Advances in Consumer Research Volume 2, 1975 Pages 507-514
EFFECTS OF POLARITY OF SEMANTIC DIFFERENTIAL SCALES IN CONSUMER RESEARCH
John Dickson, University of Connecticut
Gerald Albaum, University of Oregon
This paper examines the results of a laboratory type experiment designed to study the question of the effects of polarity of semantic differential scales. Polarity refers to the horizontal position of the positive (i.e., favorable) and negative adjective or phrase used for each bipolar scale in a semantic differential test. A total of 84 subjects rated two concepts across a set of 20 bipolar scales, which were phrases. Each subject randomly received one of three treatment variations of a semantic differential measurement instrument.
This study provides evidence which indicates that the polarity of a scale does not significantly affect the results. Since it makes little difference how groups of scales are presented, by placing all positive or negative phrases on the same side, coding, editing, and processing errors and costs should be reduced.
The use of various opinion and attitude measuring devices by marketing researchers has become increasingly popular in recent years. Of particular interest to researchers are those devices which permit a comparison of attitudes about product images, advertisements and other marketing concepts. One of the more widely used devices for collecting such information is the semantic differential, developed by Osgood and his associates (1957). Other scaling techniques such as multidimensional scaling, the Likert scale, and the Stapel scale (Hawkins, 1973) have also been used for similar purposes.
Yet, the increasing popularity of utilizing the various scaling models for measuring attitudes and opinions has led several researchers to explore more thoroughly certain methodological questions. Some of the current issues being raised include the appropriate number of points to use in presenting a scale (Green and Rao, 1970; Jacoby and Matell, 1971; Lehman and Hulbert, 1972), inclusion of "ideal" ratings (Landon, 1971), the question of forced-choice (Albaum and Munsinger, 1973; Hughes, 1969), concept-scale interaction (Sharpe and Anderson, 1972), the way in which scales are administered (Albaum and Munsinger, 1973; Hawkins, 1973), reversing the order of presentation of scales (Belson, 1966; Payne, 1972), and polarity of statements (Falthzik and Jolson, 1974). Although the published research concerning the above questions covers a variety of types of scaling techniques, all issues are directly applicable to the semantic differential.
The purpose of this paper is to present the findings of a study concerned with one specific methodological question--the polarity of a semantic differential scale. By polarity we mean the horizontal position of the positive (i.e., favorable) and negative adjective or phrase used for each bipolar scale in a semantic differential test.
Considering reading patterns of people in the United States and similar cultures, for any given set of semantic differential bipolar scales three alternative arrangements concerning polarity are possible:
1. Positive: all scales have the "positive" adjective or phrase on the left-hand side of the scale.
2. Negative: all scales have the "negative" adjective or phrase on the left-hand side of the scale.
3. Mixed: some of the scales have the "positive" adjective or phrase on the left-hand side while others have the "negative adjective or phrase on the left-hand side.
Since it is believed that the formation of position preferences may create unwanted biases, most researchers imply or state that scales representing the same underlying factor should be randomly alternated in direction of polarity (Tull and Albaum, 1973; Osgood et al., 1957). Typically, published research describes which and how many scales were reversed in polarity, thus implicitly agreeing with the above assumption. However, a thorough search of the literature yielded no information or empirical research which substantiates this assumption. Consequently, a laboratory-type experiment was conducted to test the following hypotheses:
Hypothesis I: There is no significant difference in the results obtained from a semantic differential measurement instrument in which the polarity of the scales is "positive" and one in which the polarity of scales is mixed.
Hypothesis II: There is no significant difference in the results obtained from a semantic differential in which polarity is "negative" and one in which polarity is mixed.
Hypothesis III: "Positive" and "negative" semantic differential instruments yield the results which are not significant]y different.
The treatment variable was a semantic differential measurement instrument consisting of twenty seven-point bipolar scales. Subjects were asked to rate two concepts, which were department stores. If the concepts yield similar data, increased confidence in the findings would result. In addition, two concepts of the same general type can serve as a test for the existence of an unusual concept-subject interaction.
Nine of the scales used were selected from among those previously tested and recommended by Kelly and Stephenson (1967), while eleven scales were selected from an earlier unpublished study conducted by one of the authors in which similar type concepts were analyzed. Phrases, rather than adjectives, were used to increase the relevance of the scales to the retailing context. This modification of the usual semantic differential model was suggested by Mindak (1961) for situations where concepts may provide an environment in which concept-scale interaction may occur due to the lack of relevance to subjects of the adjectives developed by Osgood et al. (1957).
Three treatment measurement instruments were administered:
1. mixed arrangement of polarity in which one-half the scales had the positive phrase, and one-half the negative, on the left side. Assignment of scales to these two groups was done on a random basis.
2. positive arrangement of scale polarity.
3. negative arrangement of scale polarity.
The vertical ordering of each scale used in the treatments was assigned randomly and was the same for all treatments. Also, for both concepts, the ordering of the bipolar scales was held constant. The order of presentation of the concepts was such that one-half the subjects rated Store A first while the other subjects rated Store B first.
Upper division undergraduate business students attending the University of Oregon were used as subjects. The concepts used were selected to insure that all of the subjects would be familiar with them. This was verified during the test. Since the function of this sample of students was to provide answers to a methodological question, there is no reason to believe such a sample would create bias problems. The treatments were assigned randomly to students in two classes. The results for each class were compared for significant differences in response distributions using the chi-square test. In addition, the Mann-Whitney U test was used to test for significant differences in location (i.e., central tendency). No significant difference was observed in terms of distribution or direction of responses so the two groups were combined for data analysis. The combined sample consisted of 84 subjects, distributed among the treatments as follows:
Treatment Number of Subjects
1 (mixed) 28
2 (positive) 27
3 (negative) 29
Precautions were taken to ensure that students enrolled in both classes participated only once.
RESULTS AND DISCUSSION
Concepts were analyzed separately in two different ways. First, each scale was treated independent of the others and treatment comparisons made as follows: (1) mixed vs. positive polarity, (2) mixed vs. negative, (3) positive vs. negative. The Mann-Whitney U test was used for the statistical analysis of these paired data. In some situations, researchers may be interested in deriving summated scale scores for each individual and then comparing the scores for one group of individuals to a second group (Osgood, et al., 1957). A second form of analysis was conducted to determine whether significant differences exist among the summated scores of individuals subjected to the three treatments. Each subject was assigned a score for each concept according to the sum of his responses on all scales. To test if there was a significant difference among treatments, the Kruskal-Wallis one-way analysis of variance test was applied to the mean summated scores for the groups of subjects exposed to the three treatments.
Individual Scale Analysis
The results of the paired-treatment comparisons for each scale are shown in Tables 1 and 2, which also present treatment mean values (scales were scored 1 to 7, positive to negative). The significance test results are summarized in Table 3, which indicates that in all situations except the positive-negative comparison for Store A the number of significant scales (a = .10) are less than the maximum number that would be expected to be significant by chance alone, on the basis of the binomial distribution. For example, the probability of finding 3 significant scales at a = .10 out of 20 possible by chance alone is .32. These data indicate that Hypotheses I and II cannot be rejected.
The results relevant to Hypothesis III are mixed. That is, for Store A the number of significant scales (a = .10) for the positive-negative treatment comparison exceeds chance expectations whereas for Store B the number can be attributed to chance. Consequently this analysis cannot be used for reaching a definite conclusion concerning Hypothesis III.
Summated Score Analysis
As indicated above, the responses for each subject were summed over all 20 scales. The mean values of summated scores for each treatment are presented in Table 4. These data were analyzed by the Kruskal-Wallis one-way analysis of variance to test for significant differences in the summed values. As shown, the probability at which significance occurs exceeds .10 for both concepts. Consequently, a test of the three hypotheses with a = .10 as the critical value would show that all three cannot be rejected.
The data in this study indicate that there are no statistically significant differences in results between a semantic differential in which all positive and all negative phrases are placed on one side and the more typical situation where each side has one-half positive and one-half negative. Both the individual scale and summated score analyses support this contention. Unfortunately, the results are inconclusive for indicating which is better, placing all the positive phrases on the left side or all the negative phrases there.
Our findings do suggest some rather pragmatic values to researchers. Since this study showed that it makes little difference how a group of scales are presented, by placing all positive and all negative phrases on the same side in a semantic differential, coding, editing, and processing errors and costs should be reduced. In addition, preparation of the instrument in the first place may be facilitated. Moreover, since many adjectives and phrases are difficult to classify as positive or negative, if polarity is not an issue then there is one less matter for the researcher to be concerned about.
As in most exploratory-type research projects,various limitations to our findings are imposed as a result of methodological problems as well as the inherent nature of the test design. First and foremost, since students were used as respondents, the findings cannot really be extrapolated beyond the student population. In essence, the convenience sample of students (randomly assigned to treatments, however) results in the test design being a fixed effects model.
TREATMENT MEAN VALUES AND PAIRED-TREATMENT COMPARISONS FOR STORE A, BY INDIVIDUAL SCALE
TREATMENT MEAN VALUES AND PAIRED-TREATMENT COMPARISONS FOR STORE B, BY INDIVIDUAL SCALE
NUMBER OF SIGNIFICANT SCALES AT A = .10, BY CONCEPT
MEAN VALUES OF SUMMATED SCORES
Secondly, no pretest was conducted using the specific questionnaire which was used in this test. Most of the bipolar phrases were selected on the basis of recommendations by other authors. However, one of the present authors has used all the scales in a previous test, although the scales were arranged in a different order.
Finally, as in the case of all semantic differentials where there are more than one scale per concept, there may be a lack of independence between scales, and there may exist a concept-scale interaction effect. However, we believe such confounding effects did not appear in this study.
Albaum, G. & Munsinger, G. Methodological questions concerning the use of the semantic differential. Unpublished paper presented at the Southwestern Social Science Association meeting, March, 1973.
Belson, W. A. The effects of reversing the presentation order of verbal rating scales. Journal of Advertising Research, 1966, 6, 30-37.
Falthzik, A. M. & Jolson, M. A. Statement polarity in attitude studies. Journal of Marketing Research, 1974, 11, 102-105.
Green, P. E. & Rao, V. R. How many rating scales and response categories to use. Journal of Marketing, 1970, 34, 33-39.
Hawkins, D. I. Utilization of the Stapel Scale in attitude measurement. Unpublished paper presented at the Southwestern Social Science Association meeting, March, 1973.
Hughes, G. D. Some confounding effects of forced choice scales. Journal of Marketing Research, 1969, 6, 223-226.
Jacoby, J. & Matell, M. S. Three-Point Likert Scales are good enough. Journal of Marketing Research, 1971, 8, 495-500.
Kelly, R. F. & Stephenson, R. The semantic differential: an information source for designing retail patronage appeals. Journal of Marketing, 1967, 31, 43-47.
Landon, E. L. Order bias, the ideal rating, and the semantic differential. Journal of Marketing Research, 1971, 8, 375-378.
Lehmann, D. R. & Hulbert, J. Are three-point scales always good enough? Journal of Marketing Research, 1972, 9, 444-446.
Mindak, W. A. Fitting the semantic differential to the marketing problem. Journal of Marketing, 1961, 25, 28-33.
Osgood, C. E., Suci, G. J. & Tannebaum, P. H. The measurement of meaning. Urbana, Ill.: University of Illinois Press, 1957.
Payne, J. D. The effects of reversing the order of verbal rating scales in a postal survey. Journal of the Marketing Research Society, 1972, 14, 30-44.
Seigel, S. Nonparametric statistics- New York: McGraw-Hill, 1956.
Sharpe, L. I. & Anderson, W. Concept-scale interaction in-the semantic differential. Journal of Marketing Research, 1972, 9, 432-434.
Tull, D. S. & Albaum, G. Survey research: A decisional approach. New York: Intext Educational Publishers, 1973.