Explore - Experience - Excel

Attitude Measurement – Reliability and Validity: Review of article by David Trafimow – Dr. S Shyam Prasad

20 April 2024

Note to the readers: This article is meant for beginners/early researchers or students. During my teaching, I noted that many have a perception that measuring attitude is a routine affair and one need not bother much about it. To remove this myth that attitude measurement is easy, I picked an article titled ‘Attitude Measurement’ by David Trafimow, published in the Encyclopedia of Applied Psychology, and summarized in a manner suitable for beginners.


To predict and influence people’s behaviors, researchers strive to measure the attitudes of the people.  However, before measuring attitude, there are several questions the researcher should clarify. They are

  1. Reliability and Validity of measurement,
  2. Whether attitude toward an object or behaviour
  3. Attitude toward a single behaviour or repeated behaviour and
  4. Can this variable (attitude) be distinguished from other variables?

Reliability and Validity

Reliability According to Trafimow, reliability can be measured in one of two ways. The first method is by doing a test-retest reliability. This is done by asking a set of respondents to complete an attitude measurement and then repeat the same process at a later date. The correlation between the two tests is the attitude measure’s test-retest reliability. The second one he says can be termed as internal consistency. If one uses several scales to measure an attitude, then all these scales should correlate with each other. He says that the scales’ correlation with each other can be termed internal consistency. Both the above reliability tests are essential before considering validity.

 Validity Valid for what? Very succinctly Trafimow says that a measure is valid “if it predicts another variable that is considered desirable to predict such as behavior and behavioral intentions.” For example, a measure of helmet-use attitudes that actually predicts their intention to use helmets or actual use of helmets can be said to have predictive validity. He also cautions that the measure should measure attitude and not something else. According to him, the measure may predict behavior, not because it measures attitude but rather because it measures something else. For example, one may prefer to use a helmet not because one has a positive attitude toward helmet usage, but due to law enforcement. For the above reason, valid measurement of attitudes depends on the theories in which attitudes play an important role. Hence, when the attitude construct is supported by a theory and the attitude measure is part of the theory, only then it is said to have construct validity.

The Problem

               For long, researchers have assumed that attitudes predicted behaviors. In contrast, many studies between 1920 and 1960 failed to predict behaviors from the attitudes. This could happen for two reasons; one, after all, attitudes do not matter in predicting behaviors, or two, attitudes do matter but attitude measures used were not valid. Hence, researchers took up the challenge and eventually, they ended up with attitude measures with greater construct and predictive validity.

Attitude Accessibility It is interesting to note that, indirectly supporting the first view stated above, Fazio believes that attitudes and behaviors are correlated only when attitudes are easy to retrieve from memory. Fazio argues that when attitudes are not accessible, there is no reason to believe that they would affect behavior. However, for unknown reasons, applied researchers have ignored the accessibility view. One of Trafimow’s conjectures is that it is not clear how the accessibility view can be applied to predict and control behaviors.

Principle of Correspondence A solution to the low attitude-behavior correlations was proposed by Fishbein, wherein behaviors have four components: action, target, time, and context.  Let us consider an example. Suppose a researcher wishes to predict whether people will give blood at the campus blood drive on Tuesday. In this example, the action is “give,” the target is “blood,” the time is “on Tuesday,” and the context is “at the campus blood drive.” If the researcher makes the following scale:

I, extremely like/quite like/slightly like/neutral/slightly dislike/quite dislike/extremely dislike giving blood,

then the results will be invalid. This is because there is no reason to expect that the above attitude measure would correctly predict whether people will “give blood at the campus blood drive on Tuesday.” The researcher should ensure that the attitude measure should correspond with the behavior measure regarding action, target, time, and context. Researchers, to predict correctly, should not measure “giving blood,” but should measure “giving blood at the campus blood drive on Tuesday.” This can be achieved by the following scale”

I, extremely like/quite like/slightly like/neutral/slightly dislike/quite dislike/extremely dislike giving blood at the campus blood drive on Tuesday.

Presently, the researchers apply the principle of correspondence to measure attitude and the results have dramatically improved in predicting behavior in different areas.

Factor analysis Earlier, we discussed that to increase the internal consistency of the attitude measure, one needs to increase the number of scales. Hence, many researchers use several scales to measure. For example, in addition to the like-dislike example given above, participants could respond to scales that include pairs such as wise-foolish, beneficial–harmful, enjoyable–not enjoyable, good–bad, and pleasant–unpleasant. Researchers take the mean of the multiple scales used to represent the attitude. But, here lies a danger. One, not all scales may be equally good in measuring the attitude. Further, some of them may even measure something other than attitude. To overcome this issue, researchers habitually conduct factor analysis of the scales. In a perfect condition, when all the scales are submitted to a factor analysis, one factor that represents attitude should emerge. Most of the time only one factor may result but if more than one factor is obtained, it falls on the researcher to identify the “true” attitude factor.

This brings us to the end of part 1. Let us take a break. In part 2, we will look into more factors in attitude measurement such as attitudes toward objects or behaviours, Open attitude measures, implicit attitude measures, direct and indirect attitudes, and a few other relevant issues. 

Assignment question

After reading the above article, students can discuss when attitudes are not good predictors of behaviors and can think of situations where attitudes may not matter. If one has a positive attitude toward a brand – say Audi – will it result in buying behavior? When will attitude result in buying behavior and when will it be otherwise?

Leave a Reply

Your email address will not be published. Required fields are marked *