Friday, December 30, 2011

Childhood Obesity? Partly The Fault of Bad Mothering, Of Course

A new study found this to be the case. Honest! But more about that later in this post. First, let's look at a few summaries of the findings:

Raw Story reported* it like this:
According to a new study in the January issue of Pediatrics, children who struggle to connect happily with their mothers are more likely to be obese by their teen years.
Ohio State University conducted the research, using almost a thousand kids born in 1991 to measure how mothers interacted with their children during various stages of childhood. Researchers studied whether children felt safe with and attached to their families.
The study found that 26.1 percent of children who reported troubled relationships with their mothers were also obese at age 15, a rate double that of children who reported close relationships to their mothers.
CNN blogs reported it like this:
The mother-child relationship has always carried a lot of weight.  Now researchers say some obese teens might be in essence, carrying the weight of their relationship with their mothers when they were younger.
A new study published in this week's edition of Pediatrics finds the type of relationship a mother has with her young child could affect that little one's chances of becoming obese as a teen.
And the New York Daily News like this:
A bad relationship with your mother can do more than leave emotional scars — it can also increase your waistline.
A new study in the January issue of Pediatrics found that children who did not have close emotional bonds with their mothers during childhood were significantly more likely to be obese as teenagers.

After those three summaries I'm sure you are ready for the necessary corollary. This one:
Anderson was quick to note that the findings should not be used to blame mothers, but should be seen as an opportunity to intervene in mother-child relationships while children are still young.
The findings should not be used to blame mothers! What a relief! For a while there I thought that this is exactly what is happening.

For something this important I had to get hold of the actual study. Which I now have read**. But before commenting on it, let's ask what the starting point of a study like this might be.

Did the researchers go out to test both fathers and mothers, for instance, to find out what the impact of both fathers and mothers might be on a child's obesity?

Can you guess the answer to that one? Yup, they only tested mothers, not fathers. So we know nothing about any possible impact the father's bad parenting skills might have on a child's later obesity, simply because fathers were not studied.

And why were fathers not studied? Because the researchers wanted to study the behavior of the main caregiver to the child! But notice the way those summaries of the study were about mothers, not about the major caregivers? That's because the study used the term "mother," not the term "caregiver."

That's a minor slippage, you might argue, because mothers usually are the major intimate caregivers to their children. But it is slippage, nevertheless, because using the term "mothers" makes us think of the family relationship between a woman and her children, not about the care-giving situation.

One final comment before I dive into the study itself: Note how negatively those summaries are framed. They essentially tell us that bad mothering produces fat children. Why not re-frame those findings by saying that good mothering protects against childhood obesity?

I think the reason for that comes from the assumption that all mothers should be perfect. If they are not, their children suffer and the mothers should shape up. Or have suitable interventions, as one of the study authors proposed.

Now to the study itself: The first question I wanted to have answered is an obvious one:

Did the findings control for socio-economic factors, especially income? This is an important variable to control for because poverty could explain both problems within the mother-child relationship AND childhood obesity. Note that this theory does not require the causality to go from bad mothering to obesity, necessarily, but argues that both could be due to the stress and limitations that low family income create.

It turns out that the results mentioned in those summaries are based on data without any control for income and other relevant factors. They are raw comparisons, if you wish. For proper comparisons, I quote from the study itself:
The prevalence of obesity in adolescence was 26.1% among children who experienced poor early maternal–child relationships (score: greater than or equal to 3) and was 15.5%, 12.1%, and 13.0% for children with better relationships (scores of 2, 1, and 0, respectively) (upper section of Table 4). After adjustment for gender and birth weight (model 2), the odds (95% CI) of adolescent obesity were 2.45 (1.49–4.04) times higher for those with the poorest relationships (score: greater than or equal to 3) compared with those with the best relationships (score: 0). With additional adjustment for race/ethnicity, maternal education, and household income-to-poverty line ratio, the OR (95% CI) was attenuated to 1.56 (0.90– 2.73), and with inclusion of maternal obesity to 1.42 (0.76–2.63). Low maternal sensitivity was more strongly related to adolescent obesity than was insecure attachment (lower section of Table 4).
That's the statistical gobbledegook. Note that those numbers are created to compare the "worst" group with the "best" group, in terms of mothering. Which is pretty much the expected thing, given that the standard for mothering is perfection.

Then note that the numbers discussed in that quote are essentially how many times more likely obesity is among the children of the "worst" mothers as opposed to among the children of the "best" mothers in the sample the researchers used. If the likelihood of obesity for the child of a "good" mother is the number x, then the quoted material tells us that the child of a "bad" mother (in that sample) has the likelihood of obesity 2.45x, or more than twice as much, assuming that only the child's sex and birth weight are held constant in the comparisons.

But if we also control for the socio-economic and demographic factors, the likelihood of obesity for the "bad" mother's child drops to 1.56x, and if we also control for the mother's own obesity, that number drops to 1.42x. Remember that 1x would mean equal odds of obesity for the children of the "best" and "worst" mothers.

If you have read my statistic series (available on the site listed at the top of this blog's front page) or are otherwise familiar with statistics, you may already have gotten an AHAH! experience from looking at those confidence intervals in the quoted material.

A confidence interval is an interval estimate, a range of values within which we believe the true value in the population to lie, with some confidence. The study values come from a sample. How well the findings of that sample apply to the general population is reflected in that interval estimate.

Let's take the income-controlled results for closer scrutiny here: The sample finding, the value that I have already cited, states that after controlling for the socio-economic variables the child of a "bad" mother is 1.56 times more likely to become obese than the child of a "good" mother. But the interval estimate on that same figure ranges from 0.90 to 2.73. Note something funny about that interval?

It covers the value one which would be the point at which the children of "bad" and "good" mothers would have an equal chance of producing an obese child. In other words, the results do not rule out the possibility that after controlling for the socio-economic factors the likelihood of obesity might not, in fact, be higher among the children of the so-called "bad mothers."

Indeed, one table in the study (Table 4) shows something quite interesting:

It compares all the ranked classes of mothering skills with the "best" skills used as the reference point. When the other three classes are compared to the "best" class, the confidence intervals for all three comparisons cover the value one if the research controls for the child's gender, birth weight, the socio-economic factors and the mother's own obesity. Remember that the value one is the referent value, applied to the "best" mothering class.

What does all this mean? Suppose that you read a poll result where Jane Smith is predicted to win some election by 5%, with a margin of error of plus/minus 7%. If those numbers use the 95% confidence interval, then the poll tells us that Jane Smith might win by as much as 5%+7%, or 12%. Or she might lose by as much as 5%-7% or -2%. The confidence interval overlaps the point where victory turns into loss.

Which isn't extremely comforting if you root for Jane Smith. Well, the findings of this study are like that, with proper controls. Nowhere near as strong as the popularized summaries suggest.
*This summary has an error. The children did not do any reporting themselves. The measures the study used were collected when the children were quite small and were based not on reporting but on observations of mothers with their children either at home or in a laboratory.
**It cost me twelve dollars to acquire. The donation button is in the right upper corner of my blog. Mmm.

Thanks to NTodd for the initial link to this study.