Palko points us to this interesting point from Josh Marhsall:
I [Marshall] am always struck by, amazed at how much of public health work involves not technology but methodicalness and record keeping. In purely technological terms much of it could have been done 100 years ago or, in outlines at least, 500 years ago. Phones and texts and emails obviously provide a critical new tool, allowing public health officials to remain in regular daily contact with hundreds or thousands of people currently self-quarantined across the country. But at its heart it’s an elementary process: find anyone who has tested positive for the infection, track down everyone with whom they had significant contact and get those people to isolate themselves (usually at home) from everyone else for 14 days.
This seems like a good insight, and perhaps it can be generalized to say that 90% of statistics is measurement, and 90% of what’s left is avoiding and accounting for selection bias. Meanwhile, we spend most of our time talking about the remaining 1%.
But that’s a bit too simplistic on my part. For example, we correct for selection bias and random variation in surveys by using MRP, but for that we need probability models, prior distributions, efficient computation, Stan, etc….
But, to get back to Marshall’s point, the first step is to measure and record.