JASP Tutorial: Bayesian Binomial Test

In this video we explain how to do a Bayesian binomial test using JASP.
Download JASP at: jasp-stats.org...
Follow JASP on Twitter: / jaspstats
Follow JASP on Facebook: / jaspstats
Follow JASP on GitHub: github.com/jas...
JASP version used for video: 0.7.5.1
Voiced by Alexander Etz (blog: alexanderetz.com/, twitter: / alxetz )
Editing and animations by Ravi Selker (twitter: / raviselker )
Table of Contents:
00:14 - Load Data
00:54 - Analysis
02:21 - Results

[@Skydmig]

Thanks for the quick tutorial and thank you for the great piece of work with JASP.
Now I don't know if you read these comments, but it struck me that there are no labelling on the two boxes representing the data available and data selected for analysis.
This may well be intuitive "enough" for most people, especially those who have used SPSS in the past, but I believe adding some kind of labelling it is worth considering.

[@ericjanwagenmakers]

Dear StiffWood,
I am not 100% clear on what boxes you refer to. Can you mention a specific time segment where this is a concern?

[@takatakiful]

JASP > SPSS
Good explanation, too, thanks for the video!

[@Bioexploration360]

can we use this software for genome seq data analysis?.....

[@Jdonovanford]

What is the difference between BF_10 and BF_01?

[@toddweaver1390]

BF_01 is the ratio of (likelihood data given H0) / (likelihood of data given H1)
(null / alternative)
BF_10 is the ratio of (likelihood data given H1) / (likelihood of data given H0)
(alternative / null)
"What are the odds? A practical guide to computing and reporting bayes factors" by Andrew Jarosz and Jennifer Wiley in the Journal of Problem Solving, does an excellent job explaining this

[@derobbertdevries]

A BF_01 indicates how likely the observed data are under the null hypothesis compared to the alternative hypothesis, whereas BF_10 indicates how likely the observed data are under the alternative hypothesis compared to the null hypothesis. For instance, BF_01=5 would mean that the observed data are 5 times more likely under the null hypothesis, than under the alternative hypothesis. This is equivalently to say that BF_10 = 1/BF_01 = 1/5 = 0.2, which would mean that the observed data are 0.2 times as likely under the alternative than under the null hypothesis.

[@leondbleondb]

Can someone please interpret the result in context of the question if the person was guessing or not? As I understand it it is interpreted that the person was not guessing?

[@xnoreq]

The null hypothesis uses a test value of 0.5 which you can think of the probability of getting a correct answer is 1/2. That is guessing.
The BF10 was 0.059 which tells you the support for the alternative hypothesis .. which in this case is negative. If you inverted it you get the BF01 which was about 17. That means the null hypothesis (guessing) fits the data way better than the alternative.
Btw, the alternative hypothesis was that the test results were better than guessing, > 0.5.

[@leondbleondb]

Ah, I understand now! Thanks a lot for clarifying.

[@xnoreq]

You are very welcome.
I personally prefer log10 of the Bayes Factor. log10(BF) above 1/2 is moderate, above 1 is strong, above 1.5 is very strong and above 2 is decisive.
It's not just neat because of the easy to remember numbers but because the inverse is just a negation.
So in the above example log10(BF10) = -1.22 so strongly against the alternative. Or log10(BF01) = 1.22 so strongly for the null.

[@leondbleondb]

That is also a very good suggestion, I will use this. I haven't looked but any references you have for those? Thank you again.

[@xnoreq]

H. Jeffreys (1961). The Theory of Probability (3 ed.). Oxford. p. 432
You can also find the table in the Wikipedia article titled "Bayes factor".
It's also approximately the same categorization that JAGS uses. See in the video around 5:18. log10(3) is .47 which is close enough to the .5 steps.
These categories just give a rough idea anyway. Depending on your application you may require a different standard.