Sorry, forgot to mention this paper, which talks about whether to do
multivariate tests or multiple univariate tests:
Huberty, C., & Morris, J. (1989). Multivariate analysis versus
multiple univariate analysis. Psychological Bulletin, 105: 302-308.
And a paper on why multivariate anova is just plain weird:
Cole, D. A., Maxwell, S. E., Arvey, R., & Salas, E. (1994). How the
power of MANOVA can both increase and decrease as a function of the
intercorrelations among the dependent variables. Psychological
Bulletin, 115(3): 465-474.
The more highly correlated your outcome variables, the higher (i.e.
less significant) your multivariate p-values will be, which is kind of
the opposite of what you might expect. You get the most power when
your outcomes are correlated negatively - but many psychologists (who
kind of blindly say "I've got more than one outcome, I'd better do a
multivariate test") will have measured two highly correlated outcomes
- like anxiety and depression, and then to a multivariate test,
because there are two of them, but that guarantees (almost) that you
won't get a significant result. Anyway, I've got off the point. I'll
stop.
J
On 19 March 2013 09:02, Jeremy Miles <[log in to unmask]> wrote:
> OK, here's a start. I don't quite understand your design, so I'm going
> to talk in general terms. People often use mixed anova when they
> should use ancova - if you have two measures of the same variable -
> e.g. pre- and post- therapy depression, and you're interested in the
> change in that depression associated with therapy, then ancova is
> better.
>
> A mixed design is the same as taking the difference and doing a
> t-test, and the problem is that this mixed anova (or t-test on the
> differences) assumes that the variables are correlated 1.00, if
> they're correlated less, a mixed anova is less powerful than an
> ancova.
>
> If your 5 measures are the big five, you should definitely not do a
> mixed anova - you don't care if people score higher on N than on E.
>
> It sounds to me like your 5 outcomes are the big five, and your
> predictor is parenting style (plus you should thrown in some other
> covariates, like age and gender.) But what your outcome (outcome is a
> better word than dependent, as it doesn't imply a causal relationship
> which you haven't established) variable is depends on your hypothesis.
> I can see that you might test whether parenting style affects
> personality, in which case personality is the outcome. Or you might
> test whether parents respond differently to different personality
> styles of their children, in which case parenting style is the
> outcome.
>
> However, MAN(C)OVA is also overused by psychologists. You should use a
> multivariate test if you're truly interested in multivariate effects,
> you're not, you're interested in univariate effects, so I'd use 5
> ancovas. (Or 5 regressions - it's the same thing).
>
> Jeremy
>
>
>
>
>
> On 19 March 2013 06:17, Nina Bay <[log in to unmask]> wrote:
>> Hi guys,
>>
>> I am sure this is going to sound like a very silly question but I am running
>> a mixed design ANOVA in SPSS, but I was told that I should probably be
>> running a mixed design MANCOVA instead.
>>
>>
>>
>> I was told that the procedure was the same in SPSS, but that I should look
>> at the Multivariate test table instead.
>>
>>
>>
>> I am using the big five inventory for personality with the 5 different
>> personality treats, and my between-subjects variable is parenting style.
>>
>>
>>
>> I was told that personality should be my dependent variable… but I am really
>> confused.
>>
>>
>>
>> Could anyone provide some guidance please?
>>
>>
>>
>> Many thanks,
>>
>>
>>
>> Nina
|