topic Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS?? in Statistical Procedures

Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

jonatan_velarde — Thu, 05 Mar 2015 18:52:34 GMT

Good day my SAS friends:

Always is a pleasure to participate here, so here we go one more time.

I'm working in some research and many authors located in a especial region of Brasil, use to be comfortable using SCOTT-KNOTT Multiple Comparison, so

I always use the other and classic procedures (Tukey, Duncan, ETC).

Could anyone tell me please!!! how to get SCOTT-KNOTT multiple comparison in SAS.

Thanks.

King regard from Brasil

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

Doc_Duke — Fri, 06 Mar 2015 20:18:56 GMT

Jonatan,

I can't, though I found lots of reference to R packages.

I found it fascinating that the only people publishing papers that use it are from Brazil. I imagine that there is one 'enthusiast' who has converted the rest!

Doc Muhlbaier

Duke

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

lvm — Fri, 06 Mar 2015 20:52:12 GMT

I am not aware of a sas macro. I wrote a FORTRAN program to do this more than 30 years ago, and used it for a few years. I know it has the intuitive advantage of 'creating' non-overlapping groups of LSMEANS, based on the means and standard errors. Basically, it is a type of cluster analysis (of means). However, early assessments showed that the experiment-wise Type I error rate can be high with this method. This would likely be the reason why the method is very uncommon, and not available within SAS procedures. I have not read anything about it in decades, so there may be some counter-arguments to the criticisms. Someone who has the time and knows PROC IML could probably take one of the multiple-comparison macros that have been written by non-SAS programmers, and add a routine for this method. I would start with the macro written by H.P. Piepho.

http://agrobiol.sggw.waw.pl/~cbcs/articles/CBCS_7_1_2.pdf

This would take some expertise with IML, and time.

Or, you could call a relevant R function for SK from within SAS IML to do this. You would have to learn about the SAS-R linkages first.

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

invisibleman — Sat, 25 Jul 2015 01:02:29 GMT

So, did you get some routine?

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

lvm — Mon, 27 Jul 2015 12:52:38 GMT

I haven't looked since I don't need it.

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

jonatan_velarde — Mon, 10 Aug 2015 16:43:43 GMT

hi there :

Unfortunately still don't have it 😕 , such a bad luck hum?

i think the person deveplops this method for SAS will be very famous

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

SteveDenham — Tue, 11 Aug 2015 12:41:22 GMT

Why? As @lvm said, the experiment wise type I error with this method is high, and there are much better methods for controlling for this sort of error already available using the ADJUST= option in the linear models procedure, and in PROC MULTTEST.

Steve Denham

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

jonatan_velarde — Tue, 11 Aug 2015 12:55:36 GMT

Hi SteveDenham:

For sure, this is the objective to be tested, could you send us an example using these methodologies.

thanks

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

SteveDenham — Tue, 11 Aug 2015 17:01:20 GMT

Start here in the documentation:

https://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_mixed_syntax08.htm

See also the Shared Concepts for LSMEANS and the documentation for PROC MULTTEST.

Steve Denham

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

LuisCarlos — Sat, 29 Aug 2015 01:58:23 GMT

Fiz uma busca ampla sobre a existência de tal rotina no SAS, no entanto, nada encontrei. Não sei qual as limitações de uso, mas sei que essa rotina está disponível no R e em um software livre chamado ASSISTAT. No entanto leve em consideração que esse teste não é propriamente para comparações de médias e sim um teste de agrupamentos a partir uma única variável, apresentando limitações teóricas como já descritas acima.

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

jonatan_velarde — Sat, 29 Aug 2015 12:50:32 GMT

Ola Luis Carlos:

obrigado pela tua resposta, nesse sentido voce tem como me informar onde tem aquela rotina de R por favor??

desde ja agradeço tua resposta

abç

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

invisibleman — Sat, 29 Aug 2015 19:49:48 GMT

/*Scott-Knott in R and SAS (integration*/

/*Tutorial by: (Alysson Jalles) Consultoria: www.alyssonjallessite.wordpress.com*/

*I have the solution;

*I've wrote a R/SAS routine for this purpose;

*You can run R inside SAS to do it!;

*To set SAS to run R codes follow my tutorial in: https://communities.sas.com/message/296841#296841;

*Copy all of this message inside SAS to perform the analysis;

*I'll provide an example with a SAS simulated dataset of a Randomized Complete Block Design (RCBD);

*=======================================================================;

*Simulating a RCBD experiment ===========================================BEGIN;

*=======================================================================;

*10 TREATMENTS;

*3 REPLICATION;

proc plan seed=123456;

factors GEN=10 ordered REP=3 ordered/noprint;

output out=RCBD_TEST

/*GEN cvals=('GEN1' 'GEN2' 'GEN3' 'GEN4' 'GEN5' 'GEN6' 'GEN7' 'GEN8' 'GEN9' 'GEN10')*/

GEN nvals=(1 2 3 4 5 6 7 8 9 10)

REP nvals=(1 2 3);

run;

*Simulating VARIABLES - BEGIN;

data RCBD_TEST; set RCBD_TEST;

call streaminit(8789); /*SEED= 8789*/

VAR1=rand('normal',110, 15); /*MEAN 110: STD: 15*/

run;

data RCBD_TEST; set RCBD_TEST;

call streaminit(907854); /*SEED= 907854*/

VAR2=rand('normal',3000, 751);run; /*MEAN 3000: STD: 751*/

*Simulating VARIABLES - END;

*=======================================================================;

*Simulating and RCBD experiment =====================================END;

*=======================================================================;

*Printing dataset, first 6 observations of simulated dataset:;

proc print data=RCBD_TEST (obs=6); run;

/*Scott-Knott statistic of VAR1 and VAR2 in R language*/

*Installing "ScottKnott" package in R;

*ScottKnott in an R package to perform Scott-Knott analysis;

*It will appear a windows, just click on the nearest mirror of your current place;

* You just need to do it ONCE!;

proc iml;

title "Installing ScottKnott package in R (integration R with SAS)";

submit / R;

#_______Begin of R code_________________

install.packages("ScottKnott")

#_______End of R code_________________

endsubmit;

*quit;

*Installing "xlsx" package in R;

*xlsx is a package to export Scott-Knott results in excel to your hard drive;

*to use this package you have to install the ultimate version of java in your computer:

https://www.java.com/pt_BR/download/

*It will appear a windows, just click on the nearest mirror of your current place;

* You just need to do it ONCE!;

proc iml;

title "Installing xlsx package in R (integration R with SAS)";

submit/ R;

#_______Begin of R code_________________

install.packages("xlsx")

#_______End of R code_________________

endsubmit;

*quit;

*****SCOTT-KNOTT ANALYSIS BEGINS HERE..*******************************************************;

*Changing treatments - GEN from numeric to factor type;

*This is necessary;

proc iml;

title "Preparing variables (integration R with SAS)";

*Exporting SAS dataset to R environment;

run ExportDataSetToR("WORK.RCBD_TEST", "RCBD_IN_R");

submit / R;

#_______Begin of R code_________________

#Class of GEN and REP

class(RCBD_IN_R$GEN)

class(RCBD_IN_R$REP)

#CONVERTING "GEN" and "REP" TO CHARACTER

RCBD_IN_R$GEN<-as.character(RCBD_IN_R$GEN)

RCBD_IN_R$REP<-as.character(RCBD_IN_R$REP)

#Class of GEN and REP

class(RCBD_IN_R$GEN)

class(RCBD_IN_R$REP)

#_______End of R code_________________

endsubmit;

*quit;

*Performing the Scott-Knott analysis;

*We'll perform the analysis with 2 VARIABLES simultaneously, but you can extend this routine to N variables;

title "Performing ScottKnott clustering in R (integration R with SAS) - by: Alysson Jalles";

submit / R;

#_______Begin of R code_________________

require(ScottKnott) #Package to perform the SK analysis

require(xlsx) #Package to perform the output in XLSX of results

#For other kind of design as CRD, factorial, etc, consult the manual of R ScottKnott package

#in: https://cran.r-project.org/web/packages/ScottKnott/ScottKnott.pdf

#Applying univariate Scott-Knott clustering

VAR1a <- SK(x=RCBD_IN_R, y=RCBD_IN_R$VAR1, model="y~GEN+REP", which="GEN", sig.level=0.05)

#summary(VAR1a)

VAR2a <- SK(x=RCBD_IN_R, y=RCBD_IN_R$VAR2, model="y~GEN+REP", which="GEN", sig.level=0.05)

#summary(VAR2a)

#Summary

VAR1b<-(summary(VAR1a))

VAR2b<-(summary(VAR2a))

#_____________________________

#RENAMING MEANS TO THE NAME OF VARIABLE EVALUATED

names(VAR1b)[names(VAR1b)=="Means"] <- "VAR1"

names(VAR2b)[names(VAR2b)=="Means"] <- "VAR2"

#RENAMING SK(5%)TO SK PLUS NAME OF VARIABLE EVALUATED

names(VAR1b)[names(VAR1b)=="SK(5%)"] <- "SK_VAR1"

names(VAR2b)[names(VAR2b)=="SK(5%)"] <- "SK_VAR2"

#MERGING SCOTT-KNOTT RESULTS IN A SINGLE DATASET

SK <-Reduce(function(x, y) merge(x, y, all=TRUE),

list(

VAR1b, #VAR1

VAR2b #VAR2

))

#RENAMING NAME OF VAR "Levels" TO GEN [TREATMENTS]

names(SK)[names(SK)=="Levels"] <- "GEN"

#_______End of R code_________________

endsubmit;

*quit;

*Exporting R results to your hard drive and SAS Environment;

title "Exporting results to Windows hard drive and SAS Environment (integration R with SAS)";

title2 "Your file: 'Scott_Knott_output.xlsx' is in C:\ of your HD";

*Exporting results from R to SAS;

run importdatasetfromR("WORK.SCOTT_KNOTT_SAS","SK");

submit / R;

#_______Begin of R code_________________

write.xlsx(SK, "Scott_Knott_output.xlsx")

#_______End of R code_________________

endsubmit;

*quit;

*Sorting treatments - GEN;

proc sort; by GEN; run;

*Printing results of Scott-Knott analysis in SAS (sig.level=0.05);

Title "Scott-Knott Analysis results to RCBD design in SAS, by: Alysson Jalles";

proc print data=SCOTT_KNOTT_SAS noobs; run;

*****SCOTT-KNOTT ANALYSIS ENDS HERE..*******************************************************;

Thanks for read my material :smileygrin:...

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

jonatan_velarde — Tue, 29 Sep 2015 17:03:42 GMT

Thank you for your contribution.

we'll test and compare this protocol and them we'll leave you any comment. Untill now we appreciate this.

Thank you

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

jonatan_velarde — Wed, 30 Sep 2015 16:59:14 GMT

Good day my friend:

As you know, ths SAS university edition is improveing each day, so i tried to use iyour code in SAS university edition and i could not finish the statement.

Anybody caould help with this???

Thank you

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

invisibleman — Tue, 13 Oct 2015 13:43:19 GMT

Hi jonatan_velarde..the bad news is that SAS university edition does not run any R code.

If you want to run this routine you should have the Commercial SAS version.

Or by other hand, you can run the second part of the routine totally in R or RStudio.

#####################################################################################

#(English title:)Scott-Knott univariate clustering in R to RCBD.
#(Portuguese-Br title): Agrupamento univariado de Scott-Knott para DBC)
#routine By Alysson Jalles.
#www.alyssonjallessite.wordpress.com

#Here we have 2 variables to evaluate.
#Creating a dataset with RCBD data...
RCBD_IN_R<-read.table(header=TRUE, text=("
GEN REP VAR1 VAR2
1 1 109.797 4296.39
1 2 109.325 3159.55
1 3 129.079 2550.13
2 1 124.041 3206.70
2 2 102.700 3498.19
2 3 90.762 2676.06
3 1 127.014 3287.07
3 2 127.461 2345.28
3 3 122.337 3649.71
4 1 95.214 2339.48
4 2 106.913 2118.81
4 3 107.182 2426.04
5 1 119.641 3811.88
5 2 94.830 4166.51
5 3 124.218 2457.73
6 1 110.261 2276.99
6 2 106.819 1494.28
6 3 103.399 1452.75
7 1 123.564 740.09
7 2 109.369 1561.06
7 3 122.680 1951.73
8 1 103.247 3259.35
8 2 103.542 3363.61
8 3 109.542 3149.99
9 1 113.026 1924.95
9 2 76.786 3041.53
9 3 91.773 1807.83
10 1 81.236 3458.72
10 2 95.125 3185.58
10 3 95.898 3039.91
                    "))

#Changing treatments - GEN from numeric to character type;
#Class of GEN and REP
class(RCBD_IN_R$GEN)
class(RCBD_IN_R$REP)
#CONVERTING "GEN" and "REP" TO CHARACTER
RCBD_IN_R$GEN<-as.character(RCBD_IN_R$GEN)
RCBD_IN_R$REP<-as.character(RCBD_IN_R$REP)
#Class of GEN and REP
class(RCBD_IN_R$GEN)
class(RCBD_IN_R$REP)

#Intalling packages to Analysis.
#Scott-Knott statistic of VAR1 and VAR2 in R language*/
#Installing "ScottKnott" package in R;
#ScottKnott in an R package to perform Scott-Knott analysis;
#It will appear a windows, just click on the nearest mirror of your current place;
#You just need to do it ONCE!;
install.packages("ScottKnott") #To Performa SK analysis

#Installing "xlsx" package in R;
#xlsx is a package to export Scott-Knott results in excel to your hard drive;
#to use this package you have to install the ultimate version of java in your computer:
#https://www.java.com/pt_BR/download/
#When you click in install.packages("xlsx") It will appear a windows,
#just click on the nearest mirror of your current place;
#You just need to do it ONCE!;
install.packages("xlsx")       #Package to export your dataset to excel

#Performing the Scott-Knott analysis;
#We'll perform the analysis with 2 VARIABLES simultaneously,
#but you can extend this routine to N variables;
#_______Begin of R code_________________
require(ScottKnott) #Package to perform the SK analysis
require(xlsx)          #Package to perform the output in XLSX of results
#For other kind of design as CRD, factorial, etc, consult the manual of R ScottKnott package
#in: https://cran.r-project.org/web/packages/ScottKnott/ScottKnott.pdf
#Applying univariate Scott-Knott clustering
VAR1a <- SK(x=RCBD_IN_R, y=RCBD_IN_R$VAR1, model="y~GEN+REP", which="GEN", sig.level=0.05)
#summary(VAR1a)
VAR2a <- SK(x=RCBD_IN_R, y=RCBD_IN_R$VAR2, model="y~GEN+REP", which="GEN", sig.level=0.05)
#summary(VAR2a)
#Summary
VAR1b<-(summary(VAR1a))
VAR2b<-(summary(VAR2a))
#_____________________________
#RENAMING MEANS TO THE NAME OF VARIABLE EVALUATED
names(VAR1b)[names(VAR1b)=="Means"] <- "VAR1"
names(VAR2b)[names(VAR2b)=="Means"] <- "VAR2"
#RENAMING SK(5%)TO SK PLUS NAME OF VARIABLE EVALUATED
names(VAR1b)[names(VAR1b)=="SK(5%)"] <- "SK_VAR1"
names(VAR2b)[names(VAR2b)=="SK(5%)"] <- "SK_VAR2"
#MERGING SCOTT-KNOTT RESULTS IN A SINGLE DATASET
SK <-Reduce(function(x, y) merge(x, y, all=TRUE),
            list(
              VAR1b, #VAR1
              VAR2b #VAR2
            ))
#RENAMING NAME OF VAR "Levels" TO GEN [TREATMENTS]
names(SK)[names(SK)=="Levels"] <- "GEN"
#_______End of R code_________________

#Exporting R results to your hard drive and SAS Environment;
#"Your file: 'Scott_Knott_output.xlsx' is in C:\ of your HD";
#Exporting results from R to SAS;
#_______Begin of R code_________________
setwd("C:/") #Assigning path to output file
getwd()       #Checking out if your file was exported sucessfully
write.xlsx(SK, "Scott_Knott_output.xlsx")
View(SK) #Once you View your table in RStudio you can copy and paste your results in excel.
#_______End of R code_________________

Re: Scott-Knott Multiple Comparison ... Is there any procedure in SAS??

tconrado — Tue, 19 Sep 2017 16:55:54 GMT

Jonatan,

You can find the SAS macro for the adjusted Scott-Knott in the link from this paper

http://ref.scielo.org/n4c75n

About the widespread use of it in Brazil, you should understand that the test by itself was proposed as an alternative to avoid ambiguity (which should not be understood as issue in comparison tests since "comparison tests were created to detect differences" - Tukey).

The Scott-Knott test was developed by two economics professors from Auckland and London and was very criticized by american researchers in the very beginning because it does not control the experiment-wise Type I Error as the MCP if there are more than a cluster of means (which means it does control it only if there is no true difference among them or if all the means are truly different). Also be aware that this statment is validy if you do evaluated the Scott-Knott test in the same manner as the MCP are, which is not the best approach since they different goals.

It is an alternative for a post ANOVA analysis, but it has a different concept than Tukey, Duncan, SNK, Dunett and my others.

Despite the lack of the control of the experiment-wise error, the Scott-Knott test has a much higher power than those listed above, since the most of MCPs lowers a lot the "real" Type I Error to guarantee it do not surpass the nominal level, therefore they have high Type II Error, which can be understood as low Power.

Thus, in a scenario were the Type I Error is not a big concern, the Scott-Knott (1974) it is a good option (i.e. preliminary experiments or studies where it is expected to have many treatments in the same cluster and it is not interesting to carry all of them to the next research stage if they show No Significant Difference. It is good fit for researches areas where there are a high number of treatments with such small variation among those and it is possible to perform a following validation, these scenarios are indeed rare if you take in account the multiple areas where statistics play a big role. I notice a lot of use of this test in agronomic research, but almost never in the other areas, so if you take in account that Brazil is country that play big almost only in agriculture it is possible to understand the wide adoption of the test in Brazil.

Regards,