search

jasp-stats / jasp-issues

This repository is solely meant for reporting of bugs, feature requests and other issues in JASP.
59 stars 29 forks source link

[Feature Request]: jackknife and leverage in linear regression + non-linear regression #1549

Open PetBuj opened 2 years ago

PetBuj commented 2 years ago

Description

I supposed to add several features to existing JASP-methods to make it more useful.

Purpose

The purpose is to incerase usability of JASP.

Use-case

No response

Is your feature request related to a problem?

no

Describe the solution you would like

  1. add feature to PCA/EFA to enable store scores into data-sheet, 2. enable to input data for contingency tables(tests) by existing counts, 3. Add to Casewise diagnostics (Lin.regression) Jack-knife residuals and leverage, 4. Nonlinear regression.

Describe alternatives that you have considered

I believe, all these features/methods are available in R packages and therefore, the process is highly possible.

Additional context

Thank authors of JASP to work on this very practical free package!

juliuspfadt commented 2 years ago

Hi @PetBuj, thanks for the reports.

PetBuj commented 2 years ago

Hello,

I mean something like this:

Minimum Iteration Section

Iteration No. of Percent of Bar Chart

No. Clusters Variation of Percent

1 2 56.39 |||||||||||||||||

4 3 33.06 ||||||||||

7 4 25.69 ||||||||

(NCSS report)

Where PercentOfVariation means how much (0-100%) the selected clusters (2-k) in k-means are possible reduce variability of the objects (when covering by k-clusters).

Best regards,

P. Bujok

From: Julius Pfadt @.> Sent: Tuesday, September 6, 2022 9:23 AM To: jasp-stats/jasp-issues @.> Cc: Petr Bujok @.>; Mention @.> Subject: Re: [jasp-stats/jasp-issues] [Feature Request]: several points to increase user-friendly approach (Issue #1549)

Hi @PetBujhttps://github.com/PetBuj, thanks for the reports.

— Reply to this email directly, view it on GitHubhttps://github.com/jasp-stats/jasp-issues/issues/1549#issuecomment-1237757629, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AVQORQPQ4K4B4CVOWXH7J23V43WNXANCNFSM5KDL5TEQ. You are receiving this because you were mentioned.Message ID: @.***>

juliuspfadt commented 2 years ago

what do you think @Kucharssim ?

Kucharssim commented 2 years ago

Dear @PetBuj,

Thanks for the points for improvement, I will break it down in separate points.

  1. add feature to PCA/EFA to enable store scores into data-sheet:

As @juliuspf wrote, this is already covered by a different issue: https://github.com/jasp-stats/jasp-issues/issues/1666

  1. enable to input data for contingency tables(tests) by existing counts:

I am not sure what you mean, the contingency analysis already allows specifying input data by using existing counts. Could you please clarify what is missing?

  1. add to Casewise diagnostics (Lin.regression) Jack-knife residuals and leverage:

I think we could do that, perhaps @fqixiang could have a look...

  1. nonlinear regression. As @juliuspf wrote, it's not clear what kind of analysis you mean by the term "nonlinear regression". Could you clarify, please?

Further, you wrote:

I mean something like this:

Minimum Iteration Section

Iteration No. of Percent of Bar Chart

No. Clusters Variation of Percent

1 2 56.39 |||||||||||||||||

4 3 33.06 ||||||||||

7 4 25.69 ||||||||

(NCSS report)

Where PercentOfVariation means how much (0-100%) the selected clusters (2-k) in k-means are possible reduce variability of the objects (when covering by k-clusters).

That seems to me that you meant to write this response to this issue instead https://github.com/jasp-stats/jasp-issues/issues/1407, is that correct? Thanks!

PetBuj commented 2 years ago

Dear Simon,

1 – thanks

2 - it is my fault I missed this possibility.

3 – Thank for your effort

4 – Nonlinear regression is ‚standard‘ procedure beside linear regression where y is dependent on Xs by a specific (no linear) model. The model contains one or more parameters which are estimated during the procedure (similarly to lin.reg but in diff.way). Some motivation is for example here (https://techvidvan.com/tutorials/nonlinear-regression-in-r/)

Best regards,

P. Bujok

From: Simon Kucharsky @.> Sent: Tuesday, September 6, 2022 3:20 PM To: jasp-stats/jasp-issues @.> Cc: Petr Bujok @.>; Mention @.> Subject: Re: [jasp-stats/jasp-issues] [Feature Request]: several points to increase user-friendly approach (Issue #1549)

Dear @PetBujhttps://github.com/PetBuj,

Thanks for the points for improvement, I will break it down in separate points.

  1. add feature to PCA/EFA to enable store scores into data-sheet:

As @juliuspfhttps://github.com/juliuspf wrote, this is already covered by a different issue: #1666https://github.com/jasp-stats/jasp-issues/issues/1666

  1. enable to input data for contingency tables(tests) by existing counts:

I am not sure what you mean, the contingency analysis already allows specifying input data by using existing counts. Could you please clarify what is missing?

  1. add to Casewise diagnostics (Lin.regression) Jack-knife residuals and leverage:

I think we could do that, perhaps @fqixianghttps://github.com/fqixiang could have a look...

  1. nonlinear regression. As @juliuspfhttps://github.com/juliuspf wrote, it's not clear what kind of analysis you mean by the term "nonlinear regression". Could you clarify, please?

Further, you wrote:

I mean something like this:

Minimum Iteration Section

Iteration No. of Percent of Bar Chart

No. Clusters Variation of Percent

1 2 56.39 |||||||||||||||||

4 3 33.06 ||||||||||

7 4 25.69 ||||||||

(NCSS report)

Where PercentOfVariation means how much (0-100%) the selected clusters (2-k) in k-means are possible reduce variability of the objects (when covering by k-clusters).

That seems to me that you meant to write this response to this issue instead #1407https://github.com/jasp-stats/jasp-issues/issues/1407, is that correct? Thanks!

— Reply to this email directly, view it on GitHubhttps://github.com/jasp-stats/jasp-issues/issues/1549#issuecomment-1238141709, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AVQORQM7TQO2KA7TITXSDWLV45AG5ANCNFSM5KDL5TEQ. You are receiving this because you were mentioned.Message ID: @.***>

TarandeepKang commented 1 year ago

Hi @Kucharssim & @juliuspfadt For a grounding in nonlinear regression see:

https://link.springer.com/book/10.1007/978-0-387-09616-2

and a great implementation:

Florent Baty, Christian Ritz, Sandrine Charles, Martin Brutsche, Jean-Pierre Flandrois, Marie-Laure Delignette-Muller (2015). “A Toolbox for Nonlinear Regression in R: The Package nlstools.” Journal of Statistical Software, 66(5), 1–21. doi:10.18637/jss.v066.i05.

Which builds on the nls function in the stats package

I was just coming to suggest this, but someone beat me to it!

tomtomme commented 9 months ago

missing part from this issue is adding more casewise diagnostics for linear regression

tomtomme commented 9 months ago

@TarandeepKang Can you have a look at point 3 from my comment above? Is that implementation of non-linear regression sufficient? Or do we need nls from stats package?

TarandeepKang commented 9 months ago

So I would suggest that in the meantime we should find some way (maybe the help file) in the regression module of saying that polynomial regression is possible with flexplot. I had no idea! In the longer term, would be great to have maximum flexibility from nls, there will obviously be much work and I think there are greater priorities. Don't you agree?

tomtomme commented 9 months ago

Agreed. Added the documentation-hint to the list: https://github.com/jasp-stats/jasp-issues/issues/2529

remaining parts from this issue:

adding more casewise diagnostics for linear regression

In the longer term, adding more flexible non-linear regression via

PetBuj commented 9 months ago

Yes, I agree.

Petr B

From: TarandeepKang @.> Sent: Saturday, February 24, 2024 6:52 PM To: jasp-stats/jasp-issues @.> Cc: Bujok Petr @.>; Mention @.> Subject: Re: [jasp-stats/jasp-issues] [Feature Request]: jackknife and leverage in linear regression + non-linear regression (Issue #1549)

So I would suggest that in the meantime we should find some way (maybe the help file) in the regression module of saying that polynomial regression is possible with flexplot. I had no idea! In the longer term, would be great to have maximum flexibility from nls, there will obviously be much work and I think there are greater priorities. Don't you agree?

— Reply to this email directly, view it on GitHubhttps://github.com/jasp-stats/jasp-issues/issues/1549#issuecomment-1962436697, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AVQORQJ5BRX5QKE7HWJSQALYVISETAVCNFSM5KDL5TE2U5DIOJSWCZC7NNSXTN2JONZXKZKDN5WW2ZLOOQ5TCOJWGI2DGNRWHE3Q. You are receiving this because you were mentioned.Message ID: @.**@.>>