logo

You are here

Warning message

Attention! This event has already passed.

Absenteeism data: new approaches to the statistical analysis of its aggregated and non-aggregated mode

Wednesday, 8 March, 2006 - 17:30
Campus: Brussels Humanities, Sciences & Engineering campus
Faculty: Psychology and Educational Sciences
auditorium Q.d
Joachim Dejonckheere
phd defence

The phenomenon of absenteeism in organizations is complex to study due to
the multiple, often coincidental, multilevel and interacting causes. Although
many explanatory models have already been presented, agreement is still
scarce. In nearly all research so far, it has been common practice to measure
absenteeism over a certain measurement period and subsequently count the
number of days absent as an aggregate measure (time lost) or the number of
non-adjacent absence periods, as a non-aggregate measure (absence
frequency). The applicability of commonly used statistical models to analyse
these data like Ordinary Least Squares (OLS) models and the Poisson
regression model (PRM) can however be questioned. An assessment of the
impact of the violations of their statistical assumptions sheds a light on, and
may question the validity of the massive body of studies in which these
previous models have been used. It can be concluded that results of the OLS
model are too conservative but can therefore be ‘trusted’, while interpretations
of the results of the PRM are more dangerous. As an alternative, the negative
binomial regression model (NBRM) is proposed since it is less restricted by its
assumptions. This could empirically be shown in a first study where various
alternative statistical models have been applied to six organizational samples
of real life absenteeism data.

The existing models are then further criticized because of their
conceptualisation of employee absenteeism as a unitary phenomenon. A new
model has therefore been advocated in which absenteeism is considered to be a
compound variable of two processes: first an employee gets ill, second a
decision is made on the number of days of absence. This new
conceptualisation of absenteeism, proposed as a ‘dual phase’ model is tested
with a zero-inflated statistical model.

The negative consequences of the usual aggregation of absenteeism data are
highlighted by contrasting the results of the NBRM and the zero-inflated
NBRM on aggregated absenteeism data with the technique of the multilevel
logistic regression model (MLRM) on non-aggregated absenteeism data. From
a second and third empirical application using absenteeism data from a sample
of 930 and 11790 employees respectively, it could be argued that the MLRM
has several advantages for the analysis of non-aggregated absenteeism data
compared to survival analysis and frailty modeling. Furthermore, we have
been able to demonstrate that the relationship of well-studied variables like age
and gender with absenteeism are not as straightforward as usually thought.
For example, it could be shown that this relationship is dependent on the
employee’s absence status of the previous day: e.g. while researchers
consistently find that women are more often absent than men, this relationship
is inversed when an absence spell is ongoing. The relevance of such findings
for further theorizing about employee absenteeism as well as for practical and
managerial applications are numerous.