Sie sind auf Seite 1von 6

Journal of Business Research 65 (2012) 874–879

Contents lists available at ScienceDirect

Journal of Business Research

Print advertising: Celebrity presenters


John R. Rossiter a,b,⁎, Ale Smidts c,⁎
a
University of Wollongong, Australia
b
Bergische University of Wuppertal, Germany
c
Erasmus University, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: This study validates Rossiter and Percy's (1987) hook theory of presenter characteristics, for celebrity
Received 1 January 2010 presenters. Firstly, by employing a product-alone control group, the study demonstrates that some celebrity-
Received in revised form 1 March 2010 product pairings have a good fit and can persuade whereas others have no effect or represent such an
Accepted 1 April 2010
obviously poor fit that they dissuade consumers from buying the product. Secondly, the study suggests that
Available online 26 January 2011
good fit, and thus persuasion, for celebrity presenters, depends on the audience immediately perceiving that
Keywords:
the celebrity is an expert user of the product (for all products) and is a positive role model (for high-risk
Celebrity presenter characteristics products). On the other hand, the study reveals that failure of any of four of the celebrity's characteristics
Hook versus linear conceptualization causes dissuasion by celebrity presenters; these failures include lack of high visibility (i.e., not widely well-
of characteristics known), perceived inexpertness as a user of the product (a strong negative hook that is probably the reason
Persuasion and dissuasion for the poor fit perception), lack of trust (though this is a weak dissuasive factor for celebrities) and,
paradoxically, the celebrity being too likable (for low-risk products). Thirdly, the hook(s) conceptualization of
presenter characteristics is superior to the conventional linear conceptualization in that a hook-scored
regression model accounted for the same amount of variance in persuasion–dissuasion as did the linear model
(adjusted R2s of 41% vs. 43%) despite the hook model's handicap of at least one-third lower possible R2 due to
trichotomization of 7-point ratings into positive, neutral, and negative hooks.
© 2011 Elsevier Inc. All rights reserved.

1. Introduction credibility (Patzer, 1985). Thus, five specific presenter characteristics


are relevant for advertising: visibility or well-knownness, expertise,
A major strategic decision in advertising is whether or not to use a trustworthiness, likability, and role-model identification.
presenter. Executionally, thereafter, the decision evolves into what The present study compares hook theory with the conventional
characteristics the presenter should have to maximally influence linear effects theory of presenter characteristics in ads. This
persuasion. Rossiter and Percy (1987) propose a hook theory of the comparison is important for theory and for practice. The conventional
way presenters' characteristics operate; in contrast with all other academic theory of presenter (or source) effects implies a “more the
theories of presenter effects, hook theory postulates that presenters' better” model of the presenter's characteristics in which any positive
characteristics work in a positive threshold manner (and sometimes in a degree of perceived possession of the characteristic is thought to be
negative threshold manner) rather than in a positive, linear manner. effective. Hook theory, on the other hand, proposes a “threshold”
Rossiter and Percy's (1987) VisCAP model of presenter character- model in which the presenter has to be perceived as very high on
istics incorporates hook theory. The theory's characteristics are relevant characteristics if they are to influence persuasion. This “very
visibility (widely well-known, necessary for celebrity presenters and high” perception has to be immediate and thus can be appropriately
creatable via advertising for “real people” presenters and animated described as constituting a “hook.” For example, if trustworthiness is a
characters), credibility (expertise and trustworthiness), attraction relevant characteristic, which it undoubtedly is for “real people”
(likability and role-model identification), and power (for coercive presenters (see especially the CESLIP model, an extension of the
persuasion that rarely applies except in public service advertising). earlier VisCAP model, in Rossiter and Bellman, 2005, pp. 177–187),
Some presenter theories postulate that physical (facial) attractiveness then the presenter must be perceived, immediately, as very highly
is also persuasive but this is merely because it heightens perceived trustworthy, otherwise the endorsement will not work. “Real people”
presenters are hired mainly because they have a positive trustwor-
thiness hook, which gives them high source credibility despite their
⁎ Corresponding authors. Rossiter is to be contacted at Faculty of Commerce, often-low expertise. Another instance of a necessary hook is a
University of Wollongong, NSW 2522, Australia. Smidts, Rotterdam School of
Management, Erasmus University, P.O. Box 1738, 3000 DR Rotterdam, The Netherlands.
presenter endorsing high-risk “transformational” consumer products
E-mail addresses: john_rossiter@uow.edu.au (J.R. Rossiter), ASmidts@rsm.nl such as expensive fashion wear or luxury holiday resorts; target-
(A. Smidts). audience consumers must perceive a very high degree of role-model

0148-2963/$ – see front matter © 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.jbusres.2011.01.010
J.R. Rossiter, A. Smidts / Journal of Business Research 65 (2012) 874–879 875

identification with the presenter before they will commit to taking 2. Hypotheses
action.
Functionally, hooks are “thresholds,” above which persuasion is
maximally influenced and below which the characteristic either does H1. The hook formulation of presenter's characteristics predicts
not have any effect on persuasion or, if very low, has a negative effect persuasion equally as well as the linear formulation.
(i.e., dissuasion). Dissuasion, first documented in advertising pretests The hypothesis of equal prediction is made because although the
of celebrity presenters by the McCollum–Spielman research company hook formulation is expected to be more predictive, hook scoring of
(1980), is the outcome when a poorly chosen presenter hurts sales of the presenter's characteristics is trichotomous for well-knownness
a new product and results in fewer people buying it than if it had been and expertise and is effectively dichotomous, due to minimal
launched on its own. Only the (positive and negative) hook theory of incidence of negative hooks, for trust, likability, and role-model
presenter characteristics postulates both persuasion and dissuasion. identification and these truncations substantially reduce the variance
In academic advertising research, failure to acknowledge threshold available for prediction. Simulations by Cohen (1983) suggest that
functions in the way variables operate results in many questionable dichotomization reduces the bivariate R2 by 35% if the predictor
and possibly misleading findings when the independent variables are variable is dichotomized at the median, and Irwin and McClelland
measured in the usual continuous manner (typically with “7-point (2003) estimate a 42% reduction in R2 if the dichotomization is at the
scales”). Using 7-point scales, many academic researchers have found 70th percentile, which is where positive hooks (6 +/ 7) were split in
what appears to be a positive linear relationship between, say, the the present study. The reduction in multivariate variance accounted
presenter's perceived expertise and brand purchase intention (e.g., for (by R2) is even worse with multiple predictors that are
Ohanian, 1991). However, the significant positive correlation could be dichotomized or trichotomized. A very conservative estimate,
the result of the often small proportion of consumers who perceive therefore, would be a one-third (33%) loss in the possibly attainable
the presenter to have very high expertise (say 6 or 7 on the 7-point R2 due to hook scoring of the predictors. Consequently, equal
scale) and who are the only individuals persuaded (and thus they prediction by R2 (adjusted for the number of predictors) would
have scores of 6 or 7 on the 7-point purchase intention scale). That is, demonstrate superior theoretical performance of the hook model
both variables could operate in a threshold manner but this is given its statistical handicap.
obscured in a linear correlation or linear regression coefficient. The The superiority of the hook conceptualization of presenter's
difference in interpretation between a threshold model and a linear characteristics should, however, be manifest in another way. The hook
model has important implications for managers. From the linear conceptualization allows both positive and negative hooks for each
correlation results, managers might conclude, as most academics characteristic and expects them to be especially prevalent for two of the
would, that any degree of perceived expertise is sufficient to characteristics, perceived expertise and role-model identification.
guarantee persuasion. This is especially likely to be the erroneous Thus, consumers who perceive that the celebrity has very low
conclusion drawn when persuasion is reported as a group mean score expertise as a user of the product should be dissuaded rather than
(e.g., 4.5 on a 1-to-7 uni-polar purchase intention scale). As Rossiter unaffected (see H3). Similarly, consumers who report very low
and Percy (1987) point out, mean purchase intention scores are identification with the celebrity may see him or her as a “negative”
uninformative about the proportion of consumers who were actually role model and thus should be dissuaded, especially if the celebrity is
persuaded (other than allowing the trivial statistical inference that endorsing a high-risk product (see H6). In the conventional linear
the higher the mean, the more people will have scored above it and conceptualization of presenter characteristics, dissuasion effects
therefore fall into the “high” persuaded category). Group mean cannot be captured because what is revealed is either a nonsignificant
persuasion scores, common in academic research, do not measure effect (i.e., no persuasion) or a significant positive effect (i.e.,
persuasion because they are the confounded mixture of the incidence persuasion).
of people persuaded and the degree to which each is persuaded. The remaining hypotheses make predictions of which particular
Most advertising managers do not actually measure the char- presenter characteristics will be significant predictors of persuasion.
acteristics that potential presenters possess when selecting a As noted, celebrity presenters were chosen for the experiment.
presenter for a campaign. Instead, most managers make an intuitive Rossiter and Bellman's (2005) CESLIP model update of Rossiter and
judgment based on other people's (such as the imagined target Percy's (1987, 1997) VisCAP model of presenter characteristics
audience's) likely extreme perceptions. For example, when managers hypothesizes the following, for celebrities:
choose a presenter because he or she is a role model for the target
H2. Visibility is a positive and negative predictor. That is, a positive
audience, such as Britney Spears may be for young women's casual
well-knownness hook is persuasive whereas a negative hook whereby
clothing or young women's cosmetic products, they base this
the celebrity is perceived as little known will be dissuasive.
judgment on the intuition that most young women in the target
audience age group identify very highly with the presenter. In doing H3. Expertise is a positive and negative predictor. Indeed, the
so, managers are implicitly using a threshold – or hook – model. expectation is that expertise is the strongest predictor because each
The present study tests the hook conceptualization vs. the linear celebrity endorses both a high-fit product and low-fit product and
conceptualization of presenter characteristics, comparing the relative perceived fit for celebrities is primarily based on perceived expertise
ability of the alternative conceptualizations to explain persuasion by as a user of the product category.
comparing (adjusted) R2s. The experiment is based on presenter-
product combinations in the merely associational format favored in H4. Trustworthiness is not a positive predictor. All celebrities are
contemporary print advertisements featuring a celebrity. In the expected to be granted only moderate, not high, trustworthiness as
experiment, each celebrity is paired with a high-fit product and a “obviously paid” spokespersons (whereas, for “real people” presen-
low-fit product to provide sufficient variance on the hypothesized ters, high trustworthiness is essential).
essential celebrity presenter characteristic of perceived user exper-
tise. User expertise is product-category specific and thus it varies H5. Likability is a positive predictor if the presenter is endorsing a
within presenter for different products (hence high-fit vs. low-fit low-risk product. Again, too few celebrities are expected to be very
products). The other characteristics – visibility, trustworthiness, disliked to make this a negative predictor.
likability, role-model identification, and (though not used here)
power – are fixed for the individual presenter and do not change with H6. Role-model identification is a positive and negative predictor if
the particular product that the presenter is endorsing. the presenter is endorsing a high-risk product.
876 J.R. Rossiter, A. Smidts / Journal of Business Research 65 (2012) 874–879

3. Experiment presenters variable. The control group participants evaluated the 16


products alone, that is, not paired with celebrities.
3.1. Overview
3.2.3. Participants
The experiment examines the persuasive effectiveness of celebrity The participants were 172 students attending a large university in
presenters in the associational advertising format, in which the The Netherlands. Their median age was 22 years and an approxi-
celebrity is paired with the product without an explicit endorsement mately equal division of sexes was achieved in the two experimental
being made. With the associational format, there is no possible groups and the control group. Participants took part in the experiment
confounding of presenter characteristics with message characteristics. during class meetings. Sample sizes (n's) were 62 in experimental
(The associational format is now common in celebrity print ads. group A, 61 in experimental group B, and 49 in the product-alone
Widely known examples are magazine ads for luxury watches where, control group. The three groups did not differ significantly in terms of
for instance, Brad Pitt is associated with a Tag Heuer watch and Nicole background demographics, suggesting that the random assignment to
Kidman is associated with an Omega watch.) A separate control group groups was successful.
rated the appeal of the product alone, without a celebrity association.
3.2.4. Experimental stimuli
3.2. Method The experimental stimuli were celebrity-product combinations
and the control stimuli were the products alone. The combinations
3.2.1. Pretests were presented in simple associational form, in words, with no
A series of pretests was conducted with Dutch university students, pictures. To minimize possible carryover, only one celebrity-product
the population from which the participant sample was drawn for the combination (for the experimental groups), or one product (for the
experiment, to select celebrities and products. Based on open-ended control group), was shown per page of the questionnaire. As in the
questioning of the students, eight celebrities were selected who pretest, because all students did not recognize some of the celebrities
varied in their well-knownness, or visibility, and differed in their by name, every celebrity's occupation was noted in parentheses. Also,
apparent appeal as role models for students (the celebrities are brief descriptive copy was added in the several cases where it seemed
named in the experimental design table, following). Participants in necessary to help differentiate an overly generic product (see
the pretest were then asked to write down, for each celebrity, a examples below).
product that the celebrity “is famous for or would be seen as an expert For those products targeted at men, the words “for men” were
user of.” Each celebrity was assigned to his or her most likely high-fit added in parentheses after the product and similarly the words “for
product, with the proviso that half the products were expensive, high- women” were added for products targeted at women. Examples of the
risk purchases and half were inexpensive, low-risk purchases. exact experimental stimuli that the participants saw are: “INGE DE
A second, low-fit product was then assigned to each celebrity, low fit BRUIJN (swimmer) for SHAMPOO (for women)”; “ANDRÉ AGASSI
meaning that the product had either very low or zero frequency of (tennis player) for COFFEE (with special blends and tastes)”; and
mentions as a product that the celebrity would endorse. The proviso was “KYLIE MINOGUE (singer) for WHITENING TOOTHPASTE.”
that if the celebrity was paired first with a high-risk product, then the
second product was also high-risk, and similarly for the first and second 3.2.5. Variables and measures
low-risk products. This consistency was built into the experimental
design so that the celebrity would have basically the same potential for 3.2.5.1. Persuasion–dissuasion. The usual persuasion variable in con-
persuasion for both products he or she was paired with, low potential for sumer behavior studies is self-reported purchase intention (e.g.,
high-risk products and high potential for low-risk products. Ohanian, 1991). However, self-reported purchase intention is likely to
be understated by participants in studies such as the present one which
3.2.2. Experimental design use such an obvious “celebrity recommends product” manipulation. In
Table 1 shows the experimental design. Participants were the often transparent laboratory investigation of celebrity presenters, it
randomly assigned to one of two experimental groups or to the is unlikely that the typical intelligent student would truthfully admit
control group. One experimental group was exposed to the celebrity- that a celebrity led them to like, try, or buy the product. Therefore, an
product combinations denoted by an “A” superscript and the other alternative persuasion measure was borrowed and adapted from a
experimental group was exposed to the combinations denoted with a study by Mittelstaedt et al. (2000) in which participants were asked to
“B” superscript, so that product fit was a within-subjects but between- rate the perceived endorsement effectiveness of the celebrity-product

Table 1
Celebrity-product combinations and experimental design.

Celebrity High fit Low fit

Low-risk products
Inge de Bruijn (swimmer) Shampoo (for women)A Chocolate cookiesB
André Agassi (tennis player) Sports deodorant (for men)B Coffee (with special blends and tastes)A
Kylie Minogue (singer) Whitening toothpasteB BeerA
Ricky Martin (singer) CondomsA Candy barB

High-risk products
Julia Roberts (actress) Hair coloring (for women)A Book club (ordering a minimum of one book, CD or DVD per quarter)B
Bruce Willis (actor) Diver's watch (with depth meter, timer, and alarm) (for men)B Art course (videocassette or DVD)A
Anita Roddick Donate €10 to “Save the Siberian Tiger” fundB Double CD pop music “Greatest Hits of 2001”A
(founder, The Body Shop)
Michael Dell Subscription to e-newsletter, “Entrepreneurs' News” High-fashion ski jacketB
(founder and CEO, Dell Computers) (26 issues per year)A
A
Product paired with celebrity at left for the first experimental group. BProduct paired with celebrity at left for the second experimental group. The control group rated both product
sets (A and B) alone.
Supplementary note: The studies were conducted prior to and with no expectation of the death of The Body Shop founder, Anita Roddick.
J.R. Rossiter, A. Smidts / Journal of Business Research 65 (2012) 874–879 877

Table 2
Estimated purchase incidence and persuasion–dissuasion in the high fit vs. low fit conditions (in percent).

Celebrity High fit Low fit

Percent would buy Percent would buy Net effect of Percent would buy Percent would buy Net effect of
without presenter with presenter presenter without presenter with presenter presenter

Low-risk products
De Bruijn Shampoo (for women) 56 39 –17a Cookies 53 28 –25a
Agassi Sports deodorant (for men) 45 54 +9a Coffee 43 26 –17a
Minogue Whitening toothpaste 43 49 +6 Beer 65 41 –24a
Martin Condoms 51 48 –3 Candy bar 60 30 –30a

High-risk products
Roberts Hair coloring (for women) 43 53 + 10a Book club 22 25 +3
Willis Diver's watch (for men) 30 40 +10a Art course 13 14 +1
Roddick Donate €10 to “Save the Tiger” fund 17 25 +8a Pop music 32 12 –20a
CDs
Dell Subscription to e-newsletter 30 46 +16a Ski jacket 32 17 –15a
a
p b .05, 2-tailed t-tests for proportions (n's were 61 and 62 for the experimental groups and 49 for the product-alone group).

pairing. This judgment presumably is “projective” of whether the 3.3. Results


participant himself or herself was persuaded. In the present study,
Mittelstaedt et al.'s measure was adapted to become a third-person 3.3.1. Persuasion and dissuasion
plural projective measure (see Soley and Smith, 2008) so that perceived An implicit hypothesis tested in the present study is that a presenter
incidence was the measure of persuasion. Immediately below each can either persuade or dissuade. This hypothesis can be tested only by
celebrity-product combination or, for the control group, below each including a product-alone control group such that persuasion is
product, participants were asked to estimate “the percentage of Dutch indicated by a net positive effect of adding the presenter and dissuasion
university students who would try [the product].” is indicated by a net negative effect. High fit of the product with the
The estimates were made on a Juster-type, 11-point answer scale presenter is expected to result in persuasion and low fit to result in
marked at intervals of 10 percentage points, that is, “0, 10, 20…80, 90, dissuasion.
100%.” For the gender-specific products, the projective question was The results in Table 2 reveal that both effects occur, although they
amended accordingly to “the percentage of Dutch male university don't inevitably occur. When there was a high fit (based on the
students…” or “the percentage of Dutch female university students.” pretest) for the celebrity-product combination, statistically significant
By comparing these estimates of incidence against the control group's (p b .05) persuasion was observed for five of the eight pairings; no
product-only estimates, this measure can record dissuasion as well as effect was found for two of the pairings; and in the other case the
persuasion attributable to the celebrity presenter. presenter, Olympic and world swimming champion at the time, Inge
de Bruijn, “hurt” the product, resulting in dissuasion, even though she
3.2.5.2. Presenters' characteristics. After judging the celebrity-product was pre-rated as having a high fit with the product, which was
combinations, participants in the experimental groups then rated the shampoo for women.
eight celebrities in terms of the five presenter characteristics on the When there was a low fit (based on the pretest), dissuasion, which
following semantic-differential scales, in this order: very unknown vs. was severe, was observed for six of the eight pairings – in other words,
very well-known (visibility), very disliked vs. very liked (likability), very in most cases the otherwise low-fit product was significantly (p b .05)
poor role model vs. very good role model (role-model identification), more appealing on its own than when introduced by a celebrity
very unreliable vs. very reliable (trustworthiness), and has very little presenter. In the other two low-fit cases the addition of the celebrity
knowledge about the product vs. has very good knowledge of the had no effect.
product (expertise). Each characteristic was rated on a single-item scale
(the pretest confirmed that the trait descriptions were unambiguous 3.3.2. Hook characteristics as explanations of persuasion and dissuasion
and “concrete,” and thus single items were sufficient; see Bergkvist and The persuasion and dissuasion effects can be explained by
Rossiter, 2007). All answer scales were numbered 1 through 7 in the presenters' characteristics if some of the characteristics have both
positive direction. The bipolar format of semantic differential scales positive and negative effects. The simplest and clearest method of
causes raters to regard the midpoint (4) as neutral and the endpoints presenting these results and testing hypotheses H2 through H6 is a
(1 and 7) as respectively negative and positive (see Heise, 1969, p. 414). univariate approach, which looks at each presenter characteristic
The bipolar-interpreted measures mean that negative hooks (if present) separately. In reality, there may be positive interactions between
as well as positive hooks could emerge. For the hook conceptualization some of the characteristics but these are neither theorized nor tested
of presenter characteristics, the continuous ratings were trichotomized: here.
negative hook = 1 to 2; neutral = 3 to 5; and positive hook = 6 to 7. For The persuasion and dissuasion effects of the presenters' character-
the linear conceptualization, the presenter characteristics were con- istics considered as hooks appear in Table 3. These results are
ventionally scored: 1 = low to 7 = high. aggregates over all celebrities and products except that low-risk

Table 3
Effects of hook characteristics on persuasion–dissuasion: signed differences in buying incidence with celebrity vs. product alone (in percent).

Type of hook Visible (well-known) Expert user of product Trustworthy (reliable) Likable (low-risk products) Role model (high-risk products)
a a
Positive hook −1 +9 0 −9 + 10a
Neutral − 9a −5 − 7a − 14a −2
Negative hook − 24a − 21a NA NA NA

NA = n b 30, so not calculated.


a
p b .05, 2-tailed t-tests for proportions (pseudo n's of participants times the number of ratings were at least 80 for experimental cells and 78 for the product-alone group).
878 J.R. Rossiter, A. Smidts / Journal of Business Research 65 (2012) 874–879

products are analyzed for presenter likability (in accordance with H5) 3.3.2.5. Role-model identification and high-risk products. H6 was
and high-risk products are analyzed for presenter role-model confirmed for the positive role-model hook with high-risk products
identification (in accordance with H6). All characteristics were (the incidence of negative role-model hooks was too low to test the
initially scored as trichotomies but the negative hook results are dual prediction made in this hypothesis). For high-risk products, very
omitted for the trust, likability, and role-model characteristics as there high personal appeal as a role model caused the celebrity's implicit
were too few cases (only 20, 9, and 25 respectively) to provide reliable endorsement of the product to increase the projected buying
estimates. incidence significantly above that of the product on its own (+10%
As in the previous analysis, persuasion was measured as the absolute). Most of the high-risk consumer products were also
difference in the projected buying incidence of the product when “transformationally motivated,” which makes this finding consistent
presented by a celebrity vs. the projected buying incidence of the with the predictions of VisCAP model and the subsequent CESLIP
product presented alone. Negative results (minus signs) signify model.
dissuasion.
3.3.3. Hook model vs. linear model
3.3.2.1. Visibility (well-knownness). The findings half-confirm H2. The The main hypothesis, H1, is that the hook model (hook scoring of
positive visibility hook had no effect on persuasion. A likely ex post presenter characteristics) is equally as predictive of persuasion as the
explanation is that high visibility is par for the course – expected – for linear model (conventional 1-to-7 scoring). Because of the automatic
celebrity presenters and therefore does not bestow a persuasive loss of variance with hook scoring, equal R2s would in fact be evidence
advantage. On the other hand, both lack of high visibility, and for the superiority of the hook model.
especially the negative hook of low visibility, was clearly dissuasive, To test H1, multivariate regressions were computed by using as the
with the latter effect predicted. In the descriptive results for the dependent variable the product-corrected projected buying incidence
individual presenters (not reported here), two of the celebrities (individuals' celebrity-product combination scores minus the prod-
selected for the experiment, Body Shop founder Anita Roddick and uct-alone control group's mean score) so as to capture the “lift” or
Dell CEO Michael Dell, had a substantially widely perceived negative “drop” in incidence caused by the ratings of the celebrity presenter's
hook for visibility, perceived by 46% and 20% of participants characteristics (as in the univariate cross-tabulation results in Table 2
respectively, and it was these two celebrities who contributed almost earlier). Also included were 16 dummy variables for the 16 products
entirely to the dissuasion result. to control for the large differences in the inherent appeal of the
products (again see the results in Table 2). To simplify the
3.3.2.2. Expertise (product-specific). H3 was fully confirmed. As presentation of the findings, the regression results are shown in
predicted, perception of a positive hook for the celebrity as an expert Table 4 without the coefficients for the product dummies, although
user of the product significantly increased persuasion (up by 9% these of course contributed to the substantial observed R2s. Also, since
absolute) and perception of a negative hook, representing an the positive and negative hooks were dummy variables, these were
“inexpert” presenter, severely reduced it (down by 21% absolute) referenced against their neutral rating, so that the regression
and was thus dissuasive. Perceived very high user expertise therefore coefficients in the hook model represent departures from neutral.
appears to be a necessary characteristic for celebrity presenters to Further, the R2s are adjusted for the number of predictors, which was
cause persuasion for all types of products with which they are paired. larger in the hook model due to the hooks being counted as separate
Very high user expertise may well be a necessary characteristic for predictor variables.
“real people” presenters, too (see Rossiter and Bellman, 2005, p. 177) As the last column in the table reports, the hook model and the
but this was not tested in the present experiment. Most striking, linear model accounted for equal amounts of variance in the
however, was the severely deleterious effect of hiring an “obviously dependent variable: R2adj = 41% for the hook model and R2adj = 43%
inexpert” celebrity, which is the negative hook effect predicted for for the linear model (p N .10, n.s.). This finding of equivalence fully
expertise in Rossiter and Percy's (1987, 1997) VisCAP model. supports H1.
Managers having only the results of the linear model would draw
3.3.2.3. Trustworthiness (reliability). H4 predicts that trustworthiness two erroneous conclusions. Managers would conclude that there is a
would not matter for celebrity presenters because “obviously paid” slight positive effect for visibility or well-knownness, whereas the
celebrities would be granted a moderate degree of trust and that the hook model shows that this is really just a negative effect of the
other component of the presenter's credibility – very high perceived celebrity being perceived as too little known. Even more important,
expertise – would be more important. Consistent with the expectation they would conclude that expertise has a large and only positive
of moderate trust, the descriptive results revealed that the average (“more the better”) effect, whereas the hook model reveals two
celebrity in our experiment was rated as neutral (a score between 3 opposing effects. Expertise was importantly demonstrated by the
and 5 on the 1-to-7 scale) by 74% of participants and only 2%
perceived a negative hook for trust. A positive hook for trust had no
Table 4
effect on persuasion, as predicted, but a neutral rating, unexpectedly, Comparison of the predictiveness of hook-coded presenter characteristics (scored − 1,
had a significant negative effect (− 7% absolute). This goes against 0, + 1) and linear characteristics (scored 1 to 7): standardized regression coefficients
Rossiter and Bellman's (2005) CESLIP model, which predicts that (Betas) and percentage R2s (adjusted for number of predictors). Dependent variable:
neutral trust is okay for celebrity presenters. The present finding product-corrected buying incidencea.

reveals that lack of very high trust of the celebrity presenter is Regression Well- Expert Trust- Likable Role R2 %
dissuasive, though possession of very high trust does not bestow a model known user worthy model (adjusted)
persuasive advantage. Hooks
Positive .08 .11b .04 .05 .11b 40.9
3.3.2.4. Likability and low-risk products. H5 was not confirmed and in Negative −.03 −.16b NA NA NA
Linear .08b .30b −.01 .01 .15b 42.8
fact a “reverse” result was obtained in that even a positive hook for
likability with a low-risk product was dissuasive (− 9% absolute). A NA = n too small to compute reliable regression coefficients for negative hooks.
a
possible explanation for this strange result may be “persuasion Buying incidence with celebrity minus control group mean buying incidence for
product alone. Product dummy variables (results not shown) were also added as
knowledge” (Friestad and Wright, 1994) whereby many participants predictors to additionally control for differences in inherent appeal between products.
may have regarded the use of an obviously highly likable celebrity as a b
p b .05, 2-tailed, by t-tests of the unstandardized regression coefficients differing
“gimmick” to sell an otherwise undifferentiated, low-risk product. from their neutral categories for hooks and in the linear model from zero.
J.R. Rossiter, A. Smidts / Journal of Business Research 65 (2012) 874–879 879

hook model to be bidirectional (a “double-edged sword”) in that very aggregate processes correspond to individual-level processes (such as
high perceived expertise persuades and very low perceived expertise “hooks”).
dissuades. Multivariate regression analyses of the 16 specific celebrity-
product combinations yield hook findings for the presenter's
4. Discussion characteristics that closely confirmed the aggregate findings reported
here (detailed results available from the authors). Thus, hook theory is
The two most important findings differ from prior expectations. remarkably well validated in this study. Hook theory predicts
The first important finding is how little effect the addition of a persuasion equally as well as the conventional linear characteristics
celebrity presenter had on persuasion, even when there was a pre- theory despite its handicap of losing at least one-third of the possibly
rated high fit of the presenter with the product. Looking back at attainable multivariate R2 due to trichotomization and dichotomiza-
Table 2, even for the high-fit pairings (left-hand side of table) the tion of the predictors.
presenter's incremental persuasion effect in the successful cases Perhaps the most important practical conclusion concerns the
averaged only about 10% projected net sales increase. However, these expertise – the perceived user expertise – of celebrity presenters,
celebrity-product pairings were hypothetical in the experiment, and which marketing managers do not always regard as essential
actual celebrity advertisements undoubtedly would be much more (Erdogan et al., 2001). Klucharev et al. (2008), in a recent brain-
effective – primarily by drawing many more consumers' attention to imaging study of celebrity-product associational pairings, reported a
the advertised product (see Holman and Hecker, 1983; Rossiter and large positive linear effect on persuasion for perceived expertise.
Bellman, 2005; Rossiter and Percy, 1987, 1997). However, the present findings for the hook model qualify that finding.
The second important finding was the unexpected severity of the Perceived expertise is bidirectional: a positive expertise hook
negative effect – dissuasion – when a celebrity with low fit with the persuades whereas a negative expertise hook dissuades.
product is hired. Low-fit pairings are not uncommon in the real world
when celebrities are hired for other reasons, such as ongoing
contractual obligations to the company or the CEO's whim. Classic References
acknowledged failures of fit were comedian Bill Cosby for the
Bergkvist L, Rossiter JR. The predictive validity of multiple-item versus single-item
brokerage firm E.F. Hutton, British dramatic actor John Houseman measures of the same constructs. J Mktg Res 2007;44(2):175–84.
for McDonald's, Celine Dion for Chrysler, and Jennifer Lopez for Louis Cohen J. The cost of dichotomization. App Psych Mmt 1983;7(3):249–53.
Vuitton luggage. Looking back at Table 1 for low-fit pairings (left-hand Erdogan BZ, Baker MJ, Tagg S. Selecting celebrity endorsers: the practitioner's
perspective. J Adv Res 2001;41(3):39–48.
side of table) it can be seen that in no cases did significant persuasion Friedman HH, Friedman L. Endorser effectiveness by product type. J Adv Res
result and that severe dissuasion was much more likely, averaging 1979;19(5):63–71.
about 20% estimated loss of potential sales. In actual advertisements, Friestad M, Wright PG. The persuasion knowledge model: how people cope with
persuasion attempts. J Cons Res 1994;21(1):1-31.
the massively increased attention effect that a well-known celebrity
Heise DR. Some methodological issues in semantic differential research. Psychol Bull
produces would possibly offset the negative attitudinal effect. 1969;72(6):406–22.
However, the pairing is still likely to be unprofitable given the cost Holman RH, Hecker S. Advertising impact: creative elements affecting brand recall. In:
Leigh James H, Martin Jr Claude R, editors. Current Issues and Research in
of the celebrity.
Advertising, Ann Arbor. MI: Graduate School of Business, University of Michigan;
Both of these important findings were detected only because of the 1983. p. 157–72.
inclusion in the experiment of a product-only control condition. Irwin JR, McClelland GH. Negative consequences of dichotomizing continuous predictor
Comparison of presenter effects with the product's appeal on its own variables. J Mktg Res 2003;40(3):366–71.
Kanungo RN, Pang S. Effects of human models on perceived product quality. J Appl
is a step rarely taken in presenter research, be it academic or applied. Psych 1973;57(2):172–8.
Only two of the many previous academic studies of presenter effects Klucharev V, Smidts A, Fernandez G. Brain mechanisms of persuasion: how ‘expert
have included a product-alone control group (Friedman and Friedman, power’ modulates memory and attitudes. Soc Cog Aff Neurosci 2008;3(4):353–66.
McCollum-Spielman & Company, Inc. Starpower: will the force be with you?
1979 and Kanungo and Pang, 1973) and both found persuasion and Topline 1980;2(3):1–8.
dissuasion, depending on the presenter. All other academic studies as Mittelstaedt JD, Riesz PC, Burns WJ. Why are endorsements effective? Sorting among
well as the available commercial market research studies (see Rossiter theories of product and endorser effects. J Curr Issues Res Adv 2000;22(1):55–65.
Molenaar PCM, Campbell CG. The new person-specific paradigm in psychology. Curr Dir
and Percy, 1987, 1997) have tested alternative potential presenters only Psych Sci 2009;18(2):112–7.
against one another and therefore could not detect whether dissuasion Ohanian R. The impact of celebrity spokespersons' perceived image on consumers'
occurred for any or all of them. Thus, in presenter tests, a product-only intention to purchase. J Adv Res 1991;31(1):46–54.
Patzer G. The physical attractiveness phenomenon. New York: Plenum Press; 1985.
control group must be used so that dissuasion as well as persuasion can Rossiter JR, Bellman S. Marketing communications: theory and applications. Sydney,
be detected. Australia: Pearson Prentice Hall; 2005.
The original purpose of the study was to test the “hook” Rossiter JR, Percy L. Advertising and Promotion Management. New York, N.Y.: McGraw-
Hill; 1987.
conceptualization of presenter characteristics (Rossiter and Percy,
Rossiter JR, Percy L. Advertising Communications & Promotion Management. New York,
1987) against the conventional linear conceptualization (the concep- N.Y.: McGraw-Hill; 1997.
tualization assumed in all other studies of presenters). As Molenaar Soley LC, Smith AL. Projective Techniques for Social Science and Business Research.
and Campbell (2009) explain, social science research must show that Milwaukee, WI.: Southshore Press; 2008.

Das könnte Ihnen auch gefallen