Chapter 16 Semantic Differential Scaling: A How To Do It Guide Semantic differential methods of scaling are a recently introduced method of attitude assessment. In 1967, Osgood, Suci and Tannenbaum introduced the method in their book "The Measurement of Meaning." Although the original purpose of semantic differential was not necessarily the assessment of attitudes, the procedure was well adapted for attitude assessment. A semantic scale is composed of polar opposite adjectives separated by a five to seven point rating scale, like this: Bad ___ ___ ___ ___ ___ ___ ___ Good To utilize the scale, the subject would be given an attitude referent. The attitude referent would be perhaps an object or event in the subject's environment, for example, the referent "mother." The subject's task would be to rate the referent "mother" on the seven point scale from bad to good. If the subject selected the middle space then it is advanced that the evaluation would be neutral. However, if the subject selected one of the spaces closer to the "good" end of the scale, then his/her evaluation was seen as a positive endorsement of the referent "mother". Conversely, selection closer to the "bad" end of the scale was seen as a negative endorsement. If you assigned numerical values of 1 through 7 to the various spaces on the scales, then a neutral score would become a 4, a very positive endorsement a 7, and a very negative endorsement a 1. To utilize these types of scales to measure an attitude it is proposed that a number of semantic scales be utilized. For example, we might assess a subject's attitude toward "mother" by using the following semantic scales: Bad Unfriendly Sad Cruel Dirty Foolish ___ ___ ___ ___ ___ ___ Please rate your Mother ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ Good Friendly Happy Kind Clean Wise The score on our attitude toward mother scale would be calculated by adding up the numerical values of the scales endorsed. The maximum score would be 42 and the minimum 6. To guard against the subjects developing some type of response bias (responding to all sales with the same ratings, leaving the impression that the subject may not have read the individual scales) the polarity of some of the scales could be reversed. Osgood, et al had their subject rate various attitudinal referents on a large number of semantic scales. These data were then analyzed by factor analytic procedures to determine possible common factors which could be identified in the collection of semantic scales. (Very simply, factor analysis identifies clusters of scales.) His analysis indicated that the majority of the semantic scales tended to fall into three principle factors. These principle factors were evaluation, potency, and activity. I have selected from the list of semantic scales presented in their book those scales which appear to load highest on each of the three factors. Semantic Scales Identifying the Factors of Evaluation, Potency, and Evaluation* Evaluation Potency Activity Good-Bad Hard-Soft Active-Passive Kind-Cruel Strong-Weak Fast-Slow Wise-Foolish Heavy-Light Difficult-Easy Beautiful-Ugly Masculine-Feminine Hot-Cold Happy-Sad Deep-Shallow Motivated-Aimless Candid-Deceitful Potent-Impotent Moving-Still Sociable-Unsociable Severe-Lenient Excitable-Calm Friendly-Unfriendly Domineering-Lax Alive-Dead Willing-Unwilling Brave-Cowardly Emotional-Unemotional Honest-Dishonest Large-Small Complex-Simple ________________________________________________________ * The left hand adjective identifies the positive end of the semantic scale. Using semantic scales from each of the factor groupings, a dimensional representation of an attitude is made possible. For example, a measure of the attitude referent can be made on the factors of evaluation, potency and activity. To demonstrate such an instrument, suppose that our attitude toward mother questionnaire was composed of the following semantic scales (notice that 3 scales were selected from each factor grouping from the listing above). Good Soft ___ ___ Please rate your mother ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ Active Cruel Strong Slow Wise Heavy Ease ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ Bad Hard ___ ___ ___ ___ ___ ___ ___ Passive Kind Weak Fast Foolish Light Difficult Notice that I have reversed the polarity of every other item, so that the first item would be scored from 7 to 1, the second item from 1 to 7, etc. Also notice that I have alternated the scales from the evaluation factor first, potency factor second, and activity factor third. This sequence of presentation is repeated through the questionnaire. Now, suppose that we administer our questionnaire to a subject and we obtain the following responses: Good _X_ Please rate your mother ___ ___ ___ ___ ___ Soft Active Cruel Strong Slow Wise Heavy Ease ___ ___ ___ ___ ___ _X_ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ _X_ ___ ___ _X_ _X_ ___ ___ ___ ___ ___ ___ ___ ___ ___ _X_ _X_ ___ _X_ ___ ___ ___ ___ ___ ___ ___ ___ _X_ ___ ___ ___ ___ ___ ___ Bad ___ ___ ___ ___ ___ ___ ___ ___ Hard Passive Kind Weak Fast Foolish Light Difficult To determine the subject's scores we will need to sum the values of the evaluation, potency, and activity scales as follows: Evaluation Value Good-Bad 7 Cruel-Kind 6 Wise-Foolish Total 7 20 Potency Soft-Hard Strong-Weak Heavy-Light Total 3 5 4 12 Activity Active-Passive Slow-Fast Easy-Difficult Total 5 3 4 12 It has been suggested that to obtain an attitude score only the evaluation factor scales should be considered. Using this suggestion, our subject would have a score of 20, out of a possible score of 21. (Minimum possible score of 3 and neutral score of 12.) This magnitude of score would indicate a very favorable attitude toward "mother". However, on the potency and activity factors our subject placed at the neutral point. In short, our subject rated the referent "mother" very high in evaluation, neutral (neither positive or negative) on the concepts of potency and activity. I have often seen researchers derive their attitude score by summing all of the semantic scales. In our example, the total score would have been 44, with the highest score being 63 and the lowest 9 (neutral point of 36). Using this procedure our subject would have a slightly positive attitude toward the referent "mother". I do not recommend the total scoring procedure and recommend that the data be scored to reveal the three factor sub-scores. I feel that the factor scoring procedures provides additional descriptive information with minimal additional computations. Critique of the Semantic Differential Scaling Process Our critique of the semantic differential scaling process will focus on 7 characteristics: zero point, equality of units, unidimensionality of items, usunidimensionalityof scale, reliability, content validity, and ease of application. In terms of a zero point, semantic differential methods provide a well defined zero. The zero point is a number of scales used times 4 (the neutral point of a given scale). In terms of our five point compliance index, ranging from 1 - no compliance to 5 - very good compliance, I would rate semantic differentials as a 4. The second criteria was equality of units. On this characteristic our semantic scaling process does not fare as well. Although there is some evidence that some of the scales contribute uniquely to the underlying factors, there is no conclusive proof that the contribution is equivalent across semantic scales. Therefore, on our five point compliance index I would have to give semantic differential scaling a 2. In terms of unidimensionality of items, the semantic differential scaling process performs satisfactorily. Each of the semantic scales represents a continuum anchored by polar opposite adjectives and appearing to assess one content. In terms of our compliance ratings, I would have to give semantic differential methods a 4. With respect to unidimensionality of scale, semantic differentials can pose some interesting problems. If the attitude score is developed by considering the three factor components of evaluation, potency and activity, then there is good unidimensionality. However, if a total attitude score is generated by combining the three factor components then we would have poor unidimensionality. Overall I rate semantic differentials as a 3 on our compliance index. Like Thurstone scaling, semantic differential processes does not include a built-in method of assessing reliability. Therefore, on our compliance scale I would have to give semantic differential scaling a 1 for reliability. The content validity of the semantic differential is not as rigorous as the previous methods of attitude scale development (Likert and Thurstone). This contention is not surprising since the original purpose of the semantic differential was not specifically attitude assessment (it has been adapted to attitude measurement). On our compliance ratings I would have to give semantic differentials a 3 in terms of compliance. Our final criteria for the evaluation of an attitude questionnaire was ease of application. This is the one area in which semantic differential surpasses all other attitude assessment procedures. I will assign a compliance rating of 5. Following is a summary of our compliance ratings. Compliance Ratings of Semantic Differentials Criteria Rating Zero Point Equality of Units Unidimensionality of Items Unidimensionality of Scale Reliability Content Validity Ease of application 4 2 4 3 1 3 5