Persuasive Advertising - J. Scott Armstrong [9]
Quasi-experimental data
I use the term “quasi-experimental” to represent an approach where the research design controls for some, but not all variables, in comparisons of different ways of advertising. For example, a quasi-experimental study might compare two ads for the same product, brand, and media. The disadvantage is that some factors are not held constant. Nevertheless, the quasi-experimental approach can be valuable to the study of persuasive advertising even though, on occasion, an uncontrolled variable might impinge on validity.
Our quasi-experimental analysis drew upon data on 240 pairs of print ads from the Which Ad Pulled Best series (Burton and Purvis 1987…). Those books provide full-page magazine ads that had appeared from the 1980s up through 2002. I refer to them as “WAPB analyses.” The ads had been tested for recall, and in some cases for persuasion, by the advertising research firm, Gallup & Robinson. Our coding of these ads was done without any knowledge of the effectiveness of the ads. The approach is briefly described in the Glossary and a more detailed discussion is provided in Appendix B. Such analyses are reported for 58 principles in this book.
We searched for pairs where one of the ads follows a principle and the other ad does not. A typical pair of advertisements is shown opposite.1 For example, to test the principle on whether the headlines mention the brand, we compared the recall score for the ads that followed the principle (ad B in this case) with the score from those that did not (ad A). Note that while the ads are for the same brand, there are many differences, such as ad A having poorer contrast between the text and the background than ad B, and ad B also having more specific benefits. We then compared the scores for all ads that followed the given principle with those that did not. In this case ad B’s recall was much better than that for ad A, 29 percent to 19 percent. By averaging across all pairs of ads that differ on this principle, we can assess the effect of putting the brand name in the headline.
The findings from the quasi-experimental analyses are valid. We were astonished to find that the directional effects of quasi-experimental findings in Persuasive Advertising agreed with those from the experimental evidence that was available. This applied to all of the 26 principles for which we also had lab experiments and all seven principles for which we also had field experiments (Armstrong and Patnaik 2009).
Given this high level of validity, we gained knowledge on 58 of the advertising principles. We also have confidence in the principles for which the quasi-experimental analyses provided the only experimental evidence.
These analyses also contributed to the formulation of the principles. Here is an example. Experts suggested that headlines should be short. But our analysis of quasi-experimental data from WAPB indicated that the number of words made little difference. Which viewpoint was correct? Sandeep Patnaik and I reexamined quasi-experimental data on tested print ads. We believed that a key condition had been overlooked: namely, whether the ad is for a high-involvement product (where the customers give much thought) or for a low-involvement product. By comparing ads separately for each group, we found that ads for low-involvement products were more effective with short headlines than long ones, while ads for high-involvement products were more effective with longer headlines.
Sample ad pair from Which Ad Pulled Best
Experimental data
In experimental studies, the values of all key causal variables are controlled. For example, an advertiser (or researcher) who wants to know whether humor is persuasive