EJ
About
- Username
- EJ
- Joined
- Visits
- 2,598
- Last Active
- Roles
- Member, Administrator, Moderator
Comments
-
Hi Michael, Correct -- the usual BF involves a null hypothesis and is therefore a test of presence/absence, not one of magnitude. Cheers, E.J.
-
We have a paper about effect size for ANOVA (minor revision); For the t-test, the standard output gives the posterior distribution for Cohen's delta (on the population level). The interpretation in terms of what is large won't differ (and depends on…
-
Good point. We hope to offer a generic solution for this issue in the future. For now, you can of course look at my own papers, and those of Dienes, Rouder, and Morey, for instance. A google search should also be effective. Cheers, E.J.
-
Hi Andrew, There is this paper: http://pcl.missouri.edu/node/133 We are currently revamping the Bayesian linear regression, and a new release (a few weeks away) will present more cool options. So stay tuned. Cheers, E.J.
-
Hi Michael, The updated priors are on the model parameters. But this happens implicitly, that is, the BF you obtain is the same one that you could also have obtained in case you had been able to properly update the prior distributions on the model p…
-
That would be your best bet. You could add my reply and ask whether he agrees.
-
Hi Kevin, You can run this past Richard Morey to be sure, but if you compare the scale for the ANOVA to that of the t-test you see that they differ by a factor of 1/2. So if you take the t-test scales and divide by 2 you should be good. Again, you …
-
Hi Michael, This is tricky. One thing you could do is compute the BF for the initial sample, and then for the complete sample; dividing them out yields the BF for the replication sample after updating the priors with the initial sample. This only w…
-
So when you look at BF01 and the "H0" is the null model, values higher than 1 are evidence for the null. So the BF01 = 43.15 is evidence for the null model that has no predictors. But the null model is not the best model -- this is the mod…
-
Hi Caeline, Yes, that's correct. There are no verbal guidelines for BF10's of 1 million. You might invent your own category -- I usually call BF's in that range "overwhelming". In general though, the verbal labels are just heuristics, and …
-
If you have a large effect, or a small effect but with large sample size, BFs can be huge. Even with N=1, you can get a BF of infinity. Example: toss a coin; H0 says theta = 1 (i.e., coin has heads on both sides); throw the coin once and observe tai…
-
Yes, you are correct. The "across all models" BFincl is also not in favor of adding the interaction, although the strength of evidence is not as compelling as for the matched models. Perhaps the interaction term features in a few models wi…
-
Do the best model has Position and irrelValue. The second best has only Position, and the third best has only irrelValue. The first model with an interaction enters at place 5, and is a factor of 3.293 worse than the best model. This is echoed by t…
-
The "Baws" method proposed by Sebastiaan should give the same result as "Analysis of effects" with the "matched models" option. Do you get a different outcome? E.J.
-
I'll pass the message on to our network expert! E.J.
-
The error is so small that I would not even report it. E.J.
-
Hi Ester, The one-sided BF is calculated by departing from the two-sided BF and then adding a correcting factor. The correction factor is close to its maximum value, which might produce the problem. I'll bring it to the addition of the programming …
-
I need a little more context, perhaps a concrete example?
-
Hi Chris, The desirable properties of the Cauchy hold for any scaling. The value of 1 was suggested by Jeffreys but this is not a principled point. The value of 1/sqrt(2) was suggested in the BayesFactor R package to be more reasonable (i.e., more …
-
Hi Anoop, One of our recent multi-million $ grants is on applications to medicine; so yes, those analyses are definitely on the agenda! Cheers, E.J.
-
Dear Clarisse, OK, let's tackle these one at a time: (Quote) Yes. (Quote) BF10 is just 1/BF01, so they provide exactly the same information. If BF10 is 0.1, say, it feels awkward to say "the data are 0.1 times more likely under H1 than under…
-
Hi Merel, I don't think you can, at least not right now. If the five conditions were between-subjects you'd just have different sample sizes in each of the conditions. But here you have a within-design, and this complicates things. From a Bayesian …
-
Hi Aram, I think this is one of the few issues where Richard and I have a different opinion. I would argue that there are multiple models to consider, and it is best to average over them. In JASP, you can do this by ticking "Effects", and…
-
https://twitter.com/AlexanderLyNL/status/918197338841190400 https://twitter.com/AlexanderLyNL/status/918195429652750337
-
Right now you'll have to create a separate column, as we do not have filtering functionality (yet). I recall that Alexander had a demonstration how this could be easily done, let me ask him... E.J.
-
Hi wendt, For a detailed explanation see for instance, on my website, the paper Wagenmakers, E.-J., Love, J., Marsman, M., Jamil, T., Ly, A., Verhagen, A. J., Selker, R., Gronau, Q. F., Dropmann, D., Boutin, B., Meerhoff, F., Knight, P., Raj, A., v…
-
MSB is spot on. See also, on my website: Ly, A., Etz, A., Marsman, M., & Wagenmakers, E.-J. (2017). Replication Bayes factors from evidence updating. Manuscript submitted for publication. URL: https://psyarxiv.com/u8m2s/ E.J.
-
The posthocness expresses itself through the prior model probabilities. The BF remains the same. Cheers, E.J.
-
Not yet. Perhaps he will respond when you send him a personal Email? Cheers, E.J.
-
Hi Anja, Usually you get this error whenever you try to estimate a model that includes interactions but not the corresponding main effects. If you just drag the variables into their boxes this should not happen. So my first question would be, when …