Monday, August 11, 2008
Statistics Question:
Quick question: Does anyone out there know how you should interpret results from an independent t-test if your sample sizes are unequal? Specifically, I am showing significant differences at Time 1 (subsequent to any treatment) with regards to a sample of completers compared to noncompleters. 20 people completed the treatment, while only 8 dropped out. Should I take the significant differences between these groups seriously, or does the fact that the non-completers group only consisted of 8 subjects undermine the strength of this finding? I am clearly hoping that I DO NOT have significant differences between completers and non-completers at Time 1, although if there are, I will need to include this in my discussion and perhaps account for any discrepancies in later analysis. Any thoughts on this will be greatly appreciated!
Subscribe to:
Post Comments (Atom)
2 comments:
Hm. The samples are small, but there's not much you can do about that. What you should consider is not the difference in sample sizes, but the possible differences in variances between the two samples. Depending on the equality of these variances, you'll be using different degrees of freedom: pooled for equal and satterthwaite for unequal. I'm not sure what stats package you're using, but if it's SAS, proc ttest will automatically test for equality of variances and then give you the pooled and satterthwaite stats. Jen says if you're using SPSS, you can request Levene's test for equality of variances, and if significant, the look to the part of the ttest output table that says "assuming unequal variances".
Ooh Simon, this is very helpful! I was using the Levine's test statistic for samples with unequal variances, but wasn't sure if that was right. The SPSS lingo is a little bit different from what I'm familiar with (we were trained on SAS), so I wasn't sure if I was conceptualizing that correctly, but from what Jen is saying it sounds like that's the way to go. Thank you both for your thoughts!
Post a Comment