bugPSPP - Bugs: bug #22788, Regression reports erroneous...

 
 

bug #22788: Regression reports erroneous degrees of freedom and parameter estimates

Submitter:  Jason H Stover <jstover>
Submitted:  Mon 31 Mar 2008 06:01:19 PM UTC
   
 
Category:  None Severity:  5 - Average
Status:  None Assigned to:  jstover
Open/Closed:  Closed Release:  Before 0.6.0
Effort:  0.00
* Mandatory Fields

Add a New Comment Rich Markup
   

Jump to the original submission

Tue 15 Apr 2008 01:34:47 AM UTC, comment #9: 

Looks good to me.

John Darrington <jmd>
Group administrator
Mon 14 Apr 2008 06:36:25 PM UTC, comment #8: 

I just checked in patch #6488 to test the QR decomposition.
If it looks OK with everyone, I'll close this bug.

Jason H Stover <jstover>
Group Member
Wed 09 Apr 2008 03:57:37 AM UTC, comment #7: 

I just submitted patch #6488 to test the QR decomposition. Let me know if it looks OK.

Jason H Stover <jstover>
Group Member
Tue 08 Apr 2008 07:57:18 PM UTC, comment #6: 

I'll add a test.

This wasn't caught before because it only shows up if the fitting method is PSPP_LINREG_QR, or QR decomposition of the covariance matrix. That isn't used unless the data set has more than 1000 cases and the number of variables isn't too large.

Jason H Stover <jstover>
Group Member
Tue 08 Apr 2008 12:13:15 AM UTC, comment #5: 

I don't quite understand why this bug wasn't caught by the existing tests.  But whatever the reason, I suggest that we add a new test to check for this problem.

John Darrington <jmd>
Group administrator
Mon 07 Apr 2008 07:23:44 PM UTC, comment #4: 

I checked in a fix for this. Coupled with John's patch, I think we can close this bug.

Jason H Stover <jstover>
Group Member
Fri 04 Apr 2008 05:55:02 PM UTC, comment #3: 

The patch fixed the problem with the degrees of freedom.

I checked the coefficients in gdb and they seem to be computed correctly (by that I mean they are not obviously wrong, as are the ones shown in the output). The problem is in reg_stats_coeff() in regression.q. The for loop is mismatching the coefficients in c->coeff and the rows of the struct tab_table *t.

I'll fix it soon.

Jason H Stover <jstover>
Group Member
Fri 04 Apr 2008 03:59:47 AM UTC, comment #2: 

The attached patch should fix the degrees of freedom problem.
I hope someone else can fix the coefficients problem, because it's probably beyond me.  What should the correct coefficients be?


(file #15400)

John Darrington <jmd>
Group administrator
Tue 01 Apr 2008 06:17:20 AM UTC, comment #1: 

Sounds bad.  I'm marking this as a pre-0.6.0 bug.

Ben Pfaff <blp>
Group administrator
Mon 31 Mar 2008 06:01:19 PM UTC, original submission:  

Using the file http://math.gcsu.edu/~jhs/regression-test.sav
run the following syntax:

get /file='regression-test.sav'
REGRESSION
/VARIABLES= Age20to29 AgeUnder20 Age30to34 Age35Plus Year
/DEPENDENT= TotalBirths
/STATISTICS=COEFF R ANOVA.

The degrees of freedom are not shown correctly in the output table, and the displayed coefficients are incorrect.

Jason H Stover <jstover>
Group Member

 

(Note: upload size limit is set to 16384 kB, after insertion of the required escape characters.)

Attach Files:
   
   
Comment:
   

Attached Files
file #15400:  regdf.patch added by jmd (1KiB - text/x-diff)

 

Carbon-Copy List
  • -email is unavailable- added by jmd (Updated the item)
  • -email is unavailable- added by blp (Posted a comment)
  • -email is unavailable- added by jstover (Submitted the item)
  •  

    There are 0 votes so far. Votes easily highlight which items people would like to see resolved in priority, independently of the priority of the item set by tracker managers.

    Only logged-in users can vote.

     

    Follow 5 latest changes.

    Date Changed by Updated Field Previous Value => Replaced by
    2008-04-15 jstover Open/ClosedOpen Closed
    2008-04-14 jstover Dependencies- Depends on patch #6488
    2008-04-09 jmd Dependencies- patch #6488 is dependent
    2008-04-04 jmd Attached File- Added regdf.patch, #15400
    2008-04-01 blp ReleaseNone Before 0.6.0

    Back to the top

    Powered by Savane 3.13-02a9.
    Corresponding source code