Recent papers document that student problem-solving competence varies (often strongly) with representational format, and that there are significant differences between the effects that traditional and reform-based instructional environments have on these competences [Kohl and Finkelstein, Phys. Rev. ST Phys. Educ. Res. 1, 010104 (2005); Kohl and Finkelstein, Phys. Rev. ST Phys. Educ. Res. 2, 010102 (2006)]. These studies focused on large-lecture introductory physics courses, and included aggregate data on student performance on quizzes and homeworks. In this paper, we complement previous papers with finer-grained in-depth problem-solving interviews. In 16 interviews of students drawn from these classes, we investigate in more detail how and when student problem-solving performance varies with problem representation (verbal, mathematical, graphical, or pictorial). We find that student strategy often varies with representation, and that in this environment students who show more strategy variation tend to perform more poorly. We also verify that student performance depends sensitively on the particular combination of representation, topic, and student prior knowledge. Finally, we confirm that students have generally robust opinions of their representational skills, and that these opinions correlate poorly with their actual performances.