Note that the intelligence tests are for people who are evaluated for conscription, not the ones who are finally conscripted, so they cover a large fraction of

Note that the intelligence tests are for people who are evaluated for conscription, not the ones who are finally conscripted, so they cover a large fraction of the population (80%-90% of males). Since not everyone was given the test, there are indeed large selection effects. The main technical contribution of the paper we are discussing (http://www.pnas.org/content/early/2018/06/05/1718793115 http://www.pnas.org/content/early/2018/06/05/1718793115) is to try to estimate what the selection effect is. They use a comparison of siblings. By comparing the score of brothers who were both given the test, one can can create a model to estimate the score of one brother from the score of the other. Then by comparing families where one brother was given the test and the other was not, one can estimate how the the probablility of the test board administering the test depends on the IQ of the testee. Then given that curve, one can estimate the average IQ of the entire population from the available test scores. Their result is that selection bias hides some of the effect. That is, as the Norwegian military got smaller, the selection board got pickier and tested fewer (smarter) people, so if you just look at the average score you will not see the full drop.