* BASIC PROGRAM TO USE PRINCIPAL COMPONENTS ANALYSIS TO IDENTIFY DIETARY
    PATTERNS.
  SPECIFICATIONS FOR INPUT DATASET:
    ONE OBSERVATION PER PERSON (ID = IDENTIFYING VARIABLE)
    ONE VARIABLE PER FOOD (FOOD1 ... FOOD100)
      GENERALLY, EACH FOOD VARIABLE IS IN FREQUENCY FORMAT, E.G., SVGS/MONTH
;


libname in 'c:\';


* PRINCIPAL COMPONENTS ANALYSIS
  NOTE THAT DATASET INCLUDES ONLY THE VARIABLES TO BE INCLUDED IN THE 
    PRINCIPAL COMPONENTS ANALYSIS
  FOR DESCRIPTION OF OPTIONS, SUCH AS TYPE OF ROTATION, SEE SAS/STAT MANUAL, OR
    HATCHER, LARRY, A STEP-BY-STEP APPROACH TO USING THE SAS SYSTEM FOR FACTOR
    ANALYSIS AND STRUCTURAL EQUATION MODELING, CARY, NC: SAS INSTITUTE INC., 
    1994;

data DATASET1A;  set DATASET1 (drop=ID);  run;
proc corr data=DATASET1A out=CORREL noprint;  run;
proc factor data=CORREL
            method=prin mineigen scree rotate=varimax reorder;
run;


* NUMBER OF COMPONENTS TO RETAIN BASED ON SCREE PLOTS, INTERPRETABILITY, AND 
  OTHER CRITERIA - SEE, FOR EXAMPLE,
    DEVELLIS, ROBERT F., SCALE DEVELOPMENT: THEORY AND APPLICATIONS, NEWBURY 
    PARK, CA: SAGE PUBLICATIONS, INC., 1991.
  IN THIS EXAMPLE, RETAIN 2 COMPONENTS;

proc corr data=DATASET1A out=CORREL noprint;  run;
proc factor data=CORREL
            method=prin scree nfact=2 rotate=varimax reorder;
run;


* SELECT ITEMS POTENTIALLY ASSOCIATED WITH EACH PATTERN, AND
  STANDARDIZE TO ACCOUNT FOR DIFFERENT RANGES OF INTAKE
  FOR EXAMPLE, IF FOODS 1-15 (FOOD1-15) ARE ASSOCIATED WITH PATTERN 1,
    AND FOODS 16-25 (FOOD16-25) ARE ASSOCIATED WITH PATTERN 2 ...
;

data DATASET2;
  set DATASET1;

* PATTERN 1;
    FOOD1_S = FOOD1;  FOOD2_S = FOOD2;  FOOD3_S = FOOD3;
    FOOD4_S = FOOD4;  FOOD5_S = FOOD5;  FOOD6_S = FOOD6;
    FOOD7_S = FOOD7;  FOOD8_S = FOOD8;  FOOD9_S = FOOD9;
    FOOD10_S = FOOD10;  FOOD11_S = FOOD11;  FOOD12_S = FOOD12;
    FOOD13_S = FOOD13;  FOOD14_S = FOOD14;  FOOD15_S = FOOD15;

* PATTERN 2;
    FOOD16_S = FOOD16;  FOOD17_S = FOOD17;  FOOD18_S = FOOD18;
    FOOD19_S = FOOD19;  FOOD20_S = FOOD20;  FOOD21_S = FOOD21;
    FOOD22_S = FOOD22;  FOOD23_S = FOOD23;  FOOD24_S = FOOD24;
    FOOD25_S = FOOD25;

run;

* TO REVERSE SCORING FOR FOOD ITEMS WITH STRONG NEGATIVE LOADINGS
  E.G., IF FOOD14 HAD A STRONG NEGATIVE LOADING ON PATTERN 1;

data DATASET2;
  set DATASET2;
  food14_sr = food14_s * -1;
run;


proc standard data=DATASET2 out=DATASET3 mean=0 std=1;
  var

    FOOD1_S  FOOD2_S  FOOD3_S 
    FOOD4_S  FOOD5_S  FOOD6_S 
    FOOD7_S  FOOD8_S  FOOD9_S 
    FOOD10_S FOOD11_S FOOD12_S
    FOOD13_S FOOD14_S FOOD15_S

    FOOD16_S FOOD17_S FOOD18_S
    FOOD19_S FOOD20_S FOOD21_S
    FOOD22_S FOOD23_S FOOD24_S
    FOOD25_S ;

run;


* CALCULATE PATTERN SCORES AS UNWEIGHTED SUM OF STANDARDIZED INTAKE FREQUENCIES.
  TWO METHODS ARE PRESENTED BELOW - ONE ASSUMING NO MISSING DATA, AND SECOND
  ALLOWING FOR SOME IMPUTATION OF MISSING VALUES FOR INDIVIDUAL FOODS WITHIN A
  PATTERN;


* METHOD 1 (NO MISSING DATA);

data DATASET4;
  set DATASET3
 (keep=ID
    FOOD1_S  FOOD2_S  FOOD3_S   FOOD4_S  FOOD5_S  FOOD6_S 
    FOOD7_S  FOOD8_S  FOOD9_S   FOOD10_S FOOD11_S FOOD12_S
    FOOD13_S FOOD14_SR FOOD15_S
    FOOD16_S FOOD17_S FOOD18_S  FOOD19_S FOOD20_S FOOD21_S
    FOOD22_S FOOD23_S FOOD24_S  FOOD25_S );

  PATTERN1 =     
    FOOD1_S  + FOOD2_S  + FOOD3_S  + FOOD4_S  + FOOD5_S  + FOOD6_S +
    FOOD7_S  + FOOD8_S  + FOOD9_S  + FOOD10_S + FOOD11_S + FOOD12_S +
    FOOD13_S + FOOD14_SR + FOOD15_S;

  PATTERN2 =     
    FOOD16_S + FOOD17_S + FOOD18_S + FOOD19_S + FOOD20_S + FOOD21_S
    FOOD22_S + FOOD23_S + FOOD24_S + FOOD25_S ;

run;


* METHOD 2 (TO ACCOUNT FOR MISSING DATA)
  THIS METHOD IMPUTES MISSING VALUES FOR FOODS WITHIN A PATTERN AS THE AVERAGE
    STANDARDIZED INTAKE FOR OTHER NON-MISSING FOODS WITHIN THAT PATTERN, BUT
    ONLY IF MOST (70-75%) OF THE FOOD ITEMS IN THE PATTERN ARE NON-MISSING;

data DATASET4;
  set DATASET3
 (keep=ID
    FOOD1_S  FOOD2_S  FOOD3_S   FOOD4_S  FOOD5_S  FOOD6_S 
    FOOD7_S  FOOD8_S  FOOD9_S   FOOD10_S FOOD11_S FOOD12_S
    FOOD13_S FOOD14_SR FOOD15_S
    FOOD16_S FOOD17_S FOOD18_S  FOOD19_S FOOD20_S FOOD21_S
    FOOD22_S FOOD23_S FOOD24_S  FOOD25_S );

  * COUNT NUMBER OF NON-MISSING FOOD ITEMS WITHIN THE PATTERN;

  COUNTP1 = 0;
  IF FOOD 1_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 2_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 3_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 4_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 5_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 6_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 7_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 8_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 9_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 10_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 11_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 12_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 13_S NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 14_SR NE . THEN COUNTP1 = COUNTP1+1;
  IF FOOD 15_S NE . THEN COUNTP1 = COUNTP1+1;

  * CALCULATE PATTERN SCORE;

  PATTERN1 = 15 * (SUM(FOOD1_S,FOOD2_S,FOOD3_S,FOOD4_S,FOOD5_S,
                       FOOD6_S,FOOD7_S,FOOD8_S,FOOD9_S,FOOD10_S,
                       FOOD11_S,FOOD12_S,FOOD13_S,FOOD14_SR,FOOD15_S)/COUNTP1);

  * IF TOO MANY MISSING VALUES, SET SCORE TO MISSING;

  IF COUNTP1 < 11 THEN PATTERN1 = .;


  * REPEAT FOR SECOND DIETARY PATTERN;

  COUNTP2 = 0;
  IF FOOD 16_S NE . THEN COUNTP2 = COUNTP2+1;
  IF FOOD 17_S NE . THEN COUNTP2 = COUNTP2+1;
  IF FOOD 18_S NE . THEN COUNTP2 = COUNTP2+1;
  IF FOOD 19_S NE . THEN COUNTP2 = COUNTP2+1;
  IF FOOD 20_S NE . THEN COUNTP2 = COUNTP2+1;
  IF FOOD 21_S NE . THEN COUNTP2 = COUNTP2+1;
  IF FOOD 22_S NE . THEN COUNTP2 = COUNTP2+1;
  IF FOOD 23_S NE . THEN COUNTP2 = COUNTP2+1;
  IF FOOD 24_S NE . THEN COUNTP2 = COUNTP2+1;
  IF FOOD 25_S NE . THEN COUNTP2 = COUNTP2+1;

  PATTERN2 = 10 * (SUM(FOOD16_S,FOOD17_S,FOOD18_S,FOOD19_S,FOOD20_S,
                       FOOD21_S,FOOD22_S,FOOD23_S,FOOD24_S,FOOD25_S)/COUNTP2);

  IF COUNTP2 < 7 THEN PATTERN2 = .;

run;