A recent medical study has gained some media attention by finding and arguing that, among other factors, consumption of margarine by infants leads to lower IQ scores. The study, based on 590 New Zealand European children claims many other factors can contribute to the IQ of children such as consuming fish weekly and grains 4 times a day. Also, the study suggests that women who drink moderate amounts of alcohol while pregnant have a positive effect on IQ.
The news article suffers from simplification and does not address some of the vital questions that are demanded of it such as the statistical analysis used, particular controls implemented, data collection, and everything else a statistician would want to know – but the article does go a bit further than other news pieces devoted to statistics based research. The article does give us enough room to pause and consider whether or not the link is merely spurious – that is low IQ and consumption of margarine may be caused by poverty (and other associated factors) rather than poverty causing margarine consumption which leads to lower IQ. This could be the case since margarine is a cheaper product than butter and other possible alternatives for consumption.
So, given the summary, the actual work becomes necessary and I believe I have found it here. The 453 page pdf contains lots of information and various tests. The main findings for margarine are presented in Table 64 (page 394) and come from two tests. First, a bivariate linear regression. Second, a multivariate regression. The first test, while useful for initial impressions, is improper for determine causation, especially without a strong theoretical consideration driving this particular variable. However, the multivariate regression still presents a robust statistical result (p = .03) (159).
The regression includes the following controls: "gestation, parity, gender, maternal school leaving age, parental occupation at birth, maternal marital status at birth, maternal BMI, children’s BMI, Stanford Binet examiner" (Table 64, 394). While the result is statistically robust, it appears the magnitude is relatively small (2.81 points) at age 3.5. By 7 years old, the correlation was significant at the .1 level only for the bivariate test and loses significance in the multivariate model. At best, then, it appears that margarine would only be a problem for kids under 4 years of age. While non-significance is not a safe bet for conclusion, an initial review of the evidence suggests the surprising result is not consistent but appears strongly in one of the four tests (two IQ tests, 2 age groups).
The control variables for the study contain some elements of socio-economic status (SES) but are not as complete as I would like. The article summary suggests that only 5% of the parents in the study were of a low SES status which may factor into who is feeding their children margarine below the age of 3.5 years. While I am not an expert in the social structure of New Zealand, this does prevent the study from being generalizable to other populations. I would guess that part of the depression in the number of impoverished children in the study results from the target sample of New Zealanders of European descent, but that is speculation on my part. As such, a relative measure of income should be included – occupation may proxy parts of this, but the diversity of pay may factor into this. Additionally, marriage status may not account for single or dual-earner income households which would boost the selection and substitution choices of such a family. Thankfully, the article also notes that exercise is not controlled for – a key component to any lifestyle choice.
The multi-variate regression may not be enough to capture the full effects of margarine. Sceptically, I doubt that margarine is an independent variable that meets the conditions of independence within the regression and some forms of simultaneity ought to be taken into account. At the very least, margarine should be treated or tested as a moderator/mediator variable and see if it enhances the independent effects of poverty on IQ.
Regardless, the author is cautious in the article and has every right to be. It is obvious more statistical refinement and further data collecting would be required before we all feed our kids butter (since it has no effect on IQ in the study) and start drinking moderate amounts of wine pregnant. As such, "What to feed your kids to make them smarter" may be a very misleading headline for a news article.
I realize this is not political science, but it is quantitative…and sometimes news articles annoy me.