0 ENGLISH LANGUAGE PROFICIENCY AND FDI

(1)

0 ENGLISH LANGUAGE

PROFICIENCY AND FDI

Abstract

This study analysed the effects of English proficiency on FDI stocks by using the newly available English Proficiency Index set together by Education First. The transaction cost reducing effect that a common language has on speakers was extended to business partners who engage in FDI transactions. It was argued that a better English proficiency makes business transactions easier and is thus desired by FDI-sender countries when considering where to invest. By means of regression analysis and through controlling for several variables, the findings indicated that English proficiency doesn’t significantly affect FDI stock variation.

Master Thesis 12-07-2015

Author: Alejandro de la Cuesta Supervisor: Prof. Dr. S. Beugelsdijk Referent: Dr. M.J. Klasing

(2)

Acknowledgements

(3)

List of Tables

Table 1: EF EPI Reports division and assumption of data for analysis ... 19

Table 2: Variables of study used per hypothesis ... 21

Table 3: Descriptive statistics of independent and dependent variables - hypothesis I . 31 Table 4: Descriptive statistics of independent and dependent variables - hypothesis II 32 Table 5: Descriptive statistics of major control variables ... 33

Table 6: Correlation table EF EPI 2013 and FDI Stock 2013 ... 34

Table 7: Model summary of regression analysis for hypothesis I ... 36

Table 8: Coefficient table of regression analysis for hypothesis I ... 37

Table 9: Model summary of regression analysis for hypothesis II ... 37

Table 10: Coefficient table of regression analysis for hypothesis II ... 38

Table 11: Model summary of regression analysis for hypothesis III ... 38

Table 12: Coefficient table of regression analysis for hypothesis III ... 39

Table 13: Robustness test based on analysis without outliers via boxplots ... 40

Table 14: Beta coefficients of the robustness test for hypothesis I ... 40

Table of Figures

Figure 1: Visualization of how English proficiency leads to more FDI stocks ... 15

(6)

1. Introduction

“[…] despite the billions of hours and dollars poured into teaching people English, there is little measurement of the success of these investments” (Education First, 2011, p. 3).

English is increasingly considered to be a core competency in a globalized economy (Education First, 2014), dominated by firms hiring employees from all over the world, countries being hosts to new foreign firms and local companies trying to expand beyond national borders. English also dominates the world landscape of business communication, as more and more companies such as Lufthansa, Nokia, Rakuten, Renault, Samsung and Lenovo officially make English their corporate’s lingua franca (Yuanqing, 2014) (Education First, 2014). This doesn’t come as a surprise, as two billion people will be learning English in the next decade (Education First, 2011). With English dominating so many aspects of our lives (Crystal, 2003) and being used in businesses, mergers, politics, schooling, etc. (Yuanqing, 2014), one might wonder if this might have an effect on a country as a whole, specifically on the attractiveness of a country for FDI investment: The main topic of this paper is to analyse this possible relationship.

The first ever attempt at a worldwide ranking of countries by their English proficiency was published in 2011 by Education First on their first edition of their Education First English Proficiency Index. They have continuously ranked 63 countries based on their English proficiency and have created an index of classification (from very low to very high proficiency). With this new ranking, new possibilities of study of the effects the English language has on countries are available.

This thesis expands the existing literature on language and FDI by being the first study that quantifies the level of proficiency of a language and correlates it with economic indicators. Previous language and FDI literature has focused exclusively on analysing whether a common language between FDI partners is beneficial, disregarding the spoken quality of the language. Proficiency lies at the core of my thesis, and is the main difference between similar studies and mine.

(7)

(8)

2. Literature review

This chapter explores the relationship of language and culture and explains to what extent the English language is an important medium of communication in the world today. A way to measure the proficiency of countries in this language is also presented.

2.1 Language as a bridge between cultures

Culture and language go hand in hand. A common culture and a common language facilitate economic exchange between two parties (Lazear, 1999). However, usually just the “cultural” aspect of individuals and of countries is analysed. Cultural differences are in the spotlight of country effect studies but the language aspect is often not given further analysis.

Much of the existing culture-related literature highlights the benefits or drawbacks of cultural differences in the workplace (Earley, 2004) (Stahl, Mäkelä, Zander, & Maznevski, 2010) (Roth, Kostova, & Dakhli, 2011), as well as the lack of cross-cultural competence of individuals (Johnson, 2006) and the existence of cultural “friction” that may difficult swift interactions (Shenkar, 2001).

It is these cultural differences, based on Hofstede’s cultural dimension studies (de Mooij & Hofstede, 2010), that indeed can affect how individuals from different nationalities interact with each other. Individuals tend to prefer to deal with cultures similar to their own (this is the usual finding of culture studies).

(9)

2.2 The worldwide reach of English

The English language is deeply rooted in our modern society, and it extends beyond the borders of almost all of the world’s countries to degrees we might not even account for. Often disregarding the official language of a country, English finds its way into many national education policy programs, as it is increasingly being taught in schools and offered as an extracurricular class. It is also present in the World Wide Web and the news media (Graddol, 2006) (Crystal, 2003).

English extends its reach especially to the business world, as English is the go-to language for several occasions such as meetings, negotiations and e-mail (Nickerson, 2005). It is a language assumed to be understood by the vast majority of people. A 2007 survey of over 10.000 non-native English-speaking employees in international corporations indicated that half of them were using English on a day-to-day basis. In fact, only 9% of employees indicated that they didn’t use English at work (Education First, 2011).

But how many people speak the language? Ethnologue is a website that actively tries to quantify the total number of speakers of any given language worldwide. They currently rank English as the third most spoken language according to native speakers (335 Million). If English as a foreign language is taken into account in the estimate, the number rises to around 840 Million total speakers, which gives it a second position in the worldwide scale of spoken languages (behind Mandarin Chinese with a total of 1.2 Billion speakers) (Lewis, 2015).

This finding is similar to that of the British Council in their 2006 Report, where English ranked as the second most spoken language worldwide. What is clear from both sources however, is that the number of native speakers does not reflect the reality of this world language, and are as equally important as the number of people who speak English as a secondary language (Graddol, 2006).

(10)

but are useful nonetheless for understanding the worldwide scale of English and its clear importance in the world.

Why English has come to this degree of universality is not studied in this thesis, but the fact that it is now the language that drives the world in terms of internationality and globalisation is a certainty. As Graddol (2006, p. 22) puts it, “On the one hand, the availability of English as a global language is accelerating globalisation. On the other, the globalisation is accelerating the use of English.”

2.3 The benefits of speaking English for the speaker

Speaking English is a modern tool of communication, often used in business-related topics. It has become the standard vehicular language for business transactions and business communication, a lingua franca (Nickerson, 2005) (Yuanqing, 2014) (Education First, 2014). Speaking English is linked to a myriad of benefits for the speaker, such as communication towards a broader public and attractiveness for job applications (Education First, 2013).

In fact, speaking English yields higher salaries. A worker in India who is fluent in English has an average hourly wage 34% higher than his peers. Even low proficiency is remunerated 13% higher than to those who have none (Education First, 2014, p. 30). These findings are in tandem with those of another study by McManus et al. (1983), where a higher proficiency in English was linked to higher earnings for Hispanic men working in the US (wage differences were completely explained by English proficiency and not by ethnicity, national origin or time in the US).

(11)

2.4 Measuring English proficiency worldwide – The EF EPI

Education First (EF) is an international education company focused on language, academics and cultural exchange programs. It is present in more than 50 countries with 500 schools and offices (Education First, 2014).

The Education First English Proficiency Index (EF EPI) is published by Education First, and it attempts to systematically quantify the level of English proficiency by country in a worldwide scale. Education First has released a total of four EF EPI Editions (the current one being the 2014 edition), in which up to 63 of the world’s countries have been ranked.

The EF EPI gives a measurable English proficiency evolution index based on the total input from more than five million adults from all parts of the world since 2007 (Education First, 2013). These are substantial amounts and should make the EF EPI a solid indicator of the English proficiency in a country.

The Report categorizes countries into five English proficiency categories, which range from very low to very high. No nations where English is the well-established native tongue participate in the ranking, as they would always be ranked first and their participation be deemed redundant. There are however countries included in which English is one of the official languages, but not-so-high EPI scores are achieved by the country’s test takers, which is the case for Malaysia, Hong Kong, Singapore, Pakistan and India.

The highest English proficiency country is currently Denmark, with an English proficiency score of 69.30. Second and third place are for The Netherlands (68.99) and Sweden (67.80) respectively. Countries like Spain (57.18), Slovakia (55.96) and Russia (50.44) neither excel nor fall behind at English proficiency levels. On the lower side of the spectrum there is Egypt on the 56th_{place (42.13), Algeria on the 60}th_{(38.51) and}

(12)

This has been done in the EF EPI in order to add meaning to the scores past their mere numeric value. For an in-depth definition of what each band means for the speaker, along with a visualization on a map, refer to appendix C. The proficiency bands (A2, B1, B2) are however not considered further in this thesis, as only the proficiency scores will be used for analysis.

Even more interesting than individual scores are score developments over time (these can be found in appendix B). The EPI Scores have changed over the years, with some countries bettering themselves and some worsening in terms of their spoken English proficiency.

(13)

3. Theoretical framework

There are several determinants of study that lead me to hypothesize that FDI is linked to English proficiency. The logical thinking of this relationship is elucidated in the following paragraphs, along with a review of some of the existing language and FDI literature.

The effects of a common language (let alone the proficiency of that language) on bilateral trade or FDI hasn’t received as much attention in the past (Melitz, 2008) (Melitz & Toubal, 2012) as other indicators, as the quote below exemplifies:

“Languages are at the very heart of business communication and negotiation. Yet, while a significant literature examines how institutions, common currency and trade agreements affect international trade and foreign direct investment (FDI) flows, few studies systematically analyse the effect of language on those flows.” (Hoon Oh,

Selmier, & Lien, 2011, p. 732).

3.1 Examining foreign direct investment Foreign direct investment is defined as:

“[…] cross-border investment by a resident entity in one economy with the objective of obtaining a lasting interest in an enterprise resident in another economy. The lasting interest implies the existence of a long-term relationship between the direct investor and the enterprise and a significant degree of influence by the direct investor on the management of the enterprise. Ownership of at least 10% of the voting power, representing the influence by the investor, is the basic criterion used”

(OECD, 2013). See also Financial Times, 2015; OECD, 2015.

FDI can thus be a measure of the investments a country receives (as an FDI-receiver) from another country, or the investments a country makes in foreign nations (FDI-sender). To measure this, the data speaks of FDI flows and FDI stocks. An FDI flow can be an inward flow (when a country receives FDI) or an outward flow (when a country invests through FDI in a foreign country). They are usually represented as yearly flows (OECD, 2015).

(14)

the reporting economy abroad) (OECD, 2013). For this thesis I will focus on FDI inward stocks.

FDI inflows (and by extension also stock) are beneficial for a country, as they have been linked to contribute to economic growth by a multitude of authors (Makki & Somwaru, 2004) (Aizenman, Jinjarak, & Park, 2011) (Borensztein, J, & J, 1998) (Weisbrod & Whalley, 2011) (Cipollina, Giovannetti, Pietrovito, & Pozzolo, 2012). Thus, with a positive correlation, more FDI inflows are better for a country’s economic growth prospects than fewer FDI inflows. This correlation is especially strong for developed countries (Beugelsdijk, Smeets, & Zwinkels, 2008).

It has been noted however by a few authors (Herzer, 2012) (Lipsey, 2007), that FDI stocks and flows are not an accurate measure of value-adding activity. In fact they can even be a biased measure of the magnitude of the value-adding activities performed by multinational enterprises abroad (Beugelsdijk, Hennart, Slangen, & Smeets, 2010). This is due to the fact that not all FDI inflows into a country result in value-adding activities, as countries can be used as tax havens (as an example).

(15)

3.2 The gravity model and its significance

Extant language and trade literature bases its analyses and conclusions on the gravity model (Egger & Pfaffermayr, 2004) (Melitz, 2008) (Fidrmuc & Fidrmuc, 2009) (Melitz & Toubal, 2012) (Goldberg, Heinkel, & Levi, 2005) (Lankhuizen, de Groot, & Linders, 2011). This model is widely used to explain bilateral trade and FDI investments between countries.

What this model’s theory states is that countries which have several characteristics in common are more likely to trade with one another. This is done via analyses including factors such as common official language, common colonial relationships and common institutions, as well as differences between countries in the sense of institutional distances, cultural distances, etc. The effectiveness of the model is well documented, as a common language between countries (Melitz, 2008), or a small geographic distance between capitals increases trade flows between countries and attracts more FDI (Egger & Pfaffermayr, 2004) (Kim et al. 2014).

Most authors use its basic assumptions and add additional variables of their own based on their specific research. Likewise, I will draw similarities from existing literature and add variables needed in my own research (such as the percentage of internet penetration). Thus, I will be applying the gravity model for this thesis based on FDI and language factors.

3.3 Language and FDI – A matter of transaction costs

(16)

robust relationship between a common language and FDI. The reason for this relationship is argued to be related to transaction cost economic theory. Through a common language, transaction costs of the FDI investment are reduced and thus a country is more attractive for these kinds of investments.

Although the study points to the fact that transaction costs are not the sole reason for FDI decisions, it shows that a language barrier generates a friction that FDI senders would seek to minimize (Kim et al., 2014). Thus, from a transaction cost economics point of view, speaking a common language is an efficient way of attracting FDI investments, by reducing communication costs. This is especially true if business partners meet face to face in order to discuss investments (Goldberg, Heinkel, & Levi, 2005), as well as by altering national language policies so that the FDI sender’s official language is taught in the schools of the FDI receiving country (Kim et al., 2014) (see the examples illustrated by the authors of the Philippines teaching Spanish, Singapore expanding its English proficiency or Indonesia teaching Chinese).

A common language promotes trade between countries (Melitz, 2008), as it makes it easier to interact and offers a degree of efficiency in any transaction that both countries undertake (Fidrmuc & Fidrmuc, 2009). This is a desired situation for FDI trade, as transaction costs would be lower thanks to the commonality of language (Williamson, 1979). In fact, two individuals who engage in a trade of any kind benefit from a common language to communicate (regardless of proficiency), because it eliminates the need for an intermediary or a translator, which would increase costs and maybe hinder otherwise beneficial mutual transactions from occurring (Fidrmuc & Fidrmuc, 2009). Extending this idea and incorporating the use of English (one of the most spoken foreign languages in the world) as a medium of communication between two business entities, we can understand the positive effect that this situation brings for both parties and for the country in which the business (assuming FDI) is taking place.

(17)

related context derives from its capacity to lower transaction costs (Selmier II & Hoon Oh, 2012). English is thus (when a common language is missing), the preferred choice of communication and the most efficient and cost-reducing language to fall back to (assuming the parties in question speak it).

As is evident from the existing literature, there is a link between a common language and FDI decisions, explained under the scope of transaction cost economics. Furthermore, language and FDI are not two unrelated variables when companies decide where to locate investments. This thesis tries to give more insight into this fact, by considering English proficiency as a means of FDI location determination.

3.4 English and FDI – Where is the relation?

The EF EPI Reports have shown that there is a strong and remarkably stable-over-time correlation between a country’s English proficiency level and its overall wellbeing (Education First, 2014). However the reason as to why better English-speaking countries enjoy a higher GNI, Human Development Index, GDP, ease of doing business, etc. is unclear. How does this economic growth and bettering of a country define itself? I wish to expand on these findings by analysing if English proficiency is linked to FDI, measured by FDI inward stock.

Taking the fact that speaking English is beneficial to the speaker (it yields higher salaries, as evidenced in the previous section), it is a logical assumption to extend this finding to the country-level. Countries are populated by individuals, and it is these people who can have a high or low proficiency. By English being beneficial for the individuals, it should be beneficial for the country as well, as similar studies point out (Kim et al., 2014) (Hoon Oh et al., 2011) (Selmier II & Hoon Oh, 2012). Furthermore, it has been proven that “US FDI into foreign countries is greater if the foreign country has English as its mother tongue” (Goldberg, Heinkel, & Levi, 2005, p. 934). I argue that it is beneficial to have a high level of English proficiency for a country (in general), as it will attract more FDI. Why this is so is explained in the following paragraphs.

(18)

attracts the attention of FDI-sender companies, which are trying to decide on where to invest in FDI.

These companies would favour to allocate FDI to countries with a higher level of proficiency and cooperate with English speaking business partners. This would decrease communication costs (Kim et al., 2014) and in turn also transaction costs (when no common official language between both parties is present).

In fact, when there exists no common language between two parties, they can always resort to their shared knowledge of English to reduce transaction costs (Williamson, 1979), thereby facilitating FDI transactions. Furthermore, English proficiency in a country is shown to correlate with the ease of doing business in that country (Education First, 2014), which is undoubtedly a factor that also adds to a country’s attractiveness for FDI.

Referring to the study results obtained by Selmier II & Hoon Oh (2012), speaking English is a language-cost reducing method of communication. This is due to the fact that high communication costs arise when two countries/individuals who engage in communication (be it for FDI purposes for example) each speak different major trade languages (like French, Arabic or Spanish). Under this circumstance, it is more beneficial to switch the conversation to a language that both speakers understand (Melitz & Toubal, 2012) (Kim et al., 2014). English is the clear winner from this holdup, as it is likely that both parties speak it (which is expected from the world’s

lingua franca) (Selmier II & Hoon Oh, 2012).

Extending this finding further than to a pair of individuals, a group of 4 people (each person from a different country) is more likely to have to switch to English as the language of communication, as there is an increased chance that only English is the shared language that they would all understand.

(19)

Figure 1: Visualization of how English proficiency leads to more FDI stocks

In my thesis, I focus on analysing solely the effects that English proficiency can have on the FDI inward stocks of countries. The argument that leads me to hypothesize that English language proficiency is linked to FDI inflows is hence based on transaction cost economics, with lower transaction costs arising through the use of the English language during FDI negotiations.

3.5 Hypotheses of study

In the following section, three hypotheses stating the relationship between FDI and English language proficiency will be formulated:

Firstly, the link between English proficiency and different economic indicators has been documented by the EF EPI (Education First, 2013). Likewise, the relationship of common languages between FDI investors has been discussed in the previous sections and is well documented in the gravity model. I extend on these effects by stating that the relationship between English proficiency and inward FDI stock is positive.

I argue that the better a country speaks English, the more attractive it becomes towards FDI investments. English proficiency thus has a positive effect on FDI stocks. This argument is based on transaction cost theory, stating that firms will be attracted to investing in countries with a good English proficiency due to decreasing transaction costs (since firms seek to lower these costs, see Williamson, 1979), brought forward by an ease of communication through the business partners’ knowledge of English.

This leads to more inward FDI stock investments

Foreign business partners are attracted by high levels of proficiency

from potential local partners, drawn by lower transaction costs

(20)

The proficiency aspect is the new variable not present in previous studies in this degree of significance and robustness of data. I expect a positive correlation between my two variables. Thus my hypothesis I proposes:

Furthermore, due to the transaction cost reducing benefits of a common language that is well spoken (high proficiency), I argue that an increase of the proficiency of English has a positive effect on FDI investments. Put differently, when a country betters its proficiency in English, it will attract more FDI investment than before. This is a natural extension of the first hypothesis but can be tested differently due to the delta variable that will be employed (further details in the next chapter).

My expectation is that as the English proficiency of a country improves over time, the attractiveness of the country for FDI stocks becomes larger due to lower potential transaction costs (Williamson, 1979). This promotes inward FDI stocks for that country. This is a new theory not present in the gravity model, which usually tests for the commonality of languages between countries. It doesn’t account for the proficiency of them and neither for the evolution of the proficiency and the degree of influence it can have on FDI.

Thirdly, in order to single-out the effect of English proficiency, I will analyse on a set of selected countries from the sample if they attract more FDI from countries with English as their native tongue. I will analyse dyads between English-speaking nations and countries with high and low proficiency scores. Thus, I am considering if FDI from English-speaking countries (i.e. countries which have English as their native tongue)

Hypothesis I: The level of English proficiency has a positive effect on the level of

FDI stocks of a country.

Hypothesis II: An increase of a country’s English proficiency is followed by an

(21)

concentrates more in countries with high English proficiency levels (instead of in countries that speak a poor English).

My expectation is that when a country has a high English proficiency, it will attract more FDI from English-speaking nations. This has the underlying logic that these FDI-senders would prefer to invest in a country which has a higher proficiency score than in a country which does not. Again, this is argued to be due to the transaction cost reducing effect that English as a medium of communication would have for English-speaking countries seeking FDI investment.

According to Goldberg (2005), the US prefers to invest in countries which have English as their mother tongue. I am extending this finding by arguing that the US (and other countries with English as their mother tongue) would also prefer to invest in countries which have a high proficiency in English. This hypothesis is thus similar to the first hypothesis, only that now dyads are considered in the analysis. Combinations of FDI-sender and FDI-receiver countries will be analysed. This wasn’t the case in the previous hypotheses.

All three hypothesis stem from the idea that English, as a vehicular language in business transactions, is a desired mean of communication when there is no common official language between business partners. English thus, based on transaction cost economics, reduces the transaction costs of the FDI investment, and better English reduces them more than average English. This is desired by both business partners.

Hypothesis III: Countries which have a high level of English proficiency attract more

(22)

4. Methodology and data

In order to tackle this topic I will focus on different aspects, presented in this chapter. It is important to note that the ultimate goal of this research is to prove or disprove the possible relation between a country’s FDI attractiveness (measured by total FDI inward stocks) and their English proficiency. Appropriate variables and statistical analyses have been selected in order to achieve this goal.

4.1 The EF EPI timeframe

The EF EPI scores will be taken as representative of a country’s whole adult population proficiency level (this assumption adds limitations to this study, which are explained in detail in chapter 8). The fact that children and teenagers are not considered in the ranking is of benefit to this study, since they would not be involved with FDI transactions anyways due to being under aged and are not part of the economically active population.

There are currently 4 different EF EPI reports (editions of 2011, 2012, 2013 and 2014). Each report presents one ranking in which each participating country is assigned one proficiency score. As an example, Japan scored 54,17 in the 2011 edition, 55,14 in the 2012 edition, 53,21 for the 2013 edition, and 52,88 in the 2014 edition for its proficiency level. There is thus no more than a total of four score values for Japan (and any other country) across all four EF EPI editions as of the time this thesis was written. I assume however, that the ranking is representative for the year before each edition’s

stated release year, since all gathered test data was taken from years prior to the

edition’s release year by EF (Education First, 2014, p. 44) (e.g. the 2014 edition presenting its newest ranking is based on 2013 test data from 750.000 test takers and is thus representative of the year 2013). Unfortunately, not all editions gather input data from just one single year prior to their release. The details are presented below:

The first edition (2011 edition) gathers test data from the years 2007-2009 and

aggregates it in order to release one score per country in its ranking (there is only 1

(23)

only one year prior to their release and also present new rankings. Please refer to the Japan example above in case of doubt of what the end result of this is.

Table 1 illustrates the data gathered by the different EPI reports and my assumptions about what years this data is representative of. In order to have a 4 year consecutive timespan, I chose the representative years to span from 2010 to 2013. I adjusted the matching FDI data I will consider accordingly.

Table 1: EF EPI Reports division and assumption of data for analysis

4.2 Sample of study

I am limited to analysing only those countries that have been present in all four EF EPI editions for the testing of my hypotheses. Due to the difficulty of retrieving the data for the percentage of internet penetration for Taiwan (along with other variable data), I have decided to exclude it from my analysis. Furthermore, I will exclude Costa Rica, The Netherlands, Panama and Switzerland from all my analyses, as they are well known tax havens (Gravelle, 2015). Including them would skew my results as they are based on FDI stocks, which can be influenced by a country’s tax policy.

The maximum number of countries analysable in this study is 37, due to the excluded countries mentioned in the paragraph above. The sample cannot be bigger because these countries are the only ones that have been present in all 4 EF EPI editions. It is necessary to remain consistent with the sample countries and thus all hypotheses are based on sample size variations of these 37 countries. The sample size is different based on which analysis will be done: Refer to appendix D for a list of the countries employed in each hypothesis.

Hypothesis I: The sample size is 37 countries, which is the maximum number of countries present in this analysis.

EF EPI Edition Relevant test data gathered for EF EPI ranking in that edition

I assume EF EPI score representative of year

Matching FDI data I will consider in my model

(24)

Hypothesis II: There were 2 outliers which had to be deleted due to the transformation not having normally distributed residuals for the dependent variable (a necessity for linear regression to produce robust results). Thus, for hypothesis II, the sample size is reduced from 37 countries to 35.

Hypothesis III: Due to the dyadic nature of this analysis, the sample size is theoretically increased to 80. This is due to having four English-speaking FDI-sender countries (USA, UK, Canada and Australia) and 20 EF EPI ranked (top 10 and lowest 10 according to their EF EPI scores) countries in the analysis (4x20 dyads). However, due to a total of 14 FDI values from the sender countries in the receiver economies being coded as zero (due to no reported FDI inward stocks by the receiver economy), the sample is reduced to a total of 66 dyads. Thus, the regression analysis will be employed with a total of 66 active dyads.

4.3 Variables of study

I will study the proposed hypotheses making use of several variables, depending on which hypothesis is being tested (both control variables, as well as dependent and independent variables, change depending on the hypothesis analysed). The dependent and independent variables used for all three hypotheses are listed in table 2, along with the corresponding control variable group.

(25)

Hypothesis Dependent variable used Independent Variable used Control variable group used

Hypothesis I FDI inward stocks 2013 EF EPI Score 2013 Group 1 Hypothesis II Delta FDI inward stocks 2010-2013 Delta EF EPI Score 2010-2013 Group 2 Hypothesis III English-speaking nation’s FDI inward

stocks in receiver country 2012 EF EPI Score 2012 Group 3

Table 2: Variables of study used per hypothesis

Regarding FDI data, FDI stocks instead of FDI inflows was chosen because previous correlation tests turned out to be not promising. This was due to the fact that the way FDI inflows are reported on all major databases such as the World Bank or UNCTAD (new investment inflows less disinvestment) allows for negative inflows on certain years. This is not in the interest of this analysis, as it is not expected that a decline in English proficiency is followed by a disinvestment in the country. Thus, FDI stocks were chosen to test my hypotheses and provide more solid results.

All employed dependent and independent variables are described below. All countries in this study have been assigned one value per variable:

 FDI inward stocks 2013: This is the dependent and continuous variable for hypothesis I. It is measured in US$ Millions. Just like in present FDI and language literature, this variable is the dependent one and thus crucial to the whole analysis.

 EF EPI Score 2013: This is the independent variable for hypothesis I. It is measured in a scale of 0 to 100. A higher value equates to a better English proficiency. The variable is thus presented in a continuous form. The fact that it is a ranking based on points is a relatively new concept for language and FDI studies.

 Delta FDI inward stocks 2010-2013: This is the dependent variable for hypothesis II. It is measured in US$ Millions and is thus continuous. It is calculated by subtracting the FDI inward stock value of 2010 from the 2013 value.

(26)

EF EPI 2013 value and subtracting the EF EPI 2010 value from it for each country.

 English-speaking nation’s FDI inward stocks in receiver country 2012: This is the dependent variable in the third hypothesis. It is measured in US$ Millions and is continuous. It is a narrowing of the FDI inward stocks values to only considering values from English-speaking nations (UK, USA, Canada, and Australia) towards FDI-receiver countries. Bilateral FDI data is important in the gravitational model and is thus included in this analysis.

 EF EPI Score 2012: This is the independent variable of the model testing the third hypothesis. The year of analysis is 2012 and the variable is presented in a continuous form. The scores are ranked in the same way as the EF EPI 2013 scores, i.e. from 0 to 100.

The most widely used model to explain differences in FDI locations is the gravitational model. I will use the model’s standard control variables but will also take others into consideration. All employed control variables are described below according to the variable group they are grouped in:

Group 1 control variables (employed for hypothesis I)

 Economic Size GDP 2013: This variable is a good predictor of FDI since it is one of the crucial components of the FDI gravity model employed by several authors. It is presented in US$ and is thus continuous.

 Internet penetration 2013: As the EF EPI scores are based partly on online internet test results, the access of a country’s population to this medium can skew my analysis (see chapter 8 for details). This variable must be controlled for this reason. It is presented in a scale from 0 to 100 (representing how many individuals have internet access per 100 people in the country of analysis), and is thus continuous.

(27)

transaction costs (Kim et al., 2014), and these are hypothetically lowered when the ease of doing business in a country is better.

 Level of Democracy 2013: This is a categorical (ordinal) variable. The scores range from 0 to 20, with 20 meaning a country is very democratic, and 0 meaning it is an autocracy. As noted by Jensen (2008), it is also important in FDI models to account for the level of democracy of a country.

Group 2 control variables (employed for hypothesis II)

 Delta GDP 2010-2013: This is a continuous variable measured in US$. It is calculated by drawing the delta/change between the 2010 and 2013 values of GDP for each country. This variable must be considered in gravity-based regression models.

 Delta internet penetration 2010-2013: This variable is the delta of the 2010 and 2013 values of internet penetration. It is a continuous variable and is needed in this analysis to account for the effect of internet penetration on the EF EPI scores.

 The ease of doing business ranking and the level of democracy control variables from group 1 are also included in group 2. They are unchanged from their values of group 1. Since hypothesis II features deltas, it is assumed that the ease of doing business and the level of democracy in countries has not shifted between 2010 and 2013 in my model. The deltas were not drawn due to the 2010 data not being available for the ease of doing business ranking, and the level of democracy data for previous years being represented in a graph by the source website and not being easy to adequately measure. Hence, just the newest 2013 level of democracy and ease of doing business data were considered for this hypothesis.

Group 3 control variables (employed for hypothesis III)

(28)

 Distance between capitals: This continuous variable is part of the gravitational model and it usually yields a negative coefficient in relation to FDI (see my analysis results as well as Lankhuizen et al., 2011 and Goldberg et al., 2005). This is expected, as the further away the capitals of two cuntries are, the less trade they will engage in according to the gravity model. This is because it can be a barrier to trade. Along with contiguous borders, this variable has important linguistic implications (Melitz, 2008), as geographically near countries might have similarities in language. Therefore, it cannot be missing in studies analysing language. It is represented as km from each nations’ capitals for each dyad.

 Economic Size receiver (GDP) 2012 and Economic Size sender (GDP) 2012: These continuous variables were used in order to account for the effect of GDP in this hypothesis, as it is a component of the gravitational model and should be present in regressions analyses involving FDI as the dependent variable. They are measured in US$.

 Internet penetration receiver 2012: Much like the previous internet penetration 2013 variable from hypothesis I, this variable is ranked from 0 to 100, with higher scores meaning there are more users with access to the internet. The variable is continuous and is required in this hypothesis due to the nature of the EF EPI tests being online-exclusive.

 Contiguous borders: This categorical (binary) variable is part of several gravitational models and has been proven to affect FDI investments (Melitz, 2008). Thus, it is included in my hypothesis as a control variable. It is coded as 0 for “non-contiguous” and 1 for “contiguous”.

 Common civilization: This categorical (binary) variable represents the differences in common civilizations that can account for FDI investment decisions between countries. The paper by Kim et al. (2014) controls for it as well, as it can be a barrier to trade and thus should be considered as part of the gravity model. It is coded as 0 for “no common civilization” and 1 for “common civilization”.

(29)

analysis. It is coded as 0 for “no common institutions” and 1 for “common institutions”.

 Common official language: This categorical (binary) variable was included in order to control for the positive effects a common language has on trade and FDI flows (Goldberg, Heinkel, & Levi, 2005) (Melitz, 2008). There is one country dyad with a common language in my analysis (Canada-Belgium). It is coded as 0 for “no common official language” and 1 for “common official language”.  The control variables WTO members and previous colonial relationship are

documented as also being important when considering FDI models and were initially considered for my analysis. However, they were found to be constant for all dyads. They have been removed from the analysis and are not presented.

4.4 Method of analysis Hypothesis I

Regression analysis will be used in order to see possible relationships between the EF

EPI 2013 (independent variable) and FDI inward stocks 2013 (dependent variable) for

the selected countries. Since this isn’t a dyad analysis, the first group of control variables will be used. Thus, gravitational model bilateral data such as common institutions, bilateral FDI stocks or geographic distance is not included in the model tested for this hypothesis. This was excluded due to the large amount of data that would have to have been gathered in order to create the coefficient table including all bilateral data. The model however includes other relevant control variables such as GDP in the receiver country. The sample size for this hypothesis is 37.

Thus, in this hypothesis, countries which do not speak English as their first language receive a ranking score of their proficiency and this is regressed with the total FDI inward stocks present in the country. I am comparing all countries in the ranking by analysing which ones have larger FDI stocks. The year of analysis is 2013.

Hypothesis II

(30)

to 2013, generating the independent variable delta EF EPI Scores 2010-2013. This will be regressed with changes in inward FDI stocks in that same timeframe. Thus the dependent variable is delta FDI stock 2010-2013. This way, I will be able to test if there is in fact a causation effect present throughout the years. The timeframe is 4 years (2010-2013) for the 35 countries of analysis. The change in EF EPI scores in this timeframe can be seen in appendix B.

Since this also isn’t a dyad analysis, the second group of control variables will be used. Similar to hypothesis I, bilateral gravity model control variables are not employed in this analysis.

Hypothesis III

To test the third hypothesis, I will look at inward FDI stocks by countries which have English as their mother tongue into different countries with both high and low EF EPI scores. Regression analysis is used again for this hypothesis. At my discretion of choice, I will analyse the following FDI-sender countries: United States of America, The United Kingdom, Australia and Canada. The dependent variable thus being English-speaking

nation’s FDI inward stock in receiver country.

I will analyse the inward FDI stock of these countries into the top 10 and bottom 10 countries with the highest/lowest EF EPI scores respectively (it is thus a dyad between English-speaking nations and non-English-speaking nations). However, it must be noted that bilateral FDI data is only available for the year 2012, and thus also the EF EPI scores of the year 2012 will be considered. The independent variable is hence EF EPI

scores 2012. The analysis year is thus 2012.

Since bilateral FDI inward stock data for Guatemala and Saudi Arabia (low EF EPI scores) isn’t available for the year 2012, I will substitute these countries with the next lowest in the EF EPI rank, which are Peru and Brazil. The countries analysed are thus Sweden, Norway, Denmark, Austria, Finland, Poland, Hungary, Malaysia, Belgium and Germany for the top 10, and Brazil, Peru, Mexico, Turkey, Chile, Colombia, Ecuador, Venezuela, El Salvador, Thailand for the lowest 10.

(31)

variables accounts for bilateral data such as common official language or geographic

distance between capitals. The sample size for this analysis is 66 (refer to chapter 4.2).

4.5 Statistical assumptions

There are certain statistical assumptions that should be met before certain analyses are made, in order to be able to draw conclusions about a population based on the regression analysis done on a sample (Berry, 1993). The four assumptions of interval data and independent observations, normality of residuals, linearity and homogeneity of variance are deemed crucial in a regression analysis (the analysis employed in this thesis), since we are talking about a parametric test (Field, 2013). The results of how the data meets these assumptions are presented below.

 Interval data and independent data

Parametric tests require at least interval or ratio data to be used for the dependent variables (i.e. continuous data) which is the case in this thesis, as equal intervals in the data represent equal differences in the scale they measure. Thus, the sampled data meets this assumption (Field, 2013).

The Durbin-Watson test shows that the errors in the regression model are uncorrelated and independent as well, since the values are above the lower bound of accepted values (Savin & White, 1977) (details in appendix E). The assumption of independence is thus met.

 Normality of residuals

The test for normality in a linear regression refers to the residuals of the dependent variable in the model, which must be normally distributed (Field, 2013). Along all three hypothesis, this assumption was met by all dependent variables, as can be seen in appendix E in the normal P-P Plot of regression of the standardized residuals. It is also worth noting that the Kolmogorov-Smirnov test for normality also shows that all dependent variables are normally distributed as is.

(32)

The models show homogeneity and linearity of variance based on the scatter graphs that can be found in appendix E. There are a few outliers, but in general this assumption is also met. These outliers will be examined in the robustness tests.

4.6 Data transformation

In order to achieve the required normality of residuals of the dependent variable (needed for regression analysis, see chapter 4.5), all variables have been transformed using a logarithmic Log10 transformation. This corrected the distributional problems. Since

every single variable received the same transformation, the data won’t change relationships between variables. It is thus a safe procedure for regression analysis (Field, 2013).

After this transformation, the data was standardised in order to prevent any distortion of the output due to the differences in magnitude of the variables (Field, 2013). Thus, the reported data are zscores (standardised data) of transformed variables.

4.7 Data sources

Regarding the EF EPI scores, the data stems from the different editions of the EF EPI. To create these country rankings, they used test data from over 750,000 adults (at least for the 2014 edition), aged 18 and above, who took their English tests in 2013. Only countries with a minimum of 400 test takers were included in the index (Education First, 2014, p. 42).

The test scores are calculated by the EF EPI as follows: “[…] each test score was normalized to obtain a percentage correct for that test according to the total number of questions. All the scores for a country were then averaged across the two tests, giving equal weight to each test”. (Education First, 2014, p. 42). Thus, an EF EPI score of 50,15 (China) indicates that the average Chinese adult has a low English proficiency, as he could only properly answer an average of 50% of the tests’ questions.

(33)

 The first is an EF online test that is open to any user for free at all times. It is a 30 question adaptive exam, which means that the questions are adjusted in difficulty according to the test taker’s previous correct and incorrect answers.  The second is an online placement test that is used internally by EF during their

enrolment process for English courses that the company offers. Its length is 70 questions.

FDI inward stocks for the period of 4 years used in my analysis were retrieved from the

UNCTAD (2013a), which features an extensive database regarding this indicator.

Inward FDI Stocks from English-speaking nations into FDI receiver (hypothesis III)

will be provided likewise by the UNCTAD (UNCTAD, 2014). For individual datasets, I will use the reported FDI inward stock by the receiver economy in the database, instead of the reported FDI outward stock by the sender economy. This is because I found the latter to lack certain countries. The UNCTAD (2013b) and the CEPII (Gouel, Guimbard, & Laborde, 2012) both state that these numbers often vary due to reporting discrepancies between countries and ways of calculation. However I find (just as the CEPII recommends) that using inward data is more precise, as the host economy is more likely to accurately report inward flows than the investing one (Gouel et al., 2012, p. 11). I will remain consistent throughout my thesis with this method of approach. For dyads where the receiver country hasn’t reported a number I will consider the value to be zero. Likewise, as proposed by the CEPII, I will deal with negative values by coding them as zero (Gouel et al., 2012, p. 12).

GDP (in nominal terms) data was retrieved from the World Bank website (World Bank,

2015a). The percentage of Internet penetration (for all years) and ease of doing business

ranking was retrieved from the World Bank website as well (World Bank, 2015b)

(World Bank, 2015c).

The CEPII provides enough information about the common official language and

geographic distances between capitals, as well as colonial information (Mayer &

Zignago, 2011).

(34)

Agency (PITF, 2014). The original ranking spanned from -10 to 10. I added 10 points to each value in order to avoid negative datasets. Hong Kong data is not available from the Polity Project. The CIA World Factbook (CIA, 2015) lists Hong Kong as a “limited democracy”. As such, I will rank Hong Kong with the lowest still-democratic score (according to polity data) of 16.

Staying in tandem with the study by Kim et al. (2014), I draw my classification for the

same civilization data from a study by Russet et al. (2000), which classifies the world’s

countries by culture.

Data for the institutional similarities was obtained from the latest institutional profiles database (the IPD 2012, sponsored by the CEPII), and takes certain concepts such as political institutions, functioning of public administrations and security of transactions and contracts into account (among others) (Bertho, 2013).

(35)

5. Empirical results

This chapter includes the empirical results to all descriptive statistics, correlation analyses and linear regression tests, with were performed for all three Hypotheses. The results of these tests are presented below. All numbers have been rounded up to maximum three decimal points. Some large values with decimal points are rounded up to cero decimal points in order simplify the tables.

5.1 Descriptive statistics

The description of basic statistical data is presented in this subchapter. It must be noted that the variables presented here are in their untransformed form (i.e. before their undergone logarithmic transformation for the linear regression).

Dependent and independent variables for hypothesis I

Key characteristics about the independent and dependent variables for hypothesis I can be observed in table 3. The average EF EPI Score for the year 2013 is around 54, which sits right in between moderate and low proficiency in the EPI ranking (Education First, 2014). FDI stock for 2013 averages at around 322 US$ Billion.

Table 3: Descriptive statistics of independent and dependent variables - hypothesis I

Dependent and independent variables for hypothesis II

Delta EF EPI and delta FDI Stock are the independent and dependent variables for

hypothesis II, respectively. Their descriptive statistics can be observed in table 4. Noteworthy is the change in English proficiency over the years, with an average increase of 3,24 percentage points in the ranking. Delta FDI stock also generally increased in that time period, with an average of 53,78 US$ Billions.

Descriptive Statistics

n=37 _Minimum _Maximum _Mean _{Std. Deviation}

EF EPI Score 2013 39,48 69,30 54,41 6,90

FDI Stock 2013 US$ Millions

(36)

n=35 _Minimum _Maximum _Mean

Delta EF EPI 2010-2013 -8,57 9,64 3,2411

Delta FDI stock 2010-2013 US$ Million 941 368976 53.784,51

Table 4: Descriptive statistics of independent and dependent variables - hypothesis II

Dependent and independent variables for hypothesis III

Due to space restraints, the descriptive statistics for the variables used in hypothesis III can be found in appendix F. The variables common civilization and English-nation’s

FDI stock in receiver country are the most noteworthy, with nearly half of all countries

sharing a common civilization (mean=0,45), and a mean of around 12 Billion US dollars of English-speaking sender country FDI stock in receiver countries. Also, receiver countries generally present lower GDPs, with their mean averaging at about 4,7 trillion US$, far below the sender’s 55 trillion. In fact, the sender’s minimum value of 15 trillion US$ (Australia) dwarf the receiver’s lowest country, with only 87 billion US$ (Ecuador).

Control variables

(37)

As can be seen from its delta, GDP has decreased for some countries during the 2010 and 2013 period and increased for others. The analysed countries have increased their average GDP in that time period however, with the mean value reaching 160 billion US$.

Table 5: Descriptive statistics of major control variables

It is also worth noting the variable Internet penetration as the EF EPI scores are based in part on internet tests taken by adults in the analysed countries. The average country had around 61 internet users per 100 citizens in 2013, although outlying values can affect this mean (such as Norway with 95,1 or India, with 15,1). Looking at the delta between 2010 and 2013, one can observe that all countries have increased their values, with the most increasing it by 21,5 percentage points (Chile).

5.2 Correlation tables and graphs

The standard Pearson’s correlation between the main variables used in this study is presented in this subchapter. To be meaningful, the data presented in the correlation is in its untransformed form.

Hypothesis I

The standard Pearson’s correlation between EF EPI Score for the year 2013 and FDI

Stocks for the year 2013 is presented in table 6. As can be noted from the table, there is

a minimal insignificant positive correlation between the two variables, r = ,056; n = 37;

p > ,05. For this thesis, I will consider p values larger than 0,05 to be deemed

insignificant (Field, 2013).

n=37 _Minimum _Maximum _Mean _{Std. Deviation}

Economic Size GDP 2013 US$ 24.259.100.000 9.240.270.452.047 1.149.384.524.763 1.749.006.539.115 Delta Economic Size GDP

2010-2013 US$

-575.824.074.624 3.309.768.181.734 160.812.744.587 554.540.032.416

Internet penetration 2013 Percent 15,1 95,1 61,211 22,8555

Delta Internet Penetration 2010-2013

(38)

Table 6: Correlation table EF EPI 2013 and FDI Stock 2013

The correlation between both variables can be observed in a scatterplot (Figure 2). As is evident, there is no apparent correlation between both variables, as the scatter dots are mostly not distributed along the normality line also depicted in the graph. This gives us an idea about the posterior regression results and the effect one variable might have on the other.

Figure 2: Scatter graph of EF EPI 2013 and FDI Stock 2013 correlation

It is worth noting that there exists also a positive and significant correlation between

Internet penetration and EF EPI scores of r = ,667 and p < ,05. This was expected, as

the EF EPI scores are based on online tests. Furthermore, there is a positive correlation between FDI stocks and GDP, of r = ,477 and p < ,05. The complete table can be found in appendix G.

Correlations

n=37 _{FDI Stock 2013 US$ Millions}

EF EPI Score 2013 Pearson Correlation ,056

(39)

Hypothesis II

Regarding hypothesis II, Delta EF EPI 2010-2013 and delta FDI stock 2010-2013 show a negative correlation of r = -,077. The significance is negligible, with p > ,05 and n = 35. As expected, delta GDP and delta FDI are highly correlated. The correlation table for all variables employed in hypothesis II can be found in appendix G.

Hypothesis III

Correlation data for selected variables used in hypothesis III can be found in appendix G. There are some significant correlations worth mentioning, such as the significant negative correlation between geographic distance between capitals and

English-nation’s FDI stock in receiver country of r = -,248; n = 66 and p < ,05.

Likewise, a significant positive correlation of r = ,360; p < ,05 can be observed between

English-nation’s FDI stock in receiver country and the Economic size sender country

(measured in GDP).

The EF EPI scores 2012 are highly correlated with internet penetration, as expected (r = ,911; p < ,05). However they are not with the dependent variable of English-nation’s

FDI stock in receiver country, were a minimal negative correlation exists (r = -,038; p >

,05).

Lastly, it is worth noting the positive and significant correlation between EF EPI Score

2012 and common civilization, with an r = ,892; p < ,05. Also, a positive relationship

between EF EPI Score 2012 and common institutions can be observed, with r = ,208; p = ,093.

5.3 Regression analyses results

(40)

Hypothesis I:

A simple linear regression was run in order to predict the dependent variable (FDI Stocks) from the independent variable (EF EPI Scores 2013). For hypothesis I the values for 2013 were taken into account. The model summary (table 7) and the coefficients table (table 8) can be found below.

As can be noted from table 7, the r square of only regressing the independent variable (EF EPI Score 2013) with the dependent one (FDI inward stock 2013) is ,07. When we add control variables, the model’s r squared rises to ,681. This is found to be the case for all three hypotheses, as when the control variables are added, the variability of the dependent variable is explained to a higher degree.

Table 7: Model summary of regression analysis for hypothesis I

Table 8 illustrates the beta values for the different variables, with the independent variable having a beta of ,264 and a significance p > ,05 in the first model. When control variables are considered along with the independent variable, it decreases to a value of ,085 with p > ,05 Furthermore, economic size (measured as zscore trans GDP

2013) has a considerable beta value of ,737 with p < ,05, while internet penetration 2013 shows a value of ,095 with p > ,05. The ease of doing business shows a negative

beta of -,188, with p > ,05. Remarkably, the level of democracy has a negative value b = -,042, with p > ,05 respectively.

Model Summary

Model (n=37) R R Square Adjusted R Square

Std. Error of the Estimate

Independent Variable ,264 ,070 ,043 ,97830512 Independent variable +

control variables (group 1)

(41)

Table 8: Coefficient table of regression analysis for hypothesis I

Hypothesis II:

In Hypothesis II, the independent variable (delta EF EPI Score 2010-2013) was regressed with the dependent variable (delta FDI Stocks 2010-2013). The independent variable has an r squared of ,010. When control variables are added into the regression calculation, they altogether account for an r squared of ,421. See table 9 for details.

Model Summary

Model (n=35) R R Square Adjusted R

Square

Std. Error of the Estimate

Independent Variable ,098 ,010 -,020 1,01011337

Independent variable + control variables (group 2)

,649 ,421 ,321 ,82422259

Table 9: Model summary of regression analysis for hypothesis II

The coefficients of the regression analysis can be found in table 10. The independent variable has a beta of ,097 with p > ,05 when regressed by itself. This is increased to a beta of ,320 when control variables are considered, with a decrease in the significance to

p > ,05. The control variable Delta GDP 2010-2013 has a sizeable positive beta of 6,58

with p < ,05. Again, the ease of doing business and the level of democracy control variables show negative beta values.

Coefficients table

Model: (n=37) Coefficients Sig.

B Std. Error

Independent Variable (Constant) 1,261E-15 ,161 1,000

Zscore (trans ef epi 2013) ,264 ,163 ,115

(Constant) -2,299E-15 ,100 1,000

Zscore (trans gdp 2013) ,737 ,105 ,000

Zscore (trans internet penetration 2013)

,095 ,144 ,516

Zscore (trans ease of doing business)

-,188 ,137 ,182

(42)

Coefficients table

Model (n=35) Coefficients Sig.

B Std. Error

Independent Variable (Constant) ,001 ,171 ,996

Zscore (trans delta ef epi 2010-2013) ,097 ,172 ,574 Independent variable

+ control variables (group 2)

(Constant) -1,080 ,383 ,009

Zscore (trans delta ef epi 2010-2013) ,320 ,189 ,102 Zscore (trans ease of doing business) -,317 ,192 ,109

Zscore (trans level democracy) -,210 ,206 ,317

Zscore (trans delta gdp 2010-2013) 6,575 2,185 ,005 Zscore (trans delta internet 2010-2013) ,127 ,180 ,487

Table 10: Coefficient table of regression analysis for hypothesis II

Hypothesis III:

The model summary and coefficient table of the regression analysis for hypothesis III can be found in tables 11 and 12 respectively. Regarding the model summary, the independent variable (EF EPI Score 2012) showed a minimal r square value of ,004 in the model without considering control variables. The model as a whole increases its r squared value to ,570 when we add control variables to the regression. The dependent variable is FDI stocks from English-speaking nations into receiver country.

Model Summary

Model (n=66) R R Square Adjusted R Square Std. Error of the Estimate

Independent Variable ,060 ,004 -,012 1,00598302

,755b _,570 _,501 _,70607544

Table 11: Model summary of regression analysis for hypothesis III

(43)

Surprisingly, common official language and common civilization have a negative beta value. This was an unexpected finding.

Coefficients

Model (n=66) Coefficients Sig.

B Std. Error

Independent Variable (Constant) -,007 ,125 ,955

(Constant) -,166 ,090 ,069

Zscore (trans distance capitals) -,363 ,095 ,000 Zscore (trans gdp sender 2012) ,466 ,086 ,000 Zscore (trans gdp receiver 2012) ,331 ,117 ,006 Zscore (trans internet penetration

receiver 2012)

,124 ,225 ,583

Zscore (trans contiguous borders) ,115 ,083 ,170 Zscore (trans common language) -,008 ,085 ,925 Zscore (trans common civilization) -,603 ,219 ,008 Zscore (trans common institutions) ,117 ,106 ,273

Table 12: Coefficient table of regression analysis for hypothesis III

5.4 Robustness test

In order to test the robustness of the model and to see if outliers were skewing results, I have analysed the boxplots of the data for hypothesis I and II in search for outliers (Field, 2013) and deleted them (those that were clearly separated from the quartiles represented in the box plots were excluded). The Log10 transformations were also

0 ENGLISH LANGUAGE PROFICIENCY AND FDI

0

ENGLISH LANGUAGE

PROFICIENCY AND FDI

Abstract

Acknowledgements

Table of Contents

List of Tables

Table of Figures

1. Introduction

2. Literature review

3. Theoretical framework

This leads to more inward FDI stock investments

Foreign business partners are attracted by high levels of proficiency

from potential local partners, drawn by lower transaction costs

4. Methodology and data

5. Empirical results