Renas, 1978, 1980, 1983; Saltz, 1998), this variable is treated as a quality-of-life control variable. As is typically the case in these studies, it is hypothesized that a warmer climate is likely to increase the inflow of migrants as a reflection of their typical preference for warmer weather, especially during the winter months. Thus, it is hypothesized that the total domestic in-migration rate to state j is an increasing function of JANTEMPj, ceteris paribus.

As a supplemental measure of climate, one that arguably reflects elevated summertime heat and elevated humidity, the model includes the variable CDDj, the average annual number of cooling degree-days in state j (1971–2000). In this case, the focus is on the discomfort typically associated with climates having much higher summer temperatures and higher humidity. Given that a higher number of annual cooling degree-days reflects conditions of greater summer heat and humidity, it is hypothesized, given presumed migrant preferences for (higher valuation of) physical comfort (Cebula & Vedder, 1973; Clark & Hunter, 1992; Renas, 1978, 1983; Saltz, 1998), that the total domestic in-migration rate to state j is a deceasing function of CDDj, ceteris paribus.

The third and final expressly quality-of-life variable is population density, POPDENj, defined as the number of people per square mile in state j. Arguably, greater population density per se can be associated with greater roadway congestion, greater congestion on public transit facilities and in the general conduct of life, such as more crowded grocery stores, theatres, restaurants, shopping malls and the like. To the extent that such congestion and crowding adds travel time to work or other destinations or simply creates discomfort in the form of waiting lines and delays, it follows that greater population density would act as a deterrent to in-migration (Cebula, 1979; Cebula & Vedder, 1973; Renas, 1978, 1983; Saltz, 1998). Hence, it is hypothesized that the total domestic in-migration rate to state j is a decreasing function of POPDENj, ceteris paribus.

The initial estimate: the Great Recession years, 2007–09

The reduced-form equation initially to be estimated is given by:(2)

where a₀ is a constant term; and u is a stochastic error term. The time period following each variable in equation (2) indicates the adoption of the value of that variable during the specified time period. As noted above, the explanatory variables are lagged to avoid potential endogeneity problems.

The study includes all 50 states, but excludes Washington, DC, as an outlier; indeed, its inclusion altered the results for several variables. Moreover, it can be argued as a legitimate omission since it is in fact not a state. In fact, inclusion of Washington, DC, could well raise the question of whether Puerto Rico should have been included in the study. Finally, omission of Washington, DC, is consistent with most previous migration studies of the United States.

The data source for the variable MIGj was US Census Bureau (2011, table 33; 2012, table 33). The source for variable PCPERSINCj, was US Census Bureau (2010, table 665), while the data source for variable COSTj was Council for Community and Economic Research (2013). The data source for variable URj was US Census Bureau (2009, table 609). The source for variable JANTEMPj was US Census Bureau (2010, table 378), whereas the data source for variable CDDj was US Census Bureau (2010, table 384) and that for POPDENj was US Census Bureau (2012, table 14). The data for the policy variables AVSTINCTRj, PCPROPTXj and ANNUALPERPUP were obtained to compute FISCSURPj from US Census Bureau (2010, table 13) and US Census Bureau (2010, table 242; 2012, table 555); the latter two variables are used to construct the fiscal surplus variable, FISCSURPj. For the convenience of readers, full definitions of all variables used in this analysis are provided in the Appendix. For interested readers, descriptive statistics in the forms of means, standard deviations, maximum values and minimum values are provided in Table 1.

Table 1. Descriptive statistics on variables, 2007–09 estimation.

Variable	Arithmetic average	Standard deviation	Maximum	Minimum
MIGj	5.958	2.055	11.9	2.5
PCPERSINCj	29 951	4425	43 047	22 092
COSTj	96.298	17.305	138.2	81.5
EMPLGRRj	7.291	3.597	19.82	0.749
AVSTINCTRj	2.898	1.962	7.851	0.00
FISCSURPj	2522	3486	5005	–1849
JANTEMPj	32.7	12.65	73.0	10.2
CDDj	1283.5	984.49	4561	0.0
POPDENj	188.82	256.45	1170	1.15

Based on the arguments provided above and the ‘conventional wisdom’ as expressed by Tiebout (1956), Tullock (1971) and Riew (1973), and more recently by Conway & Houtenville (2001) and Cebula & Alexander (2006), among others, the following coefficient signs (each case assumes ceteris paribus) are hypothesized:(3)

Formal testing for the presence of heteroskedasticity suggests that it is a problem; indeed, this problem was found to be present in all three of the estimations used in this study. Accordingly, the White (1980) heteroskedasticity correction is adopted in all three cases. Interestingly, an examination of the correlations among the explanatory variables for multicollinearity reveals that it does not seem to be a problem, given the use of the fiscal surplus variable.

The results of estimating equation (2) by OLS are provided in column (a) of Table 2. All eight of the estimated coefficients exhibit the expected signs, with three being statistically significant at the 1% level and three being statistically significant at the 5% level. The estimated coefficients on two of the explanatory variables, URj (the percentage unemployment rate in state j) and CDDj (the average annual number of cooling degree-days in state j), fail to be statistically significant at the 10% level.²

Table 2. Empirical results for total state in-migration rate. Dependent variable: MIGj.

	Study period
	2007–09	2004–06	2009–11
Variable/column	(a)	(b)	(c)
PCPERSINCj	0.0002*** (2.62)	0.00015** (2.20)	0.00012*** (2.66)
URj	–0.29 (–1.08)	–0.417* (–1.88)	–0.525** (–2.23)
COSTj	–0.014** (–2.14)	–0.019** (–2.40)	–0.008**
AVSTINCTRj	–0.357*** (–2.72)	–0.374*** (–3.33)	–0.29** (–2.60)
FISCSURPj	0.000261*** (7.41)	0.00024*** (6.77)	0.00021*** (6.68)
JANTEMPj	0.069** (2.53)	0.119*** (3.78)	0.069** (2.61)
CDDj	–0.00055 (–1.57)	–0.0009 (–1.63)	–0.00055 (–1.53)
POPDENj	–0.0021** (–2.17)	–0.0026*** (–3.10)	–0.0024*** (–2.63)
Constant	1.81	4.34	3.54
n	50	50	50
R^2	0.68	0.64	0.68
Adj. R^2	0.62	0.57	0.62
F	10.96***	9.18***	10.99***

Note:

*** Statistically significant at the 1% level;

** statistically significant at the 5% level; and

* statistically significant at the 10% level. t-values are given in parentheses.

First consider the expressly economic variables. PCPERSINCj, the variable that provides a measure for income expectations from residence in state j (the per capita personal income in state j), exhibits a coefficient that is positive and statistically significant at the 1% level. Hence, for the years of the Great Recession, this result suggests that the total state in-migration rate was positively associated with higher levels of per capita personal income in that state, i.e. with higher expected income in that state. More specifically, over the 2007–09 period (i.e. during the Great Recession), a 1-unit increase (US$1 in current United States dollars) in a state’s personal income per capita would be associated with an increase in that state’s total in-migration rate of 0.0002 or 0.02%. Alternatively stated, if a state’s personal income per capita were to increase by US$1000, its increased per capita income would in theory be associated with a 20% increase in its total in-migration rate, everything else held the same.

Next, the cost-of-living variable exhibits a negative coefficient that is statistically significant at the 5% level. Not surprisingly perhaps, given the economic hardships imposed by the Great Recession, this result suggests that the total in-migration rate was negatively associated with higher cost-of-living levels in that state. As the estimated coefficient implies, over this time period a 1-unit increase in a state’s overall cost-of-living index would be associated with a decrease in the total in-migration rate to that state of 0.014, or 1.4%.

Next, consider the purely quality-of-life variables. The coefficient on the JANTEMPj variable is positive and statistically significant at the 5% level; thus, this result suggests, as anticipated, that higher total in-migration rates were positively associated with a higher average January temperature, i.e. states with warmer climates (Clark & Hunter, 1992; Conway & Houtenville, 1998, 2001; Gale & Heath, 2000; Renas, 1983; Saltz, 1998). Thus, during the Great Recession, a 1-unit increase in a state’s 30-year average January temperature, e.g. from the sample average of 32.7 to 33.7°F, would be associated with an increase in that state’s total in-migration rate of 6.9%, everything else constant.

The estimated coefficient on the population density variable is negative and statistically significant at the 5% level. This result suggests that the total in-migration rate in this study period was negatively associated with population density. More specifically, a rise in the population density of a state by one person per square mile would be associated with a reduction in the total in-migration rate to that state of 0.0022, i.e. 0.22%. Hence, over this study period, a 10-person rise in a state’s population density would be associated with a 2.2% reduction in the total in-migration rate to that state, other things held the same. The other quality-of-life variable (CDDj) does not have a coefficient statistically significant at even the 10% level.

Finally, we focus on the public policy variables. AVSTINCTRj, the average effective state personal income tax rate in state j, has a negative coefficient that is statistically significant at the 1% level. Thus, this result suggests that as hypothesized, lower total in-migration rates were associated with higher average effective state personal income tax rates. In this case, other things held the same, a 1-unit increase in a state’s average effective percentage state personal income tax rate, e.g. from 3% to 4%, would be associated with a decrease in that state’s total in-migration rate of nearly 36% since the coefficient is –0.357. This variable appears to be a potent Tiebout (1956)/Tullock (1971)-type variable, implying that state policy-makers should be especially circumspect about using this taxing tool as a source of revenue enhancement.

As for the fiscal surplus variable, FISCSURPj, the estimated coefficient on this variable is positive and statistically significant at the 1% level; hence, this result suggests, as hypothesized, that higher total in-migration rates were positively associated with higher levels of fiscal surplus (Cebula & Alexander, 2006; Pack, 1973; Tiebout, 1956; Tullock, 1971). Regarding this variable, a 1-unit (US$1) increase in a state’s fiscal surplus would be associated with an increase in that state’s total in-migration rate of 0.000261, or 0.0261%. Hence, an increase in the fiscal surplus (as defined) in a state equal to US$100 would be associated with an increase in the total in-migration rate equal to 2.61%, on average, other things held the same.

In sum, then, for the Great Recession period 2007–09, the total state-in-migration rate was positively associated with per capita personal income, warmer January temperatures and higher levels of fiscal surplus. For the same study period, the total state in-migration rate was negatively associated with higher levels of the average overall cost of living, the average effective state personal income tax rate and population density.

The pre- and post-Great Recession estimates: 2006–06 and 2009–11

We also estimate reduced-form equations (4) and (5), which parallel equation (2), for the periods 2004–06 and 2009–11 respectively:(4) (5)

The data source for variable MIGj 2004–2006 was US Census Bureau (2008, table 33; 2009, table 32), and the data sources for variable MIGj 2009–2011 were the US Census Bureau, American Community Survey (2010,s 2011) and US Census Bureau (2012, table 19). The remaining data were also obtained from the US Census Bureau (2001, 2009, 2010, 2011, 2012) and from Council for Community and Economic Research (2013). The White (1980) heteroskedasticity-corrected estimations of equations (4) and (5) are provided in columns (b) and (c), respectively, in Table 2.

In column (b) of Table 2, all eight of the estimated coefficients exhibit the hypothesized signs, with four statistically significant at the 1% level, two statistically significant at the 5% level, one statistically significant at the 10% level (URj), and only one statistically insignificant at the 10% level (CDDj).³

The variable per capita personal income exhibits a coefficient that is positive and statistically significant at the 5% level. Hence, during the 2004–06 timeframe, this result suggests that the total state in-migration rate was positively associated with higher levels of per capita personal income. Next, the cost-of-living variable exhibits a negative coefficient that is statistically significant at the 5% level. This result suggests that the total state 2004–06 in-migration rate was negatively associated with higher cost-of-living levels. Finally, there is weak evidence of an association between the total state in-migration rate and the unemployment rate for this period at the 10% level of significance.

Among the quality-of-life variables, the coefficient on the warm weather variable, JANTEMPj, is positive and statistically significant at the 1% level. This result suggests that higher total in-migration rates were positively associated with a higher average January temperature (Clark & Hunter, 1992; Conway & Houtenville, 1998, 2001; Gale & Heath, 2000; Renas, 1983; Saltz, 1998). The estimated coefficient on the population density variable, POPDENj, is negative and statistically significant at the 1% level, suggesting a negative association between the 2004–06 total state in-migration rate and population density. CDDj (the average annual number of cooling degree-days in state j) was not statistically significant even at the 10% level.

Finally, we focus on the public policy variables. The coefficient on AVSTINCTRj, the average effective state personal income tax rate in state j, is negative and statistically significant at the 1% level. This result suggests, as hypothesized, that lower total state in-migration rates were associated with higher average effective state personal income tax rates. The coefficient on the fiscal surplus variable, FISCSURPj, is positive and statistically significant at the 1% level. This result suggests, as expected, that higher total state in-migration rates were associated with higher levels of fiscal surplus (Cebula & Alexander, 2006; Pack, 1973; Tiebout, 1956; Tullock, 1971).

The results for the period 2009–11 are very similar to those for the other two study periods. Overall, as shown in column (c) of Table 2, all eight coefficients exhibit the expected signs, with three statistically significant at the 1% level and four statistically significant at the 5% level. Only the coefficient on the CDDj variable fails (again) to be statistically significant at the 10% level.

In sum, for the 2009–11 study period, the total state in-migration rate was positively associated with per capita personal income, warmer January temperatures and higher levels of fiscal surplus. Furthermore, for this same study period, the total state in-migration rate was negatively associated with the average overall cost of living, population density, unemployment rate (as suggested by Walden, 2012), and the effective state personal income tax rate.

Conclusions

This empirical study has investigated fiscal as well as economic and non-economic (quality-of-life) determinants of total state in-migration in the United States over three time periods: the Great Recession period, 2007–09; a pre-Great Recession time period, 2004–06; and a post-Great Recession time period, 2009–11. The principal intent of investigating these three different periods has been to provide evidence regarding whether there continues to be empirical support for the Tiebout (1956)/Tullock (1971) hypothesis, be it during, before or after the Great Recession.

The findings provided in this study support the Tiebout (1956)/Tullock (1971) hypothesis in terms of the contemporary mobility of consumer-voters. For all three of these study periods, the model specification has allowed for five purely economic factors, three quality-of-life variables and two fiscal factors, one of which, FISCSURPj, embodies, i.e. is the arithmetic difference between two fiscal factors, the annual per pupil outlay on public primary and secondary education minus the per capita property tax. For each of the three study periods, it appears that the following is the case: the total state in-migration rate was positively associated with per capita personal income, warmer January temperatures and higher levels of fiscal surplus. In addition, for these same study periods, the total state in-migration rate was also negatively associated with the average overall cost of living, population density and the average effective state personal income tax rate. The results for both the effective state personal income tax rate variable and the fiscal surplus variable imply that the search for ‘fiscal surplus’ appears to be ongoing (Buchanan, 1950) and that the search was not effectively altered by the various experiences of the Great Recession.

Moreover, as observed in the above analysis, the average state personal income tax rate appears to be a potent Tiebout (1956)/Tullock (1971)-type variable, implying that state policy-makers should be especially circumspect about raising this tax rate as a source of revenue enhancement, since such a policy could very well have perverse effects on their state revenues.

Notes

1. During the pre-Great Recession period (2004–06), it is estimated that approximately 14 999 128 people in the United States moved their residence from one state to another; the corresponding figure for the post-Great Recession period (2009–11) was 13 730 645 people.

2. The coefficient of determination (R²) is 0.68, whereas the adjusted coefficient of determination (adj. R²) is 0.62, so that the model explains approximately two-thirds of the variation in the dependent variable, MIGj. Finally, the F-statistic is statistically significant at the 1% level, attesting to the overall strength of the model.

3. The coefficient of determination (R²) is 0.64, whereas the adjusted coefficient of determination is 0.57, so that the model explains approximately three-fifths of the variation in the dependent variable, MIGj. Finally, the F-statistic is statistically significant at the 1% level, attesting to the overall strength of the model.

Appendix

The purpose here is to provide a clear and simple definition for each variable considered in this study. The definition excludes years and focuses on what each variable represents:

• MIGj = total state in-migration rate to state j, defined as the total number of in-migrants to each state (j) over a specified time period divided by the total population of each state (j) at the beginning of the study period; this number is converted from a decimal to a percentage, as is standard in nearly all the migration literature.

• PCPERSINCj = personal income per capita earned in a specified year in each state (j); this number is expressed in current US dollars.

• URj = average unemployment rate of the civilian labour force in each state (j) over the course of a specified year; this number is expressed as a percentage.

• COSTj = index of the overall cost of living for the ‘average’ four-person family residing in each state (j) during the course of a given year; this number is expressed as an index, with a higher index value indicating a higher overall cost of living.

• AVSTINCTRj = average effective percentage state personal income tax rate in each state (j); this variable is the average state personal income tax paid by residents of each state (j) expressed as a percentage of the average family income in each state (j).

• PCPROPTXj = total local (city plus county) property tax liability (in current US dollars) on residential property per capita in each state (j).

• ANNUALPERPUPj = annual nominal outlay (in current US dollars) in each state (j) per pupil on primary and secondary public education from all sources, i.e. federal plus state plus local.

• FISCSURPj = ANNUALPERPUPj – PCPROPTXj = average level of fiscal surplus in each state j, i.e. the annual nominal outlay (in current US dollars) in each state (j) per pupil on primary and secondary public education from all sources, i.e. federal plus state plus local, minus the total local (city plus county) property tax liability (in current US dollars) on residential property per capita in each state (j).

• JANTEMPj = average January temperature in degrees Fahrenheit in each state (j); this average was computed over the 30-year period from 1971 to 2000.

• CDDj = average annual number of cooling degree-days in each state (j); this average was computed over the 30-year period from 1971 to 2000.

• POPDENj = average number of people residing in each state (j) per square mile in any specified year.