Date: Thu, 26 Jan 2006 09:18:49 -0500
Reply-To: Art@DrKendall.org
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Art Kendall <Art@DrKendall.org>
Organization: Social Research Consultants
Subject: Re: meaningless interaction?
In-Reply-To: <002801c621f1$9924d280$3400a8c0@NOTEBOOK>
Content-Type: text/plain; charset=us-ascii; format=flowed
to expand on the good advice Hector gave, you can get a vizualization :
do a religiosity (Y) by education (X) by gender scatter plot. rotate the
picture using the thumb wheels to see it from different angles.
also do a religiosity (Y) by education (X) with different markers for
gender and have the scatterplot fit regression lines for the groups.
rotate the picture using the thumb wheels to see it from different angles.
Try a loess line in the scatterplot instead of the regression line and
see if that changes your impression much.
Art
Art@DrKendall.org
Social Research Consultants
Hector Maletta wrote:
>Bozena,
>I do not think you can infer such conclusion. Both results are compatible,
>and the interaction may be significant while the separate regressions are
>not. On the one hand, suppose the betas are +0.2 and -0.2, and none is
>significantly different from zero, but the difference between them is still
>significant (0.4). Besides, the regression lines of educ and relig for women
>and for men could be both almost flat (i.e. betas not significant, even if
>of opposite signs) but positioned at different heights (i.e. constants in
>the equations are significantly different). These are two of the many
>situations in which the interaction gender x education could be significant
>while the separate regressions of education and religiosity yield non
>significant betas. You must take also into account the possibility of having
>non linear relationships between your variables, since all your analysis is
>only predicted on LINEAR relationships.
>
>Hector
>
>-----Mensaje original-----
>De: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] En nombre de
>Zdaniuk, Bozena
>Enviado el: Wednesday, January 25, 2006 5:39 PM
>Para: SPSSX-L@LISTSERV.UGA.EDU
>Asunto: meaningless interaction?
>
>Hello, everybody. A quick question:
>Let's assume I test the effect of gender X education interaction on
>let's say religiosity and I find the interaction term to be significant.
>And I then run the regression separately for women and men and find that
>the effect of educ on religiosity is non-significant in either gender
>group but the betas have opposite signs.
>Does that mean that the significant interaction effect was meaningless
>and simply an artifact of the non-significant betas having the opposite
>signs?
>Please let me know if I am right on that?
>Bozena
>
>Bozena Zdaniuk, Ph.D.
>
>University of Pittsburgh
>
>UCSUR, 6th Fl.
>
>121 University Place
>
>Pittsburgh, PA 15260
>
>Ph.: 412-624-5736
>
>Fax: 412-624-4810
>
>email: bozena@pitt.edu
>
>
>-----Original Message-----
>From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] On Behalf Of
>Paul Dickson
>Sent: Tuesday, May 10, 2005 6:00 PM
>To: SPSSX-L@LISTSERV.UGA.EDU
>Subject: Re: Kappa for more than two raters
>
>Hi Brian
>
>This link will hopefully help.
>
>ftp://ftp.spss.com/pub/spss/statistics/nichols/macros
>
>in nskappa.sps. The read me file is nskappa.rm. .... The macro
>computes Kappa, both ASE measures given in CROSSTABS, the
>asymptotic significance for testing population Kappa=0, and
>also adds 95% confidence interval bounds under the alternative
>(using ASE1).
>
>A second macro is also there, providing the multi-rater Kappa
>statistic and standard error discussed by Siegel and Castellan.
>This is mkappasc.sps, with read me file mkappasc.rm. I gave it
>the name mkappsc rather than just mkappa, because there are
>multiple versions for multiple rater problems, ....
>
>
>
>
>
>
>>Brian Mcmillan <b.r.w.mcmillan@LEEDS.AC.UK> wrote:
>>
>>Hello all,
>>
>>Does anyone know how to get SPSS to calculate a Kappa for more than
>>
>>
>two
>
>
>>raters? I have four raters rating about 1,000 separate events into
>>
>>
>Y/N.
>
>
>>Thanks in advance,
>>
>>Brian
>>
>>
>
>
>
>
| 
