The focus is on conducting and reading research and determining good research study practice but I think there is value in everyone knowing what confounding variables are. So what are they?
Well, read that post……
Another example that I got in my stats class (many moons ago) was the correlation of matches with cancer. Those people who tend to carry boxes of matches in their pockets also have a higher risk of cancer.
As in the Murder vs Ice-cream example given at the link above, there is no direct link between matches and cancer (though it’s obviously related), the most probable explanation is that those who carry matches are more likely to be those who smoke and it is the smoking that relates to cancer.
Smoking can then be said to be the confounding variable – the variable that explains both of the explicitly stated variables and either ties them together with a causal mechanism (Matches -> Cancer) or shows that there is no direct relation (Ice-cream -/-> Murder).
A similar effect may be seen with something like surveys, the manner in which a survey is carried out may introduce confounding variables (say a phone or internet survey which pre-selects participants by their access to said communication methods) or the questions asked may smuggle in assumptions that do not separate out confounding variables.
For example a survey may ask “Are you Religious” and “Are you Happy” (as many have). The Religious question smuggles in a number of extra factors that may contribute to a person’s level of happiness eg religions usually come with a feeling of belonging to a community, social interaction, social support networks or guilt over actions and feelings. Each of which may more directly impact happiness that religion per se.
Other areas may also suffer from the confounding variable problem, alternative medicine springs to mind. Say you suffer from a cold, you soldier through it until you can’t take it any more and start downing some homeopathic remedy. In a day or two your symptoms resolve and you feel better. Did the remedy work?
In this case the confounding variable could be the natural history of the disease. Colds don’t last for ever (it is “self limiting”), it could be that you took the remedy right before the cold would have resolved itself anyhow. If this is so the conclusion that the remedy “cured” your cold would be invalid, there would not be a causal connection between the remedy and the cold symptoms going away.
The natural history of the disease would explain the reason you took the remedy when you did (symptoms had reached a climax) and why the remedy appeared to work (the cold would have resolved anyway).
When we examine issues closely we can see that confounding variables crop up, and should be carefully considered, every time we try to determine a causal connection between two events or phenomena. This is the reason that skeptics chant “Correlation does not equal causation” like a mantra.
Just for fun, suggest some instances of confounding variables in the comments. The more obscure the better.
- I Was A Skeptic, but… (sciblogs.co.nz)
- Five Signs You Might Be Wrong (sciblogs.co.nz)
- Amber Teething Beads a Few Points to Consider (sciblogs.co.nz)