New Evidence Links Spanking to Child Behavior Problems, but Perhaps The Evidence Isn't So Clear

This post was published on the now-closed HuffPost Contributor platform. Contributors control their own work and posted freely to our site. If you need to flag this entry as abusive, send us an email.

Social science research tends to overlap with a fair number of cultural debates, often weighing in on one side or another. This isn't surprising, given that social science is often trying to give society answers to pressing questions. In the latest such example, a new meta-analytic study claims to link spanking (defined by the authors as hitting a child on the buttocks or extremities with an open hand) with a host of behavior problems in children. Yet, reading the study, it's not clear that the case is well made.

Let me say upfront, I'm not a spanking advocate (if that's a thing.) I'm perfectly happy if spanking is abandoned world-wide. So my comments here on the science should not be mistaken as an ode to spanking. But there are many realms where well-intentioned moral advocacy and actual good science may become confused.

Debates about spanking and its hypothesized harmfulness have been going on for decades. Many studies have been conducted on the issue, but some scholars have raised concerns about systematic flaws in this research. For instance, many studies don't control for other factors in the environment that might explain small correlations (and they are small) between spanking and problem behavior. For instance, family environment, parents' mental health, the child's preexisting behavior problems, etc. Even's not unreasonable to hypothesize that boys are perhaps a bit more unruly and, as a consequence, more likely to be spanked. So controlling for these other variables is crucial. Indeed, in a 2013 meta-analysis of my own, I found that controlling for such variables tended to reduce effect sizes between spanking and child problems to trivial levels.

Also of concern is the issue that spanking is often conflated with other forms of abusive behavior. For instance, parents who abuse their children probably often spank them a lot too. So asking parents the question "How often do you spank your child?" may end up inadvertently tapping into more serious forms of child abuse, even if that wasn't the intent. The ideal study would (a little statistical inside baseball here), query about both spanking and serious abuse, and, in correlational analyses, remove the explanatory variance of serious abuse and see if anything was left over to correlate spanking with problem behaviors. Unfortunately, few studies do this.

Even more unfortunately, this new meta-analysis doesn't really improve upon these problems. A main reason is that the study used (again, a little inside baseball) what are called bivariate correlations, not better controlled effect sizes. Bivariate correlations are simply the correlation between two variables without anything else controlled (like gender, parent mental health, etc.) In most research, these are well known for being spuriously high. Particularly in longitudinal studies (studies that follow kids over time), it's important to control for pre-existing behavior problems...but this meta-analysis appears to have explicitly not done this. In other words, there are still good reasons to hypothesize that the small observed correlations between spanking and outcome behaviors may be due to other, uncontrolled factors. Or that spanking is still being conflated with more serious forms of abuse.

This is a common problem for meta-analysis, spurred on by the scientific old-wives-tale that meta-analyses work best with bivariate correlations. This might be true in some medical research, using identical measures, methods and outcomes. In the messy world of social science, different studies ask questions different ways, analyze things differently, use different populations, etc., and bivariate correlations aren't really comparable to each other from study to study any more than controlled effect sizes are. As noted, scholars have been arguing for years that using better controlled effect sizes is psychometrically warranted and theoretically superior. Meta-analyses' foolish and rigid adherence to bivariate correlations has probably done more to misinform than inform the public about social science. It's theoretically possible that, in a research field, every single study finds that controlling just a few confounding variables (like gender, mental health, etc.) reduced correlations between two things of interest (such as spanking and problem behavior) to zero, but a meta-analysis of such studies, employing bivariate rather than controlled effect sizes could still claim an effect exists. To be honest, this is dumb. But it's social science as it continues to work today.

Ultimately, even the evidence from this meta-analysis is rather weak. Of the 18 outcomes the authors looked at, 14 were "statistically significant" which sounds fairly impressive. But the effect sizes were generally trivial and it's not clear that most are of "practical significance." Of the 18 reported outcomes, only 4 achieve a level I've recommended in the past for "practical significance." Other studies have suggested that these small correlations you see for spanking and problematic see almost identical correlations for almost anything else you do with children...whether time out, ignoring the child (as behavioral specialists often tell us to do), reasoning with the child, etc. It may be that small correlations are just kind of an artifact of the research linking any two things together in a survey, respondents may begin to answer them similarly. If spanking correlates with negative outcomes, but only to the degree that interacting with your child in almost any other way correlates with negative outcomes...I'm just not sure what we're supposed to do with that data. Psychological science has been historically awful about putting its data into any kind of coherent and reasonable perspective. This is just another such example.

Ultimately and intuitively, I see no reason that spanking need continue to exist. But I am troubled by the ease with which social science jumps in feet first into culture wars without restraint, armed only with moral certitude and bad data. I've said this before on different issues but...we need to do better than this.