Perceived uncertainty and the response to global warming

Jonathan Baron, Jay Schulkin, Howard Kunreuther1

1990 (unpublished)

Abstract

One group of college students read a description of the greenhouse effect that emphasized scientific uncertainties. A control group read a description containing only best estimates, without any mention of uncertainty. In the control group, those who originally did not strongly favor immediate action to combat global warming became more willing to act after reading the description, and those who strongly favored action became less willing. In the uncertainty group, subjects maintained their original willingness or unwillingness to act. When expressing uncertainty, scientists might have to warn their listeners against using it only to bolster their prior opinion.
The effects of the accumulation of greenhouse gasses in the atmosphere are highly uncertain. Some observers argue that this uncertainty must be resolved before action is taken. Others argue, on the basis of worst-case scenarios, that the very uncertainty increases the need for action.2
Both points of view are, in principle, justifiable. The argument for inaction is that the expected cost of waiting for new information is less than the expected cost of error from precipitous action. The argument for action is that delay increases expected costs.
It is possible that those who hold these contrasting positions arrive at them not through dispassionate analysis of expectations but, rather, on the basis of prior opinion. People may tend to seize on the uncertainty itself as a justification for whatever course they favored initially. When information is presented as uncertain, people may attend mainly to the possible outcomes that they most fear: being caught unprepared for serious warming, or wasting money on unnecessary preparation. Although we know of no psychological evidence for exactly this effect, a number of findings are similar. For example, conflicting evidence is interpreted so as to strengthen an initial view3, and ambiguous probability information is subjectively distorted when used in decision making 4.
To examine the effect of uncertainty, we gave each of 72 college students a description of the basic facts about global warming as reported in the press. A control group of 36 subjects read a description in which the facts were presented as best guesses, without any mention of uncertainty. An uncertainty group read an identical description in which the facts were presented with ranges so as to emphasize the uncertainty. Before and after reading the description, the subjects filled out a five-question attitude survey concerning actions about global warming.
Examples of the differences in the two forms of the description are as follows, with material in brackets added to the uncertainty version: `This increased carbon dioxide has already begun to cause a warming of the average temperature of the atmosphere of about 1 degree Celsius over the last 50 years. [(Scientists disagree on the amount of warming that has already occurred: estimates range from 0 to 2 degrees.)]' `The average estimate is that the warming over the next 60 years will be about 5.5 degrees Fahrenheit, if present trends in fuel use continue, [but the estimates range from about 2.7 to 8.1 degrees].' `The total cost of adapting to these changes is estimated at $250 billion per year for the United States alone (about half of the current military budget), [although estimates range from $50 billion to $500 billion].' Both versions ended with a statement that `changes are easier to implement when more time is available,' but the ending of the uncertainty version began with, `The uncertainties will not be resolved in the next few years.'
The attitude survey consisted of the following questions, each answered on a scale from +3 (strongly support) to -3 (strongly oppose): `The U.S. should undertake a massive program of research on new energy sources that do not yield carbon dioxide.' `The countries of the world should agree on targets for reducing the production of carbon dioxide.' `The U.S. should take unilateral steps to reduce carbon dioxide through incentives for alternative fuels' `The U.S. should take unilateral steps to reduce carbon dioxide through taxes on fuels that produce carbon dioxide, such as oil, gasoline, and coal.' `The U.S. should take unilateral steps to reduce carbon dioxide through incentives for conservation.' The total score, which could range from -15 to 15, was used in analysis.
The results are shown in Table 1, broken down according to the initial scores on the attitude survey. In the control condition, those with strong initial opinions tended to moderate them. In the experimental condition, this moderation did not occur. 
Table 1
Mean scores before and after, by condition, for subjects grouped
by initial score.

             Control group                 Uncertainty group
Initial      Mean     Mean    Mean         Mean     Mean    Mean
score     N  initial  final  change     N  initial  final  change

0         3  -2.7     3.7     6.3       3  -5.0    -2.0     3.0
1-5       4   3.5     4.3     0.8       4   3.8     4.3     0.5
6-10      9   8.0     9.3     1.3      17   8.1    10.5     2.4
11-13    13  12.4    12.1    -0.3       8  12.1    13.9     1.8
14-15     7  14.1    13.7    -0.4       4  14.3    14.3     0.0
Although these differences between conditions were small, their statistical significance was clear. The final score was regressed on the initial score, condition (control vs. uncertainty), and the interaction of control and condition.5 As hypothesized, the interaction term was significant (t=2.74, p=0.008). The effect of initial score was also significant (t=2.37, p=0.02), as was the overall regression (R2=0.79, p<.001). Condition itself had no significant effect.6
Another way to describe the results is to regress the change score (final minus initial) against the initial score in each condition. These regressions yielded the following: 
 (control)    change = 4.0 - 0.34(initial score);
 (uncertain)  change = 2.4 - 0.06(initial score).
The effect of initial score was significant in the control condition (t=4.97) but not in the uncertain condition. In sum, facts lead to moderation of opinions when the facts are presented as best estimates, but when they are presented as uncertain, people seem to use the facts as support for their initial views, whatever these views might be.
Three accounts of these results are possible. First, subjects who favor action could be justifiably (or unjustifiably) afraid of failing to prepare for the worst case, the upper end of the range of effects, and subjects who oppose action could likewise fear overreaction to the best case, the lower end.7 In the uncertain condition, these extremes remain possible. Second, subjects in the uncertain condition have less reason to question their prior views (which might, or might not, be based on reliable information). Third, subjects in the uncertain condition could attend most strongly to the end of the range that justified their earlier views, because they are reluctant to consider opposing evidence.8
Expression of uncertainty affects the interpretation of scientific evidence as a basis for action. Scientists should express uncertainty, but they might also do well to warn their audience against the possibility of biased interpretation.

Footnotes:

1Department of Psychology, Department of Anatomy, Department of Decision Sciences, respectively, University of Pennsylvania, Philadelphia, PA 19104, U.S.A.
2E. W. Coldglazier, Scientific uncertainties, public policy, and global warming: How sure is sure enough? University of Tennessee, Knoxville, 1990; E. B. Skolnikoff, Foreign Policy, 1990, 79, 77-93.
3C. G. Lord et al., J. Pers. Soc. Psychol., 37, 2098-2109, 1979.
4H. J. Einhorn & R. M. Hogarth, Psychol. Rev., 92, 433-461, 1985. H. Kunreuther, Risk Analysis, 9, 319-328, 1989.
5The groups did not differ significantly in the distributions of initial scores, final scores, or mean change.
6These results were unchanged when quadratic terms were included for the initial score. The interaction was confined to the linear term, with t=3.01, and the quadratic component itself was not significant.
7T. Page, Ecology Law Quarterly, 7, 207-243, 1978.
8J. Baron, Thinking and deciding, New York: Cambridge University Press, 1988.

File translated from TEX by TTH, version 3.61.
On 12 Nov 2004, 05:25.