Journal of International Migration and Integration / Revue de l'integration et de la migration internationale

, Volume 10, Issue 2, pp 121–143

Immigration, Trade and Home Country Development: State-Level Variation in the US Immigrant–Export Link

Authors

    • Department of EconomicsFranklin & Marshall College
Article

DOI: 10.1007/s12134-009-0096-x

Cite this article as:
White, R. Int. Migration & Integration (2009) 10: 121. doi:10.1007/s12134-009-0096-x

Abstract

This article examines the pro-trade influence of immigrants using data on state-level exports from the 48 contiguous USA to 28 countries during the year 1993. Immigrants from lesser developed countries are found to exert stronger proportional effects on state-level exports relative to the immigrants from more developed countries. Calculation of absolute immigrant effects at state, regional and national levels also reveal influences of immigrants from developing countries are of greater magnitude; however, results depend on the metric employed to categorize countries as developing or developed. The findings emphasize the importance of immigrants’ connections to business and social networks and allow for an improved understanding of the role that information asymmetries play in fomenting opportunities for immigrants to enhance trade.

Keywords

DevelopmentGravityImmigrationState-level exports

Résumé

Nous examinons l’influence pro-commerce des immigrés en utilisant des données sur les exports au niveau national des 48 états contigus des États-Unis, à 28 pays pendant l’année 1993. On trouve que les immigrés des pays moins développés exercent de plus forts effets proportionnels sur les exports au niveau national par rapport aux immigrés des pays plus développés. Le calcul des effets-immigrés absolus au niveau de l’état, de la région, et de la nation révèlent aussi que les influences des immigrés des pays en voie de développement sont d’une plus grande ampleur; les résultats dépendent pourtant de la métrique employée pour caractériser les pays comme en voie de développement ou développés. Les conclusions soulignent l’importance des liens des immigrés aux réseaux commerciaux et sociaux et permettent une compréhension améliorée du rôle joué par des assymétries de renseignment dans la création des occasions pour l’augmentation du commerce par les immigrés.

Mots clés

développementgravitéimmigrationexports au niveau national

Introduction

In recent years, as the public debate surrounding the USA immigration policy has become more heated and increasingly contentious, a growing literature has documented a positive relationship between immigrants and host-home country trade flows. Although numerous social ills, including job loss, crime, and a strain on social service provision are cited as anticipated results of immigration, the trade-enhancing influences of immigrants rarely, if ever, enter into discussions of immigration policy. Policy formulation would benefit from an accurate accounting of all effects attributable to immigration. To that end, we examine one facet of the immigrant–trade relationship: the influence of immigrants on US state-level exports and whether variation exists in the state-level immigrant–export link across home country development classifications. The findings provide for an improved understanding of the importance of immigrant networks as a means by which trade is facilitated and, in doing so, add to the relevant literature while also conferring benefits to the immigration policy debate.

There are two channels through which immigrants are thought to influence US exports to their home countries. First, immigrants may arrive in the host country in possession of information relating to home country markets that, if successfully exploited, reduces transaction costs and thus may increase exports. Such information could range from seemingly innocuous language skills to understanding the complexities of informal contracting procedures and/or expectations of culture-specific business practices. Second, exports may increase if immigrants maintain connections to business or social networks that ameliorate search costs associated with the matching of buyers and sellers or that act as a form of implicit insurance that fosters the completion of trade deals via a reputation-maintenance effect. In both instances, immigrants’ network connections would reduce the extent of asymmetric information and potentially facilitate exports.

Gould (1994), using US data, is the first to document a positive influence of immigrants on home-host country trade. Although a detailed review of the related research is beyond the scope of this paper, to provide an indication of the voluminous literature that exists on the topic, we note that White (2007a, 2009) and Mundra (2005) extend the work of Gould (1994) and report positive immigrant influences on US trade. Similarly, pro-trade immigrant influences are reported for many other home countries: White and Tadesse (2007) for Australia; Wagner et al. (2002), Head and Ries (1998), and Helliwell (1997) for Canada; White (2007b) for Denmark; Piperakis et al. (2003) for Greece; White and Tadesse (2009) for Italy; Hong and Santhapparaj (2006) for Malaysia; Bryant and Law (2004) for New Zealand; Blanes (2003, 2006) for Spain; and Girma and Yu (2002) for the UK. Focusing on the role of networks, Combes et al. (2005) examine intra-France trade and report a pro-trade influence of migrants. Similarly, Greenaway et al. (2007) and Rauch and Trindade (2002) employ Chinese population shares to represent the presence of ethnic Chinese networks and report that such networks increase bilateral trade flows. Finally, Blanes (2005) and Blanes and Martín-Montaner (2006) for Spain and White (2008) for the US report that immigrants exert positive influences on intra-industry trade.

Several recent studies that examine the influence of immigrants on US state-level exports emphasize the role of networks but fail to consider possible variation in the trade-enhancing influences of immigrants across home countries (Tadesse and White 2008; Dunlevy 2006; Herander and Saavedra 2005; Bardhan and Guhathakurta 2005). Exceptions are Bandyopadhyay et al. (2008) and Co et al. (2004). Using data on exports from each of the 50 states and the District of Columbia to 29 countries for the years 1989–1990 and 1999–2000, the authors find that ethnic networks are important for only a subset of countries but are considerably greater in magnitude for this cohort relative to effects reported in studies that assume identical immigrant–export effects for all home countries. Several explanations are offered for the observed variation. For example, cultural and institutional differences between host and home countries would likely correspond with greater information asymmetries and afford opportunities for immigrants to increase trade. Similarly, linguistic differences, at times used as a proxy for cultural dissimilarity (Dunlevy 2006; Hutchinson 2002; Boisso and Ferrantino 1997), may impede the writing and enforcement of contracts and thus hinder the initiation and completion of trade deals. Although these explanations are certainly plausible, we consider the basis for the observed variation an open empirical question and posit that lesser-developed countries are more likely than developed countries to lack the institutions and/or formal channels that facilitate trade. Accordingly, immigrants from such countries may face relatively greater opportunities to utilize their network connections and, thus, to exert positive influences on exports.

Although findings presented in this paper are relevant to Bandyopadhyay et al. (2008), in many aspects, the work is more closely linked to that of Co et al. (2004). Employing data for state-level exports to 28 immigrant home countries during the year 1993, Co et al. (2004) examine variation in the immigrant–export relationship by stratifying their sample according to OECD membership and estimating identical regression equations for the OECD member and non-member samples. The authors then construct confidence intervals for the estimated coefficients on the immigrant stock variables. Because the confidence intervals are found to overlap, the authors conclude that there is no difference between developed and developing home countries in terms of the effects of immigrants on state-level exports. The use of sample stratification and confidence intervals to discern variation in effects is problematic. While coefficients from sub-group regressions represent group-specific effects, using confidence intervals to compare immigrant effects assumes homoskedasticity in variances across the sub-group regressions. Furthermore, although few would disagree that OECD membership correlates with economic development, this metric leads Co et al. (2004) to classify Hong Kong, Mexico, and South Korea as developing countries. As Mexico and South Korea joined the OECD in 1994 and 1996, respectively, and the per capita income of Hong Kong exceeds that of many OECD members, it is unlikely that these countries are truly representative of the developing world.

Following Co et al. (2004), we use cross-sectional data for the year 1993; however, we forgo sample stratification and instead pool our data and employ interaction terms to identify variation in immigrants’ influences on state-level exports across a range of home country development classifications. In doing so, we rely upon a broader set of development classifications [OECD membership, UN Human Development Index (HDI) classifications, and World Bank income classifications] than did Co et al. (2004). Extending the related literature, we illustrate that the homoskedasticity assumption implicitly invoked by Co et al. (2004) is flawed and that there is, in fact, variation in the immigrant–export relationship. As anticipated, we find that immigrants from lesser developed countries exert stronger proportional effects on state-level exports relative to the effects of immigrants from more developed countries. Additionally, calculation of absolute immigrant effects at state, regional, and national levels suggests that the influence of immigrants from developing countries is often of greater magnitude than the absolute effects estimated for immigrants from developed countries; however, the relative magnitudes of absolute effects depend on the metric employed to categorize home countries as developing or developed. Taken collectively, the results emphasize the importance of immigrants’ connections to business and/or social networks and also allow for an improved understanding of the role that information asymmetries play in fomenting opportunities for immigrants to enhance trade flows.

Empirical Specification

We begin our analysis by first replicating the estimation strategy of Co et al. (2004). Using comparable data for the same 28 home countries and the same time period, we examine the assumption of homoskedasticity in variances across regressions.1 We then estimate an alternative specification to evaluate variation in proportional immigrant–export effects across a number of home country development classifications. The ordinary least squares technique is employed for both series of estimations. Finally, we estimate average absolute immigrant–export effects at the state-level and consider the existence of potential heterogeneity in effects across development classifications.2 Consistent with prior research, an augmented gravity specification is employed. The equation employed by Co et al. (2004) is presented as Eq. 1.3
$$ \begin{array}{*{20}c} {\ln X_{ijt + 3} = \alpha_0 + \beta_1 \ln {\text{IM}}_{ijt} + \beta_2 \ln Y_{jt} + \beta_3 \ln Y_{it} + \beta_4 \ln {\text{POP}}_{it} + \beta_5 \ln {\text{POP}}_{it} } \\ { + \beta_6 \ln {\text{DIST}}_{ij} + \beta_7 \ln {\text{OPEN}}_{jt} + \beta_8 \ln {\text{PHONE}}_{jt} + \beta_9 {\text{ADJACENCY}}_j } \\ { + \beta_{10} {\text{NBORDER}}_i + \beta_{11} {\text{SBORDER}}_i + \beta_{12} {\text{WCOAST}}_i + \beta_{13} {\text{ECOAST}}_i + \varepsilon_{ijt} } \\ \end{array} $$
(1)

Xijt+3 represents state i exports to country j during year t + 3. The lag between independent variables and exports follows Co et al. (2004), which notes that Gould (1994) finds the effects of immigrants on trade begin roughly 3.8 years following arrival. Yji represents home country gross domestic product (GDP), a measure of economic mass, with higher values implying greater potential markets for domestic goods. Similarly, gross state product (Yit) measures a state’s capacity to export. The POP variables provide additional controls for home country and host state sizes. The geodesic distance between each state and home country pair (DISTij) is a proxy for transport costs. OPENjt, which is given as the sum of imports and exports divided by GDP, is a measure of general propensity to trade and can be considered to represent a country’s integration into the global economy. The number of home country landline telephone connections per 1,000 residents (PHONEjt) reflects infrastructure development. A series of dummy variables (ADJACENCYj, NBORDERi, SBORDERi, WCOASTi and ECOASTi) capture potentially unique trading relationships between all states and Canada and Mexico and relationships that may exist between border/coastal states and all other trading partners.4

Results of estimating Eq. 1 for the full sample and for both OECD member and non-OECD member sub-groups are presented in columns a–c of Table 1. Columns d–i reproduce the results presented in Co et al. (2004). t tests evaluating differences in coefficients on IMijt variables across regressions indicate equivalence across the sub-groups. However, before drawing conclusions regarding differential effects from the sub-group regressions, it is important to examine the assumption of homoskedasticity in variances across regression estimations. F statistics corresponding to regression results presented in columns b and c, e and f, and h and i are 3.85, 4.56, and 4.62, respectively. Thus, in all cases, the null hypothesis of homoskedasticity must be rejected.
Table 1

Replication of Co et al. (2004) specification and presentation of results from Co et al. (2004)

 

Co et al. (2004), Specifications

Co et al. (2004), Estimation results, ITA exporter location series

Co et al. (2004), Estimation results, MISER exporter location series

All Countries

Developed Countries

Developing Countries

All Countries

Developed Countries

Developing Countries

All Countries

Developed Countries

Developing Countries

ln Immigrantsijt

0.298a

0.267a

0.26a

0.2993a

0.2886a

0.2698a

0.2997a

0.2908a

0.2687a

(0.052)

(0.052)

(0.067)

(0.0491)

(0.0496)

(0.0648)

(0.0495)

(0.0497)

(0.0653)

ln Gross Domestic Productjt

0.487a

6.71a

2.053a

0.8676a

4.5495a

0.6911c

0.8688a

4.4527a

0.7212b

(0.185)

(0.736)

(0.262)

(0.2423)

(0.4184)

(0.3616)

(0.2449)

(0.4203)

(0.3663)

ln Gross State Productit

1.407a

0.872b

1.963a

1.3946a

0.8134b

1.8091a

1.3537a

0.8135b

1.7496a

(0.415)

(0.413)

(0.557)

(0.3933)

(0.4041)

(0.5306)

(0.3976)

(0.4069)

(0.5371)

ln Home Country Populationjt

0.516b

−6.965a

−0.74a

0.1241

−4.9864a

0.4267

0.1008

−4.8573a

0.3693

(0.21)

(0.884)

(0.242)

(0.2733)

(0.5689)

(0.3761)

(0.2760)

(0.5707)

(0.3802)

ln Host State Populationit

−0.011

0.262

−0.41

0.0233

0.2909

-0.2001

0.0729

0.2943

−0.1274

(0.404)

(0.412)

(0.537)

(0.3826)

(0.4042)

(0.5149)

(0.3869)

(0.4070)

(0.5217)

ln Distanceij

−0.854a

−0.972a

−1.55a

−0.9882a

−0.8492a

−1.1011a

−0.9916a

0.8353a

−1.0911a

(0.117)

(0.193)

(0.2)

(0.1384)

(0.2060)

(0.2120)

(0.1408)

(0.2101)

(0.2165)

ln Openjt

0.74a

−0.562a

0.843a

0.9569b

−1.4921a

1.0136a

0.9113a

−1.4547a

0.9309a

(0.104)

(0.192)

(0.132)

(0.1503)

(0.2686)

(0.2143)

(0.1516)

(0.2690)

(0.2161)

ln Phone Connectionsjt (per 1 k residents)

0.579a

0.483

0.348a

0.2664b

0.1316

0.6095a

0.2605c

0.1034

0.6035a

(0.113)

(0.351)

(0.111)

(0.1343)

(0.3371)

(0.1635)

(0.1357)

(0.3394)

(0.1655)

Adjacencyj

0.347c

−0.572

−1.503a

0.2698

−0.8480b

−0.1970

0.3267

−0.6464

−0.2137

(0.203)

(0.357)

(0.381)

(0.2352)

(0.3891)

(0.4242)

(0.2390)

(0.3968)

(0.4318)

Border-Northi

−0.222c

0.006

−0.286

−0.2256c

0.051

−0.3415c

−0.2236c

0.0492

−0.3384c

(0.134)

(0.126)

(0.179)

(0.1329)

(0.1173)

(0.1821)

(0.1341)

(0.1180)

(0.1838)

Border-Southi

−0.968a

−0.325c

−1.196a

−0.9675a

−0.3400

−1.2497a

−0.9218a

−0.3429b

−1.1826a

(0.197)

(0.17)

(0.255)

(0.1927)

(0.1574)

(0.2564)

(0.1998)

(0.1573)

(0.2667)

Coastal-Westi

0.308

0.759a

0.273

0.3003

0.7414a

0.1151

0.3004

0.7222a

0.1201

(0.251)

(0.19)

(0.326)

(0.2494)

(0.1741)

(0.3428)

(0.2518)

(0.1758)

(0.3460)

Coastal-Easti

0.162

−0.134

0.261b

0.1751c

−0.0906

0.3181b

0.1950c

−0.0925

0.3511b

(0.102)

(0.099)

(0.133)

(0.1027)

(0.0919)

(0.1407)

(0.1042)

(0.0920)

(0.1430)

Constant

−36.847a

−65.459a

−55.652a

−40.183a

−32.000a

−45.693a

−39.288a

−31.940a

−44.775a

(4.72)

(7.131)

(6.704)

(5.0225)

(6.3446)

(6.7698)

(5.0860)

(6.4081)

(6.8506)

Adjusted R2

0.7263

0.8514

0.6961

0.7306

0.8655

0.6873

0.7255

0.8658

0.6816

N

1,344

432

912

1,344

432

912

1,344

432

912

Heteroskedastic-consistent standard errors in parentheses. Countries classified as “developed” are OECD members while “developing” countries are not

ITA International Trade Administration, MISER Massachusetts Institute for Social and Economic Research

aSignificance from zero at 1% level

bSignificance from zero at 5% level

cSignificance from zero at 10% level

While heteroskedasticity does not bias coefficients, it does bias standard errors and creates problems regarding causal inference. Application of White’s heteroskedasticity-corrected covariance matrix compensates for heteroskedasticity within regressions; however, it does not correct for unequal variances across regressions. To address this problem, we alter Eq. 1 to include a term that interacts the IMijt variable with a vector of dummy variables, LDCj, which identifies countries as lesser developed. Specifically, LDCj contains NON-OECDj, MEDIUM HDIj, UPPER MIDDLE INCOMEj, LOWER MIDDLE INCOMEj, MIDDLE INCOMEj and LOW INCOMEj. OECD membership and HDI classification, being dichotomous variables, are treated as general development measures. The United Nations classifies countries as high human development if 0.8 < HDI, medium human development if 0.5 < HDI < 0.8, and low human development if HDI < 0.5. Of the 28 home countries included in the data, none are classified by the United Nations as low human development. World Bank classifications, based on per capita gross national income, provide an arguably more detailed development measure. The Bank (1997) classifies countries as low income if 1995 GNI per capita < $765, lower middle income if $765 < GNI per capita < $3,035, upper middle income if $3,035 < GNI per capita < $9,385, and high income if GNI per capita > $9,385. Eq. 2 presents the general estimation equation.5
$$ \begin{array}{*{20}c} {\ln X_{ijt + 3} = \alpha_0 + \beta_1 \ln {\text{IM}}_{ijt} + \beta_2 \left( {\ln {\text{IM}}_{ijt} \times {\text{LDC}}_j } \right) + \beta_3 \ln Y_{jt} + \beta_4 \ln Y_{it} + \beta_5 \ln {\text{POP}}_{jt} } \\ { + \beta_6 \ln {\text{POP}}_{it} + \beta_7 \ln {\text{DIST}}_{ij} + \beta_8 \ln {\text{OPEN}}_{jt} + \beta_9 \ln {\text{PHONE}}_{jt} + \beta_{10} {\text{ADJACENCY}}_j } \\ { + \beta_{11} {\text{NBORDER}}_i + \beta_{12} {\text{SBORDER}}_i + \beta_{13} {\text{WCOAST}}_i + \beta_{14} {\text{ECOAST}}_i + \beta_{\text{LDC}} {\text{LDC}}_j + \varepsilon_{ijt} } \\ \end{array} $$
(2)

Equation 2 maintains the lag structure employed in Co et al. (2004). This is done to allow for comparison of our estimation results to those of Co et al (2004) and to avoid the potential problem of endogeneity.6 Interpretation of the coefficients on the immigrant stock variables and the interaction terms is straightforward. Coefficients on the IMijt variables, \( \hat{\beta }_1 \), apply to both developed and developing countries, while coefficients on LDCj dummy variables, \( \hat{\beta }_{\text{LDC}} \), indicate differences across development classifications. As such, coefficients on interaction terms, \( \hat{\beta }_2 \), represent differences in slopes between developing country regression lines and developed country regression lines. Thus, \( \hat{\beta }_1 \) indicates the proportional change in the dependent variable (i.e., state-level exports), applicable to developed home countries, expected to result from a 1% increase in the number of immigrants from such countries. The sum of the coefficients on the immigrant stock variable and the interaction term, \( \hat{\beta }_1 + \hat{\beta }_2 \), is thus the proportional change in state-level exports expected to result from a 1% increase in the number of immigrants from developing countries. As is evident, this approach simplifies testing whether immigrant–export effects differ across development classifications as significant coefficients on interaction terms indicate significant differences across developed and developing country classifications.

Tables  2 and 3 present descriptive statistics. Exports to OECD members, high HDI, and high-income countries are greater than exports to non-OECD members, medium HDI, and middle and lower income countries, respectively. Only lower middle income countries have average immigrant stocks significantly different from the overall mean. Developed countries are less open to trade as are lower middle income countries; however, low-income countries are more open to trade than the average countries in the sample. Regarding the infrastructure proxy variable, OECD countries, high HDI countries, and high-income countries are significantly above the average level. Developing countries generally have larger populations and smaller economies as do medium HDI as compared to high HDI countries and lower income (low, lower middle, and upper middle income) countries relative to high-income countries.
Table 2

Descriptive statistics, full sample, and by OECD and UN HDI classifications

Variable/sample

All countries

Developed (OECD member)

Developing (non-OECD)

High HDI

Medium HDI

Exportsijt+3

211,131,284

447,356,072a

99,235,332a

410,058,456a

82,413,703a

(832,714,761)

(1,204,172,940)

(545,176,862)

(1,114,665,910)

(544,976,083)

t stat

3.80

3.86

3.71

4.34

Gross Domestic Productjt (in millions US$)

428,804.66

829,041.59a

239,218.75a

715,335.68a

243,402.24a

(596,491.73)

(700,790.14)

(424,540.56)

(679,843.65)

(447,037.69)

t stat

10.69

8.82

8.49

8.21

Gross State Productit (in millions PPP$)

116,189

(140,278.01)

Home Country Populationjt

98,630,332

47,260,200a

122,963,553b

43,083,209a

134,572,588a

(252,786,629)

(36,868,073)

(302,846,558)

(35,400,723)

(318,114,512)

t stat

7.22

2.00

7.86

2.74

Host State Populationit

5,134,396

(5,450,320)

Distanceij (in miles, great circle method)

4,121.66

4,340.24b

4,018.12

4,851.78a

3,649.22a

(2,248.98)

(1,504.76)

(2,520.26)

(1,775.06)

(2,393.41)

t stat

2.30

1.00

7.40

4.55

Immigrant Stockijt

9,685.15

8,333.78

10,325.27

8,134.35

10,688.61

(75,922.20)

(19,123.53)

(91,232.03)

(19,859.18)

(96,129.78)

t stat

0.60

0.17

0.69

0.25

Opennessjt

0.6458

0.4959a

0.7168a

0.6294

0.6564

(0.4142)

(0.2256)

(0.4617)

(0.4838)

(0.3621)

t stat

9.57

3.73

0.68

0.62

Number of Landline Phonesjt (per 1 k residents)

182.73

409.11a

75.49a

403.79a

39.68a

(188.77)

(94.31)

(111.85)

(91.11)

(34.76)

t stat

32.99

16.91

34.02

27.04

N

1,344

432

912

528

816

Mean values presented. Standard deviations in parentheses

aStatistical significance between sub-group values and the full sample at 1% level

bStatistical significance between sub-group values and the full sample at 5% level

Table 3

Descriptive statistics, full sample, and by World Bank income classifications

Variable/sample

All countries

High income

Upper middle income

Middle income

Lower middle income

Low income

Exportsijt+3

211,131,284

523,901,666a

222,884,000

100,586,584a

32,643,400a

52,041,743a

(832,714,761)

(1,269,469,820)

(1,007,770,000)

(612,099,236)

(84,655,300)

(182,541,914)

t stat

4.56

0.17

3.37

7.73

6.33

Gross Domestic Productjt (in millions US$)

428,804.66

916,057.51a

209,379a

116,242.18a

64,499.60a

508,446.68c

(596,491.73)

(696,012.62)

(176,473)

(139,142.4)

(73,200.1)

(648,418.32)

t stat

12.47

11.05

18.24

21.88

1.92

Gross State Productit (in millions PPP$)

116,189

(140,278.01)

Home Country Populationjt

98,630,332

51,373,663a

29,473,400a

21,105,357a

16,456,400a

342,530,833a

(252,786,629)

(37,169,941)

(30,472,600)

(23,589,848)

(17,052,200)

(468,104,124)

t stat

6.61

9.64

11.15

11.83

8.58

Host State Populationit

5,134,396

(5,450,320)

Distanceij (in miles, great circle method)

4,121.66

4,624.28a

4,165.13

3,500.49a

3,131.24a

4,900.89a

(2,248.98)

(1,915.69)

(1,876.43)

(2,065.23)

(2,074.87)

(2,635.91)

t stat

4.35

0.32

6.18

8.45

4.67

Immigrant Stockijt

9,685.15

8,741.82

22,277.80

11,108.82

4,903.80a

7,621.01

(75,922.20)

(19,897.09)

(172,241.00)

(104,038.14)

(21,968.00)

(33,592.99)

t stat

0.41

1.11

0.32

2.06

0.72

Opennessjt

0.6458

0.6914

0.6434

0.6024a

0.5797a

0.6861c

(0.4142)

(0.5526)

(0.4042)

(0.3330)

(0.2839)

(0.3569)

t stat

1.50

0.09

2.54

3.73

1.69

Number of Landline Phonesjt (per 1 k residents)

182.73

437.63a

229.30a

110.26a

44.12a

11.95a

(188.77)

(76.63)

(115.88)

(114.02)

(22.76)

(5.35)

t stat

39.42

5.13

10.70

26.33

33.10

N

1,344

384

240

672

432

288

Mean values presented. Standard deviations in parentheses

aStatistical significance between sub-group values and the full sample at 1% level

bStatistical significance between sub-group values and the full sample at 5% level

cStatistical significance between sub-group values and the full sample at 10% level

Estimation Results

Tables 4 and 5 present results obtained from the estimation of Eq. 2. Employing OECD membership as an indication of relative development, variation in immigrant–export links is reported in column a. Coefficient estimates for the immigrant stock variable (IMijt) and the interaction term (IMijt × NON-OECDj) imply that a 1% increase in the immigrant stock from an OECD country leads to an estimated increase in state-level exports to the home country of 0.117%, while a like increase in the immigrant stock from a non-OECD member country yields an 0.33% increase in state-level exports. Since the coefficient on the interaction variable is significant, we can say that, while immigrants from both sets of home countries are found to exert positive and statistically significant influences on state-level exports, the influence of immigrants from developing countries is of greater proportional magnitude relative to the corresponding influence of immigrants from developed countries.
Table 4

Estimation results (dependent variable: ln Exportsijt+3) OECD/non-OECD and high/medium HDI home country classifications

Variable/specification

(a)

(b)

ln Immigrantsijt

0.117b

0.141a

(0.054)

(0.052)

ln Immigrantsijt × Non-OECD Memberj

0.213a

(0.054)

ln Immigrantsijt × Medium HDIj

0.221a

(0.057)

ln Gross Domestic Productjt

1.707a

0.813a

(0.2)

(0.197)

ln Gross State Productit

1.545a

1.483a

(0.405)

(0.411)

ln Home Country Populationjt

−0.579a

0.282

(0.213)

(0.212)

ln Host State Populationit

−0.134

−0.077

(0.393)

(0.401)

ln Distanceij

−1.248a

−0.902a

(0.123)

(0.122)

ln Openjt

0.656a

0.821a

(0.097)

(0.106)

ln Phone Connectionsjt (per 1 k residents)

0.412a

0.778a

(0.106)

(0.135)

Adjacencyj

−0.47b

−0.026

(0.236)

(0.224)

Border-Northi

−0.196

−0.195

(0.128)

(0.13)

Border-Southi

−0.931a

−0.974a

(0.191)

(0.195)

Coastal-Westi

0.405c

0.34

(0.242)

(0.251)

Coastal-Easti

0.131

0.151

(0.098)

(0.102)

Developing Home Countryj (Non-OECD Member)

0.269

(0.391)

Medium HDI Home Countryj

−0.372

(0.465)

Constant

−46.768a

−42.072a

(4.797)

(5.323)

Adjusted R2

0.7440

0.7336

N

1,344

1,344

Dependent variable in all specifications is ln Exportsijt+3. Heteroskedastic consistent standard errors in parentheses

aSignificance from zero at 1% level

bSignificance from zero at 5% level

cSignificance from zero at 10% level

Table 5

Estimation results (dependent variable: ln Exportsijt+3) World Bank Income Classifications

Variable/specification

(c)

(d)

ln Immigrantsijt

0.084

0.054

(0.057)

(0.056)

ln Immigrantsijt × Middle Incomej

0.213a

(0.055)

ln Immigrantsijt × Low Incomej

0.483a

0.516a

(0.1)

(0.099)

ln Immigrantsijt × Upper Middle Incomej

0.311a

(0.084)

ln Immigrantsijt × Lower Middle Incomej

0.199a

(0.06)

ln Gross Domestic Productjt

0.405c

0.788a

(0.241)

(0.274)

ln Gross State Productit

1.429a

1.447a

(0.406)

(0.394)

ln Home Country Populationjt

0.55b

0.231

(0.235)

(0.262)

ln Host State Populationit

−0.045

−0.047

(0.396)

(0.387)

ln Distanceij

−0.784a

−0.688a

(0.127)

(0.13)

ln Openjt

0.828a

0.959a

(0.122)

(0.125)

ln Phone Connectionsjt (per 1 k residents)

0.201c

0.524a

(0.12)

(0.119)

Adjacencyj

0.567b

0.938a

(0.222)

(0.237)

Border-Northi

−0.197

−0.196

(0.125)

(0.124)

Border-Southi

−0.979a

−0.985a

(0.191)

(0.189)

Coastal-Westi

0.298

0.279

(0.241)

(0.239)

Coastal-Easti

0.151

0.169c

(0.1)

(0.099)

Middle Income Home Countryj

−2.289a

(0.458)

Low Income Home Countryj

−5.489a

−3.722a

(0.776)

(0.847)

Upper Middle Income Home Countryj

−2.632a

(0.592)

Lower Middle Income Home Countryj

−0.591

(0.588)

Constant

−31.837a

−39.588a

(4.959)

(4.97)

Adjusted R2

0.7518

0.7568

N

1,344

1,344

Dependent variable in all specifications is ln Exportsijt+3. Heteroskedastic consistent standard errors in parentheses

aSignificance from zero at 1% level

bSignificance from zero at 5% level

cSignificance from zero at 10% level

Before considering alternative development classifications, it is important to note that the coefficients on the remaining variables in Tables 4 and 5 generally conform to expectations. Higher transport costs, as represented by greater geodesic distances between home countries and states, reduce state-level exports. Higher home country GDP and higher GSP values correspond with greater exports. Thus, it appears that larger economies import more from US states, and states that produce more output tend to export more. Countries that have relatively more developed infrastructure, as represented by a higher number of phone connections per 1,000 residents, also tend to import more from US states. Similarly, those home countries that are relatively more open to trade tend to import more from US states. Coefficients on variables representing home country and host state population levels, adjacency, borders, and coasts produce mixed results across the estimations presented in columns a–d. While this is likely the result of the inclusion of the various development classification dummy variables, we note that the coefficients on the immigrant stock variables and interaction terms are largely consistent across estimations in terms of sign and statistical significance and interpret this as an indication of the robustness of our principle findings.

While OECD membership is a useful indicator of a country’s level of economic development, we also consider additional indicators of development. We begin with the estimation results presented in column b, where home countries are classified as being either at high or medium stages of human development based on the UN HDI. Similar to our previous specification, the dummy variable that identifies the lesser developed country cohort (LDCj)—in this case, MEDIUM HDIj—is interacted with the immigrant stock variable (IMijt). We find that a 1% increase in the immigrant stock from a medium HDI country yields an estimated 0.362% increase in state-level exports. A similar increase in the immigrant stock from a high HDI country increases exports by only 0.141%. As with the results obtained when classifying countries by OECD membership, the immigrant–export effect is positive and significant for both classifications. Given the high correlation (0.86) between classification as high HDI and OECD membership, these results are expected. Effectively, using medium HDI instead of non-OECD membership to identify developing countries reclassifies Hong Kong and South Korea as developed countries. Since the coefficient on the interaction term is significant, the results further support the existence of variation in immigrant–export effects across home country development classifications.

To provide a more in-depth examination of heterogeneity in immigrant–export effects, we employ World Bank income classifications to categorize the countries in our sample. Column c presents results when home countries are classified as high, middle, or low income. The coefficient on the immigrant stock variable is positive but insignificant. This suggests that immigrants from high-income countries typically do not exert discernable influences on state-level exports. For middle and low-income countries, however, we do find positive and significant immigrant–export effects. More specifically, in response to an assumed 1% increase in a state’s immigrant stock from the typical middle-income country, state-level exports to the respective home country are estimated to increase by 0.297%. A similar increase in the immigrant stock from the typical low-income country increases exports by 0.567%. We consider this to be compelling support for our notion that the opportunities available to immigrants, in terms of utilizing their connections to business and social networks as a means to overcome information asymmetries and, hence, to facilitate trade, are relatively more abundant for immigrants from lesser developed countries and significantly less prevalent for immigrants from developed countries.

Modifying the specification such that the middle-income classification is further decomposed into upper middle and lower middle income groupings permits a more detailed examination of the immigrant–export relationship. Corresponding estimation results are presented in column d. As in the previous regression (column c), the coefficient on the immigrant stock variable is positive but insignificant, again suggesting that immigrants from high-income countries exert no discernable effects on state-level exports. In response to an assumed 1% increase in the number of immigrants from a low-income country, state-level exports are estimated to increase by 0.57%. As one would expect for these two development classifications, the results are nearly identical to those reported in column c. The decomposition of the middle-income classification into upper middle and lower middle income groupings, however, reveals further variation in immigrant–export effects. Assume that 1% increases in the number of immigrants from upper middle and lower middle income countries are estimated to generate increases in state-level exports of 0.365% and 0.253%, respectively. The larger proportional immigrant–export effect for the upper middle income classification relative to the lower middle income classification may seem at odds with our hypothesis that a lack of development corresponds with greater information asymmetries and, thus, increased opportunities for immigrants to facilitate trade. However, given that the upper middle income classification consists of Greece, Mexico, Portugal, South Korea, and Trinidad and Tobago, it may well be that the inclusion of Mexico and/or South Korea is responsible for the larger coefficient estimate.

Results obtained when estimating each of the four regression specifications indicate that immigrants from lesser developed countries exert positive and significant influences on state-level exports. We also find that the magnitude of proportional immigrant–export effects is inversely related to the general level of development and that, in two of the four estimations, no significant immigrant–export effect is found for the most developed countries. This is consistent with the notion that immigrants from lesser developed countries face greater opportunities to exploit their connections to business and/or social networks as a means of overcoming trade-related information asymmetries. However, it is also important to note that, while the results presented in this study may indicate general findings or tendencies regarding the presence and relative magnitudes of immigrant–export effects, as is stated in Bandyopadhyay et al. (2008), networks may also be very important for state-level exports to developed countries, and the lack of statistical significance should be interpreted as an indication of no measurable immigrant–export effects rather than as confirmation that no immigrant–export effect exists.

While we have identified variation in proportional immigrant–export effects across the home country development classifications, immigrant stock and export levels also differ across states and regions. Thus, variation in absolute effects may also exist. Table 6 presents estimates of the average marginal effects of additional immigrants. These estimates are generated, at the state, region, and national levels, using coefficients presented in Tables 4 and 5 and state-level immigrant stock and export data. Regional and national mean values, for each home country classification, are derived as arithmetic means of the corresponding estimates of state-level immigrant effects. Estimates of state-level immigrant–export effects are arithmetic averages of values calculated across relevant state-home country pairs as \( E\left( {\Delta X_{ijt + 3} } \right) = \left( {\hat{\beta }_1 \times X_{ijt + 3} } \right)/{\text{IM}}_{ijt} \) for developed countries and for developing countries as \( E\left( {\Delta X_{ijt + 3} } \right) = \left[ {\left( {\hat{\beta }_1 + \hat{\beta }_2 } \right) \times X_{ijt + 3} } \right]/{\text{IM}}_{ijt} \).
Table 6

Estimated export effect of immigration, state and regional averages

Tables 4 and 5 specifications

(a)

(a)

(b)

(b)

(c)

(c)

(d)

(d)

(d)

Home country classification

Developed countries

Developing countries

High HDI

Medium HDI

Middle income

Low income

Upper middle income

Lower middle income

Low income

 

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

All states

10,517a

13,636a

13,310a

7,105a

8,785

8,615

14,814a,a

7,572a,a

9,203a,a

New England

5,264

11,392

8,735

8,079

7,252

12,003

14,033Ba

4,206a,b

12,066Bb

 Connecticut

8,206c

16,950c

13,398

11,966

13,295

19,164

28,983Bb

6,456b,b

19,164Bb

 Maine

4,810b

27,769b

14,489

17,855

10,875c

40,475c

15,147A,B

8,577Ac

40,475Bc

 Massachusetts

6,069

5,285

7,368

4,181

4,336

3,435

10,939a,b

1,533a,c

3,435b,c

 New Hampshire

4,092

7,530

5,142

7,393

7,510b

2,620b

9,193Ab

6,412Ac

2,620b,c

 Rhode Island

3,368

3,558

4,709c

2,056c

2,229

1,564

5,412a,b

870Ca

1,564Cb

 Vermont

5,040

7,260

7,305

5,021

5,270

5,139

14,524a,b

1,390Ca

5,139Cb

Middle Atlantic

6,416

6,366

7,459

5,923

5,762

3,895

7,775c,b

4,641Cc

3,916Cb

 New Jersey

5,249

5,423

6,395

4,623

4,844

2,453

7,381Ab

3,576A,C

2,453Cb

 New York

5,699

3,081

6,327c

2,601c

2,673

1,805

3,689Ac

2,121A,C

1,805Cc

 Pennsylvania

8,299

10,596

9,655

10,545

9,769

7,490

12,254A,B

8,226A,C

7,490B,C

East North Central

14,522

13,865

16,471

13,423

14,416a

7,679a

21,681a,b

10,754Cb

7,719Ca

 Illinois

10,027

5,404

10,909c

4,927c

5,082

6,139

6,603A,B

4,191A,C

6,139B,C

 Indiana

17,510

10,200

18,959

9,691

10,816a

4,329a

16,535a,b

7,964b,b

4,329a,b

 Michigan

9,684

16,892

11,812

16,951

19,486b

5,662b

43,556a,b

9,048Cb

5,662Ca

 Ohio

20,367

19,001

22,315

19,586

18,995

11,736

23,561A,B

16,097A,C

11,736B,C

 Wisconsin

15,022

17,826

18,362

15,957

17,703

10,731

18,149A,B

16,469A,C

10,731B,C

West North Central

17,811

13,902

18,852

14,901

22,945b

7,590b

37,513b,c

15,958b,c

7,630b,b

 Iowa

11,102

7,906

12,354

7,356

8,708b

4,479b

11,971Aa

6,929A,C

4,479Ca

 Kansas

14,104

33,133

14,623

39,432

38,352

8,824

4,913c,c

48,929c,c

8,824c,c

 Minnesota

59,623

26,602

61,501

28,109

83,842

24,832

217,756b,b

27,245Cb

24,832Cb

 Missouri

7,170b

15,360b

8,884

15,820

14,254

11,561

15,051A,B

13,093A,C

11,561B,C

 Nebraska

22,426

10,746

24,354

9,451

11,536b

1,368b

9,538Aa

11,613Ab

1,368a,b

 North Dakota

2,546

2,555

2,588

3,003

2,823c

1,188c

1,423Bc

3,193b,c

1,188Bb

 South Dakota

7,706a

1,013a

7,661a

1,137a

1,098

1,157

1,942Bb

707Cb

1,157B,C

South Atlantic

10,774b

18,814b

15,994

14,144

16,493b

9,281b

20,786Ab

13,850Ab

9,330b,b

 Delaware

24,657

45,750

30,383

46,005

45,162

19,178

26,033A,B

49,819Ab

19,178Bb

 Florida

1,797a

5,215a

2,618b

4,988b

4,488

5,004

3,610A,B

4,557A,C

5,004B,C

 Georgia

7,813

11,012

9,720

10,152

9,082

11,583

7,194A,B

9,264A,C

11,583B,C

 Maryland

2,402

1,328

2,673c

1,122c

1,242

814

2,576a,b

653Ca

814Cb

 North Carolina

11,286

19,942

14,574

18,724

15,213

22,027

8,543b,c

16,869Cb

22,027Cc

 South Carolina

9,792c

19,504c

13,770

17,078

17,354b

8,259b

17,661Ac

16,194Ac

8,259c,c

 Virginia

6,512

4,750

8,092b

3,046b

2,894

4,070

5,626Ba

1,668a,b

4,070Bb

 West Virginia

21,937

43,010

46,124c

12,042c

36,508c

3,708c

95,045b,b

11,776Cb

3,708Cb

East South Central

13,233a

24,574a

16,337c

24,664c

23,997b

15,121b

24,795Ab

22,250A,C

15,201Cb

 Alabama

11,084c

21,088c

13,927

20,874

18,502

17,816

18,750A,B

17,296A,C

17,816B,C

 Kentucky

19,146

13,413

22,656c

10,168c

13,757c

5,500c

32,621a,a

5,668Ca

5,500Ca

 Mississippi

6,816b

20,994b

7,815b

23,921b

18,913

25,972

6,590b,b

22,524Cb

25,972Cb

 Tennessee

15,886b

42,802b

20,951c

43,693c

44,816a

11,517a

41,220Aa

43,512Aa

11,517a,a

West South Central

9,744

14,418

11,974

13,746

14,657c

4,854c

10,084Aa

15,539Ab

4,880a,b

 Arkansas

8,756

11,992

14,883b

4,322b

5,371

3,951

9,214a,a

3,569Ca

3,951Ca

 Louisiana

15,626

6,708

16,343

6,670

7,955a

2,486a

8,335Aa

7,332Ab

2,486a,b

 Oklahoma

6,303

30,628

6,553

36,990

36,997

8,119

10,250Bc

45,078c,c

8,119Bc

 Texas

8,291

8,343

10,117

7,002

8,304

4,964

12,535b,b

6,176Cb

4,964Cb

Mountain

5,535

7,939

9,109a

3,668a

3,575

4,406

3,630A,B

3,340B,C

4,429B,C

 Arizona

7,379

6,679

11,119a

1,804a

2,119

3,582

4,412Bb

1,109b,b

3,582Bb

 Colorado

11,794

12,367

16,796b

6,577b

6,636

5,366

6,246A,B

6,388A,C

5,366B,C

 Idaho

14,434

15,042

22,108a

5,358a

4,450

10,277

5,140A,B

3,917A,C

10,277B,C

 Montana

1,452

4,179

1,742

4,608

4,685

1,056

1,316A,B

5,701Ac

1,056Bc

 Nevada

866

588

1,065a

371a

322b

884b

561A,B

211Aa

884a,b

 New Mexico

1,426

3,254

2,952

1,420

2,744b

454b

5,084a,b

1,679Cc

454Ca

 Utah

5,603

18,580

15,241

6,638

5,479

10,832

5,183A,B

5,265A,C

10,832B,C

 Wyoming

1,330c

2,827c

1,850

2,571

2,167

2,984

1,096b,c

2,450Cc

2,984Cb

Pacific

10,016

9,490

11,903

8,267

7,105c

13,299c

7,064Ac

6,695Ac

13,369c,c

 California

4,653

2,662

5,404a

1,908a

2,129

1,464

2,131A,B

2,001A,C

1,464B,C

 Oregon

8,215

10,183

10,225

8,955

9,426

7,715

9,005A,B

9,023A,C

7,715B,C

 Washington

17,180

15,626

20,081

13,938

9,759b

30,930b

10,054Ab

9,060Ab

30,930b,b

Mean values presented with standard deviations in parentheses. Values in italics indicate statistically significant differences in mean values. The letters A, B, and C in columns 7–9 indicate there is no significant differences in mean values between the upper middle income classification as compared to lower middle income classification, upper middle income classification as compared to the low-income classification, and the lower middle income classification as compared to the low-income classification, respectively.

aSignificance at 1% level

bSignificance at 5% level

cSignificance at 10% level

t tests of differences in mean values are calculated to determine whether average estimated immigrant–export effects vary across the home country development classifications. For example, the mean values (with significance noted) for the state of Connecticut, which are listed in columns 1 and 2 of Table 6, are 8,206 and 16,950 for the developed and developing country cohorts, respectively. The corresponding t test for a difference in mean values indicates that these values are statistically different when evaluated at the 10% level of significance. Similarly, for the state of Maryland, columns 7–9 report values of 2,576, 653, and 814, respectively, for the upper middle, lower middle and low-income country cohorts. Testing for difference in mean values, again employing a t test, reveals that the value presented for the upper middle income classification is higher than that of the lower middle and low-income classifications at the 1% and 5% levels of significance, respectively. However, the value presented for the lower middle income classification is not significantly different from the value presented for the low-income classification.

At the national level, we see significant variation in immigrant–export effects. Values presented in columns 1 and 2 indicate that the average export effect of immigrants from non-OECD (developing) countries ($13,636) is significantly more than the effect attributable to immigrants from developed countries ($10,517). Variation in effects, also at the national level, is reported in columns 3 and 4. However, in this case, immigrants from high HDI (developed) countries have a larger estimated average effect on exports ($13,310) than do immigrants from developing countries ($7,105). That immigrants from developing countries are estimated to contribute relatively more to state-level exports when employing coefficients from specification (a) and to contribute relatively less when coefficients from specification (b) underscores the difficulty in employing dichotomous development classifications.

Allowing for more detailed classification of home countries by relative level of development, columns 5 and 6 illustrate that, while no immigrant–export effect is found for high-income (developed) country immigrants, positive yet statistically similar average effects are found to result from immigrants from middle ($8,784) and low-income ($8,615) home countries. Segmenting the middle-income home country classification into upper middle and lower middle income classifications, columns 7–9 indicates variation in effects across development classifications. On average, immigrants from upper middle income home countries increase state-level exports by $14,814, while immigrants from lower middle and low-income home countries increase exports by an average of $7,572 and $9,203, respectively. The remainder of Table 6 indicates that, when comparing across regions and states and considering the more-detailed classification of countries, greater variation in effects is found to exist.

Discussion of Results and Concluding Thoughts

The results of the econometric analysis indicate clear and statistically significant differences, across a series of home country development classifications, in both the proportional and the absolute effects of immigrants on US state-level exports. This corrects the finding reported in Co et al. (2004) of there being no significant difference in the immigrant–export relationship across development classifications. Additionally, the results emphasize the importance of how development is measured. While OECD membership may be a useful measure of relative development, the associated dichotomous characterization is less than ideal when one attempts to discern between group-specific effects. Use of HDI classifications as an alternative measure yields similar results and, unfortunately, suffers the same shortcoming. World Bank income classifications are more detailed and, thus, permit more in-depth examination; however, it is important to bear in mind that, while average income is correlated with non-monetary measures of development, it too is an imperfect metric. Regardless of the shortcomings of each development classification scheme, the consistency of results across all three development classifications and the four estimation equations suggests that variation in immigrant–export effects does indeed exist.

A separate finding is that immigrants from developing countries typically increase state-level exports to their respective home countries proportionally more so than do immigrants from relatively developed countries. As immigrant populations and trade patterns vary across states and regions, estimates of the absolute effects of an additional immigrant are generated. Again, significant variation is reported across geographic locales, with the average immigrant from a developing country found to exert a greater absolute influence on state-level exports when OECD membership and World Bank income classifications are used to categorize countries as developing or developed. The average immigrant from a developed country is found to exert a greater absolute influence on state-level exports when countries are classified as developing or developed based on the UN Human Development Index.

As mentioned at the outset, lesser developed countries may be characterized, relative to developed countries, as more likely lacking in terms of having established, well-functioning institutions and/or formal channels that serve to facilitate trade deals. The results presented in this paper support the notion that immigrants from developing countries may face relatively greater opportunities to utilize their connections to business or social networks as a means by which trade-related information asymmetries may be ameliorated. The importance of creating and implementing an optimal immigration policy, in terms of quelling public discontent and garnering the greatest possible gains from immigration, requires an accurate accounting of all effects of immigrants on their host countries. The findings reported in this paper inform policy on an issue that is of particular public interest; however, given the importance of immigration as a public policy issue and as the analysis is undertaken using a limited number of immigrant source countries and a single year of data, this should be considered a first step of sorts in the sense that further research, using a larger data sample and a broader array of specifications, is merited.

Footnotes
1

Appendix A lists the 28 countries included in the data set. Collectively, these countries accounted for 67.4% of US exports in 1993 (Feenstra, 1997) and were home countries for 60.2% of the 1990 US foreign-born population (US Bureau of the Census 1993).

 
2

The effect of immigrants on net exports, while an interesting issue that is particularly important for public policy, cannot be examined as import data is not available at the state level. Studies examining the immigrant–trade link using aggregate trade data, however, provide mixed results as to whether immigrants’ influences generate increases or decreases in net exports.

 
3

See Anderson and Van Wincoop (2003), Eaton and Kortum (2002), Feenstra et al. (2001), Deardorff (1998), Davis (1995), Bergstrand (1985), Helpman and Krugman (1985), and Anderson (1979) for theoretical foundations.

 
4

Appendix B lists data sources and discusses corresponding variable construction.

 
5

An F test, comparing adjusted R-squared values, indicates that in all cases estimation of Eq. 2 explains a significantly larger share of the variation in exports than does estimation of Eq. 1.

 
6

Results from estimation of ancillary regression specifications, where only the immigrant stock variable is lagged, correspond to the conclusions presented in “Estimation Results”. All estimation results and data are available upon request.

 

Copyright information

© Springer Science+Business Media B.V. 2009