[Feature Request]: General DV descriptives in ANOVA

[wpsilvestre]

Description

Adding, SEM, SD, and CV to ANOVA descriptive statistics table

Purpose

Adding the standard error of the mean, standard deviation, and coefficient of variation to ANOVA descriptive statistics, bur referring to the analysis as a whole, not individual (or level-data) results

Use-case

This feature would benefit any user that works with Jasp

Is your feature request related to a problem?

No

Is your feature request related to a JASP module?

ANOVA

Describe the solution you would like

Having the option of showing the standard error of the mean, standard deviation, and coefficient of variation to ANOVA descriptive statistics table (referring to the analysis SEM, SD, and CV - not level or individual parameters)

Describe alternatives that you have considered

I tried using the 'descriptive statistics' module, but these parameters would be related to the ANOVA analysis, rather than being conducted separately, which makes the software used truncated.

Additional context

Showing this information (SEM, SD, and CV to ANOVA descriptive statistics table, referring to whole analysis) would ease the interpretation of the results and helping users to add meaningful data to the results of the computed statsitics with Jasp.

[tomtomme]

Interesting. I teach ANOVA since >10 years and never thought about this one. Can you explain why the non-split information about the Dependent Variable is relevant in the ANOVA context? I am just curious.

[JohnnyDoorn]

@wpsilvestre can you give a concrete example? I would also think that if you want descriptives split by your factors, you can use the descriptive statistics option in ANOVA, and if you want the non-split descriptives you can use the descriptive module.

[wpsilvestre]

Dear @tomtomme, As far as I know, having this information (SEM, SD, and, especially, the CV) helps understand the variability of the dataset analysis and may be an interesting parameter to show with the anova table. But, please, correct me if I'm wrong.

@JohnnyDoorn, I am aware of the descriptive module, but having these values automatically calculated with the respective dataset may be an interesting addition to the anova module. For example, if I want to know the CV associated with the analysis, having this option with the anova table (or a general descriptive) would be interesting (and without the need to access the descriptive statistics module and risk choosing a wrong variable/dataset). Currently, the anova module shows the SEM, SD, and CV of each level in the descriptive table. I believe that bringing this information for the whole analysis would help the people using Jasp.

I hope I was clear. I am at your disposal to discuss and learn more. Thanks.

[tomtomme]

@wpsilvestre

this information helps understand the variability of the dataset

But why would I be interested in the variability of the dependent variable (DV) alone? I think this is the crucial question to answer. My understanding of ANOVA is: I have a DV and a IV (independent). I suspect a causal relationship between the two / aka a difference between the groups made by the IV. Thus I have a bivariate analysis. Now you want to look into the DV alone, unsplit, to assess its variability. This would be an univariate analysis that seems disconnected to the original goal stated above. Because I cannot learn anymore about the relationship of DV and IV anymore. Am I wrong? Do you have an example that shows that the DVs variability alone is important for the ANOVA?

[wpsilvestre]

@tomtomme, I agree with all you said, but if the dataset has a large CV, this means the data has a wide variability (and I am not talking about homoscedasticity or residuals). Data with large CV may mask treatment effects, and assessing this parameter may help the readers (and who is analyzing the data) to assess whether the data has a large variation (and more samples/runs could/would be performed, or even the quality of the analysis), or it is within a reasonable variation. Am I wrong in my assumption here?

[tomtomme]

There we have it! Sounds like a logic assessment to me! Thx for the insight. I never thought about it this way.

[wpsilvestre]

@tomtomme, I appreciate the patience and I'm happy to help. Would it be feasible to implement such a feature (and the option to have it or not shown in the results) in the anova module of Jasp?

[tomtomme]

I hope so. But I can give no timeline.

[JohnnyDoorn]

Hi @wpsilvestre,

I'm a bit hesitant to add another checkbox for descriptives, since we already provide descriptives for each level of the predictors, and if the user wants overall statistics for the DV this is straightforward to do through the descriptives module. One important thing we need to look out for is to not clutter the interface too much with redundant options. For instance, we also offer Q-Q plots of the DV in descriptives, but only Q-Q plots of the residuals of the model in ANOVA (and not Q-Q of the DV). So I would be highly in favor of having mores analysis-specific output in the ANOVA (such as factor level descriptives), while keeping the general descriptives stuff in descriptives (such as variable level descriptives).

Kind regards Johnny

[tomtomme]

@JohnnyDoorn Yes, this should definitely not be implemented via another checkbox. But we could just add an "overall" or "total" row to the descriptives table by default, similar to the "total" row in contingency tables. But I can understand if you still do not want to duplicate this stuff, since one can get it from descriptives.

[JohnnyDoorn]

That's a great suggestion @tomtomme, I'll look into it!

[wpsilvestre]

@JohnnyDoorn and @tomtomme , thanks for the feedback! The idea @tomtomme brought is excellent.

I have another doubt/request regarding the post hoc tests of anova. I can explain here, or would it be better to open a new topic?

Thanks again for the attention.

[JohnnyDoorn]

@wpsilvestre I think it's best to make a separate feature request or bug report for the post hoc test - please tag me in it, then i can take a look immediately!

[tomtomme]

@JohnnyDoorn Yes, this should definitely not be implemented via another checkbox. But we could just add an "overall" or "total" row to the descriptives table by default, similar to the "total" row in contingency tables. But I can understand if you still do not want to duplicate this stuff, since one can get it from descriptives.

I just saw this usefull difference line for the dep t-Test:

So I guess we should implement this for descriptives tables of all group comparisons, classic and bayesian:

• independent t-Test & ANOVA & RM ANOVA: total row
• one sample wilcox & z & t & all 3 t-Test: difference row

And maybe also in the other tTest modules (robust & equivalence) - to stay consistent throughout jasp.

For t-Tests this is now a separate issue: https://github.com/jasp-stats/jasp-issues/issues/2753