t-Test, P-Value of 0?

#1
Hello All,

I'm doing a paired t-test with a data set close to 1300. We're looking at pain scales from before and after treatment. When I run the test (with excel and graphpad.com) I'm getting a P-Value of 0...so I keep thinking something is wrong. I've posted the output information below from excel.

Am I doing the right test? And is it possible to have P-value of 0 (I would have expected something along the lines of 0.0008E-14 or something, but just 0 makes me a little concerned that something is wrong with either the way I set up the test in excel or with how my data is arranged).

Any help would be appreciated.

Thank you!

t-Test: Paired Two Sample for Means

Variable 1 - In Level --- Variable 2 - Out Level
Mean 6.141263238 --- 3.436649017
Variance 4.246730954 --- 3.455152805
Observations 1322 --- 1322
Pearson Correlation 0.510524044
Hypothesized Mean Difference 0
df 1321
t Stat 50.5080754

P(T<=t) one-tail 0
t Critical one-tail 1.646007938

P(T<=t) two-tail 0
t Critical two-tail 1.961761371
 
Last edited:
#2
Hello All,

I'm doing a paired t-test with a data set close to 1300. We're looking at pain scales from before and after treatment. When I run the test (with excel and graphpad.com) I'm getting a P-Value of 0...so I keep thinking something is wrong. I've posed the output information below from excel.

Am I doing the right test? And is it possible to have P-value of 0 (I would have expected something along the lines of 0.0008E-14 or something, but just 0 makes me a little concerned that something is wrong with either the way I set up the test in excel or with how my data is arranged).

Any help would be appreciated.

Thank you!

t-Test: Paired Two Sample for Means

Variable 1 - In Level --- Variable 2 - Out Level
Mean 6.141263238 --- 3.436649017
Variance 4.246730954 --- 3.455152805
Observations 1322 --- 1322
Pearson Correlation 0.510524044
Hypothesized Mean Difference 0
df 1321
t Stat 50.5080754

P(T<=t) one-tail 0
t Critical one-tail 1.646007938

P(T<=t) two-tail 0
t Critical two-tail 1.961761371
I've never done a t-test in excel and I haven't checked your maths, but you've got very large t Stats and the sample size is large, so I'd expect your P values to be very low indeed - I'd imagine they are just very very close to zero and excel gives you zero due to rounding.

On a seperate note, t-tests (the one in excel anyway), assume approx. equal variances. If you want to be a stats purist you probably want to test for this using e.g. a Levene Test.

Having said all that, I think that, even if you violate the assumptions of a t-test slightly, given the differences and sample size involved, you can be almost certain your two groups are different!
 
#3
Thank you for your reply. I attached the data below (please view the worksheet tabs at the bottom for additional information).

And if it makes any difference, when I run a Wilcoxon Signed Rank Test, I get this information:
# of cases with absolute differences greater than 0 = 1198
Smaller sum of ranks: 9951.50
Approximate normal z for test stat: T=29.150
Probability (1-tailed) of greater z = 0.000

Its been a while since I've taken & used this level of stats (more the 5 years!) so my brain is very rusty when it comes to knowing how to appropriate analyze and test this data set.

Also, I just wanted to ask this community if anyone thinks a analysis of covariance test would be appropriate (we don't have any other information about this test population though, ie no age, gender, etc). I was thinking this test wouldn't be applicable unless we could gather additional information.

Thank you, thank you, thank you!
 
#4
Hi MHoffman,

I guess you might have resolved the issue. But just thought of sharing you with some info that may be of some help in the near future. Recently I came across an estimator termed "SSMD", claimed to be an efficient tool to compare two or more groups. Wish u all the best.