Msc thesis IBM
‘’A Cross-Sectional analysis of the effect of Discrimination,
Formal Education and Informal Business Education on Ethnic
Entrepreneurship within the United States’’
‘’A Cross-Sectional analysis of the effect of
Discrimination, Formal Education and
Informal Business Education on Ethnic
Entrepreneurship within the United States’’
Wouter Veldhoen
University of Groningen, Faculty of Economics and Business,
The Netherlands
Abstract
This article describes the relationship between the perception of discrimination, formal education and informal education on the level of entrepreneurship in the United States among 8 migrant groups, which are Czech Republic, England, France, Germany, Ireland, Mexico, Poland and a group classified as Others. Particular support is found for the role of the father and the mother of migrants as statistical support is found that either or both parents are self-employed the likelihood increases that the migrant him/herself is also self-self-employed. No such results were found for the degree of formal education. And weak results were found that the perception of discrimination works as a restraining factor rather than a push factor into entrepreneurial activities among migrant groups.
Table of Content
1. Introduction……… 5
1.1 Entrepreneurship and Immigration……… 5
1.2 Problem Statement and Relevance……… 6
2. Literature Review………. 8
2.1 Explaining Factors of Migrant Entrepreneurship………. 8
2.2 Discrimination……….. 9
2.3 Formal Education………. 11
2.4 Informal Education……….. 13
2.5 Conceptual Model……… 14
3. Data & Methodology……… 15
3.1 Dependent Variable……….. 15
3.1.1 Defining Entrepreneurship……… 15
3.1.2 Ethnic Entrepreneurship………... 16
3.1.3 Corporate Entrepreneurship………. 16
3.1.4 Implications for Research………. 16
3.2 Key Independent Variables……….. 18
3.2.1 The Level of Perceive Discrimination……… 18
3.2.2 The Degree of Formal Education……….. 18
3.2.3 The Degree of Informal Education……… 19
3.2.4 Subgroups……….. 19
3.3 Control Variables……….. 21
3.4 Method……….. 21
3.4.1 The Migrant-Entrepreneurship Model………... 22
3.4.2 Multicollinearity……….. 23
3.5 Validity & Reliability……… 23
3.5.1 Selection Bias………. 23
3.5.3 Missing Values ………. 24
3.5.3 Causality……… 25
4. Results……… 26
4.1 Baseline Results………... 26
4.2 Extension of the Model………... 31
4.3 Chow Test………. 32
4.4 Robustness Checks……….. 33
4.4.1 Changing Explanatory Variables………. 33
4.4.2 Changing the Time Period……… 33
4.4.3 Excluding Country Outliers……….. 33
1. Introduction
1.1 Entrepreneurship and Immigration
Over the last decades, self-employment among migrant groups has increased significantly in the world. This has made migrant entrepreneurship an important topic on the research agenda (Janssens, 2010). Our age has become one of mass migration, both via voluntary and forced movement of people. Between 1965 and 2000, individuals living outside their countries of birth grew from 2.2 % to 2.9% of world population, reaching a total of 175 million people in 2000 (USBC, 2002) and this number has been increasing up to 231 million people or 3,2% in 2013 (United Nations, 2013).
While this number can seem insignificant the fact that individual differences exist need to be considered. The United States for example has seen an increase in immigration since the 1980’s up to an estimated 40.4 million immigrants in 2013 (United Nations, 2013). This can be considered the highest in the world both in absolute and relative terms, constituting almost 15 % of their total population in 2013 (Nwosu et al., 2014).
However, apart from the positive sides of diversity simultaneously several problems and frictions came along with it
(Baycan-Levent et al., 2005). Mexicans who are trying to cross the U.S. border in order to find employment and live the American Dream are one frequent example being covered in media reports.
The influx of migrants with different socio-cultural origins is particularly true for the industrialized world (Massey & Denton, 1993). This influx has brought about several economic advantages such as a higher level of diversity in products for consumers (Mazzolari & Neumark, 2012). On the other hand it has led to multiple social and economic tensions such as discrimination against- and high
unemployment rates among those groups (Borjas, 1990).
These high unemployment rates seem to be particularly true for low skilled migrants who generally belong to the lower segment of the socio-economic ladder, mainly due to a deficiency in skills and education (Westhead et al., 2011). It is this lower socio-economic situation that can lead to a shift into self-employment by ethnic groups (Baycan-Levent et al., 2005). This has triggered what is often called ethnic or migrant entrepreneurship (Delft et al., 2000; Masurel et al., 2002).
Controversial remains whether immigrants provide valuable contributions to
such as what happened on September 9th 2001 in the United States can have a polarizing effect for the position of
migrants and it is generally considered that migrants are in a disadvantaged position, because of discriminatory and other disadvantageous circumstances (Borjas, 1986), and with few other alternative options start their own business out of necessity. Furthermore, the results of many studies already show a tendency that immigrants are more inclined towards self-employed professions (Verheul et al., 2001; Janssens et al., 2010).
However, Quantitative studies that test the self-employment hypothesis among and between different migrant groups are scarce and most studies focus on the cultural side of self-employment (Bogan & Darity jr., 2008; DeLancey, 2014;
Hofstede et al., 2003; Schlaegel et al., 2013)
1.2 Problem statement and relevance
A quantitative approach can therefore give an insight into differences between migrant groups as the numbers of self-employment rates mentioned can differ among
migrants, for example per migrant group, per generation and per country (Andersson & Hammarstedt, 2010). A recent example is shown in research conducted by Beckers & Blumberg (2013) in the Netherlands
which shows that first generation Chinese migrants have a 60% propensity to be self-employed which is high considering the 10% native average. Compared to a 5,4% propensity of Moroccan migrants. This shows that inter-group differences can be significant.
Especially in the United States this item is becoming more important as net
1999).
Secondly, in the existing literature a lack of understanding exists in linking
migration to the entrepreneurial process (Levie & Smallborne, 2006). As many ethnic entrepreneurs are equally immigrants the separation in research between the effects of ethnic culture and migration is often vague (Levie & Hart, 2007). The question whether origin or ethnicity itself is important remains therefore unanswered. Furthermore, the origin of the individual has been neglected as an area of research in the field of migration entrepreneurship (Williams et al., 2004).
Many authors have furthermore focused on a particular country, city, or individual case and neglect the broader context (Bogan & Darity jr., 2008). This makes it interesting to look whether results also hold for other migrant groups and take a comparative approach in a country where immigration rates are relatively high. As shown previously, the United States is such a country. Furthermore, despite the vast amount of literature, there seems to be a lack of research that explicitly considers immigration or ethnicity as a factor influencing entrepreneurship (Wang, 2010).
In conclusion it becomes clear that even though some explanations exist on what and why migrants can select themselves into entrepreneurial activities
(Kloosterman et al., 1999), it mostly compares migrants to the native population and hence forgets group differences and it does not take migration and ethnicity into consideration along with other social-economic factors (Wang, 2010).
Therefore the aim of this study is firstly to find explaining factors for migrants to enter self-employment. Firstly, this article will consider the level of discrimination dealt with by migrants as a possible explanatory variable here as
self-employment can be a rational response to the obstacles in the form of perceived discrimination, secondly the level of formal education is included and a last explanatory factor is found in the level of informal education.
enclaves which can be beneficial for the group as a whole (Logan et al., 1997). The quantitative and multi-group approach is therefore another gap in literature that this study tries to fill. The central question therefore will be what effect do the
perception of discrimination, the level of formal education and the level of informal business education have on the level of ethnic entrepreneurship for different migrant groups within the United States? To address this research question this research is set up as following firstly, a literature review is provided where hypotheses will be formulated. Subsequently a data & methodology section is provided where the
methodology, the validity & reliability of the approach and the variables that will be used are discussed. Then the results will be presented, followed by a discussion. After the discussion the limitations to this study are provided and lastly a conclusion of the main findings will be given and to what extent those findings contribute to and address the main goal of this study.
2. Literature review
2.1 Explaining ethnic entrepreneurship
When looking into the literature of migrant entrepreneurship it is suggested that self-employment is a popular shared form of
work for immigrants in several immigrant receiving countries (Bernhardt, 1994; Carr, 1996; Giulietti et al., 2012).
Ho & Wong (2007) identify at least 3 general types of motivations for entrepreneurs: opportunity driven entrepreneurs, necessity driven
entrepreneurs and high growth potential entrepreneurial activities. Opportunity driven entrepreneurs are those who are driven by the perceived opportunities for opening a new venture (Westhead et al., 2011). On the other hand there exists necessity driven entrepreneurship, which is defined as those who do not have any other option to survive economically and are self-employed out of necessity.
mostly on the differential in earnings between paid employment and self-employment. Furthermore, Constant & Zimmerman (2006) find that the
expectation of higher financial earnings pulls individuals into self-employment. However, apart from the opportunity of financial rewards there are other factors that are necessity driven that can explain Ethnic Entrepreneurship too. This triggers the question: How then exactly are these migrants motivated to be entrepreneurial out of necessity?
2.2 Discrimination
A first explanatory factor which is central in this research and will be presented here is the level of discrimination that migrants have to cope with. Some of the few studies on this topic show that discrimination can work as a restraining factor. Research conducted by Cavalluzo & Cavalluzzo (1998) in the United States on
discriminated self-employed immigrants for example shows that ethnic minorities have higher loan denial rates than their majority counterparts. Blanchflower et al. (2003) found that loan denial rates are two times higher for black business owners and interest rates to be paid over the loan were charged higher too even after controlling for financial records.
Ethnic discrimination in the credit market has also been documented in more recent studies in the United States especially (Blanchard et al. 2008; Asiedu et al., 2012). Discrimination among ethnic minorities is therefore present, and at first glance can have a restraining effect on entrepreneurship.
However, the opportunity to serve their own ethnic market and a lack of other opportunities to find employment will likely make migrants more susceptible for self-employment (Wang, 2010). This can give rise to protection of a market where members belonging to the particular ethnic group by using their own language and business customs can participate in (Clark & Drinkwater, 2000). As research
conducted by Aldrich et al. (1985) shows for example that minority entrepreneurs know better about the taste and preference of the ethnic markets they are targeting. Additionally, Janssens et al. (2010) argue that credit discrimination does not restrain migrants from starting a business, but it does select them into industries which need relatively few financial investments.
Kenneth Arrow (1973) in his theory of discrimination states that basic economic theory is focused on productivity and expands this by modelling discrimination. This theory is also related to the
The disadvantage that ethnic minorities experience on the labor market in terms of unemployment, can make them more susceptible for self-employment as they lack alternatives (Clark & Drinkwater, 2000).
In the theory of discrimination, Arrow (1973) deals with the notion that characteristics other than productivity, such as race and nationality are also valued on the market and which can pose a barrier into employment for migrants. In addition, the disadvantage perspective (Wang, 2010) considers difficulty in finding employment and low wages that are induced by
discrimination as a push factor for migrants into self-employment (Mora & Davila, 2005).
An example of such push factors is
provided in a study conducted by Raijman & Tienda (2000) who show that ethnic minorities that lack linguistic skills, reported a higher threshold mobility as a direct reason for self-employment as compared to their U.S. Native
counterparts. Among the many forms of discrimination that exist unemployment due to one of the factors mentioned earlier is economically the worst that could happen (Light, 1971). And unemployment is making workers unhappy (Wielers & Van der Meer, 2013). Furthermore,
migrant workers, notably men, have been hit hard by the economic crisis.
In 2010/2011 there were 7.1 million unemployed foreign-born workers in the OECD, corresponding to an average unemployment rate of 11.6% and this rate is significantly higher than the
unemployment rates under the native population (United Nations, 2013). Hence, self-employment can provide an escape route for immigrants out of unemployment. This is also suggested in a study by Thurik el al. (2008) who have conducted research on 23 OECD countries in the time period between 1974 and 2002. The conclusion they draw is that higher unemployment rates lead to more start-up activity by self-employed individuals which they call ‘’the refugee effect’’. Additionally, in their results they find what they call ‘’an entrepreneurial effect’’ which means that high rates of self-employment tend to reduce unemployment in the subsequent period.
These difficulties to get access to a decent paid-employment alternative or other form of labour market discrimination can influence migrants to take up self-employment as a means to survive
immigrants in the U.S. to become self-employed is the perception of
disadvantages for non-business occupations, such as not speaking the language and not having an U.S. education. Faggio & Silva (2012) have conducted a similar study in Great-Britain and find that migrants who feel they lack employment options have a higher propensity to become self-employed.
Discrimination is a disadvantage which is not researched broadly in the existing literature. Therefore, in this research it is argued that discrimination itself is hypothesized to be a factor to positively influence the decision to become self-employed among migrants.
Hypothesis 1: The perception of being discriminated against among migrants will lead to higher levels of entrepreneurship among those migrants.
Hypothesis 2: The perception of feelings that racial differences exist will lead to higher levels of entrepreneurship among migrants.
2.3 Formal Education
Apart from the perceived level of discrimination other factors can equally funnel migrants into self-employment. According to neoclassical theory it is human capital in terms of high education, previous business experience and language
skills that have a significant effect on the entrepreneurial process (Fairlie, 2008). For education several effects are known in the existing literature. Firstly, education endows individuals with a skillset and knowledge that gives them an advantage in organizing and operating their business (Sanders & Nee, 1996). When applying this to immigrants in the United Wadwha et al. (2007) found that 25 % of
engineering and technology start-up companies in the United States over the past decade were founded by immigrants as compared to 14% of the native
population.
However, research conducted by Fairlie (2008) indicates that the largest
contribution of new immigrant business owners by education level is from those who have lower than a high school education. A possible explanation is provided by McMullen et al. (2008) who indicate that the opportunity costs of an educated individual are higher than those of someone with only few education and therefore highly qualified individuals are less inclined to engage in entrepreneurial activities.
separation of five types of skilled migrants. Firstly, managers and executives, secondly engineers and technical people, which is considerably determined by the intra-firm employee transfer strategy of firms. Then academics and scientists who are increasingly mobile on an international scale. Fourthly, students are constituting a major part of potential source of labor for especially the knowledge economies and lastly there is entrepreneur migration which is determined by the presence of opportunities and a sound institutional framework such as access to capital and favorable tax regimes.
The lack of job opportunities because of discrimination can therefore have a restraining factor on the willingness to go through university among migrants as job opportunities are hard to come by even with a decent education (McMullen et al., 2008) and under such circumstances entrepreneurial activities are more feasible and can provide a better opportunity. Moreover, when considering the possibility that groups of migrants often cluster
around areas of economic activity and form what is called an enclave
(Abrahamson, 1996) An enclave is defined in their research as ‘’a concentration of individuals from the same ethnic
background within a specific geographical
location’’. This can give rise to protection of a market where particularly members belonging to the particular ethnic group by using their own language and business customs have opportunities to employ themselves (Clark & Drinkwater, 2000). Hence also the broader environment can be important.
In addition, research by Evans (1989) in this context indicates that it is group fluency what is vital here as migrants who do not control the language in the country of residence form a linguistic pool of labour and it is therefore most efficient for them to get a job at a co-ethnic
entrepreneur. This indicates that the host country language is of less importance when migrants cluster. Hence, it is the network and not so much the education that is of importance.
Additionally, research conducted by Aldrich et al. (1985) shows that as ethnic entrepreneurs know better about the taste and preference of the ethnic markets they are targeting. Secondly, the point that is related to human capital theory is the point of view that language deficiencies among immigrants is another disadvantage on the labour market, but in enclave communities the host-country language is of less
importance as previously mentioned. Therefore the degree to which an
is hypothesized to be of less importance to migrants in their propensity to become self-employed.
Hypothesis 3: Migrants with lower College Degree are more inclined to be
self-employed
Hypothesis 4: The longer period in years a migrant studies the less likely he/she is to become self-employed
2.4 Informal Education
In addition to the formal educational background of migrants, the
entrepreneurial activities employed can equally be influenced by the Socio-Economic environment (Carlsson & Braunerhjelm, 2013). Explanations for Entrepreneurship given in the existing literature are among others in the intergenerational transmission of self-employment (Andersson & Hammarstedt, 2010). This is a part of ‘’social capital theory’’. Social capital is defined by Falk & Kilpatrick (2000) as an accumulation of the knowledge and identity resources drawn on by communities-of-common-purpose.
Inter-generational links in self-employment propensity fits in there and can act through a variety of channels. An individual whose parents are self-employed has an
opportunity to acquire specific human
capital on how to run a business which will make self-employment more accessible (Laferrère, 2001). Additionally, people with self-employed parents can engage in entrepreneurship by taking over the family business and by taking over the business those individuals can inherit the network that their parents have built up over the years (Dunn & Holtz-Eaking, 2000; Hundley, 2006)
Thus, self-employment can show a
correlation across generations since parents can provide their offspring with informal business experience. It is this immediate social environment that is providing social support in terms of a set of practical skills and experience for which no formal education exists (Andersson & Hammarstedt, 2010).
Furthermore, an immigrant family who has their own business tends to belong to a higher social class and has greater access to financial capital (Laferrère, 2001). This reduces the restraining loan credit
use to migrants. However, informal education via inter-generational
transmission can create awareness for the self-employment opportunities (Andersson & Hammarstedt, 2010).
Lastly, research conducted by
Blanchflower & Oswald (1998) shows that children of self-employed parents are disproportionally more inclined to be self-employed than children whose parents are not self-employed. Therefore it is expected that for immigrants the same is true.
Hypothesis 3: Migrants whose parents are self-employed have a higher propensity to be self-employed themselves.
Hypothesis 3a: Migrants whose mother is self-employed have a higher propensity to be self-employed themselves.
Hypothesis 3b: Migrants whose father is self-employed have a higher propensity to be self-employed themselves.
2.5 Conceptual model
In Figure 1 the conceptual model that is the foundation of this research is shown. The dependent variable is Ethnic
Entrepreneurship and the independent variables are labeled as formal education, which is supposed to have a negative linear effect on entrepreneurship as education increases alternatives. Furthermore, the level of perceived discrimination among migrants is expected to have a positive linear effect on entrepreneurship as it can drive migrants into entrepreneurship. Lastly, informal education also is expected to have a linear positive effect on
Entrepreneurship as it increases the opportunity to become self-employed.
3 Data & Methods
The data that will be used for testing the hypotheses come from the General Social Survey (GSS). This is a database with U.S. demographic, behavioural and attitudinal data. The dataset has a total of 5.561 variables and has 16.128 valid respondents that have an ethnic background outside the United States. Therefore this dataset is useful to gain insights into the migrant American Migrant Population over the time period from 2002 to 2012.
The data is collected every two years. How this dataset will be used is explained in the subsequent paragraphs by describing which variables will be used and also why these variables are chosen, which method will be followed to construct the models and lastly the validity and reliability will be discussed.
3.1 Dependent variable
The dependent variable will be the level of entrepreneurship. According to Carlsson & Braunerhjelm (2013) research in the field of entrepreneurship has evolved rapidly over the last years. In their article they provide an overview of the developments made in the field of entrepreneurship. Firstly, the definitions proliferated in this area of research and the history of the research field underlying entrepreneurship
research are reviewed to find out
measurement tools that are justified as a proxy to measure the level of
entrepreneurship among migrant groups. The question of what exactly is
constituting entrepreneurship is a question that has been the focus of many previous studies in which authors often created their own constructs and definitions (Evans & Leighton, 1989; Evans & Jovanovic, 1989; & Blanchflower & Oswald, 1998).
Therefore there does not seem to be a uniform definition of entrepreneurship. It is however widely accepted that
entrepreneurship involves the creation of new firms and the development of already existing firms (Westhead et al., 2011)
3.1.1 Defining Entrepreneurship
Acs & Audretsch (2003, p. 6) defined entrepreneurship as ''embracing all business that are new and dynamic,
regardless of size or line of business, while excluding businesses that are neither new nor dynamic as well as all non-business organizations''. According to Carlsson & Braunerhjelm (2013) entrepreneurship refers primarily to an economic function that is carried out by individuals,
uncertainty, by making decisions about location, product design, resource use, institutions, and reward systems. But what then constitutes ethnic entrepreneurship?
3.1.2 Ethnic entrepreneurship
In ethnic entrepreneurship literature there are three definitions being used most often: migrant entrepreneurs, ethnic
entrepreneurs and minority entrepreneurs (Chaganti & Greene, 2002). Whereas, immigrant entrepreneurs can be defined as individuals who, as recent arrivals in the country, are starting up businesses in order to survive in economic sense. This group may consist of former migrants and non-migrants who form a migration network with a common origin and destination (Butler & Greene, 1997).
Ethnic entrepreneurs are then a combination of regular patterns of
interaction and connections of persons who share a common national background or migration experience (Waldinger et al., 1990). And lastly, minority entrepreneurs are described in the literature as business owners who are not of the majority population. Chaganti & Greene (2002) define this type of entrepreneurs by looking at the U.S. Federal categories which include Hispanic or Latin American, Asian, American Indian etc. This research will focus and use the term ethnic
entrepreneurship and the national background will be central.
3.1.3 Corporate entrepreneurship
Another stream of entrepreneurship research is focused on Corporate Entrepreneurship. Corporate
entrepreneurship (CE) is a potential survival strategy for firms that operate in highly competitive business environments (Petolla, 2012). Three main forms of corporate entrepreneurship can be identified (Stopford & Baden-Fuller, 1994): firstly, the creation of new business activities that exist within the boundaries of the existing organization, secondly the transformation or the renewal of existing organization and lastly the organization changing the rules of competition in the industry in which it is active. The
overarching component in the three forms is the fact that it takes places within an existing company (Westhead et al., 2011).
3.1.4 Implications for research
with the criteria set in the definition by Carlsson & Braunerhjelm (2013) to start something new, by having the possibility to give direction to your own work and the bearing of risk. Firstly, the categorical variable of ‘’Wrkslf’’ that is the dependent variable and labeled as either working for somebody else or self-employed is changed into a binary for either 1=self-employed or 0=working for somebody else. In TABLE 1 the descriptive statistics
of this variable are shown. Over the 2002 – 2012 period a total of 1.812 observations were made for self-employment in the United States or 11.70% of the total number of observations. 13.652
observations were made as working for somebody else, or 88,30 % of the total. 664 observations are classified as missing. The consequences and countermeasures of which are discussed in more detail under the Validity & Reliability section.
TABLE 1 – Type of Employment ‘’Self-Employment vs Employment rates’’
Type of Employment Employed Share of total in %
Self-Employed 1.812 11,70% Employed Total 13.652 15.464 88,30% 100,00%
*Notes: 664 observations were reported as missing or 4.1 % of the total 16.128 observations in the 2002-2012
time-period.
TABLE 2 – Type of Employment
‘’Self-Employment vs. Employment rates distributed for the 2002-2012 period’’ Year Self- Employed Employed % Self-Employed Total respondent
2002 307 2.362 11,50 % 2.669 2004 343 2.353 12,72 % 2.696 2006 508 3.799 11,79 % 4.307 2008 226 1.732 11,54 % 1.958 2010 234 1.706 12,06 % 1.940 2012 194 1.700 10,24 % 1.894 Total obs. 1.812 13.652 11,71 % 15.464
In TABLE 2 those numbers are specified per year. It shows that percentages stay between the 10,00 % - 12,00 % boundary approximately with some decline in 2012. A possible explanation for the variance can be the lower number of respondents for this year. According to research conducted by Hipple (2012) on self-employment rates in the United States these rates are similar where the average by Hipple (2012) over the same 10 year period comes down to 11,8%.
3.2 Key independent variables
In this research three main independent variables are distinguished. These are firstly the perceived discrimination among migrants, secondly the degree of formal education a migrant possesses and lastly the degree of informal business education are hypothesized to be driving factors behind ethnic entrepreneurship.
3.2.1 Discrimination
The level of discrimination is a key independent variable under research. As the part on entrepreneurship shows, no single method is without limitations and there exist a range of perspectives that help measure, how, and to what extent
discrimination matters in the lives of immigrants within the United States.
A review of measurement tools in the existing literature proliferated at least four main perspectives which all will be elaborated on subsequently: the perceptions of discrimination by
minorities, perceptions by the majority or dominant group, research focused on outcomes of discrimination and lastly by looking at discrimination laws and lawsuits (For an elaborate literature review on the 4 perspectives see Appendix A).
The variables from the GSS that will be used are DIS1 and DIS2, where DIS1 will be used as a proxy for the perception of discrimination by looking how the question ‘’Do you have the feeling that racial
differences exist (Yes or No)?’’ relates to entrepreneurial activity. And additionally by looking how the question ‘’Do you have the feeling you are being discriminated against (Yes or No)?’’ relates to entrepreneurial activity among migrant groups.
3.2.2 Formal Education
The role of human capital will be measured by looking at separately the level of
education, by looking at their degree. Also the years of education will be taken into account. For the level of education
systems in the different countries of origin can differ in terms of quality.
According to Hoffmeyer-Zlotnik &
Warner (2006) one of the ways to measure Human Capital is by looking at education which can be measured by a metric scale of the number of years of full time
education a person has followed. However as this does not say anything about the level of their education, an additional variable that measures which degree of college the respondent holds, is added. Categories for holding a degree are lower than High School, High School, Junior College, Bachelor Degree and Graduate Degree. After tabulating these categories dummies are created for which Graduate Degree will be omitted.
To find out whether both variables can form a Human Capital scale Cronbach’s Alpha was computed (α = 0,743). This value is above the threshold of α > 0,743 and therefore both variables can form a scale. However multicollinearity can pose a threat (Hair Jr., et al., 2010) and needs to be tested and accounted for (See paragraph 3.4.5. on multicollinearity).
3.2.3 Informal Education
Furthermore, the level of informal education will be analyzed by looking at whether an individual’s propensity to
become self-employed depends on the occupation of their parents. This is in line with Nauck (2001), who calls the passing on of business related information
‘’intergenerational transmission’’.
Parents who are self-employed can provide the necessary start-up capital to overcome the financial threshold, provide specific knowledge and give the respondent access to their network (Blanchflower et al., 2003). The variables that are used are ‘’Mawrkslf’’ and ‘’Fawrkslf’’ that represent binary variables for either 0=Mother is working for somebody else and 1=Mother is self-employed or
0=Father is working for somebody else and 1=Father is self-employed respectively.
3.2.4 Subgroups
There are different types of migrants, often ethnic minorities are equally (im)migrants (Clarke & Drinkwater, 2000). Ethnic Background is the variable that is used here to create subgroups. In TABLE 3 the descriptive statistics on Entrepreneurial activity for each of a total of 8 subgroups are given. And in TABLE 4 the
TABLE 3- Separating the Subgroups
’Summary of between Country-Differences in Entrepreneurship’’
Country of origin* Employed Entrepreneur % Entrepreneur Total Observations
Ireland 1.366 188 12,10 % 1.554 France 193 31 13,83 % 224 Germany 1.587 232 12,75 % 1.819 Mexico 791 67 7,80 % 858 Czech Republic 102 17 14,28 % 119 Polen 265 33 11,07 % 298 England 1.169 195 14,29 % 1.364 Other 4.439 573 11,43 % 5.012 Total 9.912 1.336 11,87 % 11.248
*Notes: The category others includes: Austria, Belgium, Canada, China, Denmark, Finland, Greece, Hungary,
Italy, Japan, Lithuania, Netherlands, Norway, Philippines, Poland, Portugal, Romania, Russia, Scotland, Spain, Sweden, Switzerland or Yugoslavia.
**Notes: A more elaborate descriptive analysis can be found in Appendix D
TABLE 4 - Separating the Subgroups ’Summary of between Country-Differences in %’’
Ethnic Background* 2002 2004 2006 2008 2010 2012 Ireland 12.45% 8.35% 9.57% 10.77% 9.68% 9.92% France 2.48% 1.74% 1.26% 1.53% 1.32% 1.27% Germany 15.45% 11.93% 10.68% 12.16% 11.83% 11.95% Mexico 3.93% 4.73% 7.38% 6.37% 6.21% 6.73% Czech Republic 0.88% 0.07% 1.27% 0.49% 0.58% 0.30% Polen 3.13% 1.50% 1.95% 1.33% 2.15% 1.92% England 10.81% 8.35% 8.62% 8.55% 7.92% 8.10% Other 51.05% 63.29% 59.24% 58.77% 60.22% 59.77% Total 100% 100% 100% 100% 100% 100%
*Notes: The category others includes: Austria, Belgium, Canada, China, Denmark, Finland, Greece, Hungary,
3.3 Control variables
There are several variables that need to be controlled for in this study as they also determine entrepreneurial activity. Discrimination, Formal Education and Informal Business Education are only three of the explaining mechanisms.
As mentioned in the introduction cultural differences are left out on purpose, as the last decade especially has proliferated a rich stream of research that captures the cultural component as an explaining factor for entrepreneurial activity (Bogan & Darity jr., 2008; DeLancey, 2014; Hofstede et al., 2003; Schlaegel et al., 2013) Previous studies have shown that migrant entrepreneurship is also a
consequence of individual characteristics such as age and gender (Zhou, 2004). In this research therefore gender, age, and race are controlled for. The effect of ethnic background is of interest, and
discrimination can also be influenced by skin color therefore there needs to be controlled for race. The variable Race that will be used is a categorical variable with three possibilities that are White, Black or Other. Age is a variable that is adapted so that only individuals in between the 18 ≥ v. ≤ 67 age category.
The reason behind this is that respondents under 18 are considered too young to work
for somebody else or to be self-employed, whereas respondents over 67 retire
especially the ones who were working for somebody else, and thus self-employed individuals can be overrepresented if this was not accounted for.
Lastly, gender is accounted for as especially female entrepreneurship is a recent phenomenon that has proliferated a rich stream of research (Verheul et al., 2006; Rosenthal & Strange, 2012; Terjesen & Amorós, 2010; Holmén et al., 2011). A complete overview of the descriptive statistics of all variables is found in Appendix D.
3.4 Method
The model is tested using time series data from 2002 – 2012 with a delta of two years. As the dependent variable is categorical with either a migrant being ‘’self-employed’’ or ‘’working for somebody else’’ a binary logit model is constructed to make inferences about subgroup migrant populations within the United States. Furthermore, OLS Pooled estimation techniques will be used to conduct those inferences.
exp(𝑥𝛽)
1+exp(𝑥𝛽) , where π represents a link function, the cumulative logistic
distribution function. The goal is to infer an Odds Ratio to determine the likelihood per subgroup for engaging in
entrepreneurial activities.
The binary log of the odds model can be used for making such inferences and can be expressed in a log-linear form of lnΩ(x) = xβ, where Ω(x) is the odds-ratio for success (y = 1) given x (Long, 1997). The odds ratio will be used to closer examine whether change occurs in the odds when an independent variable xodds increases by δ; An odds ratio that is greater than 1 indicates that the odds increase as the variable under research increases by δ (p. 80-82).The odds can be calculated as follows: Ω(x) = 𝑃(𝑦=1│𝑥)
P(y=0│x) =
𝑃(𝑦=1│𝑥) 1−P(y=1│x)
= π(xβ)
1−𝜋(𝑥𝛽) The odds ratio then is calculated as follows: Ω(x−odds,xodds+δ)
Ω(x−odds,xodds) = exp(βoddsδ).
3.4.1 The Migrant entrepreneurship model
The following binary logistic regression model is created as shown by the following equation where Ent ie represents the level of entrepreneurship per individual(i) by looking at their ethnic background(e). Then, β0 represents the constant term,
Ent ie = β0 + β1 DIS1ie + β2 DIS2ie + β3 DEGie + β4 EDUie + β5 MASie+ β6 FASie + β7 SEXie + β8 RACie + β9 AGEie + εie Where β0 represents a constant and εie represents the error term for the unexplained variances that are not
explained by the model. The abbreviations of the variables are as follows:
Ent ie = Level of entrepreneurship per individual(i) by looking at whether that individual is self-employed or working for somebody else and taking into account their ethnic background(e)
β1 DIS1= Level of perceived
discrimination by looking at if an
individual (i) with ethnic background (e) feels discriminated or not (yes/no)
β2 DIS2= Looks whether a respondent (i)
with ethnic background (e) feels racial differences exist (yes/no).
β3 DEG= The highest degree obtained by
the respondent (i) with ethnic background (e) (Lower than High School, High School, Junior College, Bachelor and Graduate)
β4 EDU= The amount of years of
Education a respondent (i) with ethnic background (e) has had
β5 MAS= Whether mother of respondent
β6 FAS= Whether father of respondent (i)
with ethnic background (e) is
self-employed or working for somebody else
β7 SEX= Gender of respondent (i) with
ethnic background (e) (1=male)
β8 RAC= Race of respondent (i) with
ethnic background (e) (White, Black or Other)
β8 AGE= Age of the respondent (i) with
ethnic background (e) (can take on values between 18-67)
3.4.2 Multicollinearity
Looking at this base model, six
independent variables are used and three control variables are included. If the
independent variables and control variables are highly correlated it can cause a
problem for the predictability of the model. It can distort the coefficients estimations of the model and for that reason a
multicollinearity check is conducted (De Vries & Huisman, 2007).
To check if multicollinearity is a problem Variance Inflation Factors (VIFs) are computed. The boundary values for VIFs is VIF <.10, or VIF > 10, and a value higher than 4 is considered to show possible signs of multicollinearity (Pallant, 2013). In the appendix B an overview of the computed VIF values is presented. This shows that the variables ‘’Years of Education’’ and
‘’College Degree’’ show values of between 4 and 10 (Appendix B).
Therefore to look further into the possible multicollinearity issue a correlation matrix is constructed based on the bivariate correlations of the variables. For the variables College Degree and Years of Education a strong correlation is present as the correlation coefficient is 0.8740. A bivariate correlation coefficient above 0,7 is considered too high (Pallant, 2013). The correlation matrix further shows no further relationships seem to exist and after excluding the correlating variables multicollinearity is not an issue anymore (Appendix C).
3.5 Validity & Reliability
Under this section the Validity & Reliability are discussed. Firstly, the selection bias in the sampling process is discussed, subsequently the missing values that are present in the sample will be elaborated upon and lastly a paragraph about causality is created to indicate how the created structures of the relationship in the model should be interpreted.
3.5.1 Selection bias
the General Social Survey by conducting the sampling process in four separate phases and including all groups of the American Society. Until 2006 the GSS only sampled the English speaking population. Spanish speakers typically make up 60%-65% of the language exclusions.
About a dozen languages make up the remaining exclusions. Starting in 2006 the GSS samples Spanish speakers in addition to English speakers. This means that before 2006 migrants who did not speak the language did not participate and were underrepresented in the sample. A quota sampling strategy is used to have every group represented in the sample selected by among others age and education. A more elaborate discussion on the sampling strategy from the GSS website is provided in Appendix E.
3.5.2 Social Desirability bias
With specifically measuring the perceived level of discrimination, it can distort the actual level of discrimination. Social desirability bias can prevent individuals to talk about sensitive topics such as race and discrimination (Fischer, 1993). However, no single approach of the four described ways to measure discrimination is without flaws. And by taking into account several variables to measure the constructs and
relate it to the one best suiting the research question will increase the validity of the research (Pager & Shepherd, 2008). At the same time it is important to acknowledge that the nature of specifically
discrimination itself is dynamic and patterns of discrimination can be shifting over time (Massey, 2005).
3.5.3 Missing Values
With a data collection process the risk exists that some values are missing. In time series data there is always a risk that data for a particular year, subgroup or other entity can be under- or overrepresented (Fitzmaurice et al., 1996). Effectively, this means that the sample of those years are less balanced. And looking at the 2002 – 2012 period, see also TABLE 2 and TABLE 4, it can be seen that less data is gathered since 2008.
In the data researchers from NORC included these factors. In the analysis, listwise deletion of missing values is applied. With possible social desirability bias, the missing values can be non-random (Roth, 1994) and valid cases for other variables can be left out of the analysis as a consequence. Additional robustness checks will therefore be conducted.
3.5.4 Causality of Relationships in the Model
The cross-sectional approach of Pooled OLS in this research means there is only one point in time for which inferences are made, even though patterns can exist. These patterns cannot be detected by this type of research. The patterns are
depending also on the time varying character of the variables that are being studied and especially the perception of discrimination can vary significantly over time (Pager & Shepperd, 2008).
The other two main independent variables can be considered to be more time constant but can change over longer periods of time. Events such as what happened at
September 11th in New York, or what happened with Rodney King in 1991 can lead to an increased perception of
discrimination among migrants who share the ethnic background of such an
individual. Apart from additional
robustness checks therefore time dummies will be added.
By using time dummies for the two-year time intervals the migrant populations will be compared regarding their stability. A Chow-test is conducted to do so. This is a statistical test to show whether the
coefficients in two linear regressions on different datasets are equal and it can also be applied to time series analysis to test for the presence of a structural break (Howard, 1989). Therefore the following formula of Gregory Chow (1960) is used:
𝐶ℎ𝑜𝑤 = (RSS−RSS1−RSS2/K
(RSS1+RSS2)/(𝑁1+𝑁2−2𝐾) ~ Fk, n1 + n2 – 2k
To conduct the Chow-test independent regressions will be run for the separate groups under research and finally all data will be pooled for a regression and the F-statistic will be computed to look for discrepancies in the data between the groups and find out about the stability of the data over time. In TABLE 5 on the next page an overview of the time differences in employing entrepreneurial activities as a migrant is provided. Dummy variables were created to show the
TABLE 5 – Yearly Entrepreneurship propensity
‘’Yearly differences in Entrepreneurship propensity for all migrants’’ TABLE 5
Predictor* B P-value Odds ratio
Year 1 – 2002 .130 .180 1.139 Year 2 – 2004 .245 .010** 1.277 Year 3 – 2006 .159 .076* 1.172 Year 4 – 2008 .134 .196 1.143 Year 5 – 2010 .184 .074* 1.202 Constant -2.171 <.001*** .114
*Notes: Significant at the 99 % confidence level. The dependent variable is Entrepreneur (0= no, 1 = yes). The omitted variable for year is Year 6 (2012).
**Notes: Statistically significant on a 95 % confidence interval
***Notes: Statistically significant on a 90 % confidence interval
4 Results
Under this section the results of the study will be described. Firstly, two standard models are tested, one with the variable ‘’Years of Education’’ and the other with ‘’College Degree’’ for measuring the degree of Formal Education. After presenting the baseline results, the model will be elaborated by Country-Variable extensions to predict the probability of migrants to become an entrepreneur in the United States based on their ethnic
background.
4.1 Baseline model results
Binary logistic regression is performed to assess the impact that each of the
explanatory factors in the model has on predicting that respondents will be an entrepreneur.
A dummy variable was created for the dependent variable entrepreneurship ( 1 = Self-employed or 0 = Working for
TABLE 6 – Baseline Results
Base Model 1 with variable ‘’Years of Education’’ included MODEL 1
Predictor* B P-value Odds ratio
Discrimination 1 (1 = Yes) -.141 .237 .869 Discrimination 2 (1 = Yes) -.459 .051* .632 Years of Education .016 .406 1.016 Mother Self-Employed .517 .001** 1.677 Father Self-Employed .385 .002** 1.470 Age .019 <.001*** 1.019 Race - - - White -.141 .760 .959 Black -.782 .006** .482 Gender (1 = Male) .648 <.001*** 1.912 Constant -3.739 <.001*** -
*Notes: Significant at the 99 % confidence level. The dependent variable is Entrepreneur (0= no, 1 = yes). The omitted variable for degree is graduate. The omitted category for Race is Other. For the variable gender Female is the omitted variable.
Considering the full model it can be said that in both cases the model as a whole was statistically significant, in the first case with the inclusion of ‘’Years of
Education’’ χ2 (N=3638) = 81,007) and p <.001. This is indicating that the total model is able to predict correctly who will be an entrepreneur on a 99% confidence interval. When looking at the model fit the Nagelkerke R square test is 0,044 and the Cox & Snell R square is 0,022 which means the model is able to explain 4,4 % or 2,2 % in the detected variance.
TABLE 7 - Baseline Results
‘’Base Results Model 2 with variable ‘’College Degree’’ included’’ MODEL 2
Predictor* B P-value Odds ratio
Discrimination1 (1 = Yes) -.141 .236 .868
Discrimination2 (1 = Yes) -.447 .058* .640
College Degree - - -
Lower than High school -.119 .658 .888
High School .033 .852 1.034 Junior College -.029 .906 .971 Bachelor .213 .276 1.238 Mother Self-Employed .385 .002** 1.470 Father Self-Employed .517 .001** 1.678 Age .019 <.001*** 1,019 Race - - - White -.068 .740 .947 Black -.774 .007** .483 Gender (1 = Male) .649 <.001*** 1.914 Constant -3,862 <.001*** .021
*Notes: Significant at the 99 % confidence level. The dependent variable is Entrepreneur (0= no, 1 = yes). The omitted variable for degree is graduate. The omitted category for Race is Other. For the variable gender Female is the omitted variable.
**Notes: Statistically significant on a 95 % confidence interval
***Notes: Statistically significant on a 90 % confidence interval
discrimination and for the perception that racial differences exist. The variables are however not statistically significant at the 95% confidence interval. The perception that racial differences exist is however significant at a 90 % confidence interval. This is an unexpected result.
TABLE 8 – Between Country Groups Analysis
‘’Predictor outcomes of the base model extended with Country Dummies’’
MODEL 3
Predictor* B P-value Odds ratio
Discrimination 1 -.130 .333 .878 Discrimination 2 -.443 .055* .642 Years of Education -.008 .729 .992 Father Self-Employed .545 .001** 1.725 Mother Self-Employed .550 <.001*** 1.733 Age -.018 <.001*** .982 Race - - - White .316 .328 1.372 Black 1.364 .028** 3.912 Gender (1 = Male) .753 <.001*** 2.018 Country Dummies* - - - Czech Republic -.222 .774 .801 England .460 .538 1.910 France -.478 .488 .620 Germany .519 .076* 1.680 Ireland -.162 .505 .850 Mexico .326 .621 1,386 Poland -.610 .334 .543 Time Dummies* - - - Year 1 – 2002 .191 .463 1.210 Year 2 – 2004 -.271 .232 .763 Year 3 – 2006 -.110 .557 .895 Year 4 – 2008 -.131 .525 .877 Year 5 – 2010 -.045 .824 .956 Constant 2.112 .028** 8.269
TABLE 9 – Between Country Groups Analysis
‘’Predictor outcomes of the base model extended with Country Dummies’’
MODEL 4
Predictor* B P-value Odds ratio
Discrimination 1 -.128 .342 .880
Discrimination 2 -.423 .121 .665
College Degree* - -
Lower than High-School .128 .689 1.137
no linear negative trend as expected in the coefficient of College Degree. In addition has to be remarked that neither of the variables is statistically significant with an α ≤ 0,05.
Considering the full model with the variable College Degree included it can be said that in the second case the model as a whole is also statistically significant, χ2 (N=3642) = 83,471 p <.001. This is indicating that the total model is able to predict correctly who will be an
entrepreneur on a 99 % confidence
interval. In block 0 it is shown that 89,2 % of the respondents are working for
somebody else. In the block 1 omnibus test Nagelkerke’s R square is 0,046 and Cox & Snell R square is 0,023 which means that the model explains 4,6 % and 2,3 % of the variances in the dataset respectively.
4.2 Country background extensions
Apart from the explanatory powers behind ethnic entrepreneurship it is likely inter-group differences exist (Janssens, 2010). Therefore country dummies are generated for Czech Republic, England, France, Germany, Ireland, England, Mexico and a category for other countries with only few observations.
When looking at the results in TABLE 8 on page 29 it can be noted that in the block 0 outcomes for this Model 3 that includes
the country dummies and the years of education variable, it can be seen that 89,6 % of the total respondents is classified as working for somebody else. In the block 1 outcomes for Model 3 it can be noted that the model is significant at the 95%
confidence interval (N=1230, Chi-square=36,482). When looking at the explanatory power of the model
Nagelkerke’s R square is 0,059 and Cox Snell R square is 0,029. This indicates that respectively 5,9 % and 2,9 % of the
variance in the data is explained by the model. When looking at the results in the model most outcomes have the expected signs, but are not statistically significant. However, if a migrant his/her mother is self-employed the migrant is 1,896 times more likely to be self-employed too. Comparing these results with model 4 which is shown in TABLE 9 on page 30 it can be seen that the signs and significance levels do not alter, however the strength of the coefficients changes slightly.
Furthermore, for the country variables and the College Degree variable, it can be seen that in the block 0 outcomes equally 89,6% of the respondents included in the analysis are classified as working for somebody else.
(N=1230, Chi-square=37,568). When looking at the explanatory power of the model Nagelkerke’s R square is 0,062 and Cox &Snell R square is 0,031. This
indicates that respectively 6,2 % and 3,1 % of the variance in the data is explained by this model. When interpreting the results it can be noted that the findings are in line with those of model 3 and thus only partially confirming the hypotheses. The low N-value in both models can be the cause of those results.
When selecting only the cases for the countries included in the analysis
separately, it can be noted that the results of the base model also hold. Some
differences exist however in the strengths of the coefficients. For England it can be noted that a migrant whose father is self-employed is three times more likely (Odds Ratio = 2.827) to be self-employed too, than someone from England whose father is not self-employed.
A contrary effect can be found for migrants from Ireland, where the likelihood of a migrant being an entrepreneur declines when their Father is self-employed (Odds Ratio = 0.878). Furthermore, for Mexicans the role of the Mother seems to be key for engagement in entrepreneurial activities (Odds Ratio = 10.163). The models for each of the separate nationalities under research can be found in Appendix F.
4.3 Chow test
A Chow test is conducted to find out if differences between the variances of the subgroups exists. The results of the base model, with only selecting each of the separate nationalities from Appendix are used together with the pooled results from Model 3 and Model 4. The variable Race was excluded from the analysis as too few observations were included in the data too make inferences for all six years for some of the nationalities. The following formula is used for the Chow test as suggested by Gould (2011):
= EssOverall − (Ess1 + Ess2) 𝑘
𝐸𝑠𝑠1+ 𝐸𝑠𝑠_2 𝑁1 + 𝑁2 − 2 × 𝑘
Two countries will be compared at a time to the overall model. Further, for Czech Republic and France two extra variables were excluded due to a lack of respondents and for France one variable was left out the analysis for this same reason. The
In the subsequent chapters a series of three more robustness checks will be conducted to find out if signs and p-values change and to what extent the model will predict the same results under different
circumstances.
4.4 Robustness checks
To find out if the model is robust three extra tests will be conducted. Extra
explanatory variables will be added to find out if, and how the coefficients and
significance levels in the model change. Secondly, the time period will be limited to cases within the 2006 – 2012 period to see if it changes results and lastly by looking at the country outliers in the sample and by excluding them from the analysis.
4.3.1 Changing Explanatory Variables
In the original model the Degree of Human Capital and the Degree of Social Capital are highly time constant. They can change over time, but not as much as the
perception of discrimination. Therefore a new variable will be tested for the
perception of discrimination. This will be whether the ‘’respondent feels that it is important to have the right race to get ahead’’.
In Appendix G a full overview of the model is provided and it can be seen that the effect of perceived discrimination is
less negative on the chance to be self-employed, however it is not statistically significant.
4.3.2 Changing the Time Period
In addition to changing the variable for perceived discrimination also the time varying character of the perception of discrimination (Pager & Shepperd, 2008) is taken into account. It can be noted that in the United States several incidents relating to discrimination have occurred since the start of the sample in 2002. One of those events, in positive sense, is that Obama was elected the first black president of the United States in 2008. Therefore, a post 2006 analysis of the country-extension model is also conducted to find out how this has influenced the results. A full overview is provided in the Appendix H. The signs for the coefficients have not changed although the negative effect of perceived discrimination seems to be lower it is however not statistically significant.
4.3.3 Excluding Country Outliers
Germany and England together represent approximately 32% of the total
observations.
However, by selecting only the cases that exclude these two countries from the analysis does not chance the signs of the coefficients nor does it change the significance levels of the predictors included in the model. The predicting power of the model in explaining the variance in entrepreneurship rate does increase (Cox & Snell R2 = 0.036 & Nagelkerke’s R2 = 0.074). Individually excluding one of these three countries does not significantly changes signs or
significance levels of the predictors. An overview of these findings can be found in Appendix I.
5 Discussion
Looking at the results in the model, not all outcomes are as expected based on the findings from the literature. The results of the discrimination variables did for example not show the positive signs. This finding accounted for both variables that were used as a proxy for the perception of discrimination. Furthermore, it was a robust finding even after changing both variables for a third discrimination
variable. However, not in all models these findings were statistically significant.
In the base model and the extended country model however, the perception of migrants that feel racial differences exist showed a lower propensity to be self-employed and this finding was statistically significant with an α of 0,1 and this was contrary to the expectations.
A possible explanation can be that the discrimination by the institutional environment and the necessary entrepreneurial skills to become
self-employed among migrant groups can lower the perceived opportunities for engaging in entrepreneurial activities (Coate &
Tennyson, 1992) which makes them unfit to become self-employed in their own perception.
The cultural backgrounds of the selected subgroups in this research can be
unappealing towards entrepreneurial activities (Bogan & Darity Jr., 2008). Moreover, the select number of subgroups can be distorting actual results. Thus, contrary partial effects were found for the perception of discrimination, however they did not show the expected signs and hence both hypotheses 1 and 2 are not supported. Remarkable in this respect is that for French and Irish migrants at a 90 % confidence interval discrimination leads to more entrepreneurship. For French
become self-employed. The other countries however show the expected signs where the subgroup Other is also statistically significant at the 90 % confidence interval for both discrimination variables.
When looking at the degree of formal education, no statistical support was found for both ‘’College Degree’’ and ‘’Years of Education’’ the coefficients indicate that a higher college degree and more years of education do not lead to higher self-employment rates.
In the country extension model and the robustness checks no changes occurred for both the signs and the significance levels. Also do the signs not indicate that a higher formal educational level leads to lower self-employment rates. However, for Mexican migrants support is found and this finding is statistically significant at the 95% confidence interval and an Odds Ratio of 0.648.
The years of education, do equally not show a negative, but a positive coefficient for some of the countries under research. Which would indicate that the more years of education received by a respondent the less likely this respondent is to be self-employed. This is conforming the
hypothesis, however these findings are not statistically significant for most models. The results are therefore ambiguous, and
hypothesis 3 and 4 therefore are rejected. A possible explanation for the findings of the relationship between the degree of education and the length of education on entrepreneurship can be found in the complexity of entrepreneurship by perceived financial barriers, not knowing how to deal with the administrative complexities and the fear of business failure (Davidsson & Honig, 2003; Van der Zwan et al., 2013) This would mean that at least some education is necessary to overcome those barriers.
Lastly, the signs for the degree of Social Capital as an explaining factor for entrepreneurship were as expected and statistically significant at the 95% confidence interval. Both for the case where the mother is self-employed and the father is self-employed. This means that it is more likely for migrants, to be self-employed when either one, or both of their parents are equally self-employed. Parents who are self-employed can also help lower the barrier of perceived financial barriers (Dunn & Holtz-Eaking, 2000).
The results of the extended country model however show that this holds for all
his/herself is on average 1,5 times more likely. The results of the independent country subgroup variables do show some differences in the strength of the
relationship between a migrant being self-employed and one or both parents equally being self-employed. A remarkable and statistically significant finding here was that if the mother of a Mexican migrant is self-employed, this migrant him/herself is ten times more likely to become self-employed (Odds Ratio = 10.163).
Even though the signs are mostly positive and for some negative, no country was found to be significant in all models. This finding is therefore not robust and hence only partial. A possible explanatory factor is firstly the sample size and secondly an explanation can be found in research conducted by Fairlie & Meyer (2003) who indicate that the population as a whole can show a higher incentive to become self-employed because of feelings of increased competition on the labour market which is the case for the United States with
immigration rates exceeding migration rates (United Nations, 2013).
However, for migrants with an ethnic background in a relatively rich sending country in terms of Gross Domestic Product such as England or Germany a lower chance to become self-employed seems to exist which is however
insignificant at all levels and Mexican migrant seems be more likely to become self-employed with an odds ratio equal to approximately 1.5 for all models. This could be explained by the geographic proximity of Mexico to the United States. Because the proximity to the United States Mexicans may be more likely to spot business opportunities.
The other country predictors did not show statistically significant results. On the other hand the Chow statistic did show that difference exists in the propensity to become self-employed among migrants. Moreover, this finding was statistically significant for all nationalities included in the analysis.
A further explanation for these country differences can also be the effect on the ratio of the level of entrepreneurship in the home country of the immigrants to the overall self-employment rates in the United States (Yuengert, 1995). If the
entrepreneurship level of an immigrant sending country is high, it can be expected that its migrants are also more inclined to be entrepreneurial in their destination country.
robustness checks. Possibly because of the declining sample size the results changed. Apart from these changes, the model showed stable predictions for both coefficients and significance levels.
6 Limitations
This thesis has limitations that have to be considered when interpreting and using the results in practice. Firstly, the sample size for certain subgroups of ethnicities is small and can be increased for further research as this would increase the reliability of the research. This is particularly true for the country dummies, Asiatic people seem to be underrepresented in the sample and are therefore classified in the category Others. This can be a consequence of the
overrepresentation of legal migrants and not of illegal migrants. However, the probability that an illegal immigrant can start a legal business is unlikely and therefore this was a purposefully taken decision.
Secondly, the time series have fewer observations each year which is clouding results even though the dataset includes weights to counterbalance negative effects as a result of this type of bias. However, weighting essentially assumes that respondents included in the study are, given certain covariates, the same with respect to their subsequent behavior as
those who they replace from last years. This may be an unsound assumption (Fitzmaurice et al., 1996) even though there is already accounted for by the aforementioned added weighting factors in the sample by the researchers at NORC. The use of this sample was however a purposefully taken decision because of the accessibility of data in terms of costs and time restraints. Furthermore, the survey is held face-to-face and if necessary in different languages. This can lead to translation bias of the original
questionnaire. However, the interview is conducted by trained professional interviewers from the U.S. National Opinion Research Center (NORC) which minimizes this effect.
Considering the variables used in this research one could argue that a Graduate Degree received from an institution in Mexico is inferior to a degree received from a prestigious American University. This country difference in educational level is acknowledged but not taken into account in the analysis for statistical reasons and is therefore a limitation to this study.
