How Does Experience Influence Developer Perceptions of Atoms of Confusion?

Abstract

Atoms of Confusion (AoCs) are small, syntactically valid code patterns that can increase cognitive load during program comprehension. Earlier research suggested that AoCs are common and potentially harmful, but more recent studies have questioned whether their effects generalize beyond less experienced developers. This confirmatory study aims to reexamine whether the presence of AoCs slows comprehension or alters repair preferences. Moreover, we examine whether these effects are moderated by developers’ programming experience. We investigate task completion time and the kind of repairs developers prefer when interacting with code containing AoCs. We propose a two-phase study consisting of a pre-screening questionnaire and a controlled experiment. The questionnaire will function as a qualification instrument. In the experiment, participants will complete eight Java comprehension tasks, four with an AoC and four without. For each task, developers are asked to identify a seeded defect and to rank three functionally equivalent repairs differing in AoC inclusion. Task completion time and the top-ranked repair will be analyzed using mixed-effects linear and multinomial regression models, with AoC presence as the manipulated factor and programming experience as a covariate.