Steps in Building a Likert Scale

Likert’s method was to take a long list of possible scaling items for a concept and find the subsets that measured the various dimensions. If the concept were unidimensional, then one subset would do. If it were multidimensional, then several subsets would be needed. Here are the steps in building and testing a Likert scale.

1. Identify and label the variable you want to measure. This is generally done by induction—that is, from your own experience (Spector 1992:13). After you work in some area of research for a while, you’ll develop some ideas about the variables you want to measure. The people you talk to in focus groups, for example, may impress you with the idea that ‘‘people are afraid of crime around here,’’ and you decide to scale people on the variable ‘‘fear of crime.’’

Or you may observe that some people seem to have a black belt in shopping, and others would rather have root canal surgery than set foot in a mall. The task is then to scale (measure) people on a variable you might call ‘‘shopping orientation,” with all its multidimensionality. You may need a subscale for ‘‘shopping while on vacation,’’ another for ‘‘car shopping,’’ and another for ‘‘shopping for clothing that I really need.’’ (The other way to identify variables is by deduction; see box 1.1).

2. Write a long list of indicator questions or statements. This is usually another exercise in induction. Ideas for the indicators can come from reading the literature on whatever research problem has captured you, from personal experience, from ethnography, from reading newspapers, from interviews with experts.

Free lists are a particularly good way to get at indicators for some variables. If you want to build a scaling device for the concept of ‘‘attitudes toward growing old,’’ you could start by asking a large group of people to ‘‘list things that you associate with growing old’’ and then you could build the questions or statements in a Likert scale around the items in the list.

Be sure to use both negative and positive indicators. If you have a statement like ‘‘Life in Xakalornga has improved since the missionaries came,’’ then you need a negatively worded statement for balance like ‘‘The missionaries have caused a lot of problems in our community.’’ People who agree with positive statements about missionaries should disagree with negative ones.

And don’t make the indicator items extreme. Here’s a badly worded item: ‘‘The coming of the missionaries is the most terrible thing that has ever happened here.’’ Let people tell you where they stand by giving them a range of response choices (strongly agree- strongly disagree). Don’t bludgeon people with such strongly worded scale items that they feel forced to reduce the strength of their response.

In wording items, all the cautions from chapter 9 on questionnaire design apply: Remember who your respondents are and use their language. Make the items as short and as uncomplicated as possible. No double negatives. No double-barreled items. Here is a terrible item:

On a scale of 1 to 5, how much do you agree or disagree with the following statement: ‘‘Everyone should speak Hindi and give up their tribal language.’’

People can agree or disagree with both parts of this statement, or agree with one part and disagree with the other.

When you get through, you should have four or five times the number of items as you think you’ll need in your final scale. If you want a scale of, say, six items, use 25 or 30 items in the first test (DeVellis 2003:66). [1]

There is no best format. But if you ever want to combine responses into just two categories (yes-no, agree-disagree, like me-not like me), then it’s better to have an even number of choices. Otherwise, you have to decide whether the neutral responses get collapsed with the positive answers or the negative answers—or thrown out as missing data.

  • 4. Test your item pool on some respondents. Ideally, you need at least 100—or even 200—respondents to test an initial pool of items (Spector 1992:29). This will ensure that: (1) You capture the full variation in responses to all your items; and (2) The response variability represents the variability in the general population to which you eventually want to apply your scale.
  • 5. Conduct an item analysis to find the items that form a unidimensional scale of the variable you’re trying to measure. More on item analysis coming up next.
  • 6. Use your scale in your study and run the item analysis again to make sure that the scale is holding up. If the scale does hold up, then look for relations between the scale scores and the scores of other variables for persons in your study.

  • [1] Determine the type and number of response categories. Some popular response categories are agree-disagree, favor-oppose, helpful-not helpful, many-none, like me-not likeme, true-untrue, suitable-unsuitable, always-never, and so on. Most Likert scale itemshave an odd number of response choices: three, five, or seven. The idea is to givepeople a range of choices that includes a midpoint. The midpoint usually carries theidea of neutrality—neither agree nor disagree, for example. An even number ofresponse choices forces informants to ‘‘take a stand’’; an odd number of choices letsinformants ‘‘sit on the fence.’’
< Prev   CONTENTS   Source   Next >