EJ
About
- Username
- EJ
- Joined
- Visits
- 2,532
- Last Active
- Roles
- Member, Administrator, Moderator
Comments
-
Yes you can still compute the BF, that is, assess the relative predictive adequacy of the hypotheses. But what it means when the model is wildly misspecified is anybody's guess. See for instance http://www.thefunctionalart.com/2016/08/download-datas…
-
Dear JASPUser, I would have to ask the team member responsible for this, or you could check our code, but it seems to me that "t.test" in R uses the Welch test by default ("var.equal = FALSE"). You could check their documentatio…
-
Hi mspr, * The BF10 =1 because the "H0" in the comparison is the model on the first row -- in other words, the model is compared against itself. * We have some post-hoc correction for pairwise comparisons in the ANOVA. It is based on an a…
-
If I'm not mistaken this has already been implemented in the developer's version! Cheers, E.J.
-
Dear Alekhya, Thanks for your question. This is tricky! Fundamentally, there is nothing holding you back from using the regular model and denoting the missing conditions as missing data. But what to do in your specific case? Basically, my advice is…
-
Dear mspr, Yes this is possible. If I recall correctly, the Jamil paper also discussed a data set on professions held by fathers and their sons, and that was a pretty large table. Cheers, E.J.
-
We are working on integrating JASP completely with R, but it may take a while.
-
Hi Jan, I'll forward this question to our logistic regression expert, but I think the expert will probably want to know what you mean exactly. In logistic regression, your dependent variable needs to be binary; for the predictors there is no such r…
-
Hi Elien, This topic occasionally pops up on this forum, so searching for the relevant terms will bring up some relevant posts. As you suggest, the BF inclusion is the change from prior to posterior odds, where the odds concern all the models with …
-
Dear soulclimber, So Attention is a single-number, between-subjects variable? What people sometimes to is discretize the Attention variable so that they can look for an interaction in an ANOVA. Clearly this discretization is somewhat arbitrary and …
-
Dear powg, The standard output allows you to compare a main-effects only model to a model with the interaction included (so only two models are involved). In the analysis of effects, the inclusion BF compares all models with a particular term to al…
-
The Tversky & Kahneman formulation expresses the odds, not the probability. If you write that equation out in full (including the = sign) you will see that the posterior odds is 0.7058824. To get to the associated posterior probability you need …
-
That is correct E.J.
-
Hi Richard, You can multiply the Bayes factor by the prior model odds (p(H1)/p(H2)) and this then gives you the posterior model odds. The BF gives you the extent to which the data should change your beliefs. So yes, the prior model odds matter, but…
-
Yes, the Bayesian account for multiplicity is entirely in the prior model probabilities. To respond to your question: "Are you saying that we only need to adjust for multiplicity by correcting the prior odds?" Yes "In other words, …
-
I recommend the following two: 1. Etz, A., Gronau, Q., Dablander, F., Edelsbrunner, P., Baribault, B. (in press). How to become a Bayesian in eight easy steps: An annotated reading list. For the Bayesian Statistics special issue in Psychonomic Bulle…
-
I'd simply use the standard output from JASP. E.J.
-
Dear Vadim, * You have BF10 = 0.175. This means that that BF01 = 1/0.175 = 5.71, meaning that the data are about 5.7 times more likely under H0 than under H1. This is support in favor of H0. I should not that BF10 = 0.99 would also have constitute…
-
For each predictor, you can look at the posterior inclusion probability. What we will add in the next version is also the prior inclusion probability (right now this is only shown in a plot); the inclusion BF is the ratio between the posterior incl…
-
You can try it out. I recall that the results are slightly different, and this is because the underlying models are not identical (i.e., the prior predictions about to-be-observed data are not the same).
-
The posterior odds are for the effect being present, where the prior odds have been corrected for multiplicity (the post-hoc character of the test). The BF01,U gives the evidence in the data without correcting for multiplicity (the U stands for unco…
-
Yes that makes perfect sense! (note that the BF says something about the evidence from the data -- if you want the probability of a hypothesis being true then you need to take the prior model plausibility into account). So the safest conclusion is t…
-
Hi Mirjam, This looks like an issue for our GitHub page. I'll create the issue for you. Cheers, E.J.
-
Are you using a binomial model, and do you want to use a beta prior? You can set the "a" and "b" parameters of the prior distribution. If you start from a uniform distribution (big if) then, after seeing 9/10 "successes"…
-
Hi Andrew, Right now, you can't. In the future we aim to expand the output , as we have recently done for linear regression (up to 0.8.5., we only reported BFs; as of 0.8.6, we also have posterior distributions). Richard may now how you can use the…
-
You should check yourself, there is no gold standard; it also depends on the numbers themselves -- you may not care whether the BF is one or two billion. I am not sure about the difference between "samples from the posterior" and "i…
-
Thanks for bringing this to our attention. We now have a team member working to upgrade this functionality. Cheers, E.J.
-
Yes, it is based on conting! It uses Reversible Jump MCMC, and under uniform prior probabilities for the models, the number of visits relates directly to the posterior probabilities (and this also given the BF). E.J.
-
Hi Andrew, There's a series of papers on planning experiments until a BF crosses a bound. If you look on my website you'll see a few with Felix Schoenbrodt. This is the latest one: https://psyarxiv.com/aqr79 Cheers, E.J.
-
For many regression/ANOVA analyses the integral only goes over one parameter, so full MCMC is not needed. I am not sure what the BayesFactor package does for repeated measures ANOVA though -- should be in the documentation. But yes, BayesFactor asse…