Post date: Apr 20, 2019 4:26:45 PM
<<Nick Friedenberg wrote>>
I want to estimate the number of dead bats per megawatt of wind power across the US from a convenience sample of post-construction reports with inconsistent methods. Everyone I talk to goes right to the most expensive approach possible, which is to obtain the original data for each project and standardize them all with a bunch or corrections. Even then they would do regional stratification and resampling and do on and so on. None of this is a bad idea, but to me it screams false precision.
All this is prelude to saying I may finally want to sit down and study imprecise methodology for this question. Can I ask you for a short reading list?
<<Scott replied>>
I probably need to chat with you about this. What's expensive about the most expensive approach? What is the screaming false precision you are worried about?
I don't think there is a reading list for what you want. In his latest book (which I've misplaced) Sieber still believes everything he believed two decades ago.
We have addressed--though maybe not entirely solved--questions about how to handle
Incertitude (intervals) in observations,
Smallness of sample sizes,
Expert elicitation of values,
Non-random, hotspot spatial sampling
But we are only now starting to think about what I call "basal uncertainty" in translating gradations to binary or count values. And general questions about non-randomness and non-representativeness and other bad designs are still on our yeah-let's-get-to-these-real-soon list. We're thinking of a big proposal on the topic, https://sites.google.com/site/crappydata/.