Database Reference
In-Depth Information
Table 2: User modeling performance by facet in empirical studies
Batra et al.,
1990
Batra & Kirs,
1993
Bock & Ryan,
1993
Shoval &
Shiran,
1997
Liao & Palvia,
2000
Rel.
ER/
EER
Rel.
ER/
EER
ER/
EER
OO
ER/
EER
OO
Rel.
ER/
EER
OO
Entity
98.0
96.0
99.0
99.0
Identifi er
72.4 73.9
96.0
80.0
62.8
69.7
77.3
Descriptor
95.0
94.0
Category
92.0
82.0
99.0
99.0
Unary
68.3 55.2
96.0
64.0
88.0
70.0
59.9
40.0
50.0
Binary 1:M
54.4 84.9 50.6
81.2
89.0
88.0
83.0
89.0
54.2
83.8
73.9
Binary M:N
57.1 92.9 67.5
92.5
100.0
63.0
81.0
79.0
41.2
74.4
65.3
Ternary 1:M:N
8.3
41.3 46.9
60.0
47.0
44.0
85.0
68.0
Ternary M:N:O 33.3 45.2 40.6
45.6
79.0
72.0
94.0
76.0
35.4
47.7
57.5
of performance levels is, however, very large, varying from 8.3% for 1:M:N relationships
in Batra et al. (1990) to 94% for M:N:O relationships in Shoval and Shiran (1997). 2) Re-
sults are weak (below 70%) also for unary relationships, except with a semantic formalism
(ER/EER) in Bock and Ryan (1993) and Shoval and Shiran (1997). The range is large also
with this facet (from 40% to 96%). 3) With semantic and object-oriented modeling formal-
isms, users' average performance in modeling the binary relationships is consistently at a
high level (above 80%), with the exception of binary M:N relationships in Liao and Palvia
(2000). 4) Modeling identifi ers, a seemingly simple task, appears to cause diffi culties with
all modeling formalisms, with typical performance levels around 70%.
Findings related to the data modeling process.
Five of the studies included in this
review analyzed some aspect of the process that subjects followed while creating a data
model. As discussed earlier, Jarvenpaa and Machesky (1989) investigated whether the
subjects chose a top-down or a bottom-up approach when constructing data models and
whether the choice of the approach was dependent on the modeling formalism. They found
that users of the ER-based Logical Data Structure model were more likely to use a top-down
approach than the users of the relational model. Batra and Davis (1992) studied the protocol
differences between novices and experienced data modelers and found broad support for
several fi ndings from prior research regarding the differences between these two groups:
Experts had richer concept vocabulary and were better able to categorize constructs and
automate processes, whereas novices were more likely to make a range of modeling errors.
Batra and Sein (1994) analyzed at the individual level users' ability to improve the quality
of their data modeling solutions based on feedback and found out that feedback can help
users avoid errors in modeling ternary relationships. Srinivasan and Te'eni (1995) focused
entirely on the results of the process analysis of a specifi c modeling behavior. Using verbal-
ized protocols, they analyzed the use of several heuristics at various levels of abstraction to
manage the complexity of the data modeling process. The most important results reported
in Srinivasan and Te'eni (1995) were that effi cient data modelers use specifi c heuristics to
reduce the complexity of the problem, test models at regular intervals, and make orderly
