Feature request: Bayesian correlation matrix – nice to add: a tabular Bayesian correlation matrix with specific example

[Golden-Retriever-1]

Adding a tabular Bayesian correlation matrix to the analyses options as used to from factor analyses and PCAs * Enhancement: Visual analyses of a simple Bayesian correlation matrix made possible * Purpose: Showing a tabular correlation matrix of Bayesian correlations to the users as the users knows the tabular format from their statistics lectures, text books, etc. * Use-case:Several in data analysis. Tabular correlation matrices give the user a very good overview over pairs of correlations. Example Item 1 Item 2 Item 3 Item 4 Item 5 Item 6 Item 7 Item 8 Item 9 Item 1 Item 2 Item 3 Item 4 Item 5 Item 6 Item 7 Item 8 Item 9 **Is your feature request related to a problem? Please describe.** It eases data analysis and data exploration when using Bayesian correlation matrices in general. It is a well-known format and is currently not available for Bayesian correlations. **Describe the solution you'd like** The variables on the rows and columns of a correlation matrix as usual. Inside the cells several parameters can be displayed per correlation matrix such as Pearson's r with the BF below, and the lower and upper 95% interval below. I am sure it is easy to find a way to display a Bayesian tabular correlation matrix. Example Item 1 Item 2 Item 3 Item 4 Item 5 Item 6 Item 7 Item 8 Item 9 Item 1 Item 2 Item 3 Item 4 Item 5 correlations in here Item 6 Item 7 Item 8 Item 9 **Describe alternatives you've considered** There is currently no alternative in JASP to display a Bayesian correlation matrix for only this purpose. The plain and bare tabular format which is no matrix is good but does not support visual analyses. The format of the correlation matrix as value-based result is missing. **Additional context**

[boutinb]

@vandenman what do you think about this request?

[vandenman]

@Golden-Retriever-1 I'm not sure how this is different from the current Bayesian correlation, could you provide another example?

Currently, we provide these two views of the correlation matrix:

Would you like the Bayes factors to be removed from the top screenshot?

[Golden-Retriever-1]

It provides the standard layout format from textbooks and aggregates the information of one correlation in a lovely cell!

Am 02.10.2020 um 14:44 schrieb Don van den Bergh notifications@github.com:

 @Golden-Retriever-1 I'm not sure how this is different from the current Bayesian correlation, could you provide another example?

Currently, we provide these two views of the correlation matrix:

Would you like the Bayes factors to be removed from the top screenshot?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[vandenman]

something like this?

you can obtain something like that by unticking Report Bayes factors:

[Golden-Retriever-1]

It would be helpful to join the information for visual analysis, as eg correlation above, Bayes factor below in the format below.

Eg

.60 3.567

Am 02.10.2020 um 16:51 schrieb Don van den Bergh notifications@github.com:

 something like this?

you can obtain something like that by unticking Report Bayes factors:

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[vandenman]

So how is your request different from this table?

[Golden-Retriever-1]

Yes, it is. It keeps the structure for visual analyses that most know from their textbooks. In one cell, there is way to display more than one information by applying a structural approach. I am sure a key can help to read the table when adding more than the 2 suggested values.

Am 05.10.2020 um 09:11 schrieb Don van den Bergh notifications@github.com:

 So how is your request different from this table?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[vandenman]

I'm afraid I don't understand what you mean, could you provide a screenshot or word/ excel table of how you'd the output to be arranged?

[Golden-Retriever-1]

Simply keeping the layout of the usual visual representation and aggregating the different values one above each other in the places of the table:

Instead of .70 as Pearson’s r

You could write .70 as Pearson’s r above the value for the BF10 while keeping the visual structure which really most of the users are used to from their statistical textbooks. The current layout is user-unfriendly.

On 5 Oct 2020, at 11:39, Don van den Bergh notifications@github.com wrote:

I'm afraid I don't understand what you mean, could you provide a screenshot or word/ excel table of how you'd the output to be arranged?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/jasp-stats/jasp-issues/issues/986#issuecomment-703520712, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOKS3BJFXSY6FDGWCDIBLDTSJGH4FANCNFSM4RV7ESUA.

[AlexanderLyNL]

I think that what's requested is something that is implemented already. The last tweak seems to be a bit of a nuance difference and perhaps a difference in opinion regarding what's user-friendly. The current implementation allows users to only display the summary statistic without the Bayes factor, or only the Bayes factor, or both. I think it's reasonable to consider this issue solved.

[Golden-Retriever-1]

Your resistance to keeping unusual formats with a wide difference to textbooks! Enjoy.

Am 05.10.2020 um 12:51 schrieb AlexanderLyNL notifications@github.com:

 Closed #986.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[AlexanderLyNL]

I don’t see it as a matter of resistance, and I'm happy with the feedback. From what I gathered we’ve implemented exactly what you requested. A table with Pearson’s r above the Bayes factor can be requested as shown in

https://github.com/jasp-stats/jasp-issues/issues/986#issuecomment-703445379

[Golden-Retriever-1]

I just really wonder about not using the usual textbook format which is why especially beginners and learners might have difficulties in reading the current outputs! No all are really proficient which all different types of aggregating data which is why I suggested it. I extra draw a sketch of it. The most easy programming solution is often not the best visual representation of tabular data.

It could look like this:

On 5 Oct 2020, at 13:32, AlexanderLyNL notifications@github.com wrote:

I don’t see it as a matter of resistance, and I'm happy with the feedback. From what I gathered we’ve implemented exactly what you requested. A table with Pearson’s r above the Bayes factor can be requested as shown in

986 (comment)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[vandenman]

@Golden-Retriever-1 could you use the GitHub website to show us what you have in mind? Your last post doesn't show any images in case you attached any:

We prefer not to use any unusual format and would be happy to implement your request if it's reasonable, but we are having a really hard time understanding what your feature request is about.

[Golden-Retriever-1]

Okay, I have answered via email maybe the picture file was not added.

Am 05.10.2020 um 14:07 schrieb Don van den Bergh notifications@github.com:

 @Golden-Retriever-1 could you use the GitHub website to show us what you have in mind? Your last post doesn't show any images in case you attached any:

We prefer not to use any unusual format and would be happy to implement your request if it's reasonable, but we are having a really hard time understanding what your feature request is about.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[Golden-Retriever-1]

What about this link? I can later on upload the file to Github.

https://1drv.ms/u/s!AjVtS8ZiRDYiokhxlNn2QMIZGQ9n

Am 05.10.2020 um 14:09 schrieb Daniela Schmid daniela.schmid64@icloud.com:

 Okay, I have answered via email maybe the picture file was not added.

Am 05.10.2020 um 14:07 schrieb Don van den Bergh notifications@github.com:

 @Golden-Retriever-1 could you use the GitHub website to show us what you have in mind? Your last post doesn't show any images in case you attached any:

We prefer not to use any unusual format and would be happy to implement your request if it's reasonable, but we are having a really hard time understanding what your feature request is about.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[vandenman]

Thanks, that image came through. I'm also pasting it here for future reference:

[Golden-Retriever-1]

Here is the COMPLETE(!) scan – I can view the picture at github. It should be available to you as well:

Am 05.10.2020 um 14:17 schrieb Don van den Bergh notifications@github.com:

 Reopened #986.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[tomtomme]

@Golden-Retriever-1 As far as I can see, there is only one difference between JASP and your sketch. The legend or key as you call it is outside of the table in your sketch, instead of integrated into the table as in JASP. The JASP behaviour duplicates the legend multiple times. This could be considered a drawback.

However, general advice for plots also say, that you always should include the legend / key within a plot, as near as possible to the data. This increases readability, because the eye does not have to jump such a large distance between legend and plot. This general advice is certainly applicable to tables as well. If you do not like the duplication of the legend/key, you can activate the pairwise table.

Thus I, as a stats teacher, would vote against an external legend / key, even if some textbooks show this differently.