T-Test and unequal sample sizes

#1
I am currently interning for a manufacturing company that makes knee implants. I was asked to do a T-test on some test data. There was sterilization testing done on each product the company makes and I was asked to compare the results of each product to their main product, the Femur, for the T-test. I know almost nothing about statistics, I had to google T-test to figure out what it was.

The only program available to use is Excel XP, for some reason I can't install the data analysis toolpack after going to Tools>Add Ins. Because of this, all I could do is use the TTEST equation. I did one T-test for each product using its result as the 2nd array, the 1st array is always the Femur. I also used 2-tail and unequal variance type.

My problem is that the Femur has about 40 samples, but some of the other parts have 10-30 samples. I can get a P value after entering the T-Test with unequal sample sizes, but are my results going to be correct? Should I make sure I have equal sample sizes by omitting some of the Femur samples? If so, what would be a proper way to omit samples?

Thanks in advance for any help.
 

CowboyBear

Super Moderator
#2
A t test can still be performed with unequal sample sizes; but the smaller sub-sample will limit the statistical power of the test. With just 10 cases in some of your subsamples, the power to detect "statistically significant" differences will be very very low - differences will only be "statistically significant" if the group means are very different. There's no statistical cure for this problem, per se - you should probably see if you can obtain larger samples, or simply report group mean differences rather than attempting inferential testing.

T-test assumptions here.