1. huff!! i am lost!! i do not know whether this will help you or increase your confusion:
you have a variable with three categories 1,2,3 ok?
well, i know in spss, we can split file (data>>split file). after that, every analysis we perform, will result in three outputs, separate for 1,2,3.
so, this will result in 3 Fs i guess. and you can perform this for the second DV.
so you will have 2 sets of 3 Fs.
and then perform Chow!!

please do not be angry on me!! if this looks a ****, just flush it out!!

2. Bourne - how could I ever be angry with you ? ;-)

Don't know what you mean though ;-) but sounds like something I should look at ...

3. you have A,B and C as three levels of a single variable, right?: "They're different levels. Males, females, and ummm, apples."

so, while performing HLM, you have just one variable as a fixed factor. but in that variable you have three categories. so my opinion is to split the data based on this variable.

4. No HLMs. Back...away...slowly.

Okay, this is why I keep harping on what is it you're trying to test? The analysis has to match the research question and the study design.

If 1,2,3 (male, female, apples)--I'll call it Gender--is what you're trying to compare X and Y on, then you need it to be a fixed factor. There is no reason to not have a 3 level fixed factor in either an ANOVA or a regression. In a regression, you would just dummy code it.

Then you can have "Model" as a two-level fixed factor. And an interaction. That is what GLMs (anovas and regressions) were born to do. No problem.

And, Bourne, I understand what you're suggesting about splitting the file, but not why. If I understand the point of this model (and I'm not making any strong claims here about understanding it), he needs to see if Gender affects X and Y the same. He won't be able to see how Gender affects anything if he looks at each level separately.

But he could just do two regression models regressing X on Gender and Y on Gender, then do a Chow test. But my impression is that a Chow test is best suited for when there are many predictors (independent variables). Since he has only one, he might as well just do an interaction. If there were many predictors, doing interactions becomes unwieldy, thus the need for the Chow.

Have fun in Japan!

Karen

5. and good morning from Japan!
... and a good morning from my luggage still in the States ... anyhoo ...

> No HLMs.
got it ... easy to ... not do.

> Okay, this is why I keep harping on what is it you're trying to test? The analysis has to match the research question and the study design.

Research question is ... does X explain/predict "gender" better than Y?
Now, I know my DVs and IVs are a bit back to front in that regard but it's an applied world.

> If 1,2,3 (male, female, apples)--I'll call it Gender--is what you're trying to compare X and Y on, then you need it to be a fixed factor.

yeap - it IS a fixed factor.

> There is no reason to not have a 3 level fixed factor in either an ANOVA or a regression. In a regression, you would just dummy code it.

right with you ...

> Then you can have "Model" as a two-level fixed factor.

Going to try that ... not ... quite ... sure ... what it'll tell me though? Or how ... but, I will run it and see.

Thanks!
Phil

6. The main effect for Model won't tell you much. It's all in the interaction. It will tell you if the effect of Gender on Response is the same or significantly different for X and Y.

I think that's what you're trying to get at.

Karen

7. OK ... so, I've started with an easier one ...

I have two DVs (X and Y)
I combine them so that a new variable (model) is composed of zeros and ones and alongside it are the corresponding X and Y values (XY).
My purpose is to see which, X or Y, best predicts the IV - which here we shall call quality (which is a continuous variable).

So - spss ...

column 1: zeros and ones
column 2: all the Xs and all the Ys are underneath the Xs
Column 3: the continuous variable called "quality"

Univariate ANOVA
DV = quality
Fixed factors = model (Xs and Ys) and their corresponding values.

Result of interest = model * XY???
If sig then ...

what?

8. I'm not sure how well the columns will come out here, but I'll give it a try...

Quality Model XY Gender

4 0 X Male
5 0 X Female
1 0 X Apple
3 0 X Apple
5 1 Y Male
2 1 Y Male
3 1 Y Female
1 1 Y Apple

Yup, I just previewed, and the spaces don't come out. I suggest copying and pasting that into Word and putting in tabs so you can get the columns to line up. there should be 4 columns.

Model is:

Quality = Model Gender Model*Gender.

If the Model*Gender interaction is signficant, it will tell you that the effect of Gender is different for X and Y.

Karen

Hi Phil,

I wonder to know if you got some conclusion regarding comparing F's.
Is it possible to say that F1=3.4 is significantly higher than F2=3.2?

I've seen some scientific works that compute (using some bootstrapping method) the distribution of the null hypothesis: F1-F2=0, and then get a confidence..

Let me know if you have any conclusions!
Thanks,
Jaime