A Latent Class Application to the
Multidimensional Measurement of Poverty
National Research and Development Centre for Welfare and Health (STAKES) P.O. Box 220,
Fin-00531 Helsinki, Finland, E-mail: firstname.lastname@example.org
Abstract. The paper presents the multidimensional measurement as a transparent and easy-to-
interpret method to measure poverty, where poverty is measured with a set of direct and
indirect poverty indicators side-by-side. Multidimensional measurement is formalised and
compared to the traditional, one-dimensional measurement. This formalisation is based on the
idea about a set of indicators that are measuring diﬀerent manifestations of the same latent
variable. The Latent Class Model (LCM) is proposed as a method to select a valid and reliable
set of poverty indicators for multidimensional measurement. The LCM is used to test if these
diﬀerent poverty indicators really measure the same latent referent – an assumption on which
the multidimensional measurement is based. Before this method presented here, constructing
and selecting indicators for the multidimensional measurement of poverty has relied practi-
cally on theory and substance only. Naturally, the method presented here can be used gen-
erally for studying and developing multidimensional measurements.
Key words: latent class, measurement theories, validity, reliability, multidimensional mea-
surement, poverty measurement.
A measurement device is the tool that is used to translate an observed
phenomenon to the language of statistical mathematics. In this one has to
justify that the scale corresponds with the measured phenomenon and that
the measurement device is reliable and valid. Sometimes this justiﬁcation is
easy. For example, when classifying people according their gender, the scale
is dichotomous and it is easy to construct a valid and reliable measurement
device. The survey question ‘‘are you male or female’’ with two choices to
answer ‘‘Male ¼ 1, Female ¼ 2’’ is a reliable and valid measurement device
for the classiﬁcation. The value one is then treated as value that one has if
(and only if) one is a man. A small number of misclassiﬁcations are due
measurement error when a woman (or a man) has marked the value one (or
two), because of carelessness or just for fun. This is an example of one-
dimensional measurement, where there is no uncertainty about the vector of
Quality & Quantity 38: 703–717, 2004.
Ó 2004 Kluwer Academic Publishers. Printed in the Netherlands.