Master Thesis
International Business & Management
The influence of subnational cultural
variation on regional FDI location choice
Timo Dolfing
1531301
2
Table of content
INTRODUCTION ... 3
THEORY AND LITERATURE REVIEW ... 6
METHODS AND DATA ... 11
RESULTS ... 20
DISCUSSION ... 23
CONCLUSION, LIMITATIONS AND RECOMMENDATIONS FOR FUTURE RESEARCH ... 26
REFERENCE LIST ... 28
3
Introduction
The construct of cultural distance (CD) has since long maintained a strong position in International Business (IB). Presumably measuring the extent to which different cultures are similar or different, it provided an intuitively attractive explanation for differences in firm internationalization strategies (Shenkar, 2001). With the start of the new millennium, CD was increasingly criticized with respect to several aspects. One of these aspects was the assumption of spatial homogeneity that is intrinsic to CD. Assigning a single, mean index number for the cultural configuration in a country was argued to ignore the cultural variation within a country (Shenkar, 2001). In response to this critique a new strand of literature developed, promoting the use of intracultural variation (ICV) as a measure of CD instead of using a mean figure of culture as initially proposed by Kogut & Singh (1988) and then followed by many other studies (Au, 2000; Au & Cheung, 2004; Beugelsdijk, Van Hoorn, Maseland, Onrust, & Slangen, 2012; Kogut & Singh, 1988; Tung, 2008; Tung & Verbeke, 2010).
As an argument for using ICV as a measure of CD it is put forward that cultural variation within a country can be as large as and sometimes even larger than cultural variation among countries (Tung & Baumann, 2009). In addition, it is demonstrated that ICV can explain cultural differences as much as, if not more than, cultural means (Au, 2000; Au & Cheung, 2004; Beugelsdijk et al., 2012). As one of several implications of this
observation, the cultural variation on a subnational level may open a set of
opportunities to firms engaged in international business. MNESs may target a market in the host country that has the same ethnic background and in this way minimize the cultural distance (Beugelsdijk et al., 2012).
This paper empirically investigates the theoretical arguments put forward in the ICV debate and tries to push it one step ahead by looking at the actual behavior of firms to see if this matches the theory. The arguments made by Beugelsdijk et al. (2012) and Tung and Verbeke (2010) are integrated with the theory of regional FDI location choice to see whether firms are indeed driven by cultural factors in the selection of a region for FDI on a subnational level.
Therefore, the research question of this study is formulated as follows:
4 An attempt in answering this question is made by looking at the actual investment behavior of foreign multinationals in the United States regarding location choice. State-level data about the location of U.S. affiliates of foreign multinationals is obtained from the U.S. Bureau of Economic Analysis (BEA) in conjunction with cultural values data obtained from the General Social Survey (GSS) and World Values Survey (WVS) are analyzed. A subnational cultural variation (SCV) variable is constructed based on three matched survey items from both the GSS and WVS. Consequently, a linear regression is performed to see whether the variation in the number of U.S. affiliates in each state can be significantly explained by this SCV variable, controlling for both home country specific and region specific effects. Mixed evidence is found on the variation of
preferences for locating affiliates in different subnational regions among different MNE home countries. Furthermore, it is shown that SCV has a significant effect only when left uncontrolled or only controlling for region specific effects. It loses it significance when controlling for both region specific and home country specific effects.
The contribution to the literature by this study is twofold. This study is among the first to investigate the cultural variation and influence on foreign direct investment (FDI) distribution on a regional scale. Previous studies concerned with FDI patterns in relationship to cultural attributes have mainly focused on a national or cross-national level (Barkema, Bell & Pennings, 1996; Barkema, Shenkar, Vermeulen, & Bell, 1997; Drogendijk & Slangen, 2006; Tihanyi, Griffith, & Russell, 2005). Indeed, irrespective of whether these studies exploit some form of the Kogut and Singh (KS) measure, a Schwartz measure or some different measure, they are all measuring values for a national cultural system.
6
Theory and literature review
History of cultural distance in IB literature
The primary characteristic that distinguishes multinational firms from firms that do business domestically is the ability of the former to coordinate the activities of units or firms dispersed across countries. This requires them to deal with time zones, miles and long-distance communication, as well as kinds of country differences that contribute to other forms of distance, such as economic, administrative and, subject of investigation in the current study, cultural distance (Ghemawat, 2001; Zaheer, Schomaker, & Nachum, 2012). This intrinsic feature of international business have led scholars to conclude that foreign MNEs suffer from a ‘liability of foreignness’, being broadly defined as ‘all
additional costs a firm operating in a market overseas incurs that a local firm would not incur’ (Zaheer, 1995: 343).
From the late 1970s onwards, a specific stream of literature in IB has focused on the consequences of and solutions for the perceived cultural differences across nations in IB. In their internationalization theory of multinational activity, researchers of the
university of Uppsala proposed that differences between countries with regard to, for example, language and culture, constitute an important obstacle to decision making connected with the development of foreign operations (Johanson & Vahlne, 1977). They labeled these differences between countries as ‘psychic distance’, being defined as ‘all factors preventing or disturbing the flows of information between firm and market’ (Johanson & Wiedersheim-Paul, 1975: 308). Because of psychic distance, firms face difficulty in obtaining market knowledge in international operations (Johanson & Vahlne, 1977).
7 opportunistic, moral hazardous behavior on the side of the agent (Jensen & Meckling, 1976). Cultural distance is thought to aggravate both agency and transaction cost problems.
The assumption of homogeneity and ICV
The CD-concept is widely accepted in IB research, as it offers a seemingly parsimonious, quantitative expression of the intangible and complex phenomenon of culture (Luo & Shenkar, 2011). In combination with the introduction of the Kogut & Singh measure, culture became an phenomenon that was comprehensible and quantifiable (Kogut & Singh, 1988). Substantial critique on this simplification and omission of other factors was provided by Shenkar (2001) in an award winning essay, stressing that the concept in its current form omitted many aspects. As one among many points of criticism, Shenkar (2001) argued that the CD index assumes cultural uniformity within the national unit.
As a response to this criticism, several scholars proposed to replace to mean-based cultural measures by variance-based measures in an attempt to account for the
perceived cultural variation, or ICV, within a country (Au & Cheung, 2004; Beugelsdijk et al., 2012; Tung, 2008; Tung & Baumann, 2009; Tung & Verbeke 2010). Indeed, studies have shown that ICV varies between countries, that it can be substantial and that mean differences in cultural values are not more important than ICV in explaining cross-cultural differences among individuals (Au, 2000; Tung, 2008). It is shown that the cultural means for two countries can be very different, while their ICV values are more or less the same (Au, 2000). Whereas the concept of a national culture appears to be of importance (Inglehart & Baker, 2000; Minkov & Hofstede, 2012), cultures in large countries such as China, India, Russia are very diverse. But also smaller countries that consist of a very heterogeneous ethnicity or institutional make-up, such as Belgium, Surinam or Spain inhibit substantial cultural heterogeneity. In addition, it is argued that nation-states are becoming increasingly diverse due to lower barriers to the movement of people and the growing boundarylessness of the workforce (Tung, 2008).
8 different subnational regions. Finally, it presupposes that these varying choices are substantially driven by cultural regional differences. The current study will investigate these three assumptions in the context of regional FDI location choice.
Subnational regional differences matter
What factors exactly influence FDI location choice on a regional level has been studied intensively by the academic fields of economic geography and regional science. During the late 1980’s and early 1990’s multiple studies have been published investigating the factors driving the location decisions of inward FDI into the United States (Coughlin, Terza, & Arromdee, 1991; Friedman, Gerlowski, & Silberman, 1992; Luger & Shetty, 1985; Woodward, 1992). Coughlin et al. (1991) used a conditional logit model of the location decisions of foreign firms investing in manufacturing facilities in the US between 1981 and 1983. They found the size of the state land area, state per capita income, manufacturing density, local wage rates, unemployment rates, state government expediture on investment promotion, the rates of unionization and the scale of
transportation infrastructure to be of significant influence on location decisions (Coughlin et al., 1991).
Friedman et al. (1992) replicated the study by Coughlin et al. (1991) using data on the establishment of specifically new plants. They reviewed their findings parallel to those of three similar earlier studies and determined consensus on four factors: market size, manufacturing wage rate, transportation infrastructure and state promotional activities to attract foreign investment. Based on their findings, the authors suggested to
incorporate per capita state and local tax receipts, as it exerted a negative influence on location choice (Friedman et al., 1992). Unionization rate was again found to have a positive significant effect.
The established location decision factors also appeared robust to an empirical setting outside the United States. Testing them for the developing setting of Chinese regions, Sun, Tong & Yu (2002) largely confirmed the earlier results. Moreover, the authors showed the determinants of FDI to be dynamic over time, showing a changing influence of the wage rate on attracting FDI from positive to negative. Similarly, provincial GDP (market size) proved insignificant before 1991, but highly significant after (Sun et al., 2002). It is suggested that controlling for labor quality could reduce the impact of labor cost on location decisions (Cheng & Stough, 2006).
9 and therefore increases the attractiveness of a certain area (Krugman, 1990).
Theoretically, this increased attractiveness or agglomeration effect can be broadly explained by three mechanisms: inter-firm technological spill-over, specialized labour, and intermediate inputs (Head, Ries, & Swenson, 1995; Head, Ries, & Ruckman, 1998). Physical proximity to firms in the same industry may enhance knowledge flows by making causal communication less costly (Head et al., 1995). In addition, an
agglomeration creates a demand for knowledge institutes such as universities, fostering the rate innovation and supply of specialized labour in the area. Indeed, localized industry creates a pool for workers with specialized skills by mitigating the hold-up problems associated with the development of these skills (Head et al., 1995). Also, lower transportation costs and economies of scale induce firms looking for an
investment location to seect those locations with a level of yet already a lot of economic activity (Krugman, 1990). Taken together, these effect could initiate a self-reinforcing trend, or agglomerating effect, with more firms being drawn to a specific region, in turn increasing the attractiveness of this region.
Based on the theoretical and empirical contributions presented above the following hypothesis is formulated:
H1: Subnational regional differences matter for regional FDI location decisions. Varying home countries and varying location preferences
Hypothesis H1 deals with a number of generic factors determining the attractiveness of a region for foreign MNEs for locating their affiliates. What distinguishes the current study from previous studies investigating FDI location choice decisions is its attempt to determine varying preferences in location choice for MNEs from different home
countries. A first argument for differences in investment location behaviour would be the difference in geographic distance between the MNE home countries and the investment locations. Ghemawat (2001) has emphasized that geographic distance still has a significantly dampening effect on both investment flows as well as trade flows between countries. This argument was empirically verified by Nachum, Zaheer & Gross (2008), showing that the proximity of a country to the rest of the world has a positive impact on MNEs choosing that country as a location. In this study, proximity is interpreted as a function of both geographic distance and the worldwide spatial distribution of these factors (Nachum et al., 2008).
10 on a regional level, with Japanese investors setting up new manufacturing plants in China. This study showed that investments generally are located close to previous subsidiaries. New investments of MNEs from a specific home country tend to be unevenly drawn to regions in which more affiliates from this particular country are situated. It has been explained that prior firms’ investments provide information that help reduce search costs in factually opaque environments (Zhou, Delios & Yang, 2002). The influence of geographic distance and home country focused agglomeration effects on investment location choice lead to the following hypothesis:
H2: MNEs from different home countries have varying preferences for different subnational regions.
Subnational cultural variation as a location determinant for FDI
From an IB perspective, the critique on the assumption of spatial homogeneity in CD and the suggested alternative of integrating ICV, hereafter called subnational cultural
11
Methods and Data
Data: foreign MNE affiliates in U.S.
The hypotheses of this study will be tested by investigating the locations of U.S. affiliates of foreign multinationals. A dataset with data on these foreign affiliates is provided by the American Bureau of Economic Affairs (BEA), specifying the number of foreign owned establishments of MNEs per state in 2002. An establishment is considered to be foreign owned when owned 10 percent or more by a foreign person. Most of these affiliates are majority owned, i.e. owned more than 50 percent by foreign direct investors, ensuring that the analysis mostly deals with direct investments instead of portfolio investments. According to the BEA, majority owned affiliates accounted for 92 percent of all employment by U.S. affiliates in 20021. The MNE home countries that are
specified in the dataset and for which the location patterns will be analyzed are Canada, France, Germany, the Netherlands, Switzerland, the United Kingdom and Japan. Affiliates from Latin America, Africa and the Middle East are not specified on a country basis and are therefore considered too aggregate for proper analysis.
Hypothesis 1
For hypothesis 1 the distribution of the affiliates over the 51 specified states will be analyzed. A linear regression analysis will be performed, using the count data of the total number of affiliates in every state as endogenous variable and adding a dummy variable for every state. As a robustness check, the regression analysis is repeated after the affiliate count data being corrected for population size of the state, to show that the variation in the number of affiliates per state is not merely a consequence of population size. The data for the population comprises the State Population Estimates measured from the first of April 2000 until the first of July 2002 and is retrieved from the Population Division of the U.S. Census Bureau.
Hypothesis 2
Hypothesis 2 will be analyzed using a Pearson Chi-Square test. This test compares the separate distributions of the affiliates per home country over the 51 states. When these distributions deviate significantly from each other, this is considered to be evidence that indeed different home countries have varying preferences for subnational regions.
12 Hypothesis 3
Finally, hypothesis 3 is analyzed using a linear regression method, including the number of affiliates from a specific home country in a specific region as dependent variable and including a newly constructed measure of subnational cultural variation (SCV) as the independent variable. The regression controls for both home country specific as well as region specific effects.
Independent Variable: Subnational cultural variation
A variable capturing the SCV from state to state will serve as the independent variable of interest in the analysis. Based on the theory described in the previous section, this culture variable should appear significant in explaining the number of affiliates from a certain home country after controlling for standard location decision variables. The variable is constructed from three survey questions of the General Social Survey (GSS)2.
The GSS is a nation-wide U.S. social survey monitoring social change in the American society and covers a range of demographic and attitudinal questions. It had its first wave in 1972 and completed its 28th round in 2010 . The origin of the different respondents is
categorized by 9 census regions by which the United States are subdivided according to
2 Source: http://www3.norc.org/gss+website
13 the US Census Bureau. The geographical classification of states by the various census regions is shown in figure 1. These 9 census regions will serve as the unit of analysis for this study.
From the GSS, three survey items or questions are selected that are compared with three equivalent survey questions asked to respondents from the home country of the MNE. The data for the three questions asked to respondents of the home country are obtained from the database of the World Value Survey (WVS). The WVS is a worldwide network of social scientists studying changing values and their impact on social and political life3.
From 1981 to 200, the WVS has carried out five waves of representative national surveys in 97 societies covering almost 90 percent of the world’s population.
Selection of the SCV variable items
The selection of the three SCV variable items was based on earlier work by Beugelsdijk (2012) and Au (2000), two studies with an emphasis on the usefulness of ICV as a methodological construct. To keep on working in line with these studies, all WVS items used by these studies to demonstrate the existence of ICV were evaluated for use in the current study. Table 1 presented below provides an overview of the three variable items that eventually were selected plus a variable code by which they will be referred to in the text. These three variables were selected because an equivalent item was found in the GSS database that matched the meaning of the WVS as close as possible. Another criterion for selection was the availability of a Likert scale, simplifying the computation of a single unambiguous value for SCV.
LUCK describes the degree in which people feel they have control over their lives as opposed to regarding it as a matter of luck or faith. For WVS, this question yielded a 10 point Likert scale. For GSS, it yielded a 4 point Likert scale. HARDWORK describes the degree in which people think that hard work can get them ahead in life. For WVS, this question could be answered by a 10 point Likert scale ranging from complete agreement with one statement to complete agreement with the opposite statement. For GSS, this question could be answered by a 5 point Likert scale, ranging from essential to not important at all. Finally, GOVERNMENT asks for the respondents’ opinion about the role of the government in providing for people, particularly the redistribution of income. For WVS, this question contained a 10 point Likert scale, ranging from complete agreement with one side of the statement to complete agreement with the other side of the
14 statement. For GSS, this question contained a 5 point Likert scale. The use of Likert scales poses a methodological challenge. All different survey items that are taken from WVS and GSS and are matched to each other use different Likert scales. All items from WVS use a ten point Likert scale, whereas the different items from GSS use a four or five point Likert scale, complicating the comparison between the two items. To circumvent this problem, all mean scores for both the MNE home country and the US host state are converted to a standardized z-score.
Table 1: Survey items for SCV
Question WVS Question GSS
LUCK
Some people feel they have complete free choice and control over their lives, while other people feel that what they do has no real effect on what happens to them. Please use this scale where 1 means "none at all" and 10 means "a great deal" to indicate how much freedom of choice and control you feel you have over the way your life turns out.
I'm going to read some statements that give reasons why a person's life turns out well or poorly. As I read each one, tell me whether you think it is very important, important, somewhat
important, or not at all important for how somebody's life turns out?:“Its just a matter of chance.”
HARDWORK
In the long run, hard work usually brings a better life vs. hard work doesn’t generally bring success – it’s more a matter of luck and connections.
Hard work – how important is that for getting ahead in life?
GOVERNMENT
The government should take more responsibility to ensure that everyone is provided for vs. people should take more
responsibility to provide for themselves.
Some people think that the government in Washington should do everything possible to improve the standard of living of all poor Americans; they are at Point 1 on this card. Other people think it is not the
15 In sum, SCV variable describing the cultural distance between the home region and the separate states in the U.S. is constructed as follows:
ܵܥܸ ൌ หܺതሺܩܱܸሻ െ ܺത൫ܩܱܸ൯หଶ หܺതሺܮܷܥܭሻ െ ܺത൫ܮܷܥܭ൯หଶ หܺതሺܪܣܴܦܹܱܴܭሻ െ ܺത൫ܪܣܴܦܹܱܴܭ൯หଶ
As the equation reveals, all absolute differences between the value scores of the home country of the MNE and the respective states are calculated. In addition, each term is squared to emphasize this term once the difference between home country score and state score becomes larger.
Constructing the SCV items for the 9 US Census regions
The GSS equivalents for LUCK, GOVERNMENT and HARDWORK, coded respectively LFECHNE, EQWLTH and OPHRDWRK in the GSS database, have to be retrieved from different waves of the GSS survey. Therefore, averages for these three items are first calculated separately. Starting with GOVERNMENT, this question has been asked in all waves present in the GSS. Whereas it seems appropriate to select the year of interest, 2002, this particular year appears to have a large number of missing or invalid
observations with 1865 missing out of a total of 2765. It is therefore decided to switch to the 2000 wave for this item, which fares better in this respect with 968 missing out of 2817 observations. Dropping the missing observations from the dataset leaves 1849 valid cases. Inspection the distribution of observations over the regions reveals a minimum of 95 valid cases for New England and a maximum of 331 cases for South Atlantic. The LUCK item was asked in a wave held from 1993 to 1996. Dropping the missing observations reduces the number of cases with 41 to 1564. Again, New England appears to have the lowest number of respondents with 56 valid cases. HARDWORK was asked in a wave held from 1983 to 1987. Dropping the missing observations for
16 For GOVERNMENT, a figure with a minus represents a relatively egalitarian attitude towards income distribution by the government, whereas a positive figure demarcates a disapproving attitude towards income redistribution by the government. What stands out from this table is the relatively large positive number of the West South Central region, including Texas, showing a widely rejecting attitude towards income
distribution. This should not come as a surprise however, given the rejecting attitude in these states towards the central government in Washington or any government in general. Similarly, confirming expectations, the more liberal Middle Atlantic region has a more approving stance towards government income redistribution.
Table 2: SCV scores for 9 Census Regions
The LUCK item may be somewhat more difficult to explain from a sociological perspective. For this item, a negative score depicts a relative acceptance of faith and chance in life whereas a positive score depicts a rejection of faith and chance. What stands out from this stable is the relative large positive figure for the Mountain region and the cluster of negative extremes for the regions of New England, Middle Atlantic and East North Central. Interestingly, this last cluster again coincides with the more liberal region in the United States. Finally, for HARDWORK a negative score represents an attitude in which hard work is important for getting ahead in life, whereas a negative sign suggests that generally people think of hard work as being less important for getting ahead in life. What strikes from this item is the low range of scores, showing a relative consensus in this matter. Looking at the unstandardized score of 1.76 for this item for the United States on average, it is clear that most Americans think of hard work as being important for getting one ahead in life.
GOVERNMENT LUCK HARDWORK
New England -0.02448 -0.15117 0.02550
Middle Atlantic -0.19391 -0.06280 -0.06247
East North Central 0.07450 -0.06222 0.11006
West North Central -0.01510 -0.00042 0.01182
South Atlantic 0.06013 0.07291 -0.05834
East South Central -0.07452 0.07115 -0.01001
West South Central 0.13968 -0.02270 0.00411
Mountain -0.08553 0.16896 -0.00235
17 Constructing the CD items for the 7 host countries
As previously explained, the data from the World Values Survey is collected and divided into five waves: 1981 – 1984, 1989 – 1993, 1994 – 1999, 1999 – 2004 and 2005 – 2008. Each wave consists of a different set of countries investigated. Therefore, the data for WVS-equivalent for GOVERNMENT and HARDWORK, coded respectively e037 and e040, were retrieved from two different waves. For GOVERNMENT, data for all countries is obtained from the fourth wave, except for Switzerland, for which the data was obtained from the third wave. For HARDWORK, the data for all countries was obtained from the second wave, except for again Switzerland, for which the data was obtained from the third wave. For LUCK, coded a173 in WVS, all data could be retrieved from wave 2 completely.
Before any statistical analysis can be performed, the data for the GOVERNMENT item have to be rescaled. Indeed, for the GOVERNMENT item in the GSS survey, a low score represents a positive stance towards income redistribution whereas a high sore represents a negative stance towards income redistribution. This item is marked in an opposite way in the WVS and can therefore not be compared when left unchanged. Therefore, all WVS cases are inverted by subtracting their score from 11. This way, each score receives its exact opposite in the 10 point spectrum.
Inspection of the number of observations after dropping the missing values yields satisfying results for all three items. GOVERNMENT retains a minimum of 981
observations for Great Britain and a maximum of 1993 observations for Germany, LUCK retains a minimum of 899 observations for Japan and a maximum of 3370 for Germany and finally HARDWORK retains a minimum of 873 observations for Japan and a
maximum of 3272 for Germany. This is assumed to be a sufficiently large sample size to provide a meaningful and representative result.
18
Table 3: SCV scores 7 home countries
Standing out from table 3 presented above is the only negative figure for GOVERNMENT of Japan. Similarly for LUCK, Japan appears to operate on the extreme sides of both items, with a relatively high tolerance for income redistribution by the government and the highest acceptance of luck and faith in life from all 7 host countries. Another striking element is the highly rejecting attitude towards governmental income redistribution by Switzerland and, to a lesser extent, France. Both countries produce a higher score on this item than the United States. For Switzerland, the relatively high score could be linked to the decentralized nature of the country and an its averse attitude towards centralized government. What strikes for LUCK, besides the negative score of Japan discussed earlier, are also the negative scores for both the Netherlands and France. Especially the contrast with the other West-European countries Great Britain and Germany is remarkable in this respect. Finally, the highest negative score for the United States for HARDWORK should not come as a surprise, with their strong cultural belief in self-made success. What stands out on the other hand is again the diverging position of the various West-European countries, with a remarkable positive score for the
Netherlands.
As a final step in constructing the SCV variable, the absolute difference was calculated separately for each item on a host country-by-census region basis and then squared. Then the three item differences were summed for each census region, as was previously demonstrated by the shown equation. The results of this exercise are presented in tables A1 – A4, included in the appendix.
Pooling of data and control variables
Due to the relatively low number of 63 observations, i.e. 7 home countries times 9
GOVERNMENT LUCK HARDWORK
19 census regions, it is decided to create a pooled data set and not to perform multiple regressions for each home country separately. The pooled dataset contains the number of affiliates from a particular home country in a specific region as a dependent variable. The SCV value for the home country with respect to this specific region is used as the independent variable. For the same reason of the limited number of observations, it is chosen not to include separate control variables for factors such as GDP, wage rates or infrastructure, as the model would run the risk of overestimation. Instead, two generic control variables are included capturing the home country specific effects and region specific effects or agglomeration effects. The home country specific effects are estimated by using the total number of affiliates in the United States for a particular home country. The region specific effects are estimated by the total number of affiliates for all home countries in a particular region.
20
Results
Hypothesis 1
The results of the regression analysis are displayed in the first column of the table A5, exhibited in the appendix, depicting the various coefficients for the state dummy variables. What can be readily noticed from this table is the large variance in betas for the different states, showing the large variance in the number of affiliates per state. However, densely populated states such as California, Florida, Illinois, New York and Texas have a relatively large positive coefficient, whereas less densely populated states such as Alaska, Montana, Nebraska or Wyoming have a relatively large negative
coefficient. This drives the suspicion that the coefficient is primarily determined by state size in terms of population.
Therefore, a second regression analysis is performed as a robustness check, in which the number of affiliates is corrected for population size. The results of this analysis are shown in the second column of the table below. Whereas the variation in betas is maintained, the correction for population size substantially alters the results. Densely populated states such as New York and California now display a coefficient much smaller than in the first regression, albeit still positive for both. Some less densely populated states which had a large negative coefficient in the first regression, such as Delaware, Vermont or Maine now have switched signs. However, a substantial number of the states kept its negative sign after controlling for population size, indicating a more fundamental tendency for attracting less affiliates. Indeed, for states such as Mississippi or Oklahoma the negative coefficient even aggravated in the second exercise.
Hypothesis 2 Descriptive analysis
21 instead of the distribution over the 51 states as this is thought to provide a clearer picture. In addition, an average is included in the diagram, depicting the total number of affiliates in the region as a share of the total number of US affiliates as a benchmark figure.
The seven spider web diagrams clearly reveal identical behavior in the location pattern of foreign MNEs within the US, as all seven countries are broadly in line with the
diagram depicting the general distribution of all affiliates over the US. What strikes from the diagrams with regard to this general location pattern is the spike for the South Atlantic region, apparently being an attractive region for most MNEs to locate their affiliates. With respect to the deviation from the general pattern, the diagrams of Canada and Japan stand out quite clearly. For Canada, the presence in of its MNE’s affiliates in both New England and East North Central is substantially above average, whereas its presence in Middle Atlantic is below average. Japanese MNEs have a clear preference of locating their affiliates in the Pacific region, whereas New England and Middle Atlantic are preferred less.
Pearson Chi-square test
To formally determine the actual variation in investment location behavior among different MNE home countries, a Pearson Chi-square test was performed for the count data of the MNE affiliates for all 51 states. The results of this test are summarized in table 4 below. The result of this test is highly significant on all levels, proving that the distribution of affiliates over the different states indeed varies between different home countries.
Table 4: Results Pearson Chi-Square test
Value df Asymp. Sig.
(2-sided)
Pearson Chi-Square 7483,339a 300 ,000
Likelihood Ratio 6830,942 300 ,000
22 Hypothesis 3
Regression analysis
A linear regression of the pooled data set of affiliates data is performed to determine whether the variation in preferences for different location regions among MNE home countries can be explained by SCV. The results of the analysis are presented in table 5 for four scenarios: SCV included as independent variable without control variables, SCV included both controlled for only home country specific effects or region specific effects and all three variables included in the model together. The table displays the estimated beta and associated p-value for all three independent variables as well as the adjusted R2
for all four scenarios as an indication of the explanatory power of the model.
Table 5: Results regression analysis SCV
Scenario SCV Region
specific effects
Home country
specific effects Adj. ࡾ 1 -0,286 (0,023)* - - 0,067 2 -0,201 (0,039)* 0,622 (0,000) - 0,444 3 -0,051 (0,640) - 0,610 (0,000) 0,379 4 0,054 (0.373) 0,657 (0,000) 0,651 (0,000) 0,812 * significant at α = 0.05
A few aspects can be remarked from these results. In the first two scenarios, the SCV variable bears the expected negative sign and the coefficient is significant in both models at a five percent confidence level. However, the explanatory power of SCV remains low with an adjusted R2 of 0.067. When controlling for home country specific
effects, SCV keeps its expected negative sign, but loses its significance. Finally,
23
Discussion
In this study the analysis of 93371 US affiliates from 7 different foreign home countries yielded some interesting results that partly confirm earlier established insights as well as provide some interesting new insights.
First, from the diverse sizes and signs of the dummy variable coefficients in the first analysis it can be safely concluded that 51 states differ in their attractiveness for MNEs to locate their affiliates. When correcting the number of affiliates in each state for population size, the coefficients change in size and, in some cases, sign. However, the diversity in coefficients is maintained and therefore confirms the earlier found results. This considered to be substantial proof for H1, the notion that subnational regional differences matter for regional FDI location decisions, which is accepted. Given the broad body of empirical work that has been done earlier on this subject, this should not come as a surprise.
However, the novel aspect of this study is that it breaks up the generic group of foreign MNEs and tries to analyze differing behavior within this group, a notion that is
expressed by H2. The analysis of H2 yields two results that are on the one hand striking and on the other hand strongly contradicting each other. The descriptive analysis of the distribution of each home country’s affiliates over the 9 Census regions clearly shows that most home countries follow quite closely a general pattern when locating their affiliates. Presumably, this pattern is largely determined by the factors also causing the significance of H1. Two interesting deviations from this pattern are the location
preferences of Japan and Canada. For Canada, this is presumably caused by the shared border between these regions and Canada, confirming the distance argument of Ghemawat (2001). For Japan, the striking presence of its MNE’s affiliates in the Pacific region intuitively again confirms the influence of geographical distance, as also its presence in the Atlantic regions New England and Middle Atlantic is below average. On the contrary, South Atlantic does not fit in with this explanation. Closer research is required to find a more satisfying explanation for this remarkable location pattern.
24 accepting H2. Replacement of the Pearson Chi-square test by a different statistical technique testing the deviation in the distribution of affiliates between home countries could verify whether the contradiction is caused by the selection of this technique or whether this is a more fundamental theoretical issue.
For now assuming the appropriateness of the Pearson Chi-square test, H3 hypothesized that the difference in affiliate distribution are driven by SCV. Based on the results of the regression analysis no evidence was found that enables acceptance of this hypothesis. Regressing this variation in the number of affiliates on a self-constructed measure of cultural distance, controlled for both region and home country specific effects yields no significant results. H3 is therefore rejected. Nonetheless, the significance of the SCV variable in combination with region specific effects is a surprising result. It suggests that, when looking at SCV and region effects alone, agglomeration effects apparently do not entirely explain all variation in the number of foreign affiliates per region.
From a methodological perspective, a few aspects should be mentioned that possibly frustrated the finding of any significant, meaningful result. One such a methodological pitfall could be the difference in timeframes between the different CD items and also the timeframe of the affiliate data. In an ideal experimental set-up, all affiliate data would stem from 2002, that is, only data about affiliates that were established in 2002, and all CD items from both WVS and GSS would stem from a few years earlier, reflecting the time for decision making process of the affiliate establishment. However, due to data limitations, cultural data was obtained ranging from 1983 (HARDWORK, GSS) to 2004 (GOVERNMENT, WVS), resulting in a maximum time range of 21 years.
Nonetheless, it was decided to continue with this data. As a main reason for this, cultural characteristics do not change drastically overnight (Minkov, 2013). Therefore, no such changes are expected within the limited timeframe of 21 years that they would
substantially alter the outcome of this investigation. Within the separate SCV-items LUCK, GOVERNMENT and HARDWORK, the WVS and GSS components are matched as much as possible, to maintain the experimental set-up as rigid as possible. Furthermore, it was decided to use the aggregate affiliate data of 2002, including affiliate
establishments from all earlier years, so that influences from early SCV components on earlier establishment decisions are incorporated in the figures as well.
25 United States, and the current MNE home countries, Germany, the Netherlands,
Switzerland, United Kingdom, France, Japan and Canada, no evidence is found that subnational cultural variation substantially influences location decisions of MNEs on a regional level. Changes of recipient countries or MNE home country in future research could substantially strengthen the validity of this study and possibly shed some
26
Conclusion, limitations and recommendations for future research
The purpose of this study was to investigate whether the cultural variation on a subnational level of a host country influences the regional location choice of MNEs investing abroad. This research question was investigated in three steps, each being formulated in a testable hypothesis. The hypotheses were tested on data on the number of foreign affiliates of 7 different home countries in 9 US census regions, combined with cultural data from both GSS and WVS. From the WVS and GSS data a subnational cultural variation variable was constructed based on three equivalent survey items from both surveys to provide an indication of cultural distance.
As a result of this investigation, it was reestablished that subnational regions differ in their attractiveness for MNEs to locate their affiliates. Consequently, the generic group of MNEs was disaggregated by their home country to see whether differences in affiliate locating behavior can be perceived among the different home countries. This yielded mixed results. Descriptive evidence delineated a clear pattern of locating affiliates in particular states, which is similar for most MNE home countries. However, a formal test failed to confirm this identical pattern and did detect significant variance in the
distribution of affiliates over the different regions among the different home countries. Finally, an SCV variable was regressed on the foreign affiliate data in each of the 9 census regions, to see whether the variability in affiliate distribution could be explained by cultural (dis)similarity of the MNE home country with the host region. Controlling for home country and region specific effects, this exercise failed to establish a significant result.
27 existence of ICV in itself, which has been clearly demonstrated by previous studies (Au, 2000; Au & Cheung, 2004). However, for ICV to be usefully integrated in the CD concept with respect to IB, it is strictly necessary to demonstrate that MNEs indeed somehow make advantage of the cultural variation on a subnational level. No such pattern came forward from this study.
A second set of limitations in this study pertains to the use and operationalization of the cultural distance concept. This concept has been subject of an intense debate and essentially, the same critique that was raised against the K-S measure can be raised as well to the SCV variable used in the current study (Shenkar, 2001). The illusion of symmetry and the assumption of equivalence can be argued to still apply to the SCV variable in this study. On the contrary, the very motive for undertaking this study was to explore the possibility for including intracultural variation in a CD measure and thereby tackling an important flaw of the original K-S measure, the assumption of spatial
homogeneity. Therefore, however still imperfect, the currently used measure can be argued to suffer from at least one less shortcoming.
28
Reference list
Au, K. Y. 2000. Intra-cultural variation as another construct of international management: A study based on secondary data of 42 countries. Journal of international management, 6(3): 217-238.
Au, K., & Cheung, M. W. 2004. Intra-cultural variation and job autonomy in 42 countries. Organization Studies, 25(8): 1339-1362.
Barkema, H. G., Bell, J. H., & Pennings, J. 1996. Foreign entry, cultural barriers and learning. Strategic Management Journal: 151-166.
Barkema, H. G., Shenkar, O., Vermeulen, F., & Bell, J. H. 1997. Working abroad, working with others: How firms learn to operate international joint ventures. Academy of Management journal, 40(2): 426-442.
Beugelsdijk, S., Van Hoorn, A., Maseland, R., Onrust, M., & Slangen, A. 2012. Cultural distance in business and management: From mean based to variance-based measures. MIMEO working paper, Rijksuniversiteit Groningen
Cheng, S., & Stough, R. R. 2006. Location decisions of japanese new manufacturing plants in china: A discrete-choice analysis. The annals of regional science, 40(2): 369-387. Coughlin, C. C., Terza, J. V., & Arromdee, V. 1991. State characteristics and the location of foreign direct investment within the united states. The review of economics and statistics: 675-683.
Drogendijk, R., & Slangen, A. 2006. Hofstede, schwartz, or managerial perceptions? the effects of different cultural distance measures on establishment mode choices by multinational enterprises. International Business Review, 15(4): 361-380.
Friedman, J., Gerlowski, D. A., & Silberman, J. 1992. What attracts foreign multinational corporations? evidence from branch plant location in the united states. Journal of Regional Science, 32(4): 403-418.
Ghemawat, P. 2001. Distance still matters. the hard reality of global expansionHarvard Business School Publication Corp.
Head, K., Ries, J., & Ruckman, K. 1998. Industry agglomeration and the location of foreign affiliates. Research in Global Strategic Management, 6: 53-85.
Head, K., Ries, J., & Swenson, D. 1995. Agglomeration benefits and location choice: Evidence from japanese manufacturing investments in the united states. Journal of International Economics, 38(3): 223-247.
29 Jensen, M. C., & Meckling, W. H. 1976. Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of Financial Economics, 3(4): 305-360. Johanson, J., & Vahlne, J. 1977. The internationalization process of the firm-A model of knowledge development and increasing foreign market commitments. Journal of International Business Studies, 8(1): 23-32.
Johanson, J., & Wiedersheim-Paul, F. 1975. The internationalization of the firm -- four swedish cases. Journal of Management Studies, 12(3): 305-322.
Kogut, B., & Singh, H. 1988. The effect of national culture on the choice of entry mode. Journal of International Business Studies: 411-432.
Krugman, P. 1990. Increasing returns and economic geography.
Luger, M. I., & Shetty, S. 1985. Determinants of foreign plant start-ups in the united states: Lessons for policymakers in the southeast. Vand.J.Transnat'l L., 18: 223. Luo, Y., & Shenkar, O. 2011. Toward a perspective of cultural friction in international business. Journal of International Management, 17(1): 1-14.
Mariotti, S., & Piscitello, L. 1995. Information costs and location of FDIs within the host country: Empirical evidence from italy. Journal of International Business Studies: 815-841.
Minkov, M., & Hofstede, G. 2012. Is national culture a meaningful concept? cultural values delineate homogeneous national clusters of in-country regions. Cross-Cultural Research, 46(2): 133-159.
Minkov, M. 2013. Cross-cultural analysis: The science and art of comparing the world’s societies and their cultures. Thousand Oaks, CA: Sage
Nachum, L., Zaheer, S., & Gross, S. 2008. Does it matter where countries are? proximity to knowledge, markets and resources, and MNE location choices. Management Science, 54(7): 1252-1265.
Shenkar, O., Luo, Y., & Yeheskel, O. 2008. From “distance” to “friction”: Substituting metaphors and redirecting intercultural research. Academy of Management Review, 33(4): 905-923.
Shenkar, O. 2001. Cultural distance revisited: Towards a more rigorous
conceptualization and measurement of cultural differences. Journal of International Business Studies, 32(3): 519-535.
Srivastava, S., & Sen, R. 2004. Competing for global FDI: Opportunities and challenges for the indian economy. South Asia Economic Journal, 5(2): 233-260.
Sun, Q., Tong, W., & Yu, Q. 2002. Determinants of foreign direct investment across china. Journal of International Money and Finance, 21(1): 79-113.
30 mode choice, international diversification, and MNE performance: A meta-analysis. Journal of International Business Studies, 36(3): 270-283.
Tung, R. L. 2008. The cultural research imperative: The need to balance cross-national and intra-cross-national diversity. Journal of Intercross-national Business Studies, 39(1): 41-46.
Tung, R. L., & Baumann, C. 2009. Comparing the attitudes toward money, material possessions and savings of overseas chinese vis-à-vis chinese in china: Convergence, divergence or cross-vergence, vis-à-vis ‘one size fits all’human resource management policies and practices. The International Journal of Human Resource Management, 20(11): 2382-2401.
Tung, R. L., & Verbeke, A. 2010. Beyond hofstede and GLOBE: Improving the quality of cross-cultural research. Journal of International Business Studies, 41(8): 1259-1274. Wei, Y., Liu, X., Parker, D., & Vaidya, K. 1999. The regional distribution of foreign direct investment in china. Regional Studies, 33(9): 857-867.
Williamson, O. E. 1979. Transaction-cost economics: The governance of contractual relations. Journal of Law and Economics, 22(2): 233-261.
Williamson, O. E. 1981. The economics of organization: The transaction cost approach. American journal of sociology: 548-577.
Woodward, D. P. 1992. Locational determinants of japanese manufacturing start-ups in the united states. Southern Economic Journal: 690-708.
Zaheer, S. 1995. Overcoming the liability of foreignness. Academy of Management journal, 38(2): 341-363.
Zaheer, S., Schomaker, M. S., & Nachum, L. 2012. Distance without direction: Restoring credibility to a much-loved construct. Journal of International Business Studies, 43(1): 18-27.
31
Appendices
Table A1: LUCK
Table A2: HARDWORK
U.S. Region Canada France Germany Japan The
Netherlands Switzer land Great Britain New England 0,4834 0,0941 0,1592 0,4341 0,1223 0,3976 0,2372 Middle Atlantic 0,0941 0,1825 0,0709 0,5225 0,2107 0,3092 0,1488 E. Nor. Central 0,1592 0,1831 0,0703 0,5231 0,2112 0,3087 0,1483 W. Nor. Central 0,4341 0,2449 0,0085 0,5849 0,2731 0,2469 0,0865 South Atlantic 0,1223 0,3182 0,0648 0,6582 0,3464 0,1735 0,0131 E. South Central 0,3976 0,3165 0,0631 0,6564 0,3446 0,1753 0,0149 W. South Central 0,2372 0,2226 0,0308 0,5626 0,2508 0,2691 0,1087 Mountain 0,4921 0,4143 0,1609 0,7542 0,4424 0,0775 0,0829 Pacific 0,1512 0,2500 0,0034 0,5899 0,2781 0,2418 0,0814
U.S. Region Canada France Germany Japan The
32
Table A3: GOVERNMENT
U.S. Region Canada France Germany Japan The Netherlands Switzer
land Great Britain New England 0,3642 0,6544 0,4088 0,2342 0,4268 0,9708 0,5005 Middle Atlantic 0,6544 0,8238 0,5782 0,0648 0,5962 1,1402 0,6699 E. Nor. Central 0,4088 0,5554 0,3098 0,3332 0,3278 0,8718 0,4015 W. Nor. Central 0,2342 0,6450 0,3994 0,2436 0,4174 0,9614 0,4911 South Atlantic 0,4268 0,5697 0,3242 0,3188 0,3422 0,8862 0,4159 E. South Central 0,9708 0,7044 0,4589 0,1842 0,4768 1,0208 0,5505 W. South Central 0,5005 0,4902 0,2447 0,3983 0,2626 0,8066 0,3363 Mountain 0,5476 0,7154 0,4699 0,1731 0,4879 1,0318 0,5615 Pacific 0,0245 0,6012 0,3557 0,2873 0,3737 0,9177 0,4474
Table A4: Final SCV calculation
U.S. Region Canada France Germany Japan The
Netherlands
Switzer
33
Table A5: Regression coefficients H1
State Standardized Coefficients Standardized Coefficients, corrected for pop. State Standardized Coefficients Standardized Coefficients, corrected for pop.
Beta Beta Beta Beta
Alabama -,020 -,007 Montana -,088 -,078 Alaska -,086 ,056 Nebraska -,077 -,074 Arkansas -,076 -,144 Nevada -,061 -,005 California ,626 ,034 New Hampshire -,056 ,262
Colorado ,017 ,119 New Jersey ,128 ,127
Connecticut ,016 ,235 New Mexico -,075 -,075
Delaware -,056 ,574 New York ,359 ,085
District of Columbia
-,073 ,474 North
Carolina
,099 ,086
Florida ,247 ,036 North Dakota -,091 -,061
Georgia ,128 ,127 Ohio ,168 ,078
Hawaii -,052 ,314 Oklahoma -,062 -,113
Idaho -,082 -,074 Oregon -,025 ,047
Illinois ,219 ,107 Pennsylvania ,177 ,062
Indiana ,013 ,000 Rhode Island -,069 ,171
Iowa -,057 -,056 South Carolina -,005 ,075 Kansas -,057 -,036 South Dakota -,093 -,137 Kentucky -,015 ,039 Tennessee ,025 ,051 Louisiana -,035 -,059 Texas ,337 ,025 Maine -,074 ,033 Utah -,053 ,030 Maryland ,036 ,101 Vermont -,084 ,143 Massachusetts ,122 ,250 Virginia ,085 ,109 Michigan ,086 ,001 Washington ,030 ,047
Minnesota -,011 -,008 West Virginia -,078 -,095
Mississippi -,071 -,126 Wisconsin -,008 -,023
34
Figure A6.1: Affiliate distribution Canada
35
Figure A6.3: Affiliate distribution Germany
36
Figure A6.5: Affiliate distribution Switzerland
37
Figure A6.7: Affiliate distribution Japan
