tag:blogger.com,1999:blog-3240271627873788873.post5186609722753366675..comments2024-03-26T06:46:11.752-04:00Comments on Doing Bayesian Data Analysis: Is it legitimate to view the data and then decide on a distribution for the dependent variable?John K. Kruschkehttp://www.blogger.com/profile/17323153789716653784noreply@blogger.comBlogger5125tag:blogger.com,1999:blog-3240271627873788873.post-6701378970092583552016-12-02T08:12:35.155-05:002016-12-02T08:12:35.155-05:00My view on this is here: http://simkovic.github.io...My view on this is here: http://simkovic.github.io/2016/11/21/With-pre-registered-fully-bayesian-analyses-against-the-pre-registered-replications.htmlmatushttp://simkovic.github.ionoreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-90948741903370841952016-11-20T12:44:11.905-05:002016-11-20T12:44:11.905-05:00If you have data to spare, holding out a test set ...If you have data to spare, holding out a test set is always s good way to sanity check your choices.Carlos Cinellihttps://www.blogger.com/profile/08852810848505628328noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-36885334111568419992016-11-17T09:53:47.906-05:002016-11-17T09:53:47.906-05:00In addition, through modeling, one can vary assump...In addition, through modeling, one can vary assumptions to see how robust the findings are. <br /><br />A similar argument could be made about prior-should these be registered as well ? Outside of a registered replication, many models could be fit to see how the evidence varies with varying priors, or whatever else assumption.<br /><br />Anonymoushttps://www.blogger.com/profile/13045015065678590524noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-82886035121550776812016-11-17T09:29:43.910-05:002016-11-17T09:29:43.910-05:00This comment has been removed by the author.Anonymoushttps://www.blogger.com/profile/13045015065678590524noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-23094928977136083942016-11-17T09:29:13.365-05:002016-11-17T09:29:13.365-05:00Hi John,
In my opinion, ensuring a well fitting mo...Hi John,<br />In my opinion, ensuring a well fitting model is very important. In the multilevel frquentist literature one often uses model selection to select the most appropriate distribution. If my data is zero inflated, do I really want a poisson model that does not take this into account? In bayes, LOO can be used to produce a measure of fit, while posterior predictive checks can provide visual checks.<br /><br />Like all things, this approach can be abused. For instance, selecting a distribution in which some interval now excludes zero.<br /><br />I end with stating, strongly, that I really think researchers should spend more time modeling their data.Anonymoushttps://www.blogger.com/profile/13045015065678590524noreply@blogger.com