Article View: net.general
Article #4697Fallacies of testing
From: Larry Yang
Date: Mon, 22 Sep 1986 18:52
Date: Mon, 22 Sep 1986 18:52
65 lines
3464 bytes
3464 bytes
Now that we're on the subject of drug testing, I thought that it would be
appropriate to share some information I have on testing in general.
Anyone who has had any training in statistics or probability will probably
be familiar with this stuff already, so you needn't read on.
Let us assume that we have this wonderful test that:
will test positive 99% of the time if the subject has the condition
will test negative 99% of the time if the subject doesn't have it
Let us suppose that in the population 1% of the people have the condition.
Let there be 10,000 people in our study. Below is a chart showing
have/don't have vs. positive test/negative test:
#have | #don't have
----------------
pos. 99 | 99
----------------
neg. 1 | 9801
----------------
total 100 9900
Note that the columns should add up to be 100/9900, since we assumed
that 1% of our population had the condition.
The interesting result is that of the people who tested positive,
99 had the condition and 99 didn't. Thus, if a person tested positive,
***** HE/SHE HAS A 50/50 CHANCE OF HAVING THE CONDITION!! *****
What does this mean? Basically, the moral of the story is that
no matter how good your test is, if the condition you are testing for
is sufficiently rare, you will have a very large number of
false positives (People who test positive but do not have the condition).
In medicine, doctors face this dilemma all the time. The perform a wonderful
test for a rare disease and end up performing surgery unnecessarily
may times.
How does this relate to drug testing? I think that it's fairly
obvious. In this example, I chose a really good test to test for
a fairly common condition, and the result was that there was a 50%
chance of being wrong. Drug abuse is a condition that hopefully
exists in less that 1% of the population; that would make the chance
of error even greater.
Just a random thought.....
--
==============================================================================
/-\ /\ |
-- Larry Yang __|_/ | |
Sun Microsystems, Inc. | | | /--\ /\/-\/--\
Mountain View, CA 94043 /-|\ \ / | | || || |
lyang@sun.COM \_/ \ \__/| /--/\/| \/--| /
{ihnp4|hplabs|seismo}!sun!lyang __|/ __|/
/ / / /
(a REAL signature) \_/ \_/
==============================================================================
"If you don't like the news, go out and make some of your own..."
"If you don't like the junk on usenet, post junk of yo.. <FLAME> AAAAGH!!.."
==============================================================================
Technical Details
Message-ID:
<7498@sun.uucp>Path:
arc1-ca.pugleaf.net!utzoo!watmath!clyde!cbatt!ihnp4!qantel!lll-lcc!pyramid!decwrl!sun!lyangShow Full Headers
Complete Article Headers
Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: arc1-ca.pugleaf.net!utzoo!watmath!clyde!cbatt!ihnp4!qantel!lll-lcc!pyramid!decwrl!sun!lyang From: lyang@sun.uucp (Larry Yang) Newsgroups: net.general Subject: Fallacies of testing Message-ID: <7498@sun.uucp> Date: Mon, 22-Sep-86 14:52:50 EDT Article-I.D.: sun.7498 Posted: Mon Sep 22 14:52:50 1986 Date-Received: Sun, 28-Sep-86 20:45:28 EDT Reply-To: lyang@sun.UUCP (Larry Yang) Followup-To: net.followup Organization: Sun Microsystems, Inc. Lines: 65