# Sample Sizes

#### Stuck

##### New Member
Oks. I have stressed a bit and am going to have another go.

I have concluded that however I divide the data I have, I am going to end up with unequal sample sizes. How important is this statistically?

Previously I had n1=24 and n2=16. JohnM very kindly helped me out and highlighted the reasons why my tests would be unfair with these figures.

I have another idea involving two groups where n1=43 and n2=55. How similar do sample sizes have to be for a statistical test to produce valid results?

Thanks very much for reading and helping the statistically challenged such as myself, this is a great forum.

#### JohnM

##### TS Contributor
As sample sizes increase, differences in the sample sizes aren't as problematic. In other words, 12 vs 4 would be a problem, but 70 vs 62 wouldn't be a big deal.

One way to deal with it is to randomly select a number of items from the larger set that is equal to the smaller set (i.e., randomly select 62 from the 70).

Repeat the exercise several times and run the analysis each time you do it to see how often you get significant or non-significant results.

#### Stuck

##### New Member
JohnM said:
As sample sizes increase, differences in the sample sizes aren't as problematic. In other words, 12 vs 4 would be a problem, but 70 vs 62 wouldn't be a big deal.

One way to deal with it is to randomly select a number of items from the larger set that is equal to the smaller set (i.e., randomly select 62 from the 70).

Repeat the exercise several times and run the analysis each time you do it to see how often you get significant or non-significant results.
I didn't think of that.

I'll try doing that. My problem is that I have all this data and have to do a report on it but I am struggling to decide what to investigate. I thought I had it with that t-test but trying to start again now!

Thanks for the help, you're the greatest.

#### Stuck

##### New Member
I have 40 cows, divided into group A in which n=24 and B in which n=16.

Can I select 15 cows at random from each group to produce two equal sized samples on which I can perform an unpaired t-test?

It's annoying that my sample sizes are so small but I only have so many cows to test.

Would this be a valid experiment?

Last edited: