Thursday 4 November 2010

Cargo cult reading

Just a reminder that folks ought read with caution arguments of the form
  • Vocabulary at 22 months predicts educational attainment;
  • Reading to kids can help vocabulary;
  • Therefore, read to kids to help them in the long run.
Hey, reading's great. And fun. Ira now recites Where the Wild Things Are back to me, and that's cool. But thinking that moving from not-reading to reading to your kids will ensure outcomes commensurate with those of heavily reading kids is cargo cult thinking.

High IQ folks tend to have lots of books in the house and tend to read to their kids 'cause they enjoy it. They have high IQ kids because of assortative mating and intelligence's heritability. Even if they never read a word to their kids, those parents' kids would probably wind up with pretty decent outcomes just from innate IQ and hearing their parents talking: higher IQ folks have a vocabulary range maybe twice or three times the average. And I know I've seen studies suggesting that, if you control for number of books in the house, the amount of reading a parent does with his kid doesn't have much effect on the kid's outcomes.

So then vocabulary at 22 months may predict educational attainment because they've gotten a taste for reading from their parents, because they've inherited good genes from their parents, or a combination of both. Simply looking at differences in mean educational attainment by differences in mean vocabulary at 22 months doesn't tell you much at all about the effect of reading on educational attainment.

Similarly, sorting kids by income and vocabulary scores tells you little about the effect of income on kids' ability if IQ predicts income and if IQ is heritable.  The report linked above contains this graph:
The report notes that some of the effect is mean reversion, but also points to the different slopes as suggestive of the effects of income.  An entirely plausible alternative reading is that it's reflecting mean reversion to different means.  If it's more likely that high IQ parents have higher income, then a kid from high income parents who tests badly at age 3 is more likely to have had an off day than a kid who presents the same score but who comes from a lower income cohort.

Income may well be doing some of the work here, but ignoring IQ seems a bit odd.  Especially since the Millennium Cohort Study included measures of parental educational achievement, as seen in the study of the effects of prenatal maternal drinking that was also based on MCS data.  Sounds like there's fodder there for a future honours research project.


  1. If I understand what the graphs are presenting, kids were assessed only twice, so each 'slope' is actually a line joining only two points. In addition, there is no indication of standard errors around those two points. Thus for all purposes there may not be any significant difference between the lines.

    We read a lot with my son and the adventures of Tintin are a blast. I want to inculcate a love for reading and learning that, I hope, will result in a good education (in the broad sense of the word). Test scores are not our objective though.

  2. Yup. Two points make a line. No t-tests.

  3. 1 year old Isla was shoving books in my face as soon as I got home tonight...I guess that's a good sign. Hopefully it means a good income to support me when I get old.