Main Content Area
Don't Just Measure Half of an Attitude
The semantic-differential is among the two most popular forms of scales used in marketing research, at least, that is true in published academic research. (The other popular form is the Likert-type scale.) A person does not have to look any further to see how popular that format is than noting its use in measuring brand attitude and attitude toward the ad. Those are the two most measured constructs in published marketing studies. Although they have been measured several ways, the overwhelmingly most popular form to use is the semantic-differential. Despite its popularity with scholars and practitioners, a typical problem occurs in its use. The semantic differential should be phrased such that the opposing verbal anchors (adjective pairs) truly represent the extreme opposites of the semantic continuum. Yet, that rule is routinely broken and only a half of a semantic range is measured.
Part of the problem lies with the English language; there may not be a good way to express the complete opposite of some words. For example, if we were examining Attitude Toward the Ad, some facets we might want to specifically examine are how irritating an ad is and how entertaining it is. “Irritating” and “entertaining” would be great poles for one end of their respective scales, but what would be the semantic opposites? You could try to find words that are opposites of these terms but you may conclude there are none that do it well.
That is, indeed, a problem with the English language but what can make it worse is to use simplistic techniques for creating pseudo-opposites. Here are examples of reasonably good semantic differentials: good/bad, happy/sad, interesting/boring. But, as noted above, when a good polar opposite can not be found there is a tendency to use the simple form "X/not X" as if “not X” is the semantic opposite of "X." But, is that true? For example, is “not interesting” the same as “boring?” Let's examine it this way, on a seven point scale with “interesting” at one extreme and “not interesting” at the other, what would be the meaning of the mid-point? Wouldn't it mean “somewhat interesting?” This would appear to violate the assumption of the semantic differential that the midpoint of the scale “is clearly meant to be used when the rater associates the concept with neither pole of the adjective pair” (Dawes and Smith 1985, p. 534). Further, on a scale where “interesting” is scored as a 7 and “boring” is scored a 1, then “somewhat interesting” is probably a 4 or 5. However, in a scale where “not interesting” is the anchor receiving the score of 1 then the “somewhat interesting” anchor could be scored as 1, 2, or 3. The point is that “not interesting” does not mean the same as “boring.” Thus, if your goal is to determine how intellectually stimulating something is and you use this one item or a scale made up of items similar to it in form, it is doubtful that your measure is capturing the full range of the construct.
If just one item out of many is of this form (X/not X) in a summated scale then it probably won't cause a problem, especially if the items as a set are unidimensional and have high internal consistency. On the other hand, if many or all of the items are of this form, even if they are unidimensional and have strong internal consistency, the scales are only measuring a portion of a construct rather than the full construct. Admittedly, the practical implications of this issue have yet to be examined in detail.
Dawes, Robyn M. and Tom L. Smith (1985), "Attitude and Opinion Measurement," in Handbook of Social Psychology, 3rd., Vol. 1, Gardner Lindzey and Elliot Aronson, eds., New York: Random House, 509-566..