Entrepreneurial success
– A comparative approach on German and Swedish entrepreneurs during
the nineteenth and twentieth century
Nada Faraj Farhijo Hashi
Supervisor: Hans Sjögren and Joakim Persson
Linköpings universitet SE-581 83 Linköping, Sweden 013-28 10 00, www.liu.se
Title:
Entrepreneurial success
- A comparative study on German and Swedish entrepreneurs during the nineteenth and twentieth century Authors: Nada Faraj Farhijo Hashi Supervisor: Hans Sjögren Joakim Persson Publication typ: Master’s thesis in Economics Master’s program in Economics
Advanced level, 30 credits Spring 2018
Linköping University
Department of Management and Engineering (IEI) www.liu.se
interesting discussions regarding entrepreneurship. During this process, we had the possibility to visit Bocconi University in Milan, we would like to thank Andrea Colli for new and interesting viewpoints.
1 Abstract
This paper examines the similarities and differences between German and Swedish entrepreneurs born in the nineteenth and twentieth century. It is a replicated study by Vasta et al. (2015). The study is a part of an international project, where the characteristics of entrepreneurs’ is reviewed in multiple countries. The information of entrepreneurs is collected from biographical dictionaries and the qualitative data is later on converted into quantitative data, where several different variables are measured. The applied method, Principal Component Analysis (PCA) generated three different dimensions of entrepreneurial success. The components are the following: celebrity (captures entrepreneurs’ visibility in the various sources), economic success (describes the economic performance of an entrepreneur) and social mobility (measures entrepreneurs’ social class improvement). By using a prosopographical approach, we found among other things that a level of innovation intensity is required in order to reach a higher level of economic success for both German and Swedish entrepreneurs.
Keywords: Germany, Sweden, entrepreneurs, entrepreneurship, economic business history, founder, inventor
2
Table of content
1. Introduction 4 1.1 Problem discussion 5 1.2 Purpose 7 1.3 Research questions 7 1.4 Limitations 7 1.5 Implementation 8 1.6 Research contributions 9 2. Theory 102.1 The Schumpeterian entrepreneur 10
2.2 Risk and uncertainty 11
2.3 Previous studies 13
3. Sources and data 17
3.1 Biographical studies of German entrepreneurs 17
3.2 Biographical studies of Swedish entrepreneurs 18
3.3 Description and selection of variables 19
4. Method 26
4.1 Definition of Principal Component Analysis (PCA) 26
4.1.1 Example: Data reduction using Principal Components 27
4.2 Properties of Principal Components 29
4.3 Detection of Outliers and Robust Estimation 30
4.4 Rotation method and interpretation of principal components 31
4.5 Regression model 31
4.6 Criticism of the method 32
5. Results 34
5.1 Dimensions of entrepreneurial success for the German sample 34
5.2 Dimensions of entrepreneurial success for the Swedish sample 41
5.3 Multiple linear regressions 47
5.3.1 Regressions for the German sample 47
5.3.2 Regressions for the Swedish sample 48
5.4 Detecting heteroscedasticity and multicollinearity for both samples 50
6. Discussion 51
6.1 A comparative analysis of the components for Germany, Italy and Sweden 51
6.2 A comparative analysis of the results from regression models 54
6.3 A comparative analysis of American, British, French, German Italian, Spanish and
3
7. Conclusion 65
References 67
4
1. Introduction
Entrepreneurship in Europe is evident and several entrepreneurs throughout history have led by example and been role models for many. A few successful entrepreneurs that were active in Germany respectively Sweden during the industrialization were Hugo Stoltzenberg, Otto Meyer, Axel Wenner-Gren and Alfred Nobel (Deutsche Biographie and Swedish Biographic Lexikon, 2018). In connection to the industrialization in the late nineteenth century, institutions gave individuals the possibilities of making success based on their individual drive and knowledge. The establishment of firms across Europe contributed to economic growth which would not have been possible without the help of financial institutions (Sylla and Toniolo, 1991). According to German economic history, Germany had by 1890 one of the largest economies in the world and was among the leading countries in the industrial sector (Plumpe, 2016). In the beginning of the twentieth century, German industrial sector was very successful internationally due to high production standards and low costs. Germany was also able to use skilled labors more efficiently than any other country in Europe (ibid).
From 1850 to 1913 Sweden underwent industrialization and experienced economic growth at an accelerating rate (Myhrman, 2003). Between 1870 and 1970, Sweden’s economy was amongst the fastest growing economies in the world second to Japan (Johnson, 2006). The success of Sweden was mainly due to political reforms that occurred throughout the nineteenth century, for example the formation of the joint stock corporation law (Aktiebolagslagen) which took in effect 1848. The implementation of a renewed and more modern bankruptcy law in 1862 which simplified the reconstruction of firms for business owners. As well as the introduction of an improved patent and trademark law to secure innovations from possible imitation in 1884 (ibid).
The impact entrepreneurship has on economic growth has been examined from different perspectives such as the relationship between entrepreneurship and job creation (Malchow-Møller et al., 2015). In high income countries entrepreneurship have shown to have a positive impact on macroeconomic growth, due to entrepreneurs’ efforts to maintain and develop their firm (Stam et al., 2011). Also, the entrepreneur attracted the attention of management which in turn forced owners to find the appearances that distinguish entrepreneurs from managers (Hébert and Link, 2009).
5
Previous literature has examined the traits and personality characteristics of entrepreneurs to find specific traits that defines an entrepreneur. The entrepreneur has been defined as a risk taker, innovator, manager of a business venture, decision maker, coordinator and asset owner. (Gartner, 1988). According to Schumpeter (1911) the entrepreneur is an innovator who identifies and introduces new products, production methods, resources and organizational structure (Henrekson and Stenkula, 2015). For many other researchers, entrepreneurs differ from each other and each of them are considered as a unique person whose qualities need to be examined (Gartner, 1988).
In this study, we have examined 100 German and 100 Swedish entrepreneurs born between 1800 and 1970. Germany and Sweden stand out mainly due to their similarities in recent technological innovation (Government of Sweden, 2017). Taking this into account we find it suitable to examine entrepreneurs that are founders which have contributed to the establishment of enterprises and inventors. This study is a replicate of a previous study conducted by Nuvolari, Toninelli and Vasta (2015), “What makes a successful (and famous) entrepreneur? Historical evidence from Italy (XIX-XX centuries)”.
1.1 Problem discussion
According to the entrepreneurship barometer1 from 2016, nearly half of Sweden’s population
in working age have a positive attitude towards becoming entrepreneurs, in terms of starting or managing a business. The results have also showed that the distance between having a positive attitude towards entrepreneurship and starting a business is vast. This is partly because many individuals lack a business idea or believe the risks are too big (Swedish Agency for Economic and Regional Growth, 2017)2. It is therefore important that the public
sector can lead the next generation of entrepreneurs to their goals, by influencing laws and regulations.
Recently, studies have chosen to examine entrepreneurs’ biographies to find the factors and characteristics of individuals that led them to their entrepreneurial success. The first study
1 The entrepreneurship barometer is an attitude survey about Swedish people´s views on starting and managing
a business. In the survey for 2016, 10 300 people in working age have participated.
2 The Swedish Agency for Economic and Regional Growth (Tillväxtverket) is a government agency under the
6
was conducted by Nicholas (1999) who examined British entrepreneurs with a lifetime wealth accumulation as a measure of entrepreneurial success. In another study by Vasta et al. (2015), researchers have examined the determinants of entrepreneurial success for Italian entrepreneurs in nineteenth and twentieth century. The idea of studying the entrepreneurs’ biographies and transforming the qualitative data into quantitative but on different samples and frameworks has spread amongst other authors. There are four replicated studies of Vasta et al. (2015) that has used a similar sample of American entrepreneur (Magenes, 2015), British entrepreneurs (Piantanida, 2017), French entrepreneurs (Bonsignore, 2018) and Spanish entrepreneurs (Zollet, 2018) for the same time period. There is not a similar comparative study conducted for German and Swedish entrepreneurs trying to explain the main factors and differences between their success.
In contemporary times both Germany and Sweden are categorized as innovation-driven economies based on reports from The Global Entrepreneurship Monitor (2016). According to the German Federal Ministry for Economic Affairs, start-up companies and entrepreneurs are necessary to create a lasting competitive market and to develop job opportunities. Also, both countries offer government state-support for new start-ups and sizable information on how to start a successful business (Federal Ministry of Finance3 (2018) and Swedish National Audit
Office (2016)).
The Government of Sweden and the Federal Government of Germany have agreed on a unique partnership for innovation in January 31, 2017. The agreement on cooperation is to promote innovative community solutions, launch new products to export and to strengthening the competitiveness of the countries. One of the cooperation is about strengthen the digitalization of small and medium sized (SME) companies and the purpose is to match Swedish and German SMEs to exchange experiences through the countries’ agencies (Government of Sweden, 2017). We found it relevant to implement a comparative approach of German and Swedish entrepreneurs due to the partnership for innovation that will bring governments, businesses and institutions together, hopefully. When examining the underlying determinants of entrepreneurial success, we have studied the following factors for both German and Swedish entrepreneurs; education level, experience abroad, innovation intensity, involvement in politics, level of risk taking and scientist.
7
1.2 Purpose
The aim of the study is to examine the underlying determinants and differences between German and Swedish entrepreneurial success from 1800 to 1970.
1.3 Research questions
● Which factors have a significant impact upon the different dimensions of entrepreneurial success, regarding German and Swedish entrepreneurs?
● What does German and Swedish entrepreneurs have in common with the international studies based on American, British, French, Italian and Spanish entrepreneurs?
1.4 Limitations
Our sample consists of founders and inventors born no earlier than 1800 and who have been and are operating in the following sectors: agriculture or mining, industry, services and financial services. The chosen set of variables in this study differ slightly from the study by Vasta et al. (2015) because we found them to be relevant for the study’s aim (see more in Section 3.3). Furthermore, we chose to search with following keywords: gründer (grundare) and erfinder (uppfinnare) in the Deutsche Biographie (DB) and the Swedish National Biography (Svenskt Biografiskt lexikon, SBL). Due to the lack of time, we selected 100 German and 100 Swedish entrepreneurs. Regarding German entrepreneurs, we were able to sort by relevance in the DB, meaning that individuals were sorted according to their contribution for German history. The DB provides information of individuals in the German speaking areas from the Middle Ages to the present (DB, 2018). We chose to proceed with only German individuals, which means individuals from Austria and Switzerland are excluded from our sample.
The main source regarding Swedish entrepreneurs is SBL. For the Swedish entrepreneurs’ further limitations was made due to keywords being found in individual’s biographical information based on father’s or relative’s occupation. Those individuals have been excluded from the sample. Since we want a sample of 100 entrepreneurs and the search in SBL resulted in 75 individuals, we have completed with some additional individuals from Swedish entrepreneurial literature. The literatures used are Johnson (2006), Peterson (2012) and a
8
series of news articles regarding entrepreneurs from the Swedish daily newspaper (Svenska Dagbladet, SvD).
1.5 Implementation
We have replicated the method used by Vasta et al. (2015), by using a prosopographical approach. Firstly, we have transformed qualitative data into quantitative in order to analyze the data. Biographical information of German entrepreneurs has been translated from German to Swedish in order to encode the information. Google Translate has been used to translate the German text. We have checked the translated texts with the content provided by Encyclopedia Britannica which showed to be similar. Biographical information of Swedish entrepreneurs has been obtained from different relevant sources as described above, and the qualitative information has been transformed into quantitative data in the same way as for German entrepreneurs.
Secondly, we performed Kaiser-Meyer-Olkin (KMO) test that indicate the suitability of our samples for structure detection and it examines the proportion of variance in the variables that possibly can be caused by the underlying factors. KMO generates a value between 0 and 1, values less than 0.50 imply that the sample is not suitable for structure detection and higher values imply the sampling adequate. Additionally, Bartlett’s test of sphericity has been used to test if the selected variables are related, which is required for structure detection. It tests the hypothesis that the correlation matrix is an identity matrix (variables are uncorrelated) and in order to proceed it has to generate p-values less than 0.05 (IBM, 2018).
Thereafter, the Principal Component Analysis (PCA) has been employed to reduce the dimensionality of the data set and with help of the method, different dimensions of entrepreneurial success were derived. Before we continued to estimate the regressions with the components as dependent variables, we studied the retained components to detect any outliers. The regression method used is multiple linear regression because we found it to be the most appropriate method. The most important assumptions for a multiple regression model is homoscedasticity, no multicollinearity and that the model is correctly specified. The Ordinary Least Square (OLS) method minimizes the sum of squares of the residuals and the estimator is efficient compared to the others, see more in Section 4.5 (Gujarati, 2004 p. 202 - 208). The regression models have been estimated with different dependent variables similar
9
as in the study by Vasta et al. (2015). The variables used as proxies differ due to the different sample applied. The approach is explained in section 5.1 and 5.2.
In order to maintain robust regressions, we have performed tests for heteroscedasticity and multicollinearity. Regarding heteroscedasticity, Breusch-Pagan-Godfrey test has been conducted, which is appropriate when dummy variables are included in the model. It tests the hypothesis of no heteroscedasticity (Gujarati, 2004 p. 400). The second test involves identifying multicollinearity, by examining the variables Variance Inflation Factor (VIF). According to the rule of thumb, a VIF-value that exceeds 10 indicates multicollinearity problems (Gujarati, 2004 p. 359). Also, the correlation matrix of the variables has been examined to identify highly correlated variables. After we conducted the test, we re-estimated all regressions with robust standard errors due to the uncertainty of having heteroscedasticity. In this study, we have used the statistical software programs Eviews and SPSS.
1.6 Research contributions
Our contribution to this field is mainly the comparison of Germany and Sweden, which have not been previously studied or compared. We believe that our aim of focusing on entrepreneurs who are founders and inventors are relevant for the studied time period. The study is based on a new dataset that we have created with the help of biographical information from the mentioned sources in section 1.4.
10
2. Theory
The definition of the concept entrepreneurship is a highly discussed topic, it has been frequently discussed in the fields of business, sociology and economic psychology. There is a theoretical difference between social, political and institutional entrepreneurship (Henrekson and Stenkula, 2016). Even if the topic has been discussed over a long period in various fields, there is yet not a common definition of an entrepreneur. The factors that form an entrepreneur is a question of interpretation. Hébert and Link (2009) estimated that there are approximately 12 different interpretations of the term, given that all theoretical understandings are considered. Entrepreneurship can generally be explained by four classic entrepreneurial factors; the innovator, the arbitrage, the uncertainty carrier and the coordinator (Henrekson and Stenkula, 2016). In section 2.1 and 2.2, we define the differences and what distinguishes the four classical functions.
2.1 The Schumpeterian entrepreneur
Amongst the many interpretations in economic literature, the Schumpeterian entrepreneur is the most frequently mentioned (Henrekson and Stenkula, 2016). Joseph Schumpeter, an economist active in Austria and later on in the United States during the twentieth century drew the conclusion that an entrepreneur is someone who is an innovator. His economic theory on entrepreneurship focused on the development of ideas, the characteristics and timeframe used by an individual (Hébert and Link, 2008). Schumpeter referred to entrepreneurs as individuals that during a certain period are perceived as inventors during the process of the innovation (Schumpeter, 1911). Hence, entrepreneurship is a creating process which has a time frame and because of that, entrepreneurship is a scarce resource. Schumpeter viewed entrepreneurs as “primus motor”, a way to describe the entrepreneurs as head of the economic development (Henrekson and Stenkula, 2016). According to American economist William Baumol, the impact entrepreneurship has on economic growth is unquestionable, which is in line with Schumpeter’s definition of the entrepreneur being the “primus motor” of economic growth (Baumol, 2010).
In neoclassical growth theory, economic growth tends to come from knowledge as an exogenous factor. The impact entrepreneurs have contributed is barely recognized. Entrepreneurs connect knowledge with growth, by using knowledge and turning it to services
11
and products (Braunerhjelm, 2008). The Schumpeterian entrepreneur influenced many including Swedish economist Erik Dahmén, who studied the impact of entrepreneurship on the Swedish industrial development. Dahmén believed entrepreneurs to be the driving force of the economy. This because of the initiatives that entrepreneurs took, which in turn helped the development and creation of commercial business possibilities (Johansson and Karlson, 2002).
Along with other economists such as Joseph Kirzner, Schumpeter did not acknowledge the risks associated with entrepreneurship (Braunerhjelm, 2008). Kirzner described an entrepreneur as someone with the ability to identify and discover new opportunities. He believed in order to detect the many possibilities available which are to be utilized, an entrepreneur must have characteristics like alertness. Kirzner often used the term alertness in context with entrepreneurship, more specifically he described an entrepreneur as an arbitrage. An individual that utilizes the imbalances between markets and the market price differences
(Henrekson and Stenkula, 2016). In conjunction with alertness which raises competition, the entrepreneur is responsible for moving the economy towards the equilibrium (Braunerhjelm, 2008).
2.2 Risk and uncertainty
In many cases entrepreneurship is strongly associated with profit making and enterprises. Richard Cantillon was the first economist to study the entrepreneurs’ central role in economic development and interpreted the establishment of companies as a vast part of the idea behind entrepreneurship (Henrekson and Stenkula, 2016). According to Cantillon the market economy consisted of three classes. The entrepreneur was one of the three classes which was explained as an economic agent together with landowners and laborers. In the same direction, French economist Jean-Baptiste Say used the term coordinator to describe the responsibilities bestowed upon entrepreneurs. The responsibilities involved overseeing production as well as making decisions regarding capital and division of labor.
An entrepreneurs’ willingness to take risks by investing in new markets as well as being a decision maker are the most important aspects of Cantillon’s theoretical point of view (ibid). An individual will in search for profit encounter uncertainty, Cantillon focused on the actions made by the entrepreneur and not their personalities (Hébert and Link, 1989). To determine
12
who is an entrepreneur, he distinguished between individuals and categorized them into two separate groups, those who worked for a certain income and those which had an uncertain income. Based on that narrative Cantillon concluded that individuals that operate in uncertain markets and make risk full choices can be perceived as entrepreneurs. Hence, entrepreneurs are risk takers (Boutillier and Uzunidis, 2016). Later, economists with inspiration from Cantillon further developed theories on risk taking, amongst them was the German economist Johann Heinrich von Thünen. He believed that in order to be accepted as an entrepreneur one should always accept a part of the risk if complications arise. This due to insurance companies who may not cover all expenses causing the entrepreneur to accept an uninsurable risk (Hébert and Link, 2008). He also agreed with the certain vs uncertain income statement, where he explains how an entrepreneur differs from individuals with managerial profession.
Managers according to von Thünen should not be mistaken for entrepreneurs, since they have a secure income source. He believed that individuals with a secured income do not have to worry while an entrepreneur is constantly concerned and it is affected by sleepless nights. Furthermore, Von Thünen explained the period of concern that an entrepreneur endures, as a period of a productive stage where ideas are further developed to innovations and the risk taker develops to becoming an inventor (Hébert and Link, 2009). In 1921, the American economist Frank Knight published the book Risk, Uncertainty and Profit, where he reviewed entrepreneurs as uncertainty carriers. The focus was on individuals that make decisions during uncertainty. Knight’s theory differs from the risk-taking entrepreneur by only referring to the decisions that solely are made due to genuine uncertainty. Knight distinguished between risk and genuine uncertainty by claiming that unlike risk during genuine uncertainty individuals are not able to calculate the likelihood for a successful or a failed decision (Henrekson and Stenkula, 2016).
Cantillon’s definition on risk takers and Say’s description of a coordinator are in line with our sample of founders, an individual that creates jobs and can estimate customer demand as well as the markets. This can be described as an economic catalyst, a form of industrial leader, hence the relevance of the word founder (Hébert and Link, 2009). Our definition of an entrepreneur in this study can be summarized as an individual who introduces new products, a market or methods to a previously underdeveloped market. Meanwhile having the expectations to grow as well as contributing to a country’s macroeconomic growth.
13
2.3 Previous studies
The pioneer of applying a quantitative prosopographical approach4 is Nicholas (1999) who
explained entrepreneurial performance in Britain since 1850. By assuming the profit as the main motive for entrepreneurship, he used information on lifetime wealth accumulation as a measure of entrepreneurial performance. The wealth accumulation outcome is determined by initial wealth, profit (equivalent to the entrepreneurs’ income), consumption and rate of return. Moreover, entrepreneurial type was categorized as either inheritors or non-inheritors entrepreneur (such as firm founders or managers) in order to distinguish value of wealth obtained by inherited wealth from values of wealth obtained by entrepreneurial activities. The result specified two important determinants of entrepreneurial performance; entrepreneurial type and education. Both determinants indicated to have a negative impact on entrepreneurial performance. Other findings such as region of activity, religious affiliation and industry of occupation has shown not to be determinants of entrepreneurial performance.
The French economic literature in the early nineteenth century, considered entrepreneur as a vital component of a market economy who assumes risk associated with uncertainty (Hébert and Link, 2009). Foreman-Peck, Boccaletti and Nicholas (1998) examined the determinants of nineteenth century French entrepreneurship and management by a sample consisting of 244 French businessmen. They constructed two models for demand and supply of entrepreneurship and management success. The demand for entrepreneurs occurs due to their productivity, which in turn depends upon the opportunities given by the country’s economy. While, the supply for entrepreneurs is based on the assumption of free entry and exit of firms and it includes several determinants of starting a business. The findings on the demand side specified textiles to be the most promising industry rather than iron and steel, for increasing individual wealth. The findings on the supply side implied that secondary and University education have a negative effect on the probability of starting a firm.
Recently, researchers have focused on the relationship between education and development, since more knowledge and skill causes entrepreneurs to implement new methods more efficiently (Paul and Siegel, 2000). A comparative approach conducted by Tortella, Quiroga
4 The method is based on biographical studies of individuals and proceeds by analyzing the quantitative data statistically.
14
and Moral (2010) explored education’s effect upon entrepreneurial success by comparing Spanish with British entrepreneurs of the nineteenth and twentieth centuries. With economic versatility5 as a measure of entrepreneurial success, the positive effect of education showed to
be more evident for British than for Spanish entrepreneurs. This is due to the countries different educational systems. A remarkable finding from this study is that wealth measured by entrepreneurs’ family income had negligible impact on economic versatility for both samples. Nevertheless, Nicholas (1999) found entrepreneurial type (inheritors or non-inheritors) and education as the determinants of entrepreneurial performance in case of British entrepreneurs.
Toninelli, Vasta and Zavarrone (2014) explained entrepreneurial success of Italian entrepreneurs based on a new data set provided by Toninelli and Vasta (2010). They selected firm’s growth as a proxy of entrepreneurial success. Recently, Vasta et al. (2015) examined the determinants of entrepreneurial success in Italy by adding two additional components for entrepreneurial success. The authors began with coding biographical information into a series of categorical variables and observed eight variables6 which considered to be theoretically
relevant. Afterward, they performed factor analysis, to explore the relationship between the chosen variables and to create measures of entrepreneurial success. The study contributes with three dimensions of entrepreneurial success which were later used as dependent variables. The first dimension of entrepreneurial success is celebrity and it captures the visibility of entrepreneurs in various sources (number of rows in Wikipedia and Italian dictionary). The second, economic dimension is measured by the firm’s growth, geographically expansion of business and by innovation. Lastly, social mobility and entrepreneurial type was selected as proxies for the social mobility dimension of entrepreneurial success. Thereafter, the authors performed multiple linear regressions to find the determinants for the different dimensions of entrepreneurial success. The following variables are included in their model; education, experience abroad, innovation intensity, involvement in politics and scientist. The determinants of Italian entrepreneurial success have indicated to be education level and political connections when celebrity was set as a
5 The entrepreneur is considered to be adaptable if he operated in several sectors and not adaptable if he
operated in only one sector.
6 Employment growth, expansion of business at geographical growth, introduction of successful brand or
product, social class improvement, entrepreneurial type, number of rows in Italian dictionary, number of words in English and Italian Wikipedia.
15
dependent variable. Both factors have a positive impact on celebrity. Italian entrepreneurs with higher level of education gained more attention in the various sources, and it can also describe their social class in society. Similarly, being politically involved increased their visibility and it indicates that being politically involved played a role during the Italian capitalism. With economic success as dependent variable, the result specified following significant variables; innovation intensity, education level and experience abroad. The factors have also a positive impact on economic success. This finding confirm that Italian entrepreneurs were innovators, in accordance with Schumpeter (1911) definition of an entrepreneur. Regarding the last dimension of entrepreneurial success, social mobility, the authors did not present any results of the determinants (Vasta et al., 2015).
There are four replicated studies of Vasta et al. (2015) which examine the drivers of entrepreneurial success for the same time period but on different samples. The studies have replicated the method used by Vasta et al. (2015) and, also the same explanatory variables have been used. The following countries have been studied: France (Bonsignore, 2018), Great Britain (Piantanida, 2017), Spain (Zollet, 2018) and United States of America (Magenes, 2015). The determinant for the first dimension of entrepreneurial success celebrity, is innovation intensity and it have a positive impact in all four studies. Findings imply that American, British, French and Spanish entrepreneurs became more visible on the various sources by being more innovative and it is also consistent with Schumpeter’s (1911) definition of an entrepreneur as an inventor. The common determinant for American, British and Spanish entrepreneurs is experience abroad. The effect of experience abroad in terms of broader network and increased level of knowledge is shown to be positive in which increased entrepreneurs’ visibility on the various sources. Another common determinant for British, French and Spanish entrepreneurs is involvement in politics, with a positive impact on celebrity. British entrepreneurs’ visibility in the public domain has increased with higher level of education, while American entrepreneurs became less visible on sources. They also became less visible if they were educated or trained in science or engineering (captured by the variable scientist), in contrast it increased the visibility of Spanish entrepreneurs.
The common determinant for economic success in the mentioned studies is innovation intensity, with a positive impact on the economic performance of the entrepreneur. The ability to innovate and identifying opportunities occurring on a market are the fundamental driver of celebrity and economic success dimension of entrepreneurial success. The common
16
determinant of economic success with a negative impact is education which is found in all studies except in the Spanish sample, where it is positive. The result indicates that the more educated the American, French and British entrepreneur was, the less successful he or she became. The same reasoning applies to Spanish entrepreneurs but with the opposite effect. Moreover, other findings imply that American, French and Spanish entrepreneurs became less successful due to being politically involved. When British and French entrepreneurs were educated or trained in science or engineering, they also became less successful.
In the social mobility dimension, the common negative associated determinant for all four studies is education. As discussed earlier, pursuing higher education was less preferred by individuals that had potential successful ideas, which they rather focused on. The determinant scientist has a positive impact for American and French entrepreneurs which implies that the drivers of social mobility is same for both type of entrepreneurs. British and Spanish entrepreneurs social class did not improve when they were politically involved, but it did improve when they became more innovative. Therefore, the determinants of social mobility are the same for British and Spanish entrepreneurs.
17
3. Sources and data
The method of using biographical information with quantitative approach has spread progressively amongst business historians and economists. Biographical studies of entrepreneurs provide a better understanding of historical motivations, strategies and the development of entrepreneurs. Additionally, we get a better understanding of how society and institutions have influenced entrepreneurship over time and their impact on society as a whole (Mokyr et al., 2012). The entrepreneurial biography from reliable sources has been used to identify the different characteristics and the entrepreneurial activities for German and Swedish entrepreneurs. Generally, the method of collecting our dataset can be described as primary. According to Hox and Boeije (2005) primary data is described as original data which is collected for a specific research purpose by the researchers. As previously mentioned this study is limited to entrepreneurs that are founders and inventors. In below sections 3.1 and 3.2 follows a description of how the search for the observations was conducted, and in section 3.3 information regarding the selected variables are presented.
3.1 Biographical studies of German entrepreneurs
The Deutsche Biographie is funded by the German Research Foundation and it is managed by the Historical Commission and Bayerische Staatsbibliothek. Since 2010, it has been providing digital full texts of more than 130,000 individuals from the German-language areas. The service provides more than 48,000 biographical articles obtained from the Allgemeine Deutsche Biographie. It also provides 24 volumes (since 1953) of the Neue Deutsche Biographie. The DB provides well founded articles by experts and the dictionary contains deceased individuals whose achievements influenced the social development in Germany. We believe that the DB is a reliable source, since the information is provided by experts in the field. It is also considered to be the most relevant biographic source of the German-language area (DB, 2018).
As mentioned earlier, the DB consist of people from all German-language areas and we have chosen to focus on individuals only from Germany. We conducted an advanced search to find founders and inventors. In the advanced search selection of gender, beliefs and professional fields can be made. Due to time constraint, we selected the first 50 founders and the first 50 inventors from the search results in the fields of business, industry and technology. With the
18
keyword “gründer” we received 1534 individuals and the second keyword “erfinder” gave 441 individuals. Therefore, we have not supplemented with additional sources. The figure below shows the number of German entrepreneurs by year of birth.
Figure 1: Number of German entrepreneurs by year of birth
Source: Deutsche Biographie. Own illustrated figure.
3.2 Biographical studies of Swedish entrepreneurs
Our primary source, the dictionary of Swedish National Biography (SBL) is a government funded project and is a part of the National Archives of Sweden (Riksarkivet). SBL consists only of deceased individuals who during their lifetime were active in Sweden or abroad but remained ties with their home country. It includes those who had a significant impact on Swedish history and contributed to the different stages in the development of the Swedish society. The first publication of the dictionary occurred in 1917 and the process of digitalization started in 2012. The publication is an ongoing work (SBL, 2018). SBL provides the ability to perform text search in their database and with the help of keywords we searched for founders and inventors. The search resulted in 47 individuals who are recognized as “founders” and 28 individuals as “inventors”, all born between 1800 and 1970. The information provided by SBL are from first-hand sources with high credibility and the authors of the articles are qualified experts in the fields. Since our search in SBL led to a total of 75 individuals, we have supplemented our sample with additional individuals. This by using literature of entrepreneurship by Johnson (2006), Petersson (2012) and as well as a
0 1 2 3 4 5 6 N umb er of ent re pre nue rs Year of birth
19
series of news articles published in the Swedish daily paper (Svenska Dagbladet, SvD). This completes our sample of 100 individuals. Figure 2 displays the number of Swedish entrepreneurs by year of birth.
Figure 2: Number of Swedish entrepreneurs by year of birth.
Source: Swedish National Biography and Nationalencyklopedin (NE). Own illustrated figure.
3.3 Description and selection of variables
In this study, we have collected a total of 18 variables as shown in table 1. The set of variables has been chosen according to their relevance for the study’s aim and they are also theoretically relevant. Out of the total, we have focused on 7 main variables for derivation of the different dimensions of entrepreneurial success. The variables are the following: entrepreneurial type, geographical growth, growth in terms of employment, number of rows in the dictionaries (DB and SBL), number of words in Wikipedia (English, German and Swedish) and social mobility. The set of variables are identical to the variables used by Vasta et al. (2015) except the exclusion of the variable introduction of successful brand or product. This depends on findings made by Vasta et al. (2015), which implies that the variable holds a greater bias when encoding than the remaining variables. As previously stated we have transformed qualitative data into quantitative, then later on constructed a number of categorical variables as well as dummy variables. Majority of the variables are set up as categorical and are measured in scales, which have been defined by Vasta et al (2015).
0 1 2 3 4 N umb er of ent re pre nue rs Year of birth
20
It is important to highlight that we have from the best of our ability encoded the biographical information for each observant equally. This by using the same criteria and interpretation for each variable. We have taken this into consideration, regardless there may still appear biasness in the samples.
The biographical sources have given us clear demographic description of each individual, as well as an overview of their lifetime and achievements. With the help of the location of birth we created the variable area of birth, which is based on the geographical location in each country. This variable was executed separately for Germany and Sweden. For Sweden, the following categories were created: north (Norrland), middle (Svealand) and south (Götaland) and for Germany: north, west, east and south. If an individual was born in another country, it was categorized as abroad. The difference between the categorization in our case depends on a countries geographical appearance, since Sweden is an elongated country it was categorized as stated and vice versa for Germany. Following variables, area of birth, gender and year of birth are relevant for our descriptive statistics and it is also relevant when studying the demographics of the samples.
An interesting variable is father’s main activity, which is defined as the father's occupation. This information simplifies the determination for the variable social mobility of the entrepreneurs. Hence, the variable social mobility was later on created with the help of father’s main activity. The variable is scaled in a range from 0 to 2, where 0 indicates that no improvement was made, 1 from middle class to upper class and 2 from lower to upper class. With the help of these variables, we take into account the social development each individual experienced over the time. In the field of social science, researchers have made repeated discoveries that a child’s level of achievements can be based on their social environment as well as their parents educational and occupational background (Davis-Kean, 2005). Taking this into consideration the outcome of the mentioned variables is of great interest.
In contemporary times, many may argue that education level is vital in order to reach high achievements. In order to chart how educational background can differentiate, we collected the highest level of education reached. The following categories was available; illiterate/primary school, middle school, high school/college or University. From the data we collected regarding education, we found that a number of individuals completed internships
21
or studies abroad. This information led us to our next variable experience abroad, which is set as a dummy variable. The variable reveals whether an individual during his or her lifetime travelled overseas for educational purpose or for work. The remaining variables regarding experience and social engagements are scientist (education or training in science) and politically involved, if an individual was politically involved during his career7.
To determine what type of entrepreneur an individual is, we differentiate between if an individual is a founder, co-founder, co-founder with family, inheritage or by purchasing along with which sector they operated in. There is a variety of sectors in which entrepreneurs was or still active in. With the help of the variable main sector of activity, we are able to distinguish them. An entrepreneurs’ level of risk taking, product innovation level and process innovation level are also included in the dataset and are categorized according to a scale of 0 to 1.5. The variable level of risk taking is a measurement based on individuals’ entrepreneurial decisions. If an entrepreneur formed an enterprise by investing in new markets, he or she obtains a value of 0.5. The formation of an enterprise in an uncertain market or the introduction of an invention gives a higher level of risk taking, a value of 1. The highest value of 1.5 implies that an individual has spent many resources or have introduced a new product in a highly risky market. Risk taking is heavily discussed in theory and according to Cantillon and Von Thünen entrepreneurial activity induce risk and create uncertainty. An inclusion of the aspect in our empirical study is essential.
Product and process innovation level can be described as a measurement for Schumpeter's theory and the belief that innovation is the central role of an entrepreneur. In the study by Vasta et al. (2015), they created an additional variable named innovation intensity which is the sum of product and process innovation. It measures how well an entrepreneur is in introducing both a process and a production innovation. The level of exposure an entrepreneur has experienced for their innovation or enterprise around the world is described by the variable geographical growth. The variable growth in terms of employment can explain the size of enterprises and the development achieved by an entrepreneur. A few enterprises may have participated in horizontal integration, it is the acquisition of a
7Overall for German entrepreneurs, it was difficult to find information whether they were politically involved.
When there was not any information mentioned in DB, we assumed that the entrepreneur was not politically involved.
22
competing company that is at the same level and the variable is set up as a dummy variable. Finally, we look at visibility and popularity of each individual based on their accomplishments and history by observing the number of words in Wikipedia (English, German and Swedish versions) and number of rows in dictionary (DB and SBL).
From the descriptive statistics in table 1, it is possible to identify the possible effect of the variables from the regression analysis section. We believe that the effect of education level will be more evident for German entrepreneurs than for Swedish entrepreneurs. Majority of German entrepreneurs (63 %) have obtained a University degree, while the largest group of Swedish entrepreneurs (32 %) accomplished only high school/college diplomas. Also, none of the German entrepreneurs attained primary school meanwhile 8 % of Swedish entrepreneurs did. This strengthens our belief that education may have a positive impact on entrepreneurial success regarding German entrepreneurs. A negative impact or a non-significant result can possibly be obtained for Swedish entrepreneurs. We also believe early introduction of a compulsory school attendance law may generate a higher educated population. The compulsory school attendance law took effect in 1763 in Germany (Spielvogel, 1999) and 1842 in Sweden (Johnsson, 2006). Reviewing experience abroad, majority of both German and Swedish entrepreneurs (78 % and 74 %, respectively) travelled overseas for educational purpose or for work. Therefore, the variable is excepted with great probability to have a positive impact on entrepreneurial success for both samples. Regarding innovation intensity, the proportion of German entrepreneurs who obtains the highest value of 3.0 is higher (39 %) than Swedish entrepreneurs (24 %). German and Swedish entrepreneurs have introduced both a process and a production innovation during their career. Thus, innovation intensity should have promoted entrepreneurs’ success as Schumpeter's theory of entrepreneurship.
Majority of German and Swedish entrepreneurs (73 % and 80 %, respectively) was not politically involved during their career. It was difficult to find information on DB and SBL whether both type of entrepreneurs were politically involved. When there was no information mentioned in the sources, we assumed that the entrepreneur was not politically involved. We believe that being political engaged does not have a positive impact on entrepreneurial success. According to the results all entrepreneurs have acquired some level of risk, although it appears that majority (82 %) of the German sample and (44 %) of Swedish sample maintained a low level of risk.
23
None of the entrepreneurs were free of risk therefore we have a strong believe that any form of risk taking has a positive effect on entrepreneurship. The last variable scientist, may have a positive impact for German entrepreneurs (71 % were educated or trained in science) and a negative impact for Swedish entrepreneurs (26 % were educated or trained in science).
24
Table 1. Descriptive statistics (in percent), 100 observations.
Germany Sweden Germany Sweden
Gender Scientist (education or training in science or engineering)
Female 1 9 No 29 74
Male 99 91 Yes 71 26
Area of birth Involvement in politics
North 19 North 8 No 73 80
West 26 Middle 42 Yes 27 20
East 15 South 40
South 24 Abroad 10 Entreprenurial type
Abroad 16 Founder 67 71
Co-founder 20 18
Fathers main activity Co-founder with family 6 6
Farmer 8 14 Inheritage 6 3
Laborer 1 9 Purchasing 1 2
Manager 37 5
Technician 1 3 Main sector of activity
Craftsman 12 11 Agriculture, fishing and
mining 5 7
Entrepreneur 15 1 Industry 60 46
Freelance 4 2 Service (not financial) 32 41
Employee 10 14 Financial service 3 6
Merchant 10 23
Church minister 2 4 Geographical growth
Military/politician 0 14 Local 0 8
National 39 60
Social mobility International 61 32
No improvement 6 7
From middle to upper class 75 64
From lower to upper class 19 29 Growth in term of employment
Increasing growth 28 34
Education level Workforce doubled 1 1
Primary school/illiterate
0 8 Workforce more than doubled 71 65
Middle school 23 29 High school/College 14 32 Horizontal integration University degree 63 31 No 93 87 Yes 7 13 Experience abroad No 22 26 Yes 78 74
25
Germany Sweden Germany Sweden
Level of risk taking Numbers of words Wikipedia (English)
0.5 (low risk) 82 44 0 45 62
1.0 (medium risk) 16 48 1-200 4 9
1.5 (high risk) 2 8 201-400 4 2
401-600 3 5
Innovation intensity (sum of product and process
innovation) 601-1000 4 5
0.5 0 9 >1001 40 17
1.0 4 17
1.5 2 16 Number of dictionary rows (DB respectively SBL)
2.0 20 26 0-100 84 68
2.5 35 8 101-200 12 24
3.0 39 24 201-300 3 4
Numbers of words Wikipedia (German and Swedish) 301-400
1 2 0 6 12 401-500 0 2 1-200 5 9 >501 0 0 201-400 8 17 401-600 5 11 601-1000 9 4 >1001 67 47
26
4. Method
Factor Analysis and Principal Component Analysis (PCA) are two different techniques for reducing original variables into a reduced set of variables that are easier to interpret. Principal component analysis has often been described as a special case of factor analysis and statistical programs treat it as an option for factor analysis. The differences between these techniques is in the definition of their model, the estimation process and the dimensionality of the model can result in more radical effects on factor analysis than in principal component analysis8. The present chapter focuses on principal component analysis as the chosen method
of this study.
4.1 Definition of Principal Component Analysis (PCA)
Pearson (1901) was first to introduce the Principal Component Analysis and it is a well-known technique of multivariate analysis. PCA is a method for reducing the dimensionality of a data set, containing a large number of correlated variables and it generates a new reduced set of uncorrelated variables, the principal components. The components are ordered in such a way so that the first components retain most of the variation from the original variables and the method focuses mainly on the total variation among the variables. The PCs can be derived from covariance and correlation matrices, in which both are based on the eigenvectors and eigenvalues of the matrices. However, the PCs does not provide same information due to using different transformations technique of the variables (Jolliffe, 2002 p. 1)9.
The principal component, denoted as PC are a function of p random variables, denoted as x (see equation 1) and it is the covariances or correlations between p variables that are of interest. Unless the number of variables are small, or the structure of the covariances or correlations between p is simple, it will be difficult to analyze the structure of the variables. Through PCA, we maintain few derived variables with as much information as possible given by the variables covariances and variances. Firstly, we select a linear function w´1 x (PC1) of a
vector of variables that have maximum variance and the second linear function w´2 x chosen
has to be uncorrelated with w´1 x and also have maximum variance. The functions proceed to
8 Further discussion of differences and similarities is found in p. 150 – 161 in ” Principal Component Analysis”, by Jolliffe.
27
be generated in the same way, so that the last linear function w´k x having maximum variance
are uncorrelated with w´1 x, w´2 x, …, w´k-1 x. The derived PC can be as many as p but the
goal is to retain m PCs, where m < p, that account for most of the variation in the variables. The PCk is given by w´k x, where w´k is an eigenvector of a covariance or correlation matrix
which in turn is consistent to its largest eigenvalue. Further explanation of eigenvalues is given below in section 4.2. Additionally, it is clear from the equation below that PCA assumes that the measurement is without error due to the exclusion of an error term (ibid).
PC1: w´1 x = w11 x1 + w12 x2 + … + w1p xp = ∑𝑝𝑗=1𝑤1𝑗 𝑥𝑗
PC2: w´2 x = w21 X1 + w22 x2 + … + w2p xp =∑𝑝𝑗=1𝑤2𝑗 𝑥𝑗 (1)
:
PCk: w´k x = wk1 x1 + wk2 x2 + … + wkp xp =∑𝑝𝑗=1𝑤𝑘𝑗 𝑥𝑗
k: the number of principal components p: the number of random variables PCj: principal component
xj: a random variable
x: a vector of xj
wkj: the weight of xj that maximizes the ratio of the variance of PCj to the total variation,
i = 1, 2, …, k, and p = 1, 2, … , j
w´k: a vector of xj coefficients and ´ denotes a vector
4.1.1 Example: Data reduction using Principal Components
The reduction attained by transforming the original variables to PCs, can be illustrated by considering two variables, x(1) and x(2) containing totally 50 observations. Figure 3 shows a
plot of 50 observations on x(1) and x(2) in two dimensions and due to their concentration near
the value of 1 and -1, it indicates that the variables are highly correlated. There is also a great variation in both variables but the variation is greater in x(2) than in x(1). Figure 4 provides a
plot of the 50 observations after transforming the variables to PCs: z(1) and z(2). The variation
is greater in z(1) than the variation of the original variables, while the variation in z(2) is
negligible. Generally, if a set of variables (more than two) are correlated with each other, then the first PCs will account for most of the variation so that as much information as possible got retained from the original variables. In contrast, the last PCs will account for those with
28
negligible variation in which it identifies near constant linear relationship among the original variables. The last PCs are not impractical, they can be useful in regressions or in case of detection of outliers (Jolliffe, 2002 p. 3 - 5), see more in section 4.4.
Figure 3: Figure 4:
Plot of 50 observations on x(1) and x(2) Plot of 50 observations with transformed PCs,
z(1) and z(2)
Source: Jolliffe, 2002.Own illustrated figures.
How to find Principal Components?
The PCk is given by w´k x, where w´k is an eigenvector of a covariance or correlation matrix,
as mentioned earlier. If wk is equal to unity, then the variance of w´k x will be equal to its
eigenvalue. The derivation can easily be presented by considering w´1 x as an example, in
which the vector w1 maximizes the variance of w´1 x by using the eigenvalue as a Lagrange
multiplier10. The approach approves that the coefficients and variances of PC
k are the
eigenvectors, denoted as k, of a covariance matrix. Hence,
var(w´k x) = k (2)
29
4.2 Properties of Principal Components
Consider the derivation of PCs given in Section 4.1 and by denoting the transformed vector of variables as z, we obtain
z = A´ x (3)
where A consists of columns of the eigenvector of a correlation matrix and x* consists of (original) standardized variables. The purpose is to find PCs of x*, which is a standardized (transformed) version of x and the covariance matrix of x* is the correlation matrix of x. This implies that PCs are defined as a function of x* and they are invariant in orthogonal transformations (i.e. preserves lengths and angles between vectors) of x. When interpreting the functions, the standardized variables can easily be transformed back to x, by multiplying x* with the standard deviation of x. An important property of PCs is that they depend on the correlation ratio11 and not on their absolute values. Also, probability distributions for the
eigenvectors and eigenvalues of a covariance matrix, are generally asymptotic and they assume the multivariate normal distribution for the original set of variables (Jolliffe, 2002 p. 30 - 47).
How many components to retain?
It is important to know how many m PCs can be selected without any information loss and there are several rules for choosing an appropriate amount of m. The most used criterion is the cumulative percentage of total variation, referring to the first m PCs who accounts for the largest possible variance in terms of percentage of total variation. The total variances of PCs are equal to the total variances of the elements of x and it is influenced by the number of observations included. As the number of observations increases, the variance becomes smaller. Furthermore, it has shown that a sensible cut-off is somewhere between 70 % and 90 %, although the limit can change depending on the particular dataset.
The second rule for computing PCs is described as following; if the elements of x are uncorrelated, then the number of PCs will be as many as the original variables and all having unit variances. When m PCs are equal to p variables, it becomes meaningless to retain. The
30
rule, Kaiser´s rule (introduced by Kaiser, 1960)12 is to retain only PCs whose variances is
greater than one. As mentioned earlier, the variances of PCk are the eigenvectors (see
equation 2) and in the forthcoming, we refer to the eigenvalues instead of variances. Thus, components whose eigenvalues is less than one will be excluded because they contain less information than one of the original variables (Jolliffe, 2002 p. 111 - 115).
An alternative approach involves studying the scree graph, which displays a plot of eigenvalues against m components. The number of components to be retained is decided by selecting the components whose eigenvalues is greater than one. In the graph, the plotted eigenvalues are steep to the left of the component axis (lk > 1) and then the slope remains less
steep to the right of the component axis (lk > 1). When the slope of the line changes, it refers
to the next component and those break-points tells us how many components m to retain. This approach is a complement to the upper approach, as the scree graph mainly contributes to better understanding of how the eigenvalues of the components appears graphically (Jolliffe, 2002 p. 115 - 118).
4.3 Detection of Outliers and Robust Estimation
To make the analysis more robust, it is important to detect outliers in the dataset to determine which outlying observations have or do not have a large effect. The later, are influential observations and they can be handled by either removing the observation or by diminishing the effect. Furthermore, an observation is considered to be an outlier if its value is greater or lower than 3.5 units (Engineering Statistics Handbook, 2018).
An observation can have different effects for different types of analysis or parameter, such as the coefficients, the variances and the PCs. In PCA, influential observations that have effect on the coefficients of a PC does not necessarily mean that they also are influential for the variance of the same PC, and vice versa. It is mentioned in section 4.2 that the distribution of a variable is not important, instead it is important to classify whether an outlier is influential or not. Since, the effect of influential observations on the PCs are inconsistent and if the analysis proceeds without considering if there are any influential observations, then the result is with great probability determined by such observations. Thus, a comparison between
12 The rule is constructed for PCA based on correlation matrices, but it can in some circumstances be used for some types of covariance matrices.
31
estimated and actual influence of observations should be conducted and one approach is to observe changes in eigenvalues. The estimated changes in eigenvalues are calculated from theoretical influence functions, while the actual changes in eigenvalues compare the change in eigenvalue with and without an observation. By removing a particular observation, it is possible to estimate its influence (Jolliffe, 2002 p. 232 - 250, 263 - 264).
4.4 Rotation method and interpretation of principal components
The interpretation of PCs as a linear function of all the random variables is not easy and an alternative for simplifying interpretation is to rotate the PCs. There are several techniques that provide replacements for PCs which simplifies the interpretation of the components retained. The most used approach is to rotate the PCs in which the axes of the first m components that account for most of the variation is rotated so that it clarifies the interpretation of the axes. Furthermore, by rotating the components it becomes clear which variables are most important, they have the largest values for their loadings and the less important variables have loadings near zero. The most used simplicity criterion is the varimax criterion (rotation method) which takes into account the variance maximization. It has been found that the different simplicity criterion, such as quartimax have no major differences (Jackson, 1991). Hence, the choice of simplicity criterion is not very important (Jolliffe, 2002 p. 269 – 272).4.5 Regression model
The regression method used is multiple linear regression because we found it to be the most appropriate method. The theoretically relevant explanatory variables are included in our regression model. The variables are the same as in the study by Vasta et al. (2015) but with one additional variable, level of risk taking. The variable is relevant due to explaining the three different dimensions of entrepreneurial success and, also the variable is theoretically relevant (see more in Section 3.3). In order to maintain robust regressions, we have estimated all regressions with robust standard errors which handle heteroscedasticity. In case of absence of heteroscedasticity, the robust standard errors are appropriate. The following model has been used for each dependent variable (the dimensions retained), denoted as Y, for both German and Swedish sample,
32
Y = c + 1*education level + 2*experience abroad + 3*innovation intensity
+ 4*involvement in politics + 5*level of risk taking + 6*scientist +
(4)
where c is the intercept term, (1-6) is the coefficients of the variables and is the error term.
4.6 Criticism of the method
In the following section, we discuss advantages and disadvantages of the used method and also the alternative approaches. The discussion criticizes the whole study’s approach. Firstly, the use of biographical studies of entrepreneurs to further analyze different research questions is the only method used so far by other researchers. The reason for transforming qualitative information into quantitative, is due to the fact that there is no complete dataset of entrepreneurs. We believe that the use of a prosopographical approach is appreciated given that the information of entrepreneurial biography is obtained from reliable sources.
Secondly, the PCA is an appropriate method for the study's aim because we needed to reduce different measures into a composite component. The PCs is derived based on correlations matrices because our chosen variables are measured in different units. In this case, it is not possible to conduct a covariance matrix which requires that the variables are measured in same units (Jolliffe, 2002 p. 22-24). Another reason for not using covariance matrices is because the differences between the variances of the variables are large, in terms of high standard deviation (see more in Section 5.1 and 5.2). Because those variables with the largest variance are likely to dominate the first few PCs. An advantage of using correlation matrices is that the results obtained from different analyzes are comparable (Jolliffe, 2002 p. 22).
Thirdly, the purpose of PCA is to receive uncorrelated components who accounts for most of the variation from the original variables. An alternative approach is the Maximum-likelihood extraction method, which is used to extract parameters that have caused the correlation between the set of variables. The method requires that the sample has a multivariate normal distribution (ibid). Taking this into consideration, our conclusion is that the PCA is a more appropriate method for our dataset, partially due to our dataset not fulfilling the requirement for a multivariate normal distribution.
