76 P. Cerchiello, M. Iannario, and D. Piccolo
[22]. However, there is no consensus on the relationship between personality and risk
perception [5].
Another fundamental dimension related to the concept of risk deals with the
dichotomy between experts’ perceptions and those of the common people. The role of
experts is central in several fields, especially when quantitative data are not sufficient
for the risk assessment phase (i.e., in operational risk). Typically, experts’ opinions
are collected via questionnaires on ordinal scales; thereby several models have been
proposed to elaborate and exploit results: linear aggregation [9], fuzzy methods [2,
25] and Bayesian approaches [4].
Our contribution follows this research path, proposing a class of statistical model
able to measure the perceptions expressed either by experts or common people. In
particular we focus on the problem of risk perception related to the workplace with
regards to injury. Thus, some studies focusing on the relationship between organisa-
tional factors and risk behaviour in the workplace [21] suggest that the likelihood of
injuries is affected especially by the following variables: working conditions, occupa-
tional safety training programmes and safety compliance. Rundmo [20] pointed out
how the possibility of workplace injuries is linked to the perception of risk frequency
and exposure.
2 CUBmodels: description and inference
A researcher faced with a large amount of raw data wants to synthesise it in a way
that preserves essential information without too much distortion. The primary goal of
statistical modelling is to summarise massive amounts of data within simple structures
and with few parameters. Thus, it is important to keep in mind the trade-off between
accuracy and parsimony. In this context we present an innovative data-reduction
technique by means of statistical models (CUB) able to map different results into
a parametric space and to model distinct and weighted choices/perceptions of each
decision-maker.
CUBmodels, in fact, are devoted to generate probability structures adequate to
interpret, fit and forecast the subject’s perceived level of a given “stimulus” (risk, sen-
sation, opinion, perception, awareness, appreciation, feeling, taste, etc.). All current
theories of choice under risk or uncertainty assume that people assess the desirability
and likelihood of possible outcomes of choice alternatives and integrate this informa-
tion through some type of expectation-based calculus to reach a decision. Instead, the
approach ofCUBmodels is motivated by a direct investigation of the psychological
process that generates the human choice [15].
Generally, the choices – derived by the perception of risk – are of a qualitative
(categorical) nature and classical statistical models introduced for continuous phe-
nomena are neither suitable nor effective. Thus, qualitative and ordinal data require
specific methods to avoid incongruities and/or loss of efficiency in the analysis of real
data. With this structure we investigate a probability model that produces interpretable
results and a good fit. It decodes a discrete random variable (MUB, introduced by