How to arrange data in rows and columns when inputting when you only have totals?

[akimme]

Hi everyone, I'm trying to:

1. run a chi square on proportions of people who did and did not have a cardiovascular (CVD) event, comparing those who did and did not take aspirin
2. and determine if a positive result on one test vs another are associated

I've tried a few ways to arrange the rows and columns, but none of them produce any kind of useful output from PROC FREQ (see spoiler). I only have the totals: I don't have a file with all 19934 lines of data, just the information that 477 of them had a CVD. Is there any way to use SAS for this calculation, short of inputting 40,000 lines of data?

Spoiler

DATA ASPIRIN; 
   INPUT Treat $ Total CVD Divided; 
   DATALINES; 
   1=Aspirin 19934 477 .0239
   2=Placebo 19942 522 .0262
;
run;

PROC FREQ DATA=ASPIRIN;
TABLES Total*CVD /CHISQ RELRISK ;
TITLE 'Relationship between treatment and CVD';
RUN;

DATA TEST; 
   INPUT Q $ Yes; 
*I initially also tried 
putting in 4 variables (Q1andQ2, etc) 
with 1 value each but this also seemed not to work? 
What am I missing?;
   DATALINES; 
Q1andQ2 172 
NEITHER 15  
OnlyQ1 7    
OnlyQ2 6   
;
run;

PROC freq data = test;
run;

Thank you!!

[Reeza]

DATA ASPIRIN; 
   INPUT Treat $ Total CVD Divided; 
   DATALINES; 
   1=Aspirin 19934 477 .0239
   2=Placebo 19942 522 .0262
;
run;

data aspirin_long;
set aspirin;
treatment = input(scan(treat, 1, "="), 8.);
Disease=1;
N=CVD;
output;
Disease=0;
N=Total-CVD;
output;
run;

proc format;
value treat_fmt
1 = 'Aspirin'
2 = 'Placebo'
;
value disease_fmt
1 = 'CVD'
0 = 'Non-CVD';
run;

proc freq data=aspirin_long;
table treatment*disease;
weight N;
format treatment treat_fmt. disease disease_fmt.;
run;

You can use the WEIGHT statement to use aggregate data but you do need to have the data structured a bit differently. 

Hopefully this works for you.

---

@akimme wrote:

Hi everyone, I'm trying to:

1. run a chi square on proportions of people who did and did not have a cardiovascular (CVD) event, comparing those who did and did not take aspirin
2. and determine if a positive result on one test vs another are associated

I've tried a few ways to arrange the rows and columns, but none of them produce any kind of useful output from PROC FREQ (see spoiler). I only have the totals: I don't have a file with all 19934 lines of data, just the information that 477 of them had a CVD. Is there any way to use SAS for this calculation, short of inputting 40,000 lines of data?

Spoiler

DATA ASPIRIN; 
   INPUT Treat $ Total CVD Divided; 
   DATALINES; 
   1=Aspirin 19934 477 .0239
   2=Placebo 19942 522 .0262
;
run;

PROC FREQ DATA=ASPIRIN;
TABLES Total*CVD /CHISQ RELRISK ;
TITLE 'Relationship between treatment and CVD';
RUN;

DATA TEST; 
   INPUT Q $ Yes; 
*I initially also tried 
putting in 4 variables (Q1andQ2, etc) 
with 1 value each but this also seemed not to work? 
What am I missing?;
   DATALINES; 
Q1andQ2 172 
NEITHER 15  
OnlyQ1 7    
OnlyQ2 6   
;
run;

PROC freq data = test;
run;

Thank you!!

---

[Tom · Accepted solution]

You need N*M observations to represent data when have two factors that have N and M possible values.

So you need 2*2=4 observations.

You can get it from your source text.

DATA ASPIRIN; 
   INPUT Treat :$20. Total CVD Divided;
   HAS_CVD=1;
   COUNT=CVD;
   output;
   HAS_CVD=0;
   count = total-cvd;
   output;
   keep treat has_cvd count; 
DATALINES; 
1=Aspirin 19934 477 .0239
2=Placebo 19942 522 .0262
;

proc print;
run;

proc freq ;
  tables treat*has_cvd /chisq;
  weight count;
run;

The FREQ Procedure

Statistics for Table of Treat by HAS_CVD

Statistic                     DF       Value      Prob
------------------------------------------------------
Chi-Square                     1      2.0606    0.1511
Likelihood Ratio Chi-Square    1      2.0613    0.1511
Continuity Adj. Chi-Square     1      1.9697    0.1605
Mantel-Haenszel Chi-Square     1      2.0606    0.1512
Phi Coefficient                       0.0072
Contingency Coefficient               0.0072
Cramer's V                            0.0072


       Fisher's Exact Test
----------------------------------
Cell (1,1) Frequency (F)     19457
Left-sided Pr <= F          0.9289
Right-sided Pr >= F         0.0802

Table Probability (P)       0.0091
Two-sided Pr <= P           0.1586

Sample Size = 39876
 

[akimme]

Okay, it looks like the COUNT= step was the big thing I was missing. That worked, thank you so much!