A new study (summarized here) casts doubt on the popular notion that exposure to sex in the media is linked to earlier sexual activity. “There is a common problem in social science research called the third variable problem,” said Laurence Steinberg, one of the study’s authors. “When looking at the relation between a given behavior and given experience, it could look like there is a correlation, when in fact the relationship is dependent on something else entirely.” In order to address the “third variable problem,” Steinberg re-analyzed existing data, controlling for “adolescents’ propensity to be exposed to sexualized media.” In Steinberg’s more conservative analysis, the relationship between media exposure and early sexual activity disappeared. “There are many reasons to find the portrayal of sex in mass media objectionable,” says Steinberg. “But let’s not confuse matters of taste with matters of science.” (HT: Carl Beyer) [%comments]
So Maybe Sexy Media Doesn't Lead to Teen Sex?
and 
. You can also watch it on 
Instances of sexual activity may seem on the rise in general, but a large part of that is the liberation or volume increase of sexually related activity in general and more accurately, in the media. In saying that, clearly there are correlations between exposure and activity but the causative factors of sexual activity is numerous and intricate given the nature of human relationships.
Media portrays sexual activity in such vast volumes but the reality is that when the actual nature of a sexual encounter is shown, it is almost exclusively sexual intercourse which is shown or suggested. The transparency of oral sex in particular may enable the extreme reduction in teenage pregnancy, intimacy in relationships and true liberation and enjoyment particularly for females. This may seem like a crude or drastic idea, but the truth is: sex is mainstream but how it is done is not. Intercourse is far from the only option.
The correlation and causation argument is clearly debatable. Teenage hormones is not. There shouldn’t be media volume without transparency.
-BillyM
Delighted to see “third variable problem” appear here! Generally news stories go to far in inferring causality,. which is the case in, I think, only a riny percentage of correlations – these third variable problems are there in a great majority of cases.
Yet, you still are left with the question – if only teenagers with propensity toward seeing prurient material are affected, is that cause enough for concern? Sorf ot like alcohol and gambling where while most do not abuse them, some do, but in this case perhaps it’s a majority instead of a minority..
@frankenduf
Well, unless I misinterpreted your post, you are indirectly proposing something interesting here. You might be viewing the conclusion brought forth by this study with disapproval because you fear it gives ground to marketing tobacco to children.
I’d argue that this sort of scientific approach works the other way around.
Suppose there is someone out there who argues that celebrity advertisement of tobacco for children isn’t harmful due to this third variable problem, i.e., children who consume tobacco would have done so either way because they just like tobacco from the beginning or have a natural propensity to enjoy smoking.
This scientific approach would not, however, be satisfied with such an argument: it would also demand testing to be done to qualify it.
So suppose you do calculate the correlation between tobacco consumption and celebrity advertising of tobacco while controlling for an underlying natural propensity for interest in tobacco.
Some economist out there may have done this already, but I’d bet that this controlled correlation between tobacco consumption and celebrity advertising WILL NOT disappear! Meaning that celebrity advertising does harm children. If the correlation DOES disappear however, we’ll have to face the FACTS and accept that such form of advertising doesn’t have the adverse effects we thought it had, and instead direct our efforts to variables that do affect premature tobacco consumption, such as education (those things that affect the propensity in the first place!).
The article does not advocate that you can make correlations disappear if you control for underlying propensities. It argues that you need to control for it to test if the correlation survives or not before you make any further conclusions.
I’m a media effects researcher and out of curiosity I looked over the Steinberg and Monahan article and, at least at first glance, it’s not very convincing to me.
At least in the white sample, they controlled for 12 covariates, which seems unreasonably conservative to me especially given the sample size (dropping to 176-186 in critical analyses). Even with those covariates, they still found sexy media to marginally predict age of sexual debut (regression coefficient of 0.357). When they did their critical analysis of also matching for propensity to be exposed to sexy media (without replacement) the coefficient barely changes (dropping from 0.357 to 0.308).
However because the statistical power was so low with all those covariates even an effect size estimate that many researchers would be describe as large is non-significant by a fairly small drop in the coefficient.
In the model that uses all the covariates and matching with replacement, the coefficient drops to the point of being trivially small (0.013) however the standard error also more than doubles in that model (going from 0.194 to 0.480), which can happen with analyses that match with replacement.
The more important point to be made is that the initial analysis with 12 covariates was so conservative and yielded an effect size estimate that, despite being statistically large, was so close to non-significance that just about anything could push it over the edge. A drop from 0.357 to 0.308 is not meaningful (p values notwithstanding), which seriously undermines the explanation that sexy media exposure does not affect age of sexual debut when propensity to be exposed to sexual media is accounted for. It seems to me more like they are relying on a misleading statistical trick.
As with any analysis deemed too conservative, the results remain open to interpretation but I certainly wouldn’t reject the original authors’ (Brown et al., 2006) interpretation on the basis of these results.