The Misuse of Institutional Theory in Institutional Distance; an Empirical Analysis.

The Misuse of Institutional

Theory in Institutional Distance;

an Empirical Analysis.

Economy and Business, Research School SOM Supervisor Dr. A.A.J. van Hoorn

2

ABSTRACT

It is well known that the key driver of liability of foreignness is institutional distance. However, the empirical results of fields that heavily rely on institutional distance tend to be equivocal and contradictory, despite many attempts to improve institutional distance metrics. The core of the problem lies at the introduction of new institutional distance metrics, where any form of validation consequently misses. This deeper discussion about the appropriateness of the introduced metric is the missing link in understanding the cause of the mixed results. In this paper I propose that institutional distance can be divided into two broad areas (formal institutional distance and cultural institutional distance), and that the two areas contain multiple dimensions, rather than a single dimension. These propositions are underlined by empirical tests of the validity of the most used unidimensional operationalizations within institutional distance. The findings show that the operationalizations either suffer from face validity problems or suffer from construct validity problems. At the basis of these validity problems lies the theoretical problem that institutional theory has been misused in the institutional distance literature through taking a one-on-one approach, rather than discussing the relevance of the institutional theory for the desired research and approaching its use though an appropriate lens.

1 INTRODUCTION

Whenever firms operate abroad, they face certain cost that are not occurred by domestic firms (Hymer, 1960). These costs are known as the costs of doing business abroad (or CDBA) and derive from the difficulty of operating in a distant environment that differs from the home environment both institutionally, and culturally (Zaheer, 1995; Ghemawat, 2001; 2011; Eden and Miller, 2004). CDBA can be defined into two broad categories: activity-based costs and liability of foreignness (LOF) (Calhoun, 2002; Zaheer, 2002; Eden and Miller, 2004). Activity-based costs are economic costs (such as production, and distribution) that are driven by geographic distance. LOF on the other hand refers to costs that are inferred due to “being a stranger in a strange land” and can be considered the core strategic issue for MNE managers, since geographic costs are well understood (Eden and Miller, 2004, p. 2, 10). The key driver behind LOF is the institutional distance between the host and the home country (Hymer, 1960; Kostova and Zaheer, 1999; Ghemawat, 2001; Eden and Miller, 2004; Perkins, 2009). Examples of LOF leading to market failures include the US video rental chain Blockbuster, who suffered a big blowback between 1991 and 1999 when it entered (and

3 exited) the Japanese market; here they found out that the Japanese did not much care for videos as “wholesome family entertainment”, but rather preferred adult and extreme horror movies instead (US-Tradeshow, 2013). Similarly, Google exited the Chinese market in 2010, due to disagreements with the Chinese government. The pinnacle of these disagreements was the ‘Google Suggest incident’ (Levy, 2011), where the Google Suggest option suggested sexual content for certain search terms and therefore was banned by the Chinese government. As a result, Google was not able transfer their unique selling points to Google.cn and (combined with backlash from the international community) decided to exit the Chinese market.

The problems faced by blockbuster in Japan and by Google in China are not unique. Rather inefficiencies or even failure of a MNE to transfer their unique selling points (e.g., wholesome family entertainment and advanced search algorithms) are commonplace, exemplifying the kind of challenges that MNE’s face in dealing with the distance between their domestic institutional environment and the institutional environment of the host country. More generally, the idea of institutional distance (i.e., the dissimilarity between the institutions of the home and the host country of the MNE) is central to studies considering firms that perform activities across boarders (Ghemawat, 2001; 2011). Cultural distance, for instance, is considered an important variable for any international marketing research (Cho and Padmanabhan, 2005). Also, it is considered a decisive factor for marketing managers who are choosing whether to standardize of adapt their international marketing activities (Karande et al., 2006; Larimo and Kontkanen, 2008; Moon and Park, 2011), and is of increasing interest within advertising and global branding (de Mooij and Hofstede, 2010). The importance of distance is even more central in the field of international management, where distance is considered to be the defining concept (Xu and Shenkar, 2002; Zaheer et al., 2012). Examples of core issues in international management in which the institutional distance construct features prominently includes: entry mode selection (Harzing, 2004; Dow and Larimo, 2009), foreign market selection (Flores and Aguilera, 2007), foreign market sequence (Johanson and Vahlne, 1997), MNE performance (Hutzschenreuter and Voll, 2008), diversification (Tihany et al., 2005).

4 and Kontkanen, 2008; Reus and Rottig, 2009)1. Many researchers have subsequently sought to improve institutional distance research, particularly by adding explanatory variables, assigning the problems as a result to the narrowness (or omitted variable bias) of the currently used metrics (Brouthers and Brouthers, 2001; Xu and Shenkar, 2002, p. 615). Others have taken a less theoratical approach, scrutinizing the institutional distance construct and empirical distance research. Van Hoorn and Maseland (2013), for instance, lament the lack of diversity in the institutional distance literature (which overwhelmingly uses a single base country). Other work has found that distance research imposes symmetry (i.e., the distance from country A to country B is assumed to be the same as the distance from country B to country A), where the actual effect of distance is asymmetric (e.g., Wang and Schaan, 2008). Shenkar (2001) famously criticizes commonly used metrics of institutional distance, particularly cultural distance, noticeably these metrics’ neglect for within-country heterogeneity, such as subcultures and regional policy differences (see also Tung, 2008; Tung and Verbeke, 2010).

Not yet considered in the literature, however, is the more fundamental issue of the construct validity (i.e., “does the set of measured variables actually represent the theoretical latent construct that they are designed to measure; Hair et al, 2010) of institutional distance metrics: are the commonly used indexes actually up to their task of adequately capturing the problems that a firm faces when operating in a foreign context and the challenges that this entails. Although institutional distance is widely considered to be the main driver of LOF (Hymer, 1960; Kostova and Zaheer, 1999; Ghemawat, 2001; Eden and Miller, 2004; Perkins, 2009), simple questions about the correctness of the distance metrics operationalizations and the appropriateness of a direct, one-to-one translation of institutional theory to cross-country distances remain unaddressed. Nevertheless, serious and valid concerns regarding such issues readily present themselves, especially considering that the institutional theories used (i.e., North, 1990; Scott, 1995) mainly concern institutional change and, as such, should not automatically be considered as appropriate for dealing with the specific difficulties and challenges that firms might face when operating abroad. Moreover, the field has developed somewhat of a blind spot, in that they have developed a standard of using composite distance metrics (based on secondary data) without any systematic assessment of the validity of such

1

5 item metrics (i.e., without any evidence that show that these metrics actually possess construct validity)2. For institutional distance the scope has shifted to four different data sources (thoroughly discussed in the chapter Hypotheses and empirical approach), though a lack of discussion about the validity of the metrics that are used to measure the institutional distances, remain. Meanwhile, the introductions of these metrics have created a precedent for repeated usage of them in other research, or as stated by Kogut (2009, p. 785): "It is one of the best‐kept secrets of research that a methodological contribution is the most powerful engine for the replication and diffusion of an idea". However, if such a methodological contribution lacks a serious discussion about the its validity and its appropriateness with regard to the literature; invalid and unreliable metrics of a construct can become the norm in the literature which in turn would lead to questionable and/or contradictory results; a consequence that is observed with regard to current institutional and cultural distance measures. This paper explores the cause of the contradictory results by challenging the basis of the construct in question; the application of institutional theory with regard to the creation of the construct and the validity of the used metrics. Put differently:

This research provides a theoretical and empirical assessment of the validity of the institutional and cultural distance construct, and the metrics that are commonly used to operationalize these constructs in order to measure the costs of doing business abroad.

This paper makes several contributions to the current literature. Theoretically, it sorts out the similarities between the key institutional frameworks of North (1990) and Scott (1995) with regard to both the concept of institutional distance and the operationalizations the institutional distance construct. Similarly, the paper contributes to the ongoing debate in the literature about the difference between institutional distance and cultural distance (Peng and Pleggenkuhle-Miles, 2009). A proposal is derived to distinguish mutually independent areas of institutional distance, namely

formal institutional distance and cultural institutional distance. Also theoretically, the paper further

uncovers fundamental flaws in the metrics that are typically used in the literature to measure institutional distance. Specifically, the paper revisits the institutional theories by North (1990) and

2

6 Scott (1995), and contrasts them with other institutional and cultural perspectives (e.g., Hofstede, 1980; Schwartz, 1994; Hall and Soskice, 2001; House et al. 2004; Acumoglu and Robinson, 2012) to question the appropriateness of unidimensional operationalizations of distance (e.g., Kogut and Singh distance or Mahalonobis distance (Berry et al., 2010)) in different institutional areas. Two propositions are formulated to improve the application of institutional theory in the measurement of institutional distances. Finally, this paper confirms these two propositions, empirically showing that the unidimensional metrics that prevail in the literature suffer severe validity problems. In the end, the paper does not dismiss institutional distance research on principle. Rather, it highlights the need for the field to deal with some basic issues, not only as a way to entangle inconsistent results, contradicting theories, and paradoxes. But also in order to make much-needed progress on an issue that is at the core of many disciplines in business studies.

7

2 INSTITUTIONAL THEORIES AND INSTITUTIONAL DISTANCE

IN INTERNATIONAL MARKETING AND BUSINESS

In this chapter I will address the institutional theories that are currently used as a base for theorizing about institutional distance. An understanding about the underlying theory is needed before being able to address the validity issues regarding the operationalizations of institutional distance metrics. There are two different institutional approaches towards measuring institutional distance; through the use of the framework of North (1990) and through the use of the framework of Scott (1995). First, the similarities between both frameworks are discussed, as well as the link between cultural distance and institutional distance. Secondly, two propositions are posed for the proper use of the institutional theory with regard to distance.

2.1 INSTITUTIONAL THEORIES

2.1.1 North’s framework of formal and informal institutions

8 conduct, norms of behavior and conventions.” Thus, informal institutions can be defined as: humanly devised patterns of interaction that are not formally codified but embedded in the shared norms, values and beliefs of a society (North, 1990, 2005). Following the analogy, the formal institutions then are the (official) rules of the game in a competitive team sport. For North (1990 p. 47), formal constraints derive from formal rules, meaning “political (and judicial) rules, economic rules and contracts”.

2.1.2 Scott’s three institutional pillars

The framework of Scott (1995) consists out of three pillars: regulatory, normative, and cultural-cognitive (henceforth referred to as the cognitive pillar). These pillars are based on the three isomorphic processes described in the famous work by Dimaggio and Powell (1983): Coercive, normative, and mimetic. The regulatory pillar reflects “the existing laws and rules in a particular national environment and promote certain behavior and restrict others” (Kostova, 1999, p. 314). Scott (1995) states that the explicit regulatory processes are: rule setting, monitoring, and sanctioning activities. The basis of the legitimacy of the regulatory pillar is ‘legally sanctioned’, and the mechanism is coercive. Therefore, institutions in the regulatory pillar are exerted through force, fear and expedience (Scott 1995). The normative pillar refers to “the values and norms held by the individual in a given country” (Kostova, 1999, p. 314) and it is “rooted in societal beliefs and norms and prescribes desirable goals and the appropriate means of attaining them” (Xu & Shenkar, 2002, p. 610). The basis of legitimacy of the normative pillar is ‘morally governed’ and the mechanism is normative. Thus institutions in the normative pillar are exerted through social obligation (i.e., the way a specific actor is expected to behave by other salient actors) (Scott, 1995). The cognitive pillar reflects “the cognitive categories widely shared by the people in the particular country such like schemas, frames, inferential sets and representations affect the way people notice, categorize and interpret stimuli from the environment” (Kostova, 1999, p. 314) or stated differently: “the social knowledge, national symbols and the way and ability that the people understand and interpreted things in a certain country” (Xu & Shenkar, 2002, p. 610). The basis of legitimacy of the cognitive pillar is ‘comprehensible, recognizable, and culturally supported’ and the mechanism is mimetic. Therefore, institutions in the cognitive pillar are exerted through their ‘taken-for-grantedness’ and through a ‘shared understanding’ (i.e., the way we do things around here) (Scott 1995).

2.1.3 Integrating institutional theories

9 institutional environment of a country. As a result, although they are different, there is a large overlap between the two frameworks. Scott’s regulatory pillar matches North’s idea of formal institutions, while the normative and cognitive pillar corresponds to North’s idea of informal institutions (see figure 1).

2.1.3.1 Formal institutions vs. the regulatory pillar

The similarities between the definitions of formal institutions and of the regulatory pillar are rather large. Scott speaks about ‘the existing laws and rules’ while North in terms calls it the ‘official rules of the game’. In both cases, the mechanism behind compliance is coercive (i.e., the individual is forced to comply with the rules) and the processes involved include rule-setting, monitoring, and sanctioning activities (Scott, 1995 p. 52; North, 1990, p. 58 (refers to it as formal third-party enforcement)). Furthermore, Scott warns that the actual interpretation of laws and rules by humans resides with the other two pillars. This demarcation fits with the distinction made by North, who forwards the idea that formal rules are underlined and supplemented by informal codes.

2.1.3.2 Informal institutions vs. the normative and cognitive pillar

10 therefore should be viewed as a breakdown of informal institutions, rather than supplementing them.

2.2 CURRENT USE OF INSTITUTIONAL DISTANCE

2.2.1 Data sources and institutional dimensions

Since the similarities between both frameworks are rather large, it should not be surprising that the metrics that are used to measure both frameworks tend to be similar. Typically, a composite metric that is based on either the Worldwide governance indicators (WGI) or the Economic freedom project (EFP) is both used to measure the regulatory distance (e.g., Ionascu et al., 2004; Pogrebnyakov and Maitland, 2011; Estrin et al., 2007) and to measure the formal distance (e.g., Estrin et al., 2009; Abdi and Aulakh, 2011; Dikova, 2009). With regard to informal distance and normative,- and cognitive distance, the typical operationalization is through a composite metric based on Hofstede and/or GLOBE, which are both most commonly used to measure cultural distance (Ng. et al. 2006; Hofstede, 2006). The use of these cultural distance metrics with regard to institutional distance is widespread and varied, where it either proxies normative distance, cognitive distance, informal distance, institutional distance, or is added (as suggested by Xu and Shenkar, 2002) as a cultural dimension, supplementing institutional distance (see table 1).

In current practice, papers that operationalize institutional distance do so either through the framework of North, Scott, or no framework at all. Papers that use North as the institutional framework for institutional distance, operationalize this framework through a formal/regulatory metric and a cultural metric.3 Papers that do not use an (explicit) institutional framework conform to

3 _{with exception of Arslan and Larimo (2010) who operationalizes a formal and an informal metric based on}

world economic forum questions

11 this practice and (without exception) use one or more composite metrics based on the four above mentioned data sources. However, papers that use Scott as theoretical basis diverge. They consider cultural and normative distance to be the same (Bea and Solomon, 2010), with noticeable exceptions of Gaur and Lu (2007) and Gaur et al. (2007), whom consider cultural distance to correspond with cognitive distance, and Xu et al. (2004) who argue that the regulative and normative distance “in particular reflect the institutional perspective’s roots in economics and sociology” (Xu et al. 2004, p. 288), thus excluding the cognitive dimension which they consider hard to operationalize on a country level, however adding cultural distance as a control.

I however would like to suggest that it is not that the cognitive pillar is less important than the other two pillars or that cultural distance converges to either the normative pillar or the cognitive pillar. Rather I suggest that the division of institutions as suggested by Scott (1995) is not the correct distinction to make with regard to institutional distance. Considering that the interest of distance measures is to gain insight in the costs of doing business abroad, the pillars should be viewed through that lens. Eden and Miller (2004) state that the reason MNE’s have troubles with regard to LOF, is that they are: ‘a stranger in a strange land’. As such, the normative and cognitive pillars of the host country become indistinguishable for the foreign MNE. Again, consider the sport analogy; if a player is not familiar with the informal constraint to throw the ball back to the opposing team after an injury treatment, and subsequently breaks the rule. Then it does not matter for a foreign player whether the rule was embedded in the normative pillar or in the cultural-cognitive pillar. In both cases, the responses towards the failure to comply with the informal rules of the game are the same (i.e., outrage and a more aggressive approach towards the player of both the opposing team and its audience). However, would the rule be embedded in the regulatory pillar, than the consequences would diverge (i.e., referee awards a free kick to the opposing team and perhaps a yellow card). Cognitive institutions than can be regarded as normative institutions that are so embedded in a culture that they do not need to be communicated within the society in order to be enforced and violations are (as a consequence) punished in a similar fashion. This makes it is not only hard for a stranger to distinguish between normative and cognitive institutions of a host country, it makes it irrelevant.

2.2.2 Institutional distance vs. cultural distance

12 metrics used to measure informal institutional distance (as discussed above). One of the prominent current debates in the Global Strategy literature is whether cultural distance and institutional distance should be distinguished (Peng and Pleggenkuhle-Miles, 2009). Xu and Shenkar (2002, p. 615) state that “institutional distance complements, rather than replaces, the cultural distance construct”, and they consequently argue that previous problems with results from cultural distance measures are a result of not including institutional distance. Hofstede et al. (2002, p. 800), on the other hand, find that “institutions are the crystallizations of culture, and culture is the substratum of institutional arrangements”, implying that any claim concerning the link between culture and institutions becomes “hairsplitting” (Peng and Pleggenkuhle-Miles, 2009, p. 55). In this paper, a third viewpoint is proposed: With regard to intended purposes, cultural distance is interchangeable with informal institutional distance.

13 Considering that formal distance deals with rule-setting, monitoring, and sanctioning activities; it is safe to say that the formal distance construct is not captured in the cultural distance construct. To exemplify that formal distance is not captured by cultural distance; consider the cultural distance and the formal distance between North Korea and South Korea, or the difference in culture and formal institutions that has grown in Greece over the last couple of years due to austerity measures. Therefore, the similarities between cultural distance and institutional distance need to be discussed with regard to informal distance. Informal institutions can be defined as humanly devised patterns of interaction that are not formally codified but embedded in the shared norms, values and beliefs of a society (North, 1990, 2005). As a result, the definition of informal institutions and the definition of culture are hard to distinguish from each other. It should be acknowledged that there is a gap between culture and informal institutions; for instance, Helmke and Levitsky (2003) present several different sources for informal institutional change, under which cultural evolution is one. This leads to interesting questions about the difference between culture and informal institutions with regard to the evolution and change of them over time. However, since distance calculations are not interested in the (slow) process of change, the question how culture drives informal institutional change is not of interest when there is no clear indication that there is a significant gap between both. Therefore, when interested in distance, informal institutional distance and cultural distance (though conceptually different) can be regarded interchangeable.

15

2.3 PROPOSED USE OF INSTITUTIONAL THEORY IN MEASURING

INSTITUTIONAL DISTANCE

2.3.1 Cultural institutional distance

Considering the both the current practice of interchanging informal distance with culture distance, and the similarities in the definitions (as discussed above), the current explanation of the equivocal and contradictory results does not suffice anymore. As stated by Xu and Shenkar (2002) among others, the use of the cultural distance construct has (consistently) led to inconsistent results. Xu and Shenkar (2002) find that this ambiguity in distance research is caused by an omitted variable bias and suggest extending distance research by considering institutional distance in addition to cultural distance. Therefore, using informal institutions and culture interchangeably with regard to distance, contradicts the explanation of Xu and Shenkar (2002), since the extension becomes an integration. This leads to the question: if not omitted variable bias, what than causes the inconsistent results of cultural distance measurements?

Cultural distance metrics have been often criticised in the past, most noticeably the metrics that are based on the Hofstede data. Shenkar (2001) notes that cultural distance methods assume that the distant is symmetric (i.e., distance between Netherlands and Japan is assumed to be equal to the distance between Japan and Netherlands), there is cultural homogeneity within a country, and homogeneity over corporate culture, and the methods are subject to the same critiques that apply for the dataset that it is based on. Similarly, Zhao et al. (2004) question the measurement accuracy of these cultural distance measures. Harzing (2004) mentions that there is almost a blind confidence in the Kogut and Singh index and the absolute validity of Hofstede (both discussed in the following chapters). She shows several examples of studies that have used flawed proxies in their researches as a result of the desire to use this method. Shenkar (2012, p.12) states “I have gradually come to realize that what we have been doing was not only superficial, lacking in substance and rigor, but was perhaps invalid or, at the very least, seriously flawed”.

16 Vermeulen, 1997; Berry et al. 2010; Miller et al. 2013). As a result, it is arguable that the current cultural distance metrics do not actually measure cultural institutional distance at all (i.e., lack construct validity).

One of the main theoretical problems with current cultural institutional distance metrics is that they are operationalized as unidimensional constructs. Culture as a unidimensional construct is however a stretch. First of all, a unidimensional operationalization of the cultural institutional distance construct has a meaningless scale on which the countries are ordered (i.e., what does a low cultural country score means compared with a high cultural country score?), thus resulting in serious face validity (i.e. the extent to which the content of the items is consistent with the construct definition, based on the researcher’s judgement, Hair et al., 2010) issues. Secondly, leading frameworks of national culture include six or more independent dimensions of national culture4. The idea to capture a complex construct as culture in a single dimension than seems doomed to render inconsistent results with the resulting metrics, almost certainly lack construct validity. In fact, treating culture as a unidimensional construct actually implies that these frameworks have been wrong in identifying multiple dimensions of culture.

The issue of dimensionality in any construct is a fundamental one and one that must be resolved before the construct can be employed in any theoretical network (Mittal, 1989), where assuming unidimensionality in a multidimensional construct could severely skew the parameter estimates (Yang, 2007). To exemplify the consequences of the unidimensionaly assumption, consider that individualistic societies need an approach that is based on persuasion, with regard to advertising, and collectivistic societies need an approach that is based on trust (de Mooij and Hofstede, 2010). As a result, MNE’s that advertise in foreign countries with large individualistic/collectivistic distance will deal with LOF due to unfamiliarity. However, the same effect on advertising preference is not expected (or smaller) for the other items of culture (as proposed by Hofstede, 1980). Therefore, a composite metric of these culture dimensions will have a skewed parameter estimate when the cultural institutional distance scores per (Hofstede) dimension are not highly correlated.

4

17

Proposition 1: With regard to distance, culture (and informal institutions) should be

considered a multidimensional construct. Thus measuring cultural institutional distance through a unidimensional metric will lack construct validity.

2.3.2 Formal institutional distance

The same question that has been asked for cultural institutional distance should be asked for the other side of the coin: is the formal institutional distance construct a unidimensional construct? There is some reason to assume unidimensionality of the construct. First of all, in contrast with the construct culture, ordering the regulations of a country can be a logical and meaningful activity. For instance, one could order countries with regard of the quality of their rule-setting, monitoring, and sanctioning activities (where (e.g.) Denmark governs more effectively than Iraq). Similarly, Acemoglu and Robinson (2012), suggest that countries can be sorted in a continuum that ranges from extractive,- to inclusive economic (and political) institutions. Here, countries have extractive economic institutions when the institutions are designed to extract incomes and wealthy from one subset of society to benefit a different subset, while inclusive economic institutions are designed to do the opposite. As such, the assigned score to a country would have meaning (i.e., one could state that an increase in the country score from 2 to 4 would represent a doubling of their regulatory quality, where the same change in a culture score is meaningless). Secondly, both the literature (as shown above) as well as the typical metric used to measure formal institutional distance is based on a single dimension.

18 0.22 distance) Through the use of this unidimensional metric, it is predicted that a Norwegian MNE would experience more formal institutional LOF when entering the Danish market, than it would when entering the UK market. It is however more likely that a Norwegian MNE would experience more costs due to the unfamiliarity with the (liberal) system of regulations of the UK than it would when confronted with the (coordinated) style of Denmark. Similarly, in a unidimensional scale of formal institutional distance, Haiti and North Korea both score very low (WGI 2003 unidimensional standardized country scores: -1.580 vs. -1.579). However, the reason behind the low scores are opposite with regard to the rule-setting, monitoring, and sanctioning activities. Consider that Haiti is perhaps the country with the least government control in the world and North Korea is perhaps the country with most government control in the world.

In my opinion, the current literature typically uses formal institutions and the outcome of these institutions interchangeably, which results in faulty measurements of formal institutional distance. The use of the WGI is an example of this practice, where this measure ranks countries with respect to good governance (Kaufmann et al. 2007). Good governance however is an outcome of rule-setting, monitoring, and sanctioning activities, rather than a part of them. As such, governments can use different philosophies to come to a similar quality of governance. These different philosophies in term create large distances with regard to formal institutions, however are not reflected in the measurement method that only regards the outcome (as shown above). Similarly, the distinction suggested by Acemoglu and Robinson (2012) allows for a free interpretation on how to reach either extractive or inclusive economic institutions and in which form it manifests. Considering the differences between the distance calculation of UK and Norway and the actual distance, I propose the following:

Proposition 2: With regard to distance, formal institutions should be considered a

19

3 HYPOTHESES AND EMPERICAL APPROACH

In order to underline the stated propositions, I will discuss the construct validity (i.e., do the set of measured variables actually represent the theoretical latent construct) of current institutional (and cultural) distance measures in this chapter. Through this discussion, it is shown whether the data supports the propositions, and (by extend) if mistakes in the operationalizations of the current metrics are to blame for the inconsistencies in the results of the literature. As mentioned in the introduction; the standard approach to measure the institutional distance between a host and a home country is to combine a set of indicators from secondary data sources (e.g., Hofstede, GLOBE, WGI) into a single composite metric. However a sound discussion of the validity of the measurement typically lacks (e.g., Kogut and Singh, 1988; Barkema and Vermeulen, 1997; Berry et al. 2010; Miller et al. 2013). Considering the difficulties that the mixed results have caused in the fields that use these measures, and the lack of proof of the validity and fit of these methods; a plausible explanation of the mixed results over different researches can be found in invalid measurement methods. Therefore, it is time to take a step back and address the issue of construct validity of distance measures that are used in the institutional approach towards doing business abroad.

20 Figure 2 expends on Figure 1 from the previous chapter. It depicts the two main institutional frameworks by North and Scott and the dimensions of institutional distance that derive from these frameworks. However, Figure 2 additionally includes information on the various distance metrics used in the literature as well as the most often used data sources that the basis for these metrics. In this chapter, the most commonly used metrics of both formal institutional distance as well as cultural institutional distance are discussed5.

3.1 VALIDITY AND UNIDIMENSIONALITY OF FORMAL INSTITUTIONAL

DISTANCE METRICS

3.1.1 Worldwide Governance indicators

One of the most commonly used data sources used to construct a measure for formal institutional distance is the Worldwide governance indicators6. The World Government Indicators (WGI) is based on the aggregation of perceptions of governance from 31 different data sources that are provided by 25 different organizations. It ranks countries with respect to six dimensions of good governance (excerpt from Kaufmann et al., 2010, p. 4):

(1) Voice and Accountability (capturing perceptions of the extent to which a country's citizens are able to participate in selecting their government, as well as freedom of expression, freedom of association, and a free media),

(2) Political Stability and Violence (capturing perceptions of the likelihood that the government will be destabilized or overthrown by unconstitutional or violent means, including politically-motivated violence and terrorism).

(3) Government effectiveness (capturing perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government's commitment to such policies).

5 _{Only the metrics that have been operationalized more than two times are included in figure 2.} 6

21 (4) Regulatory quality (capturing perceptions of the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development). (5) Rule of law (capturing perceptions of the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence).

(6) Control of Corruption (capturing perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as ‘capture’ of the state by elites and private interests).

These aspects are derived from the data sources through the use of a cluster analysis (Kaufmann et al., 2010). With these six dimensions, three different aspects are included; first of all, “the process through which governments are selected, monitored, and replaced” is represented in the first two dimensions. Secondly, “the capacity of the government to effectively formulate and implement sound policies” is represented through the third and fourth dimension. And finally, “the respect of citizens and the state for the institutions that govern economic and social interactions among them” is represented in the fifth and sixth dimension of good governance (Kaufman et al., 2010, p. 3).

22 an effect, (s)he should also explain why this effect is caused by only this dimension and not any of the other dimensions or even multiple ones. To test this, the research cannot just add multiple dimensions in an analysis, because this would lead to large multicollinearity problems.

One interesting (but faulty) methods that has been used in order to prevent the multicollinearity problems is by estimating different models for each dimension (e.g., Slangen and van Tulder, 2009; de Groot et al., 2004). Though a creative solution, the use of this method automatically leads to endogeneity problems. For instance, Slangen and van Tulder (2009) show that all the dimensions have a significant and similar impact on their dependent variable, the exclusion of these variables in other models (in order to combat multicollinearity) than automatically causes the error and the independent variables to be related. Therefore it is more common that the WGI is included as a composite metric of all dimensions as an independent variable in their research (Beugelsdijk et al. 2004; Linders et al., 2005a; Linders et al., 2005b; Dikova, 2009; Slangen and Beugelsdijk, 2010; Ando, 2011; Pogrebnyakov and Maitland, 2011; Ando, 2012; Abdi and Aulakh, 2012; Dikova, 2012; Aleksynska and Havrylchyk, 2013; Ando and Paik, 2013;; Pérez-Villar and Seric, 2013). However (as is a common theme), no proof of the validity of averaging the dimensions is provided (or a reference to a paper that shows the validity of the method used).

Considering the practice of using a unidimensional metric, constructed through the WGI as a measurement for formal institutional distance; face validity issues occur. As argued above, the WGI does not actually measure formal institutions; rather it measures the outcome of these formal institutions. When contrasting the WGI scores and the actual difference between Norway vs. the UK and Haiti vs. North Korea, questions arise about the applicability of the WGI scores for measuring formal institutional distance, since the actual formal institutional distance construct does not seem to be represented in the WGI scores. As argued above, formal institutional distance should be a multidimensional construct in order to be able to capture the complexity of the construct. As a consequence, a valid unidimensional construct is not an adequate measure of formal institutional distance. Considering the critique of Thomas (2006) about the discriminant validity of the use of separate WGI dimensions (thus indicating that the WGI dimensions are highly correlated), it is hypothesised that:

H1: The use of a unidimensional metric based on the worldwide governance indicators is

23

3.1.2 Economic freedom project

The purpose of the economic freedom project (EFP) is to reflect the economic environment in every country studied in as balanced a way as possible (Miller et al., 2013). The EFP is constructed through the analysis of 10 components of economic freedom. These 10 components can be grouped into four pillars: Rule of law (property rights, freedom from corruption); Limited government (fiscal freedom, government spending); Regulatory efficiency (Business freedom, labor freedom, monetary freedom); and Open markets (trade freedom, investment freedom, and financial freedom). All of the components are measured on a 100 point scale, and a total score is given by averaging the scores of all components (Miller et al., 2013). Even though an overall score is provided in the economic freedom project, no other rational is given then the one given above. Even more so, the authors state that while it is clear that the ten economic freedoms interact, a thorough investigation of these interactions is beyond the scope of their research (Miller et al., 2013). Several researchers have used a composite metric (of either the full EFP or of several dimensions of it) in order to measure institutional distance (Ionascu et al., 2004; Estrin et al., 2007; Estrin et al., 2009; Meyer et al., 2009; Gubbi et al., 2010).

The EFP is a subjective scale that not only looks at the regulatory constraints that a government places on its citizens through rule-setting, but also considers the enforcement of these rules. For instance, the efficiency of the court system is taken into account as well as whether private property is guaranteed for the construction of the dimension property rights (Miller et al., 2013). Therefore, the fit with the regulatory pillar, as specified by Scott (1995) (i.e., rule-setting, monitoring, and sanctioning activities) can be considered good and no face validity problems can be detected. Despite current practices of researcher to use the overall score to measure formal institutional distance, it is expected that distance measures based on transformations of the EFP will exist out of multiple dimensions (in accordance with proposition two).

24 investment capital (e.g., Chile and Peru – Miller et al., 2013) and vice versa (e.g., Maldives and Solomon Islands – Miller et al., 2013). As briefly mentioned in the previous chapter, reasons for a different composition of the formal institutions in a country can be expected when considering the different styles of governing. As a result, it is hypothesized that:

H2: The use of the overall score of the economic freedom project lacks construct validity due

to a lack of unidimensionality.

3.2 VALIDITY AND UNIDIMENSIONALITY OF CULTURAL INSTITUTIONAL

DISTANCE METRICS

3.1.3 Hofstede

Cultural distance, as a construct, has been operationalized many times (Xu and Shenkar, 2002; Zaheer et al., 2012). The operationalization that have had the most impact and is currently most commonly used for institutional cultural distance, are the metrics which are based on the specification of national culture by Hofstede (Beugelsdijk et al., 2004; Linders et al., 2005b; Ionascu et al., 2004; Jensen and Szulanski, 2004; Xu et al, 2004; Estrin et al., 2007; Flores and Aguilera, 2007; Gaur and Lu, 2007; Gaur et al., 2007; Parboteeah et al., 2008; Dikova, 2009; Estrin et al., 2009; Slangen and van Tulder, 2009; Dikova et al. 2010; Slangen and Beugelsdijk, 2010; Ando, 2011; Abdi and Aulak, 2012; Ando, 2012; Dikova 2012; Higon and Antolin, 2012; Salomon and Wu, 2012; Ando and Paik, 2013). In his research, Hofstede (1980) defines 4 dimensions of national culture: power distance, uncertainty avoidance, individualism-collectivism, and masculinity-femininity. Currently, both a fifth dimension is added: long-term / short-term orientation (Hofstede and Bond, 1998), and a sixth dimension is added: Indulgence vs. constraint (Minkov and Hofstede, 2011).

25 original construction of the dataset does not allow for the use of a composite measure, and the validity of this practice has already been openly questioned, it is hypothesized that:

H3: The use of a composite distance metric based on Hofstede data lacks construct validity

due to a lack of unidimensionality.

3.1.4 GLOBE

Finally, the GLOBE (Global Leadership and Organizational Behavior Effectiveness) research (House et al. 2004) has been used as basis for determining both cultural distance as well as institutional distance (Schwens et al., 2007; Parboteeah et al., 2008; Estrin et al., 2009; Dikova et al., 2010). The GLOBE research distinguishes two times 9 dimensions of culture: Uncertainty avoidance, power distance, institutional collectivism, in-group collectivisism, gender egalitarianism, assertiveness, future orientation, performance orientation, humane orientation (House et al. 2004). For each of these dimensions they distinguish in the culture’s values (i.e., how members of the culture believe that the culture should be) and a culture’s practices (i.e., how members of the culture believe that the culture currently is). Hofstede (2006) states that the GLOBE research closely resembles the original Hofstede model, as is confirmed by the research of Hutzschenreuter and Voll (2008), who found similar effects for their cultural distance measure based on Hofstede and based on GLOBE. Therefore, the same results for composite metrics that are based on the GLOBE dataset are expected that are expected for the composite metrics that are based on the Hofstede dataset:

H4: The use of a composite distance metric based on the GLOBE dataset lack construct

26

4 DATA AND METHODS

In this chapter I will discuss data and methods that are needed to test the hypotheses. First the three different sorts of transformations that are performed on the four data sources are discussed. Secondly, the data collection methods and the choices in the construction of the twelve tested metrics are discussed. Finally, the measurements that are used to analyze the results and the methods that are used to gain these results are addressed.

4.1 DATA

4.1.1 Data transformation methods

Four different sorts of transformations are used on the aforementioned data sources in order to gain a composite distance metric. These are: averaging the actual distances, the Kogut and Singh index (KSI), Euclidean distance as proposed by Barkema and Vermeulen (1997), and the mahalanobis distance (Mahalanobis, 1936; Berry et al., 2010). Both of these methods perform each a horizontal transformation of the data (i.e., reducing the number of variables) and a vertical transformation of the data (i.e., calculating a distance between two observations).

The use of the first method, averaging the actual distance per dimension, is typical for the transformation of WGI (Pogrebnyakov and Maitland, 2011) and EFP data (Estrin et al. 2009). In this method, first the vertical transformation takes place by calculating the actual distance of the home and host country, for each dimension. Secondly, these scores are horizontally transformed by averaging them.

27 has not only been used on the Hofstede data, but has also been used on the other data sources (e.g., Beugelsdijk et al. 2004; Ando, 2012).

Barkema and Vermeulen (1997) proposed the use of the Euclidean distance measure. Here, the horizontal transformation of the data is a square root of the sum of the dimensions, rather than an average. Drogendijk and Slangen (2006, p. 367) state that the main difference with the KSI is that there is not assumed that each score on Hofstede’s dimension is equally important, rather “it computes their distance in a four dimensional space as the square root of the sum of the squared differences in the scores on each cultural dimension”.

Berry et al. (2010) suggest that distances are better calculated through the Mahalanobis method for calculating distance (Mahalanobis, 1936). This method differs from the Euclidean distance measure, where it takes the correlation between the variable input indicators into account, it accounts for the variance of the variables and it is not as sensitive to the scale of measurement (Berry et al. 2010). A simple method to calculate the Mahalanobis distance is to use a principle-component analysis to create standardized values of the principle principle-component, and use these values in a Euclidean distance calculation (De Maesschalck et al., 2000). An extensive explanation of the Mahalanobis method is given in McLachlan (1999).

4.1.2 Horizontal vs. vertical data transformations

In order to gain distance scores out of any of the above mentioned datasets, two transformations need to take place: (1) the number of observations (i.e., dimensions) need to be reduced (horizontal transformation), and (2) the distance between host and home country needs to be calculated (vertical transformation). While all three methods are adequate to perform the vertical transformation, they only assume that the horizontal transformation is allowed. The implicit assumption that is made through the use of these methods is debatable (Shenkar, 2001), since it contradicts the separation that is made in the initial research.

28 data through taking standardized residuals of a principle component analysis) and secondly on the vertical axis (using Euclidean distance). The vertical transformation for the Euclidean Distance measure and the KSI is the same (as discussed in the above). It is in the horizontal transformation that the difference between the two methods is made. However, the validity of these measurement methods are tested before the horizontal transformation is made. Results of these tests show whether a horizontal transformation is allowed, or if the separate items are not meant to be reduced to a single variable. Therefore, a total of three metrics are created per data source (Mahanalobis/profile; KSI/Euclidean (weighted) distance; Actual (unweighted) distance), resulting in a total of (3*4) 12 datasets.

4.1.2 Data collection

In order to perform the WGI transformations, the WGI scores have been downloaded for the years 2003 till 2011 (available at: databank.worldbank.org). Every country that is complete in the observations on all six dimensions (Voice and Accountability; Political stability and violence; Government effectiveness; Regulatory quality; Rule of law; Control of corruption) and on all years, are included in the dataset. This results in 195 included countries with each nine lines of observations. Therefore, a total of (195*9 =) 1,755 observations are included in the Mahanalobis/profiles metric. For the other metrics all distances are calculated, resulting in a total of (195*195*9 =) 342,225 observations. Here, 1,755 observations are deleted because they have the same host and home country, leaving (342,225-1,755 =) 340,470 observations. For the weighted distance metric, the variance of the original data has been added in the distance calculation (V&A: 1.010 – PS/VN: 0.993 – GE: 1.010 – RQ: 1.013 – RoL: 0.999 – CoC: 1.016).

29 calculation (PR: 583.732 – FC: 440.671 – FisF: 180.162 – GS: 618.899 – BF: 333.592 – LF: 291.567 – MF: 110.223 – TF: 158.205 – IF: 555.814 – FinF: 373.717).

In order to perform the Hofstede dataset transformations, the cultural country scores of Hofstede are downloaded (available at: geerthofstede.nl). In this research, every country that has all observations on the items: Power Distance, Individualism, Masculinity, and Uncertainty avoidance, are included. The choice has been made not to include the other two dimensions, where a result that does not converge on these four dimensions automatically does not converge when adding extra dimensions. Next to that, Belgium and Switzerland are deleted because the inclusion of Belgium – French and Belgium – Netherlands, as well as Switzerland – German and Switzerland – French cause these observations to be double. Therefore, a total of 76 observations are included, leading to a total number of 76 observations in the Mahanalobis/profiles metric. For the other metrics all distances are calculated, resulting in a total of (76*76 =) 5776 observations. Here, 76 observations are deleted, because the home country cannot equal the host country. Therefore, these metrics have 5700 observations. Finally, for the weighted distance metric, the variance of the original data has been added in the distance calculation (PDI: 454.596 – IDV: 570.653 – MAS: 377.547 – UAI: 532.220).

In order to perform the GLOBE dataset transformations, the GLOBE country scores are downloaded (available at: Harzing.com). The dataset contains 62 countries which all have two times nine dimensions (power distance; uncertainty avoidance; humane orientation; institutional collectivism; in-group collectivism; assertiveness; gender egalitarianism; future orientation; and performance orientation). For this research, only the practices are used, since these are typically used in institutional distance calculations (Estrin et al., 2009). This leads to 62 observations in the Mahanalobis/profiles metric. For the other metrics all distances are calculated, resulting in a total of (62*62 =) 3844 observations. Here, 62 observations are deleted, because the home country cannot equal the host country. Therefore, these metrics have 3782 observations. For the weighted distance, the variance of the original data has been added in the distance calculation (UNCS: 0.376 – FUTS: 0.265 – POWS: 0.162 – INDS: 0.248 – HUMS: 0.129: ACHS: 0.319 – TRIS: 0.164 – MALS: 0.236 – AGGS: 0.412).

4.2 METHOD AND MEASUREMENTS

30 al., 2010, p. 631, 664). In order to analyze whether the metrics fit their theorized latent construct, Hair et al. (2010) suggests testing the goodness-of-fit, reliability, convergent validity, discriminant validity, nomological validity and face validity. The first three measures are tested and discussed. 7

To be able to measure the convergent validity, reliability and the goodness-of-fit of the twelve different metrics of institutional distance, a confirmatory factor analysis (CFA) is performed. A CFA is a statistical technique that is used to verify the factor structure of a set of observed variables. It allows for testing how well the observed variables represent the latent construct that they form. “CFA’s ultimate goal is to obtain an answer as to whether a given measurement model is valid.” Hair et al., 2010, p. 711). The CFA is performed with maximum likelihood estimation (MLE), through the use of LISREL software. The MLE is a flexible approach in which the ‘most likely’ parameter values to achieve the best model fit is found (Hair et al., 2010). Furthermore, a principle component analysis is performed to show the robustness of the results. It allows to test whether the data provides an alternative solution for the stated problem. Also, a reliability analysis is performed to test the internal consistency of the observed variables.

4.2.1 Reliability

“Reliability concerns the degree to which the scores are free from random measurement error” (Kline, 2005, p. 58). The most widely used objective measurement of reliability is the Cronbach’s alpha (Tavakol and Dennick, 2011), as introduced by Lee Chronbach (1951). It measures the internal consistency of a construct, and is expressed as a number between 0 and 1. The internal consistency of a construct is the degree to which responses are consistent across the included items; it describes the extent to which all items measure the same construct. Internal consistency is a necessary, but not sufficient condition for measuring homogeneity or unidimensionality for the included items (Cortina, 1993). When constructs are not internally consistent, the content of the items may be so heterogeneous that the total score is not the best possible unit of analysis for the

7 _{face validity is discussed above, however discriminant validity and nomological validity are not}

31 measure (Kline, 2005). The numerical value of the Chronbach’s alpha can be used to explain the percentage of the observed variance that is due to random error, by taking 1 - the squared value. For instance, a Chronbach’s alpha of .70 means that (1 – 0.70 * 0.70 =) 51 percent of the variance is due to random error (Nunnally and Bernstein, 1994).

Some discussion exists about the value of the Cronbach’s alpha needed for an item to show reliability of a factor. Hair et al. (2010) notes that, as a rule of thumb, the Cronbach’s alpha should exceed the threshold of .70. Other papers and textbooks suggests either the same threshold, or higher. In general, .70 is considered adequate, .80 is considered good/very good, and .90 is considered excellent. Scores below .70 become questionable (especially for confirming constructs) and scores below .50 are considered to be unacceptable (e.g., Cortina, 1993; Kline, 2005; Hair et al. 2010).

Finally, an extra construct reliability measure (CR) is added as a calculation of internal consistency, which is calculated by dividing the squared sum of standardized factor loadings by the squared sum of standardized factor loadings plus the sum of errors. A threshold value of .7 is used (Hair et al. 2010).

4.2.2 Goodness-of-fit

Goodness-of-fit (GOF) is a measure that indicates how well a specified model reproduces the observed covariance matrix among the indicator variables. Put differently: “model fit compares the theory to reality by assessing the similarity of the estimated covariance matrix (theory) to reality (the observed covariance matrix)” (Hair et al., 2010, p. 665). Thus, with GOF measures it is possible to test whether the theory fits the reality. When a lack of fit is detected, the plausibility of the model is questioned. Therefore a good fit is required before interpreting the validity of a construct. It should be noted that, while a good fit is a prerequisite for a valid model, it is not sufficient proof of it.

32 Incremental Fit Index (IFI) (Marsch et al., 1988; Bollen, 1990; Gerbing and Anderson, 1993; Hu and Bentler, 1995). Therefore, the NNFI, IFI, and the CFI (Comparative fit index) are reported. For all, a threshold value of .90 is used8.

4.2.3 Convergent validity

When a metric is deemed to be unreliable; checking the validity becomes a moot point, whereas a construct cannot be valid unless it is reliable (Nunnally and Bernstein, 1994). The reverse however is not true, where reliable metrics can be invalid (Cortina, 1993). Convergent validity is defined as the “extend to which indicators of a specific construct converge or share a high proportion of variance in common” (Hair et al., 2010, p. 689). The convergent validity of the construct is tested by discussing the standardized factor loadings, and the average variance extracted (AVE).

The factor loadings of the CFA show to which extend an item correlates with the latent construct. Similar to the Chronbach’s alpha, the squared sum of a factor loading is the variance that is explained by the latent variable, the rest is error variance. Interpreting the factor loadings allows discussing whether each item is represented in the latent variable, and to which degree. As a rule of thumb, a factor loading of .5 is deemed acceptable (for CFA), though higher is desired (Hair et al. 2010). When factor loadings do not meet this threshold, it becomes questionable whether the item should be included in measuring the latent construct.

Closely related to the factor loadings is the AVE, which is calculated as the average of the squared sum of the standardized factor loadings. The AVE shows whether the included items are adequately represented in the latent construct. Again a threshold of .5 is used.

4.2.4 Robustness

For robustness of the results, a principal component analysis (PCA) is used as an alternative analysis to test whether a decent factor solution can be found. The main differences between these two methods are that “CFA analyzes only the reliable common variance of data, while PCA analyzes all the variance of data. An underlying hypothetical process or construct is involved in CFA but not in PCA. PCA tends to increase factor loadings especially in a study with a small number of variables

8

34

5 RESULTS

The twelve created datasets are tested on their reliability, goodness-of-fit, and convergent validity. Furthermore, a Principal component analysis is performed for robustness. First the results of the six created metrics for formal institutional distance are discussed; secondly, the results of the six created metrics for cultural institutional distance are discussed.

5.1 VALIDITY OF FORMAL INSTITUTIONAL DISTANCE METRICS

Three metrics are created based on the Economic freedom project (EFP), and three metrics are created based on the World Government Index (WGI). The results are presented in table 2. First of all, the solutions can be deemed reliable for all six metrics (scores all exceed the threshold of 0.7). The results however diverge for the goodness-of-fit and validity. The metrics that are based on the EFP all lack goodness-of-fit, where the NNFI, IFI and CFI do not meet their thresholds of 0.90. Also convergent validity lacks in these metrics, with an AVE score below 0.50, factor loadings below 0.50 for several items (three items in profiles, six items in actual, and four items in weighted), and (in the case of profiles) even some negative correlations. The principal component analysis (where all variance is taken into account and no underlying factors are assumed up front) gives similar results. Here it is shown that the data does not suggest a unidimensional solution, since the variance extracted by a single factor solution does not meet the threshold of 65%, and in all cases the eigenvalue of the second factor exceeds 1.00. This is perhaps best shown in the communalities of the items Fiscal Freedom and Gov’t spending at the profiles metric. These are relative 0.000 and 0.061, meaning that these items do not share any variance with the other items that are included in the analysis. Considering that the metrics of formal institutional distance that are based on EFP data lack goodness-of-fit, validity, and unidimensionality, hypothesis 2 is accepted: The use of the overall score

of the economic freedom project lacks construct validity due to a lack of unidimensionality.

35 construct is not deemed to be specific enough to capture all the relevant differences between multiple countries. Thus, hypothesis 1 is accepted: The use of a unidimensional metric based on the

worldwide governance indicators is valid, which indicates that the WGI is not able to adequately represent formal institutional distance.

RESULTS

FORMAL INSTITUTIONAL DISTANCE

Economic Freedom Project Worldwide governance indicators Profiles Actual Weighted Profiles Actual Weighted

Reliability Chronbach's Alpha 0.81

V 0.79 V 0.79 V 0.96 V 0.91 V 0.91 V construct reliability 0.82 V 0.72 V 0.82 V 0.96 V 0.92 V 0.92 V Goodness- of-fit NNFI 0.77 N 0.64 N 0.51 N 0.90 V 0.91 V 0.90 V IFI 0.82 N 0.72 N 0.62 N 0.94 V 0.95 V 0.94 V CFI 0.82 N _0.72 N _0.62 N _0.94 V _0.95 V _0.94 V Validity factor loadings

Property Rights 0.98 V 0.96 V 0.70 V V&A 0.82 V 0.63 V 0.66 V

Freedom from Corruption 0.94 V 0.89 V 0.62 V PS/V 0.74 V 0.48 N 0.49 N

Fiscal Freedom -0.17 N 0.22 N 0.46 N GE 0.97 V 0.93 V 0.94 V

Gov't spending -0.29 N _0.19 N _0.25 N _{RQ 0.93} V _0.85 V _0.86 V

Business Freedom 0.73 V 0.47 N 0.68 V RoL 0.97 V 0.93 V 0.94 V

Labor Freedom 0.38 N 0.19 N 0.45 N CoC 0.96 V 0.91 V 0.91 V

Monetary Freedom 0.50 V 0.21 N 0.48 N Trade Freedom 0.54 V 0.25 N 0.53 V Investment Freedom 0.74 V _0.47 N _0.67 V Financial Freedom 0.76 V 0.53 V 0.75 V AVE 0.43 N 0.27 N 0.33 N 0.81 V 0.65 V 0.67 V Principal Components

Eigenvalue factor one* 4.96 3.61 4.00 5.07 4.20 4.28

Eigenvalue factor two 1.71 N 1.42 N 1.85 N 0.44 V 0.73 V 0.76 V Extracted variance factor one 49.60 N 36.09 N 39.98 N 84.44 V 69.96 V 71.27 V

factor loadings

Property Rights 0.90 V 0.75 V 0.68 V V&A 0.87 V 0.93 V 0.75 V

Freedom from Corruption 0.85 V 0.69 V 0.60 V PS/V 0.80 V 0.57 V 0.57 V

Fiscal Freedom -0.02 N _0.47 N _0.62 V _{GE 0.96} V _0.92 V _0.93 V

Gov't spending -025 N 0.33 N 0.33 N RQ 0.94 V 0.88 V 0.89 V

Business Freedom 0.82 V 0.49 N 0.74 V RoL 0.98 V 0.94 V 0.95 V

Labor Freedom 0.50 V _0.41 N _0.57 V _{CoC 0.96} V _0.91 V _0.91 V

Monetary Freedom 0.66 V 0.54 V 0.62 V

Trade Freedom 0.73 V 0.58 V 0.68 V

Investment Freedom 0.85 V _0.69 V _0.67 V

Financial Freedom 0.87 V 0.73 V 0.73 V

*The eigenvalue of the first factor is per definition larger than one / V _{The score meets the threshold. /} N _{The score does not meet}

the threshold.

Thresholds: reliability = 0.70 / Goodness-of-fit = 0.90 / Factor loadings and AVE = 0.50 / Eigenvalues = 1.00 / Extracted variance

of the first factor = 65%.

36

5.2 VALIDITY OF CULTURAL INSTITUTIONAL DISTANCE METRICS

Three metrics are created based on Hofstede data, and three metrics are created based on GLOBE data. The results are presented in table 3. The most noticeable results that are presented in this table are that some results with regard to the CFA solutions based on the Hofstede data seem illogical. Considering that the presented factor loadings are standardized, scores that exceed the scale (either higher than 1.00 or lower than -1.00) should not occur and are considered important indicators of problems with the data (Hair et al., 2010, p. 713). The cause of these occurring problems at both Hofstede profiles and Hofstede actual can be found in the error variance estimates of the CFA. In both cases, the power distance error variance estimate is negative (relatively -1.98 and -0.27) which implies that more than 100% of the variance of power distance is explained (Hair et al., 2010). As a result, both cases can be considered so-called Heywood cases, of which structural misspecification is among the most important causes (Kolenikov and Bollen, 2008). Misspecification as a cause for the Heywood cases is supported by the principal component analysis, where neither the eigenvalues, nor the extracted variance of the first factor, nor the factor loading scores supports a unidimensional solution. The negative variance in the error and the factor loadings (for profiles), lead to corrupted reliability, goodness-of-fit, and validity scores.

The weighted solution does not experience this problem, however in this case the correlation table is negative. Due to negative factor loadings, the construct reliability measure is not a valid estimate (i.e., it uses a squared sum of factor loadings); The use of a weighted cultural institutional distance construct, based on Hofstede data lacks reliability (CA of 0.212), convergent validity (AVE of 0.03) and unidimensionality (Eigenvalues indicate two factors, first factor extracts less only 31.89%). Interestingly, the goodness-of-fit measures seem to indicate a perfect fit of the CFA solution. This can be explained by considering that the weighted solution (almost) completely exists out of the power distance (see factor loadings). Thus, the observed covariance matrix (consisting almost completely out of power distance) and the latent covariance matrix is the same. This however does not indicate a good fit of the solution; rather it indicates severe specification (i.e., identification) issues. It is important to note that for both the described problems with the Hofstede data solutions that these problems only occur when there are serious specification issues. Would the (often used) solutions actually possess validity, than these problems would not occur in the estimation of the CFA. As a result, hypothesis 3 is accepted: The use of a composite distance metric based on Hofstede data lacks