on an 11-point scale? If you have more points than real differences,then you have too many points.
Conversely, you must be sure to include enough points to see those real differences.Consider an
extreme example of a two-point scale: agree or disagree.While this might work on a simple issue,
how many more examples can you think of where there is a grey area that is neither full agree-
ment nor full disagreement? In those cases you would need more points to get an accurate pic-
ture of the issues.One point to remember: you can always collapse categories together during data
analysis if you need to,but you cannot add extra points after the fact (after you already have col-
lected the data).As a result,some researchers err on the side of slightly more rather than less points
on a scale.On the other hand,Dillman (2007) reports that he has,over the years,encouraged the
use of fewer points (i.e., four or five points) because of their simplicity and their easier compre-
hension by respondents.We recommend starting with the commonly used (i.e.,“standard”) rat-
ing scales,such as the ones provided in Exhibit 7.1,and adjusting them only if needed.
An empirical factor that can affect the number of response categories needed is how your
respondents actually rate the attitudinal objects. For example, when pilot testing a question-
naire designed to measure researchers’ methodological beliefs, I (Burke Johnson) found that
the traditional four-point agreement scale (strongly disagree, disagree, agree, strongly agree)
was not working well. My participants sometimes complained that they didn’t fully agree; at
other times they complained that they didn’t fully disagree. Therefore, I shifted to a six-point
scale (strongly disagree, disagree, somewhat disagree, somewhat agree, agree,strongly agree)
to provide more choices.
You might also wonder whether you should include a center or middle category in your
rating scale.Research suggests that omitting the middle alternative (“neutral,”“about the same,”
“average,”“no difference,”etc.) does not appreciably affect the overall pattern of results (Converse
& Presser, 1986, pp. 36–37; Schuman & Presser, 1981; see Chapter 6). As a result, some
researchers choose to include a middle alternative, and others choose not to include it. Both
practices can be defended.You can see in Figure 7.1 that Rosenberg used four-point rating scales
(i.e., he omitted the middle alternative) in his Self-Esteem Scale. Some researchers, such as
Rosenberg,prefer to omit the middle alternative because it forces research participants to lean
one way or the other.
Take a moment now to examine Exhibit 7.1, and you will see some rating scales that
researchers and practitioners commonly use.You can use these in your questionnaires.Note that
the ordering of the categories does not appear to affect response patterns (Barnette,1999;Weng
& Cheng, 2000).For example, both of these patterns are commonly used and have been shown
to work: (a) strongly disagree,disagree,agree,strongly agree and (b) strongly agree,agree,dis-
agree,strongly disagree.Also note that both four-point and five-point rating scales are commonly
used by survey research experts.As you can see in Exhibit 7.1, you can construct rating scales
for many dimensions,such as agreement,approval,importance,and satisfaction.When you con-
struct your own rating scales,you will find additional dimensions that you are interested in,and
you will need to construct similar (i.e.,analogous) response categories for those dimensions.
Rankings
Sometimes you might want your research participants to rank order their responses. A
ranking indicates the importance or priority assigned by a participant to an attitudinal object.
Rankings can be used with open-ended and closed-ended questions. For example, you might
first ask an open-ended question such as, In your opinion, who are the three top teachers in
your school? Then you could follow up this question with a ranking item such as, Please rank
CHAPTER 7: HOW TO CONSTRUCT A QUESTIONNAIRE 181
RRaannkkiinngg
The ordering of
responses in
ascending or
descending order