Jeromy Anglim's Blog: Psychology and Statistics


Friday, September 25, 2009

Spanking Lowers Child IQ | Correlation is not Causation

As I was waiting for a coffee this morning, I had a cursory glance at the paper and found The Herald Sun reporting on a study stating that:
"A SPANK on the bottom, long used by parents to discipline a naughty child, could cause more than tears... It's now thought the age-old disciplinary method may also lower a child's IQ, with those spanked up to three times a week having a lower IQ due to psychological stress."

I then said to myself,  "If these researchers were able to determine that spanking has a causal effect, I guess they must have done an experimental study?" But how did they manage to get that past ethics? The hypothetical experiment that would allow for an unambiguous causal inference on the effect of spanking would involve assigning one-group of children to a no-spanking group and one group to a "spanking up to three times a week group". The mean IQ of the two groups could then be compared. And a logical inference could be drawn about he mean causal effect of regular spanking.
I read on:
"After studying 800 toddlers aged between two and four for four years, [the researcher] found those who were subjected to spanking had an IQ five points lower than that of a child who wasn't physically disciplined."
It becomes clearer. It's an observational study, albeit longitudinal. Thus, the research shows an association between child IQ and spanking. Last I checked, an association in an observational study does not mean causation. However, an association is often one piece of evidence in an argument about a causal effect.

To make the leap from an association to a causal inference it is useful to show that there are no alternative causal explanations, or at least argue that the alternatives are less plausible. In the presently discussed research on spanking, let's look at a few alternative causal explanations:

  1. Less intelligent parents have less intelligent children and less intelligent parents are more likely to smack their children.
  2. Less intelligent children behave in ways that provoke smacking responses in parents more often.
  3. Low socio-economic status and a generally poor parenting environment leads to both a greater likelihood of smacking and reduces intelligence in the child.
The above three alternative causal models assume that any association between child smacking and intelligence is, or is largely, epiphenomena. I find all of them more plausible than the claim presented in the newspaper. There is strong evidence for the genetic component of intelligence. And there are no doubt associations between socio-economic status and quality of parenting. I'm no expert on this literature, but there's a general principle at work:
To cause a big Dependent Variable to change, you need another big Independent Variable.
What do I mean by this?
Intelligence is famous for being difficult to modify. It has a strong genetic component and major interventions have often struggled to make appreciable changes. It's a big variable in that it represents the overall cognitive functioning or capacity of an individual. However, without wanting to weigh into debates about when smacking is or is not detrimental to a child's development, child abuse seems like the big variable, whereas smacking in and of itself seems like only a moderate variable. Thus, prima facie I am suspicious of the causal inference.

Thus, the core points:
  • Correlation does not prove causation
  • Correlation is consistent with multiple candidate causal inferences
  • To prefer one causal claim over another requires: a) the identification of alternative causal claims; b) making reasoned arguments about why the alternatives are less likely to explain the observed association
  • Causal effects on big dependent variables are more likely to result from other big variables; and if a dependent variable is small, causal effects are likely to flow from other small variables measured in the same setting or context.
This all got me thinking: Are there any interventions which could exert a causal effect to improve the reasoning about causality in popular newspapers?