Developing and Improving Measurement Methods - Evaluation Methods in Biomedical Informatics - page 178

In this example, there is too much semantic space between “generally” and
“not at all.” There are three response options that reflect positive views of
the resource and only one option that is negative. To rectify this problem,
a response option of “slightly” or “modestly” might be added to the exist-
ing set.
4. Include an appropriate number of response options. Although it may
seem tempting to use a large number of response options to create at least
the appearance of precise measurement, the results might prove illusory. In
general, the number of response options should be limited to a maximum
of seven. 26 * For most purposes, four to six discrete options suffice. Using a
five-option response format with a bipolar semantic axis allows a neutral
response. We can offer arguments for and against neutral responses. A
potential benefit is that a respondent whose true belief is neutral has a
response option reflective of that belief. In the opposing view, a neutral
response option plays to the central-tendency problem. It provides a
way to respond that is safe and noncommittal, even though it may not be
reflective of the respondent's actual belief.
5. Invite a nonresponse. Giving respondents permission to decline to
respond to each item also contributes to successful measurement. When
using rating forms, for example, respondents may offer uninformed opinions
based on insufficient experience because they feel they are expected to com-
plete every item on the form. If an “unable to respond” category is explicitly
available, respondents are more likely to omit items on which they do not
feel confident or competent, which of course is what they should do. If an
“unable to respond” category is offered, it should be in a different typeface
or otherwise visually apart from the continuum of informed responses.
6. Request elaborations. Asking respondents specifically for verbal elab-
orations or justifications of their responses can serve multiple purposes. It
often forces them to be more thoughtful. Respondents may check off a spe-
cific option and then, when trying to elaborate on it, realize that their deeper
beliefs differ from what a first impression suggested. Elaborations are also
a source of valuable data, particularly helpful when the items are part of a
rating form that is in the early stages of development, to help validate the
form. Elaboration can also be informative as a source of evaluation data.
If the purpose of a study is to understand “why,” in addition to “how much,”
these verbal elaborations may even be essential. Chapters 9 and 10, where
we discuss subjectivist approaches to evaluation, indicate that these verbal
comments can become the data of primary interest to the investigator.
7. Address halo effects. There are two major ways to minimize halo
effects through item design. The first is to include, within a set of items
* It is generally known that humans can process about seven (plus or minus two)
items of disparate information at any one time. 27 The practical upper limit of seven
response options may be attributable to this feature of human cognition.