@amorphous-glob.bsky.social
But late to the party, but there's probably a case to make for using pooled SD with some algebraic tricks. E.g. if the % diff is D:
(Pre-post)/pre = 1-post/pre=D
Post/pre = (1-D)
Post = (1-D)pre
Post - (1-D)pre = 0
Now add some distributional assumptions and take expectations.
(Pre-post)/pre = 1-post/pre=D
Post/pre = (1-D)
Post = (1-D)pre
Post - (1-D)pre = 0
Now add some distributional assumptions and take expectations.
October 5, 2025 at 3:13 AM
But late to the party, but there's probably a case to make for using pooled SD with some algebraic tricks. E.g. if the % diff is D:
(Pre-post)/pre = 1-post/pre=D
Post/pre = (1-D)
Post = (1-D)pre
Post - (1-D)pre = 0
Now add some distributional assumptions and take expectations.
(Pre-post)/pre = 1-post/pre=D
Post/pre = (1-D)
Post = (1-D)pre
Post - (1-D)pre = 0
Now add some distributional assumptions and take expectations.
Reminds me of a line I read in a textbook once, "Everyone's a Bayesian in the design phase".
March 1, 2025 at 12:01 AM
Reminds me of a line I read in a textbook once, "Everyone's a Bayesian in the design phase".
Assuming it's not terrible to do, I'd crunch the forecasts out by hand to see if they agree with either stata or statsmodels.
(In my experience, it'd be neither and I'd have a small panic attack before realizing I forgot a negative sign somewhere)
(In my experience, it'd be neither and I'd have a small panic attack before realizing I forgot a negative sign somewhere)
February 21, 2025 at 12:43 AM
Assuming it's not terrible to do, I'd crunch the forecasts out by hand to see if they agree with either stata or statsmodels.
(In my experience, it'd be neither and I'd have a small panic attack before realizing I forgot a negative sign somewhere)
(In my experience, it'd be neither and I'd have a small panic attack before realizing I forgot a negative sign somewhere)
I reckon you've fit a time series model? It's been a spell, but if I had to hazard a guess the difference is probably in the implementation of how forecasted terms are fed back into subsequent forecasts (deffo if you have autoregressive terms).
February 21, 2025 at 12:41 AM
I reckon you've fit a time series model? It's been a spell, but if I had to hazard a guess the difference is probably in the implementation of how forecasted terms are fed back into subsequent forecasts (deffo if you have autoregressive terms).
IMO, I would love to see a JMP style graph builder. The flexplot in the visual modeler is great, but it it'd be nice to have some of that drag-and-drop functionality.
February 15, 2025 at 12:36 AM
IMO, I would love to see a JMP style graph builder. The flexplot in the visual modeler is great, but it it'd be nice to have some of that drag-and-drop functionality.