topic Re: Manually Performing a Grubb's Test in Discussions

Manually Performing a Grubb's Test

cwestberg — Thu, 12 Jun 2014 17:06:20 GMT

Howdy Everyone,

I am in the process of creating a JMP appendix for one of our processes and have found it necessary to explain how to manually set up a column formula to perform a Grubb's Test. I have done this in Excel, but i want to replicate that in JMP and have had a hard time setting up the formula correctly. We have 4 columns of independent data (1a,2a,3a,4a) that each need a Grubb's Test performed on them. Each has a max of 40 values. I woudl like to create four more columns (1b,2b,3b,4b) that each scan the appropriate column (1-1, 2-2 and so forth). I was thinking of setting up an IF formula so that i can do a column min and a column max calculation and depending on the Texp value for each, it would spit out Outlier as a result. In my excel spreadsheet i it set up this way:

have the Max and min calculated

Texp calculated: (Max/Min - column avg)/ Column Std Dev = Texp

Then the cell below it compares that Texp to the Tcrit for an N=40 (2.866), and if Texp > Tcrit it says "Yes"

In my head i can see how to do this in JMP, but i am unsure how to set it up correctly with parenthesis and so forth. If i have not eplained it very well I apologize. Sometimes the brain has a hard time describing what appear easy.

Thanks,

Re: Manually Performing a Grubb's Test

stan_koprowski — Thu, 12 Jun 2014 19:04:54 GMT

Hello Charles,

Please use the following link for using an add-in from the file exchange for performing Grubbs' Outlier Test.

Please let us know if this works for your situation or if you need additional assistance.

Stan

Re: Manually Performing a Grubb's Test

ms — Thu, 12 Jun 2014 21:03:45 GMT

Here's an attempt to use a single formula column to perform the Grubb's test (g > g0) and, if an outlier is detected, pinpoint the row with "Yes".

The script creates an example table with a numeric column "Data" and a formula column with Grubb's test at alpha 0.05. The data is from the N(0,1) distribution plus one extreme value (50). Run the script, the extreme should be marked with "yes", and then change one or more numbers in the data column to observe how it affects the output in the formula column.

New Table( "Grubbs",

New Column( "Data", numeric, values( 50 || J( 1, 25, Random Normal( 0, 1 ) ) ) ),

New Column( "Grubbs Outlier Test",

Character,

Formula(

If(

And(

Col Maximum( Col Standardize( :Data ) ) ==

Col Standardize( :Data ),

Col Maximum( Col Standardize( :Data ) ) > ((Col Number( :Data ) -

1) / Sqrt( Col Number( :Data ) )) *

Sqrt(

Abs(

t Quantile(

0.05 / (2 * Col Number( :Data )),

Col Number( :Data ) - 2

)

) ^ 2 / ((Col Number( :Data ) - 2) +

Abs(

t Quantile(

0.05 / (2 * Col Number( :Data )),

Col Number( :Data ) - 2

)

) ^ 2)

)

"Yes"

)

);