Outlier

#1
Hallo!

I have the following problem:

I have this data
44
41
90
55

What I have tried so far (with R):

Chi-squared test for outlier
X-squared = 2.0888, p-value = 0.1484
alternative hypothesis: highest value 90 is an outlier

Dixon test for outliers
Q = 0.7143, p-value = 0.1544
alternative hypothesis: highest value 90 is an outlier

I also tried an Grubbs test (I found a tool online at: http://www.graphpad.com/quickcalcs/Grubbs1.cfm )
I got a critical Z value of 1.45 (1.48 would be significant).


I want to get the value of 90 out by an outlier test. I am not too experienced in outlier testing (I have no experience :) ). So could anybody name me a test, that is more "sensitive" concerning outliers so that th value of 90 is thrown out of my data? Thanks a lot!

Yours Hubert
 
Last edited:

Dason

Ambassador to the humans
#2
I don't know too much about tests for checking if something is an outlier but I'm guessing you won't find one that shows that 90 is an outlier when you only have 4 data points.

But you might be able to increase power by making certain distributional assumptions? How did you get this data?
 
#3
But you might be able to increase power by making certain distributional assumptions? How did you get this data?
Thanks so far! I got it by quantitative PCR. This method is normally very reproducible and I am sure that it is an Outlier. Usually values just differ by +/- 10 or something like this...
But I can't say "I feel, this is an outlier, so I leaf it out". Therefore I have to proof it somehow...

Thanks a lot!

Yours
Hubert
 

gianmarco

TS Contributor
#4
Hi!

I am not very familiar with the tests you quoted, but I would like to point out a method that is based on median and median absolute deviation (MAD).

Basically, it is a robust method which try to circumvent the issue called "masking" (outliers tending to mask themselves inflating the standard deviation).

A data point is flagged as outlier if:
value - median > 2*(MedianAbsoluteDeviation/0,6745)

You will find info about this method in this book: http://www.amazon.co.uk/Fundamental...1663557?ie=UTF8&s=books&qid=1261934308&sr=1-1

I implemented this method in a free Excel template that you can download from my website. See this previous post for a previous discussion on this topic and for download location http://www.talkstats.com/showthread.php?t=10461

By the way, I tried it on your dataset and 90 was declared outlier.


Hope this helps,
bye

Gm
 
#5
Hallo gianmarco!

Thanks so much for that. It really helped me alot so far.

I have downloaded your file and tried to find out, where you have filled in my data, but I doesn't find the right place. Could you maybe name to sheet and the area (eg "A2" in sheet "CI-Median_Table"). Thank you so much!

For those who will also read this postings later. The book is also available online (google books). You can find it here: http://books.google.at/books?id=uUN...num=1&ved=0CCoQ6wEwAA#v=onepage&q=MAD&f=false

On page 33 is a detailed description of the methode.

Yours
Hubert
 

gianmarco

TS Contributor
#6
Hi,
you can download my template and open the file. Then, go to the worksheet sample 1 (or sample 2, it is the same) and enter your values in the first left column. Choose the appropriate method to detect outlier: there is a specific menù for this. Please, read the guideline that comes along with the template.

I hope you manage to use the template.

For any further help, feel free to write to' me.

All the best,
gm