# IT; statistics

**URL:** <http://discourse.iapct.org/t/it-statistics/5479>\
**Category:** CSG1994\
**Created:** [July 25, 1994, 1:16pm UTC](http://discourse.iapct.org/t/it-statistics/5479 "1994-07-25T13:16:40Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![William\_T\_Powers2](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/w/c67d28/32.png) [@William\_T\_Powers2](http://discourse.iapct.org/u/William_T_Powers2)\
**Post date:** [July 25, 1994, 1:16pm UTC](http://discourse.iapct.org/t/it-statistics/5479/1 "1994-07-25T13:16:40Z")

</div>

[From Bill Powers (940725.0600 MDT)]

Bill Leach (940724c.xxx) --

I liked your replies to Bob Clark and Martin Taylor. When Martin says

"The claim is that all the information that allows the output to oppose  
the disturbance is obtained through the perceptual input function that  
acts on the fluctuations of the CEV induced by the disturbance. "

he is really declaring his faith that somehow the information needed for  
achieving control is coming from the disturbance, which is the basic S-R  
faith. He goes so far as to say that this amounts to saying that control  
is a physical process and not a magical one. This is claiming far too  
much, in my opinion. We explain the operation of a control system on  
perfectly valid physical grounds, and in the process demonstrate that  
the control system uses only information about the state of the  
controlled variable; it does not require any internal computations that  
deduce the state of the disturbance. The state of the controlled  
variable, furthermore, is not a unique indicator of the state of the  
disturbance, because it is simultaneously affected by the system's own  
output, which in turn is a function of the controlled variable AND the  
reference signal. As far as I am concerned, the control-system analysis  
gives a complete account of how control works and why, leaving nothing  
for information theory to explain. NOTHING. So there.

Your tale of a false-positive drug test is far from unusual. A great  
part of the problem comes from tests for rare conditions -- that is,  
conditions that hold only in a small part of the population. Let's see  
if I can remember how this problem goes:

Suppose we have a test for cancer that is 98% effective both in  
indicating the existence of a cancer and in indicating its absence. That  
is, for people who actually have cancer, it will pick up 98% of the  
cases that later prove to be real cancer, and it will reject 98% of the  
cases that later prove not to have been cancer. The actual incidence of  
cancer in the population in question is known, furthermore, to be 1%:  
100 people of every 10000.

The question is, if the test indicates that you have cancer, what are  
the chances that you actually do have cancer?

The trap here lies in the fact that 99% of the people do not have  
cancer. The test has a false-positive rate of 2%. So in a population of  
10000 people, it will indicate that of the 9900 who do not actually have  
cancer, 198 people have it. It will also pick up 98% of those who really  
do have it, meaning 98 people.

Therefore, if you test positive for cancer, the chances are 2:1 that you  
do NOT have cancer. And this is using a test that can be described as  
"98% accurate."

Apply this to the "second-hand smoke" problem. It is claimed (as the  
news media report it) that 3000 people per year (out of 250,000,000) die  
from the effects of second-hand smoke. That is an incidence of 0.000012.  
A test that determines whether a given person died from this cause might  
possibly, given extreme faith, be 99.9% accurate both ways, false  
positives and false negatives. But a test with 0.1% false positives  
would indicate 250,000 people per year dying of second-hand smoke. What  
would the accuracy have to be for a test that shows only 3000 per year?  
It would have to be 99.99%+, even if most of the positive results were  
false positives. So the claim that 3000 people die every year from  
second-hand smoke is ludicrous. No phenomenon with that low an incidence  
can possibly be verified as existing with any sort of medical test that  
exists now or is likely to be invented.

So how can this silly claim be made? If you read carefully, you will see  
that no authority claims that second-hand smoke causes 3000 deaths per  
year. What is claimed is that in 3000 deaths per year, second-hand smoke  
is determined to be a contributing factor. Determinations don't make  
themselves. The statement is that of all the deaths that do occur, 3000  
of them are claimed by someone to have been contributed to by second-  
hand smoke. It's sort of like an FBI file: the allegation becomes part  
of the factual record. There is certainly no claim that if non-smokers  
were totally protected against second-hand smoke for the rest of their  
lives, the death rate from cancer would be measurably lower. To say that  
second-hand smoke was a contributing factor is not to say that the  
person wouldn't have died anyway, at the same time. The implication is  
that there were \_other\_ contributing factors as well, such as heart  
failure or bullets. You'll see this sort of weaseling all the time.  
Doctors say that excess iron can cause heart problems, but they hasten  
to say that there's no reason to reduce your high iron intake, because  
they have no proof that reducing iron intake will help any with reducing  
heart problems. So go on eating your spinach, one of them said.

All this medicostatistical mumbo-jumbo about health is simply a vehicle  
for forcing one's preferences on others. Medicine has had very little  
impact on increasing longevity; the primary increases have been due to  
increases in cleanliness of homes, water supplies, and food, and to a  
few antibiotics (which are now becoming ineffective as bacterial control  
systems learn to cope with them). Most of the vast costs of medicine are  
due to taking credit for cures that would have occurred without  
treatment, offering false hopes to a nation of hypochondriacs, and  
keeping people miserably alive for a short time against the natural  
tendency to give up when all the systems are failing.

I suppose I shouldn't be so shy about saying what I really think.

> **···**
>
> ---------------------------------------------------------------------  
> Best to all,
> 
> Bill P.

---

<div class="post-metadata">

**Author:** ![Bill\_Leach1](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/848f3c/32.png) [@Bill\_Leach1](http://discourse.iapct.org/u/Bill_Leach1)\
**Post date:** [July 26, 1994, 2:32am UTC](http://discourse.iapct.org/t/it-statistics/5479/2 "1994-07-26T02:32:09Z")

</div>

\<[Bill Leach 940725.19:19 EST(EDT)]

> [Bill Powers (940725.0600 MDT)]

Yes, it is not magic of course and yes since these are finite reaction  
time functions with finite gain changes in the disturbance effect the  
perception changing the error signal value and generating an output that  
not quite "compensates" exactly for the delta perception.

But unfortunately for the idea of extracting useful information  
concerning the disturbance, the "not quite exactly compensates" also  
includes sometimes over compensating, sometimes undercompensating,  
sometimes exactly compensating and if the delta due to disturbance is  
small enough (itself in the noise) no changes will occur.

The real issue is that "of course the effect of the disturbance is 'in'  
the perceptual signal" but with a good high gain control system the  
magnitude is divided by the available gain and is in the noise. Thus,  
impractical for use.

A simple "unity gain op-amp" provides a good example of the problem. Set  
a reference, use a very accurate and solid power supply, stay well within  
the op-amps output ratings and vary the load. Watch the inverting input  
for change. Will you see any change? Well maybe, if your test equipment  
is significantly more sensitive than the op-amp and the noise is not too  
great. Add a few somewhat random variations to the power supply and it  
becomes about hopeless. Of course this example does introduce another  
problem possibly not present in the biological system and that is that  
the noise current of the load itself and the noise of the later stages of  
the op-amp will make the noise a function of output current which means  
that the results will also be a function of the initial conditions.

> Your tale of false-positive drug test is far from unusual. ...

Another example (that strikes at another area of concern for me) is the  
use of those "techniques" with nuclear power in general.

Even the Cherynoble power plant disaster which was a better example of a  
rather extreme force "controlled" society vs. a not quite so extreme one  
as opposed to saying anything about nuclear reactor plants in general.

Regardless, the media "dutifully" spooned out the cancer death  
predictions for the projected health effects estimated to be directly  
attributable to the accident. This was done without including the  
responsible data showing that the estimate was significantly LESS than  
the normal expected variance within that population for the total number  
of deaths expected due to cancer from all causes other than the accident  
much less mentioning that total.

Nor did that same media mention the additional predictions that indicated  
that orders of magnitude greater effects would result from either  
replacement power from coal plants or the really staggering death rate  
estimates based upon increase in rate due to NOT replacing the power loss  
at all.

When there is a release of radioactive material from a plant that exceeds  
the limits set by the government, our high integrety media of course  
always points out that should these same regulations be applied to  
things other than nuclear power plants:

All coal fired plants would be shut down permenantly as their routine  
operational release rate for radioactive nuclieds exceeds those limits by  
orders of magnitude. Most all oil fired plants would likewise be shut  
down. Many gas fired plants as well and all have exceeded the limits  
many times.

All petrol fueled vehicles and stationary engines release radioactive  
materials at a unit energy rate that would be considered catastrophic for  
a nuke plant.

Grand Central Station would be closed as a "dangerous" uncontrolled  
radioactive material surface contamination site as well as for exceeding  
airborne radioactive gas limits.

All geothermal plants would be shut down and their operators bankrupted  
by the fines.

And finally the same intelligent folks that brought you "pi=22/7" by  
force of law would mandate that as the largest clandistine and unlicensed  
radioactive material mover on the North American continent, the  
Mississippi river would be required to not transport radioactive material  
anymore.

-bill