Database Reference
In-Depth Information
each other's fi ndings. The only result related to user attitudes in these studies was made by
Shoval and Shiran (1997), who found that ER/EER users' quality perceptions were higher
than those of OO users.
Effects of user characteristics on performance and attitudes.
Seven empirical studies
have signifi cant results regarding the effects of user characteristics on performance and at-
titudes, and all of them have focused on some type of task-related experience. The results
do not, unfortunately, build a highly consistent image because every study has investigated
a different aspect of experience. Therefore, the studies will be discussed here in chronologi-
cal order. Batra and Davis (1992) confi rmed that well-known process differences between
novices and experts could be observed also within this domain. Siau et al. (1995) found out
that domain familiarity did not have an impact on the choice between optional and manda-
tory relationships; subjects (experts) almost invariably chose to use an optional relationship
construct. According to Agarwal et al
.
(1996), subjects with experience in modeling with a
process focus are able to utilize this experience when they are modeling behavior but not
with data structures. Weber's (1996) results in his experiment using a recall task suggest
that although NIAM experts' ability to recall model elements was slightly better than that
of novices, their memory structures and recall strategies were the same. Lee and Choi's
(1998) results regarding the differences between experienced ER modelers and novices are
somewhat diffi cult to interpret, but it appears that in most respects ER experience led to
higher performance with the other methods, too, although experienced modelers used more
time. In all cases but one (ORM), experienced ER modelers perceived the methods to be
easier to use than inexperienced modelers did. According to Ramesh and Browne (1999),
“database-naive” subjects were better able to express causal relationships than “database-
knowledgeable” subjects, and they attribute this to the inability of commonly used modeling
formalisms to support the expression of causal relationships. Finally, Burton-Jones and Weber
(1999) studied the effects of domain knowledge and ontological clarity of a representation
on the subjects' ability to answer problem-solving questions. Their results provide limited
support to the claim that ontological clarity is particularly important in cases when domain
knowledge is low.
Effects of task characteristics on user performance and attitudes.
None of the studies
have directly focused on the effects of task characteristics on the main dependent variables,
although four of them (Hardgrave & Dalal, 1995; Liao & Palvia, 2000; Shoval & Even-
Chaime, 1987; Weber, 1996) used task complexity as an independent variable and all of
them found a main effect for complexity on performance (in practice, this means that the
experimental manipulation worked). This is understandable because in most cases, the fo-
cus is on the moderating effects of task characteristics on the effects of other variables on
performance, particularly the model formalism and user characteristics.
Differences between facets.
As discussed above, most of the studies have used some
version of the facet structure for analyzing user performance since Batra et al. (1990) origi-
nally presented it. Five of them have analyzed user performance in one or several of these
facets with measures that are similar to each other and give us an opportunity to review users'
relative performance with various facets. The performance data per facet from these studies
are included in Table 2; no aggregate data is presented here because it is not in all cases
clear whether or not the methods have been similar enough to justify the use of composite
measures. This data does, however, lead to the following observations: 1) Identifying and
modeling ternary relationships correctly is diffi cult for novice users, and even in the relatively
simple experimental tasks users' average performance level is often below 50%. The range
