Accident risk

[From Bill Powers (990328.1115 MST)]

Here is the abstract of that paper on accident risk (too lazy to store this
and go see who it was) that Jeff Vancouver sent us.

···

==========================================================================
Previous experimental tests of risk homeostasis theory (RHT) have failed to
manipulate both motivational and nonmotivational variables in an
ecologically valid within-subject design. In this study, 24 Ss operated an
interactive driving simulator under varying levels of a within-subject
motivational factor (monetary accident cost), a within-subject
nonmotivational factor (speed limit), and a between-subjects
nonmotivational factor (speeding fine). Consistent with RHT, increased
speed limit and reduced speeding fine significantly increased driving speed
but had no effect on accident frequency. Moreover, increased accident cost
caused large and significant reductions in accident frequency but no change
in speed choice. Results suggest that in contrast to motivationally based
accident countermeasures, regulation of specific risky behaviors such as
speed choice may have little influence on accident rates.

It seems to me that the author's description of speeding as a "risky
behavior" is contradicted by the findings. Notice edspecially:

"..increased speed limit and reduced speeding fine significantly increased
driving speed but had no effect on accident frequency."

This says that an increase in driving speed (no matter how it is created)
has no effect on accident frequency within a given subject. Therefore
driving faster is not risky. I suppose bad drivers have most of the
accidents no matter how fast, or slowly, they drive.

I think the author was thinking that increased driving speed must _a
priori_ be risky, and failed to see what his own data told him.

Best,

Bill P.

The authors do think driving fast is risky, but so are other behaviors.
Bill, I am not sure why you are not seeing this. You seem to assume that
other variables like driving close to another car, looking at the floor,
etc., are not available behaviors to compensate for the slower driving
speed. Now granted, it would be nice if these behaviors could have been
identified and measured, but it would have been nice if the test were used.
Nonetheless, they are talking about a hypothesized CV!

Sincerely,

Jeff

···

At 11:23 AM 3/28/1999 -0700, you wrote:

[From Bill Powers (990328.1115 MST)]

Here is the abstract of that paper on accident risk (too lazy to store this
and go see who it was) that Jeff Vancouver sent us.

Previous experimental tests of risk homeostasis theory (RHT) have failed to
manipulate both motivational and nonmotivational variables in an
ecologically valid within-subject design. In this study, 24 Ss operated an
interactive driving simulator under varying levels of a within-subject
motivational factor (monetary accident cost), a within-subject
nonmotivational factor (speed limit), and a between-subjects
nonmotivational factor (speeding fine). Consistent with RHT, increased
speed limit and reduced speeding fine significantly increased driving speed
but had no effect on accident frequency. Moreover, increased accident cost
caused large and significant reductions in accident frequency but no change
in speed choice. Results suggest that in contrast to motivationally based
accident countermeasures, regulation of specific risky behaviors such as
speed choice may have little influence on accident rates.

It seems to me that the author's description of speeding as a "risky
behavior" is contradicted by the findings. Notice edspecially:

"..increased speed limit and reduced speeding fine significantly increased
driving speed but had no effect on accident frequency."

This says that an increase in driving speed (no matter how it is created)
has no effect on accident frequency within a given subject. Therefore
driving faster is not risky. I suppose bad drivers have most of the
accidents no matter how fast, or slowly, they drive.

I think the author was thinking that increased driving speed must _a
priori_ be risky, and failed to see what his own data told him.

[From Rick Marken (990329.1530)]

Jeffrey B. Vancouver (29 Mar 1999 17:25:59) --

Nonetheless, they are talking about a hypothesized CV!

A CV is something that can be perceived by both the behaving
system and the observer of the behavior system. In order for an
observer to determine whether or not "accident risk" is a CV
s/he has to observe "accident risk" while it is being disturbed.
The authors of the study you discussed never did this. They
simply manipulated all kinds of variables (stimuli) and looked
for reactions (R) to these variables. This has nothing to do
with testing hypotheses about CVs. The (largely statistical)
relationships between S and R that are found in studies like
this are consistent with an infinite number of possible CVs.
Unless you monitor the behavior of a hypothetical CV while it
is being disturbed you cannot possibly determine whether a
particular variable is or is not a CV.

The "accident risk" study you posted is no more a study of the
CVs controlled by people than is Bruce Abbott's "signal shock avoidance"
study a study of the CVs controlled by rats. It's
just S-R psychology using cybernetic terms (like "homeostasis").

Best

Rick the Cynic

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

[From Bruce Gregory (990330.0642 EST)]

Rick Marken (990329.1530)

A CV is something that can be perceived by both the behaving
system and the observer of the behavior system.

It might be slightly clearer to say, "In order to test for a CV, it must
something that can be observed by both the behaving system and the observer
of the behaving system." There can be CV's that the observer is not in a
position to perceive.

Bruce Gregory

[From Bill Powers (990330.07556 MST)]

Jeff Vancouver (990329) --

I said

[The data say] that an increase in driving speed (no matter how it is
created)has no effect on accident frequency within a given subject.

Therefore

driving faster is not risky. I suppose bad drivers have most of the
accidents no matter how fast, or slowly, they drive.

And you said

The authors do think driving fast is risky, but so are other behaviors.
Bill, I am not sure why you are not seeing this. You seem to assume that
other variables like driving close to another car, looking at the floor,
etc., are not available behaviors to compensate for the slower driving
speed. Now granted, it would be nice if these behaviors could have been
identified and measured ...

I am going strictly by the data that were reported, not the data that were
imagined. If you're allowed to imagine data, you can come up with any
conclusion you please. This is not how to do science -- that is, it's not
how to find out about natural phenomena in a way that you can rely on.
Whether you realize it or not, your paragraph above condemns the paper
we're talking about to the realms of pseudo-science.

Best,

Bill Powers

, but it would have been nice if the test were used.

···

Nonetheless, they are talking about a hypothesized CV!

Sincerely,

Jeff

[from Jeff Vancouver 990330.1020 EST]

[From Rick Marken (990329.1530)]

Jeffrey B. Vancouver (29 Mar 1999 17:25:59) --

Nonetheless, they are talking about a hypothesized CV!

A CV is something that can be perceived by both the behaving
system and the observer of the behavior system. In order for an
observer to determine whether or not "accident risk" is a CV
s/he has to observe "accident risk" while it is being disturbed.
The authors of the study you discussed never did this. They
simply manipulated all kinds of variables (stimuli) and looked
for reactions (R) to these variables. This has nothing to do
with testing hypotheses about CVs. The (largely statistical)
relationships between S and R that are found in studies like
this are consistent with an infinite number of possible CVs.
Unless you monitor the behavior of a hypothetical CV while it
is being disturbed you cannot possibly determine whether a
particular variable is or is not a CV.

The "accident risk" study you posted is no more a study of the
CVs controlled by people than is Bruce Abbott's "signal shock avoidance"
study a study of the CVs controlled by rats. It's
just S-R psychology using cybernetic terms (like "homeostasis").

I said they were talking about a CV (but it is probably more reasonable to
say they are talking about controlling a perception), not that they were
doing the TEST. One of my bugaboo's with some on this list is that
discrete labels are put on continuous phenomenon. The discrete labels are:
is or is not a study/theory of control. I am trying to educate the
psychological community in PCT (among other goals). When I see someone
close, I think like a teacher instructing a pupil "That was very good, you
are almost there, now if you just ..., you will have accurately applied the
concept of control." This is what happened to me (although I did not have
the encouragement).

It is true that the study has vestiges of S-R psychology. But it is also
true that it has vestiges of cybernetics (not just the words, but the
concepts). I think these people should be taken under our wing, not
denigrated. It appears Wilde has come to his theory independent of Bill P.
He might be happy to see there are some well defined protocols for
studying control that he was unaware of. But never mind, I just thought it
was interesting given the conversation around speed limits.

Later

Jeff

A great many people think they are thinking when they are merely
rearranging their prejudices.
                -- William James

Data is not information any more than 50 tons of cement is a skyscraper.
                  -- Clifford Stoll

[From Rick Marken (990330.0730)]

Me:

A CV is something that can be perceived by both the behaving
system and the observer of the behavior system.

Bruce Gregory (990330.0642 EST)--

It might be slightly clearer to say, "In order to test for a
CV, it must something that can be observed by both the behaving
system and the observer of the behaving system." There can be
CV's that the observer is not in a position to perceive.

That would put CVs into the same class as god, the tooth fairy
and Bruce Abbott's commitment to PCT. If there are CVs that an
observer is not in a position to perceive then they exist only
in the "realms of pseudo-science" (Powers, 990330.07556 MST).

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

[from Jeff Vancouver 990330.1040 EST]

[From Bill Powers (990330.07556 MST)]

I am going strictly by the data that were reported, not the data that were
imagined. If you're allowed to imagine data, you can come up with any
conclusion you please. This is not how to do science -- that is, it's not
how to find out about natural phenomena in a way that you can rely on.
Whether you realize it or not, your paragraph above condemns the paper
we're talking about to the realms of pseudo-science.

Do you re-establish, with data, the laws of heat transfer when you report
about temperature control systems. Did you establish, with data, the
effect of the joy stick on the cursor in your '78 data or did you just
report the nature of the relationship as a given (the answer is the later).
You are confusing the findings of one study with the findings of a program
of research. No one study establishes much but the most trivial. It is
with the combination of studies that we begin to understand a phenomenon.

Now if you argue that the combination of studies that are usually presented
as demonstrating some phenomenon are lacking in substantial ways, I will
not disagree. What I was hoping for (way to optimistically), is that you
would recognize that with some help, these people may be able to develop
the constillation of studies that would truly support their theory (or deny
it as the case may be). Clearly, there is no interest there.

BTW, I do believe that all-else-being-equal, speed is related to accidents
because of its effect on environmental lags (it shortens them). But all
else is not equal. That fact creates a potentially interesting phenomenon.
But never mind.

Later,

Sincerely,

Jeff

[From Bruce Gregory (990330.1050 EST)]

Rick Marken (990330.0730)

That would put CVs into the same class as god, the tooth fairy
and Bruce Abbott's commitment to PCT. If there are CVs that an
observer is not in a position to perceive then they exist only
in the "realms of pseudo-science" (Powers, 990330.07556 MST).

Balderdash! There are no doubt countless CVs that we are unaware of. How
do we regulate the levels of hormones in our blood streams? Because you
don't know are you prepared to label our conviction that there is a
mechanism "pseudo-science"? Considering the facility with which you
imagine CV's I nominate you chief charlatan.

Bruce Gregory

[From Rick Marken (990330.0800)]

Me:

That would put CVs into the same class as god, the tooth fairy
and Bruce Abbott's commitment to PCT. If there are CVs that an
observer is not in a position to perceive then they exist only
in the "realms of pseudo-science" (Powers, 990330.07556 MST).

Bruce Gregory (990330.1050 EST)--

Balderdash! There are no doubt countless CVs that we are
unaware of.

Oh, calm down, Bruce. "Unaware" is not the same as
"unobservable". We are unaware of countless CVs. That's
why we have The Test. In order to do The Test you have to
be able to observe the variable(s) that you believe to be
under control. If you suspect that the level of a particular
blood chemical (hormone) is a controlled variable, for example,
then you have to Test to determine whether the _observed_ level
of the hormone is protected from disturbance.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

[From Bruce Gregory (990330.1230 EST)]

Rick Marken (990330.0800)

Oh, calm down, Bruce. "Unaware" is not the same as
"unobservable". We are unaware of countless CVs. That's
why we have The Test. In order to do The Test you have to
be able to observe the variable(s) that you believe to be
under control. If you suspect that the level of a particular
blood chemical (hormone) is a controlled variable, for example,
then you have to Test to determine whether the _observed_ level
of the hormone is protected from disturbance.

I think I've gotten control of myself again. Must have been a flashback.
It seems to me that I probably have perceptions at the upper levels of
the hierarchy that you do not share. You may conjecture what they are
like and perform The Test. But simply because The Test turns out
negative, you would not assume that I am not controlling any perception.
You'd simply say that you hadn't figured out what it was. I definitely
have a perception something like "perceived risk". For example, most of
the time I am on the highway, I let the cruise control maintain my
speed. However, where cars are entering the highway and lane-changing
occurring (in particular where the MassPike and I 495 meet), I often
disengage the cruise control in order to exercise finer control of the
car's speed. When the "perceived risk" is again lower, I re-engage the
cruise control. I conjecture that my "perceived risk" is a CV composed
of more than one lower level perceptions (condition of the road, density
of traffic, speed of merging...). So even though I do not know exactly
what the perception is, I'm sure it exists and it is under my control.
You are free to call this is unscientific if you like.

Bruce Gregory

[From Rick Marken (990330.1400)]

Bruce Gregory (990330.1230 EST)--

But simply because The Test turns out negative, you would
not assume that I am not controlling any perception.

Of course not. I would assume that I just haven't yet figured
out what perception(s) you are controlling.

The Test is an iterative process, not a one shot affair (like
most conventional psychology experiments). The Test is used to
1) home in on what the CV actually is and 2) prove that a
particular definition of the CV is actually the CV. Both of
these processes involve more than just applying a disturbance
and seeing if anything happens to the hypothetical CV. You can
quickly _reject_ a variable as a CV if a continuous disturbance
has the expected continuous effect on the variable. But the Test
still isn't over even when you conclude that variable X is
not a CV. The next step is to _keep looking_ by trying other
variables to see if they seem to be controlled. Once you find
what seems to be a controlled variable (such as optical velocity
in fly ball catching behavior) then you move into phase 2) where
you try to prove to yourself that this variable really is under
control. This means subjecting the variable to all kinds of
different disturbances and seeing if it is protected from them.

I definitely have a perception something like "perceived risk"...
I conjecture that my "perceived risk" is a CV composed
of more than one lower level perceptions (condition of the road,
density of traffic, speed of merging...). So even though I do
not know exactly what the perception is, I'm sure it exists
and it is under my control.

The job of the scientist would be to try to determine what
variable you are controlling that you call "perceived risk". Maybe
it will turn out that you are controlling some calculable function
of the variables you describe (condition of the road, density of
traffic, speed of merging...). Or is may be that the variable
you are controlling is not easily measured or quantified; the only
way to measure the variable may be by using the perceptual
capabilities of another human (the observer). The Tester would
then look to see if there were any aspect of his own perceptual
experience that seems to be protected from disturbance by the
actions of the person under study.

Anyway, my point is that you can't study controlled variables
that you cannot perceive in _some_ way. Indeed, I don't see
how you can have a science of _any_ phenomenon that can't be
perceived by (or made perceptible to) at least one other person.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

[From Bruce Gregory (990330.1725 EST)]

Rick Marken (990330.1400)

Anyway, my point is that you can't study controlled variables
that you cannot perceive in _some_ way. Indeed, I don't see
how you can have a science of _any_ phenomenon that can't be
perceived by (or made perceptible to) at least one other person.

Indeed.

Bruce Gregory

from [ Marc Abrams (990330.1702) ]

Hi Jeff, Glad to see your still around. Hope all is well.

[from Jeff Vancouver 990330.1040 EST]

Now if you argue that the combination of studies that are usually presented
as demonstrating some phenomenon are lacking in substantial ways, I will
not disagree. What I was hoping for (way to optimistically), is that you
would recognize that with some help, these people may be able to develop
the constillation of studies that would truly support their theory (or deny
it as the case may be). Clearly, there is no interest there.

Jeff, if this were 2 months ago I would have largely agreed with you. But
it's not and I don't. One of the problems faced by PCT is that PCT "looks",
"feels", "smells". and "tastes", _like_ alot of things to alot of people.
Depending upon where you are coming from, It all _seems_ so damn close. But
it's not.

An example: if I told you that every time you took a step you were utilizing
the laws of gravity. You would say "Sure, I know that". If I asked you to
explain _how_ gravity works, could you do that? What _seems_ to be obvious
is not. People who understand something ( in this case Psych ) one way, have
to give up those ideas, because things just don't "fit" the same way. No
matter how hard you try. In trying to explain PCT to others, if the meaning
of "perception" is not misunderstood then the meaning of "conflict" will be,
if not that, then "reference level", on and on it goes ad nauseum.

Jeff if your truly interested in "spreading PCT" then I think you should
fully understand it yourself first. Sorry in advance if this comes off as
slamming your efforts. I know how enthusiastic you are about PCT I just
could not think of another way of saying it.

Marc

[From Bill Powers (990331.1218n MST)]

Bruce Gregory (990330.1230 EST)--

Rick Marken (990330.0800)
For example, most of
the time I am on the highway, I let the cruise control maintain my
speed. However, where cars are entering the highway and lane-changing
occurring (in particular where the MassPike and I 495 meet), I often
disengage the cruise control in order to exercise finer control of the
car's speed. When the "perceived risk" is again lower, I re-engage the
cruise control. I conjecture that my "perceived risk" is a CV composed
of more than one lower level perceptions (condition of the road, density
of traffic, speed of merging...). So even though I do not know exactly
what the perception is, I'm sure it exists and it is under my control.
You are free to call this is unscientific if you like.

So? Lots of what we think and do is unscientific. Fortunately, most of the
time it doesn't matter.

Your experience goes against the Wilde data, which appears to say that the
subjects in the experiment do not try to _reduce_ perceived risk; they
maintain it at a positive value. Of course there's a big assumption
involved, which is that subject using a driving simulator experience risk
in the same way drivers of real cars do. I'm pretty sure they know they
won't get killed in the simulator, so that assumption is pretty dubious.

In answer to Jeff Vancouver's comments, yes, in the tracking experiments we
ALWAYS keep track of ALL the variables we use in the simulation --
specifically, the effect of the handle on the cursor is always part of the
data, as is the disturbance pattern and the handle and cursor positions.
The data wouldn't mean anything if this weren't true.

Best,

Bill P.

···

Bruce Gregory

[From Bruce Gregory (990332.0525 EST)]

Bill Powers (990331.1218n MST)

Your experience goes against the Wilde data, which appears to say that the
subjects in the experiment do not try to _reduce_ perceived risk; they
maintain it at a positive value.

If I wanted to minimize perceived risk, I'd probably stay at home! Clearly
there is some degree of risk that I am willing to tolerate, I can believe
that I drive in a way that keeps this perception above zero and below some
threshold. I'll bet this is true of skiers too.

Of course there's a big assumption
involved, which is that subject using a driving simulator experience risk
in the same way drivers of real cars do. I'm pretty sure they know they
won't get killed in the simulator, so that assumption is pretty dubious.

Yes, I think that's a very weak point. It would be interesting to know if
subjects in a simulator have heart-rates, for example, that correlate with
those on the road in similar circumstances.

Bruce Gregory

[from Jeff Vancouver

From [ Marc Abrams (990330.1702) ]

Jeff, if this were 2 months ago I would have largely agreed with you. But
it's not and I don't. One of the problems faced by PCT is that PCT "looks",
"feels", "smells". and "tastes", _like_ alot of things to alot of people.
Depending upon where you are coming from, It all _seems_ so damn close. But
it's not.

I will admit that I have moved further in that direction over the years,
but I still think in continuous terms, not discrete.

Jeff if your truly interested in "spreading PCT" then I think you should
fully understand it yourself first.

I am sorry, I do not think Bill P. fully understands it himself. This is
not meant as any disrespect of Bill, but instead a healthy respect for the
complexity of the role of control systems in humans. Just to be clear, the
"it" I refer to is "the role of control systems in humans." PCT is an
imperfect representation of that. I don't think Bill would disagree.
Meanwhile, my understand of PCT is an imperfect representation of PCT. A
subset within a subset. I suspect that just as PCT will only approach
reality, I (or anyone) will only approach PCT. I have learned to live with
that. No offense taken.

I think my primary problem is that I am interested in problematic control
systems. The CV that creates perceptions of risk (which I am really not
that interested in) would be a difficult CV to test. As pointed out, the
simulation is not reality - you will not be killed. Meanwhile, the lack of
control in real settings (not to mentioned the ethical issues - imagine
blocking the CV that determines perceived risk) makes anything approaching
the rigor of the Test very problematic. Hence, I cut a little slack.

Perhaps Bill is correct. All the contingencies in all his experiments are
completely verified. I certainly believe that more of the contingencies in
the accident risk study could have been verified. But I do not think that
if one wanted to test this risk homeostasis theory, one would do it with
all the controls of a tracking task. The question is, does PCT want to be
a science of tracking tasks.

But, alas, I do not think that is the issue. Awhile back someone (I do not
remember who) posted the results of a study they did of some higher-level
control system. My read of it was that it lacked a huge number of controls
(i.e., there were several alternative explanations available).
Nonetheless, Bill endorsed the study. In other words, lack of experimental
control did not create the disturbance. I guess I am still trying to
figure out what the CV is (or are) that Bill is controlling for.

To i., thanks for the references. I will seek them out, but I may owe you
a beer.

To Bruce A., I do believe that one can have a commitment to a theory. It
was the commitment to a theory that led to the discovery of Pluto (rather
than tossing the theory because of observed deviations in Neptunes orbit
from it). However, it is a fine line.

To Bruce G., Rick is correct. Tolerating risk is different from seeking
risk. You tolerate risk to achieve other goals. You seek risk when you
get on a roller coaster (presumably). However, it may be that tolerating
risk is the issue. This is an interesting, different (and more
complicated) model. Alas, another alternative explanation to be tested. I
liked the heart rate idea, too. That might lend some credence to the
psychological meaningfulness of the simulation.

Well, I have to get back to work. If I get some done, I might be ready to
talk about decision making and PCT (gasp).

Sincerely,

Jeff

[From Bruce Gregory (990331.1013 EST)]

Jeff Vancouver

But, alas, I do not think that is the issue. Awhile back
someone (I do not
remember who) posted the results of a study they did of some
higher-level
control system. My read of it was that it lacked a huge
number of controls
(i.e., there were several alternative explanations available).
Nonetheless, Bill endorsed the study. In other words, lack
of experimental
control did not create the disturbance. I guess I am still trying to
figure out what the CV is (or are) that Bill is controlling for.

Explicit acknowledgement that the proper model of behavior is based on
the control of input.

Bruce Gregory

[From Bill Powers (990401.0803 MST)]

Bruce Gregory (990332.0525 EST)--

It would be interesting to know if
subjects in a simulator have heart-rates, for example, that correlate with
those on the road in similar circumstances.

Actually, it wouldn't interest me. By this time we're examining the dots in
the half-tone picture. Specific examples of behavior are of interest to me
only if they cast light on the organization we're studying in PCT, either
by bringing out something new or by raising doubts about something we've
already accepted. Just finding example after example that illustrates the
theory doesn't seem to me very useful.

Best,

Bill P.

[From Bruce Gregory (990331.1021 EST)]

Bill Powers (990401.0803 MST)

Bruce Gregory (990332.0525 EST)--

>It would be interesting to know if
>subjects in a simulator have heart-rates, for example, that
correlate with
>those on the road in similar circumstances.

Actually, it wouldn't interest me. By this time we're
examining the dots in
the half-tone picture. Specific examples of behavior are of
interest to me
only if they cast light on the organization we're studying in
PCT, either
by bringing out something new or by raising doubts about
something we've
already accepted. Just finding example after example that
illustrates the
theory doesn't seem to me very useful.

Different strokes for different folks.

Bruce Gregory