dealing with unequal sample sizes... can anyone help please?

#1
Hello,
I am comparing men and women on outcomes of their diagnostic assessment (autism). there are approx 600 men and 200 women. I have used t-tests adn chi-squared analyses to compare outcomes, and then a multivariate ANOVA with sex and diagnostic outcome (also unequal sample sizes) as fixed factors to compare scores on individual symptom measures. The data is normally distributed, but I don´t know whether the unequal sample sizes are problematic here.
a colleague suggested using weighted means in SPSS - if I choose the 'weighted means - weight by sex' option in SPSS it seems to simply double the sample size of the women group, so teh groups are still not actually even (600 : 400). Maybe it is more complex than this, but I don't really understand if this is a sensibe thing to do, and if it is how do I report the N values for these analyses?
any advice at all would be greatly appreciated!
Thanks - ellie
 

Karabiner

TS Contributor
#2
The data is normally distributed,
Certainly not, but with n=800 it need't be.
I don´t know whether the unequal sample sizes are problematic here. a colleague suggested using weighted means in SPSS
Hm. You don't know whether there is a problem at all, or what
kind of problem, but nevertheless you start fixing it instantly.

AFAICS there is no basic problem arising from unequal group
sizes in your study, if you want to perfom significance tests
(by the way, did you mean multivariate ANOVA or multifactorial
ANOVA?). Or, do you have a research question which perhaps
might require equal groups?

With kind regards

K.
 
#3
Dear Karabiner,

Thanks for taking the time to comment on this.

The research questions I wanted to answer were:
1) Is there a difference in the proportion of males and females who receive an autism diagnosis (I used chi-squared here)
2) Within the autism group, are there sex differences on scores of diagnostic instruments (I used t-tests here)
3) Do sex differences on scores of diagnostic instruments vary according to subdiagnostic group (there are 2 subdiagnostic groups). (I used a multivariate ANOVA here).
 

Karabiner

TS Contributor
#4
So there's no need for weighting. If you look into your
literature/reference studies, you certainly will find
that weighting was nowhere performed.

With kind regards

K.
 
#5
Who are those people that have done an autistic assessment?

I would say that it might be because it was suspected that they are autistic. So they are not representative for the general population at all. (Have you ever heard or an autism investigation among people in general?) So you can not make any conclusions at all about women and men in general (or any other subgroup), no matter what statistical method you are using.
 
#8
that´s right - I´m not trying to make any conclusions about men and women in general, just the population that these individuals were sampled from.