semester, we administered the survey by email to faculty and in paper form to students in three CS courses: first-semester introductory (CS1), second-semester introductory (CS2), and senior-level capstone design.

Discussion

We obtained responses from 13 faculty (of a total of 25). For student surveys, we received 71 surveys from CS1, 48 from CS2, and 41 from senior capstone. The survey was voluntary, though no more than one or two students in each class declined to participate. No surveys contained a response to the catch item, but we did reject one survey because the last three pages had identical responses for each item.

We tallied responses by grouping “strongly disagree” and “disagree” as negative responses, “strongly agree” and “agree” as positive responses, and all other responses, including omitted responses, as neutral. We examined the responses by faculty to classify the items as either rejected or endorsed by faculty. Using the criterion that 75% or more of faculty had to agree to reject or endorse an item, we excluded five items as not showing consensus among the faculty (see cluster 2 in the table).

We placed the remaining 27 items in thematic categories using a combination of what Adams et al. 1 called “predeterminism” and “raw statistical” grouping. We first sorted them into groups reflecting our sense of the relationships among them, without reference to the data ( predeterminism) and used hierarchical cluster analysis, a statistical technique, to identify items participants commonly responded to in the same way (using the SPSS 16 package2).

Before we performed cluster analysis, we transformed responses for items in the same thematic category so answers reflecting a related underlying attitude would be coded the same. For example, we transformed the responses to item 64 (“If you can do something you don’t need to understand it”) so a negative response would match a positive response to item 12 (“I am not satisfied until I understand why something works the way it does”).

We used the results of the cluster
analysis to modify the groupings to
bring the resulting categories in line
with the data, where appropriate. That
is, where the data showed the partici-
pants commonly answered two items
the same way, we grouped these items
together, even if they were not grouped
together in our original classification.
In other cases, where the data showed
that two items we thought were related
were actually commonly answered dif-
ferently, we adjusted the grouping to
reflect that fact.