Optimistic Bias
William M. P. Klein

Suggested Scales to Measure Optimistic Biases

Optimistic Bias at the Group Level
To establish an optimistic bias at the level of the group, one might use a version of the following direct comparative question and then determine whether the mean response deviates from the midpoint:

1. How do you think your chances of getting lung cancer compare with those of the average smoker of your age and sex? Your chances are:

One could also ask a standard risk perception question, yet have respondents answer the question for both themselves and others in two separate items. A difference score deviating from zero would suggest an optimistic bias. Some risk perception scales elicit perceptions of numerical risk such as the following:

2a. Suppose you had to estimate your chances of getting lung cancer on a percentage scale. What would your estimate be? You can give any number between 0% and 100%. Please try to be as exact as possible, and use any number between 0% and 100%.

2b. Using the same scale, what would you estimate the risk of the average smoker to be?

One problem with the preceding questions is that many people might respond with "50%" because they view such a response as equivalent to "unsure." Thus, a more exact approach is to give participants a graduated nonlinear scale such as the following:

.1 .2 .3 .4 .5 .6 .7 .8 .9   1 2 3 4 5 6 7 8 9 10   15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

People have a great deal of difficulty working with numerical information, so it is important not to make too much of their response to any one item. For example, most individuals greatly overestimate the risk of rare hazards. However, if we assume that people use the scale similarly when comparing ratings of their own risk with that of a comparative target, the difference score in these ratings should be a good indication of their perception of risk. One can also get around the difficulties with numerical information by having participants use scales with verbal scale points such as:

3. How likely do you think you are to get lung cancer at some point in the future?

A disadvantage with this approach is that people's use of verbal labels (and the way in which they map verbal labels onto numerical risk) varies considerably. Fortunately, given that establishment of optimistic bias only requires the computation of a difference score between estimates of own risk and others' risk, this is less of a problem when measuring optimistic bias at the group level than when measuring it at the individual level. Scales such as these vary a great deal in the number of scale points and use of scale labels, so the above items are mere examples.

In lieu of having participants answer these items for both own risk and others' risk, one can also have separate samples answer the two items to ensure that responses to one item do not influence responses to the other item (although evidence of such a carryover effect is mixed, with only a handful of studies reporting such effects, e.g., Hoorens & Buunk, 1993 xClose
Hoorens, V., & Buunk, B. P. (1993). Social comparison of health risks: Locus of control, person-positivity bias, and unrealistic optimism. Journal of Applied Social Psychology, 23, 291-302. ).

For all items, identification of the comparative target is important, given that bias occurs for some targets (e.g., the average peer) and not others (e.g., a close friend).

Optimistic Bias at the Individual Level
In order to establish optimistic bias at the individual level, one must use a scale for which there is a credible criterion for accuracy. The numerical items above (#2a and #2b) are problematic because of people's misuse and misunderstanding of numerical information and particularly small probabilities; most people will appear to be unrealistically pessimistic because they overestimate small risks. If numerical information is to be used, it is helpful to "anchor" participants by telling them the numerical risks of other, similar hazards, and by using the graduated scale above rather than an open-ended "0-100%" scale.

Several investigators have used the comparative item (#1) instead, because there are several risk engines that can compute a person's comparative risk of having a particular problem. For example, one can ask participants to estimate their comparative risk of heart disease and then use the risk engine at www.yourdiseaserisk.harvard.edu to compute the person's actual comparative risk. Usually people are categorized as believing their risk is below average, average, or above average (irrespective of the number of scale points) and are similarly categorized with respect to actual risk. Of course, this approach makes it impossible to identify some groups of individuals, such as those who believe their risk is only slightly above average when in fact it is well above average. To deal with this problem, one can ask participants to estimate the numerical magnitude of their comparative risk (e.g., 50% lower than average, 10% higher than average), and check the accuracy of this estimate using a risk engine that provides numerical comparative risk (as in many Health Risk Appraisals). In this case, it is conventional to allow some margin of error; for example, Kreuter and Strecher (1995) xClose
Kreuter, M. W., & Strecher, V. J. (1995). Changing inaccurate perceptions of health risk: Results from a randomized trial. Health Psychology, 14, 56-63. allowed a 10% margin of error when categorizing participants as optimistically biased or not. Of course, a weakness of this approach is people's difficulty with the use of percentages and other numerical information.

When actual event data are available, it is easiest to simply ask participants whether an event will occur (or whether they are more or less likely than others to experience the event) and then follow up to determine whether the event does or does not occur.