Question:
We can look at this from the other direction too. What would the
accuracy of the test have to be if we wanted to have a 99% probability
that the person tesing positive in the above scenario is truly an
illegal drug user? Out of the 9,900 people in the non-illegal-
drug using group who are tested we can accept no more than one(1)
false positive so that our ratio of true positives to false is
99 to 1, or approximately 99%. Well to get this the testing
procedure can accept no more false positives than 1/9900 or
.01% false positives. Or to put it in the way in which we
began the test has to have an accuracy of 99.9899% in order
for a company to be able to say that a person testing positive
from the above hypothetical has a 99% probability of actually
being an illegal drug user. Now I wonder how many drug testing
companies can claim this kind of accuracy day in and day out.
The importance of this can not be emphasized enough. In the
above example we knew exactly what the distribution of illegal
drug users was. We knew exactly what the accuracy of the test was.
In real life, we don't know shit. The drug testing companies
don't seem to be willing to give us accuarcy information, and
likewise how are we to determine the % of people who use illegal
drugs on the job?
Answer:
In Greg's model we assume 10% of the work force use drugs. We assume 1%
of the workforce will test positive even if they are clean. Unless
idiosyncratic individual effects resulting in two false positives
account for the majority of positive tests, then I fail to see how a
test that separates a single group of 10,000 employees into two groups,
one of 9802 containing one undetected drug user and another group of 198
containing even 80 suspected drug users is equivalent to flipping a
coin.
(deep breath)
Even if you fire all 198 suspected drug users you would not be reducing
your labor force as drastically as you would if you flipped a coin to
determine each employees fate. I'm not good with numbers but drug
testing might better be compared to having your employees roll dice and
firing those who roll doubles. From a managerial stand point, rolling
dice is less expensive and less intrusive than testing bodily fluids.
No doubt accurate drug testing would fail a cost/benefit analysis.
There are less Orwellian, more effective quality control techniques
anyway.
One factor that is never taken into account when considering the wisdom
of drug testing is the impact of drug testing on employee moral, job
stress and job security. On-the-job-stress is the leading cause of work
place violence and absenteeism due to minor illness. Job insecurity
destroys quality and productivity.
I agree so much with what follows that I'll repost it without comment
except to reiterate that, using Greg's model, drug testing is better at
detecting drug users than flipping a coin. I agree that testing cannot
be motivated by productivity and safety concerns.
While the second test would supposedly mathematically mean that there
would be 98 true positives and one false positive, I strongly doubt it
for one reason: Why wouldn't they just have you take the test twice in
the first place if that was all they had to do to get from a mediocre
99% accuracy to an awesome 99.99%? This is just speculation, but I
suppose whatever screwed with someone's results the first time would do
it the second time as well.
Not that I think drug testing is worthwhile anyway... the whole concept
of drug testing employees because supposedly the users wouldn't work as
hard just doesn't make since, as an employee could be fired for
incompetence anyway.
