I want to offer positive reinforcement for your model-generating algorithm, especially the “notice that I am confused by my model’s retrodictions, question everything” step. More posts like this, please!
On this particular topic, I want to emphasize the massive amount of uncertainty about the numbers you get; the factors you can’t control for could make a 10x difference on the x-axis, and of course the data we care about (organization survival in the future) need not resemble historical data very much at all. The model is essentially only good for an outside view / sanity check. (Of course, that’s the best we can get in complicated cases, and it’s a good improvement on ignorance priors! I’m just saying to everyone, please don’t go around citing “22.8% chance of institutional survival for 415 years” in cryonics debates.)
I want to offer positive reinforcement for your model-generating algorithm, especially the “notice that I am confused by my model’s retrodictions, question everything” step. More posts like this, please!
On this particular topic, I want to emphasize the massive amount of uncertainty about the numbers you get; the factors you can’t control for could make a 10x difference on the x-axis, and of course the data we care about (organization survival in the future) need not resemble historical data very much at all. The model is essentially only good for an outside view / sanity check. (Of course, that’s the best we can get in complicated cases, and it’s a good improvement on ignorance priors! I’m just saying to everyone, please don’t go around citing “22.8% chance of institutional survival for 415 years” in cryonics debates.)