It's becoming increasingly concerning to me that people are staking out philosophical and mathematical positions (for political reasons) that are increasingly incompatible with a healthy public discourse on data. https://twitter.com/seerutkchawla/status/1345813080782082062

Normally, I'd ignore something like this but it's a key part of my intellectual mission in being on the internet.

Before we begin, let me clearly state that I don't agree with the particular narrative presented so technically I agree with @seerutkchawla that this particular narrative should be rejected. However, I think that the reasoning is deeply flawed in a concerning way.

ONE. Framing the use of "narrative" as somehow in opposition to the acquisition and analysis of "facts" is wrongheaded in my opinion.

Consider the scientific process. You have to believe that the scientist constructed the experiment and apparatus without bias, that the apparatus remained completely under their control for the duration of the experiment, that observations were obtained at random or...

observations from opposing treatment groups were appropriately matched, that the analysis presented is the totality of the analyses performed (no p-hacking), and that the review process proceeded without bias. In short, you must assume a narrative of impartiality.

Failure to appreciate that scientific facts assume a certain narrative about the origin of such facts leaves us open to being duped when the narrative fails to be accurate. This is the type of error that people make when they assume everything written in a book must be true.

TWO. I would argue narrative is essential to causal interpretation of data.

Causal reasoning is based on narrative. In order to interpret "A causes B" precisely you need to set out a narrative about how A leads to B. If you don't set out a narrative that just means "A causes B" is partially undefined (and is dependent what's going on in your data.)

As a statistician, failure to adequately specify a narrative is a very common weakness in purely associational studies of human data because associational investigations are inherently non-narrative.

With such analyses, you often end up knowing a lot about how certain variables are related to other variables with no clear idea of what the people inside the dataset are experiencing as they move through the system.

All I'm saying is NARRATIVES ARE IMPORTANT. We scientists can learn a lot from the humanities! Thanks for coming to my TED talk!