speaker-0: One thing we like to say is that even if you knew ahead of time what speaker-1: model you wanted to end up with, you'd still want to be fitting simpler models to understand what you got out of fitting the more complicated model. And you'd also want to fit a more complicated model just to see that you didn't get more out of that. Yeah. speaker-2: And a key part speaker-0: So Bayesian data analysis like back when it came speaker-1: out in nineteen ninety five was kind of revolutionary and I called it basing data analysis rather than basing inference. speaker-3: Because we our big thing was that there are three steps. You put the model, you fit the model, and then you check the model. And then you can explain. speaker-1: Build Bandit. Didn't talk about building the model and they didn't talk about speaker-0: Before that people really speaker-3: Yeah. speaker-0: The th Bayesian speaker-2: workflow or statistical speaker-1: workflow goes beyond what we had in Bayesian data analysis because that's speaker-3: Because you fit the you build the model, you f you fit it and you check it, maybe you expand it. speaker-2: all about. a lot of workflow. speaker-1: is going from model to model. And a lot of the operations statistical workflow can be framed fitting additional models. Even gathering new data could be thought of as fitting a new model to data in the sense that if I have a bit of data and I fit a model, then I get more data and fit a bigger model. You could think of the original version simple model fit to all the data. speaker-2: and LUCK speaker-3: That we do into speaker-0: As speaker-3: It just didn't use the new stuff. speaker-2: flow has Like, how do you like. speaker-1: If you have an iterative algorithm, how long do you run it? So some people say, ⁓ well just to be safe, run it speaker-3: Everything overnight. speaker-2: Well, that doesn't really work with workflow. speaker-1: We'll reduce the number of model we learned so much by fitting model after model after model. And then being able to simulate from a generative model. speaker-3: means that you can speaker-0: Well let's say this. If you think about speaker-2: like I like to say that like all statistics. speaker-1: is frequentist including over speaker-3: Bayesian because we're averaging the prior, which is a a frequency. Thing Ben Goodrich doesn't like when I say this, but I I I think this is basically right. speaker-0: But if you think about speaker-2: ⁓ in old school textbook what speaker-3: properties. There's been a lot of people have can't do that, so they do speaker-2: papers for people. Papers by speaker-3: I like Stanford types. speaker-1: Which will have tables at the end with a bunch point nine three between point nine three and point nine six, which are ninety five percent interval. speaker-3: Where supposed to demonstrate the coverage of under various conditions. speaker-0: But what's really c that you don't need the I mean, the theory helps because gives you speaker-2: What? speaker-1: Some mathematical intuition. speaker-2: but. speaker-0: You don't need the theory and you don't actually speaker-1: Actually like if you have a ⁓ if you're gonna use a method, a model, traditionally you would say, Well let's use a method or a model that's been published somewhere and then let's hopefully somebody wrote a paper. speaker-2: need that published paper. speaker-3: Demonstrating that it has good frequency properties. speaker-0: Now speaker-1: Well what do you do? Well I'm gonna fit a model and it's a new model, never got created. speaker-2: before I could. speaker-3: Can it reconstruct how well can the it parameters can be be estimated? And speaker-0: So I can build this. speaker-2: Cool. ⁓ speaker-1: Giving me more confidence. speaker-3: in my own work. So we try to demonstrate that to people. speaker-1: Is not just get the good speaker-3: Just to answer, but to get an answer that you know is good or you have some degree of confidence that it's good.