When trying to interpret study results, statisticians have to take into account the fact that trials don’t happen in a vacuum. That is, when they want to ask whether a certain factor affects a certain disease, it is not enough to merely ask whether people with that factor are more likely to have that disease, because there may be other confounding factors at play as well. For example, a study to look at whether uterine irritability in pregnant women contributes to low birth weight of their babies should also take into account the confounding factor of smoking, which could cause both uterine irritation and low birth weight.
A standard statistical method to measure the effect of the confounding factor exists, but may be misleading in some cases when the cofounding factor does not have a linear effect – for example, if smoking two packs of cigarettes a day did not double the chance of low birth weight over smoking one pack a day. VIDI assistant member Dr. Holly Janes and colleagues created a new statistical method to measure the effect of confounding in nonlinear cases such as these.
On quantifying the magnitude of confounding. Janes H, Dominici F, Zeger S. Biostatistics. 2010 Mar 4.