The relationship between sibship size and intelligence is not that simple. Photo: Wavebreakmedia
The relationship between sibship size and intelligence is not that simple. Photo: Wavebreakmedia

For many years, researchers have been interested in whether there is a correlation between the number of siblings in a family and the children's intelligence. Two recent researchers in this field are Linda Wänström, lecturer in statistics and previously employed at the Department of Statistics, Stockholm University, and Gebrenegus Ghilagaber, professor and head of the Department of Statistics.

Linda Wänström, who wrote part of her thesis on the subject, tell us that observational studies are necessary to investigate the correlation.  This is because it is not possible, of course, to carry out experiments where you say to one group of families not to have any more children, and to others that they should have more children, to see if there are changes in the children's intelligence.

Instead, we must make measurements of children with different numbers of siblings, and compare results between them.

"And even then we don't know what influenced the differences, or if they would have been the same if they had not had more siblings. So there are many methodological problems here," says Linda Wänström.

More siblings - but no decrease in cognitive ability

In order to draw causal conclusions and demonstrate cause-and-effect, which can normally only be done with experiments, we need to use the right statistical methods. Gebrenegus Ghilagaber had previously worked with such methods, and he and Linda Wänström worked together to implement them to the relationship between the number of siblings and their intelligence. The result is presented in the article, "Adjusting for selection bias in assessing the relationship between sibship size and cognitive performance", published in the Journal of the Royal Statistical Society - Series A (Statistics in Society).

The data that were used are from a study that made measurements on thousands of families in the United States between 1979 and the 2010s. The data here were taken from the period 1986 to 2006.

Selection bias occurs when parents who decide to have large, or small, families differ systematically from the rest of the population of parents, according to the article. Thus, they are not a representative sample of the whole population of parents.

Gebrenegus Ghilagaber. Photo: Department of Statistics
Gebrenegus Ghilagaber. Photo: Department of Statistics

"After we had applied a model that takes the selection bias into account, we found that it is not actually true that cognitive ability decreases with the size of a family," says Gebrenegus Ghilagaber.

Disentangling cause and effect

More siblings do not automatically bring about lower intelligence in children. Instead of a simple causal link, there are more complex relationships that are intertwined.

Gebrenegus Ghilagaber explains that the decision to have more children in itself may be correlated to the family's characteristics, such as education, income and the parents' age. Mothers who were older when they had their first child were less inclined to have more children, for example.

The study also shows that families with children who perform below average are over-represented among families with more children. The reverse also applies: families with children who perform above average are over-represented among families with fewer children. This is where the selection bias arises:

"You can't just say that one affects the other, you have to realise that both affect each other," explains Gebrenegus Ghilagaber.

Here is how the method works

The method we use to check for selection bias is called "multiprocess-multilevel modelling". It means that you evaluate parameters from two statistical models simultaneously. One model (the process) is about the cognitive performance of children, while the other is about the family's inclination to have more children. In this way, the selection bias is taken into account.

Linda Wänström believes that the most important conclusion from this study is showing that when you are dealing with observational data, it is difficult to trust the effects you obtain - there may be methodological errors, or selection bias, or both.

"Using this method we can avoid some of the errors. Even if it is not possible to control all the factors, we still believe it is better than not doing it."

Translation from Swedish by Språkservice AB