Testing, Testing

[From Rick Marken (951030.0900)]

CHUCK TUCKER (951030.10:50 EST)


--select certain words (the same ones all of the time) that another speaks
and "spell" them "out loud" after each time the word is spoken by the other.

The Test is not just a matter of "applying disturbances". The first, and most
important, thing to do when testing for controlled variables is GUESS what
variable is under control. In the example above it is not clear why spelling
words out loud should be a disturbance; are you guessing that people are
controlling for responsive talk or for being able to hear the words they are
saying or for not being mocked or what? If you don't guess at a controlled
variable it is impossible to tell whether a disturbance is effective or
not. How do you know whether or not a disturbance like "spelling out loud" is
having the expected effect on a controlled variable if you don't know what
the controlled variable might be? It's true that a "response" to your
disturbance (like the subject telling you to "stop!" each time you spell a
word) suggests that spelling out loud is a disturbance; but a disturbance to
what? You can't determine what's under control without guessing first. Also,
you might not even notice a response to you disturbance even when there is
one; if, for example, the spelling is a disturbance only to the speaker's
ability to hear himself then he may simply raise his voice slightly while
you are spelling a word-- a feature of the person's behavior that you might
not even notice.

Since one's first guess about a variable under control is likely to be wrong,
you really can't tell much about what a person is doing (controlling) by just
"applying a disturbance" and seeing if it has an effect. You have to test
your hypothesis about the variable under control by producing many different
disturbances that SHOULD affect that variable if the person is NOT
controlling it. If you guess that the subject is controlling the signal to
noise (S/ N) level of his own speech, for example, then he should compensate
by talking louder (raising the signal level) when there is any added noise
(like spelling some of his words out loud, turning on a fan, driving by the
airport, etc). If (as is likely) it turns out that your guess about what is
controlled is wrong (the disturbance has the expected effect on the
hypothesized controlled variable) you have to abandon the hypothesis and try
a new one; the Test is an ITERATIVE process. The iterative nature of the Test
is captured rather well in your description of testing for the controlled
variable in the "coin game".

Even when your guess about a controlled variable seems right -- a disturbance
has far less than the expected effect on the hypothesized controlled variable
-- you have to try several other kinds of disturbance to make sure that you
have characterized the controlled variable accurately. For example, you might
find that the subject does respond to your "spelling out loud" disturbances
by maintaining S/N level. But to make sure S/N level is controlled you might
try this again by turing on a fan. Now suppose you find that S/N level is NOT
defended; S/N level goes down when the fan goes on. This suggests that your
guess about the controlled variable was wrong; the subject is controlling
something other than S/N level; now you have to guess what the person might
be controlling that would lead him to compensate for "spelling out loud" but
not for the continuous drone of a fan. Perhaps you would guess that the
controlled variable is maintaining the S/N level within a spoken phrase. And
the Test goes on.

(This is just a made up example but, of course, but I hope you see the point).