Paired T-test

[zerostin]

Hello,

I want to perform a paired T-test. I try to do this in Excel 2007. The only "problem" is, is that I already have my std. deviation and means. So, excel thinks my n=1, which is not the case. Is there a program or a way to do manually input the std. deviation, means and N so that I can calculate this?

[Dason]

You have more than one mean for a paired t-test?

[zerostin]

I read some old statistics books and came to the conclusion that this was the best option.

H0: U1 - U2 = 0
Ha: U1 - U2 != 0

So, I thought that the paired T-test can check if the differences between the two means are different than zero.

                               Sample 1                 Sample 2
N:                            11368                     9706      
Mean:                      .1119241                 -.047241
Std. deviation            .352279                  .2427075

[trinker]

i don't have the answer to your excel problem but your last statement about Ho: mean difference for pre / post is equal, is correct provided you only have one group no control.

[zerostin]

It's the same group in a different year! Therefore the N is also a bit different!

[Dason]

I'm still hung up on the fact that you have two means for a paired t-test. You should only have one mean. The mean of the differences between the pairs. I don't know exactly how your pairing was done. Let's assume it was a pre-post test situation. This implies you took the mean for the pre group and the mean for the post group. That wouldn't be what you would do for a paired t-test. For a paired t-test you would compute the post-pre differences and use the mean and standard deviation from that data.

[zerostin]

But in Excel you can give the values of sample 1 and the observations of sample 2. And then he compares those two samples. Like this --> http://www.qimacros.com/qiwizard/t-t...variances.html

As I know the mean, SD, Variance of the sample (pre/post) I thought there would be a way to calculate it with only that information. The reason that I am trying to want to do it like this is that it would save me a lot of time. Otherwise I need to create approx. 250 variables.

edit: I see that it should use an unpaired T-test (because N is not the same).

[Dason]

edit: I see that it should use an unpaired T-test (because N is not the same).

Definitely. How exactly were you doing the pairing before if N isn't the same?

[zerostin]

I have no idea It is a long time ago for me that I had these classes about T-tests I didn't try it, because I first need to figure out how to do it.
Anyway, I did the unpaired T-test manually in excel. I created an excel sheet (see attachment) were you can fill in the mean, SD, and N. Excel calculate the T-statistic and P-value.

However, I have a question about this. You have an unpaired t-test with equal variances and an unpaired T-test with unequal variances.
How do you decide if your variances are unequal?

[zerostin]

I created variables and performed an unpaired t-test in another statistical program. However, I have to reject the null hypothesis for every ttest I performed. All the p-values are very significant. So, that's not a good thing. My variables are not normally distributed, but N is at least 3000 and at most 15000 so then I can assume that my data is still normally distributed, right? So, anybody got some hints to fix this? Is a Wilcoxon rank-sum test the best alternative in this case?

edit: I performed a WR sum test and almost everything is still very significant. Is this still an indication that there is something wrong?