Study II: Reconstructed RCMs

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.