Anova with data reduction? (Experimental design)

#1
Hi, this is my first post, I hope to do nothing wrong like posting in the wrong area.:pag

It happens that I did an experiment where I have a control group and an experimental one, I am comparing means with ANOVA, and I want to reduce the data by eliminating 5% of the lower and higher data.

I need to support the reduction of data through some book or paper, but I could not find the reference, am I doing wrong with the reduction? Could any of you provide me with an academic reference to support my procedure? Or a good website to look for information? Please, and thank you very much.
 

Miner

TS Contributor
#3
As you described you requirement, it is called trimming, which discards the data. A different yet similar process is called winsorizing, which replaces the data.

As Masteras asked, why do you want to do this? Outliers often tell you that there are lurking variables out there that you have not considered.
 
#4
Thanks to both, I will investigate these concepts.

I want to reduce data because I did 4 tests, with a different question for each one, as I said before with control group and experimental (With augmented reality).

Hypothesis:
• H0: μ1 = μ2, there is no difference between the means of the score of the groups.
• H1: μ1 ≠ μ2, there are differences between the means of the score of the groups.

The ideal would have been that with all the tests the H0 was rejected and thus to prove that the experimental group has significant differences on the control group.

I had many tests scored zero, so those students with or without experimental method do not make a difference, to be more balanced is that I also want to reduce high grades, since I assume that these students with or without new method would have good score.

Results (alfa=5%):
Test 1:
With all data -> H0 is accepted
Without 5% higher and without 5% lower of the data -> H0 is accepted

Test 2:
With all data -> H0 is accepted
Without 5% higher and without 5% lower of the data -> H0 is Rejected

Test 3:
With all data -> H0 is accepted
Without 5% higher and without 5% lower of the data -> H0 is accepted

Test 4:
With all data -> H0 is rejected
Without 5% higher and without 5% lower of the data -> H0 is rejected
 

Masteras

TS Contributor
#5
This is up to you then to explain this particular reason. You do not need bibliography validation. This reason sounds good enough. But, on the other hand some people will disagree with this, even if you have bibliography in statistics. Best thing to do is to look for bibliography in your field. Is this a common strategy? Other people do it? But again, it doe snot mean you must do it, or must not do it. In general with statistics, there is no clear answer, statisticians are like lawyers, here yes, there no.