Different answer formats lead to different risk estimates (Symposium: Psychological aspects of risk perception and behavior)

Measuring risk estimates is important for various scientific disciplines, the economy as well as daily life. However, to date knowledge on the question of how risk estimates should be measured is ambiguous and the underlying... [ view full abstract ]

Measuring risk estimates is important for various scientific disciplines, the economy as well as daily life. However, to date knowledge on the question of how risk estimates should be measured is ambiguous and the underlying psychological processes are still not well understood. In a previous study we examined the impact of different answer formats (rating scale, percentages, open and closed frequency scale) and risk perspectives (target person e.g., abstract or specific person) on risk assessment (Lermer, Streicher, Sachs, & Frey, 2013). Results showed that risk estimates are strongly influenced by the used questionnaire. We assume that the questionnaires trigger different systems of probabilistic reasoning: a distributional approach (for abstract targets) leading to higher risk estimates and singular approach (for specific targets) leading to lower estimates. Moreover, the risk perspectives’ impact varied with the used answer format. In order to explore in which situations which instrument is most useful two further studies were conducted. In these studies the influence of different answer formats was explored by seven different scales (seven-step verbal scale, -3 to +3 scale, 1 to 7 scale, percentages, visual analogue scale, open and closed frequency scale). In the first study the target person was abstract. In order to explore the questionnaires’ impact on risk estimates’ accuracy we asked participants to estimate the probability for a person in Germany to die from different causes (e.g., stroke). Participants’ estimates were compared with cause of death statistics from the Federal Statistical Office of Germany. Results showed that risk estimates’ accuracy varied with the used scale. In the second study we used the same set of scales and asked participants to estimate personal risks. Another modification was that we used a broader set of items from different domains (e.g., household or illness) and probability ranges (from rare to frequent events, e.g., hepatitis b or power failure). Again, results showed that different answer formats lead to different risk estimates. Thus, it might be assumed that findings in literature regarding risk estimates are biased by the used measurement. By combining the results of our studies we aim to develop an answer to the question under which circumstances which answer format is advisable and when it should not be considered.