The tradition they now represent has centered its chief inquiries around the two humble questions, “What do you mean?” and “How do you know?”
Herbert Feigl, “Logical Empiricism”
Christian Frankel sent me this quote the other day knowing it would remind me of Wayne Booth’s mythical Oxford tutorials. As a statement of the underlying attitude of the logical positivists, it’s quite nice, and I remember Steve Fuller once telling me that, whatever we may think of positivism as a philosophy of science, positivists were often excellent dissertation supervisors. Indeed, it’s not difficult to see how it might be helpful if your supervisor patiently and insistently asked you mainly, “What do you mean?” and “How do you know?” Imagine having someone give every single one of your paragraphs that treatment!
I’m going to spend some time this weekend reading Feigl and relating his ideas to the issues I’ve been raising this week. Have a great weekend also!
I’m reminded of the two general statistical questions:
1. What would you do if you had all the data?
2. What would you do if you had no data?
For example, suppose a study is performed on two groups of patients: one group gets the control and the other gets the treatment, and the treatment group has a higher five-year survival rate, but there are problems with missing data, selection into one group or another, measurement error, etc.
Question 1 asks: if you could apply both treatments (that is, observe the potential outcomes under both conditions) for the entire population, then what would you want to know? Presumably it would not be five-year survival rates; that’s just a convenient thing to measure. But how would you summarize such results? What would you do with them? The point here is to separate the “science” or “decision” questions (what is going on in an underlying sense, and what would you do about it) from the “statistics” or “data” questions (how can we analyze existing data and gather new data to aim to answer the deeper questions). So much of statistics starts with the data, and it can be helpful to step back and ask what we are trying to learn.
Question 2 asks: What would you do, what would you believe, without looking at these data? From a Bayesian perspective, this is the question of what is your prior, but if we forget about Bayes for a moment, it’s a more general question that reminds us that we would be forming beliefs and making decisions absent this particular study. Such an exercise can help us better see how the data at hand fit into the larger questions.