Re: [bug #48040] GLM produces wrong output

[Alan Mead]

On 5/27/2016 11:24 AM, John Darrington wrote:

Follow-up Comment #1, bug #48040 (project pspp):

Thanks for the bug report.

PSPP (unlike SPSS) does not show the "intercept" if the model is
unbalanced, since it has no meaning (just what calculation SPSS does in this
case is a mystery).

The intercept has never been of interest to me, so I don't have much experience. However, I thought the intercept was simply the value of the dependent variable when the composite of independent variables was zero?  Being unbalanced prevents the composite of independent variables from having a zero point? Or am I misunderstanding?

By "unbalanced" do you mean that some cells are entirely missing? Or that cells and not all the same size? (Both of which are true in these data.)

The negative F value (and consequently NaN df) are more of a concern, possibly
also related to missing values.

From the output, it's possible this is all (or much) of the issue is due to the df being calculated wrong.  The error df somehow came out negative, which caused the error MS to be negative, which caused all the F's to be negative.

Although that explanation doesn't explain how the second analysis failed to find significant main effects for both independent variables.

Also it's interesting that the total SS match between PSPP and SPSS precisely in both analyses, but the error SS are off by a bit in the first analysis (where things went really wrong).

Also the speed issues concern me too.

Possibly a windows issue as I cannot replicate on the same/comparable hardware running Linux. PSPP becomes unresponsive when running this analysis (i.e., in response to mouse events, Windows adds "(Not Responding)" to the title bar--a sign that usually means that the application is crashing or fallen into an infinite loop or some similar failure).  I was surprised when the output appeared while I was preparing the bug report.

-Alan

-- 

Alan D. Mead, Ph.D.
President, Talent Algorithms Inc.

science + technology = better workers

+815.588.3846 (Office)
+267.334.4143 (Mobile)

http://www.alanmead.org

I've... seen things you people wouldn't believe...
functions on fire in a copy of Orion.
I watched C-Sharp glitter in the dark near a programmable gate.
All those moments will be lost in time, like Ruby... on... Rails... Time for Pi.

          --"The Register" user Alister, applying the famous 
            "Blade Runner" speech to software development