help
help > RE: t-test and p-value in graph theory
Jan 31, 2017 12:01 PM | Pravesh Parekh - University of Oslo
RE: t-test and p-value in graph theory
Hello,
1. Are you sure you selected the correct version of test when running it in SPSS (making sure that you have made the correct choice between one sample, two sample, and paired t tests)? How are you cross-checking the values with the p values from Conn? Do note that irrespective of the test selected in the GUI, Conn stores one sided results in the .mat files.
2. Uncorrected p-values means that the test is being performed by setting the alpha at 0.05 (or some other fixed number) i.e. out of 100, you might declare a false positive 5 times. However, when doing multiple tests (as is the present case), the number of false positives no longer remain curtained to this alpha. Let's say you did 2 t tests...then the number of false positives would be 1-(0.95)^2 = 0.0975 or approx. 10 out of 100. This 'multiple comparison' problem has to be corrected for. Therefore, we penalise the alpha value and reduce it so that the overall rate of error is not more than 5%. There are two methods to do so: Family-wise error rate (FWER) correction and False discovery rate (FDR) correction. So the t values represent p-uncorrected for multiple comparison and p values after correcting using the FDR method.
3. If I remember correctly, during the thresholding step, the sided-ness refers to the positive, negative, or both values. Let's say you are thresholding based on the absolute Z values. You could look at the graphs made from just choosing positive Z scores (one-sided positive), negative Z scores (one-sided negative), or both (two sided). The threshold would then apply in the direction you have indicated. Regarding the sidedness for a statistical test like t test, the side indicates whether you are expecting a deviation on both sides of the curve or just one side. As a simplistic example, testing between two groups, you could say if A>B (one sided positive), A
Hope this helps
Best
Pravesh
Originally posted by snowa wo:
1. Are you sure you selected the correct version of test when running it in SPSS (making sure that you have made the correct choice between one sample, two sample, and paired t tests)? How are you cross-checking the values with the p values from Conn? Do note that irrespective of the test selected in the GUI, Conn stores one sided results in the .mat files.
2. Uncorrected p-values means that the test is being performed by setting the alpha at 0.05 (or some other fixed number) i.e. out of 100, you might declare a false positive 5 times. However, when doing multiple tests (as is the present case), the number of false positives no longer remain curtained to this alpha. Let's say you did 2 t tests...then the number of false positives would be 1-(0.95)^2 = 0.0975 or approx. 10 out of 100. This 'multiple comparison' problem has to be corrected for. Therefore, we penalise the alpha value and reduce it so that the overall rate of error is not more than 5%. There are two methods to do so: Family-wise error rate (FWER) correction and False discovery rate (FDR) correction. So the t values represent p-uncorrected for multiple comparison and p values after correcting using the FDR method.
3. If I remember correctly, during the thresholding step, the sided-ness refers to the positive, negative, or both values. Let's say you are thresholding based on the absolute Z values. You could look at the graphs made from just choosing positive Z scores (one-sided positive), negative Z scores (one-sided negative), or both (two sided). The threshold would then apply in the direction you have indicated. Regarding the sidedness for a statistical test like t test, the side indicates whether you are expecting a deviation on both sides of the curve or just one side. As a simplistic example, testing between two groups, you could say if A>B (one sided positive), A
Hope this helps
Best
Pravesh
Originally posted by snowa wo:
Thank you for your reply!
But i have a question about your answer.
1. The p-value of the data exported by conn in SPSS and p-value of t-test in conn is very different.
I can't understand why. Is the t-test method in conn different from the method in SPSS?
2. And the p-value at the top of the t-test(top line in the results table of the graph theory window)
Can uncorrected p-value and FDR p-value be equal?
When i choose uncorrected p-value, there is a significant roi but not FDR.
Both total p-values are the same.What is the difference between uncorrected p-value and FDR in t-test?
3. What are the means of 3 ways(one sided(positive), one sided(negative), two sided) in Network edges(adjacency matrix threshold)?
And what are the three methods(one-sided(positive), one-sided(negative), two sided)at the analysis threshold(p-value)?
What is the difference between the two?
Please answer my question.
Thanks as always.
Best regards.
But i have a question about your answer.
1. The p-value of the data exported by conn in SPSS and p-value of t-test in conn is very different.
I can't understand why. Is the t-test method in conn different from the method in SPSS?
2. And the p-value at the top of the t-test(top line in the results table of the graph theory window)
Can uncorrected p-value and FDR p-value be equal?
When i choose uncorrected p-value, there is a significant roi but not FDR.
Both total p-values are the same.What is the difference between uncorrected p-value and FDR in t-test?
3. What are the means of 3 ways(one sided(positive), one sided(negative), two sided) in Network edges(adjacency matrix threshold)?
And what are the three methods(one-sided(positive), one-sided(negative), two sided)at the analysis threshold(p-value)?
What is the difference between the two?
Please answer my question.
Thanks as always.
Best regards.
Threaded View
| Title | Author | Date |
|---|---|---|
| snowa wo | Jan 25, 2017 | |
| Alfonso Nieto-Castanon | Jan 27, 2017 | |
| snowa wo | Jan 28, 2017 | |
| Pravesh Parekh | Jan 31, 2017 | |