Quality & Quantity 38: 603–620, 2004.
© 2004 Kluwer Academic Publishers. Printed in the Netherlands.
Using Standardised Tables for Interpreting
Management Studies Group, Wageningen UR, Hollandseweg 1, 6706 KN Wageningen, The
Netherlands, E-mail: John_Hendrickx@yahoo.com
Abstract. Loglinear models are a useful but under-utilised research tool. One of the reasons for
this is the difﬁculty of explaining model coefﬁcients to others. Presenting coefﬁcients in the form
of standardised tables can help communicate the results to readers with little or no background in
loglinear models. The strength and direction of association is conveyed in the familiar form of a
percentage table, allowing an intuitive grasp of the model results.
Key words: loglinear models, standardised tables, coefﬁcients, interpretation
The basic idea for this article originated in 1993 while working on an article us-
ing loglinear models to analyse religious intermarriage (Hendrickx et al., 1994).
Professor Schreuder was unfamiliar with loglinear models and I tried to explain
them to him, pointing out that a key advantage for the problem at hand was that
they are based on probabilities. Schreuder wondered why he had to struggle with
the confusing coefﬁcients of loglinear models – why couldn’t I just present the
probabilities? I didn’t really have a good answer at the time but this is a renewed
The coefﬁcients of loglinear models are difﬁcult to comprehend, if only because
there are so many variations. The coefﬁcients can be (log)linear or multiplicative,
can have a ﬁxed reference category, sum to zero or have a product of 1. Other
restrictions can be imposed as well, although these are less common. Another
source of confusion is that the coefﬁcients are relative measures. The odds-ratio,
a common type of loglinear coefﬁcient, summarises the relationships among 4
probabilities in one number. This can be very confusing and requires a careful
explanation, certainly to an audience with little statistical background.
Interpretation is somewhat simpler in the case of logit/logistic and related mod-
els, which are closely related to loglinear models and also present the results as
(log) odds-ratios. This is because logistic models have a dependent variable, which
is not necessarily the case for loglinear models. Given a dependent variable, the
researcher can present predicted probabilities for given values of the independ-