Statistical Disclosure

[Simon Moore]

Dear Statalist,

I am automating some reports using putdocx that are based on clinical data. Reporting these data are subject to requirements concerning statistical disclosure. Counts <10 need to be reported as “<10”, rather than the actual value.

I tried searching Statalist, but “disclosure” is a common term here, so apologies if this has been covered.

My question: is there an easy way to replace values in outputs that meet a certain criterion? For example,

clear all
input str20 condition count
"injury" 20
"overdose" 5
"other" 15
end
table condition, stat(sum count) stat(percent count)

I would like to replace the count 5 with “<10” and percent 12.50 with “<20.00”.

Simon

[Nick Cox]

I haven't heard of a way to do this.

Even your very simple example shows the problem here, which is that suppressing a display of 5 still lets it be inferred because the total will be 40 and the other counts add to 35.

[David Radwin]

suppressing a display of 5 still lets it be inferred because the total will be 40 and the other counts add to 35.

Agreed, which is why some statistical agencies also require "complementary suppression" for published results. The publication below explains how and is also a contender for longest title for a brief.

Chen, X., Bersudskaya, V., & Cubarrubi, A. (2011). Statistical Methods for Protecting Personally Identifiable Information in the Disclosure of Graduation Rates of First-Time, Full-Time Degree-or Certificate-Seeking Undergraduate Students by 2-Year Degree-Granting Institutions of Higher Education (NCES 2012-151). Technical Brief. US Department of Education, National Center for Education Statistics.

One possible workaround, which may not be satisfactory, is to search-and-replace each single digit value, using the "whole words only" setting, in Word or other word processor, replacing it with "<10." Even if feasible, this approach is not ideal because you have to do that nine or ten times and there may be false positive values that should not be suppressed.