Question about SS errors in GLM Module

Discuss statistics related things

by pennefather » Mon May 06, 2019 6:14 pm

Hello,
I am a first time user of the program and really like it. I teach a Honor's statistics class teaching from the GLM perspective and was hoping to move to Jamovi. However, I am unable to find a way to view the total SS and corrected SS using GLM. As the introductory part of the course teaches the basics using a simple model comparison (using the mean versus using zero) to teach the basics of model building those two errors are necessary. Is there a way to output them?

Thanks,
Jordan
pennefather
 
Posts: 9
Joined: Mon May 06, 2019 5:20 pm

by pennefather » Tue May 07, 2019 8:14 pm

Also a related issue. I am unable to run the equivalent of a single sample t-test either against zero or another a priori value using GLM (which would again require SS total and SS corrected values). My hope is to run all analyses for the class using only the GLM module for simplicity and clarity of instruction if possible.
pennefather
 
Posts: 9
Joined: Mon May 06, 2019 5:20 pm

by mcfanda@gmail.com » Wed May 08, 2019 11:36 am

Hi Jordan,
thanks for considering our software. Here are some thoughts.
1) single sample t-test. Within the GLM, a single sample t-test is simply the t-test associated with the intercept of a intercept-only model. To test that a variable X mean is different from zero use GLM, put X as a dep variable, no independent variables. Here is an example

Image

Image

For comparing the men of X with an arbitrary value mu, simple run the same model with X-mu as dependent variable.

2) I'm afraid that the SS you mentioned are not output by default by GLM. I see their use in teaching, but they are less useful in practice. However, one thing we can to do to come in your direction, is to add SS of error also for only-intercept models. Thus you can get the error SS for the mean with a intercept-only model, and compare that with the SS error of more complex models. This will help understanding model comparison. I believe that this will make the GLM useful for learners and it would not change the output of normal users (who usually do not have this models in mind). Would that be useful?
mcfanda@gmail.com
 
Posts: 145
Joined: Thu Mar 23, 2017 9:24 pm

by mcfanda@gmail.com » Wed May 08, 2019 2:28 pm

Dear Jordan,
I got hook by your request so I've tried to implement it. It will be in version 1.5.2, which has also some bug fixes, due soon (end of the week). The small change I made can, I think, solve your case. The Anova table in GLM now shows the total sum of squares and the error sum of squares for all model, including the intercept-only model. I did not implement the SS corrected and not corrected as SPSS does because I think it is just confusing for the general users. Nontheless, I think you can use the new info for your reasoning with the students.

One can now show what is the error SS if one only use the mean of y to predict it, thus one uses an intercept-only model

Image

Then one shows that by adding a variable as predictor, the error SS decreases whereas the total SS stays the same
Image

From there, one can show all the coefficients (R^2, peta^2 and F-tests) with a model comparison approach.

If you really wants to show how the intercept compares with the null-model, than you can still to that by showing the error SS of a model with no intercept and no predictor. You do that in GLM by deflaging the option "Fixed Intercept" in the "Model" option panel. This would estimate a model were the predicted values are all zero.

Image

The Error SS is the "uncorrected" SS that produces SPSS.

I hope it is usefull
mc
mcfanda@gmail.com
 
Posts: 145
Joined: Thu Mar 23, 2017 9:24 pm

by pennefather » Wed May 08, 2019 5:39 pm

mc,
Thanks you so much that is excellent. Once the new version is released I will run through the analyses from my first 3 weeks of class (the simple model sections) and make sure I am able to get the output I need and report back.
pennefather
 
Posts: 9
Joined: Mon May 06, 2019 5:20 pm

by pennefather » Tue May 28, 2019 10:54 pm

Has this change been implemented yet?
pennefather
 
Posts: 9
Joined: Mon May 06, 2019 5:20 pm

by mcfanda@gmail.com » Wed May 29, 2019 11:24 am

Yes, it will be available in a few days, just the time to update the library.
mcfanda@gmail.com
 
Posts: 145
Joined: Thu Mar 23, 2017 9:24 pm

by pennefather » Mon Jul 08, 2019 9:32 pm

Hi MC,
I downloaded the newest version today and am not seeing the total SS in the output. Has it been implemented?
pennefather
 
Posts: 9
Joined: Mon May 06, 2019 5:20 pm

by mcfanda@gmail.com » Tue Jul 09, 2019 1:20 pm

Hi,
since 1.5.3 there're residual and total SS, also for intercept-only models.
Image
mcfanda@gmail.com
 
Posts: 145
Joined: Thu Mar 23, 2017 9:24 pm

by pennefather » Wed Jul 10, 2019 6:39 pm

Great thanks. I just needed to reinstall the Module.
pennefather
 
Posts: 9
Joined: Mon May 06, 2019 5:20 pm


Return to Statistics