1
Probability-Based versus Parametric Methods of Estimating Stress Determinants:
Models for Russian Federation
Babichenko J. S.
PhD student
Tallinn University, Institute of International and Social Studies, Tallinn, Estonia
In the situation of unstable business environment, being a result of economic volatility,
the behavioral patterns of an individual can become especially interesting, having consequent
influence on market behavior. One of the factors, which can be of special interest, is the degree
of work-related stress and fatigue, influencing individual perceptions and, correspondingly, the
behavior of an individual in relation to the market. Here, two types of relations can be taken
into account: the influence on organizational effectiveness (or internal type) and the influence
on the behavioral habits of an individual on the market (external type), analyzed mainly in
behavioral finance discipline. The objective of the current paper is the try to elaborate the
system of work-related stress determinants (defining a set of indicators and their classification
in groups) and model data for Russian Federation from ISSP databases with the aim
to determine significant dependencies (determinants' relationships with work-stress factor
behavior) for the first type of relations.
One special question is the choice of appropriate statistical procedures, as in the current
research the data is the survey-based Likert-scale type variables, when the assumptions of
multivariate normality and linearity may not hold. The character of raw data imposed a set of
limitations to reliable mathematical analysis, as: a) most of the variables are non-normally
distributed, b) no predetermined exact theoretical model specification exists, c) non-linear
dependencies can hold, d) data consists of nominal and ordered categorical variables. Taking
into account the data set characteristics and the absence of mathematically predetermined
model type, two methods were chosen for the analysis with the aim to combine different
approaches to mathematical formulation of the problem: probabilistic approach and parametric
approach (ordered models). Namely, the first approach is the Bayesian modeling, designed to
analyze discrete categorical variables with no limitations of sample size and no linearity
assumptions. Here, Bayesian classification networks (analyzed by the mean of B-course tool)
were constructed to test the significant (or, “more probable”) determinants, with Bayesian
classification modeling meaning searching for the best predictors for group memberships and
testing for class similarity, or estimating the probability that variable belongs to the certain
class. Here, with the best model searching algorithm the models are compared by calculating
the ratio of the probabilities, and the predictive accuracy of each model is estimated on the base
of leave-one-out cross validation. The second approach included the estimation of ordered
regressions with different link functions tested. The best type of ordered model for the current
data set was ordered logit (testing as well probit and extreme value functions for each sample).
The combination of the different categorical variables compatible methods has allowed making
a comparative significance analysis of work-related stress determinants.
The phenomenon of work-related stress can be directly or indirectly influenced by
multiple factors, so the structure of system of determinants can be quite complicated. In
general context, work-related stress can emanate from two types of sources: a) direct stressing
factors, b) factor of “realization” (balance of importance/presence of certain factor). Here, the
first group can be divided into two sub-groups: internal and external factors. The first sub-
group included real data (characteristics) of the workplace, such as: job characteristics group (8
variables) (income, advancement, independence, interesting job...); work climate group (2
variables) (relations between colleagues and management/employees); workplace
characteristics group (2 variables) (working hours, place); stress group (3 variables) (hard
physical work, work in dangerous conditions, possibility to loose the job); socio-demographic
group (age, sex, position). The second sub-group included different aspects, representing