Small Effects, Social Problems, and Change

Divorce kills!  Bad mothers breed killers!  Really!

The experience of divorce is associated with a statistically significant increased risk for early death from all causes. Quantitative synthesis of 32 prospective studies involving more than 6.5 million adults and over 755,000 divorces yielded an overall (multivariate adjusted) mean RH of 1.23 for the association between martial separation/divorce and early death . . . For men, the divorce–mortality effect approximates the meta-analytically derived risks for all-cause mortality associated with two important public health outcomes: having an elevated body mass index (McGee, 2005) and failing to get regular aerobic exercise (Nocon et al., 2008).

Now bad mothers and their bad kids.

The infants’ observed aggressiveness was significantly correlated with mothers’ mood disorder during pregnancy and with mothers’ history of conduct problems.  Infants’ observed aggressiveness was correlated with parents’ ratings of infants’ anger and aggression, which were also predicted by mothers’ mood disorder and history of conduct problems.  Our findings indicate that infants at risk for serious aggression can already be identified when the motor ability to use physical force first enters the human repertoire.

Here’s the Table that displays the results to back that.  Click to enlarge if needed.

The divorce study is a well done meta analysis while the bad mothers study is a well done longitudinal observational study.  The divorce meta contains studies similar in method to the bad mothers research.  The meta authors thoughtfully and carefully selected only particular, high quality studies for the meta.  Let me highlight just 3 of the 8 selection criteria that I think are most compelling.

  • Prospective cohort design involving a community sample of healthy adults.
  • Exclusion of cross-sectional studies (that use discrete-time modeling procedures and typically rely on the odds ratio statistic to quantify a cross-sectional association).  !!!!!!!
  • Exclusion of overlapping studies that report data from the same samples. In the case of redundant samples, the most comprehensive papers on marital status were selected.

The nature of these topics precludes experimental methods, but at least we’re dealing with strong longitudinal (prospective) designs, specific exclusion of those one-shot cross sectional designs, and carefully inspected publications to spot repeated use of the same dataset for different purposes.  So, this is about as good as the science of divorce and bad mothers is going to get.

Now realize the Small Windowpanes here.  In the divorce meta, that RH of 1.23 is less than half of a Small Effect.  The bad mothers outcomes are Smallish Windowpanes, in the neighborhood of 43/57.  And realize that these are observational studies with uncontrolled assignment to the assumed independent variables.  Sure, we’ve got large samples, but that invents statistical significance which is not the same thing as clinical, practical, or even scientific significance.  Yet, the authors of both reports argue as if we are dealing with a major threat to health and safety.

But both as a scientist and a persuasion guy, I see these outcomes as No News while the authors try to make something closer to a Doomsday Defense.  All Bad Science Is Persuasive!  Just that comparison to obesity and exercise in the divorce meta makes the thing overwrought for me.  Let’s spend billions of dollars on a National Divorce Intervention the same way we’re spending billions on Lifestyle and getting nothing for it.  When even good Observational Research is delivering Less Than Small Effects, it’s time to move on.

Just think like a persuasion maven for a minute.  Are there any proven interventions that reduce divorce rates at a national level?  How about this one:  Tough divorce laws.  Hey, remember the Bad Old Days when no one got divorced (the kids, baby, the kids).  Were we living longer back then?  Of course not.  And remember the carnage of all those lives lived in Quiet Desperation?  Laws don’t seem to be the answer here.

And with bad mothers who smoke, suffer mood disorders, and find themselves in a bad social index, what shall we do?  Mandatory anti-psychotic drug therapy?  Reversible sterilization for women who smoke?  Perhaps a swell persuasion campaign about Do It For The Kids?

We’re in the days of Hope and Change, so you could do that wide-spectrum policy thing, also known as the Kitchen Sink at the Table of Brotherhood, wherein you contact all the factions for their preferred Intervention, put them all in the Law, and away we go to re-election, if not the Shining City on the Hill.  We’ll even include more funding at NIMH for observational studies and for training APA and APS certified marriage and family counselors.  If your science isn’t getting where you want to go, get a law passed.  That’s what the Food Police do.

Thinking through these studies with a persuasion orientation reveals serious problems in their conclusions.  So even if a Little Small Effect is real, there’s little we can do to moderate it.  Which points me back to my old professor admonition that if you think you understand it, change it.  And if you can’t change it, then maybe you don’t understand it.

David A. Sbarra, Rita W. Law, and Robert M. Portley.  (2011). Divorce and Death: A Meta-Analysis and Research Agenda for Clinical, Social, and Health Psychology.  Perspectives on Psychological Science, 6, 454-474.


Hay DF, Mundy L, Roberts S, Carta R, Waters CS, Perra O, Jones R, Jones I, Goodyer I, Harold G, Thapar A,  and van Goozen S. (2011). Known Risk Factors for Violence Predict 12-Month-Old Infants’ Aggressiveness With Peers.  Psychological Science, Aug 18. [Epub ahead of print].

No doi available.