1
Bachelor Thesis
Is there an impact of month of birth on the success of male
Dutch soccer players?
Final version
15 – 07 – 2015
Cas Weijers
10003136
Faculty of Economics & Business
BSc. Economics & Business
Track: Economics & Finance
Supervisor: L. Geijtenbeek
2
1. Abstract
The relative age effect assumes a bias towards those born early in the selection period. The predetermined cutoff point within sports could therefore be of influence for the spread of monthly births. This thesis analyzes the relative age effect within the Dutch professional soccer league. First, previous research is summarized and their testing methods are described. Then the variables used for the test of this thesis are shown. The created database consists of all players within the first and second Dutch professional league. Then a weighted average month of birth is created in order to compare both distributions. The actual test conducted is the chi2-test, to see whether the distributions differ from each other. The results of comparing the distributions show significant, except for goalkeepers. Finally, graphs are created and all have downward slopes. This understates the relative age effect since there are more players in the first three months.
3
2. Introduction
Nowadays almost every child competes in a certain discipline of sport. Youth sports provide an attractive tool to improve your physical fitness and teach youth players to interact in team understanding. The increased focus on sports however, has a downturn. According to Kirkendall (2014) youth sports get increasingly organized. There is a starting shift from the original recreational character of sports to a more competitive character. In this way parents and children are increasingly focusing on achieving sportive goals. Each year new goals are set and every year the bar is set higher.
Talented children are getting selected for more advanced training and receive better coaching on increasingly younger ages with an ultimate goal of solely becoming the best. In order to achieve this, a greater focus is on the physique of those young athletes. Some argue that nowadays, young athletes are selected on their physical abilities and tactical awareness as opposed to their talent.
Players compete at different divisions and leagues. The young players however, are divided among these competitions according to their designated ages. This is done in order to maintain a certain standard of equal competition.
However, some argue that this way of creating a fair and equal competition by certain age groups is a relevant problem. Musch and Grondin (2001) compared the birth distribution of several different young professional athletes and revealed that there was a certain skewness in this distribution. This is called the relative age effect. The relative age effect assumes a certain bias towards those born early during the selection period. This bias is considered a problem since some argue that players will not be noticed, despite their greater potential opposed to their teammates. The actual relevance of this problem is considered by Bandura (1986). He argues that this effect could lead to a downfall in confidence for not selected players and a lower self-esteem. Moreover, this downfall in confidence could precede in further disciplines, for example school or potential careers.
The relative age effect is considered among several disciplines of sports but also on the field of schooling. A well-known example is young versus late students. According to Tanner (1978) the age difference of at most 12 months may seem as a small difference for adults, for the youth however, a difference of twelve months may result in significant physical differences.
This thesis will address the issue whether or not the relative age effect is indeed of interest with respect to the Dutch professional leagues. The following research question is formulated: ‘Is there an impact of month of birth on the success of male Dutch soccer players?’ In order to test
4 this, a database is created of all Dutch first league (Eredivisie) and second league (Eerste Divisie) players. Also a weighted average distribution of the month of birth is created, in order to compare both distributions and show any inequalities. Since this research has not been conducted for the Dutch sporting climate this could be considered relevant.
Within sports, children are grouped in accordance with their chronological age. According to Malina (1994) this is done in order to ensure a fair and equal competition among young athletes. Due to the assumption that ability development is age related, dividing children into their designated age group, these children will be spread fair among competition groups.
Musch and Grondin (2001) compared the birth dates of several young athletes within different sport disciplines and concluded that the distribution of birth data had biased skew towards athletes born early in the selection period. Malina (1994) stated that this bias could be explained due to the fact that during the mid-adolescence, early maturing athletes have a great physical advantage within their age group and therefore will be more likely to be over represented. According to Tanner (1978) the age difference of at most 12 months may seem as a small difference for adults, for youth players however, a difference of twelve months may result in significant physical differences.
On the other hand, Kirkendall (2014) stated that the main difference between the early maturity players and late maturity players in advanced teams is their endurance. Also physical maturation in the form of weight, height and muscle measurements influence the performance of players. Kirkendall (2014) believes that smaller players, when not selected for advanced teams, will fall behind their similar peers because they do not receive the advanced training standards and coaches. By getting early selected for more advanced teams, players will develop a comparative advantage opposed to peers. This will result in a higher degree of dropout for the sport and for Kirkendall (2014) this is not consistent with the goal of youth sports.
Vaeyens et. al (2005) tested the relative age effect for the Belgian second and third league in five consecutive seasons, as well as one year of the fourth national league was used. They created two studies since the Royal Belgian Football Association altered the cutoff date in 1997. First the cutoff date was August and after 1997 it was altered to January. They created the expected birth date distribution and tested their created database distribution against it. They conducted a Pearson chi2-test in order to see if the both distribution were aligned. Eventually they used a regression analyses to see in what extent the number of soccer players corresponds to the spread of the month- and quarter of birth.
The Pearson chi2-test of Vaeyens et. al (2005) resulted significant, so the distributions were not the same. Moreover, the regression analyses demonstrated a bias towards higher
5 amount of players born during the early part of the selection period. In order to overcome the bias of the relative age effect, they state that there should become more awareness of coaches about the relative age effect. In this way, Malina (2004) says that there will be more opportunities for late-maturing athletes and greater development.
Helsen et. al (2005) examined the relative age effect in correspondence to youth selections for the 1999-2000 season. They examined the birth date distribution of 10 youth teams respectively under-15, under-16, under-17 and under 18. Secondly they looked at the birth distribution of the youth teams playing at international tournaments, namely: 16, under-18, under-21 and womens under-18.
The first selection month was January and the twelfth selection month was December. They tested the distributions against the birth dates distributions of children in Belgium. First they used the Kolmogorov-Smirnov one sample test and eventually a regression analyses was used to examine a possible relationship between both distributions.
Both the Kolmogorov-Smirnov and the regression analyses proved significant in nearly all the teams. Implying that a high number of players is in the first quarter of the selection period and decreasing numbers in the last quarter.
However, as early research of Helsen et. al (2000) proved, a smaller effect was found for teams under-18 and under-21. There were no differences in the distribution of date of birth and the distribution of players for 16- to 18-year-old players. This is explained by the change of the cutoff date from August to November. Since these players have overlap during the alternation of the cutoff date, there are players who initially were in selection quarter 1, but under the new cutoff date, are in selection quarter 3.
For women however, Malina (1994) found a different explanation. Since women mature earlier than men, on average at the age of 18. The relative age differences are less significant for this group (women under-18). In addition, Giacomini (1999) showed the same conclusion for female tennis players. The reasoning for this conclusion is that females presumed to be selected more on their technique rather than their physical abilities.
Kirkendall (2014) examined the relative age effect within the North Carolina Youth Soccer Association. Kirkendall (2014) used data for teams from under-11 up to under-18 for both male and female. Different from previous research, the main focus was on the performance of teams with respect to the relative age effect. Certain points were allocated for a win or draw and the amount of scored goals in order to compare the variable ‘performance’. The relative age effect was tested using birth fractions according to actual calendar counts. For the actual testing the Pearson chi2-test was used on the observed distribution and the expected distribution in order to examine
6 the goodness-of-fit. Differences between birth quarters were examined by the use of a 95%-confidence interval. Interaction between the average team age and performance were tested by the use of simple correlation methods. The range for significance was set on p-value < 0,05.
Only the female group of under-15 did not show a significant Pearson chi2-test. However, for testing the performance level related to the relative age effect, only three of the 12 groups were significant. The results showed a lack of correlation between the average team birth date and the amount of points won and goals scored. Moreover, Anderson and Sally (2013) analyzed several variables for determining match outcomes and concluded that nearly 50% of the explanation of match outcomes come from random chance, therefore age differences account for only a very small fraction.
The third chapter explains the data which is used to conduct the test for the relative age effect. The fourth chapter describes the research methodology of the created data and explains the testing techniques which are used. The fifth chapter shows the results of the tests and gives the interpretations of these results. The last chapter gives the conclusion of the thesis and discusses the limitations of the research and provides opportunities for further research.
7
3. Dataset
3.1 Variables
The research is focused on the Dutch soccer environment. Success is defined as reaching a professional level of soccer. A professional level of soccer is defined as reaching the first team of the Dutch first league (Eredivsie) and second league (Eerste Divisie) with respect to the players of these teams for 2014-2015. The first and second Dutch national leagues consist of respectively 18 and 20 teams making a total of 38 teams.
3.2 Database of players
The database consists of all players selected for the first teams of the Dutch first and second league according to their official club sites. The number of goalkeepers within the database is 131, of defender 330, of midfielders 307 and of strikers is 281. Making a total of 1049 observations. To have a representative test, there should be at least 1000 observations.
These variables are chosen in order to test whether the relative age effect is of impact in the Dutch soccer environment. Since the parameter success is defined as reaching a professional level of soccer, the first and second Dutch league were chosen. Also extra features were added to make a distinction whether the impact of the relative age effect is greater in different positions or different leagues.
Furthermore, the players’ date of birth is added and are numbered from 1 to 31. According to the guidelines of the KNVB (Koninklijke Nederlandse Voetbalbond) the selection period starts at the 1st of January and ends at the 31st of December therefore months are accounted for as
January=1, February=2 up to December=12. The database includes years of birth from 1974 up to 1998. Figure 3.1 describes the spread of players within the database across the months of birth.
The dataset includes several features of players. The first feature is the league in which the players appear. Since the first national league is expected to have better players on average, this could indicate that the first Dutch league has more early selection period players. In other words, it relies more on the impact of the relative age effect. In order to test this there should be looked at the players’ leagues separately as well.
The second feature is position. This feature is included in order to test if there is a relationship between the relative age of a player and his position. Since every position has his own important aspects, this should be tested in accordance to each individual position. For example, goalkeepers have on average less physical contact, so this could indicate that the relative age effect for goalkeepers could be of less importance.
8
Figure 3.1: month of birth of observed players
Source: Player database
0 20 40 60 80 100 120 140 1 2 3 4 5 6 7 8 9 10 11 12
Nu
mb
er o
f ob
servat
ion
s
MonthNumber of observations per month
Next, date of birth is sorted quarterly; with quarter 1 including January, February, March. Quarter 2 includes of April, May and June. Quarter 3 includes of July, August and September. Quarter 4 includes October, November and December. Eventually players are also sorted in which league they play.
Figure 3.2: Overview database
Source: Database
League 1 League 2 Total
# of clubs 18 20 38 # of players 490 559 1049 # of goalkeepers 60 71 131 # of defenders 153 177 330 # of midfielders 140 167 307 # of attackers 137 144 281
9 Figure 3.3
Source: player database and CBS
month % observed % expected difference
January 12,68% 8,28% 4,40% February 10,10% 7,75% 2,35% March 11,06% 8,49% 2,57% April 10,49% 8,21% 2,28% May 8,48% 8,59% -0,11% June 7,34% 8,34% -1,00% July 6,20% 8,69% -2,49% August 8,96% 8,69% 0,27% September 7,34% 8,56% -1,22% October 5,43% 8,44% -3,01% November 6,77% 7,79% -1,02% December 5,15% 7,99% -2,84%
3.3 Overview of annual month of birth
The distribution of the year of birth of the player database runs from 1974 up to 1998. In order to test the player database an overview is created of the Dutch annual month of birth according to the CBS-Statline and the CBS archive. Here the same years are taken into account (1974 till 1998). Figure 3.3 shows the expected and observed data in percentages. The expected data is created in order to be able to test the observed data. The test conducted will be the Pearson chi2-test. The next chapter will describe this test and will explain how the expected data was calculated.
10
4. Methodology
4.1 Hypothesis
The construction of the hypothesis is based on the previous research conducted by Vaeyens et. al (2005). In their research they tested the relative age effect for the Belgian second, third and fourth league. Vaeyens et. al (2005) used the Pearson chi2-test to compare the distribution of the natural birth rate with the distribution of Belgian players. This thesis will conduct the same research methods, however the test will compare the distribution of Dutch players with the natural birth rate in the Netherlands. With respect to the previous formulated research question: ‘is there an impact of month of birth on the success of male Dutch soccer players?’ the following hypothesis is constructed:
H0: There is no relative age effect. The distribution of the month of birth is the same as the distribution of players in the Dutch professional football league.
H1: There is a relative age effect. The distribution of the month of birth is different from the distribution of players in the Dutch professional football league.
The rejection region of this test will be α = 5%.
4.2 Selection of comparing dataset
Based on primary research by Vaeyens et. al (2005), the distribution of the dataset will be compared with the actual birth date distribution. Since the birth rate differs each month and each year, this could explain a difference in the absolute amount of players for a given month in the dataset. Therefore a weighted average birth rate is created. The weighted average birth rate is calculated with the following formula:
𝑁𝑚𝑗 = 𝑎𝑎𝑛𝑡𝑎𝑙 𝑔𝑒𝑏𝑜𝑜𝑟𝑡𝑒𝑠 𝑁𝑒𝑑𝑒𝑟𝑙𝑎𝑛𝑑 𝑛𝑚𝑗 = 𝑎𝑎𝑛𝑡𝑎𝑙 𝑣𝑜𝑒𝑡𝑏𝑎𝑙𝑙𝑒𝑟𝑠 𝑝𝑒𝑟 𝑚𝑎𝑎𝑛𝑑 𝑛𝑚= ∑ 1998 𝑗=1974 𝑟𝑚= ∑ 𝑁𝑚𝑗 𝑁𝑗 1998 𝑗=1974 ∗ 𝑛𝑗
11 In this way the expected data distribution can be compared to the observed distribution. Figure 3.3 shows the output of the expected data. Adjustments are made for testing the extra features of the players since these tests will not contain the whole dataset but only the observations that contains these features. For example, testing the relative age effect for defenders will only contain players marked as defender so the number of observations for that test will be n=330.
4.3 Chi2-test of dataset
To conduct an empirical test on the hypothesis mentioned in section 4.4 the Pearson Chi2-test should be conducted. The Pearson chi2-Chi2-test is used to compare two distributions. It Chi2-tests whether the observed frequencies are aligned with the expected frequencies. The created dataset of players is considered as the observed data. The monthly birth rate is considered as the expected data. This test was chosen since it compares the distributions with each other, there might be evidence that these two distributions are not the same. From this we can conclude that the distribution of players in the Dutch professional leagues do not have the same spread as the birth rates.
4.4 Chi2-test of dataset, extra features
This test follows the same reasoning as in 4.5, the difference however is that this test will look at the quarterly distributions. To do so, the monthly distribution is put into the 4 quarters of the year. Quarter 1: January-March, Quarter 2: April-June and so on.
The same is done with the monthly distribution of births. Three months are accumulated and by this manner the year is narrowed down into 4 quarters instead of the previous 12 months. This is done in order to have a more general view to what extent the relative age effect has impact, since there is more grouped data. The results are expected to be more significant since the differences between the groups are bigger due to the accumulations of three months per quarter. Another test will look at the differences in league 1 and league 2. First the expected dataset is altered by the amount of players in each league. League 1 consists of n=490, league 2 consists of n=559. Secondly, the Pearson chi2-test is conducted on league 1 versus the new expected database, then the same test is conducted for league 2.
Moreover, both leagues will be subjected to the Pearson chi2-test to see whether they differ from one another in order to see if there is more impact in one of the two leagues. This is done because in general the first league consists of better quality players compared to the second league. Therefore the relative age effect is expected to show a bigger significant impact in the first league compared to the second league.
Furthermore a tests will be conducted to look at the differences between positions. Positions are divided into 4 categories. Position 1 consists of goalkeeper with observations of
12 n=131, position 2 consists of defenders with observations of n=330, position 3 consists of midfielders with observations of n=307 and position 4 consists of attacker with observations of n=281. The reasoning behind testing for each position is because each position on the field has its own features. For example it could be that goalkeeper are expected to have less physical encounters and therefore the physique of a keeper is of less importance compared to a defender. Therefore it could be that the relative age effect is of less significance for goalkeepers compared to other positions.
13
Figure 5.1: The expected frequencies versus the observed frequencies
Source: Dataset and CBS
0 20 40 60 80 100 120 140
Jan Feb Mar Apr Mai Jun Jul Aug Sep Oct Nov Dec Observed Expected Evenly
5. Results
5.1 Main outcomes
The first chi2-test reflects the distribution of all players’ month of birth against the expected distribution of month of birth. Figure 5.1 shows the observed frequencies and the expected frequencies. The horizontal axe shows each month, the vertical axe shows the frequency of observations in each month. According to the graph the observed frequencies tend to be higher than the expected frequencies at the beginning of the selection period and lower at the end of the selection period. The line evenly explains the total amount of players spread evenly across all months, this amount is 1049 observations divided by 12 months, so 87,412 per month.
Figure 5.2 shows the actual output of the Pearson chi2-test. As show in figure 5.2, the test has a Pearson chi2-value of 811,649 and a p-value of 0.000. This indicates that, with α=5%, that the test is significant. Therefore the H0 hypothesis will be rejected and the alternative H1 hypothesis will be accepted. This concludes that there is significant evidence that the distribution of the expected frequencies of players differ from the distribution of the observed frequencies of players.
14
Figure 5.2: The Pearson chi2-test
Source: Database and CBS
month Observed Expected Obs-Exp Pearson
January 133 86,874 46,126 4.949 February 106 81,294 24,706 2.740 March 116 89,08 26,92 2.852 April 10 86,114 -76,114 2.574 May 89 90,133 -1,133 -0.119 June 77 87,439 -10,439 -1.116 July 65 91,145 -26,145 -2.739 August 94 91,204 2,796 0.293 September 77 89,777 -12,777 -1.349 October 57 88,558 -31,558 -3.353 November 71 83,565 -12,565 -1.375 December 54 83,815 -29,815 -3.257 Chi2 811,649 p-value (0.000) Chi2 (quarterly 63,8756 p-value (0.000)
Figure 5.2 show the distribution of the frequencies of both expected and observed dataset. Since the Pearson chi2-test concluded significant, the distributions differ from each other. However, the assumptions of the relative age effect state that most observed data should be in the beginning of the selection period.
From the graph in figure 5.1 we can see that from January up to Mai, the observed frequencies are substantial above the expected frequencies. From Mai up to December, with an exception for August, the observed frequencies are below the expected frequencies. The peak of the observed data in August could be explained by a difference in previous selection periods, since the KNVB has altered the selection period in the past.
From this graph we can conclude, considering the significant Pearson chi2-test, that there is an impact caused by the relative age effect, due to the substantial greater frequencies of observed data in the beginning of the selection period.
The quarterly test shows that the Pearson chi2-test value is 81.1649 with again a p-value of 0.000. This indicates, with respect to α=5%, that the test is significant. Therefore the H0 hypothesis will be rejected and the alternative H1 hypothesis will be accepted. Since H0 is rejected and the alternative hypothesis H1 is accepted, it can be stated that there is significant evidence that the distribution of the observed quarterly frequencies differ from the distribution of the expected quarterly frequencies. As shown in figure 5.2; the value of the Pearson chi2 was
15 81.1649. However, in figure 5.2 the value of the Pearson chi2-test is 63.8756. This indicates that the test of the dataset sorted by month is more convincing then the dataset sorted into 4 quarters. However, both are significant.
This could be explained by the fact that the power of the second test is reduced. Since the months are accumulated, there are only four categories instead of the original twelve categories.
5.2 Pearson chi2-test observed month distribution versus expected month distribution, per league
This Pearson chi2-test reflects the observed monthly distribution against the expected monthly distribution with respect to both leagues. Since there are quality differences between both leagues, the relative age effect should be of bigger impact in the first league compared to the second league. Figure 5.3 describes lists all output of both leagues.
From figure 5.3 we can see that the value of the Pearson chi2-test for league 1: 33,5586 and for league 2: 54,0848 with a p-value of both 0.000. This indicates, with respect to α=5%, that the test is significant. This means that H0 is rejected and the alternative hypothesis H1 is accepted. Hence the distribution of the month of birth is not the same as the distribution of players in the both Dutch professional football league.
The value of the Pearson chi2-test with respect to comparing the distributions of both leagues shows a value of 20,6762 with an p-value of 0,037. With respect to α=5% we can conclude that both distribution differ from one another. However, the evidence is less compelling compared to the other two p-values.
16
Figure 5.3: the Pearson chi2-test, league 1
Source: Dataset and CBS
month League 1 League 2 League 1 vs 2
pearson chi2 output pearson chi2 output pearson chi2 output
January 2,586 4,358 -2,179 February 1,472 2,375 -1,562 March 1,910 2,119 -1,016 April 2,163 1,501 -0,267 May 0,179 0,004 -1,01 June 0,032 -1,558 0,833 July -1,920 -1,954 -0,845 August 0,367 0,057 -0,571 September -0,456 -1,420 0,162 October -2,544 -2,212 -1,237 November -1,923 -0,083 -2,563 December -1,465 -3,090 1,25 Chi2 33,5586 54,0848 20,6762 p-value 0,000 0,000 0,037
The observed frequencies, are above the expected frequencies in the beginning of the selection period up to Mai and fall under the expected frequencies in Mai until December, with an exception for August. This could be explained by the fact that the KNVB altered the beginning of the selection period in the past. This rise in August is less noticeable for league 2 players. This can be explained by the fact that this league contains relative younger players. Therefore it is possible that the shift of the cut-off date is less influence.
5.3 Pearson chi2-test observed month distribution versus expected month distribution, positions
This section will compare the observed distribution with the expected distribution, focusing on the positions. Since there are differences in the requirements of a good physique per position, this will be tested separately for each position. The expected data is adjusted to have the right distribution of the expected players according to the birth ratios according to the CBS.
17 The value of the Pearson chi2-test for goalkeepers is 14,4578 with a p-value of 0.209. With a α=5%, we can conclude that the test is not significant. In other words; there is not enough evidence to reject the hypothesis that the distribution of month of birth is the same as the distribution of goalkeepers. Therefore, H0 will not be rejected.
This could have two possible reasons. Since there are considerable less observations (n=131) of solely goalkeepers, this test has less power compared to the whole dataset.
Secondly, goalkeepers endure less physical contact during a match, therefore the impact of the relative age effect could be less noticeable when it comes to the physique of goalkeepers.
As shown above in figure 5.4, the value of the Pearson chi2-test for defenders is 31,7451 with a value of 0.001. For midfielders the value of the Pearson chi2-test is 28,5349 with a p-value of 0.003. For attackers we see that the Pearson chi2-test p-value is 32,6975 with corresponding p-value of 0.001.
From this, we can conclude that, with respect to α=5%, the test is significant. This results in the rejection of H0 and acceptance of H1. Said differently, there is considerable evidence to conclude that the distribution of the expected data differs from the observed data.
Figure 5.4: the Pearson chi2-test, per position
Source: Database and CBS
month Keepers Defenders Midfielders Attackers
Pearson output Pearson output Pearson output Pearson output
January 0,967 2,427 3,677 2,428 February -0,045 1,862 1,892 1,329 March 1,166 2,063 0,77 1,674 April 1,283 1,902 0,361 1,659 May -0,074 0,121 0,508 -0,843 June 1,239 -1,809 -0,914 -0,087 July -2,191 -1,058 0,065 -2,715 August -0,406 -0,32 -0,328 1,532 September -0,667 -0,046 -1,024 -1,029 October -1,821 -1,108 -1,949 -1,998 November 0,179 -1,418 -1,308 0,125 December 0,463 -2,602 -1,718 -1,996 Chi2 14,4578 31,7451 28,5349 32,6975 p-value (0.209) (0.001) (0.003) (0.000)
18 The observed frequencies are above the expected frequencies from January up to May. Then the observed frequencies fall below the expected frequencies but tend to rise from June up to September. This is partly explained by the change of the selection period in the past.
The difference in the chi2-test output for defenders, midfielders and attackers, is only a slight difference compared to the chi2-test of players from position 1. This can be explained by the fact that these positions are positions in which physique plays a bigger part compared to position 1. Since the position 1 players, goalkeepers, endure less direct physical contact and could therefore be less sensitive for the relative age effect.
However, the chi2-test output for midfielders is the lowest among the three significant positions. This is remarkable since midfielders are expected to rely the most of all positions on their physical abilities since they run the highest distance of all players and encounter relatively high physical duels.
5.4 Summary of results
As previous tests concluded, the observed distributions did not match the expected distributions in nearly all the tests, with exception of the distribution of goalkeepers. Figure 5.5 summarizes the results. Furthermore, looking at graphs of all the observed distributions, in comparison with the expected distributions, there was a high degree of observed data in the beginning of the selection period.
This bias of births towards the beginning of the selection period resemblance the assumptions of the relative age effect. The next chapter will describe the conclusion with respect to the results found in chapter 5.
Figure: 5.5: Summary results
Source: Database and CBS
Chi2-test P-value Dataset 811,649 (0.000) Dataset (quarterly) 63,8756 (0.000) league 1 33,5586 (0.000) league 2 54,0848 (0.000) keepers 14,4578 (0.209) defenders 31,7451 (0.001) midfielders 28,5349 (0.003) attackers 32,6975 (0.001)
19
6. Conclusion
6.1 Summary
Chapter 5 discussed the results of the expected and the observed data. Most of the output displayed a significant Pearson chi2-test, with exception of the goalkeepers.
First the dataset as a whole was tested monthly and quarterly. Both sets were considered significant. Surprisingly however, the value of the Pearson chi2-test of the quarterly sorted data was ‘less’ significant then the monthly data. This was explained by the fact that the ‘strength’ of the quarterly test was diminished by grouping the data in even less categories.
When looking at the graph of the whole dataset, there was clear evidence of the assumed biased distributions. Considering the assumption of the relative age effect; a bias of births in the beginning of the observed data with respect to differences in the observed and expected distributions, there can be concluded that there is a certain degree of bias from the relative age effect within the Dutch professional football leagues.
6.2 summary extra features
The Pearson chi2-test resulted significant for both leagues. However, both leagues did not have the same distribution when tested. Both leagues, however, did not have the same distribution. Concluding, it is plausible argued that there is an impact of the relative age effect within both leagues.
With exception of position 1: goalkeepers, all the Pearson chi2-tests displayed differences in the expected and the observed distributions. This is explained by the fact that the observations were relatively low to give a good view and by the fact that goalkeepers encounter less physical duels.
There can be concluded that there was a certain bias towards the first 3 selection months therefore concluding an impact of the relative age effect
6.4 Conclusion research question
With success defined as reaching a professional level, the originally formed research question: ‘Is there an impact of month of birth on the success of male Dutch soccer players?’, can be answered.
Since the relative age effect is related to the month of birth and tested significant, there is enough evidence provided to conclude, that the month of birth does indeed have impact on the success of male Dutch soccer players. Despite the fact that goalkeepers did not prove significant,
20 the overall dataset proved significant. Therefore there can be stated that the relative age effect has an impact in the Dutch professional soccer environment.
However, since the test group consist of all players in the first and second professional Dutch league, the test group does not only consists of Dutch players. Nearly every team has at least one foreign player. Since the cutoff date is not universal for all countries, players should be listed according to their own countries cutoff date, in order to have a more relevant view of the relative age effect.
6.5 Suggestions for further research
Firstly, since the cutoff date of the selection period has been altered over the years, further research could make a database dividing players into 2 groups. The first group consists of players before the altered cutoff date, the second group should consist of players after the altered cutoff date.
Secondly, since the relative age effect comes from the differences in maturity, further research should be conducted at younger ages. Especially the comparison of the age-classifications of the so called Dutch ‘E-level’ and ‘D-level’. When playing at the E-level, children play at half a field. When they get older and have to go to the D-level, they first encounter a whole field. In this transition the relative age effect could be more noticeable since playing on the bigger field relies more on the physical abilities of players.
21
BIBLIOGRAPHY
Anderson C, Sally D. (2013). The numbers game: why everything you know about soccer is wrong. New York: Penguin Books
Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory. Englewood Cliffs, NJ: Prentice Hall.
Giacomini, C.P. (1999). Association of birthdate with success of nationally ranked junior tennis players in the United States. Perceptual and Motor Skills, 89, 381–386.
Helsen, Werner F ; Van Winckel, Jan ; Williams, A Mark (2005). The relative age effect in youth soccer across Europe. Journal of Sports Sciences, 23, 629-636
Helsen, W. F., Starkes, J. L., & Van Winckel, J. (2000). Effect of change in selection year on success in male soccer players. American Journal of Human Biology, 12, 729–735.
Helsen WF, Baker J, Michiels S, Schorer J, van Winckel J, Williams AM. (2012). The relative age effect in European soccer: did ten years of research make any difference? J Sports Sci, 30, 1665-71
Kirkendall, Donald T.(2014) The relative age effect has no influence on match outcome in youth soccer. Journal of Sport and Health Science, 4, 273-278
Malina, R. M. (1994). Physical growth and biological maturation of young athletes. Exercise and Sports Science Reviews, 22, 389– 433.
Malina, R. M., Bouchard, C., & Bar-Or, O. (2004). Growth, maturation, and physical activity. Champaign, IL: Human Kinetics.
Musch, J., & Grondin, S. (2001). Unequal competition as an impediment to personal development: A review of the relative age effect in sport. Developmental Review, 21, 147–167
Quaquebeke N van, Giessner SR. (2013). How embodied cognitions affect judgement: height-related attribution bias in football foul calls. J Sports Exerc Psychol, 32, 3-22
Tanner, J. M., & Whitehouse, R. H. (1976). Clinical longitudinal standards for height, weight, height velocity and weight velocity and the stages of puberty. Archives of Disease in Childhood, 51, 170–178.
Vaeyens, Roel ; Philippaerts, Renaat M ; Malina, Robert M. (2005) The relative age effect in soccer: A match-related perspective, Journal of Sports Sciences, 7, 747-756
