Page 214 - The Handbook of Persuasion and Social Marketing

206 The Handbook of Persuasion and Social Marketing

price increases reduce demand and price decreases increase demand.
Advertising elasticities are typically positive, such that changes in demand are
in the same direction as changes in advertising. Exceptions are possible—for
example, advertising volume could reach the point of annoyance and reduce
demand—but the results reported in the tables are all in the usual direction.
Standardized mean differences, often symbolized d, summarize differ-
ences between groups in terms of the variability within groups. They thus
indicate the magnitude of group differences independent of measurement
scales. Cohen (1992) suggests that d values of .2, .5, and .8 represent
small, medium, and large effects, respectively. Roughly 95% of a variable’s
observations are within ±2 standard deviations of its mean, so that a me-
dium standardized difference between groups is equal to about one-eighth
of the variability within groups.
The odds ratio OR is the proportion of observations that meet a crite-
rion in one group, such as behavior change given an intervention (e.g.,
quitting smoking after receiving counseling), relative to another group
(e.g., quitting smoking after not receiving counseling). Odds ratios are
widely used in research areas involving discrete outcomes, such as con-
tracting or succumbing to a disease. Nevertheless, Fleiss and Berlin (2009,
p. 244) suggest that “the meaning of the odds ratio is not intuitively clear.”
Considering the specific proportions involved may be useful. For example,
Sussman, Sun, and Dent (2006; see Table 8.1) report an odds ratio of 1.46
for teen smoking cessation following an intervention. The percentages in-
volved are 9.14% versus 6.24%, an absolute difference of less than three
percentage points. The costs and benefits of the intervention may be
clearer in terms of the absolute results than in terms of the odds ratios.
Elasticities are interpreted in terms of percentage changes and are diffi-
cult to compare with the other measures of effect sizes. However, results
given in terms of correlations, standardized mean differences, and odds
ratios are interchangeable (Borenstein, 2009). Specifically, ln(OR)/1.81 = d,
2
2
where ln indicates the natural logarithm, d = 2r/√(1 – r ), and r = d/√(d +
4). Thus, the value of OR = 1.46 in Sussman, Sun, and Dent (2006) is
equivalent to d = .21 and r = .10, a small effect according to Cohen’s (1992)
guidelines. This small effect may be important nevertheless, given the po-
tential health benefits of not smoking.

Results for Empirical Generalizations

Given the breadth of this review, obtaining a complete set of relevant meta-
analytic studies is not a realistic objective. Just the topic of intervention ef-
fects, for example, is huge; Google Scholar reports more than 16,000 results