Recent bug report on pspp-users

[John Darrington]

On pspp-users, Doug Bonnet reported the following, all of which I think are 
valid:

* Change ???Significance??? to ???Sig.??? in linear regression and one-way 
ANOVA output to be consistent with SPSS and other PSPP analyses.
* A vertical line is needed after the DV variable name in the output of the 
independent t-test (see output of the linear contrast table in one-way ANOVA as 
a guide)
* ???pair 0??? is printed instead of ???pair 1??? in the paired-samples t-test

These three were trivial, so I have pushed a fix.


* Only integer percentage confidence levels are allowed. Non-integer values 
(e.g. 97.5%) are needed for Bonferroni adjustments (like SPSS).

The Bonferroni option is in the /POSTHOC subcommand of ONEWAY - I checked this 
and it certainly accepts non-integer values.  But perhaps we ought to do an 
audit
of all procedures to check that we haven't used lex_integer where lex_number is 
needed.


* Information in all analysis-specification screens is lost when returning to 
those screens (as if Reset had been selected).

A number of people have complained about this.  I started looking at it in 
September;  maybe I should pick it up again.  There are some questions to be 
answered though -  for example presumably the state of each dialog should be
cached on a per dataset basis - otherwise bad things could happen if a dialog
popped up with variable A selected, but there is no variable A in that dataset.
This means that we need to invalidate the cache if a procedure changes a dataset
in any way.  I'm a bit worried that if we don't do this carefully, crashes could
occur ...


* When entering k columns of data one column at a time, an error message occurs 
when entering a value in the second row of column 2, 3, ..., k.

Confirmed.  Any ideas anyone?

* Confidence intervals are needed in the linear regression output (like SPSS).

Makes sense.  Currently there are a lot of subcommands in REGRESSION which are 
accepted but ignored.


* Certain selected measures of association are ignored in Crosstabs. 

I find the code in crosstabs.q hard to follow.  Is the time right to replace 
crosstabs.q with crosstabs.c like (most of) the other commands?


The following easily-programmed features are not in SPSS but are very important 
and would make PSPP more useful than SPSS for classroom use now that APA is 
requiring confidence intervals in all of its journals:

* Confidence intervals for linear contrasts in one-way ANOVA
* Confidence intervals for Pearson correlations
* Confidence intervals for measures of association in Crosstabs for those 
measures where standard errors are computed.

Makes sense.  However we need someone qualified, who can suggest how to present 
these, and to provide some sample datasets and cheeck our calculations.


J'




-- 
PGP Public key ID: 1024D/2DE827B3 
fingerprint = 8797 A26D 0854 2EAB 0285  A290 8A67 719C 2DE8 27B3
See http://sks-keyservers.net or any PGP keyserver for public key.

signature.asc
Description: Digital signature