In order for students across the sciences to avail themselves of modern data streams, they must first know how to wrangle data: how to reshape ill-organised, tabular data into another format, and how to do this programmatically, in languages such as Python and R. Despite the cross-departmental demand and the ubiquity of data wrangling in analytical workflows, the research on how to optimise the instruction of it has been minimal. Although data wrangling as a programming domain presents distinctive challenges – characterised by on-the-fly syntax lookup and code example integration – it also presents opportunities. One such opportunity is how tabular data structures are easily visualised. To leverage the inherent visualisability of data wrangling, this dissertation evaluates three types of graphics that could be employed as scaffolding for novices: subgoal graphics, thumbnail graphics, and parameter graphics. Using a specially built e-learning platform, this dissertation documents a multi-institutional, randomised, and controlled experiment that investigates the pedagogical effects of these. Our results indicate that the graphics are well-received, that subgoal graphics boost the completion rate, and that thumbnail graphics improve navigability within a command menu. We also obtained several non-significant results, and indications that parameter graphics are counter-productive. We will discuss these findings in the context of general scaffolding dilemmas, and how they fit into a wider research programme on data wrangling instruction.