| This paper focuses on robust inference on the shape parameter of a composite transformation model. Large sample robust tests and conÞdence intervals are derived from a quasi proÞle likelihood ratio statistic. This statistic is obtained from a suitable bounded proÞle estimating function, which deÞnes a robust estimator for the shape parameter, while the remaining parameters of the model are treated as nuisance. This test has desirable robustness properties and leads to more reliable inference then the classical robust Wald test statistic. |