1
Who becomes a CEO?
Mark Vermeulen, 11766611
University of Amsterdam
April 23, 2020
Abstract
This paper uses a regression discontinuity design to measure the increase in CEO potential due to relative-age in school cohorts. It uses a sample of 482 Dutch executives to show that those who are relatively old in school also have a higher chance of becoming a CEO. The cut-off date to start school is the first of October and creates a sharp discontinuity in who is the oldest in the cohort. There is a sharp discontinuity in the probability of becoming a CEO at the cut-off, however this discontinuity is not unique, which discredits the long-run relative age effects hypothesis. Furthermore, there is no evidence of gender differences.
2
Statement of Originality
This document is written by Student Mark Vermeulen who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
1
Introduction
This research focusses on whether the development of leadership traits at an early age increases the chances for an individual to become an executive in its later career. As it is widely argued that being the oldest in a school cohort could bring substantial benefits to an individual in the short-run, some researchers argue that there is little evidence of these benefits sustaining in the long-run (Black, Devereux and Salvanes (2011), and Bedard and Dheuy (2006)). However, other researchers state that the relative-age effect has a long-lasting influence on the successes on an individual’s career (Du, Gao and Levi (2012), and Page, Sarkar and Silva-Goncalves (2019)). In addition, Oosterbeek, ter Meulen and van der Klaauw (2020) argue that due to a flexible education system, such as the Dutch, the relative-age effects fade out.
This paper aims to add insights to the ongoing discussions about whether or not the relative-age effects in a school cohort are significant in the long run or not.
3
The data for this paper is hand-scraped from the internet. It contains the birth dates of a sample of executives in a professional setting. These birth dates are used, such that their (potential) relative age in months can be determined and as such researched.
This knowledge could be beneficial for parents. They could postpone the enrolment of their child by one year, if this child would initially be one of the youngest in their school cohort, to increase the child’s potential career successes. The research could also provide some insights on the discussion about whether leadership traits are typically born with, or can be learned. According to Page, Sarkar, and Silva-Goncalves (2019), the oldest children in a school cohort typically tend to have more self-esteem and higher academic achievements, which makes them develop leadership traits at an early age. They see this as an explanation of why individuals who were the oldest in their school cohort are over-represented in occupations such as being a CEO or politician. Secondly, in a high school setting, the relatively older students tend to be 4-11 percent more likely to be high school leaders (Dheuy and Lipscomb (2008)).
The sample consists of Dutch executives in a professional setting, who at least followed their primary education in the Netherlands. The data is gathered using convenience sampling. This is done by searching for executives on the internet and reporting their date of birth, function, industry, and gender. The downside of this sampling procedure is a higher probability for a potential bias in the taken sample. Furthermore, with this sampling procedure, a sampling error is more likely to occur. Such a sampling error combined with the potential bias decreases the external validity of this research.
4
In order to determine the relative-age in a school cohort, the Dutch school starting age requirement is applied to the gathered data. In the Netherlands, the school-age requirement demands that you are six on the first of October in order to start school in that year. The exogenous variation in relative cohort age, created by this age requirement, is studied using a regression discontinuity design.
The regression discontinuity design estimates shows some evidence for long-run relative age effects among executives. There is a 14.6% increase found in represented executives in the month October, compared to the month September. However, the estimated regression also shows April, May, June and July to be more represented birth months among executives. There are no differences found for long-run relative age effects between the two genders.
This study deviates mostly from previous literature in its sampling criteria and the empirical strategy. Whereas most studies, which study career effects of relative-age in a school cohort, analyze data that does not specifically concern executives (Bedard and Dhuey (2006), Black Devereux and Salvanes (2011), Page, Sarkar and Silva-Goncalves (2019), and Frederikson and Öckert (2014)), whereas this study does. The research of Du, Gao, and Levi (2008) has a sample of only CEOs of S&P500 companies and studied their relative-age effects. However, the study of Du, Gao, and Levi (2008) only compares the months June and July to all other ten months across different countries and their educational tracks. This study will provide insights of relative-age effects in a school cohort on career successon all twelve months, within only the Netherlands and its educational tracks.
In the following section, the sampling and the gathered data will be addressed. The third section will provide an extensive overview of the Dutch
5
educational system. Following the section about the Dutch education system, the empirical strategy of this research will be discussed. In the fifth section the results are presented and in the sixth section these results will be interpreted. These interpretations will provide an answer to the research question on whether relatively old children in a school cohort have a higher probability on becoming an executive in its future career. Lastly, the results of this research will be compared to other findings, the flaws of this research will be addressed and suggestions for future research will be made.
2
Data
The gathering of data is done by means of convenience sampling. A sample of 482 executives is gathered by hand from the internet. The dataset records the date of birth, the function, the industry, and the gender of the executives. Each data-entry is also checked on the birthplace to exclude any individuals who might have followed (part of) their primary education in a different country from the Netherlands, which has other age requirements. In this dataset, 71.8% is male. The years of birth range from 1939-1987 (M=1962.56, SD=8.54), 46.3% of the executives in the data are born in the range of October-March, who are considered to have been the relatively old children in their school cohort. Table 1 shows the number of births in the dataset for each month and Figure 1 shows the density of each birth month in the dataset. The largest part of the data entails mayors, politicians and board chairmen, with 302, 69 and 63 entries, respectively. The most common industry in the data are governmental, finance and employment , with 384, 29 and 9 entries, respectively.
6
The hand-gathered data is matched with the number of births per 1,000 citizens of the Netherlands, for each month. This number of births per month is taken from Statistics Netherlands. Statistics Netherlands provides the data of the number of births per month from January 1995-Febuary 2020. Because this range of available data is outside of the birth range of the gathered data, the mean of births per 1,000 citizens of the Netherlands for the available data will be matched to the gathered data for each month. The mean birth numbers were determined by taking the population size and the number of births for every month of 1995-2019. To check for robustness, the same data was also gathered for every month of 1995-1999. Table 1 shows the calculated mean birth numbers for each month.
Table 1: Numbers on births per month, centered around October.
Month Na M 95-99b M95-19c April 51 1.057 0.918 May 40 1.042 0.957 June 43 1.035 0.947 July 45 1.100 1.013 August 41 1.086 1.009 September 39 1.086 0.995 October 37 1.059 0.973 November 46 0.990 0.913 December 23 1.000 0.913 January 43 1.036 0.952 February 36 0.965 0.877 March 38 1.037 0.945
Notes. aN = number of individuals in the hand-gathered dataset born in each month. bM
95-99 = mean of births per month per 1,000 citizens calculated over the year of 1995-1999. cM
95-19 = mean of births per month per 1,000 citizens calculated over the year of
7
8
3
Institutional setting
Figure 2: Dutch education system.
Notes. Figure retrieved form Oosterbeek, ter Meulen and van der Klaauw (2020). The
blocks in this figure each represent a stage of the Dutch education system, with the arrows as possible transitions. The age bar on the right represents the age at which students go through with each stage of education.
Figure 2 shows a concise overview of the Dutch education system. Children can enter kindergarten at the age of four, where they spent their first two years of the educational track. When children turn six, they are put into cohorts to start their primary education. After six years, they take an exit test to determine their level of secondary education. When students complete their secondary education, they can access the connecting higher education tracks. The rest of this section will describe these different educational stages and transition moments in more detail.
The age requirement of starting primary education is that a child should be six on the first of October. When two identical children are born on September 30th
9
and October 1st, only one of them is allowed to start its primary education on the first of October in the year they turn six, the other will have to wait for another year and stay in kindergarten.
This age requirement was heavily enforced from 1937-2008, whereas after this time period the policy became less clear. The first entrance rule was institutionalized in 1920 (Koninklijk besluit nr 37, 1920, Art. 1). The rule was harshly enforced by the minister, prosecuting non-complying schools (Inspectie Onderwijs, 1936). The structure undersaw some changes in 1985, when the kindergarten and primary education merged to elementary education. This merger did not come with a specific cut-off date to start education, this was unnoticed for at least 8 years (van Dam, 1994). After 2008, some schools deviated from the first of October as the cut-off date, to the first of January. Nonetheless, Smeets and Resing (2013) argue that most schools still used October 1st as the age requirement in 2013.
After six years of primary school, students take a nationwide exit test before they enter secondary education. The results of this test range from 500-550. Based on the result of a student, the student gets an advice from the testing bureau to proceed to one of the three ability tracks of secondary education. However, besides the advice of the testing bureau, the teacher can provide a different recommendation, based on the exit test and a subjective estimate of a student’s probability of completing an ability track. In the end, the school offering secondary education determines which students get enrolled. The correlation between exit test and teacher recommendation is 0.85 (Oosterbeek, ter Meulen and van der Klaauw (2020)).
10
There are three ability tracks in the Netherlands, vocational (vmbo), college (havo) and academic (vwo). Ability tracks consist out of students with the same ability. During the ability track, a student chooses its curriculum varying from academic courses to vocational courses. The academic track consist solely on academic courses and the vocational track is solely made up of practical vocational courses. The college track has a mix of theoretical and practical courses.
After students finished their secondary education, they can enroll in higher education. Every ability track gives access to a corresponding higher education degree, including the vocational track. Education is compulsory until a student reaches the age of 18. However there are incentives to finish higher education until the age of 27.
The Dutch education system allows for flexibility. As in some secondary schools the educational tracking allows for a one or two years delay. Furthermore, when students excel in their current ability track, they can often move to a higher ability track, with the same vice versa. Students can also move to a higher ability track by graduating in its current ability track, instead of moving directly to higher education. Lastly, the Dutch education system also allows for some switching in higher education. As a student could enroll in university after finishing its first year of college.
4
Empirical strategy
In order to estimate the effect of an individual’s birth month on its chances on becoming an executive in its future career, the following linear regression is estimated:
11
where 𝑦𝑖 is the outcome variable for month 𝑖. 𝑦𝑖 is computed by dividing the number of executives in the sample by the mean of births per 1,000 citizens of the Netherlands for each month. This mean of births is computed over the years 1995-2019. The mean of births is also computed over 1995-1999 to check for robustness of the gathered mean birth numbers. The robustness-check will be operationalized by a regression (2), which is similar to regression (1), but 𝑦𝑖 will be replaced with 𝑤𝑖, which is computed with the mean of births over 1995-1999. 𝐴𝑓𝑡𝑒𝑟𝑖𝑗 is a dummy
variable which has the value 1 if an individual 𝑗 is born in the range October-March and has a value of 0 if an individual is born in the range April-September. This is the variable of interest, because it measures the jump of the probability of becoming an executive in an individual’s future career, when this individual is a member of the relatively older group of a school cohort. 𝑀𝑜𝑛𝑡ℎ𝑗 is a continuous variable that assigns a value of -6 if an individual 𝑗 is born in April to a value of 5 for the month March, centering around the first of October. Lastly, 𝛼𝑖 is a constant which shows the starting value of the estimated regression.
However, next to relative-age, gender also plays a significant role in assigning a leader (Alexander and Anderson (1993), Sczesny et al. (2004), and Ayman and Korabik (2010)). To check on whether long term relative-age effects are different for women compared to men, the following regression is estimated: (3) 𝑦𝑖 = 𝛼𝑖 + 𝛽1𝐴𝑓𝑡𝑒𝑟𝑖𝑗+ 𝛽2𝑀𝑜𝑛𝑡ℎ𝑗+ 𝛽3𝐺𝑒𝑛𝑑𝑒𝑟𝑗+ 𝛽4𝑍𝑖𝑗+ 𝜀𝑖
where 𝐺𝑒𝑛𝑑𝑒𝑟𝑗 is 0 if an individual 𝑗 is male and 1 if the individual 𝑗 is female. 𝑍𝑖𝑗 is the interaction variable 𝐴𝑓𝑡𝑒𝑟𝑖𝑗 ∗ 𝐺𝑒𝑛𝑑𝑒𝑟𝑗. If 𝛽4 is different from 0, this will indicate a stronger or weaker relative-age effect on becoming an executive for women. The other variables are unchanged.
12
Lastly, to check the robustness of the estimated model as a whole, regression (1) will be repeated with a dummy variable for each month as a cut-off date. This will result in 12 regressions, where in the first regression the dummy variable 𝐴𝑓𝑡𝑒𝑟𝑖𝑗 will have a value 1 for the months January-June. The second regression
will be estimated with 𝐴𝑓𝑡𝑒𝑟𝑖𝑗 measuring 1 for the months February-July and so on. The estimated 𝛽1’s for all 11 newly estimated regressions will then be compared to regression (1), to check for the uniqueness of this estimated regression. Afterwards, a robustness analysis will be performed to test the estimated model for its linearity. This is done by means of a multiple polynomial test for the variable 𝑀𝑜𝑛𝑡ℎ𝑗, with regression (4), where 𝛽3𝑀𝑜𝑛𝑡ℎ𝑗2 will be added to regression (1), and
regression (5), where 𝛽4𝑀𝑜𝑛𝑡ℎ𝑗3 will be added to regression (4).
5
Results
Table 2: Estimated β’s for regressions (1), (2) and (3).
Variable (1) (2) (3) Afterij 5.760*** (1.065) 5.084*** (0.968) 5.488*** (1.125) Monthj -1.518*** (0.136) -1.356*** (0.119) -1.484*** (0.137) Genderj 1.374*** (0.524) Zij 0.374 (1.175) Constant 39.508*** (0.404) 36.317*** (0.361) 39.226*** (0.415)
Notes. N = 482. Dependent variable in (1) & (3) = 𝑦𝑖. Dependent variable in (2) = 𝑤𝑖. *p
13
Table 3: Robustness-check outcomes.
Cut-off month Constant Afterij Monthj
January 40.446*** (1.073) 5,450*** (1.548) -0.181 (0.180) February 42.685*** (0.819) 1,554** (1.232) 0.408*** (0.118) March 39.930*** (0.673) 6,194*** (1.135) -0.390*** (0.126) April 36.162*** (0.499) 12,451*** (0.890) -1.518*** (0.136) May 45.627*** (0.778) -4,308*** (1.144) 0.170 (0.186) June 47.885*** (0.720) -7,965*** (1.144) 1.367*** (0.193) July 44.813*** (0.728) -3,283*** (0.799) -0.181 (0.180) August 46.689*** (0.428) -6,453*** (0.409) 0.408*** (0.118) September 43.785*** (0.340) -1,516** (0.728) -0.390*** (0.126) October 39.508*** (0.404) 5,760*** (1.065) -1.518*** (0.136) November 42.338*** (0.701) 2,269* (1.366) 0.170 (0.186) December 48.123*** (0.756) -8,440*** (1.429) 1.367*** (0.193)
Notes. N = 482. Every row shows the estimated regression (1) with a different cut-off
month. In the estimated regressions, the indicated cut-off month and the five following months get a value of 1 for the dummy-variable 𝐴𝑓𝑡𝑒𝑟𝑖𝑗, the other months six month get a value of 0. *p < 0.10, **p < 0.05, ***p < 0.01. The standard errors are indicated within the brackets.
Table 4: Multiple polynomial test for linearity.
Variable (4) (5) Afterij -7,546*** (1.169) -17,246*** (1.029) Monthj 0,020 (0.101) 3,688*** (0.131) Monthj2 0,489*** (0.0511) 0,606*** (0.041) Monthj3 -0,142*** (0.004) Constant 40,722*** (0.361) 43,262*** (0.287)
Notes. N = 482. *p < 0.10, **p < 0.05, ***p < 0.01. The standard errors are indicated within
14
Figure 3: Graphical non-linear representation of regression (1)
Notes. The probability of birth month is defined as the number births per month in the
gathered dataset divided by the number of births per 1,000 citizens of the Netherlands over the years 1995-2019.
Table 2 shows the estimated 𝛽’s for regression (1). The constant 𝛼𝑖 for regression (1) shows an estimate of 39.508 executives for each child born per 1,000 heads of the Dutch population for the month October. The estimated 𝛽2 for the variable 𝑀𝑜𝑛𝑡ℎ𝑗, shows that for every month following April there is an indication for a downward trend of 1.518 in represented executives for each month. However, with the estimated 𝛽1 for the variable 𝐴𝑓𝑡𝑒𝑟𝑖𝑗 being 5.760, there is an indication of long-run relative-age effects. As this estimated 𝛽1 represents a 14.6% increase in represented executives for the month October, compared to September. However, the estimated regression shows 23.0% more represented executives for the month April, compared to October.
15
The results of regression (2) in Table 2 show somewhat similar results as regression (1). Regression (2) is estimated the same as regression (1), but with 𝑤𝑖 as the dependent variable instead of 𝑦𝑖 to check for the robustness of 𝑦𝑖. 𝑤𝑖 is computed in the same way 𝑦𝑖 is, except 𝑤𝑖 is computed over the years 1995-1999, whereas the interval is 1995-2019 for 𝑦𝑖. The trends in regression (2) are mostly similar to those of regression (1), except for the estimators being somewhat smaller. This could be explained by the fact that the mean number of births for 1995-1999 are larger for any month, compared to those of 1995-2019, as indicated in Table 1.
Furthermore, regression (3) shows no significant difference in relative-age effects between the two genders. As the estimated value of 𝛽4 for variable 𝑍𝑖𝑗 is small and is not significant, with a p-value > 0.10, this cannot be considered as evidence for gender difference in relative-age effects.
Lastly, Table 3 shows the results for the robustness-check of the variable 𝐴𝑓𝑡𝑒𝑟𝑖𝑗 and Table 4 shows the robustness analysis on the linearity of the estimated
model. When comparing regression (1), with October as the cut-off month, to the other estimated regressions with a different cut-off month in Table 3, there is no unique value for the variable 𝐴𝑓𝑡𝑒𝑟𝑖𝑗 in regression (1). This lack of uniqueness for the month October as the cut-off month, means the results of the estimated regressions should be interpreted with caution, as it provides no strong evidence for the hypothesis to be true. The assumption of linearity of the estimated model should also be taken with caution, as the estimated 𝛽’s for the variables 𝑀𝑜𝑛𝑡ℎ𝑗2 and 𝑀𝑜𝑛𝑡ℎ𝑗3 in regression (4) and (5), shown in Table 4, are significantly different from 0.
16
6
Conclusion
In this research, some evidence was found on long-run relative-age effects among executives born around the cut-off date. However, when taking all 12 months into account, no evidence is found of executivesbeing underrepresented when they are allegedly being considered to be part of the youngest half of a school cohort. On the contrary, this research found the executives born in April to be relatively most represented. These results go against the findings of Du, Gao and Levi (2012), and Page, Sarkar and Silva-Goncalves (2019).
An argument that could explain these contradictory findings, is the lack of accuracy of the data used in this sample. As the data in this sample was not checked on whether each individual started school on the minimal required starting age, but rather a year later. Also some of the individuals might have had to redo a grade during its educational career, which automatically would make them one of the oldest in a school cohort in their following years of education, when they originally were not. These variables could not be included in the research due to the unavailability of these measures.
Another reason for the differences in results, could be the flexibility of the Dutch educational track. As many recent studies from the tracking literature emphasize the benefit of providing a second chance to students in early years of education (Oosterbeek, ter Meulen and van der Klaauw (2020), Buchholz and Schier (2015), Biewen and Tapalaga (2017), Hillmert and Jacob (2010), and Dustmann, Puhani, and Schönberg (2017)), which the Dutch education system provides, might be the cause of no strong evidence on long-run relative age effects among executives in the Netherlands.
17
References:
Alexander, D., & Andersen, K. (1993). Gender as a Factor in the Attribution of
Leadership Traits. Political Research Quarterly, 46(3), 527-545.
Ayman, R., & Korabik, K. (2010). Leadership: Why gender and culture matter.
American psychologist, 65(3), 157.
Bedard, K., & Dhuey, E. (2006). The persistence of early childhood maturity:
International evidence of long-run age effects. The Quarterly Journal of Economics, 121(4), 1437-1472.
Biewen, M., & Tapalaga, M. (2017). Life-cycle educational choices in a system with early tracking and ‘second chance’ options. Economics of Education Review, 56 , 80–94.
Black, S. E., Devereux, P. J., & Salvanes, K. G. (2011). Too young to leave the
nest? The effects of school starting age. The Review of Economics and
Statistics, 93(2), 455-467.
Buchholz, S., & Schier, A. (2015). New game, new chance? social inequalities and
upgrading secondary school qualifications in west germany. European
Sociological Review, 31 (5), 603–615.
Dhuey, E., & Lipscomb, S. (2008). What makes a leader? Relative age and high
school leadership. Economics of education Review, 27(2), 173-183.
18
Evidence from corporate CEOs. Economics Letters, 117(3), 660-662.
Dustmann, C., Puhani, P. A., & Schönberg, U. (2017). The long-term effects of
early track choice. The Economic Journal, 127 (603), 1348–1380.
Fredriksson, P., & Öckert, B. (2014). Life‐cycle effects of age at school start. The
Economic Journal, 124(579), 977-1004.
Hillmert, S., & Jacob, M. (2010). Selections and social selectivity on the academic
track: A life-course analysis of educational attainment in germany. Research
in Social Stratification and Mobility, 28 (1), 59–76.
Oosterbeek, H., ter Meulen, S., & van der Klaauw, B. (2020). Long-term effects of
relative age in school.
Page, L., Sarkar, D., & Silva-Goncalves, J. (2019). Long-lasting effects of relative
age at school. Journal of Economic Behavior & Organization, 168, 166-195.
Sczesny, S., Bosak, J., Neff, D., & Schyns, B. (2004). Gender stereotypes and the
attribution of leadership traits: A cross-cultural comparison. Sex roles,
51(11-12), 631-645.
19
