[from Jeff Vancouver 990329.1630 EST]
[From Bill Powers (990326.1428 MST)]
If they are controlling for a constant level of experienced accident risk,
then if something reduced the accident risk below that level they would
change their actions so as to increase the risk again. Is that what Wilde
found?
me:
This is what he seems to be trying to find (or explain). The data (what
has been found) is ambiguous. There seems to be a constant accident rate,
despite incentives to reduce specific risky behaviors.
Bill:
You don't seem to understand what I said. ...
But controlling for a certain level of perceived accident _risk_ does not
mean controlling for a constant accident _rate_.
I think I understand perfectly. That is why I (and Wilde) say the rate
data is ambiguous. The relationship between rate and risk is ambiguous.
It is possible that the
so-called measure of accident risk has nothing to do with accident rate,
which is why changing the risk has no effect on the rate. I would choose
this guess as the one most worth testing first. The experienced risk
depends on the person's way of evaluating situations in terms of risk,
which could easily be radically incorrect.
They are not attempting to change the risk, but merely one behavior which
is assumed to be related to risk (speeding). Indeed, they can influence
the behavior, but not the rate of accidents. This is _presumably_ because
the individuals do other things that counteract the changes in speed such
that the end result is the same level of experienced risk. These other
things _are_ risky, such that the accident rate does not change either.
Hence, a great intervention would be to encourage a behavior that increases
perceived risk (I use perceived and experienced interchangeably), but is
not actually risky. This would be an attempt to change overall risky
behavior.
He is arguing that
is because people are controlling for accident risk.Are you just saying that they are controlling for accident risk in general,
without specifying how much accident risk (it could be zero), or are you
saying they are trying to maintain some positive amount of accident risk,
rather than a zero amount?
I think it would have to be a positive amount. I think they assume (or
have evidence that) speeding is actually related to accidents, yet
interventions which do in fact reduce speeding do not reduce accidents. If
individuals can compensate for how the lower speed reduces the perception
of risk by engaging in other dangerous acts, there would be no change in
accidents. But I am not sure what negative risk would mean, so I assume
that they assume individuals (at least some) are controlling for a positive
value.
The study (cited above) looked at the lack of
affect of speed limits on accidents in a within-person simulation study.I'm not interested in this kind of fact. You can find studies to show that
lowering speed limits _does_ lower the accident rate. That is because all
such studies are statistical, and admit almost as many negative as positive
results. If you want to prove that there is a positive relationship, you
just go shopping in the literature and pick out all the studies with a
positive outcome. If you want to prove the opposite, pick the excluded set
of experiments instead. It's disgusting, but that's how behavioral science
works.
Oops, I have been doing it wrong then. I tend to look at both positive and
negative studies.
I think this idea is analogous to the idea that one can engage in several
acts to achieve a body temperature. If one act is constrained, another
will emerge to get the CV to its desired state. Mathematically, let me
revise equation 2 in the 78 paper:
qi = g1(qo1) + g2(qo2) + h(qd)
If qo1 is limited, qo2 must take up the slack.
I am certainly not arguing that they are correct. Or that they have
unambigously made their point. I will react to some of the other comments
in response to those posts.
Sincerely,
Jeff