
119Testing the Parallel Regression Assumption
Although the score test is easier to compute, it also has several well-known limitations.
Models based on large sample sizes with many continuous independent variables are
more likely to violate the parallel assumption according to the score test (Allison 1999,
p.141; O’Connell 2006, p. 47). In addition, a signicant Brant test does not necessarily mean
that the partial or nonparallel model should be preferred over the parallel model (Greene
and Hensher 2010, p. 187). There are a number of factors that could lead to a signicant
Brant test statistic, including a misspecication of the linear predictor ...