Discussion: View Thread

This message has been cross posted to the following eGroups: Young Professionals Group and Statistical Consulting Section .
-------------------------------------------
Greetings Everyone,

I need to use proc npar1way to find the difference between the amount of protein expressed between two groups: diseased versus normal. Specifically, I am looking at the amount of proteins expressed on immune system cells.
 
But I am confused with proc npar1ways score rank thing and how it affects my p-values.

Also, how do I interpret the results when a score rank is used? What is the benefit of using my raw data as the scores? (feeling utterly confused right now)

Here is my code:
proc npar1way data=_14INT.set;  
class Lupus;
var _14intTyro3DMFI;
exact wilcoxon;
run;

Thanks in advance for the suggestions 
-------------------------------------------
Margaret Linan
Graduate Research Intern
Temple University School of Medicine
------------------------------------------- 

Margaret,
The measure of difference between groups associated with the Wilcoxon rank-sum test is the Hodges-Lehman, estimator.  It it the median of all the possible (n*m) pairwise differences.  It is computed by many software packages, so must be available in SAS somewhere.  Conover's Practical Nonparametric Statistics or other references, including Wikipedia  http://en.wikipedia.org/wiki/Hodges%E2%80%93Lehmann   will give you more info.

-------------------------------------------
Dennis Helsel
Practical Stats
-------------------------------------------

SAS does indeed give Hodges-Lehman estimates.  Just say "hl" in the Proc Npar1way statement. 



-------------------------------------------
Eric Siegel
Biostatistician
Univ of Arkansas for Medical Sciences
-------------------------------------------





Original Message:
Sent: 07-18-2012 09:39
From: Dennis Helsel
Subject: PROC NPAR1WAY: To Score or Not to Score?

Margaret,
The measure of difference between groups associated with the Wilcoxon rank-sum test is the Hodges-Lehman, estimator.  It it the median of all the possible (n*m) pairwise differences.  It is computed by many software packages, so must be available in SAS somewhere.  Conover's Practical Nonparametric Statistics or other references, including Wikipedia  http://en.wikipedia.org/wiki/Hodges%E2%80%93Lehmann   will give you more info.

-------------------------------------------
Dennis Helsel
Practical Stats
-------------------------------------------

The Hodges-Lehmann estimator was available since SAS 9.2.  Only the Wilcoxon test is currently available for the HL estimators.

To call it, simply include HL as the proc NPar1way option (e.g., 'proc NPar1way Wilcoxon HL').

You can save the HL estimators (and CI) by:
      ods output WilcoxonTest=Wilcoxonp HodgesLehmann=HLEstimator;


-------------------------------------------
Allen Fleishman
Allen Fleishman Biostatistics Inc.
-------------------------------------------



Original Message:
Sent: 07-18-2012 09:39
From: Dennis Helsel
Subject: PROC NPAR1WAY: To Score or Not to Score?

Margaret,
The measure of difference between groups associated with the Wilcoxon rank-sum test is the Hodges-Lehman, estimator.  It it the median of all the possible (n*m) pairwise differences.  It is computed by many software packages, so must be available in SAS somewhere.  Conover's Practical Nonparametric Statistics or other references, including Wikipedia  http://en.wikipedia.org/wiki/Hodges%E2%80%93Lehmann   will give you more info.

-------------------------------------------
Dennis Helsel
Practical Stats
-------------------------------------------

Hi, Margaret, I have two more questions:
(1) for this data, are you merely the analyst, or did you also generate the data?
(2) do you have additional code that computes the median amounts of protein in each group?

-------------------------------------------
Eric Siegel
Biostatistician
Univ of Arkansas for Medical Sciences
-------------------------------------------