Facilitating API lookup for novices learning data wrangling using thumbnail graphics

Figure 1.  Kelleher & Ichinko's Collection and Organization of Information for Learning (COIL) model [20,15]

Figure 2.  The platform is split into three parts. Part 1 introduces the user to an ontology of data wrangling operations. Part 2 introduces programming. Part 3 contains 18 programmatic data wrangling exercises

Figure 3.  The sidebar menu and a Part 1 operation card under the two conditions

Figure 4.  A snippet from the sidebar menu. Calculate is one of five top-level categories, while the next level is split by data structure (e.g. dataframes). In TG, each leaf node has a thumbnail graphic

Figure 5.  Three examples of graphical thumbnail, using color in different ways to convey operation semantics

Figure 6.  Part 1 contains a series of exercises in which the user selects operations from the menu and drags it to the corresponding subgoal

Figure 7.  Part 3 involves programming exercises. The user is guided by a list of subgoals, each of which has associated hints. The sidebar menu serves as a menu for looking up API documentation (shown above). In reality, the menu, subgoals and documentation are all tabs within the same sidebar panel

Figure 8.  Participants were asked to rate their experience with Excel, Python, and R on a five-point Likert-scale (1 = Not at all to 5 = Advanced). The distribution among people who started and completed each part is illustrated and does not provide any visual evidence for differences that would reflect that more experiences participants are more likely to persevere

Figure 9.  The distributions in the number of tooltip events per person, grouped by condition, in Part 1 (left) and Part 3 (right). Dashed lines indicate medians. In both cases, the TG group uses the tooltip significantly less often

Figure 10.  Total number of menu clicks per participant in Part 3, grouped by condition. Dashed lines indicate medians

Figure 11.  Total reading times of operation cards per person, grouped by condition. Dashed lines indicate medians. The TG group is quicker on average, but the difference is non-significant

Figure 12.  Total time on task per person for Part 1 (left) and 3 (right), grouped by condition. Dashed lines indicate medians. The median differences are in both parts negligible

Figure 13.  Number of incorrect attempts per person for Part 1 (left) and 3 (right), grouped by condition. Dashed lines indicate medians. In both parts, the TG group makes fewer incorrect attempts, but the difference is not significant

Figure 14.  Responses to the evaluation survey item asking participants how helpful they found the thumbnail graphics and tooltips (1 = Not at all, 5 = Very much) in Part 1 (N = 187) and Part 3 (N = 115). This survey was given both after Part 1 (left) and Part 3 (right)