proc freq: "exact riskdiff" vs "exact riskdiff(method=fmscore)"

MRB3855 — Thu, 15 Nov 2018 11:58:09 GMT

Good day All: Suppose I have the data (x_data) below, and further suppose I want to calculate confidence intervals for the difference of two binomial proportions (treat A vs treat B for each trial, respectively) via proc freq as below:

proc freq data=x_data;
by trial;
tables treat*outcome /riskdiff(cl=(exact));
weight count;
exact riskdiff;
run;

proc freq data=x_data;
by trial;
tables treat*outcome /riskdiff(cl=(exact));
weight count;
exact riskdiff(method=fmscore);
run;

trial treat outcome count

1	A	x	56
1	A	y	14
1	B	x	48
1	B	y	32
2	A	x	9
2	A	y	1
2	B	x	3
2	B	y	7
3	A	x	10
3	A	y	0
3	B	x	0
3	B	y	20

I expected to get different results from the two different procedures, but I don't. The confidence intervals are exactly the same. My understanding is the former should use the Santner-Snell method ("By default, PROC FREQ uses the unstandardized risk difference as the test statistic in the confidence limit computations.") and the latter should use the Chan-Zhang method ("If you specify the RISKDIFF(METHOD=FMSCORE) option, the procedure uses the Farrington-Manning score statistic (Chan and Zhang 1999).").

What gives? BTW, I'm currently using SAS 9.4 TS Level 1M5.

Thanks

Re: proc freq: "exact riskdiff" vs "exact riskdiff(method=fmscore)"

MRB3855 — Thu, 15 Nov 2018 13:16:13 GMT

Ah, I've found the problem (via a careful read of log file). Stat v 14.3 switches the default from previous versions.

From the log file:

NOTE: The default method for EXACT RISKDIFF is now METHOD=SCORE. To use the previous default
      method (in SAS/STAT 14.2 and earlier releases), you can specify EXACT
      RISKDIFF(METHOD=NOSCORE).

Problem solved!

As you were 😉

topic proc freq: "exact riskdiff" vs "exact riskdiff(method=fmscore)" in New SAS User

proc freq: "exact riskdiff" vs "exact riskdiff(method=fmscore)"

Re: proc freq: "exact riskdiff" vs "exact riskdiff(method=fmscore)"