Wednesday 11 December 2013

Does it really matter if you talk to your babies?

For many mothers and fathers, it is the most natural thing in the world to talk to their baby. Babies may not understand what you say, not initially at least, but why not talk to them anyway? It gets you in the habit of explaining things to them, and they get in the habit of hearing you talk. This habit should lead to happiness and learning and the transmission of culture. Hard to see anything wrong with it.

However, it is another step to attribute a great deal to this verbosity. Some talking may be enough, and more talking may not add much. Sometimes a parent may not feel like talking, perhaps because they have been up all night dealing with a crying baby and find that, as the famous Guardian editorial put it, “a period of silence would be welcome”.

However, there is a body of research which seems to suggest that talking has a strong causal effect, and that failure to talk enough to a baby can have a deleterious impact on their linguistic and cognitive development. The argument makes sense. Children need stimulation, and need to learn the intricacies of the local language.

Is it true? A paper in Psychological Science suggests that it is. Adriana Weisleder and Anne Fernald “Talking to Children Matters: Early Language Experience Strengthens Processing and Builds Vocabulary”

However, on closer examination it begins to fall apart somewhat. The obvious major confounder is that brighter, more verbally competent parents are likely to give birth to brighter children, given that intelligence is roughly 60% heritable when assessed in maturity, though it will appear to be less than that when assessed on younger children. Any researcher needs to be alive to that possibility, and find a way of dealing with it. The authors mention Oliver and Plomin (2007) so they are aware of the issue. A very, very obvious way of dealing with this is to give the mothers (and the fathers) an IQ test. Even a simple vocabulary test, plus a digit span test or speeded coding task would provide a working answer to this question. The authors did not do this, or did not report it. So we are left with an observational study which did not control for parental intelligence, so it is difficult to sustain a causal interpretation of any results.

The authors chose to study low SES Latino families. They have data on maternal education, which ranged from 4 to 16 years (M = 10, SD = 3) and that was used as the primary index of SES which was “controlled in all analyses”. (My emphasis). Incidentally, this is a very big range, which makes it better as a predictor, but is also something of a worry. It suggests that some mothers simply did not see the need to stay in school for very long. Mexican schooling has been free, secular and accessible to all since 1917.  It is also said to be not very good, but that is another matter. The US also has a free public school system. In both systems brighter children with brighter parents tend to stay longer in school, but length of schooling is a crude measures. A test of written Spanish or some other simple scholastic test would be a better measure. The authors do not report such assessments.

Anyway, years of maternal education, the only proxy for intelligence, was controlled for, which means controlled OUT! Classifying education as SES imposed from outside obscures the potential contribution of genetic intelligence revealed by maternal education. It is the sociologist’s fallacy again and again. You might wish to stop reading now.

The authors found enormous variation in the number of words heard by babies, particularly those directed at the baby. Of this very interesting finding they say: “These differences in parental engagement were uncorrelated with maternal education (r = .29, p = .13)” So, they accept the null hypothesis of no correlation in this case.

At this stage I have to point out that 29 subjects is too low for rejecting the influence of education at r = .29.  The study does not have sufficient power to determine whether that correlation can be set aside. In fact, given that length of education is a weak measure of intelligence, and that the link between maternal IQ and child IQ is lower in infancy, and grows with age (and grows more when children leave home!) it is likely that 0.29 is probably as sizeable a correlation as one would be likely to find if there was a true relationship tested at this age. Furthermore, even if one accepts the small sample, the correlation is close to significant on a one tailed test, which is a valid assumption because in virtually all studies vocabulary is associated with higher intelligence. Enough special pleading: the sample is too small. Years of education and talking to babies are probably weakly correlated, which is informative.

Here are their rather bold conclusions: “Our results reveal that caregiver talk has direct as well as indirect influences on lexical development. More exposure to child-directed speech not only provides more models for learning words but also sharpens infants’ emerging lexical processing skills, with cascading benefits for vocabulary learning. If increased opportunities for verbal interaction can strengthen critical processing skills that enable more efficient learning, then interventions aimed at increasing parents’ verbal engagement with their infants have the potential to change the course of vocabulary growth and, in turn, to improve later outcomes for disadvantaged children.”

All of this may be true, for all I know, but the paper does not prove their case. It is observational, not experimental, and they haven’t bothered to collect that little bit of extra data on the mother’s intelligence which would show a possible maternal IQ/child vocabulary link. If they were to go back to the mothers and quickly do the tests I had mentioned, the paper could be improved considerably. Worth a try? I think so.

In the mean time, talk to your baby if you have something interesting to say, but please do not prattle at them out of a sense of social duty. Babies may prefer to concentrate on something else, without background chatter.


  1. On a related note, the social science industry that does not control for IQ (prime example here) can actually have some rather unpleasant real world consequences, namely wasted time, wasted money, and unnecessary, unpleasant guilt-tripping of working-class (often black) mothers. All of which you can read about courtesy of Slate Magazine here:

    1. Yes, I had seen the Slate article. I cannot be sure how big the separate effects are. Of course language must be heard to be learned, but the effect may plateau, and more talk may not lead to more anything, absent intelligence.

  2. I think the comment about a period of silence being welcome originated with Clement Attlee.

    That said, a period without click-bait, Gladwell-friendly articles in Psychological Science would certainly be very welcome.

    1. Attlee to Laski, “I can assure you there is widespread resentment in the Party at your activities and a period of silence on your part would be welcome.” Thanks.

  3. Indeed. With N=29 and r=.29, the statistical power to reject the null is about 34% (alpha=.05, two-tailed test). In other words, even if the real population correlation is .29, with that sample size the sample correlation is non-significant most of the time. Psychological Science is supposed to be the "leading empirical journal in psychology", so why does it keep publishing stuff like this?

    Not only is there a genetic confound in this study because parents and children share genes, but additionally the propensity of parents to talk to their kids is influenced by the kids' innate differences.

  4. Yes, if a child is unresponsive, then with the best will in the world parents will interact less. Shows up in large families, in which mothers report these temperamental differences.