Is this possible: group passes normality, but outlier is present?!

#1
Hi!

I have posted last night something about this issue, but now I will reformulate the question.

I have data from animal study in which we had 6 animals in both of 2 groups: control and test. We treated them and I measured amount of enzyme in both of them. I expected that in a test group amount of enzyme is going to be higher than in control group.

1. I did normality testing by GraphPad Prisms ( Kolmogorov-Smirnov test) , and both groups passed.
2. In test group there was one value which was approximately 7x higher than others so I run Grubbs test for outliers. It showed that the 7x higher value is outlier.

3. I excluded outlier, applied one-tailed t-test and I got that there is significant difference between those two groups. But, the mean of test group was Lower than the control group's mean (the opposite than expected).
4.If I keep outlier, run one-tailed t-test there is no significance. But, the mean of test group is in this case HIGHER than the control group's mean.
5. If I apply two-tailed t-test, there is no significance, no matter whether outlier is excluded or not.

I am wondering what would be the best approach: to keep that outlier or not?

And I am wondering if in my case the one-tailed t-test is correct?

I think that the most correct would be to exclude outlier an run two-tailed t-test, but I am not quite sure about that.

I will appreciate any comment, suggestion, advice or discussion. Thanks
 

Karabiner

TS Contributor
#2
1. I did normality testing by GraphPad Prisms ( Kolmogorov-Smirnov test) , and both groups passed.
You surely cannot describe this as "passed". With such a small sample size
as n=6, even marked deviatons from normality will go unnoticed by the K-S-test.
The power of the test is very, very low with n=6. Your outlier test is much more
convincing here.
5. If I apply two-tailed t-test, there is no significance, no matter whether outlier is excluded or not.
Just don't apply one-tailed tests.
I am wondering what would be the best approach: to keep that outlier or not?
Outliers distort the t-test. One question is: why does this outlier
exist, which process did produce it? Was it e.g. measurement errror? Cf.
http://pareonline.net/getvn.asp?v=9&n=6 for some conceptual issues and
possible solutions.

With kind regards

K.
 
#3
One question is: why does this outlier
exist, which process did produce it? Was it e.g. measurement errror?
Actually, when we sacrificed the animal we realized that it seems that part of the food instead to end up in animal's stomach ended up in lungs :(. So the measuring itself is not the reason, but probable mistake/defect in procedure, yes.

Thank you so so much.

And do you think that two-tailed t-test is more correct than one-tailed in this situation? I have issue with this.

Usually biologists in similar cases of research one-tailed, but from my point of view it should be two-tailed

Once again, thank you very much.
 

Karabiner

TS Contributor
#4
And do you think that two-tailed t-test is more correct than one-tailed in this situation?
The one-tailed test has more power than the
2-tailed test, which could be nice to have with
just 11 valid observations. But apart from that
I don't see a reason for using a one-tailed test.

With kind regards

K.