The Dark Age of Football

Following my post on the all-time best football players aged 21, a discussion started of why there are so few players between 1987 and 2007 on this list. The time period is nearly 50% of all the total period covered, but only two, Roque Santa Cruz and Ronaldo, out of 25 players fall into those two decades. This can't be incidental, yet it is unclear what causes this 'Dark Age'.

It could be purely artificial. For example, there could be a data selection bias by including or excluding different leagues in the different time periods. Indeed I have much more leagues covered in the latest years starting from 2007 than I had in the 1980s. However, there are no discontinued national leagues. Every country's top national league that was once covered is covered until today. So there are only additions of leagues and no removals. Any addition of leagues, however, will broaden the base of players to select from and will increase the likelihood of a great player in the sample. Adding leagues should therefore lead to an increasing number of players in the list instead of a decreasing.

Another artificial source of the Dark Age could be a quirk in the Goalimpact algorithm. It is difficult to argue for or agains that. Unless the cause of a potential quirk identified, it is just hypothesizing.

A potential alternative explanation was given by an anonymous reader in a comment.

Ich schätze damals war man halt im vergleich zu heute mit 21 noch nicht zu alt ;) Denke es liegt daran, dass man in dieser Zeit einfach erst in einem höheren Alter seine Einsätze bekam...

I figure, back then 21yo players weren't considered as too old ;) I think it is caused by the fact that players got playing time a higher age those times.

Following that line of thought, young players were less good between 1987 and 2007 because they couldn't play at matches top level and hence developed slower. I like that hypothesis because it is testable. So here is the chart.


Blue line shows the average Goalimpact of all players that became 21 that year. It was stable between 95 and 96 from the sixties until the mid eighties. Then there is a sudden drop to 94 at the end of the eighties. It drops even further end of the nineties until it bottoms out in the mid 2000s and it has been on a rise since then.

As comparison, the blue line shows the average minutes players have played already in the years before they turned 21. Here we see a drop that coincides with the drop in Goalimpact at the end of eighties. When it used to be just below 1500 minutes, it dropped blow 1000. It is a drop of nearly 40% average playing time. It never picked up again until 2007, but this increase is, at least partly, because of the addition of various youth leagues to the database.

So the anonymous reader may have a point. At least the drop in Goalimpact at the end of the eighties might have been caused by the football clubs. They didn't give enough playing time to young players anymore. As a consequence the players may have developed at a higher age than in the previous decades. The big teams rather bought 'complete' players from selling countries than to educate their own youth.

However, the drop of average GI in 2000s is clearly not explained by this. Here I assume the addition leagues to be the main cause. The leagues added are mainly second tear national leagues and the average player playing there will have a lower Goalimpact than in the Big5 leagues that have previously dominated the sample.

The recent increase of the average GI, in contrast, can't be explained by the addition of new leagues. So, if indeed the effect is real, and if players did develop slower between 1987 and 2007 than before, we have an indications that this 'Dark Age of Football' is over by now.

Appendix: The data for the chart

Year Number Players Avg Goalimpact Avg #Games Avg #Minutes
1966 71 95,4 12,8 1189
1967 65 95,8 12,4 1148
1968 89 96,0 12,0 1072
1969 97 96,5 14,8 1282
1970 88 94,7 18,2 1476
1971 88 96,7 16,7 1317
1972 89 96,4 15,1 1161
1973 86 98,2 26,5 2101
1974 97 95,8 20,7 1581
1975 145 95,5 16,0 1159
1976 177 94,8 21,2 1632
1977 145 95,8 22,5 1668
1978 175 96,4 20,3 1531
1979 156 95,9 21,4 1604
1980 181 94,8 20,0 1441
1981 200 95,1 22,2 1657
1982 175 95,6 19,8 1453
1983 194 95,3 18,6 1360
1984 158 93,6 18,7 1353
1985 199 96,0 18,2 1344
1986 172 96,0 20,9 1510
1987 180 95,9 21,1 1523
1988 212 94,8 15,7 1108
1989 241 93,9 14,4 1040
1990 236 93,9 14,6 1013
1991 216 94,8 11,4 815
1992 244 93,2 12,4 871
1993 187 93,8 11,1 766
1994 221 93,5 8,6 592
1995 266 94,3 13,5 957
1996 247 93,5 11,4 758
1997 252 94,8 13,9 877
1998 324 93,4 13,2 842
1999 293 92,9 13,2 896
2000 360 93,7 12,7 826
2001 378 92,2 13,5 893
2002 422 92,4 14,4 949
2003 513 92,7 13,9 925
2004 685 91,7 12,1 787
2005 824 92,2 14,0 903
2006 968 92,4 13,8 908
2007 1151 92,2 16,6 1103
2008 1579 92,3 18,4 1226
2009 2080 92,7 18,3 1191
2010 2308 92,9 20,5 1342
2011 2171 93,2 23,3 1535
2012 2666 92,9 23,6 1535
2013 1976 93,8 26,4 1769