As discussed in chapter two the thesis focuses on how human capital and size affect the success of a newly started business. The previous literature focused thereby on survival of a company whereas this thesis focuses on the impact of human capital and size on later turnover. I men-tioned three different concepts of human capital and two different measurements for size in the previous section which lead to following hypotheses: The later turnover of a new venture is higher if the owners have former experience in line with their business. The later turnover of a new venture is higher the faster the percentage of owners holding an academic degree of more than 3 years education grows. vThe more the percentage of human capital in a company increases the higher is the later turnover. The higher the initial size in terms of employees is the higher is the later success meas-ured as the turnover in later years. The faster the company grows the higher is its later turnover. All of the hypotheses are expected to have a positive relationship between the independent and dependent variables.
Introducing the Data
The data of this thesis is data from ‘Statistics Sweden’ which holds a wide range of data concern-ing the Swedish Economy. The regression in this thesis is run on a population which is defined by the following features: the company is officially listed since 1997, it still exists in 2005 which encompasses also the span of time where all (growth) rates are based on and it is a limited com-pany. In Sweden limited companies are called ‘Aktiebolag’, in short ‘AB’. It is a population be-cause all companies in Sweden with these characteristics are included in this data set. The number of observations therefore is determined by the selection process enforced by the features just mentioned. The data the thesis is build upon is originally on firm and individual level. The regres-sion was run on the firm level whereby individual characteristics where computed to firm level variables and merged to the firm level data set. One should mention at this point that there is no available information about the legal stati of the firms before they were officially listed. This causes biases due to that the actual start of a compa-ny could have been several years earlier. To understand what the labels of the variables stand for the next table (table 1) explains shortly their meanings. More detailed explanations and indications for interpretation for each variable follows later on. In this thesis I will use three concepts of higher education in a company; i. e. work experience of the owners in line with their new venture, growth of the percentage of owners with more than three years of higher education and the change of the percentage of the overall higher education of the whole labor force in one company. One concept will reveal the impact of the percentage of owners that worked earlier in the same sector (‘PercSameSec’) on turnover (‘Turnover2005’) eight years after starting up the business following studies by for example Colombo et al. (2004), Bosma et al. (2004) or Cressy (1996). The second concept for human capital (‘OwnerMoreThan3’) will measure how the dependent variable is affected by the growth of the percentage of owners with higher education. The third concept (‘DiffMoreThan3’) will measure how the overall change of percentage of higher educated employees between 1997 and 2005 af-fects the turnover rate. Åstebro and Bernhardt (1999) for example controlled for higher educa-tion also by years of education. A dummy for higher education was introduced by Bosma et al. (2004). Boden and Nucci (2000) differentiated higher from lower education by if an owner went to college for more than four years or not. The common measurement for size is the number of employees. With that I follow for example studies by Colombo et al. (2004), Audretsch (1995), Mata and Portugal (1994) or Cefis and Marsili (2005). On the two-digit level of the SNI-code of an establishment a variable was generated to check whether an owner worked in the same sector before. In this context it is checked for five years earlier. The value of this ‘PercSameSec’-variable is the percentage of owners that worked in the same sector (five years earlier than the startup) as the company. This is the first concept of hu-man capital introduced in section 2.1.4. The selection of owners that were part of the actively working population in Sweden in 1992 might cause a bias in the results due to that there might be young academics that started a company right after graduation without having work experience in line with their business. The second concept as mentioned above is the variable ‘OwnerMoreThan3’ that represents the growth of percentage of owners that have a higher education of more than three years where lev-els that were from ‘Statistics Sweden’ reported as unspecified are counted as lower education. Additionally the third variable ‘DiffMoreThan3’ was computed where missing observations are counted as lower education. The variable is the difference of the percentage of employees with more than three years of higher education between the years 2005 and 1997. The initial size of a company is measured by the amount of employees in the year of foundation. The name of this variable is ‘InitialSize’. ‘GrowthSize’ is a variable which measures how much a company grew from 1997 to 2005 in terms of employees. The ‘PercSex’ variable stands for the percentage of female owners and it controls for the differ-ence of the performance of a company due to if the owner is male or female. To control for the potential difference that might occur due to that the owner was born outside of Sweden a variable ‘PercForeign’ was included. The value of this variable represents the per-centage of owners in their company that were born outside of Sweden. The ‘Productiveness’ variable was computed by the value added of a company divided by the number of employees and of that the growth rate again to control for different levels of produc-tivity between the two years 1997 and 2005. To control for the different market potentials in Sweden a proxy ‘RegDiffProxy’ was introduced. On the level of ‘AstKommunC’, which is a coded variable for the establishment where an indi-vidual is working, the income of the individuals are summed up and merged to the company lev-el. The income of individuals living in the municipality where the company is located in was summed up to control for the different chances for a company to develop in different locations. This variable was included to the model to avoid the implicit assumption of even chances for start-ups all over Sweden. The sector dummy (‘SectorDum’) was generated by ‘Stata’ on the level of ‘SNI’ (Swedish Stand-ard Industrial Classification) which is a coded variable for the sector a company is operating in. In order to keep the number of dummies low the initially five-digit code was decreased to only one digit. Very few observations missed data for its sector and were added to sector zero. Due to the little amount of missing values the bias is minimal and will not be noticeable. The sectors of the tenth dummy ‘Other Community, Social and Personal Service Activities & Activities of House-holds & Extra-Territorial Organizations and Bodies’ are the sectors all others are compared with.
The following table (table 2) lists all variables used for the regression with the main descriptive statistics to see first indications of non-normal behavior of the data. The dependent variable of this dataset varies a lot. The mean is relatively low compared to the maximum value which implies that the data is skewed in its distribution and that a ‘high’ turnover in 2005 is very rare. Secondly the standard deviation is more than double as high as the mean which implies that the observations differ a lot from each other. If the skewdness and spread of observations are too high the estimations get unstable which leads to insignificant or unreliable results. For the variable ‘PercSameSec’ similar is true. The standard deviation is nearly the same amount as the mean which implies a wide spread of the data. But for this variable the mean of 50.78 of 100 in total is reasonable. There are no conspicuous values for ‘OwnerMoreThan3’ or ‘DiffMoreThan3’. The initial size and the growth rate on the other hand are apparently very skewed which will make reliable estimations difficult. To emphasize the problem of multicollinearity the plots of the variables of interest are depicted which is partly a reason of the poor performance of the regression on the data at hand. In the section of the descriptive table one could already find first hints of weak distributions of the data which lead to biased results due to skewness of the data. The graphs show the scatter plots for each variable of interest against the dependent variable. It is very obvious that the data neither contains useful information for a general statement about the impact of the independent variables on the dependent variable nor does it contain useful infor-mation for forecasting which means there is no linear relationship between the variable of ‘PercSameSec’ and the turnover of 2005. To actually be able to get an indication of a relationship of ‘PercSameSec’ and ‘Turnover2005’ to some degree the data points of the different levels of ‘PercSameSec’ would need to group around different levels of ‘Turnover2005’. Here are the data points for all levels of ‘PercSameSec’ on a similar low level of the turnover. The same is true for ‘OwnerMoreThan3’ and ‘DiffMoreThan3’, only to some weaker extent. Interesting to see though, is that the turnover is highest in the companies with either 100% of the owners having experience in line with their business or none at all has worked five years earlier in the same business. For the variable ‘OwnerMoreThan3’ is no linear relationship apparent, either. One could say that there is somehow regularity in that sense that the overall distribution is more bundled than in the previous plot but there is definitely no linear relationship between how many owners of a new business have a higher education of more than three years and the turnover in 2005. To some ex-tent one can see a similar pattern as in the graph of ‘PercSameSec’ which might be due to the fact that both measures are a kind of human capital. Again there is no obvious relationship between ‘DiffMoreThan3’ and the turnover in 2005. One can detect by just looking at the plots that there are high correlations of the variables. In the table of correlation below there will be a statistical test on correlation to show formally that the varia-bles suffer from high correlation between variables. The negative numbers of ‘DiffMoreThan3’ indicate that by the companies by growing and stand-ardizing their processes they employ less academically educated but more non-academic workers. The plots for the two variables measuring the impact of size show similar patterns. There is a cloud of data in the lower left corner and some data points spread all over the scatter plot. The high Pearson correlation coefficient of the initial size and the growth rate of size (compare with section 5.5) are easy to retrieve. The high correlation of the two ‘size’-variables, as already noted, will lead to biased results. Due to that the data cloud is very localized and in the left lower corner and additionally some da-ta points that are very far away from the main data points the regression is very unstable and might suffer from the leverage effect. This means that the slope of the regression line changes easily if only one or few data points are excluded from the regression. Again in these plots there are no apparent linear relationships between the levels of initial size in year 1997 or the growth rate of size and the turnover 2005. To some extent the plots indicate that small initial size leads to small later turnover.
2 Human capital and size of newly started businesses Error! Bookmark not defined.
2.1 Human Capital
4.1 Introducing the Data
4.2 Descriptive Data
6.1 Summarized results
6.2 Interpretation of results
GET THE COMPLETE PROJECT
How Does Human Capital and Size Affect Later Turnover of New Ventures?