M. Wilke et al.: Risk Specificity for Risk-Defusing Operators
35
Table 2
Explanation of Coding Categories Shown With Examples From the Scenario “Avian Flu Pandemic” (see Appendix)
|
Category |
Explanation |
Example |
|
Situation |
Information regarding the decision |
“How many people are already affected in |
|
Background knowledge and |
Background information, which does not |
“To my knowledge, the quarantine measure |
|
Probabilities |
Likelihood of an event or prognosis for the |
“How probable are the brain-related side |
|
Negative consequences (extent of |
Negative consequences for choosing a |
“What brain-related side effects are we |
|
Positive consequences (advantages / |
Positive consequences for choosing a |
“Would a quarantine be cheaper as compared |
|
New alternatives |
Options that point to the investigation of |
“Are there any alternative precautions, for |
|
Control |
Control of the event or the negative |
“Could the danger of side effects be reduced, |
|
Worst-case plans |
Anything that can be done in case of |
“Could they treat the side effects in the brain |
|
Information about RDO Attitudes / rules / principles |
Receive more information about the RDO. Statements concerning the content of a |
“How would they treat the side effects in the “Basically, I am not in favor of mass vaccination.” |
Logit analyses describe a direct relation between inde-
pendent and dependent variables. The aim of the estimation
of multiple influence factors (type of risk and risk domain)
is to make their specific effects visible (Urban, 1993). For
example, the natural logarithm of the ratio between the fre-
quencies of both categories of dependent variables (no vs.
at least one question or statement concerning new alterna-
tives, possibilities for control, or worst-case plans) are rep-
resented as a sum of effect parameters (λ coefficients) un-
der the influence of selected categories of independent
variables (e.g., of the interaction between the normal type
of risk and the risk domain of politics). The explained quan-
tities are therefore not the cell frequencies themselves, but
the ratio between two probabilities of specific expressions
of variables, so-called “odds ratios” (Andreβ, Hagenaars,
& Kühnel, 1997).
This way, the directed hypotheses regarding the influ-
ence of type of risk as well as other possible differences in
the search for RDOs could be tested. Based on the compu-
tation of λ coefficients, the rank order of the influence fac-
tors type of risk, risk domain, and interaction on the influ-
ence of the three RDOs new alternatives, possibilities for
control, and worst-case plans was computed. The estima-
tion of parameters and effect sizes from the logit model is
based on maximum likelihood. Using one-dimensional χ2
tests, the frequency distributions of different variables like
background knowledge, attitude, situation, probability, and
positive and negative consequences were analyzed in rela-
tion to the four types of risks.
Results
Mean frequencies can be misleading if one participant asks
a lot of questions in one category and other participants ask
no questions at all (see Huber et al., 2001). Therefore, we
counted the number of participants with at least one state-
ment or question in a given category.
Data Analysis Based on Logits
First, we specified a model which identified the main ef-
fects of type of risk and risk domain as well as the interac-
tions between them. The resulting Pearson χ2 (p = 0.64) and
the likelihood ratio (p = 0.34) showed that the observed and
expected values were congruent. Thus, the specified mod-
el can be used to explain our data.
Testing for risk specificity of active risk defusing re-
vealed a significant negative λ coefficient for possibilities
for control and worst-case plans for normal risks: λ = -0.85,
p < 0.05. For normal risks, fewer participants formulated
questions or statements for possibilities for control and
worst-case plans than for new alternatives. For medium
Swiss J Psychol 67 (1), © 2008 by Verlag Hans Huber, Hogrefe AG, Bern