Confounding Variables

Did you know smoking while driving reduces your chance of dying in an auto collision? That's right, cigarettes, when combined with a healthy routine of staying within speed limits, wearing safety belts, and abstaining from driving while intoxicated, lowers your car accident mortality a great deal!

It sounds ridiculous but this is how a lot of diet studies are done. Instead of being controlled (where only one variable is being tested), the experimental group in the trial changes a number of factors in addition to the one ostensibly being tested. They might promote their study for its reduction in saturated fat intake, but will also reduce trans fat intake, increase fruit and vegetable intake, and reduce refined carbohydrate intake as well. These bad studies can often be spotted because while a particular cause of death goes down in the experimental group, the rate of death over all stays the same. If go from driving without smoking or seatbelts to driving with smoking and seatbelts, your chance of car accident mortality will probably go down, but your chance of cancer mortality will go up.

Keep this in mind when reading meta-analyses (top level studies of multiple individual experimental trials), in which you're unlikely to get the whole story.