Splitting Longitudinal Data into Training & validation Sets

[anu1999]

Hi,

Can anyone please suggest a way how can I randomly split the longitudinal data into training (60%) and validation (40%).

In my case, I'd like split on a data set where each individual has more than one observation, in such a way that if an individual is in one of the training/validation sets, then all of their observations are in that same set.

Example data (BMILONG) below:

I want to split on BMILONG dataset generated in the second step.

DATA BMI;
CALL STREAMINIT(12345);
DO ID = 1 TO 100;
GENDER=(MOD(ID,2)=0);
TREAT=( ID>50);
BASELINE = ROUND(RAND('NORMAL',35,2),.1);
IF GENDER=1 AND TREAT=0 THEN DO;
GROUP = 'FEMALE - PLACEBO';
MONTH3 = ROUND(BASELINE - .25 + RAND('NORMAL',0,1),.1);
MONTH6 = ROUND(MONTH3 + .25 + RAND('NORMAL',0,1),.1);
MONTH9 = ROUND(MONTH6 - .25 + RAND('NORMAL',0,1),.1);
MONTH12= ROUND(MONTH9 + .25 + RAND('NORMAL',0,1),.1);
END;
IF GENDER=0 AND TREAT=0 THEN DO;
GROUP = 'MALE - PLACEBO';
MONTH3 = ROUND(BASELINE - 1 + RAND('NORMAL',0,1),.1);
MONTH6 = ROUND(MONTH3 - 1 + RAND('NORMAL',0,1),.1);
MONTH9 = ROUND(MONTH6 + 1 + RAND('NORMAL',0,1),.1);
MONTH12= ROUND(MONTH9 + 1 + RAND('NORMAL',0,1),.1);
END;
IF GENDER=0 AND TREAT=1 THEN DO;
GROUP = 'MALE - TREAT';
MONTH3 = ROUND(BASELINE - 1.5 + RAND('NORMAL',0,1),.1);
MONTH6 = ROUND(MONTH3 - 1.5 + RAND('NORMAL',0,1),.1);
MONTH9 = ROUND(MONTH6 - 1.5 + RAND('NORMAL',0,1),.1);
MONTH12= ROUND(MONTH9 - 1.5 + RAND('NORMAL',0,1),.1);
END;
IF GENDER=1 AND TREAT=1 THEN DO;
GROUP = 'FEMALE - TREAT';
MONTH3 = ROUND(BASELINE - 1 + RAND('NORMAL',0,1),.1);
MONTH6 = ROUND(MONTH3 - 1 + RAND('NORMAL',0,1),.1);
MONTH9 = ROUND(MONTH6 - 1 + RAND('NORMAL',0,1),.1);
MONTH12= ROUND(MONTH9 - 1 + RAND('NORMAL',0,1),.1);
END;
OUTPUT;
END;
RUN;

DATA BMILONG;
SET BMI;
TIMEPT=0; BMI=BASELINE; OUTPUT;
TIMEPT=3; BMI=MONTH3; OUTPUT;
TIMEPT=6; BMI=MONTH6; OUTPUT;
TIMEPT=9; BMI=MONTH9; OUTPUT;
TIMEPT=12; BMI=MONTH12; OUTPUT;
DROP BASELINE MONTH:;
RUN;

[PGStats · Accepted solution]

Thanks for a well presented problem. Here is a solution using surveyselect:

proc surveyselect data=bmiLong seed=8588 samprate=40 outall
    out=bmiLongGroups(rename=selected=validation) ;
cluster id;
run;

[anu1999]

This works perfect!! Appreciate your help.