Unequal sample sizes, unequal variances, help my ANCOVA!

Hi everyone,

I am trying to test to see if there is a difference between the means in two populations.

Population A: 100 people, controls
Population B: 330 people, anorexic

I want to see whether they differ on a measure of body checking (ex: pinching fat). I'm using SPSS.

Originally I did an independent samples t-test. Levene's test was significant, so I just read along the line for 'equal variances not assumed'.

But now I want to control for age and BMI (Body Mass Index). I was going to use an ANCOVA for this. BUt my variances are very unequal, and there is no option in the output for when that happens.

From everything that I've read, an ANCOVA is not appropriate due to my large discrepancy in variances.

WHat are my options now? Is there any way to compare the means of these two populations with unequal variances? My prof mumbled something about a regression, but I'm not sure that would be appropriate due to the unequal variances?

Thank you!


Ambassador to the humans
Are you going just based off the levene's test? You might want to compare the residuals after you fit the ancova to see if the unequal variance is still an issue. It could be that not controlling for those factors is partly why there was unequal variance in the first place.
I did do some math that was recommended in a guide - I'm not sure if it qualifies as 'comparing the residuals'? Forgive my ignorance…

What I did was I got the standard deviation of each sample from the ANCOVA output. I squared it to get the variance. I then divided the larger variance by the smaller variance. My guide said that if this then equalled less than 3, I didn't have to worry about unequal variances. But it came out to more than three, so I figured ANCOVA was therefore inappropriate.


Ambassador to the humans
The standard deviation of what exactly? The response? The residuals? We care about the variance of the residuals (the error term actually but the residuals are the drop in replacement since we can't actually observe the error directly).
The standard deviation on the questionnaire that the participants were given. Ex: population A had a mean of 24 and an SD of 13 on the questionnaire, while population B had a mean of 8 and an SD of 7.

By variance of residuals are you referring to the standard error of the estimated marginal means? You noted that I could compare the residuals to see if unequal variance was still an issue…I'm not quite sure how to do that but I am very intrigued. Thank you so much for your help!


TS Contributor
Population A: 100 people, controls
Population B: 330 people, anorexic
But now I want to control for age and BMI (Body Mass Index).
One of your groups is more or less defined by having an extremely low BMI,
and you want to "control" for this very feature which constitutes this group.
Why is this done? And why age, do the groups markedly differ with regard to age?

With kind regards