Please help!

#1
I would be SO grateful if anyone could help me with my poor stats know-how. My experiment involves data from patient tissue before and after treatment to asses whether certain immune cells increase or decrease, with the null hypothesis that there is no change....Unfortunately due to time constraints I have extremely low sample size, n=3 in some cases! Example below:

Untreated:
1. 2374.591497
2. 2784.754691
3. 12653.68345

treated:
1. 2729.185784
2. 7233.945032
3. 17401.62162

I can see after treatment an increase in cell number in each patient.
I have done a paired T-test on xcel (p=0.154).
Is there another statistical test I can do which might highlight significance in this difference?
 

Miner

TS Contributor
#2
You could perform a 1-sample Wilcoxon on the delta, but will get similar results. With n = 3, you have a woefully low powered test.

I recommend that you decide a priori what difference is of practical significance then calculate the sample size necessary to detect that difference.
 
#3
Thank you for your reply Miner. I will take your advice on board, unfortunately my time to continue this experiment has run out so I will have to earmark this as a limitation to the study.
Meanwhile if my untreated tissue cell numbers have a distribution approximate to normal (according to SPSS) whilst my treated cells do not...(or vice versa) is it still okay to perform a T-test or 1-sample Wilcoxon?

Thanks again : )
 

Miner

TS Contributor
#4
For the paired t-test, it is the differences that must be normal, not the groups. The 1-sample Wilcoxon is a nonparametric test.
 

Karabiner

TS Contributor
#5
Meanwhile if my untreated tissue cell numbers have a distribution approximate to normal (according to SPSS) whilst my treated cells do not...
SPSS didn't tell you that your sample is from a normal distribution.
With n=3, the normality test is simply too powerless to reject the
normality distribution hypothesis. With n=3, you can test for any
distribution and SPSS will not reject the Null hypothesis that the
sample data come from that whatever-distribution.

With kind regards

K.
 
#7
Thank you for confirming something I was wondering Karabiner, very glad you brought that up. I actually did the normality test on a larger set (n=6), is this still too small to test for normality? If what Miner said (that it is only the normality of the difference between tissues that is important) is true then my question doesn't really matter any more. However on a Youtube video (link below) at 0.49 it says that to perform a parametric T-test the dependent variable (cell number in my experiment) should be approximately normally distributed for each category of independent variable (my treated and untreated tissues) so I am wondering how these seemingly conflicting bits of information fit together?
Charlotte

https://www.youtube.com/watch?v=IiedOyglLn0&index=4&list=UUcF71p7RVOVLVjanRsouQ_A
 

Miner

TS Contributor
#8
You must distinguish between a 2-sample t-Test, which requires independent samples that assume normality of both groups, and a Paired t-test of dependent samples that assume normality of the differences.
 

Stuby

New Member
#9
Please help

UP UP
So no one for the first tune...I know its tough & scarce but surely someone got the the knowledge, please help me
What i can tell is this a blank dont have matix & a very rare tune