ANCOVA when variances are not equal?


New Member
I am currently writing a 4th year freshwater biology report analysing data collected on the length and weight of fish from 2 sites, collected over 3 years.

We need run an ANCOVA checking Weight v Length/Site(SPSS, weight=dependent, site=factor, length=covariante).

After checking assumptions of equal variance we had to log transform the data. Variances were checked again yet were still not equal p<0.05.

Any ideas on how we should attempt to analyse this data as litereature seems to rule out the idea of using any standard non-parametric tests.




Fortran must die
If your sample sizes are relatively equal between levels you might be able to get away with unequal variances. If they are not the situation is much more difficult.

Which I am sure is what Dason was pointing out :)


New Member
sample sizes are fairly equal, approx 100 and 110 so hopefully running with unequal variances will be ok. Thanks for the help guys.
HI all!
I am having the same problem for a project that i am doing (unequal variances: variance for one group is more than 3 times smaller than for the other group), only that sample sizes are quite unequal: 25 vs 75. So i think i definitly cannot do ANCOVA. Does any of you know, if there is a way of still testing a group comparison while accounting for covariets and how i should do it?


New Member
i guess u can select randomly 25 out of 75 cases, so the the sample size becomes equal. then u can go on doing ANCOVA. U can take significant at p-value of <.001, so that committing typeI error is very less.


TS Contributor
I cannot imagine that throwing away information would be of any help, though I admit the idea tempted me as well. You could try some other transformations beside the log, like Box-Cox or Johnson.