From PolicyMic:
For decades, a research team at King’s College London has been tracking a group of 7,771 people from England, Scotland and Wales who were all born in the same week in 1958. At age 7 and 11, their parents were asked to provide information on whether the children had been bullied. More than one in four had occasionally been bullied, while 15% were bullied frequently. Then later in life, the participants were asked to check in periodically and provide feedback on their health.
What the researchers found was shocking: At age 50, those who had been frequently bullied as a child were much more likely to have depression, anxiety, suicidal thoughts and poor physical health than those who were not. In addition, they were less likely to have higher education degrees, high-paying jobs, romantic partners or social support. Even when childhood intelligence levels were taken into account, these adults also scored lower on cognitive IQ tests. In general, they reported lower quality of life and satisfaction.
"The impact of bullying is persistent and pervasive, with health, social and economic consequences lasting well into adulthood," said Ryu Takizawa, the lead author of the paper.
At the risk of being a killjoy, does the study control for reverse causation?
That is, are the social scientists assuming that bullies pick victims at random? Have they controlled for the possibility that bullies are disproportionately likely to pick on kids that show signs of weakness — weakness such as depression, anxiety, suicidal thoughts, poor physical health, academic difficulties, low IQ, or a difficulty with social situations and making friends?
Here’s the article itself (no pay wall!).
I haven’t read it yet, but I wondered the same thing as nobody really did. However, just from glancing at the abstract: “and the effects remained significant after controlling for known correlates of bullying victimization.”
Thanks for the link, Charles S!
The relevant section is on page 2 under “childhood confounders.” They looked at IQ and behavior problems. They also assessed class based on the father’s occupation in broad categories (professional, other nonmanual labor, skilled manual labor, unskilled manual labor) and used both questionnaires at the time and retrospective questionnaires to figure out adversities such as childhood poverty, low parental involvement, or abuse (as well as several other adversities). If you want to see the effect of controlling for these factors, compare the first set of columns from Table 2 with the columns in Table 3. The results remain statistically significant, for the most part, but the effect gets somewhat smaller. I’m not a statistician so I can’t say if this method is the best for what they’re trying to accomplish, but based on my casual perusal of such articles, their methods seem standard for the field. They note that there may be other markers of adversity they didn’t have measurements for which could also be confounding factors; they tried several additional characteristics including physical disability but controlling for those didn’t make a difference in the final results.
They point out something interesting in their discussion:
I read it after posting the link, and found that note interesting as well.
The paper they are citing there is also not pay-walled (I haven’t read it yet).
From discussions around the Lancet Iraq death toll studies, my recollection is that the 1.5 relative risk (RR) factor was considered to be about the lower bound of a meaningfully measurable RR. This paper reports an OR (odds ratio) of 1.5. This isn’t my area of statistics (and I’m not a statistician, just someone who uses statistical methods for work), but wikipedia supports my belief that OR and RR are closely related statistical measures, with RR being a more conservative measure than OR. On the other hand, this study had a quite large sample, so it looks like they are justified in considering their results significant. The one statistical problem I see with this study is that they don’t seem to be doing anything to control for the problem of having a grab bag of possible effects, where some show up as statistically significant and some don’t. The chance of having some of your tested measures show up as statistically significant goes up rapidly as you increase the number of measures you test, and it if they controlled for that, I missed it.
And now I’ve gone and read the wikipedia article on the Lancet study and made myself angry and sad.
Good find, I’ll have to go over the study to see if they worked the data for all it was worth. For instance in a good 30 year longitudinal study they could also look at what factors might have helped minimize long term effects of childhood bullying. Compare the family support systems of those who had minimum/maximum long term negative effects, what therapy did they have over the 30 years, faith in a higher power, etc.
I have a hunch they focused on easier things to measure; percentage of suicide attempts, addictive patterns, failed marriages…