Return to Stat Tools Home

Title: Example Testing SAS Format

Evaluate this tool by clicking on the stars:

%VOTE{id="EvalStatTools" stars1="Tools Overall Rating" stars2="Code Works Well" stars3="Good Instructions" global="off" open="off" secret="off" format="| $key | $small
Score: $score, My vote: $mylast, Total votes: $sum |" }%

Add Comments and then click on "Add comment" button

Please note any errors found in the macros.

%COMMENT%

StatToolsForm edit

Type of Tool SAS Macro
Title Example Testing SAS Formats
Programmer/Email Rochester
Contributing Site Rochester
Description Here is an example of attaching the code or cut and pasting the code directly into the form.
Example Example code
Classification Graph Type
References
Datasets
Data
Software Program SAS
Software
Macro Parameters
R-Code - Attachment
R-Code
SAS-Code - Attachment Rochester_EGTesting_SASmainmacro
SAS-Code * Macro logistic invokes SAS proc logistic and organizes the output. It has options for automatic creation of quartile
variables and/or quadratic variables.
The macro parameters are:

DSName = name of SAS data set
Outcome = name of outcome variable. This variable is assumed to be coded 0, 1 and the model predicts the
probability of a 1 outcome.
Predictors = list of predictor variables. This list should include class variables and variables to be 'quartiled.'
Interaction variables may be included in which case the predictor list should be quoted, e.g.
Predictors = %quote(Age Gender Age*Gender). Note: all variables appearing in interaction terms
should also appear as main effects.
Classvar = classification variables with reference level. Use %quote function. For example,
Classvar=%quote(Gender(ref='Male')Race(ref='Asian')) Gender and Race could be numeric variables with
formatted values of Male or Asian or they could be alphanumeric variables with string values of 'Male'
and 'Asian'. Do not include variables for which quartiles are to be generated.
Quartiles = list of numeric variables which should be entered into the model as class variables with four quartile
levels. Variables in this list will not be included in the model as linear predictors.
Quad = Y/N for creating and testing quadratic terms (default = Y)
Out = name of output SAS data set containing logistic estimates, p-values, etc.
Print = Y/N (default = N)
Include = F,P,FP,or N
F = frequencies of outcome variable are displayed for each class variable value, e.g., Sex Female 31/100
P = percentages are displayed with class variable values, e.g., Sex Female 31.0%
FP = frequencies and percentages are displayed with class variable values, e.g., Sex Female 31/100 (31.0%).
N = no frequencies or percentages are displayed with class variable values (default = FP)
Obs = Y/N If Y, then the number of observations in each outcome group and the total are displayed. (default = Y)
LL = Y/N If Y, then -2log(likelihood) is displayed. (default = N)
HL = Y/N If Y, then the Hosmer-Lemeshow p-value is displayed along with the number of groups of estimates
(default = N).

Special features:
The macro will automatically test a quadratic term for numeric (non-class) predictors with more than two values.
The macro will automatically output odds ratios that have been rescaled until the units > 1/10 of the IQR.

The macros called by this macro are %words, %checkvar, %univariate, and %quartiles. The global macro Model
is incremented by one with each call to the logistic macro. The model number is printed and/or saved to the output file.
To reset the model counter to start at one, use the command: [percent sign]Let _Model = 0;

filename words URL "http://ctspedia.org/twiki/pub/CTSpedia/UtilityMacros/words.sas";
%include words;
filename chkvar URL "http://ctspedia.org/twiki/pub/CTSpedia/UtilityMacros/CheckVar.sas";
%include chkvar;
filename quart URL "http://ctspedia.org/twiki/pub/CTSpedia/UtilityMacros/quartiles.sas";
%include quart;
filename univar URL "http://ctspedia.org/twiki/pub/CTSpedia/UtilityMacros/Univariate.sas";
%include univar;


%macro logistic(DSName=,Outcome=,Predictors=,ClassVar=,Quartiles=,Quad=Y,Out=,Print=N,Include=FP,Obs=Y,LL=N,HL=N);

%Local AFormat BadOutcome ClassVariables Flag Has0 Has1 I J K L MaxAfter MaxBefore NAst NClass NextVar NInt NPred NQuart
NNumVars NumVars NNumVars2 NObs NumVars2 Status TempVars TestVar ThisVar Unique Var VarList VarName VarType ;

%Global _ClassVar _ClassVariables _Error _Model;

Make sure just one leading and trailing blank and replace multiple blanks with single blanks in all lists.
Convert all text to uppercase;
%Let Predictors = %str( )%upcase(%cmpres(&Predictors))%str( );
%Let Quartiles = %str( )%upcase(%cmpres(&Quartiles))%str( );
%Let NQuart = %words(&Quartiles);
%Let Include = %upcase(&Include);
%Let LL = %upcase(&LL);
%Let HL = %upcase(&HL);
%Let Print = %upcase(&Print);
%Let Quad = %upcase(&Quad);
%Let Obs = %upcase(&Obs);

* Look for interactions and make sure asterisks have one blank on each side;
%Let TempVars = &Predictors;
%Let Predictors = ;
%Let NAst = 0; * number of asterisks in predictor list;
%do %while(%index(&TempVars,%str()) > 0);
%Let K = %index(&TempVars,%str());
%Let NAst = %eval(&NAst + 1);
%Let Predictors = &Predictors %substr(&TempVars,1,%eval(&K-1))%str( )%str()%str( );
%Let TempVars = %substr(&TempVars,%eval(&K+1));
%end;
%Let Predictors = &Predictors &TempVars;
%Let Predictors = %str( )%cmpres(&Predictors)%str( );
%Let NPred = %words(&Predictors); * This will count 3 predictors for Age * Sex;

Create a list of class variable names in ClassVariables;
%Let ClassVariables ;
  %Let TempVars = %upcase(&ClassVar)%str( );   * Make a copy that can be modified. Then remove reference values spece.;
  %do %while(%index(&TempVars,%str(%()) > 0);
      %Let K = %index(&TempVars,%str(%());
      %Let L = %index(&TempVars,%str(%)));
      %if &L = 0 %then
        %do;
            %put;
            %put Error: Unbalanced parens in classvar parameter.;
            %put;
            %goto ExitLogistic;
        %end;
      %else
        %do;
            %Let TempVars = %substr(&TempVars,1,%eval(&K-1))%str( )%substr(&TempVars,%eval(&L+1));
            %Let TempVars = %cmpres(&TempVars)%str( );
        %end;
  %end;
  %Let Flag = 0;                         * Flag = 1 indicates we have all the class variables;
  %Let TempVars = %cmpres(&TempVars)%str( );
 
  %do %while(&Flag = 0);
      %Let K = %Index(&TempVars,%str( ));    * look for a blank;
      %if &K = 0 or &K > %length(&TempVars) %then
        %do;
            %Let Flag = 1;
        %end;
      %else
        %do;
            %if &K > 1 %then
              %do;
              %Let VarName = %substr(&TempVars,1,%eval(&K-1));
              %Let ClassVariables = &ClassVariables%str( )&VarName;
              %end;
            %Let K = %eval(&K+1);
            %if &K < %length(&TempVars) %then
%do;
%Let TempVars = %substr(&TempVars,&K); * Delete this portion of TempVars;
%Let TempVars = %cmpres(&TempVars)%str( );
%end;
%else %Let TempVars=;
%end;

%end;
%Let ClassVariables = %str( )%cmpres(&ClassVariables)%str( );
%Let NClass = %words(&ClassVariables);

* Create a list of unique variable names in the predictor list;
%Let Unique ;
  %do i = 1 %to &NPred;
      %Let ThisVar = %scan(&Predictors,&i);
      %Let K = %index(&Unique,%str( )&ThisVar%str( ));
      %if &K = 0 %then
        %do;
            %Let Unique = &Unique%str( )&ThisVar%str( );
        %end;
  %end;
  * Check that the outcome and all variables in the predictor list are in the data set;
  %Let VarList = %upcase(%cmpres(&Outcome%str( )&unique));
  %CheckVar(CVDsname=&DSName,CVVars=&Varlist,Message=Y);
  %if &_Error = 1 %then %goto ExitLogistic;

  * Check that outcome is not in the predictor list;
  %if %index(%str( )%upcase(&Predictors)%str( ),%str( )%upcase(&Outcome)%str( )) > 0 %then
      %do;
        %put;
        %put Error: Outcome variable cannot be in the list of predictors.;
        %put;
        %goto ExitLogistic;
      %end;

  * Check that variables in the ClassVariables and Quartiles lists are all in the Predictor list;
  %Let VarList = %str( )%cmpres(&ClassVariables &Quartiles)%str( );
  %do i = 1 %to %words(&Varlist);
      %Let TestVar = %scan(&Varlist,&i);
      %Let K %Index(&Predictors,%str( )&TestVar%str( ));
%if &K = 0 %then
%do;
%put;
%put Error: Variable &TestVar is not in the list of predictors.;
%put;
%goto ExitLogistic;
%end;
%end;

* Check that variables in the predictor list that are not in ClassVariables are in fact numeric.
Variables in main effects and interaction terms will be checked more than once, but that is ok;
%do i = 1 %to %eval(&NPred - &NAst);
%Let TestVar = %scan(&Predictors,&i);
%Let K = %Index(&ClassVariables,%str( )&TestVar%str( ));
%if &K = 0 %then
%do;
data _null;
set &dsname (obs = 1);
VarType = Vtype(&TestVar);
call symput('VarType',VarType);
run;
%if &VarType ne N %then
%do;
%put;
%put Error: Variable &TestVar is not numeric and is not in the list of class variables.;
%put;
%goto ExitLogistic;
%end;
%end;
%end;

* Create a a series of macro variables LogDV1,_LogDV2,...,_LogDV&NPred that will be 1 if that numeric variable is a date variable
otherwise 0.;
data _null;
set &dsname (obs = 1);
%do i = 1 %to %eval(&NPred - &NAst);
%Let ThisVar = %scan(&Predictors,&i);
Fmt = VFormatN(&ThisVar);
Call Symput("_LogDV&I",0);
if Fmt in('DATE','NENGO','DDMMYY','JULIAN','MMDDYY','YYMMDD','WEEKDATE','WEEKDATX','WORDDATE','WORDDATX') then Call Symput("_LogDV&I",1);
%end;
run;

* Create dataset nomiss that eliminates all observations with any missing data on the predictors or the outcome.
Also check that the outcome variable has only values of 0 and 1 (and at least one of each).;
data _nomiss;
set &dsname (Keep = &Outcome &Unique) end = done; * Keep only the needed variables for efficiency;
if _n = 1 then
do;
Outcome0 = 0;
_Outcome1 = 0;
_BadOutcome = 0;
end;
retain _Outcome0 _Outcome1 _BadOutcome;
AnyMissing = 0;
if missing(&Outcome) then AnyMissing = 1; * Missing() returns 1 if either a character or numeric var is missing;
%do i = 1 %to %eval(&NPred - &NAst);
%Let ThisVar = %scan(&Predictors,&i);
if missing(&ThisVar) then AnyMissing = 1;
%end;
if AnyMissing Then Delete;
if &Outcome = 0 then _Outcome0 = _Outcome0 + 1;
if &Outcome = 1 then _Outcome1 = _Outcome1 + 1;
if &Outcome ne 0 and &Outcome ne 1 then _BadOutcome = 1;
if done then
do;
call symput('BadOutcome',_BadOutcome);
call symput('Has0',_Outcome0);
call symput('Has1',_Outcome1);
end;
run;
%if &BadOutcome = 1 %then
%do;
%put;
%put Error: Outcome variable &Outcome has nonmissing values other than 0 and 1;
%put
%goto ExitLogistic;
%end;
%if &Has0 = 0 %then
%do;
%put;
%put Error: Outcome variable &Outcome has no zeros;
%put;
%goto ExitLogistic;
%end;
%if &Has1 = 0 %then
%do;
%put;
%put Error: Outcome variable &Outcome has no ones;
%put;
%goto ExitLogistic;
%end;

* Create a list of numeric predictors that are not class or quartile variables;
%Let VarList = %str( )%cmpres(&ClassVariables &Quartiles)%str( );
%Let NumVars =;
%do i = 1 %to %words(&unique);
%Let ThisVar = %scan(&unique,&i);
%Let K = %Index(&VarList,%str( )&ThisVar%str( ));
%if &K = 0 %then
%do;
%Let NumVars = &NumVars%str( )&ThisVar;
%end;
%end;
%Let NumVars = %str( )%cmpres(&NumVars)%str( );
%Let NNumVars = %words(&NumVars);

* Check that each of these numeric predictors has more than two levels.
Put numeric variables with more than two levels into &NumVars2;
%Let NumVars2 =;
%if &NNumVars > 0 %then
%do i = 1 %to &NNumVars;
%Let ThisVar = %scan(&NumVars,&i);
proc freq noprint data = _nomiss; tables &ThisVar/out=_counts;
run;
data _null;
set counts (obs = 1) nobs = last;
call symput('NObs',Last);
run;
%if &NObs > 2 %then %Let NumVars2 = &Numvars2%str( )&ThisVar;
%end;
%Let NumVars2 = %str( )%cmpres(&NumVars2)%str( );
%Let NNumVars2 = %words(&NumVars2);

* Create squared variables for all variables in NumVars2;
%if NNumVars2 > 0 and &Quad = Y %then
%do;
data _nomiss;
set _nomiss;
%do i = 1 %to &NNumVars2;
%Let ThisVar = %scan(&NumVars2,&i);
_&ThisVar.2 = &ThisVar * &ThisVar;
%end;
run;
%end;

* Create quartile variables and add them to the classvar and classvariables lists;
%if &NQuart > 0 %then
%do;
%Let _ClassVar = &ClassVar;
%Let _ClassVariables = &ClassVariables;
%Quartiles(QDsname=_nomiss,QVars=&Quartiles)
%Let ClassVar =&_ClassVar;
%Let ClassVariables = &_ClassVariables;
%end;

%Let NClass = %eval(&NClass + &NQuart);

* Now fit the basic model without quadratic terms;
ods listing close;
proc logistic data = _nomiss order = internal descending;
%if %length(&ClassVar)> 0 %then %do; class &classvar/param=ref order=internal desc; %end;
model &Outcome = &Predictors/lackfit;
ods output ConvergenceStatus = _cs;
ods output FitStatistics = _fs;
ods output LackFitChiSq = _hl;
ods output ParameterEstimates = _pe;
%if %length(&Classvar) > 0 %then %do; ods output Type3 = _t3; %end;
ods output ResponseProfile = _rp;
run;

data _null;
set cs;
call symput('Status',Status);
run;

* Get frequencies and percentages for class variables. Do this even if Include = N so that we get the reference levels.;
%if &NClass > 0 %then
%do;
proc datasets nolist; delete _allcounts; quit;
%do i = 1 %to &NClass;
%Let ThisVar = %scan(&ClassVariables,&i);
proc freq data = _nomiss noprint; tables &ThisVar * &Outcome/out = _counts;
proc sort data = _counts; by &ThisVar &Outcome;
data _null;
set Counts (obs = 1);
Call symput('AFormat',VFormat(&ThisVar));
run;
data _counts2; * reduce _counts to one obs per value of thisvar;
set _counts (drop = percent); by &ThisVar;
if first.&ThisVar then Total = 0;
retain Total;
Total = Total + count;
if last.&ThisVar then
do;
Length Var $ 32 ClassVal0 $ 50 ;
Var = "&ThisVar";
ClassVal0 = Trim(Left(Put(&ThisVar,&AFormat)));
Freq = count;
if &Outcome = 0 then Freq = 0; * if last value of &This var is 0 then no events;
if Total > 0 then Percent = 100 * Freq / Total;
Keep Var ClassVal0 Freq Percent Total;
output;
end;
run;
proc append base = _allcounts data = _counts2;
run;
%end;
* Remember the order in _allcounts;
data _allcounts;
set _allcounts;
Order = _n;
run;
data pe; * remember original order of variables;
set _pe;
Length PrevVar $ 32;
if _n = 1 then OrderBetween = 0;
Retain OrderBetween PrevVar;
if n = 1 or Variable ne PrevVar then OrderBetween = OrderBetween + 1;
PrevVar = Variable;
Length Var $ 32;
Var = upcase(Variable);
run;
proc sort data = pe; by var classval0;
proc sort data = _allcounts; by var classval0;
data _pe2 (drop = PrevVar);
merge _pe _allcounts; by var classval0;
run;
proc sort data = _pe2; by Order;
data _pe3 (drop = Order);
set _pe2;
Length FreqPct $ 40;
if Freq ne . then
do;
%if &Include = F %then %do; FreqPct = trim(left(put(Freq,8.)))||'/'||trim(left(put(total,8.))); %end;
%if &Include = P %then %do; FreqPct = Trim(left(put(percent,5.1)))||'%'; %end;
%if &Include = FP %then %do; FreqPct = trim(left(put(Freq,8.)))||'/'||trim(left(put(total,8.)))||' ('||trim(left(put(percent,5.1)))||'%)'; %end;
if index(Variable,'*') > 0 then FreqPct = '';
end;
run;
ods listing close;
* If there are numeric predictors, then use FreqPct to hold the overall proportion (%) of events;
%if &NNumVars > 0 %then
%do;
data _null;
set rp;
Length FreqPct $ 40;
if _n = 1 then Freq = Count;
retain Freq;
if n = 2 then
do;
Percent = 100 * Freq / (Count + freq);
%if &Include = F %then %do; FreqPct = trim(left(put(Freq,8.)))||'/'||(left(put(count+Freq,8.))); %end;
%if &Include = P %then %do; FreqPct = trim(left(put(Percent,5.1)))||'%'; %end;
%if &Include = FP %then %do; FreqPct = trim(left(Put(Freq,8.)))||'/'||trim(left(put(count+freq,8.)))||' ('||trim(left(put(percent,5.1)))||'%)'; %end;
call symput('Overall',FreqPct);
end;
run;

data pe3;
set _pe3;
if FreqPct = '' and Var ne 'INTERCEPT' and index(var,'*') = 0 then FreqPct = "&Overall"; * must be a numeric variable if freqpct is blank;
run;

%end;

* Create an order within each variable so the levels can later be sorted properly;
proc sort data = _pe3; by Var;
data _pe3;
set _pe3; by Var;
if first.Var then OrderWithin = 0;
OrderWithin = OrderWithin + 1;
Retain OrderWithin;
run;
proc sort data = _pe3; by var descending OrderBetween;
data _pe3 (drop = SaveOrderBetween SaveVariable);
set _pe3; by var;
Length SaveVariable $ 32;
if first.var then do; SaveOrderBetween = OrderBetween; SaveVariable = Variable; end;
retain SaveOrderBetween SaveVariable;
if OrderBetween = . then OrderBetween = SaveOrderBetween;
if Variable = '' then Variable = SaveVariable;
if df = . then df = 1; * must be the reference category;
run;
%end;
%else
%do;
data _pe3;
set _pe;
Length Var $ 32 FreqPct $ 40;
FreqPct = '';
Var = upcase(Variable);
OrderBetween = _n;
OrderWithin = 1;
run;
%end;

* For numeric non-quartile variables, add quadratic term to model to get p-value for quadratic trend;
%if &NNumVars2 > 0 %then
%do;
%if &Quad = Y %then
%do;
proc datasets nolist; delete allquad; quit;
%do i = 1 %to &NNumVars2;
%Let ThisVar = %scan(&NumVars2,&i);
proc logistic data = _nomiss order = internal descending;
%if %length(&ClassVar)> 0 %then %do; class &ClassVar/param=ref; %end;
model &Outcome = &Predictors _&ThisVar.2;
ods output ParameterEstimates = _pe;
run;
data oneline;
set _pe (rename = (ProbChisq = PQuad));
Length Var $ 32;
if Variable = "_&ThisVar.2" then
do;
Var = "&ThisVar"; * so we can merge back into pe3;
Keep Var PQuad;
output;
end;
run;
proc append base = _allquad data = _oneline force;
run;
%end;
proc sort data = _allquad; by var;
proc sort data = _pe3; by var;
%end;
%univariate(dsname=_nomiss,vars=&NumVars2,Print=N,out=_irq) * get q1 and q3 for possible rescalings;
data _irq (Keep = Var IRQ);
set _irq (keep = Variable q1 q3 rename = (Variable = Var));
Var = upcase(var);
IRQ = q3 - q1;
run;
proc sort data = _irq; by var;
data _pe3;
merge _pe3 _irq; by var;
if OrderBetween > 1 then IsDate = input(Left(symget('_LogDV'||left(trim(put(OrderBetween-1,5.))))),1.);
run;
%if &Quad = Y %then
%do;
data _pe3;
merge _pe3 _allquad; by var;
run;
%end;
proc sort data = _pe3; by OrderBetween OrderWithin;
data _pe4;
set _pe3 (rename = (Variable = OldVar));
Length Variable $ 50;
Variable = OldVar;
Unit = 1;
output;
Count = 0;
if IRQ ne . then
do;
Length BaseVar $ 32;
BaseVar = Variable;
do while(Unit < .1 * IRQ and Count < 2);
Count = Count + 1;
if IsDate = 0 then
do;
Unit = 10 * Unit;
Estimate = 10 * Estimate;
StdErr = 10 * StdErr;
Variable = trim(left(BaseVar))||' (per '||trim(left(put(Unit,8.)))||' units)';
end;
else
do;
if Unit = 1 then NewUnit = 30;
if Unit = 30 then NewUnit = 365;
if Unit = 365 then NewUnit = 3650;
Unit = NewUnit;
Estimate = Unit * Estimate;
StdErr = Unit * StdErr;
if Unit = 30 then Variable = trim(left(BaseVar))||' (per month)';
if Unit = 365 then Variable = trim(left(BaseVar))||' (per year)';
if Unit = 3650 then Variable = trim(left(BaseVar))||' (per decade)';
end;
ProbChiSq = .;
%if &Quad = Y %then %do; PQuad = .; %end;
OrderWithin = OrderWithin + 0.1;

FreqPct = '';
output;
end;

Unit = 1;
do while (Unit > 10 * IRQ and Count < 2);
Count = Count + 1;
Unit = Unit / 10;
Estimate = Estimate / 10;
StdErr = StdErr / 10;
ProbChiSq = .;
%if &Quad = Y %then %do; PQuad = .; %end;
OrderWithin = OrderWithin + 0.1;
Variable = trim(Left(BaseVar))||' (per '||trim(left(put(Unit,10.2)))||' units)';
FreqPct = '';
output;
end;

end;

%end;
%else
%do;
data _pe4;
set _pe3 (rename = (Variable = OldVar));
Length Variable $ 50;
Variable = OldVar;
Unit = .;
run;
%end;
* Merge in Type 3 p-values. Convert Estimates to Odds Ratios;
%if %length(&ClassVar) > 0 %then
%do;
data _t3;
set _t3 (rename = (Effect = OldVar ProbChisq = PType3 DF = DF3));
run;
proc sort data = _t3; by OldVar;
proc sort data = _pe4; by OldVar;
data _pe4 (drop = DF3);
merge _pe4 _t3; by OldVar;
if not(first.OldVar) or df3 = 1 then PType3 = .;
OddsRatioMeasures = .;
LowerCL = .;
UpperCL = .;
if Estimate = . then OddsRatioMeasures = 1; else
do;
if Estimate <25 then OddsRatioMeasures = exp(Estimate);<br /> X1 = Estimate - 1.96 * StdErr;
X2 = Estimate + 1.96 * StdErr;
if StdErr ne . then
do;
if X1 < 25 then LowerCL = exp(X1);
if X2 < 25 then UpperCL = exp(X2);
end;
end;
if Percent in(0,100) then ProbChiSq = .;
;
run;
%end;
%else
%do;
data _pe4;
set _pe4;
OddsRatioMeasures = .;
LowerCL = .;
UpperCL = .;
if Estimate = . then OddsRatioMeasures = 1; else
do;
if Estimate < 25 then OddsRatioMeasures = exp(Estimate);
X1 = Estimate - 1.96 * StdErr;
X2 = Estimate + 1.96 * StdErr;
if StdErr ne . then
do;
if X1 < 25 then LowerCL = exp(X1);
if X2 < 25 then UpperCL = exp(X2);
end;
end;
run;
%end;

* Add, # observations, -2log likelood and Hosmer-Lemeshow statistics;
data _extra1;
set _rp;
Length Variable $ 32 Pre $ 3;
Variable = "Outcome";
if _n = 1 then Pre = '(+)'; else Pre = '(-)';
ClassVal0 =Pre||"&outcome "||outcome||'='||left(trim(put(count,10.)));
OrderBetween = 1000;
OrderWithin = n;
Keep Variable ClassVal0 OrderBetween OrderWithin;
run;

data extra2;
set _fs;
Length Variable $ 32;
if Criterion = '-2 Log L' then
do;
Variable = Criterion;
ClassVal0 = left(trim(Put(InterceptAndCovariates,10.3)));
OrderBetween = 2000;
OrderWithin = 1;
Keep Variable ClassVal0 OrderBetween OrderWithin;
output;
end;
run;

data _extra3;
set _hl;
Length Variable $ 32;
Variable = 'H-L Test';
If df = 0 then ClassVal0 = '< 3 groups'; else ClassVal0 = Trim(Left(Put(df+2,3.)))||' groups';
OrderBetween = 3000;
OrderWithin = 1;
Keep Variable ClassVal0 Probchisq OrderBetween OrderWithin;
run;

%if &Status ne 0 %then
%do;
data _extra4;
Length Variable $ 32;
Variable = 'Convergence problems';
ClassVal0 = "Status = %trim(&Status)";
OrderBetween = 4000;
OrderWithin = 1;
run;
%end;

data _pe4;
set
%if &Obs = Y %then %do; _extra1 %end;
%if &LL = Y %then %do; _extra2 %end;
%if &HL = Y %then %do; _extra3 %end;
%if &Status ne 0 %then %do; _extra4 %end;
_pe4 ;

run;

%if %symexist(_Model) = 1 %then
%do;
%Let _Model = %eval(&_Model + 1);
%end;
%else
%do;
%Let _Model = 1;
%end;
proc sort data = _pe4; by OrderBetween OrderWithin;

proc format; value _pfmt . = ' ' low-.00995 = [pvalue6.4] .00995-.0995 = [6.3] .0995-1 = [6.2];
value _orfmt . = ' ' low-2 = [6.2] 2-20 = [6.1] 20-high = [6.0];
value _missint . = ' ' other = [3.];
data _pe5 (drop = pre post);
set _pe4 end = done; by OrderBetween;
if _n = 1 then do; MaxBefore = 0; MaxAfter = 0; end; * find max digits before, after slash so freq could be aligned;
retain MaxBefore MaxAfter;
Length ProbChiSqAlpha P3Alpha PQuadAlpha $ 6;
ProbChiSqAlpha = left(put(ProbChiSq,_pfmt.));
%if &NClass = 0 %then %do; PType3 = .; *Length ClassVal0 $ 40; %end;
P3Alpha = left(Put(PType3,_pfmt.));
%if &NNumVars = 0 or &Quad ne Y %then %do; PQuad = .; %end;
PQuadALpha = left(put(PQuad,_pfmt.));
if OrderBetween = 2 and OrderWithin = 1 then Model = &_Model;
%if (&Include = F or &Include = FP) %then
%do;
K = index(FreqPct,'/');
if K > 0 then
do;
Pre = K-1;
L = index(Substr(FreqPct,K+1),' ');
If L > 0 then Post = L - 1; else Post = 0;
If Pre > MaxBefore then MaxBefore = Pre;
if Post > MaxAfter then MaxAfter = Post;
end;
%end;
label ProbChiSqAlpha = 'P-Value'
ProbChiSq = 'P-Value'
PType3 = 'Type 3 P-Value'
P3Alpha = 'Type 3 P-Value'
PQuad = 'Quadratic P-Value'
PQuadAlpha = 'Quadratic P-Value'
ClassVal0 = 'Value'
LowerCL = 'Lower 95% CI'
UpperCL = 'Upper 95% CI'
%if &Include = FP %then %do; FreqPct = "# &outcome/# subgroup (% &outcome)" %end;
%if &Include = F %then %do; FreqPct = "# &outcome/# subgroup" %end;
%if &Include = P %then %do; FreqPct = "% &outcome" %end;
;
output;
if last.OrderBetween then
do;
Variable = '';
ProbChiSqAlpha = '';
P3Alpha = '';
PQuadAlpha = '';
FreqPct = '';
OddsRatioMeasures = .;
LowerCL = .;
UpperCL = .;
ClassVal0 = '';
Model = .;
output;
end;
if done then
do;
call Symput('MaxBefore',MaxBefore);
call Symput('MaxAfter',MaxAfter);
end;
run;

%if (&Include = F or &Include = FP) %then
%do;
data pe5 (drop = K L M MaxBefore MaxAfter);
set _pe5;
K = index(FreqPct,'/');
if K > 0 then
do;
FreqPct = Repeat(' ',&MaxBefore-K+1)||FreqPct;
%if &Include = FP %then
%do;
K = index(FreqPct,'/');
L = index(Substr(FreqPct,K+1),' ');
M = index(Substr(FreqPct,K+1),'(');
if L > 0 then FreqPct = Substr(FreqPct,1,K+L-1)||Repeat(' ',&MaxAfter-L+1)||Substr(FreqPct,M+K);
%end;
end;
run;
%end;

data _pe5;
set _pe5;
OddsRatioAlpha = left(put(OddsRatioMeasures,_orfmt.));
UpperCLAlpha = left(put(UpperCL,_orfmt.));
LowerCLAlpha = left(Put(LowerCL,_orfmt.));
label OddsRatioAlpha = 'Odds Ratio'
UpperCLAlpha = 'Upper 95% CI'
LowerCLAlpha = 'Lower 95% CI';
if Variable = 'Intercept' then delete;
run;

%if &Print = Y %then
%do;
ods listing;
proc print data = _pe5 noobs label uniform; var
OddsRatioAlpha LowerCLAlpha UpperCLAlpha ProbChiSqAlpha
%if &NNumVars2 > 0 and &Quad = Y %then %do; PQuadAlpha %end;
%if &NClass > 0 %then %do; P3Alpha %end;
;
id Model variable
%if &Obs = Y or &HL = Y or &LL = Y or %length(&ClassVar) > 0 %then %do; classval0 %end;
%if (&Include = F or &Include = P or &Include = FP) and %Length(&ClassVar) > 0 %then %do; FreqPct %end;
;
format Model _missint.;
run;
%end;

%if &out ne %then
%do;
data &out (Type = 'Logistic' drop = OldVar BaseVar Var IRQ Count X1 X2 ClassVal0);
set _pe5;
Length Value $ 40;
%if &NClass > 0 or &Obs = Y or &HL = Y or &LL = Y %then %do; Value = ClassVal0; %end;
if Variable = '' then delete;
%if &NNumVars = 0 %then %do; PQuad = .; %end;
%if &NClass = 0 %then %do; FreqPct = ''; Total = .; Freq = .; Percent = .; PType3 = .; %end;
Status = &Status;
Model = &_Model;
Length VarType $ 20 Dependent $ 32;
Dependent = "&Outcome";
if Variable in('Outcome','-2 Log L','H-L GOF','Convergence problems') then VarType = Variable; else VarType = 'Logistic Model';
run;
%end;

%ExitLogistic:
proc datasets nolist; delete _allcounts _allquad _counts _counts2 _cs _extra1 _extra2 _extra3 _extra4
_fs _hl _irq _nomiss _oneline _pe _pe _pe2 _pe3 _pe4 _pe5 _pe5 _rp _test _t3; quit;

%mend logistic;
*Stata-Code - Attachment
Stata-Code
Other Code - Attachment
Other Code
Called Data Manipulation
Called Tool/Utility
Called Checking Macro
Called Other
Creation-Date
Revision-Date
Example Code
Example Output
SAS Examples
SAS_Example_Code
SAS_Example_Output
R Examples
R_Example_Code
R_Example_Output
Stats Examples
Special Features
Special Features Attached
Special Features Text
Notes1-Legend Section with calls to needed macros
Notes1
Notes2-Legend Macro call with fully specified parameters
Notes2
Notes3-Legend Section with definitions of global macro parameters
Notes3
Notes4-Legend Section with definitions of local macro variables
Notes4
See Also
Checklists
Stat Tools Disclaimer
Discliamer The views expressed within CTSpedia are those of the author and must not be taken to represent policy or guidance on the behalf of any organization or institution with which the author is affiliated.
Permission Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF ERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT OLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Reference Image
This topic: CTSpedia > WebHome > StatToolsHome > StatToolsTopic054
Topic revision: 21 Jun 2012, MaryBanach
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback