A scientific approach to what happend at LTCM [what Graham would have called lack of "margin of safety" --''''> ]:
"When Scientific Predictions Are So Good They're Bad"
By WILLIAM K. STEVENS -- 9/29/98
Noah had it easy. He got his prediction straight
from the horse's mouth and was left in no doubt
about what to do.
But when the Red River of
the North was rising to record
levels in the spring of 1997,
the citizens and officials of
Grand Forks, N.D., were not
so privileged. They had to rely
on scientists' predictions about
how high the water would
rise. And in this case, Federal
experts say, the flood forecast
may have been issued and
used in a way that made
things worse.
The problem, the experts said,
was that more precision was
assigned to the forecast than was warranted. Officials
and citizens tended to take as gospel an oft-repeated
National Weather Service prediction that the river
would crest at a record 49 feet. Actually, there was a
wider range of probabilities; the river ultimately
crested at 54 feet, forcing 50,000 people to abandon
their homes fast.
The 49-foot forecast had lulled the town into a false
sense of security, said Dr. Roger A. Pielke Jr. of the
National Center for Atmospheric Research in Boulder,
Colo., a consultant on a subsequent inquiry by the
weather service.
In fixating on the single number of 49 feet, the people
involved in the Grand Forks disaster made a common
error in the use of predictions and forecasts, experts
who have studied the case say. It was, they say, a case
of what Alfred North Whitehead, the mathematician and
philosopher, once termed "misplaced concreteness." And
whether the problem is climate change, earthquakes,
droughts or floods, they say the tendency to overlook
uncertainties, margins of error and ranges of
probability can lead to damaging misjudgments.
The problem was the topic of a workshop this month at
Estes Park, Colo. In part, participants said, the
problem arises because decision makers sometimes want
to avoid making hard choices in uncertain situations.
They would rather place responsibility on the
predictors.
Scientifically based predictions, typically using
computerized mathematical models, have become pervasive
in modern society. But only recently has much attention
been paid to the proper use -- and misuse -- of
predictions. The Estes Park workshop, of which Pielke
was an organizer, was an attempt to come to grips with
the question. The workshop was sponsored by the
Geological Society of America and the National Center
for Atmospheric Research.
People have predicted and prophesied for millenniums,
of course, through means ranging from the visions of
shamans and the warnings of biblical prophets to the
examination of animal entrails. With the arrival of
modern science, people teased out fundamental laws of
physical and chemical behavior and used them to make
better and better predictions.
But once science moves beyond the relatively
deterministic processes of physics and chemistry,
prediction gets more complicated and chancier. The
earth's atmosphere, for instance, often frustrates
efforts to predict the weather and long-term climatic
changes because scientists have not nailed down all of
its physical workings and because a substantial measure
of chaotic unpredictability is inherent in the climate
system. The result is a considerable range of
uncertainty, much more so than is popularly associated
with science. So while computer modeling has often made
reasonable predictions possible, they are always
uncertain; results are by definition a model of
reality, not reality itself.
The accuracy of predictions varies widely. Some, like
earthquake forecasts, have proved so disappointing that
experts have turned instead to forecasting longer-term
earthquake potential in a general sense and issuing
last-second warnings to distant communities once a
quake has begun.
In some cases, the success of a prediction is near
impossible to judge. For instance, it will take
thousands of years to know whether the environmental
effects of buried radioactive waste will be as
predicted.
On the other hand, daily weather forecasts are checked
almost instantly and are used to improve the next day's
forecast. But weather forecasting is also a success,
the assembled experts agreed, because people know its
shortcomings and take them into consideration. Weather
forecasts "are wrong a lot of the time, but people
expect that and they use them accordingly," said Robert
Ravenscroft, a Nebraska rancher who attended the
workshop as a "user" of predictions.
A prediction is to be distrusted, workshop participants
said, when it is made by the group that will use it as
a basis for policy making -- especially when the
prediction is made after the policy decision has been
taken. In one example offered at the workshop, modeling
studies purported to show no harmful environmental
effects from a gold mine that a company had decided to
dig.
Another type of prediction miscue emerged last March in
connection with asteroids, the workshop participants
were told by Dr. Clark R. Chapman, a planetary
scientist at the Southwest Research Institute in
Boulder. An astronomer erroneously calculated that
there was a chance of one-tenth of 1 percent that a
mile-wide asteroid would strike Earth in 30 years. The
prediction created an international stir but was
withdrawn a day later after further evidence turned up.
This "uncharacteristically bad" prediction, said Dr.
Chapman, would not have been issued had it been
subjected to normal review by the forecaster's
scientific peers. But, he said, there was no
peer-review apparatus set up to make sure that
"off-the-wall predictions don't get out." (Such a
committee has since been established by NASA.)
Most sins committed in the name of prediction, however,
appear to stem from the uncertainty inherent in almost
all forecasts. "People don't understand error bars,"
said one scientist, referring to margins of error.
Global climate change and the Red River flood offer two
cases in point.
Computer models of the climate system are the major
instruments used by scientists to project changes in
climate that might result from increasing atmospheric
concentrations of heat-trapping gases, like carbon
dioxide, emitted by the burning of fossil fuels.
Basing its forecast on the models, a panel of
scientists set up by the United Nations has projected
that the average surface temperature of the globe will
rise by 2 to 6 degrees Fahrenheit, with a best estimate
of 3.5 degrees, in the next century, and more after
that. This compares with a rise of 5 to 9 degrees since
the depths of the last ice age. The temperature has
increased by about 1 degree over the last century.
But the magnitude and nature of any climate changes
produced by any given amount of carbon dioxide are
uncertain. Moreover, it is unclear how much of the gas
will be emitted over the next few years, said Dr. Jerry
D. Mahlman, a workshop participant who directs the
National Oceanic and Atmospheric Administration's
Geophysical Fluid Dynamics Laboratory at Princeton,
N.J. The laboratory is one of the world's major climate
modeling centers, and the oldest.
This uncertainty opens the way for two equal and
opposite sins of misinterpretation. "The uncertainty is
used as a reason for governments not to act," in the
words of Dr. Ronald D. Brunner, a political scientist
at the University of Colorado at Boulder. On the other
hand, people often put too much reliance on the precise
numbers.
In the debate over climate change, the tendency is to
state all the uncertainties and caveats associated with
the climate model projections -- and then forget about
them, said Dr. Steve Rayner, a specialist in global
climate change in the District of Columbia office of
the Pacific Northwest National Laboratory. This creates
a "fallacy of misplaced confidence," he said,
explaining that the specific numbers in the model
forecasts "take on a validity not allowed by the
caveats." This tendency to focus unwisely on specific
numbers was termed "fallacious quantification" by Dr.
Naomi Oreskes, a historian at the University of
California at San Diego.
Where uncertainty rules, many at the workshop said, it
might be better to stay away from specific numbers
altogether and issue a more generalized forecast. In
climate change, this might mean using the models as a
general indication of the direction in which the
climate is going (whether it is warming, for instance)
and of the approximate magnitude of the change, while
taking the numbers with a grain of salt.
None of which means that the models are not a helpful
guide to public policy, said Mahlman and other experts.
For example, the models say that a warming atmosphere,
like today's, will produce heavier rains and snows, and
some evidence suggests that this is already happening
in the United States, possibly contributing to damaging
floods. Local planners might be well advised to
consider this, Mahlman said.
One problem in Grand Forks was that lack of experience
with such a damaging flood aggravated the uncertainty
of the flood forecast. Because the river had never
before been observed at the 54-foot level, the models
on which the prediction was based were "flying blind,"
said Pielke; there was no historical basis on which to
produce a reliable forecast.
But this was apparently lost on local officials and the
public, who focused on the specific forecast of a
49-foot crest. This number was repeated so often,
according to the report of an inquiry by the National
Weather Service, that it "contributed to an impression
of certainty." Actually, the report said, the 49-foot
figure "created a sense of complacency," because it was
only a fraction of a foot higher than the record flood
of 1979, which the city had survived.
"They came down with this number and people fixated on
it," Tom Mulhern, the Grand Forks communications
officer, said in an interview. The dikes protecting the
city had been built up with sandbags to contain a
52-foot crest, and everyone figured the town was safe,
he said.
It is difficult to know what might have happened had
the uncertainty of the forecast been better
communicated. But it is possible, said Mr. Mulhern,
that the dikes might have been sufficiently enlarged
and people might have taken more steps to preserve
their possessions. As it was, he said, "some people
didn't leave till the water was coming down the
street."
Copyright 1998 The New York Times Company |
