Solving parsons problems versus fixing and writing code

159 recruited, 135 analysed (45 fix, 44 Parsons, 46 write)

Undergraduates in CS1 for computing majors, Python, at a US research-intensive university.

Three-arm random assignment with a pretest, an immediate post-test, and a one-week retention post-test isomorphic to the first — the retention measure is what makes this worth citing. Recruitment was by extra credit, so there is self-selection into the study though not into condition. The authors report a suspected CEILING EFFECT on the post-tests, with many participants at maximum score, which is exactly the condition under which a no-difference result is least informative. Researcher-designed instruments throughout.

Parsons problems with distractors took significantly less time than fixing or writing the equivalent code, F(2,133) = 10.835, p < 0.001, significant against both alternatives, with no time difference between fixing and writing (for example problem one: 84.2 seconds for Parsons against 114.5 for fixing and 171.6 for writing). There was NO significant difference between conditions on the immediate post-test, on any post-test subtype, or on the one-week retention post-test. Same learning in roughly half the time — under a ceiling effect the authors flag themselves.

Low. Random assignment to condition.

The efficiency-without-loss result was reproduced by the same team with adaptive Parsons problems (Ericson, Foley & Rick 2018), which is internal rather than independent replication. Du, Luxton-Reilly & Denny (2020) explicitly flag a lack of replicated research across the Parsons literature.