Study II: Reconstructed RCMs
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118
119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
136 137 138 139 140 141 142 143 144 145 146
Recall that the factor analysis of the survey data yielded five factors: Nesting, Object
Characteristics, Behavior II, OO Modeling/Analysis II, and OO Development Concepts
II. The reconstructed map can be seen in Figure 2(b). The density of the overall map is
1.80. The centrality of the five constructs varied from a low of 0.11 (OO Development
Concepts II), 0.33 (Object Characteristics), to 0.44 (Behavior II) to a high of 0.56 (OO
Modeling/Analysis II, and Nesting). Table 6 also presents the reachability matrix for the
reconstructed (aggregated construct level) RCM based on the survey data. The aggregated
RCM of the OO software developers reveals that developers learning OO tech-
Figure 2(a). Aggregated construct level revealed causal map, interview data
Figure 2(b). Data aggregated construct level reveal causal map
Interview A B C D
A. Structure 0.104 - 0.260
B. Behavior 0.094 - 0.219
C. OO Modeling/Analysis I 0.042 0.042 0.188
D. OO Development Concepts 1 - - -
Survey A1 A2 B C D
A1. Nesting - - 0.025 -
A2. Object Characteristics 0.047 - - -
B. Behavior II 0.074 0.036 0.045 -
C. OO Modeling/Analysis II 0.098 0.045 0.031 -
D. OO Development Concepts II 0.029 - - -
Table 63. Reachability matrices for aggregated construct level RCM
niques (with the exception of OO Development Concepts II being a cause construct) do
not see clear cause and effect constructs within the OO software development approach.
The reachability values between the constructs were fairly consistent (mean .05), with
the linkages into the Nesting construct (effect) slightly stronger.
A Comparative Analysis of Revealed and Reconstructed
RCMs
There are similarities and differences between the cognitive structures that constitute
OO software development expertise yielded by revealed and reconstructed RCMs. See
Table 7 for comparison of the concepts and constructs. From the demographics of the
two samples we can see that the respondents for the interview data had about one
additional year of OO experience, but had completed twice the number of OO projects as
the respondents for the survey data. Based on the differences in experience, the
difference between the interview (expert) cognitions and the survey cognitions should
be quite large. When comparing the OO interview RCMs to the survey RCMs there should
be differences because the respondents are at different places in the learning process.
Recall that the factor analysis yielded five constructs, one more than the theoretical
framework that best fit the evoked concepts. As shown in Table 7, there was significant
overlap between the Behavior, OO Modeling/Analysis, and OO Development Concepts
constructs across the two methods. The survey responses separated into two constructs
(Object Characteristics and Nesting) what the best fitting theoretical framework combined
into one (Structure). This is consistent with general theories of expertise on
Table 7.4 Comparision of Study I versus Study II concepts and constructs
Study I Construct Study I Concepts Study II Constructs Study II Concepts
Structure Abstraction Nesting Inheritance
Attribute OO Development
Class Polymorphism
Encapsulation Object Characteristics Attribute
Information Hiding Class
Inheritance Encapsulation
Instantiation Instantiation
Object Method
Behavior Collaboration Behavior II Collaboration
Message Passing Message Passing
Method Relationship
Polymorphism
Relationship
OO Modeling / Analysis Identifying Objects OO Modeling / Analysis II Identifying Objects
Noun-Verb Analysis Noun-Verb Analysis
Object Model Object
OO Development Concepts Patterns OO Development Concepts II Abstraction
Layer Patterns
OO Development Information Hiding
Layer
Object Model
software development that asserts that as developers gain expertise they create larger
chunks of information with more abstract representations (e.g., Adelson, 1981;
McKeithen, Reitman, Rueter & Hirtle, 1981; Murphy & Wright, 1984; Pennington, 1987).
In addition to the basic structure of the maps there are several similarities and differences
in the linkages of the maps. As shown in Table 6, the reachabilitiy matrices demonstrate
that there is a common linkage from the Behavior construct to the Structure (Nesting and
Object Characteristics) constructs. In addition, the OO Modeling/Analysis construct
has two linkages in common across the maps, with the Structure (Nesting and Object
Characteristics) and Behavior constructs. The centrality measures highlight some
differences in the maps. The interview data indicated two layers of centrality, with layer
one including the Structure and Behavior constructs, and layer two including the OO
Modeling and OO Development constructs. This indicates a relatively flat cognitive
structure. In contrast, the survey data indicated three layers, with layer one including the
OO Development construct, layer two including the Behavior and Object Characteristics
constructs, and layer three including the Nesting and OO Modeling constructs. In
addition to more layers, the survey data map had a wider range of centrality scores than
the interview data map. This indicates a more hierarchical and complex cognitive
structure in which some concepts are more central and others more ancillary to the
domain.
The reachability measure provides another mechanism for comparison. In the interview
data map the OO Development construct has by far the highest reachability with a large
variance among the reachability values. In contrast, in the survey data map the reachability
values have a much smaller range with the Nesting construct having the highest
reachability. Thus while the content of the maps is quite similar, the main difference in
the two maps is in the linkages.
Discussion
Before discussing our results, we should place them in context, noting that OO development
techniques have only come into vogue in recent years. This low level of maturity
has three correlates: lack of theoretical parsimony, standardization and
professionalization. As a field matures, developments are likely to result in parsimonious
theory and consequently fewer clearly articulated concepts. This makes possible
standardization of approaches, very much akin to the emergence of technical standards.
In turn this allows transmission of explicit knowledge thus facilitating professionalization
of practitioners. In our search for an adequate theoretical scheme to categorize the
evoked concepts in OO, we found a variety of frameworks. Even the best fitting framework
could embrace only slightly over half the evoked concepts. The tacit knowledge of the
experts seems to be more extensive than represented by the available theoretical
frameworks. Clearly, theoretical development in OO software development is far from
complete.
Under these conditions, evocative approaches such as RCM appear not merely to be an
adornment, but a necessity in research to capture the “true” phenomena under investigation.
Against the backdrop of the lack of consensus among theoretical frameworks
describing OO, the correspondence between the RCM and survey responses augurs well
for the former, newer method, which is alleged to be a more appropriate tool for cognition
research (Huff, 1990). This is one of the first studies designed to develop a cognitive
representation of expertise in OO software development and then empirically test that
representation. Its results, consisting of the concepts, constructs and the linkages
among them — in short the cognitive structuring of expertise in OO — provide a starting
point for empirically representing knowledge structures.