Deviance mining is an emerging area in the field of Process Mining, with the aim of explaining the differences between normal and deviant process executions. Deviance mining approaches typically extract representative subprocesses characterizing normal/deviant behaviors from an event log and use these subprocesses as features for classification. Existing approaches mainly differ for the employed feature extraction technique and, in particular, for the representation of the patterns extracted, ranging from patterns consisting of sequence of activities to patterns explicitly representing concurrency. In this work, we perform a latitudinal study on the use of sequential and concurrency patterns in deviance mining. Comparisons between sequential and concurrency patterns is performed through experiments on two real-world event logs, by varying both classification and feature extraction algorithms. Our results show that the pattern representation has limited impact on classification performance, while the use of concurrency patterns provides more meaningful insights on deviant behavior.