planned comparison (post hoc) with fewer than 3 groups?

Say you've found that the IV "sex" has a significant effect on your DV, but you want to know the nature of that effect.

If your IV were something like "gender identity", you could run post hoc tests in ANOVA, because you would have 4 groups. But "sex" only has 2 groups. To wit, SPSS will not run a post hoc.

So, how do you analyze the relationship between variables beyond just determining significance?



Super Moderator
There is no point if you have a two level factor becasue the nature of the difference has to be female and males are significantly different. If you want to explore the nature of these differences, try plotting the data either as boxplots or are means with errror bars.
That's new to me. Not that I can't figure it out, but a little direction may help. Let me give some more detailed information.

I have Male/Female (categorical) and I know that there is a significant difference on one of my 5 (continuous) IVs. Specifically, the IV has to do with the likelihood that the individual will employ a "physical" style of relationship initiation. I have a strong intuition about WHY there is a significant difference, but, of course, I need to establish the nature of the relationship objectively before I can say "males are more likely to employ physical styles" and discuss it further (even though this is already established in the literature, I need to speak specifically concerning my sample).

With gender identity, Tukey HSD tells me the nature of the relationship (positive or negative), so I can base my "masculine and androgynous individuals are more likely to employ physical styles" off of some objective results. Something that tells me, not just that there is a significant difference, but what that difference means; the direction (positive or negative) of that relationship.

I understand that you are probably telling me that - and thank you very much. I'm only offering the specifics so that any subsequent responses are better informed. Also, I have to admit, I do not know what box plots or error bars are and I'd like to keep this as simple as possible. Nothing I've learned so far has been overwhelming, so I'm ready to tackle whatever. But, is there a third or fourth alternative that will do what I need?

Oh, and I'm using SPSS. On a scale of 1-10, I rate myself a 3 with SPSS... but, not being a 10, I'm probably not qualified to rate myself. :-/

Thanks so much!


Fortran must die
I have a somewhat different question. If male/female is one variable and physical style is another one, then I would think you could only comment on gender and physical style the way you did after you tested an interaction effect between the two (and its not signficant).
The effect of sex (male/female) on style is different than the effect of gender identity (masculine, feminine, androgynous, undifferentiated) on style.

They are interrelated (thus the use of one as a covariate in another part of my study), but sex is biologically determined, whereas gender identity is a social construct that is influenced by sex.

In this study, gender identity is the IV (literature suggests that, especially in communication formats, gender is a stronger indicator for style than sex) and the IVs are 5 different styles.

That said, in order to show that gender identity is a stronger indicator of style, I slid biological sex into the IV spot and found that it was only significant for one of the DVs (a relationship that the literature already assumed). I'd just like to say more about it than "there is a significant difference between males and females..."; specifically, I'd like to say "and the difference is ______".


Super Moderator
You may want to read a little on effects sizes. Also, what type of measurement is your response? (maybe I missed it in this thread) - at work right now, so time is limited.