The Structure of Objectivist Studies - Evaluation Methods in Biomedical Informatics - page 104

Descriptive Studies
A descriptive design seeks only to estimate the value of a dependent vari-
able or set of dependent variables in a selected sample of subjects. Descrip-
tive designs have no independent variable. If a group of nurses were given
a rating form (previously validated through a measurement study) to ascer-
tain the “ease of use” of a nursing information system, the mean value of
this variable would be the key result of a descriptive study. If this value were
found to be toward the low end of the scale, the researchers might conclude
from this descriptive study that the system was in need of substantial revi-
sion. Although they seem deceptively simple, descriptive studies can be
highly informative. Teach and Shortliffe's 12 examination of physicians' atti-
tudes toward medical decision support is an example of a descriptive study
that has had substantial impact “simply” by asserting that physicians, as a
group, appear to share certain beliefs. Studies of the quality of health infor-
mation on the Internet also illustrate the importance of well-conducted
descriptive studies. 13 Descriptive studies also can be tied to the “objectives-
based” approach to evaluation described in Chapter 2. When an investiga-
tor seeks to determine whether a resource has met a predetermined set of
performance objectives, the logic and design of the resulting demonstration
study may be seen as descriptive.
Comparative Studies
In a comparative study, the investigator typically creates a contrasting set
of conditions. After identifying a sample of subjects for the study, the
researcher either assigns each subject to one of these conditions or classi-
fies them into one of the conditions based on some predetermined charac-
teristic. Some variable of interest is then measured for each subject, and the
measured values of this variable are compared across the conditions. The
contrasting conditions comprise the independent variable(s) for the study,
and the “variable of interest” is the dependent variable. The study by
MacDonald and colleagues 14 of the effects of reminder systems is a classic
example of a comparative study applied to informatics. In this study, groups
of clinicians (subjects) either received or did not receive computer-
generated reminders (conditions comprising the independent variable), and
the investigators measured the extent to which clinicians took clinical
actions consistent with the reminders (dependent variable). Comparative
studies are aligned with the “comparison-based” approach to evaluation
introduced in Chapter 2.
When the experimental conditions are created prospectively by the inves-
tigator and all other differences are eliminated by random assignment of
subjects, it is possible to isolate and explore the effect due solely to the dif-
ference between the conditions. Under this design, the investigator also may