rural bound:determinantsof metroto non-metro … · 2017-10-02 · this article investigates the...

RURAL BOUND:DETERMINANTS OF METRO TO

NON-METRO MIGRATION IN THE UNITED STATES

ANIL RUPASINGHA, YONGZHENG LIU, AND MARK PARTRIDGE

A general global precept is that agglomeration forces lead to migration from rural to urban areas.Yet for much of the time since the early 1970s, more people have moved from metro to nonmetroU.S. counties. The underlying causes of this pattern have changed over time with economic shocksand changing household preferences. For instance, the post 2000 period has seen a significantdecline in domestic migration rates, a significant increase in commodity prices that favor ruralareas, and potential changes in the valuation of natural amenities that would affect migration.This article investigates the determinants of U.S. gross migration from metro to nonmetro countiesand nonmetro to metro counties for the 1995–2000 and 2005–2009 periods in order to comparethe differences in rural to urban and urban to rural migration, as well as compare the 1990s tothe 2005–2009 periods. More specifically, the present study extends the literature by more broadlyexamining the underlying factors associated with deconcentration and economic restructuringarguments of metro to nonmetro migration. The article uses (1) extensive county-to-county migra-tion flows and (2) the utility maximization theory that extends the framework of a discrete choicemodel. The results show that population density, distance to urban areas, industry mix employmentgrowth, natural amenities, and percentage of older people are key factors underlying these migra-tion patterns. We also find a slight fading of effects of natural amenities and population density,and a slight increase in the effects of wage and employment growth from 2005–2009.

Key words: Metro to nonmetro migration, urban to rural migration, domestic migration, county-to-county migration, USA counties, natural amenities, industry-mix employment growth, retireemigration, Poisson regression.

JEL Codes: J11, J61, R11.

Agglomeration economies are attractingpeople from rural to urban settings, with nowmore than 50% of the world’s populationresiding in urban areas, and expectationsthat the urban share will rise to 70% by 2050(China Development Research Founda-tion 2010). The historic direction of internalU.S. migration was also rural-to-urban, ornonmetro-to-metro. However, the prevailingnonmetro-to-metro trend reversed duringthe 1970s, and mostly held thereafter; basedon the USDA metropolitan classification, the2000 Census data show that between 1995

Anil Rupasingha is economist at the Economic ResearchService, U.S. Department of Agriculture, Washington, DC.Yongzheng Liu is an assistant professor in the School ofFinance at Renmin University of China. Mark Partridge isthe Swank Chair in Rural-Urban Policy in the Department ofAgricultural, Environmental, and Development Economics atthe Ohio State University. Correspondence may be sent to:[email protected].

The views expressed here are the authors’ and not nec-essarily those of the Economic Research Service, the U.S.Department of Agriculture, Renmin University of China, orthe Ohio State University.

and 2000, about 220,000 more people movedto nonmetropolitan areas from metropolitanareas than the reverse. Recently releasedAmerican Community Survey (ACS) datashow that between 2005 and 2009, net domes-tic migration to nonmetropolitan areas inrelation to metropolitan areas totaled about100,000 annually.1 Yet these patterns areunevenly distributed. About one-half ofnonmetro counties lost population between2005–2009, and 57% lost population over the1995–2000 period.

Understanding the causes of relativelyfavorable U.S. nonmetropolitan net-migra-tion patterns and their changes over time

1 A word of caution when comparing the data for the two timeperiods: While 1995–1999 U.S. Census data are 5-year aggregates,the ACS 5-year estimates are not five years of aggregated data.Rather, they are a 5-year period estimate from 2005–2009 usingannual data (see Benetsky and Koerber 2012). The general patternof recent positive net-migration to nonmetropolitan areas didreverse itself in 2011 and 2012 (Cromartie 2013).

Amer. J. Agr. Econ. 97(3): 680–700; doi: 10.1093/ajae/aau113Published online January 20, 2015

© The Author 2015. Published by Oxford University Press on behalf of the Agricultural and Applied EconomicsAssociation. All rights reserved. For permissions, please e-mail: [email protected]

Downloaded from https://academic.oup.com/ajae/article-abstract/97/3/680/67201/Rural-Bound-Determinants-of-Metro-to-Non-Metroby University Libraries | Virginia Tech useron 05 September 2017

Rupasingha, Liu, and Partridge Determinants of Metro to Non-Metro Migration 681

is important for assessing whether the gen-eral urbanization trend will slow as incomesincrease, and for crafting better regionaldevelopment policies aimed at reducingregional inequities. For example, if U.S. ruralareas have primarily benefited from com-modity booms or it is mainly high-naturalamenity areas that are gaining population, itwill be harder to develop effective policiesaimed at improving rural economic prospects.Supporting the possibility that U.S. ruralareas can remain competitive in terms ofmigration, Partridge et al. (2010b) find thatwhile firms increasingly prefer to locate nearagglomeration economies, households preferto be more distant from urban areas.

Several conceptual frameworks have beenadvanced to explain the reversal that tookplace in 1970s and 1990s. The main expla-nations for the reversal in the 1970s were“period effects,” the “regional restructuringperspective,” and “suburbanization.” Theperiod effects follow the unique circum-stances of the 1970s, such as the 1973–74 oilcrisis and subsequent recession (Frey andJohnson 1996). The regional restructuringperspective is based on the structural changesthat led to the transformation of the urbaneconomy from traditional heavy industriesto the service economy (Frey and Johnson1996) and a boom in extractive and manufac-turing activities in nonmetro areas (Fuguittand Beale 1996). Kim (1983) contends the1970s reversal was due to expanding subur-ban development and increased retirementmigration to rural areas.

A primary conceptual framework for thereversal in the 1990s is deconcentration,which is attributed mainly to movement ofpeople and firms to low-density and high-amenity locations. A further perspectiveis regional restructuring, which refers tochanges in economic opportunities (Freyand Johnson 1996; Partridge et al. 2010b).Although traditionally agglomeration econo-mies have been positively associated within-migration, the deconcentration hypoth-esis is associated with the following: retireemigration to nonmetro locations (Frey andJohnson 1996; Nelson and Nelson 2011);migration to metro-adjacent locations,which then requires an out-commute to work(Cromartie 1998; Partridge, Ali, and Olfert2010a); rural gentrification, which is tied toeconomic restructuring due to advances intelecommunications (Nelson, Oberg, andNelson 2010); and amenity-based migration

(Rudzitis 1999; Nelson and Nelson 2011;Kahsai, Gebremedhin, and Schaeffer 2011).Yet, other studies stress structural economicchanges as being partly responsible for thereversal (Frey and Johnson 1996; Ghatak,Levine, and Price 1996).

An unexplored aspect of the reversal lit-erature is the systematic integration andanalysis of the deconcentration and regionalrestructuring perspectives. There is also aneed to know how sensitive these key argu-ments are to different time periods, and howdifferently these forces describe metro-to-nonmetro migration and nonmetro-to-metromigration. By comparing the set of deter-minants for these types of migration, onecould answer the question of whether metro-to-nonmetro migrants consider differentfactors than their nonmetro-to-metro coun-terparts. Therefore, the present study extendsthe literature by more broadly examin-ing the underlying factors associated withdeconcentration and economic restructuringarguments of metro-to-nonmetro migration.Employing a more unified framework, thisarticle determines the extent of the effectof these arguments on metro-to-nonmetromigration. It also tests whether the recentattraction of nonmetro counties is derivedfrom more economic factors such as wages,industry mix, and proximity to urban areas,or other reasons such as natural amenitiesand retirees relocating. Thus, in order toform better policy, this analysis enhancesour understanding of metro-to-nonmetromigration while addressing many lingeringquestions in the literature.

Moreover, we investigate whether theeffects of these factors differ over time andwhether these determinants vary betweenmetro-to-nonmetro and nonmetro-to-metroflows. For example, the 1990s were charac-terized by a strong national economy (withweak rural commodity markets) versus thesluggish national economic environmentpost-2000 (with strong rural commoditymarkets), in which gross migration flowsgreatly diminished (Partridge et al. 2012).Considering net migration, Partridge et al.(2012) found that economic migration ingeneral greatly declined after 2000, butthey did not consider urban-rural migrationpatterns. Likewise, population growth andnet migration studies find that the effectof natural amenities may be diminishing innonmetropolitan counties (Rickman andRickman 2011; Partridge et al. 2012), but


682 April 2015 Amer. J. Agr. Econ.

these studies do not specifically considermetro-to-nonmetro flows. Moreover, retireemigration may have strengthened in thesecond time period as more and more babyboomers are attaining retirement age.

Comparing both metro-to-nonmetro andnonmetro-to-metro migration flows usingthe same set of factors is vital from a policyperspective because such an analysis shedslight on factors associated with in- and out-migration in metro and nonmetro localities,as well as whether these factors are impor-tant in light of recent migration patterns.It is also essential to know for policymak-ing whether the role of the determinantsvary depending on whether the nonmetrocounties are located near metropolitan areasor whether they are located remotely. Forexample, Partridge et al. (2008) and Wu andGopinath (2008) study the effects of prox-imity to urban areas on population growthin rural U.S. counties and find that there arestrong negative growth effects of distances tohigher-tiered urban areas. Yet while the eco-nomic gains in nonmetro counties are visiblenear metro counties, many nonadjacent, ruralcounties have seen increases.

This study further strengthens the liter-ature by utilizing econometric modelingthat uses gross county-to-county migrationflows to estimate a spatial interaction model(Greenwood 1985; Cushing and Poot 2004;Etzo 2010). The current approach allows usto employ the utility maximization theory ofmigration using the discrete-choice frame-work based on random utility maximization(Davies, Greenwood, and Li 2001), whiletaking advantage of an equivalent relationbetween the conditional logit model (CLM)and Poisson regression (Cushing and Poot2004; Arzaghi and Rupasingha 2013). Weutilize aggregate county-to-county migrationflows from the U.S. Census Bureau for 1995–2000 and 2005–2009 and consider a metrohousehold’s opportunity to move to all possi-ble non-metro counties (and vice versa). Thisyields a very large number of observationsfor each time period (more than two million),which allows for high statistical power whilemitigating endogeneity problems. Likewise,the related literature reviews emphasizethe importance of distance between ori-gin and destination in migration decisions(Greenwood et al. 1991; Cushing and Poot2004). Yet distance is typically not con-sidered when using population growth ornet migration. As a result of incorporating

origin to destination choices, the presentstudy will be able to assess the effect of dis-tance. Another advantage of our empiricalapproach is that we can directly considerwhether distance effects wane over timewith newer information technologies, andhow distance interacts with job opportu-nities and amenities in shaping migration.Hence, our approach has both theoretical andmethodological extensions.

Our findings support both the deconcen-tration and regional restructuring hypotheses,but the effects of some deconcentration mea-sures may have diminished over time. Whilenatural amenities are a very strong predictorin both metro-to-nonmetro and nonmetro-to-metro migration flows, the effect in nonmetrocounties may have also diminished over timecompared to metro counties. Distance is adeterrent for both metro-to-nonmetro andnonmetro-to-metro flows, with little changeover time. Amenities also become relativelymore important for long distance moves.The results suggest that migrants movingfrom metro-to-nonmetro areas are morelikely to settle in densely-populated non-metro counties than less dense nonmetrolocations, suggesting some minimal thresholdsize effects for nonmetro areas. Yet the effectis diminishing over time in nonmetro areas,implying that some agglomeration constraintsmay be overcome. Results also lend supportto the suburbanization hypothesis that moremigrants are attracted to rural counties thatare adjacent to metro areas.

We next present an overview of the con-ceptual model and econometric approach,followed by a description of the data, and ourestimations. The results are then presented,followed by a conclusion and a discussion ofthe rural development implications.

The Model and Econometric Approach

Our conceptual approaches follow Goetz(1999): individuals maximize utility (Ui),which is defined by place characteristics,i = 0, 1, 2, . . . , p, where utility can be affectedby numerous factors such as real income(y) and amenities (a); prospective migrantsevaluate the expected utility of residing indifferent places over a given planning hori-zon; more specifically, they compare theutility derived at their current location (U0)with the utility that can be derived from



other locations, net of costs of migrating (ci),between 0 and i. Based on all the informa-tion available to migrants, they are able torank any two locations, using the locations’attributes. This process can be depicted as:

(1) �i = U(yi − ci, ai) − U(y0, a0).

Utility-maximizing individuals will migratewhenever �i > 0; otherwise they stay in place.

Basing our model on location charac-teristics and costs allows us to utilize therandom utility approach developed byMcFadden (1974). This approach is widelyused in the empirical industrial organiza-tion literature on firms’ location decisions(Guimarães, Figueirdo, and Woodward2003; Guimarães, Figueirdo, and Woodward2000; Arauzo-Carod, Liviano-Solis, andMonjón-Antolín 2010), but it is relativelynew to the migration literature (Davies,Greenwood, and Li 2001; O’Keefe 2004;Christiadi and Cushing 2008; Arzaghi andRupasingha 2013). The random utility modelleads to the application of the CLM for var-ious destination choices. Based on equation(1), a potential migrant will choose a par-ticular location if expected (net) utility inthat location is greater than utility in thecurrent and other potential locations. For-mally, consider a resident at metro countyi aiming to relocate to non-metro countyj, where i = 1, . . . , 1090 (total number ofmetro counties in the 1995–2000 sample) andj = 1, . . . , 2052 (total number of non-metrocounties in 1995–2000 sample). As usual,for any alternative county choice, the util-ity derived for this individual (Uij) can bewritten as:

(2) Uij = β′Xij + εij

where Xij is a vector of the choice-specificattributes, including those of the current loca-tion and migration-related costs (Arzhagiand Rupasingha 2013); εij is a random errorterm. Thus, the utility for an individual forlocating at j is composed of a deterministicand a stochastic component. Following utilitymaximization, the individual will choose thelocation that yields the highest utility (i.e.,Uij > Uik, for all k �= j). So the probabilitythat an individual in county relocates tocounty j is

(3) Pij = P(Uij > Uik) for all k �= j.

McFadden (1974) shows that if εij are inde-pendent and identically distributed andextreme value type-I distributed,2 then theprobability Pij can be rewritten as

(4) Pij = exp(β′Xij)∑2052

j=1 exp(β′Xij).

Equation (4) expresses the familiar CLMformulation. With independent observations,the corresponding log likelihood function forall the individuals moving from any metrocounty i to a specific non-metro county j is

(5) log L =∑

i

dij log Pij = nij log Pij

where dij = 1 if a resident in metro county ichooses to reside in non-metro county j, andzero otherwise; nij is the number of individu-als moving from metro county i to non-metrocounty j.

Since residents in all 1,090 metro coun-ties can possibly migrate to any of the 2,052non-metro counties, the coefficients β′ canbe estimated by maximizing the followinglog-likelihood function:

(6) log Lcl =1090∑

i=1

2052∑

j=1

nij log Pij.

It is well-recognized in the CLM literaturethat estimating equation (6) is cumbersomeand even infeasible for a large number ofalternative choices.3 An alternative proposedby Guimarães, Figueiredo, and Woodward(2003) is to estimate the CLM using an“equivalent” standard Poisson regressionmodel (PRM). These authors prove thatunder certain conditions, the log-likelihoodfunctions of the conditional logit and the

2 This assumption implies the Independence of Irrelevant Alter-natives (IIA) property, which requires that, for any household, theratio of choice probabilities of any two alternatives is independentof the utility of any other alternative.

3 However, the ability to include a large number of spatialalternatives is important because factors usually identified as beingthe most relevant for location decisions are at a small geographicallevel, which cannot be adequately captured by large areas in thespatial choice sets (Gabe and Bell 2004; Guimarães, Figueirdo, andWoodward 2003). Following a suggestion by McFadden (1978),one solution is to estimate the model using a randomly selectedsub-sample (Friedman, et al. 1992; Guimarães, Figueirdo, andWoodward 2000; Hansen 1987; Woodward 1992). Although theresulting estimators are consistent, the efficiency is reduced due todropping some information, and also the small sample propertiesare unknown (Guimarães, Figueirdo, and Woodward 2003).



Poisson regression are identical, whichin practice implies that the coefficients ofequation (6) can be equivalently estimatedby estimating a standard PRM with tak-ing nij as a dependent variable, and Xij asexplanatory variables.4 To see the equiva-lence more clearly, let nij be independentlyPoisson-distributed with conditional mean

(7) E(nij) = μij = exp(α + β′Xij).

Then, the standard log-likelihood function ofthe PRM can be written as

(8)

log Lp =1090∑

i=1

2052∑

j=1

(−μij + nij log μij − log nij!).

As shown in Guimarães et al. (2003), aftertaking the first-order condition of equation(8) with respect to α and inserting back thederived expression of α to equation (8), theconcentrated log likelihood function canbe simplified to

log Lp =1090∑

i=1

2052∑

j=1

nij log Pij − N(9)

+ N log N −1090∑

i=1

2052∑

j=1

log nij!

where the first term in expression (9) is thelog likelihood of the CLM, and the otherthree terms are constants.

The most important advantage of thePRM over the CLM is its superior compu-tational ability in handling a large numberof spatial alternatives that comprise an indi-vidual’s choice. Beyond this, the PRM is alsoeffective in controlling for the violation ofthe Independent of Irrelevant Alternativesassumption, which is especially problem-atic for the CLM when a large number ofnarrowly-defined spatial alternatives areinvolved in the decision making (Guimarães,Figueiredo, and Woodward 2004).

4 See, for example, Arauzo-Carod and Antolín (2004), ArauzoCarod (2005), and Gabe and Bell (2004) for applying the Poissonapproach as a substitute to estimate the CLM.

Variables, Data, and Estimation Issues

We separately examine the 1995–2000 and2005–2009 periods to assess how migrationpatterns changed over the two decades. Theadvantage of these two periods is that theyare about 10 years apart and both occur ina general positive net migration period fornonmetro areas, and overall gross migrationflows were relatively constant during eachrespective period.5 One concern is that thelatter period includes the housing crisis andsubsequent Great Recession. However, aspointed out in footnote 5, the housing crisisand Great Recession had remarkably lit-tle effect on overall gross migration flows,which remain at historically low levels post2000. Likewise, the Great Recession was theculmination of slow job growth in the post2000 period. Indeed, U.S. Bureau of LaborStatistics data suggests that the pre-GreatRecession period 2000–2007 were the sevenslowest years of nonfarm job growth of anyseven-year period dating back to the late1930s (or at least until the Great Recessionbegan). Yet as described below, we will takespecial care to avoid having the housingcrisis (and Great Recession) confound ourresults by accounting for factors such as con-trolling for nonmetro counties adjacent toa metropolitan area to control for the pat-tern that the housing crisis was more severein far-suburban and exurban locations, andwe also account for lagged median hous-ing prices to control for places with greaterthan expected market prices.6 Likewise, weaccount for underlying industrial demandshocks to address differential demand effectsfrom the Great Recession and housing crisis.

5 Cromartie (2013) shows that net migration rates for non-metropolitan areas was positive during both sample time periods,though nonmetropolitan net migration rates did turn negative in2011 and 2012. U.S. Census Bureau (2014) reports that overallgross-migration rates across state and/or county borders, respec-tively, equaled 5.6% and 6.4% between 1995–96 and 1999–00.The corresponding figures for 2005–06 and 2008–09 were 4.7%and 3.7%. The Great Recession seemed to have relatively littleinfluence on these gross migration flows, as gross migration flowsacross state/county borders, respectively, still only totaled 3.9%and 3.8% in 2011–12 and 2012–13, suggesting that in terms ofoverall migration patterns, using periods after the Great Recessionmay not yield very different patterns.

6 We do not expect the housing crisis to have tangibly affectedaggregate migration flows of given metropolitan areas, which iswhat we are primarily interested in. Yet we expect that intra-metropolitan area patterns were affected, as more distant suburbswere particularly hard hit, but this was offset as central areasfared relatively better. For the metropolitan area as a whole, thenet migration rates would not be measurably affected.



We separately appraise nonmetro-to-metromigration and metro-to-nonmetro migrationto assess differing causes for their respectivepatterns. Namely, heterogeneity implies thatpeople who move one direction (say to anurban area from a rural area) would likelyhave very different preferences and abilitiesthan those who migrate the other way. Forone, there is some tendency for higher-abilitypeople to sort into metro areas, implyingthat returns to agglomeration differ acrosspeople (Combes et al. 2012). Likewise, weexpect that nonmetro-to-metro migrants mayrelatively value urban amenities associatedwith population density, whereas metro-to-nonmetro migrants may place a higherweight on other rural features of quality oflife. This sorting and preference heterogene-ity implies that not only do metropolitan andnonmetropolitan characteristics (X) vary,but likely so do the underlying regressioncoefficients.

As discussed in the introduction, themigration reversal of the 1990s is explainedmainly by the deconcentration and eco-nomic restructuring perspectives. Althoughnot mutually exclusive, the deconcentrationargument may be manifested in amenity-based or quality-of-life migration, retireemigration, and preference for lower densityor low agglomeration locations, while theeconomic restructuring argument may beexpressed in industry structure, jobs, andwage related migration. The studies discussedabove strongly support the amenity-basedor quality-of-life argument in which somemetro-based workers choose to forego higherearnings in exchange for the quality of lifefound in nonmetro localities. This qualityof life is mainly attributed to natural andman-made amenities (Knapp and Graves1989; Mueser and Graves 1995; Deller et al.2001). We use the natural amenity index(amnscale) developed by McGranahan (1999)and hypothesize it to be positively associatedwith in-migration.

Another argument is that metro areas havesimply expanded or people moved to metro-adjacent counties. Partridge et al. (2008) findthat the proximity to larger metropolitanareas has been an important driving forcefor rural population gains since at least 1950.Following Wu and Gopinath (2008), weinclude an indicator variable (metroadj) inour full-sample metro-to-nonmetro modelfor counties that are adjacent to metro areas,and expect it to be positively associated

with nonmetro in-migration. Likewise, ifmetropolitan migration to adjacent nonmetrocounties is in large part driven by commutingback to the metro area, we would expectlocal labor market conditions to matter lessto possible migrants than local rural labormarket conditions in remote nonmetro coun-ties. Yet this does not mean that adjacentcounty in-migrants do not care about localeconomic conditions, simply because not allof them are going to commute to the metroarea. Moreover, even for commuters, theirspouses may want to work locally or theymay want the option of being able to worklocally in the future.

Gravity models of migration use pop-ulation density (popden) and distance asstandard pull factors. We use populationdensity per square mile as both a measureof agglomeration and as an attraction forcein gravity model formulations, which alsodirectly relates to the deconcentration per-spective. Traditionally, population densityis found to be positively associated with in-migration, but in case of metro to nonmetromigration, the deconcentration hypothesisargues that migrants may prefer low densityor low agglomeration locations.

Migration is costly for financial, informa-tion, and personal reasons. Migration costsalso rise as the moving distance increases.The deconcentration perspective suggeststhat one motive for metro residents tomove to suburbs is to live in more openlandscape and then commute to work inmetro locations. Distance plays a key rolein this kind of movement as it becomes aprimary deterrent (Wu and Gopinath 2008;Partridge, Ali, and Olfert 2010a). Our dis-tance variable is calculated using the distancebetween each pair of county centroids viahighway (divided by 100). Following Davies,Greenwood, and Li (2001), we conjecturethat the deterring effects of distance maydecline as distance increases, and we includea distance-squared term to capture thesenonlinearities, which is expected to have apositive coefficient.

As discussed above, retirement-basedmigration to nonmetro counties is saidto be growing as increasing numbers ofbaby boomers reach retirement age. Whileour migration data do not identify retireemigrants from other migrants, we test thegeneral hypothesis from past studies that onereason that nonmetro counties gain migrantsmay be due to their retiree attraction. We



include the percentage of the populationthat is 65 and over (elder) for each base yearand expect a positive relationship (Jensenand Deller 2007; Rayer and Brown 2001),postulating that retiree migrants may beself-sorting to nonmetro counties that havea higher percentage of their age group (per-haps reflecting better public and privateservices for retirees).

Though the economic factors are down-played as a pull factor to rural areas (focusingmore on rural quality of life), some studiesstress structural economic changes as partlyresponsible for the reversal. Indeed, thecommodity booms of the 1970s (Ghatak,Levine, and Price 1996) and post 2000 periodhave bolstered certain rural economies. Totest the validity of these claims, we incorpo-rate the average county wage (wage) andindustry-mix employment growth rate (indus-mix). Empirical results on the relationshipbetween per capita income or earnings andin-migration have been mixed; some studiesfound a positive link (Davies, Greenwood,and Li 2001). Consistent with a spatial equi-librium view, Markusen and Schrock (2006)find that migrants will accept lower wages tolive in locales with higher amenities—thatis, relative wage levels are a compensatingdifferential.7 Due to these offsetting effects,the expected effects of the wage variable areambiguous.

As a sign of employment availability, forboth periods we use the 1990–2000 and 2000–2007 industry-mix employment growth ratefrom shift-share analysis, which is routinelyused as an exogenous instrument for jobgrowth by previous studies (Bartik 1991;Blanchard and Katz 1992; Partridge et al.2012) as an exogenous measure of local

7 Even though wages are lagged five years before the initialperiod of the dependent variable, there is a chance they hadbegun to adjust in anticipation of future migration behavior.However, as discussed below, it is unlikely that this would tangiblyoccur because the dependent variable is migration for county-to-county pairs and it is doubtful that wages tangibly adjust fromone of the county-to-county pairs when each county is pairedwith over 3,000 county pairs. In addition, our regression modelsinclude origin-county fixed effects, which removes any omittedbias due to time-invariant omitted variables in the origin county.We experimented with using a 15-year lag of wages to furthermitigate any fear of endogeneity, but the results were essentiallyunaffected suggesting endogeneity is not a major concern. We alsoreplaced the 15-year lag of wages with 15-year lagged per capitapersonal income because per capita personal income should beless affected by endogeneity (not shown), but the general patternof results were also unaffected.

demand conditions.8 The industry mix vari-able is the “share” variable from shift-shareanalysis, which captures the fact that, nation-ally, some industries grow faster or slowerthan others, and these structural differencesaffect local labor markets through their dif-ferential industry composition. This indexis constructed by summing the products ofthe initial industry shares in the county atthe beginning of each time period (1990 and2000) at the four-digit level with the corre-sponding national U.S. industry growth rates,thus producing an exogenous measure oflocal labor demand shocks.9

We also include several other county char-acteristics that past research has shown tobe associated with U.S. domestic migration.These factors include economic variablessuch as median housing value (mhv) andvolatility of local economies (cvurate), aswell as government policy variables suchas per capita taxes (pctax) and governmentexpenditures (pcgexp). Higher housingprices may discourage in-migrants, thoughthey also may reflect unmeasured amenities(Jeanty, Partridge, and Irwin 2010; Murphy,Muellbauer, and Cameron 2006), and theycould reflect housing market conditions.Likewise, following the Roback (1982) spatialequilibrium model, populous metropolitanareas that lack large-scale natural amenitiesand have weak zoning would have lowerhousing prices (e.g., Dallas or Atlanta), illus-trating how housing prices differ from thepopulation density measure.

Several studies have considered whethermigration is also associated with riskand uncertainty (Daveri and Faini 1999;Rosenzweig and Stark 1989; Stark andLevhari 1982; Arzhagi and Rupasingha 2013).We incorporate the volatility of unemploy-ment in the destination county to control forrisk. Specifically, we use the coefficient ofvariation of the unemployment rate between1990 and 1999, and 2000 and 2009 for eachdestination county as a measure of risk and

8 A direct incorporation of unemployment rate in the empiricalmodel, for example, can be problematic due to endogeneity of theunemployment rate, which may be simultaneously determinedwith migration (Etzo 2010).

9 The result is the predicted growth rate if all of the county’sindustries grew at the national growth rate. The level of detailused in the calculation is the four-digit level using proprietaryemployment data from the EMSI consulting company. Note thatwe end the 2000–2007 industry mix variable before the GreatRecession in order to not mix the cyclical effects of the recessionwith the persistent growth effects reflected in the 2000–2007period beforehand.



hypothesize it to be inversely associatedwith in-migration. The Tiebout hypothesis(Tiebout 1956) suggests that if mobility iscostless, individuals will “vote with theirfeet” by moving to a locality that providesthe optimal mix of local public goods. Thus,we include per capita local taxes and percapita local government expenditures, whichinclude intergovernmental transfers. Taxesare hypothesized to be negatively associ-ated with in-migration, while governmentexpenditures are positively associated within-migration.

This article utilizes county-to-county migra-tion data from the 2000 decennial census forthe population aged 5 and over for the periodbetween 1995 and 2000, and the same datafrom the American Community Survey(ACS) for 2005 to 2009. A key differencebetween the two surveys is in how pastmigration is defined. The 2000 Census askedwhere a resident lived five years prior, whilethe ACS asked where a resident lived oneyear prior. Therefore, the 2000 Census datainclude those who moved over the previous5-year time span and the ACS data includesonly people who moved during the previ-ous year. Based on this, even though the2005–2009 ACS is a 5-year dataset, it is a5-year estimate using 1-year datasets. Docu-mentation is available at the Census Bureauwebsite on the compatibility of the databetween the 2000 Census and the 2005–2009ACS (Benetsky and Koerber 2012).

Benetsky and Koerber (2012) analyze therelationship between ACS and 2000 Censusmigration data and show that the flows inthe 2005–2009 ACS are highly correlatedwith the 2000 Census flows, with a Pearson’sr of about 0.94. They also regress 2005–2009ACS flows on the 2000 Census flows andfind that the ACS flows account for about89.0% of the flows in the 2000 Census. Basedon these findings, these authors concludethat the ACS flow data is a good estimate ofmigration relative to the 2000 Census data,declaring, “[d]espite comparing two differ-ent surveys utilizing two different migrationquestions, there is congruence in the rela-tive magnitude of county-to-county moversfound between the surveys,” (Benetsky andKoerber 2012). The main way our resultswould be tangibly affected is if, conditionalon our control variables for demographicsand economic conditions, in-migration ratesare systematically different across countiesfor one-year and five-year flows beyond a

simple scaling effect where the sum of theone-year flows may be larger than the five-year flows (which would simply change thescaling of the regression coefficients). Themigration data is cross-sectional, providing ann by n matrix of internal migration flows forall U.S. counties.

A major concern with migration modelsis possible endogeneity, in which the errorterm may be correlated with some of theexplanatory variables. A key cause of suchendogeneity is that labor demand-shift vari-ables are jointly determined with migration.Our use of the industry mix variable as anexogenous proxy for demand shifts greatlymitigates this concern. We also use five-year lagged period values (1990 and 2000)for explanatory variables as in the “weaklyexogenous” regressors assumed by Levine,Loayza, and Beck (2000). This approachimplies that future migration does not affectcurrent levels of explanatory variables. Tofurther account for omitted variable bias,we include origin fixed effects. Finally, theresearch design is less exposed to endogene-ity concerns than in standard models of netmigration for an individual county. Specif-ically, in the individual county models usedin most of the literature (e.g., populationgrowth or net migration), job growth andnet migration are jointly determined. How-ever, when estimating migration betweenmore than 3,000 counties’ pairs, the relativeshare of total migration for a county thatis explained by one county pair is typicallyquite small, meaning that shifts in migrationbetween a single county pair would have amuch smaller influence on a county’s overalleconomic activity—reducing the severity ofany endogeneity. We discuss other ways wemitigate endogeneity below.

Estimation and Results

We estimate a PRM by taking advantage ofthe equivalence relation between the log-likelihood functions of the CLM and thePRM (Guimaraes, Figueirdo, and Woodward2003). To ensure compatibility betweenthe conditional logit and Poisson models,it is necessary that we incorporate locationfixed effects in our empirical application(Guimaraes, Figueirdo, and Woodward 2004).Ideally, both origin and destination countyfixed effects must be incorporated, but due



to limited computational power, we only useorigin-county fixed effects. To assess whetherclose proximity to metropolitan areas leadsto different results, in the metro to nonmetromodel, we also separate nonmetro countiesinto two sub-samples: nonmetro-adjacent(rural urban continuum codes 4, 6, and 8)and nonmetro-nonadjacent (rural urbancontinuum codes 5, 7, and 9).

The descriptive statistics are summari-zed in table 1 for both metro to nonmetroflows and nonmetro to metro flows for bothtime periods. The estimation procedure formetro to nonmetro flows employs a fullsample model (about 2.2 million metro tononmetro county-to-county flows for eachtime period) and the two subsamples formetro flows to nonmetro-adjacent (around1.13 million county-to-county flows, deno-ted as subsample 1) and metro flows tononmetro-nonadjacent (around 1.04 millioncounty-to-county flows, denoted as sub-sample 2). Then we estimate the model formigration from all nonmetro to all metrocounties. In all cases, we report hetero-skedasticity-robust standard errors. This isparticularly important for Poisson regres-sion, because while we expect that thecoefficients of the Poisson model mainlyremain consistent in the presence of over-dispersion, the standard errors may be hea-vily underestimated. We also include loglikelihood values of each specification inorder to show the appropriateness of eachspecification, as well as the results of aWald test to indicate the suitability of thefixed-effects Poisson models.

Table 2 presents fixed effect Poisson esti-mation results for metro-to-nonmetro countymigration for the full sample, two sub sam-ples, and nonmetro-to-metro migration forboth time periods.10 As suggested in pre-vious studies, our results show that naturalamenities are a strong predictor in metro-to-nonmetro migration in both time-periods.The coefficients for all samples in bothperiods are highly significant and positive.

10 When comparing the results between two time periods, wecaution the reader that, despite assurances given in Benetsky andKoerber (2012), there might be differences between one migrationmeasure and the other associated with the variables in the analysis.One reviewer pointed out that bias may be introduced from thedifferences in the way the migration question was posed. Forinstance, young people may make several moves over a 5-yearperiod as they search for jobs they like, making yearly ratesmore responsive to area employment opportunities than 5-yearrates. People may move more often in short-distance moves thanlong-distance moves.

All else being constant, for one standarddeviation increase in the natural amenityindex, a nonmetro county’s in-migrationincreases by 23% for the full sample from1995–2000.11 However, the effect seems tohave weakened in the second period in allsamples. The value of the amenity coefficientfor the full sample decreased from 0.091in the first period to 0.065 in the second.The respective figures for the nonmetro-adjacent are 0.075 and 0.025 and for thenonmetro-nonadjacent they are 0.100 and0.088. Accordingly, for a one standard devi-ation increase in the natural amenity index,a nonmetro county’s in-migration increasesby 16% for the full sample in the 2005–2009 period. This finding suggests that eventhough natural amenities are still a key deter-minant of metro-to-nonmetro migration,its overall effect on this migration direc-tion may have diminished, which supportsthe findings of Partridge et al. (2012). Thepositive and significant coefficient for thenatural amenity variable in the nonmetro-to-metro model suggests that natural amenityis a strong factor in rural-to-urban migra-tion, and the results for 2005–2009 showno notable temporal change in the over-all effect of natural amenities. For a onestandard deviation increase in the naturalamenity index, a metro county’s in-migrationfrom nonmetro counties increases by 59%and 58%, respectively, for the 1995–2000and 2005–2009 models. For example, for agiven average metropolitan-nonmetropolitancounty pair between 2005–2009, table 1 showsthat on average, about 0.86 migrants movedfrom the nonmetropolitan county to themetropolitan county. Thus, a one standarddeviation increase in amenities would leadto an increase of 58% in migration, or therewill be an additional one-half of a migrantmoving to the metropolitan county fromeach nonmetropolitan county annually overthe five-year period (or by about 1,000 newmigrants in total from the approximately2,000 possible nonmetropolitan counties).

The non-metro adjacent to a metropoli-tan area coefficient is highly significant andpositive in both time periods, supporting thesuburbanization hypothesis. The size of itsestimated parameter increased in the secondtime-period, indicating that the influence of

11 This is calculated using 100 ∗ ((eβk∗δ) − 1), where δ indicatesstandard deviation or a factor change in the covariate (see Long1997).


Rupasingha,L

iu,andP

artridgeD

eterminants

ofM

etroto

Non-M

etroM

igration689

Table 1. Variable Description and Descriptive Statistics: Metro to Nonmetro vs. Nonmetro to Metro Migration Full Samples

Metro to Nonmetro Migration Nonmetro to Metro Migration

1995–2000 2005–2009 1995–2000 2005–2009

Variable Description Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.

inflows(dependent)

Migration from metro to nonmetro(or nonmetro to metro) counties

2.45 32.95 0.90 13.82 2.35 28.45 0.86 13.15

amnscale Natural amenities index −0.05 2.25 −0.05 2.25 0.25 2.32 0.25 2.32metroadj Nonmetro counties that are

adjacent to a metro area (0,1)0.52 0.50 0.52 0.50

distance/100 Actual distance between an origincounty and destination county onaverage, in miles

9.10 5.96 9.10 5.96 9.10 5.96 9.10 5.96

distance_sq Distance squared 118.41 181.64 118.41 181.64 118.41 181.64 118.41 181.64popden/1000 Population per square mile 0.04 0.11 0.05 0.11 0.56 2.40 0.62 2.79wage Wage and salary per job 16.47 3.26 23.20 3.80 19.67 4.01 28.58 6.63indusmix Industry mix employment growth,

calculated by multiplying eachsum across all industries of theproduct of the industry’s nationalemployment growth (1990 - 2000,and 2000–07) with the initialperiod (1990 2000) industryemploy share in the county

13.76 4.89 9.12 2.51 16.43 5.12 10.85 2.72

elder Percentage of population over 64years old

16.24 4.12 16.07 3.91 12.54 3.64 12.48 3.40

pctax/1000 Per capita local taxes 0.65 0.49 0.96 0.81 0.68 0.41 1.03 0.60pcgexp/1000 Per capita local government

expenditure1.90 0.86 2.96 1.22 1.81 0.68 2.87 1.23

mhv/1000 Median housing value 43.56 20.36 70.55 36.16 72.70 42.63 109.41 54.90cvurate Coefficient of variation of

unemployment rate0.21 0.09 0.24 0.09 0.23 0.08 0.29 0.08

Obs. 2,103,245 2,109,260 2,128,421 2,089,500


690A

pril2015

Am

er.J.A

gr.E

con.

Table 2. Fixed Effect Poisson Estimation Results

Nonmetro to MetroMetro to Nonmetro Migration Migration

1995–2000 2005–2009 1995–2000 2005–2009

Full Metro to Metro to Full Metro to Metro to Full FullSample adjacent nonadjacent Sample adjacent nonadjacent Sample Sample

amnscale 0.091∗∗∗ 0.075∗∗∗ 0.100∗∗∗ 0.065∗∗∗ 0.025∗∗∗ 0.088∗∗∗ 0.200∗∗∗ 0.198∗∗∗(0.008) (0.009) (0.007) (0.008) (0.009) (0.009) (0.008) (0.011)

distance −0.819∗∗∗ −0.895∗∗∗ −0.656∗∗∗ −0.896∗∗∗ −0.966∗∗∗ −0.729∗∗∗ −0.752∗∗∗ −0.808∗∗∗(0.044) (0.042) (0.059) (0.050) (0.050) (0.058) (0.081) (0.093)

distance_sq 0.026∗∗∗ 0.029∗∗∗ 0.019∗∗∗ 0.029∗∗∗ 0.031∗∗∗ 0.022∗∗∗ 0.022∗∗∗ 0.024∗∗∗(0.002) (0.002) (0.003) (0.002) (0.002) (0.003) (0.005) (0.006)

popden 7.796∗∗∗ 7.464∗∗∗ 9.207∗∗∗ 0.345∗∗∗ 0.193∗∗∗ 4.763∗∗∗ 0.020∗∗∗ 0.022∗∗∗(0.306) (0.352) (0.330) (0.022) (0.025) (0.193) (0.005) (0.005)

wage 0.006∗ −0.009∗∗ 0.050∗∗∗ 0.044∗∗∗ 0.029∗∗∗ 0.067∗∗∗ 0.099∗∗∗ 0.055∗∗∗(0.003) (0.004) (0.004) (0.002) (0.003) (0.004) (0.002) (0.002)

indusmix 0.035∗∗∗ 0.029∗∗∗ 0.041∗∗∗ 0.070∗∗∗ 0.058∗∗∗ 0.065∗∗∗ 0.071∗∗∗ 0.145∗∗∗(0.002) (0.003) (0.002) (0.004) (0.005) (0.006) (0.002) (0.006)

elder −0.050∗∗∗ −0.033∗∗∗ −0.068∗∗∗ −0.082∗∗∗ −0.066∗∗∗ −0.102∗∗∗ −0.064∗∗∗ −0.058∗∗∗(0.004) (0.004) (0.004) (0.005) (0.006) (0.005) (0.003) (0.005)

pctax −0.244∗∗∗ −0.198∗∗∗ −0.332∗∗∗ −0.312∗∗∗ −0.210∗∗∗ −0.550∗∗∗ −0.143∗∗∗ −0.347∗∗∗(0.036) (0.051) (0.036) (0.031) (0.036) (0.043) (0.038) (0.036)

pcgexp −0.088∗∗∗ −0.055∗∗ −0.110∗∗∗ −0.099∗∗∗ −0.136∗∗∗ 0.029 0.139∗∗∗ 0.137∗∗∗(0.024) (0.028) (0.023) (0.016) (0.021) (0.018) (0.022) (0.016)

mhv 0.004∗∗∗ 0.006∗∗∗ 0.004∗∗∗ 0.003∗∗∗ 0.009∗∗∗ 0.001∗∗∗ −0.006∗∗∗ −0.006∗∗∗(0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) (0.000)

cvurate −0.361∗∗ −0.013 −0.808∗∗∗ −1.160∗∗∗ −1.131∗∗∗ −1.175∗∗∗ −0.340∗∗ −1.454∗∗∗(0.169) (0.185) (0.178) (0.175) (0.197) (0.209) (0.162) (0.212)

metroadj 0.324∗∗∗ 0.356∗∗∗(0.025) (0.029)

Fixed Effect Yes Yes Yes Yes Yes Yes Yes YesLog L −9356425 −6388894.8 −2717527.7 −5051182.9 −3344577.6 −1562524.7 −10056934 −5224726.2Wald 9230.28 6720.33 17017.11 7955.29 5394.61 8499.52 13722.21 8354.70Obs. 2,137,330 1,110,772 1,025,595 2,197,914 1,146,845 1,006,495 2,141,009 2,180,850

Note: Robust standard errors appear in parentheses; ∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1.



locating in nonmetro adjacent counties mayhave grown, ceteris paribus. A nonmetro-adjacent county has a 38% greater numberof expected in-migrants, holding all othervariables constant from 1995–2000. Therespective number for the latter period is43%.

The estimated coefficients for the dis-tance variable is negative and highly sig-nificant in both metro-to-nonmetro andnonmetro-to-metro models for all samplesfor both periods, implying that longer dis-tance is associated with lower migrationflows. For example, a 100 mile increasein the distance between metro-nonmetrocounty pairs is associated with a nonmetrocounty’s in-migration rate decreasing by56%. Comparing the two sub-samples inmetro-to-nonmetro flows, the nonmetro-nonadjacent sample seems to have a smallerdistance effect than in the nonmetro-adjacentsample. This implies that distance is moreimportant to migrants moving into adja-cent counties as many of them tend to becommuters who have economic and noneco-nomic links to metro counties, but preferto live in more distant counties. The coef-ficient on the distance-squared variable ispositive, suggesting that the deterring effectsof distance declines as distance increases.12

A comparison of the two periods shows thatthe absolute value of the distance coefficientslightly increases in the second period forall samples. This result may conflict withsome of the claims in the previous literature(e.g., Juarez 2000) that distance may be lessof a migration barrier than in the past, butsupports the view that the cost of movingto remote locations increases because tech-nological advances may increase the valueof other types of agglomeration economiesfound in metro areas (Partridge et al. 2008).This result may also be at odds with the argu-ment that telecommuting has aided longerrural-urban commutes. Rather, it may beprimarily facilitating longer commutes withinlarge urban areas because the worker canoccasionally telecommute.

The population density results in all spec-ifications confirm the traditional gravitymodel hypothesis that migrants are attracted

12 Using the marginal formula for the Poisson function, themarginal distance effect reaches zero at 1,575 miles in the metro-to-nonmetro 1995–2000 sample and 1,544 miles in the 2005–2009sample. The corresponding zero marginal distance effects are at1,709 and 1,683 miles in the nonmetro-to-metro sample.

to more populous locations. Results in themetro-to-nonmetro specifications are consis-tent with the view that these migrants prefermore populated rural locations. In terms ofmagnitude, a 50-person increase in popula-tion per square mile is associated with a 48%increase in a nonmetro county’s in-migrationrate in the first time-period. The effect is evenmore notable for remote nonmetro countieswhen comparing the size of the estimatedcoefficients. Although the significance andthe sign of population density coefficientremains the same for the second period inthe metro-to-nonmetro specifications (2005–2009, Columns 5–7), the absolute value ofthe coefficient decreases substantially in thesecond period. The regression coefficientin the full sample decreased from 7.796 inthe first period to 0.345 in the second. A 50-person increase in population per square mileis associated with only a 2% increase in anonmetro county’s in-migration. The tempo-ral differences are also clearly visible in themetro-to-nonmetro subsamples. The advan-tages of population appear to have declinedfor metro-to-adjacent nonmetro migrants, inwhich improved information technologiesmay have reduced the need for local agglom-eration economies associated with populationdensity. Despite the temporal decline, the“high” returns to population density formetro-to-remote rural migration continued inboth periods, suggesting that existing agglom-eration economies remain an importantconsideration when moving to remote areas.

Compared to the metro-to-nonmetroresults, the size of the coefficient for pop-ulation density is markedly smaller inthe nonmetro-to-metro-county migrationresults, though some of this is scaling inthat population density is much higher inthe metropolitan destination compared tothe nonmetro destination in the metro-to-nonmetro results. However, there wasvirtually no change in in-migration in a typ-ical metro county (0.1%) for a 50-personincrease in population per square mile. Thesize of the coefficient remains relativelyunchanged between the two time periods.Thus, at the margin, returns to populationdensity for potential metro-to-remote-nonmetro migrants is much larger than fornonmetro-to-metro migrants, consistent witha threshold level of population density beinga key draw in more remote settings.

The coefficient of wage and salary per jobis significant and positive in all specifications



(except in the metro-to-nonmetro-adjacentsample in the first period, which has anunexpected negative sign) for both periods,indicating that, ceteris paribus, migrants aremore likely to move to counties that havehigher wage rates, confirming the labor mar-ket theory that in-migration is more likelyfor regions experiencing relatively high wagelevels. On average, a $1,000 increase in non-metro county salary is associated with a 1%increase in in-migration to that county, andthe same increase in a metro county is asso-ciated with a 10% increase in in-migration tothat county.

The results between two time periods inthe metro-to-nonmetro model show con-siderable nominal variation for the wagevariable’s results. For example, while theestimated coefficient for the full metro-to-nonmetro sample increased from 0.006 inthe first period to 0.044 in the second period,the estimated coefficient for the metro tononmetro-nonadjacent sample increasedfrom 0.050 in the first period to 0.067 in thesecond period. However, the coefficient fornonmetro-adjacent is negative and significantin the first period but is positive and signifi-cant in the second period. The smaller migra-tion response to wages for in-migration tometro adjacent counties compared to remotenonmetro counties is expected because locallabor market conditions should play a smallerrole for those who commute back to metroareas.13 The notable temporal change in thewage coefficient suggests that this labor mar-ket factor may have become an even moreimportant determinant of migration frommetro to nonmetro counties in the 2000s thanin the 1990s, downplaying the claim that eco-nomic factors may have become less impor-tant in metro-to-nonmetro migration. Forexample, a $1,000 increase in nonmetro salaryis related to a 4% increase in in-migration tothat county in the second period. However,a decrease in the size of the coefficient in thesecond time period in the nonmetro-to-metroflows indicates that the effect of the wagevariable may have weakened over time, sug-gesting a smaller role for economic effects (atleast through wages).

The estimated coefficient for the industrymix variable that measures labor demand

13 Future research should identify the responsiveness of com-muting from adjacent counties to wages in the nearest urbanarea. In the metro-to-nonmetro case, this effect is controlled forwith the origin county fixed effect.

shocks is positive and statistically significantin all specifications for both time-periods. Allelse being equal, a one standard deviationincrease in industry mix employment growthin a given nonmetro county was related toa 19% increase in the population movingto that county from a given metro countybetween 1995–2000, and the correspond-ing figure for nonmetro-to-metro modelwas 44%. This result supports the economicrestructuring argument of reversal and showsthat positive demand shocks are a strongpull factor in metro-to-nonmetro migration,indicating that job availability is a key fac-tor in rural areas, perhaps due to thin labormarkets. There are also noticeable differ-ences with regard to the magnitude of thecoefficient of this variable between samplesin the first period in the metro-to-nonmetromodel: the coefficient for the nonadjacentsample is larger than that for the adjacentsample. This relative difference also indicatesthe possibility that many migrants to adjacentcounties are commuters who have jobs inthe nearby metro area, though the statisticalsignificance in adjacent counties suggests thatlocal conditions play a role. The differencesare also visible between the time periods.The numerical size of the coefficient in allspecifications increased noticeably. Theseresults run counter to Partridge et al.’s (2012)findings, that employment-related migrationresponses declined after 2000—though a keydifference is that they were concerned withnet migration from all sources (especiallymetro-to-metro). In summary, the effectthat industry mix employment has on bothmetro-to-nonmetro and nonmetro-to-metromigration flows is highly significant and theeffect seems to be increasing over time.

One main explanation given in the earlyliterature for the rural “reversal” is thatretirees moved to nonmetro areas. Our proxyto measure this argument is to include theshare of population who are over 64 yearsold. The estimated coefficients for this vari-able do not have the hypothesized positivesign, although they are highly significant inall samples in both periods for metro-to-nonmetro model. In other words, nonmetrocounties that have a higher concentration ofolder people are not attractive to migrantscoming from metro counties. A comparisonbetween the two time periods shows thatthere are visible temporal changes in thecoefficients. The absolute value of the coef-ficients in all samples has increased from



the first period to the second. To furtherconsider this issue, in results not shown were-estimate the models by replacing the shareof those aged over 64 with the percentageof households with retirement and socialsecurity income, but the results are similarwith the same temporal tendencies betweenthe two time-periods. We also test anotherspecification using a dummy variable forretirement-destination counties developedby the USDA Economic Research Service(the number of residents 60 and older grewby 15% or more between 1990 and 2000due to in-migration). Not surprisingly givenits construction, the estimated coefficientof this variable is positive and highly sig-nificant in the first period for all samplesin metro-to-nonmetro flows. However, theresults change in the second period: the esti-mated coefficient is not significant for thefull sample, is negative and significant for thenonmetro-adjacent sample, and is positiveand significant for the nonmetro-nonadjacentsample. Even though this coefficient con-tinues to be positive and significant for thenonmetro-nonadjacent sample from the firstto the second period, the size of the coef-ficient decreases from 0.518 to 0.174 in thesecond. The coefficient of the retiree attrac-tion variable is negative and significant inthe nonmetro-to-metro estimation, but thetemporal changes seem to have reversed: theabsolute value of the coefficient in the sec-ond period in the nonmetro-to-metro modeldecreased, indicating a weakening effect.

The results show that the lower per capitalocal taxes coefficient is highly significantand negative in all specifications for bothperiods in the metro-to-nonmetro model,though the absolute value of the coeffi-cient seems to have increased in the secondperiod. The estimated coefficient of localgovernment expenditure per capita in themetro-to-nonmetro model is highly signif-icant but has an unexpected negative signacross the specifications in the first period.The results are similar in the second periodfor the full and metro-adjacent sample butthe coefficient is not statistically significant inthe nonadjacent sample. Though we controlfor economic conditions, government expen-ditures are affected by economic conditions,which may underlie some of this pattern. Asfor nonmetro-to-metro results, the coefficientof the government expenditure variablesis highly significant and has the expectedpositive sign for both time periods.

The estimate for the median housingvalue variable is highly significant but has anunexpected positive sign in all samples forboth time periods in the metro-to-nonmetromodel, consistent with unmeasured ameni-ties affecting the results.14 Indeed, we wouldnot expect households moving from metrolocations to have large negative marginalresponses to what should be relatively lownonmetro housing prices. Conversely, thisestimate is negative and highly significantin the nonmetro-to-metro model for bothperiods, suggesting that higher housing costsdeter nonmetro migrants to metro areas. Theunemployment risk variable has the expectednegative sign and is statistically significant inall specifications, indicating some evidencethat a stable job market at the destination isimportant for in-migrants, ceteris paribus.15

Note the relatively smaller marginal responsein the adjacent sample, which supports thenotion that local labor market conditionsmatter less in adjacent counties to metromigrants. The absolute size of this coefficientincreases significantly in the second period,indicating the increasing importance of jobmarket stability in the destination location.

Distance and the Role of Job Opportunitiesand Amenities

In this sub-section, we assess whether theattraction of job opportunities and nat-ural amenities vary between long- andshort-distance moves from metropolitanto nonmetropolitan locations. Greater dis-tance can reduce the amount of informationthat migrants have on potential destina-tions, including whether there are suitablejob opportunities (Brown and Scott 2012).We expect that potential economic migrantswould have better labor market informationabout nearby locations, suggesting that thepotential effects of job growth as an attrac-tive force diminish with greater distance.Regarding amenities, we expect that nearbylocations would have similar packages ofnatural amenity bundles, even if the amenityscores differ between two relatively nearby

14 Jeanty, Partridge, and Irwin (2012) contend that a positivemigration coefficient on median housing values may suggest thatthere are omitted amenities capitalized into housing values. Theseauthors suggest possible solutions for future research.

15 This is also true in the metro-to-adjacent-nonmetro sub-sample in which many of the in-migrants are expected to becommuters.


694A

pril2015

Am

er.J.A

gr.E

con.

Table 3. Fixed Effect Poisson Estimation Results with Interaction Term for Distance and Industrial Mix


1995–2000 2005–2009 1995–2000 2005–2009


amnscale 0.091∗∗∗ 0.074∗∗∗ 0.100∗∗∗ 0.065∗∗∗ 0.026∗∗∗ 0.088∗∗∗ 0.200∗∗∗ 0.197∗∗∗(0.008) (0.009) (0.007) (0.008) (0.009) (0.009) (0.008) (0.010)

distance −0.787∗∗∗ −0.856∗∗∗ −0.633∗∗∗ −0.863∗∗∗ −0.917∗∗∗ −0.705∗∗∗ −0.760∗∗∗ −0.850∗∗∗(0.053) (0.052) (0.068) (0.055) (0.056) (0.068) (0.074) (0.078)

distance_sq 0.026∗∗∗ 0.029∗∗∗ 0.019∗∗∗ 0.029∗∗∗ 0.032∗∗∗ 0.022∗∗∗ 0.022∗∗∗ 0.024∗∗∗(0.002) (0.002) (0.003) (0.002) (0.002) (0.003) (0.005) (0.006)

popden 7.766∗∗∗ 7.368∗∗∗ 9.223∗∗∗ 0.339∗∗∗ 0.181∗∗∗ 4.775∗∗∗ 0.019∗∗∗ 0.022∗∗∗(0.301) (0.342) (0.340) (0.023) (0.026) (0.199) (0.005) (0.005)

wage 0.005∗ −0.009∗∗ 0.050∗∗∗ 0.044∗∗∗ 0.029∗∗∗ 0.067∗∗∗ 0.099∗∗∗ 0.055∗∗∗(0.003) (0.004) (0.004) (0.002) (0.003) (0.004) (0.002) (0.002)

indusmix 0.044∗∗∗ 0.040∗∗∗ 0.048∗∗∗ 0.084∗∗∗ 0.077∗∗∗ 0.077∗∗∗ 0.069∗∗∗ 0.133∗∗∗(0.004) (0.004) (0.004) (0.006) (0.007) (0.009) (0.003) (0.008)

indusmix*distance −0.002∗∗∗ −0.003∗∗∗ −0.002∗ −0.004∗∗∗ −0.006∗∗∗ −0.003 0.000 0.004∗(0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.001) (0.002)

elder −0.049∗∗∗ −0.033∗∗∗ −0.068∗∗∗ −0.082∗∗∗ −0.065∗∗∗ −0.102∗∗∗ −0.064∗∗∗ −0.059∗∗∗(0.004) (0.004) (0.004) (0.005) (0.006) (0.005) (0.003) (0.005)

pctax −0.233∗∗∗ −0.186∗∗∗ −0.326∗∗∗ −0.312∗∗∗ −0.207∗∗∗ −0.550∗∗∗ −0.142∗∗∗ −0.345∗∗∗(0.037) (0.052) (0.034) (0.031) (0.036) (0.044) (0.037) (0.035)

pcgexp −0.090∗∗∗ −0.058∗∗ −0.111∗∗∗ −0.098∗∗∗ −0.135∗∗∗ 0.029 0.139∗∗∗ 0.138∗∗∗(0.024) (0.028) (0.023) (0.016) (0.021) (0.018) (0.022) (0.016)

mhv 0.004∗∗∗ 0.006∗∗∗ 0.004∗∗∗ 0.003∗∗∗ 0.009∗∗∗ 0.001∗∗∗ −0.006∗∗∗ −0.006∗∗∗(0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) (0.000)

cvurate −0.362∗∗ −0.019 −0.805∗∗∗ −1.156∗∗∗ −1.145∗∗∗ −1.159∗∗∗ −0.346∗∗ −1.466∗∗∗(0.168) (0.185) (0.179) (0.176) (0.199) (0.214) (0.163) (0.211)

metroadj 0.322∗∗∗ 0.355∗∗∗(0.025) (0.029)

Fixed Effect Yes Yes Yes Yes Yes Yes Yes YesLog L −9345874 −6378391.3 −2716053.8 −5048510 −3340895.4 −1562135.5 −10056635 −5223313Wald 10140.13 7095.98 17492.35 8254.32 5694.52 8846.96 13700.80 8085.21Obs. 2,137,330 1,110,772 1,025,595 2,197,914 1,146,845 1,006,495 2,141,009 2,180,850

Note: Robust standard errors appear in parentheses; ∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1.


Rupasingha,L

iu,andP

artridgeD

eterminants

ofM

etroto

Non-M

etroM

igration695

Table 4. Fixed Effect Poisson Estimation Results with Interaction Term for Distance and Natural Amenity Scale


1995–2000 2005–2009 1995–2000 2005–2009


amnscale 0.061∗∗∗ 0.040∗∗∗ 0.062∗∗∗ 0.058∗∗∗ 0.017 0.062∗∗∗ 0.034 0.054∗∗(0.013) (0.015) (0.014) (0.012) (0.018) (0.016) (0.024) (0.025)

amnscale*distance 0.005∗∗∗ 0.006∗∗∗ 0.006∗∗∗ 0.001 0.001 0.004 0.028∗∗∗ 0.025∗∗∗(0.002) (0.002) (0.002) (0.002) (0.003) (0.003) (0.005) (0.006)

distance −0.815∗∗∗ −0.892∗∗∗ −0.651∗∗∗ −0.895∗∗∗ −0.965∗∗∗ −0.724∗∗∗ −0.728∗∗∗ −0.787∗∗∗(0.046) (0.043) (0.059) (0.051) (0.050) (0.060) (0.055) (0.071)

distance_sq 0.025∗∗∗ 0.028∗∗∗ 0.018∗∗∗ 0.028∗∗∗ 0.031∗∗∗ 0.022∗∗∗ 0.016∗∗∗ 0.018∗∗∗(0.002) (0.002) (0.003) (0.003) (0.003) (0.003) (0.004) (0.006)

popden 7.925∗∗∗ 7.666∗∗∗ 9.366∗∗∗ 0.349∗∗∗ 0.198∗∗∗ 4.833∗∗∗ 0.026∗∗∗ 0.029∗∗∗(0.314) (0.367) (0.325) (0.023) (0.024) (0.191) (0.005) (0.005)

wage 0.004 −0.010∗∗∗ 0.048∗∗∗ 0.044∗∗∗ 0.029∗∗∗ 0.067∗∗∗ 0.102∗∗∗ 0.056∗∗∗(0.003) (0.004) (0.004) (0.002) (0.003) (0.004) (0.002) (0.002)

indusmix 0.035∗∗∗ 0.029∗∗∗ 0.040∗∗∗ 0.070∗∗∗ 0.058∗∗∗ 0.065∗∗∗ 0.071∗∗∗ 0.143∗∗∗(0.002) (0.003) (0.002) (0.004) (0.005) (0.005) (0.002) (0.006)

elder −0.049∗∗∗ −0.033∗∗∗ −0.068∗∗∗ −0.082∗∗∗ −0.065∗∗∗ −0.102∗∗∗ −0.066∗∗∗ −0.059∗∗∗(0.004) (0.004) (0.004) (0.005) (0.006) (0.006) (0.004) (0.005)

pctax −0.230∗∗∗ −0.165∗∗∗ −0.327∗∗∗ −0.308∗∗∗ −0.203∗∗∗ −0.544∗∗∗ −0.004 −0.245∗∗∗(0.036) (0.053) (0.038) (0.030) (0.037) (0.043) (0.029) (0.029)

pcgexp −0.100∗∗∗ −0.075∗∗∗ −0.119∗∗∗ −0.101∗∗∗ −0.140∗∗∗ 0.025 0.054∗∗∗ 0.086∗∗∗(0.024) (0.028) (0.022) (0.016) (0.021) (0.018) (0.017) (0.012)

mhv 0.004∗∗∗ 0.006∗∗∗ 0.004∗∗∗ 0.003∗∗∗ 0.009∗∗∗ 0.001∗∗∗ −0.008∗∗∗ −0.008∗∗∗(0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) (0.000)

cvurate −0.333∗∗ 0.026 −0.766∗∗∗ −1.157∗∗∗ −1.127∗∗∗ −1.148∗∗∗ −0.014 −1.216∗∗∗(0.168) (0.185) (0.178) (0.175) (0.198) (0.211) (0.159) (0.238)

metroadj 0.328∗∗∗ 0.357∗∗∗(0.026) (0.030)

Fixed Effect Yes Yes Yes Yes Yes Yes Yes YesLog L −9346479 −6379587.7 −2712982 −5050993 −3344416.2 −1561784.5 −9752745 −5138697Wald 8676.02 5986.72 16210.56 8000.65 5372.07 8100.88 10480.42 6622.02Obs. 2,137,330 1,110,772 1,025,595 2,197,914 1,146,845 1,006,495 2,141,009 2,180,850

Note: Robust standard errors in parentheses; ∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1.



locations. Thus, we expect the draw of ameni-ties to be stronger in more distant locationsbecause different regions offer entirely dif-ferent bundles of amenities. To assess thesepossibilities, we add both interactions ofindustry mix job growth and the amenityscore with distance; the respective results arereported in tables 3 and 4.16

Table 3 shows a significant and negativecoefficient on the interaction between indus-try mix job growth and distance, supportingour hypothesis that the pull effects of jobgrowth diminish with distance. The effectis more prominent for metro-to-nonmetromigration in the first period, though theeffect becomes insignificant in the metro-to-nonadjacent model in the second period.The larger negative magnitude of the adja-cent distance × industry mix job growthvariable is a little surprising, suggesting thatlocal job market conditions matter less atgreater distances (which would be moredifficult to commute). This pattern may bebecause natural amenities matter more forthe more distant moves in general, whichis discussed below. The magnitudes of thedistance interaction coefficient are larger inthe latter period, which suggests that shorterjob-related moves were more of the normduring the sluggish post-2000 environmentincluding the Great Recession, perhaps dueto greater risk aversion (though we cautionthat the coefficient is imprecisely measured inthe metro to nonadjacent model). Nonmetro-to-metro results suggest that local job marketconditions can be an attractive force formigration regardless of the distance in thefirst time-period, but they matter more atshort distances in the second time-period.

Table 4 shows that the coefficient for theinteraction between distance and the natu-ral amenity scale has the expected positivecoefficient. However, the coefficient is onlystatistically significant in the earlier periodand the magnitude of the coefficient is alsosmaller in the latter period. So at least for

16 We did not include the respective interactions with dis-tance squared because it would have greatly complicated theinterpretation of the results. In sensitivity analysis, we includedthe distance-squared interaction with amenities and industry mixjob growth. In this case, the marginal effect of amenities andindustry mix was mostly unaffected by including their respectiveinteraction with distance squared. However, when we includedthese interactions with distance squared, the resulting coefficientson the distance squared interaction were negative and statisti-cally significant, suggesting some attenuation at greater distancesregarding how amenities and industry mix job growth affectmigration.

the earlier period, while having strong natu-ral amenities is important regardless of thedistance of the metro to nonmetro move,the amenity pull effect is stronger for moredistant moves. The fact that the interactivevariable in the second time period is notstatistically significant relative to the slopeof distance indicates that distance was notan intervening factor when migrating foramenity purposes in the second time period.

Conclusions

The study provides new insights into thechanging U.S. migration patterns, wheremigration has historically been from non-metro areas to metro areas, but has changedto show more migration from metro to non-metro areas during the last two decades. Wefind support for both the deconcentrationand economic restructuring perspectivesput forth by previous studies, with someexceptions and temporal changes. Morespecifically, key destination county character-istics such as natural amenities, populationdensity, distance, wage and salaries, indus-trial mix, adjacent to metro counties, andshare of population aged over 64 are shownto be significantly associated with metro tononmetro migration. All these factors havehypothesized outcomes except the share ofpopulation over 64-years old variable, whichis negative.

While our results suggest that attemptsby local policymakers to improve and pro-mote local natural amenities to attract peopleand businesses may still be good policy, thepersistence of such policy over time maybe questionable. Nonmetro-to-metro flowresults suggest that natural amenities are stillimportant in retaining population in non-metro locations because these movers stillprefer to locate in high amenity metro loca-tions, and the effects show no change overtime. We also find that migration respondsto agglomeration economies even in non-metro areas, suggesting that some thresholdof agglomeration economies are necessary,though it is also contrary to the claims madeby the deconcentration perspective that thesemovers prefer less dense areas. This may bedue to the people enjoying easy access torural amenities, but at the same time enjoy-ing some level of urban amenities, includingaccess to technology. The effects seem tobe somewhat muted over time. We also find



that distance from metro counties negativelyaffect in-migration in nonmetro areas, butthis effect is more pronounced in nonmetro-adjacent than in nonmetro-nonadjacentcounties, indicating the possibility that manyof the migrants to adjacent counties considercommuting. There are no notable temporalchanges in the distance effects, which down-plays the argument that people may moveout to rural areas and then telecommute. Insummary, results for both population densityand distance show that urban amenities inrural areas and proximity to metro areas areimportant if nonmetro areas are to attractand retain migrants.

While our results show that nonmetrocounties with very high retiree growth rate(15% of more) tend to attract more migrantsfrom metro locations, the overall resultsdiminish the claim that retiree attractionmay be good policy for nonmetro countiesin general. Results show that the economicrestructuring argument has some validityin metro-to-nonmetro migration and thatlabor market opportunities play a significantrole. The effect of the industry mix variableseems to be unchanged in nonmetro-to-metromigration over the two periods, but this effecthas clearly increased in metro-to-nonmetromigration, indicating that successful gov-ernment policies that are geared towardscreating jobs tend to attract more and morepeople into nonmetro areas.

Further analysis suggested that the jobgrowth effects for metropolitan to non-metropolitan migration declined with greaterdistance from the origin. Indeed, there isevidence that the role of distance in affectinghow job growth affects migration increasedin the post 2000 period, perhaps suggestinggreater risk aversion. In a similar analysis,the draw of natural amenities also increasedwith distance, thus explaining metropolitanto nonmetropolitan migration, but the effectwas only statistically significant in the earlierperiod. We also confirm that local taxes arestill a deterrent for in-migration, whetherthe migration is from metro to nonmetro ornonmetro to metro areas. One importantexception is housing values at the destination,where we find that while higher values attractmore migrants from metro to nonmetro areas(perhaps due to unmeasured amenities),the opposite is true for nonmetro-to-metromigration.

Finally, it is important to draw attention tothe role of context in shaping the substantial

changes in coefficients over time. Forinstance, the coefficients of several variablesthat measure economic restructuring, such asvariability of unemployment, wage rates, andjob growth or industry mix, increased in mag-nitude between the periods. Even though thegeneral patterns of migration stayed the sameas indicated above, economic shocks such asthe significant drop in employment in bothmetropolitan and nonmetropolitan areas in2007–2009 seem likely to make economicfactors more relevant to migration patterns inthe second period.17

References

Arauzo-Carod, J.M., and M.C.M. Antolín.2004. Firm Size and GeographicalAggregation: An Empirical Appraisalin Industrial Location. Small BusinessEconomics 22 (3/4): 299–312.

Arauzo Carod, J.M. 2005. Determinants ofIndustrial Location: An Applicationfor Catalan Municipalities. Papers inRegional Science 84 (1): 105–20.

Arauzo-Carod, J.M., D. Liviano-Solis, andM. Monjón-Antolín. 2010. EmpiricalStudies in Industrial Location: An Ass-essment of Their Methods and Results.Journal of Regional Science 50 (3):685–711.

Arzaghi, M., and A. Rupasingha. 2013. Migra-tion as a Way to Diversify: Evidencefrom Rural to Urban Migration in theUS. Journal of Regional Science 53 (4):690–711.

Bartik, T.J. 1991. Who Benefits from State andLocal Economic Development Policies?Upjohn Institute, Kalamazoo, MI.

Benetsky, M., and K. Koerber. 2012. Howdo the ACS Five-Year Migration DataCompared to the 2000 Census Migra-tion Data? http://www.census.gov/hhes/migration/files/PAA_2012_County-County.pdf (accessed on December 31,2014).

Blanchard, O.J., L.F. Katz, R.E. Hall, andB. Eichengreen. 1992. Regional Evolu-tions. Brookings Papers on EconomicActivity 1992 (1): 1–75.

Brown, W.M., and D.M. Scott. 2012. HumanCapital Location Choice: Accounting

17 We are thankful to an anonymous referee for raising thispoint.


http://www.census.gov/hhes/migration/files/PAA_2012_County-County.pdf




for Amenities and Thick Labor Mar-kets. Journal of Regional Science 52 (5):787–808.

China Development Research Founda-tion. 2010. Trends in Urbanisation andUrban Policies in Organization for Eco-nomic Cooperation and DevelopmentCountries: What Lessons for China?Organization for Economic Cooper-ation and Development Publishing.www.oecd.org/dataoecd/2/18/45159707.pdf (accessed on December 31, 2014).

Christiadi, and B. Cushing. 2008. The JointChoice of an Individual’s Occupation andDestination. Journal of Regional Science48 (5): 893–919.

Combes, P.P., G. Duranton, L. Gobillon,and S. Roux. 2012. Sorting and LocalWage and Skill Distributions in France.Regional Science and Urban Economics42 (6): 913–30.

Cromartie, J.B. 1998. Net migration in theGreat Plains Increasingly Linked toNatural Amenities and Suburbaniza-tion. Rural Development Perspectives 13:27–34.

Cromartie, J. 2013. Nonmetro Areas as aWhole Experience First Period of Popu-lation Loss. Washington DC: EconomicResearch Service, U.S. Department ofAgriculture. http://www.ers.usda.gov/amber-waves/2013-may/nonmetro-areas-as-a-whole-experience-first-period-of-population-loss.aspx#.U0Reo1fVu2X(accessed on April 8, 2014).

Cushing, B., and J. Poot. 2004. CrossingBoundaries and Borders: Regional Sci-ence Advances in Migration Modeling.Papers in Regional Science 83 (1): 317–38.

Daveri, F., and R. Faini. 1999. Where DoMigrants Go? Oxford Economic Papers51 (4): 595–622.

Davies, P.S., M.J. Greenwood, and H. Li.2001. A Conditional Logit Approach toU.S. State-to-State Migration. Journal ofRegional Science 41 (2): 337–60.

Deller, S.C., T.H. Tsai, D.W. Marcouiller,and D.B.K. English. 2001. The Role ofAmenities and Quality of Life in RuralEconomic Growth. American Journal ofAgricultural Economics 83 (2): 352–65.

Etzo, I. 2010. The Determinants of theRecent Interregional Migration Flowsin Italy: A Panel Data Analysis. Journalof Regional Science 51 (5): 948–66.

Frey, W.H., and K.M. Johnson. 1996. Con-centrated Immigration, Restructuring

and the Selective Deconcentrationof the United States Population.Population Studies Center ResearchReport Series No. 96–371, The Popula-tion Studies Center at the University ofMichigan.

Friedman, J., D.A. Gerlowski, andJ. Silberman. 1992. What AttractsForeign Multinational Corporation?Evidence from Branch Plant Locationin the United States. Journal of RegionalScience 32 (4): 403–18.

Fuguitt, G.V., and C.L. Beale. 1996. RecentTrends in Nonmetropolitan Migration:Toward a New Turnaround? Growth andChange 27 (2): 156–74.

Gabe, T.M., and K.P. Bell. 2004. Tradeoffsbetween Local Taxes and GovernmentSpending as Determinants of BusinessLocation. Journal of Regional Science 44(1): 21–41.

Ghatak, S., P. Levine, and S.W. Price. 1996.Migration Theories and Evidence:an Assessment. Journal of EconomicSurveys 10 (2): 159–98.

Goetz, S.J. 1999. Migration and Local LaborMarkets. In Web Book of Regional Sci-ence. West Virginia University, RegionalResearch Institute. http://www.rri.wvu.edu/WebBook/Goetz/contents.htm(accessed on December 31, 2014).

Greenwood, M.J. 1985. Human Migration:Theory, Models, and Empirical Stud-ies. Journal of Regional Science 25 (4):521–44.

Greenwood, M.J., G.L. Hunt, D.S. Rickman,and G.I. Treyz. 1991. Migration, RegionalEquilibrium, and the Estimation ofCompensating Differentials. AmericanEconomic Review 81 (5): 1382–90.

Guimarães, P., O. Figueiredo, and D. Wood-ward. 2000. Agglomeration and theLocation of Foreign Direct Investment inPortugal. Journal of Urban Economics 47(1): 115–35.

———. 2003. A Tractable Approach to theFirm Location Decision Problem. Reviewof Economics and Statistics 85 (1):201–4.

———. 2004. Industrial Location Modeling:Extending the Random Utility Frame-work. Journal of Regional Science 44 (1):1–20.

Hansen, E.R. 1987. Industrial LocationChoice in São Paulo, Brazil: A NestedLogit Model. Regional Science andUrban Economics 17 (1): 89–108.


http://www.oecd.org/dataoecd/2/18/45159707.pdf

http://www.oecd.org/dataoecd/2/18/45159707.pdf

http://www.ers.usda.gov/amber-waves/2013-may/nonmetro-areas-as-a-whole-experience-first-period-of-population-loss.aspx#.U0Reo1fVu2X




http://www.rri.wvu.edu/WebBook/Goetz/contents.htm

http://www.rri.wvu.edu/WebBook/Goetz/contents.htm


Jeanty, P.W., M. Partridge, and E. Irwin. 2010.Estimation of a Spatial SimultaneousEquation Model of Population Migrationand Housing Price Dynamics. RegionalScience and Urban Economics 40 (5):343–52.

Jensen, T., and S. Deller. 2007. Spatial Mod-eling of the Migration of Older PersonsWith a Focus on Amenities. Review ofRegional Studies 37 (3): 303–43.

Juarez, J.P. 2000. Analysis of InterregionalLabor Migration in Spain Using GrossFlows. Journal of Regional Science 40 (2):377–99.

Kahsai, M.S., T.G. Gebremedhin, and P.V.Schaeffer. 2011. A Spatial Analysisof Amenity and Regional EconomicGrowth in Northeast Region. Review ofUrban & Regional Development Studies23 (2/3): 77–93.

Kim, J. 1983. Factors Affecting Urban-to-rural Migration. Growth and Change 14(3): 38–43.

Knapp, T.A., and P.E. Graves. 1989. On theRole of Amenities in Models of Migra-tion and Regional Development. Journalof Regional Science 29 (1): 71–87.

Levine, R., N. Loayza, and T. Beck. 2000.Financial Intermediation and Growth:Causality and Causes. Journal ofMonetary Economics 46 (1): 31–77.

Long, J.S. 1997. Regression Models for Cate-gorical and Limited Dependent Variables:Advanced Quantitative Techniques in theSocial Sciences, Vol. 7. California: SagePublications, Inc.

Markusen, A., and G. Schrock. 2006. TheArtistic Dividend: Urban Artistic Spe-cialisation and Economic DevelopmentImplications. Urban Studies 43 (10):1661–86.

McFadden, D.L. 1974. Conditional LogitAnalysis of Qualitative Choice Behav-ior. In Frontiers in Econometrics, ed.P. Zarembka, 105–42. New York:Academic Press.

McFadden, D. 1978. Modelling the Choiceof Residential Location. In Spatial Inter-action Theory and Planning Models, ed.A. Karlqvist et al., 1–24. Amsterdam:North-Holland.

McGranahan, D.A. 1999. Natural Ameni-ties Drive Rural Population Change.Washington DC: U.S. Department ofAgriculture, Economic Research Service,No. 33955.

Mueser, P.R., and P.E. Graves. 1995. Examin-ing the Role of Economic Opportunityand Amenities in Explaining Popula-tion Redistribution. Journal of UrbanEconomics 37: 176–200.

Murphy, A., J. Muellbauer, and G. Cameron.2006. Housing Market Dynamics andRegional Migration in Britain. Depart-ment of Economics Discussion Papers,University of Oxford, 275.

Nelson, L., and P.B. Nelson. 2011. The GlobalRural: Gentrification and Linked Migra-tion in the Rural USA. Progress inHuman Geography 35 (4): 441–59.

Nelson, P.B., A. Oberg, and L. Nelson. 2010.Rural Gentrification and Linked Migra-tion in The United States. Journal ofRural Studies 26 (4): 343–52.

O’Keefe, S. 2004. Locational Choice ofAFDC Recipients within California: aConditional Logit Analysis. Journal ofPublic Economics 88 (7–8): 1521–42.

Partridge, M.D., D.S. Rickman, K. Ali, andM.R. Olfert. 2008. Lost in Space: Popula-tion Growth in the American Hinterlandsand Small Cities. Journal of EconomicGeography 8 (6): 727–57.

Partridge, M.D., K. Ali, and M.R. Olfert.2010a. Rural to Urban Commuting:Three Degrees of Integration. Growthand Change 41 (2): 303–35.

Partridge, M.D., D.S. Rickman, K. Ali,and M.R. Olfert. 2010b. The SpatialDynamics of Factor Price Differentials:Productivity or Consumer AmenityDriven? Regional Science and UrbanEconomics 40: 440–52.

Partridge, M.D., D.S. Rickman, M.R. Olfert,and K. Ali. 2012. Dwindling US InternalMigration: Evidence of Spatial Equilib-rium or Structural Shifts in Local LaborMarkets? Regional Science and UrbanEconomics 42: 375–88.

Rayer, S., and D.L. Brown (2001). Geo-graphic Diversity of Inter-CountyMigration in the United States, 1980–1995. Population Research and PolicyReview 20 (3): 229–52.

Rickman, D.S., and S.D. Rickman. 2011.Population Growth in High-AmenityNonmetropolitan Areas: What’s thePrognosis? Journal of Regional Science51 (5): 863–79.

Roback, J. 1982. Wages, Rents, and the Qual-ity of Life. Journal of Political Economy90: 1257–78.



Rosenzweig, M.R., and O. Stark. 1989. Con-sumption Smoothing, Migration, andMarriage: Evidence from Rural India.Journal of Political Economy 97 (4):905–26.

Rudzitis, G. 1999. Amenities IncreasinglyDraw People to the Rural West. RuralDevelopment Perspectives 14: 9–13.

Stark, O., and D. Levhari. 1982. On Migrationand Risk in LDCs. Economic Devel-opment and Cultural Change 31 (1):191–6.

Tiebout, C.M. 1956. A Pure Theory ofLocal Expenditures. Journal of PoliticalEconomy 64 (5): 416–24.

United States Census Bureau. 2014.CPS Historical Geographical Mobil-ity/Migration Tables. Available at: http://www.census.gov/hhes/migration/data/cps/historical.html (accessed on December31, 2014).

Woodward, D.P. 1992. Locational Deter-minants of Japanese ManufacturingStart-ups in the United States. SouthernEconomic Journal 58 (3): 690–708.

Wu, J.J., and M. Gopinath. 2008. What CausesSpatial Variations in Economic Devel-opment in the United States? AmericanJournal of Agricultural Economics 90 (2):392–408.


http://www.census.gov/hhes/migration/data/cps/historical.html



rural bound:determinantsof metroto non-metro … · 2017-10-02 · this article investigates the...

Documents