|Date: ||Thu, 26 Jun 2003 11:02:05 +0200|
|Reply-To: ||Asesoría Bioestadística
|Sender: ||"SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>|
|From: ||Asesoría Bioestadística
|Subject: ||Re: Interaction term significant but sub-group p-value is not|
|Content-Type: ||text/plain; charset=us-ascii|
First of all, I suppose the two variables you talk about (affiliation &
staff) are the IV, and there is a 3rd variable, not mentioned by you, that
is the DV.
I think you have missunderstood the concept of interaction, or
"heterogeneity of effects". If an interaction term is significant, it means
that the OR across sugroups are statistically different from each other, but
it doesn't mean that a certain OR has to be or not significant.
In your case, you have: Affiliation C( 2)*staff (1) p=0.020. This means that
the OR for staff in affiliation=C is different from the OR for staff in
affiliation=A (reference group), but it doesn't mean that the OR has to be
I recommend you to try a stratified analysis with crosstabs:
/TABLES="staff" BY "dv" BY "affiliation"
/FORMAT= AVALUE TABLES
/STATISTIC=CHISQ RISK CMH(1)
/CELLS= COUNT .
Replace the names in brackets by the actual names of your variables.
You will get 3 OR, one for the association of "staff" with the outcome
("dv") in the subgroup affiliation=A, another OR for subgroup Affiliation=B,
and a last one for Affiliation=C.
If a significant interaction exist, it will mean that the 3 OR are different
from each other, but it will not means that they are different from the null
value (null OR=1).
Christina Cutshaw ha escrito:
> List Members,
> Does anyone know why one would have a significant interaction term but
> the odds ratios of the specific subgroup is not significant?
> I have two variables: affiliation (3 nominal groups A, B, C) and staff
> (0=0, 1=>0).
> In my binary logistic model I ran: affiliation, staff, and an
> interaction term affiliation*staff.
> In my output I have:
> Affiliation B (1) ignore
> Affiliation C (2) ignore
> Staff (1) use info for affiliation reference group A
> Affiliation B(1)*staff (1) p=0.101 (ignore OR, etc.)
> Affiliation C( 2)*staff (1) p=0.020 (ignore OR etc.)
> The odds ratio, CI and p-value of staff (1) the Affiliation (2) group is
> OR=4.50, 95% CI=0.89, 22.64, p=0.068
> Why would the interaction term be significant at p=0.020 but the p-value
> for the actual group is 0.068?
> Chris Cutshaw