Some industrial applications lead to observations that are characterized by
a bivariate response where one observation is an angle or direction and th
e other is linear. These data are frequently called cylindrical data since
they can be represented by points on a cylinder of radius one. In these ind
ustrial applications when several related, but distinct groups produce data
of this variety, the analyst often nees to study the common and distinct f
eatures of the data across the groups. This paper presents a method for tes
ting whether different functional forms of the relationship between compone
nts are required and two strategies for determining an appropriate final mo
del. The goal of both of these strategies is the parsimonious characterizat
ion of the data, while preserving the unique characteristics of the groups.
A constructed example is presented to illustrate the methodology.