The most interesting hypotheses are the ones that describe a causal relationship. If we know what causes an effect, we can predict it, influence it, better understand it. So, how do we identify a causal relationship? Well, it was David Hume. With a little help from John Stuart MIll, who first listed the criteria that we still use today. These are the four essential criteria. Number one, the cause and effect are connected. The first criterion means there has to be a way to trace the effect back to the cause. If a patient was not exposed to a virus, then we can't argue that the virus caused the patient's death. Number two, the cause precedes the effect, I hope this is obvious. Number three, the cause and effect occur together consistently. This means cause and effect should go together, or covary. When the cause is present, we should see the effect, and if the cause is not present, then the effect should be absent. If the cause influences the effect to a certain degree, then we should see a consistently stronger or weaker effect, accordingly. Criteria number four, alternative explanations can be ruled out. Okay, so let me illustrate these criteria with an example. Suppose I hypothesize that loneliness causes feelings of depression. I give some lonely depressed people a cat to take care of. Now, they're no longer lonely. If my hypothesis is correct, then we would expect this to lower their depression. The cause and effect, loneliness and depression, are in close proximity. They happen in the same persons, and fairly close together in time. So we can show they are connected. The cause, a decrease in loneliness, needs to happen before the effect, a decrease in depression. We can show this because we can control the presence of the cause, loneliness. The cause and effect should occur together consistently. This means that less loneliness should go together with lower depression. I could find a comparison group of lonely, depressed people that do not get a cat. Since the cause is absent, their loneliness doesn't change. There should be no effect. Now, this is all easy. The real difficulty lies in the last criterion. Excluding any alternative explanations. Other possible causes. Let's look for an alternative explanation in our example. Maybe the increased physical activity required to take care of a cat, actually caused lower depression, instead of the reduction in loneliness. Alternative explanations form threats to the internal validity of a research study. An important part of methodology is developing and choosing research designs that minimize these threats. Okay there's one more point I want to make about causation. Causation requires correlation. The cause and effect have to occur consistently. But correlation doesn't imply causation. I'll give you an example. If we consistently observe aggressive behavior after children play a violent video game. This doesn't mean the game caused the aggressive behavior. It could be that aggressive children seek out more aggressive stimuli. Reversing the causal direction. Or, maybe children whose parents allow them to play violent games aren't supervised as closely. Maybe they're just as aggressive as other children. They just feel less inhibited to show this aggressive behavior. So remember, correlation does not imply causation.