The point of averages is extremely important. I always hear from kids (and also some adults) that "so-and-so billionaire dropped out of college and look how successful they are!" Yeah, and some daily smokers live to be 100 years old. The point is that they are one in a billion in their luck. Though I will say, I got a good laugh about it once -- one time I explained this whole life expectancy thing to my senior students (12th grade), and a kid said "don't you have a master's degree?" "Yes," I said. "Well but...what if you walk outside right now and get hit by a bus?" With an n=1, I guess that would mean that having a master's degree makes bus-death more likely...
Good points, Greg. We need a sensible jargon for times when heaps of correlations point the same way. A common term is 'strong inference' which can imply the judgement of 'provisionally correct.' I suspect that would be a red rag to some bulls. My suggestion is 'PLAUSIBLE INFERENCE' Should anyone tut-tut about that term, then a suitable response might to ask them for empirical evidence that points to weak inference, or none.
I’d quibble with your comment that not agreeing with the evidence for smoking causing cancer is not a mistake. As you say it could be a semantic distinction- smoking won’t definitely cause cancer. But what good correlational data with the sort of analysis you suggest to eliminate third factors does is tell you the best estimate of the probability of a causal link. That’s all we have with any prediction, even that the sun will rise tomorrow. If someone chooses to ignore this sort of analysis for a particular topic that is definitely a mistake on their part.
The point of averages is extremely important. I always hear from kids (and also some adults) that "so-and-so billionaire dropped out of college and look how successful they are!" Yeah, and some daily smokers live to be 100 years old. The point is that they are one in a billion in their luck. Though I will say, I got a good laugh about it once -- one time I explained this whole life expectancy thing to my senior students (12th grade), and a kid said "don't you have a master's degree?" "Yes," I said. "Well but...what if you walk outside right now and get hit by a bus?" With an n=1, I guess that would mean that having a master's degree makes bus-death more likely...
Good points, Greg. We need a sensible jargon for times when heaps of correlations point the same way. A common term is 'strong inference' which can imply the judgement of 'provisionally correct.' I suspect that would be a red rag to some bulls. My suggestion is 'PLAUSIBLE INFERENCE' Should anyone tut-tut about that term, then a suitable response might to ask them for empirical evidence that points to weak inference, or none.
I’d quibble with your comment that not agreeing with the evidence for smoking causing cancer is not a mistake. As you say it could be a semantic distinction- smoking won’t definitely cause cancer. But what good correlational data with the sort of analysis you suggest to eliminate third factors does is tell you the best estimate of the probability of a causal link. That’s all we have with any prediction, even that the sun will rise tomorrow. If someone chooses to ignore this sort of analysis for a particular topic that is definitely a mistake on their part.