On Friday, my wife sent me this study. James Bruce Tomblin and colleagues followed a group of people for about thirty years. They observed them from kindergarten until their mid-thirties. The researchers wanted to know whether differences in oral language ability at a young age could predict the socioeconomic position people would reach much later as adults. What did they find? The short answer: quite a lot. Read the interesting, but once again nuanced, story.

The researchers had data from 114 people who, as children, had participated in a much larger longitudinal study. This was 19 per cent of the original sample. Their oral language ability was extensively measured in kindergarten and again in second grade. Testing included vocabulary, sentence comprehension, listening skills, and the ability to tell stories. In addition, researchers collected data on their non-verbal intelligence, the socioeconomic situation of their parents, and their later school achievement. What was already quite a substantial dataset was expanded when the participants were between 35 and 38 years old. The researchers then looked again at their education, income, and occupational status.

This brought them to this striking finding. When the researchers took differences in non-verbal intelligence into account, early language ability was associated with about 24 per cent of the differences in socioeconomic position thirty years later! For comparison, socioeconomic background in studies such as PISA is around 11 per cent. In Flanders, figures sometimes reach around 20 per cent. So you know this is large, even though we are talking about a different measure here.

The researchers then tried to unravel this relationship and distinguished between two processes.

The first is what they call persistence. Children grow up in families with a particular socioeconomic position, and part of this is passed on from generation to generation. Think of intergenerational poverty, or the other side of the story, nepotism. In this study, the socioeconomic position of the family explained about 12 per cent of the differences in the children’s later position. Interestingly, about two-thirds of that relationship could statistically be explained by early language ability. According to the researchers’ model, language accounted for about 7.8 per cent of the total variance in socioeconomic position as an adult. That is less than what we just noted. But it is still quite a lot.

But the researchers also looked at mobility: who does better or worse as an adult than you would expect based on his or her background? Again, early language ability was quite strongly associated with this. After the researchers took parental SES and non-verbal IQ into account, language still explained about 16 per cent of the differences in this kind of social mobility. That is more than twice as much as the share they found for simply passing on the social position of parents to their children.

Or, in other words, language ability does not just predict the reproduction of social differences. It predicts even more strongly who later moves above or below the position you would expect based on their family background.

There is another striking result in this study. As I already wrote, the researchers also had data on school achievement around tenth grade. As you would expect, this was also associated with later socioeconomic position. In a simple model, about 35 per cent of the intergenerational association in SES appeared to run statistically through school achievement.

But… language ability and school achievement are very strongly related in this study: r = .73, with 1 meaning a perfect correlation. When the researchers included them both in the model simultaneously, little reliable unique contribution from school achievement remained. Language did remain a clear predictor. Mind you, this does not mean that school achievement apparently does not matter.

First, the sample is simply too small for that conclusion. This is an incredibly difficult dataset to collect. There are a lot of data points in this study. But they still come from only 114 people. The authors themselves stress that they cannot reliably determine, based on these data, how much school achievement adds to language ability.

Second, the school achievement measures may contain more measurement noise. The researchers measured language ability directly and at several points in time, whereas they measured school achievement later, combining, among other things, reading tests and ratings by the participants’ own teachers. More measurement error in that second measure could make its unique contribution look smaller. The researchers, of course, acknowledge this themselves.

The study also provides some additional support for considering language a core subject or skill. Language is not simply one school skill among many. It is also a tool we use to do much of our other learning. We listen to explanations, understand texts, formulate thoughts, ask questions, negotiate, and explain things to others. Later, we use language again to learn new things at work. So it should hardly surprise us that language is associated with so many things!

But if you thought I had now mentioned all the nuances with the limitations concerning school achievement, nope. What follows is almost a tradition: correlation, not causation. So this study does not show that better language ability causes social mobility.

It remains an observational study. Still, the thirty-year time span is a major strength. The researchers measured language ability when the participants were still very young children and socioeconomic outcomes about thirty years later. But even thirty years cannot make all possible third variables disappear. The authors are clear about this: genuinely causal conclusions would require experimental or at least quasi-experimental research. Of course, that immediately raises ethical and financial challenges. Sometimes you simply have to work with what you have.

But there are other reasons to remain cautious. As I already wrote, they were only were able to test 19 per cent of the original longitudinal cohort. And this group showed a clear bias as the people who continued to participate were, on average, somewhat more advantaged than those who dropped out. The researchers tried to correct for this. But you can never completely make thirty years of attrition disappear statistically.

Also, almost all participants were White, lived in the American Midwest, and grew up as monolingual English speakers. That last element is not an unimportant detail. This study may not only be about ‘language ability’ in an abstract sense, but also about how well a child’s language abilities fit the language used at school, at work, and the demands of wider society. The authors explicitly point this out themselves. For multilingual children and in other education systems and labour markets, the story may therefore be different.

I hope I haven’t nuanced the study to death in your eyes. Realise what this study does show: differences in language ability at a very young age strongly predict life trajectories that only become visible decades later. More strikingly, language in this study not only predicts the transmission of social differences but also predicts even more strongly whether people deviate from what you would expect based on their background.