Behavioral Research

Table of Contents
1 Definition and History
2

Measurement and Methodological Issues

3

Utility of Construct

4

Related Constructs

5

References

6

Measures Appendix

7 Published Examples

Download Full Text (PDF)

Other Constructs
 

Barriers

 

Dispositional Optimism

 

Environments

 

Illness Representations

  Implementation Intentions
  Intention, Expectation, and Willingness
  Normative Beliefs
  Optimistic Bias
  Perceived Benefits
  Perceived Control
  Perceived Severity
  Perceived Vulnerability
  Self-Efficacy
  Self-Reported Behavior
  Social Influence
  Social Support
  Stages
  Worry

Optimistic Bias
William M. P. Klein

<< Previous

2

Measurement and Methodological Issues

Bias at the group level. The simplest method of establishing an optimistic bias (see Appendix) is to ask a sample of individuals to estimate their risk relative to that of other members of the sample (or the population from which that sample is taken). This is called the "direct" method of elicitation. For example, a respondent might be asked to "compare your risk with that of the average person of your age and sex" on a scale that ranges from "below average" to "above average" with "average" as the midpoint. Investigators have generally used odd-numbered scales (e.g., 5-pt. or 7-pt. scales) to ensure that "average" is in the middle of the scale. If the mean response is higher or lower than this midpoint, one has demonstrated an optimistic bias (assuming that the sample is fully representative of the reference group, and that actual risk is not highly skewed). One might also ask respondents to compare others' risk to their own risk, which turns out to elicit less bias (Otten & van der Pligt, 1996).

Another approach is to ask participants to make two judgments - an estimate of their own risk (on a likelihood scale, for example), and an estimate of the risk of the average peer (see Appendix). These ratings can then be subtracted, and if the mean difference is not zero, a bias can be said to exist. This is called the "indirect" method of measuring optimistic bias. The attractiveness of such an approach is that it is possible to assess whether a given moderator influences estimates of personal risk or the comparative target's risk. In a review of studies using the indirect method, Helweg-Larsen & Shepperd (2001) showed that negative affect influences personal risk estimates whereas positive affect influences target risk estimates, a finding that would have been obscured had comparative risk not been assessed with separate items. Finally, separate samples can be asked to make the two judgments; for example, Weinstein, Marcus, & Moser (2005) asked separate groups of smokers to estimate their own or other smokers' risks of experiencing tobacco-related illnesses, and again observed an optimistic bias when assessing the difference in estimates between the two groups. Interestingly, some studies show that the magnitude of bias is greater when using the direct method (e.g., Goszsczyska & Roskan, 1989) yet others show the opposite pattern (e.g., Sutton, 2002).

Bias at the individual level. Although the above methods are effective when evaluating optimistic bias at the level of the group, they cannot be used to determine which members of a group are biased. A woman who believes her risk of breast cancer is below average, for example, may be quite accurate if she has no risk factors for breast cancer. In fact, this woman may be unrealistically pessimistic if her comparative risk is even more below average than she thinks it is. It is important to be able to identify which members of a sample are biased, however, in order to determine whether biases are correlated with other individual-level variables such as personality and behavior (see Appendix). Many studies attempting to link optimistic biases and related "positive illusions" with other variables such as health behavior simply define bias as a tendency to make self-serving judgments, without taking the important step of assessing the accuracy of these judgments. Consequently, although we know that optimistic beliefs are related to precautionary behaviors and ultimately to a more adaptive psychophysiological profile (e.g., Taylor, Lerner, Sherman, Sage, & McDowell, 2003), we do not have sufficient data to determine whether such beliefs are adaptive when they are illusory.

A small number of studies have attempted to use objective criteria to assess individual bias. Several of these studies use experimenter-initiated models to determine which members of the sample are at higher risk (e.g., Gerrard & Luus, 1995; Wiebe & Black, 1997). Others use "risk engines" to compute a person's risk based on epidemiological models (which are built from large epidemiological data sets such as the Framingham study) and then determine how participants' estimates compare with values computed by these risk engines (e.g., Kreuter & Strecher, 1995; Radcliffe & Klein, 2002). Very few studies measure actual outcomes to determine accuracy, and such studies are needed. In one example, college students estimated their comparative risk of having unplanned sexual intercourse in the next year, and reported six months later whether such an event had occurred (Klein, Geaghan, & MacDonald, 2005).

Absolute vs. comparative optimistic bias. There is no reason, of course, to limit optimistic biases to comparative beliefs. The use of a comparative measure was initially based on the ease of demonstrating optimistic bias at the group level (Weinstein, 1980). However, if a man predicts that he will not get prostate cancer and then he does, he would clearly be optimistically biased. Similarly, most HIV-seropositive individuals who do not believe they will succumb to AIDS are optimistically biased (Taylor, Kemeny, Aspinwall, Schneider, Rodriguez, & Herbert, 1992). Whether an investigator measures optimistic bias based on comparative or absolute measures should depend on the hypothesis being tested. For example, given findings that comparative risk perceptions are more predictive than absolute risk perceptions of colorectal cancer screening (Blalock, DeVellis, Sandler, & Afifi, 1990), research on screening behaviors may benefit from the use of comparative measures. Absolute and comparative risk perceptions are not redundant; each explains independent variance in worry, behavior, and other related constructs (Lipkus et al., 2000).

Cross-sectional and prospective designs. An important methodological issue one faces when attempting to link optimistic biases with other constructs such as risk-reducing behavior is the type of design in which these constructs are measured. Assessing any type of risk perception and behavior in a cross-sectional design makes it difficult to determine whether bias influences behavior, behavior influences bias (or both), or whether a third variable (such as education or negative affectivity) influences both (Gerrard et al., 1996; Weinstein, Rothman, & Nicolich, 1998). The same problem applies when attempting to link biased risk perceptions with other constructs. Although there is now a growing literature using prospective designs to assess the link between risk perceptions and behavior, very few of these studies evaluate the accuracy of these risk perceptions.

Reliability. Given the difficulty of measuring optimistic biases at the level of the individual, there are few if any studies that determine the test-retest reliability of optimistically biased judgments. Moreover, because bias is usually established for single events, no data are available to determine whether bias is consistent across multiple events, so there are no published scales that measure a general form of the optimistic bias. Although some studies have collapsed comparative ratings across multiple events based on high reliability coefficients and identified the collapsed index as a generalized measure of optimistic bias (e.g., Davidson & Prkachin, 1997; Taylor et al., 2003), these measures are better characterized as generalized risk beliefs rather than biased risk beliefs per se. However, it is worth noting that comparative risk judgments have been shown to be reliable over time (Shepperd, Helweg-Larsen, & Ortega, 2003), suggesting that biases in these judgment may also be reliable.

<< Previous

Search | Help | Contact Us | Accessibility | Privacy Policy

DCCPSNational Cancer Institute Department of Health and Human Services National Institutes of Health USA.gov

Health Behavior Constructs: Theory, Measurement, & Research  You can Quit smoking Now! smokefree.gov