The difference between a statistical relationship and a causal relationship is one of the few things I will always remember from college statistics. Articles like this constantly remind me why it’s so important for people to know the difference.

Teens whose iPods are full of music with raunchy, sexual lyrics start having sex sooner than those who prefer other songs, a study found.

Songs depicting men as “sex-driven studs,” women as sex objects and with explicit references to sex acts are more likely to trigger early sexual behavior than those where sexual references are more veiled and relationships appear more committed, the study found.

The study was done via a telephone, which means it is an observational study, not an experiment. Causality can not be determined in an observational study. If you ever read a study that’s conducted by survey or outside observation, just remember that one can never conclude causation from it. This study certainly does not prove that songs “trigger” early sexual behavior. What it says is that they’re statistically related – nothing more.

The reasons for this are precisely why scientists do experiments. The reasons an observational study can’t prove causation is because an observational study cannot account for:

1. The direction of causation. Do listening to sexual songs cause people to have sex earlier, or do people who have sex earlier have more of an interest in listening to sexual songs?
2. Confounding variables – a third variable that causes both symptoms. For example, maybe a rough upbringing causes kids to both enjoy sexually explicit songs and partake in sex at an early age.
3. Bias – This problem isn’t unique to observational studies, but it’s still readily present. Were the people they sampled sufficiently randomized or stratified? Did some group of people choose not to participate (for example, perhaps people who had sex at an early age but don’t care for sexually-oriented music were also more reluctant to participate or answer truthfully)? Did the interviewers know what hypthesis they were testing (the general problem of the observer effect)?

Sadly, I’m not sure a lot of people think about these things. Data like this can then be misinterpreted by writers, who create articles like this which mislead the general public.