The effect of exposure to sexual appeals in advertisements on memory, attitude, and purchase intention: a meta-analytic review

ABSTRACT

Researchers have often attempted to answer the question, ‘Does sex sell?’ In this article, we present a meta-analysis of studies that used an experiment to test the effect of sexual appeals in ads on memory, attitude, and purchase intention. Our analysis revealed a significant positive effect for sexual appeals on ad recognition and recall (weighted Cohen's d = .38, p < .001), but the effect on brand recognition and recall was not significant (d = .09, p = .30). We also found that the effect of sexual appeals on attitude towards the ad was not significant (d = −0.07, p = .26); however, additional analysis showed that males (d = .27, p < .01) evaluate ads with sexual appeals significantly more positively than females (d = −.38, p < .001). Finally, we found a small significant negative effect on brand attitude (d = −.22, p < .05), but no effect on purchase intention (d = .01, p = .94).

KEYWORDS:

• Sex in advertising
• meta-analysis
• sexual appeal
• ad recognition and recall
• brand recognition and recall
• attitude toward ad
• attitude toward brand
• purchase intention

Acknowledgements

The authors would like to thank the three anonymous reviewers for their constructive feedback on the original manuscript and revised versions of the manuscript. The first author would also like to thank Daniel O'Keefe for comments on a previous version of the manuscript and Michael Borenstein for his good humour and advice on addressing significant heterogeneity of effects. The length of time between the initial submission and the submission of the final revision was due to personal circumstances rather than problems with the research project itself.

Disclosure statement

There are no potential conflicts of interest.

Notes

1. Lull and Bushman (2015) recently published a meta-analysis that examines the effects of exposure to violent and sexual media content on individuals, which included some of the same variables in this meta-analysis. While there is some overlap, our meta-analysis focuses more specifically on advertising research and outcomes specific to advertising, such as ad recall and recognition and attitude towards the ad.

2. For simplicity sake, we will use the term ‘control condition’ throughout the rest of the article to refer to the non-treatment sexual appeal condition. Some authors use the term ‘comparison condition’, and others refer to an ad with a pastoral scene as a control ad and an ad with fully clothed models in non-sexual behaviours as a comparison condition.

3. A number of studies have tested whether TV programme–sexual appeal congruity or incongruity influences ad and brand recall and recognition (e.g. Bushman 2007; Furnham and Mainaud 2011; Parker and Furnham 2007). While conceptually the same, these studies have generally not tested product–appeal congruity, which is the focus of our moderator test. See Lull and Bushman (2015) for a review and discussion of the programme–appeal congruity research.

4. Each website differed in the years that papers and abstracts were available online. For the AEJMC website, the available years were 1997–2016. For the ACR website, available years were 1974–2016. For the AAA website, the years checked were 2007–2016. We also conducted a hand search of printed abstracts from AAA conferences between 1979 and 2006. A majority of the abstracts and conference papers that we located had been published subsequently in a journal or book chapter. Additional information about this part of the search is available from the first author.

5. A complete list of the publications providing data for the meta-analysis appears in the References section of the paper. Studies in the meta-analysis are preceded by *.

6. We defined a publication as a published or unpublished manuscript reporting the results of one or more sex in advertising studies. We defined a study as an experiment using a unique sample and method and reported as such by the authors (e.g. Putrevu and Swimberghek 2013; Dahl et al. 2009). When authors used the same method and stimulus materials to collect data from multiple samples, we counted this as a single study (e.g. Liu et al. 2009; Yan et al. 2010).

7. The issue of whether or not to include data from the “sexual programme” conditions in the TV programme–appeal congruity studies was one that we considered carefully. Bushman (2007) hypothesized a negative effect on memory for ads featured in programmes with sexual themes and content. Our position is that by including only data from the low-sex programmes, we provide a “cleaner” test of the direct effect of sexual appeals on memory. However, in light of the findings presented in Lull and Bushman (2015), we wanted to eliminate the possibility that this decision would cause us to overstate or overestimate any significant effects. Therefore, we conducted a second set of analyses with data from the sexual and non-sexual TV programme conditions. Those results are summarized in Note 14.

8. We used univariate F values only when there were two levels of sexual appeal (e.g. appeal type: sexual, non-sexual). This allowed us to compare treatment and control conditions directly rather than collapsing three or more levels of sexual appeals into one effect size (e.g. level of nudity: demure, partial nudity, full nudity). This approach was also consistent with the way we collected other data.

9. Products were coded as congruent if they fit in any of the following categories: (a) cosmetics or makeup; (b) perfume or body fragrance; (c) oils and lotions (e.g. suntan lotion, body, massage oil); (d) undergarments (e.g. bras, underwear); (e) luxury clothing and accessories (e.g. designer jeans, jewellery); (f) alcohol or liquor (e.g. beer, whiskey); (g) cigarettes; or (h) products or brands associated with fashion or sex (e.g. American Apparel, Details, Vogue, Playboy). All other products were coded as not congruent.

10. The Krippendorf's alpha statistics for the coding for descriptive variables were last name first author 1.0; publication type 1.0; publication year 1.0; studies per publication 1.0; sample size 1.0; sample type 1.0; per cent of female participants .91; levels of sex appeal .87; products appearing in ad .73; model/s in treatment ad 1.0; type of control/comparison .84.

11. The Krippendorf's alpha statistics for the data for the main effects and moderator analyses for the first 24 studies were number of relevant outcome variables .92; separate data for male and females 1.0; means and standard deviations (combined across dependent variables) .96; other data used to calculate Cohen's d (combined) .76; product–appeal congruity .88; ad memory (recognition, aided recall, unaided recall) .52; brand memory (recognition, aided recall, unaided recall) .44.

12. Violating the assumption of independent effects can produce inflated estimates of population effect sizes and biased results of tests for heterogeneity of effects, which can occur when data provided by the same participants on two or more dependent variables are combined (Hedges and Olkin 1985 Scammacca, Roberts, and Stubing 2014). To avoid introducing statistical dependence into the data, we used the ‘shifting unit of analysis’ approach recommended by Cooper (2010), which entails conducting separate analyses for each dependent variable. In situations where studies reported data on two memory outcomes (e.g. ad recognition, ad recall; Samson 2016), only the data from ad or brand recall was included in the combined test, and then the ad or brand recognition data were included in the subgroup analysis.

13. Moderator analyses for participant gender, product–appeal congruity, and model type were conducted using only the data that were available within each subgroup. A second approach is to include all data and to treat any data not fitting in a category as an additional category (e.g. female participants, male participants, other). We chose the first approach, because it fits the assumptions of a mixed-effects model and because it provides a direct test between each subgroup of a moderator variable. See Pigott (2012) for additional discussion on this topic.

14. The weighted mean Cohen's d for ad recognition and recall was calculated after adding data from the sexual programme conditions from the relevant studies. There were k = 23 effect sizes, and the result indicated a positive significant result for sexual appeals, d = .37 (95% CI [.15, .58]; z = 3.34.; p < .001; I² = 81.83). The weighted mean Cohen's d for brand recognition and recall was calculated after adding data from the sexual programme conditions from the relevant studies. There were k = 33 effect sizes, and the result indicated no significant effect for sexual appeals, d = .16 (95% CI [−.02, .33]; z = 1.78.; p = .08; I² = 84.02). Additional details about this analysis can be obtained from the first author.