You've probably seen this caveat pretty often: Just because a study that uses mice as subjects produces a specific result, doesn't mean you'd get the same result using human subjects. Mice are handy research animals, but they aren't perfect analogues to humans. A mouse study is a stepping stone towards better evidence. It is something we do because there are potentially useful ideas that we should not try out on humans first. But mouse studies should not count as incontrovertible proof of anything.

Usually, when that caveat comes up, the person giving it is talking about fundamental differences between mouse biology and human biology. For instance, a mouse might only need one copy of a genetic factor to grow normally. Meanwhile, a human needs to have both copies or risk altered sexual development.

But there are other problems with mice, problems that have more to do with how we select, breed, and raise mouse models. In a fascinating three-part series on Slate.com, Daniel Engber looks at how we undermine the usefulness of our own lab mice, and the risks we take when we do so.

If you put a rat on a limited feeding schedule—depriving it of food every other day—and then blocked off one of its cerebral arteries to induce a stroke, its brain damage would be greatly reduced. The same held for mice that had been engineered to develop something like Parkinson's disease: Take away their food, and their brains stayed healthier.

But Mattson wasn't so quick to prescribe his stern feeding schedule to the crowd in Atlanta. He had faith in his research on diet and the brain but was beginning to realize that it suffered from a major complication. It might well be the case that a mouse can be starved into good health—that a deprived and skinny brain is more robust than one that's well-fed. But there was another way to look at the data. Maybe it's not that limiting a mouse's food intake makes it healthy, he thought; it could be that not limiting a mouse's food makes it sick. Mattson's control animals—the rodents that were supposed to yield a normal response to stroke and Parkinson's—might have been overweight, and that would mean his baseline data were skewed.

Part 1: The unhealthy lives of industrialized lab mice
Part 2: The trouble with focusing so much research on one single mouse species
Part 3: Why the naked mole rat (and the Burmese python) can help