# P Values/T-test

1. ## P Values/T-test

So, I've been given a data set containing the median number of total clinical trial procedures and total unique clinical trial procedures. I've been asked do a t test and find the p values to see if there is any relationship between the median number of unique procedures and total procedures.

Basically, it's a table containing the numbers as shown in the image. I had to cover the numbers because this is confidential information. Anyway, my stats is failing me, and I wasn't sure how to go about calculating the p value.

What would make sense?

Also, would you recommend using the TTEST function or the data analysis function? For some reason, I get different p-values when I use the two different options.

Any suggestions will be appreciated.

Capture.PNG  Register To Reply

2. ## Re: P Values/T-test

Excel is not very good with accuracy in statistical analysis. I highly recommend using something like minitab for this.

That said:

1. Do F-test first to see if variances are equal
2. Use TTEST function to get just your P-Value or, better yet, use data analysis function to get a lot more info (in each case you will need to use your result of the F-Test)

When done properly, P-Value returned by both function is identical (at least for all scenarios I have tried)  Register To Reply

3. ## Re: P Values/T-test

Hi Vandan_tanna
Thank you for your suggestion. I've summarized the different therapeutic areas under each phase 1-4 by year. Do I calculate a P value for each Phase, comparing the data for Unique Procedures and Total Procedures? Or am I missing something? My PI has asked me to calculate the P value for this table, and I'm not sure if he's looking for a single P value or multiple P values for each phase. What do you think?Capture2.PNG  Register To Reply

4. ## Re: P Values/T-test

Since I don't fully know/understand your data I can only offer few suggestions:

1. t-test/F-test is for comparing two groups assuming both are normally distributed. When comparing more than two groups simultaneously, use Levene's test for variance and Anova for means

2. For non-normal distributions, use Levene's and Mann-Whitney for pairwise comparison and Levene's/Kruskal-Wallis to compare multiple groups simultaneously

Use Normal probability plot to see if your data is normally distributed.

Remember, torture numbers enough and they will confess to anything. So know your data, know what you are doing, and use the correct statistical test to support your conclusion.

I use minitab for my 'important' statistical analysis. For a quick glace I use excel. I have an excel add-in for this quick-glance (t-test, F-test, Anova, Kruskal-Wallis, Mann-Whitney, box plots, normal probability plots, etc etc) at www.vtools.net

Word of caution: This is written by an engineer with limited statistical knowledge!  Register To Reply

5. ## Re: P Values/T-test

hallo,
I saw your discussion of 2012, and I was wondering if these statistical tests, of which you speak, can be found in an excel file, they are not very practical and insert only the values ​​would be useful  Register To Reply