isbn9789526046266

are “somewhat contrary to the classic work of Spohrer and Soloway”: students are sometimes able to
create buggy template-based (schema-based) code, but do not necessarily understand the code that they
themselves produce, and cannot trace its execution to fix the bugs (Thomas et al., 2004).
Overall, recent research does not disagree with Spohrer and Soloway’s identification of problem-solving
schemas as important, but does emphasize that lower-level issues are also worthy of attention.
Garner, Haden, and Robins categorized the problems that CS1 students needed help with in class and
examined the resulting distributions (Garner et al., 2005; Robins et al., 2006). Many of the problems had
to do with program design, but many also involved various construct-related issues. Trivial mechanical
problems (e.g., missing semicolons) were also common. The distribution pattern was similar across course
offerings; high-, medium-, and low-achieving students also all exhibited similar patterns. Fitzgerald et al.
(2008) report that novice programmers with 15 to 20 weeks of programming experience behind them
had more trouble finding non-construct-related bugs than construct-related ones. Ko and Myers (2005)
presented a framework for analyzing the causes of software errors; applying it, they found that many bugs
were rooted in cognitive difficulties with particular language constructs, although other causes of difficulty
were common as well. Denny et al. (2011) demonstrated that novices have great trouble in producing even
small programs that are syntactically correct. McCauley et al. (2008) point out that construct-related
bugs feature even in a revered pedant’s classification of the bugs in TeX (Knuth, 1989).
Drawing on the BRACElet studies (Chapter 3), Lister (2011b,c) has recently emphasized that although
struggles with syntax initially contribute to students’ cognitive load – and are important to address –
the great cognitive load inherent in program writing remains a major concern even when those earliest
problems have been overcome. Kranch (2011) studied novice programmers taught using three different
topic sequencings, including one that started with higher-level schemas before moving to their elements,
and one that did the opposite. He found that irrespective of ordering, the higher-level schemas were
always the most difficult topic (both in terms of achievement and student-given difficulty ratings), a
finding that he attributes to higher intrinsic cognitive load. Kranch argues that novices should be given
ample opportunity to practice lower-level fundamentals before they are taught higher-level schemas.
The studies from the 1980s that pointed to the relatively minor importance of misconceptions used
simple imperative programs. It is not obvious to what extent the findings are generalizable to objectoriented
programming, which has brought a slew of new concepts and misconceptions into many CS1
courses. Many OOP misconceptions are about the fundamental nature of pivotal concepts such as object
and class (see Appendix A).
Conclusion: they all matter
My conclusion from the literature is that all five challenges – cognitive load, plan schemas, misconceptions,
tracing skills, and notional machines – are significant. Furthermore, none of these challenges is independent
of the others. Of particular interest to the present work is the fact that the other four challenges appear
to be affected by the fifth: the novice’s ability to understand the notional machine. Understanding the
role of the machine in program execution:
• prevents and corrects numerous misconceptions;
• serves as a basis for the formation of low-level schemas which in turn form the basis for the successful
formation and application of higher-level ones;
• thereby indirectly reduces cognitive load as one is able to rely on increasingly complex problemsolving
schemas, and
• is key to the skill of tracing programs, which is useful in both program authoring and comprehension
and in finding bugs whether they be misconception-related or the results of high-level schema
failures.
The notional machine is not a singular bottleneck responsible for students’ struggles, but it does appear
to be one of several main sources of difficulty.
70