You are on page 1of 14

Journal of Consumer Behaviour

/, Consumer Behav. 9: 32-44 (2010)


Published online 26 October 2009 in Wiley InterScience
(www,interscience,wüey,com) DOI: 10,1002/cb,291

Primacy effect or recency effect?


A long-term memory test of Super
Bowl commercials
Cong Li*
School of Communication, University of Miami, USA

The serial position effects for television commercials were tested within a naturalistic
setting in this study, at both the micro level and the macro level. Television viewers' brand
memory (recall and recognition) for the 2006 Super Bowl commercials were analyzed. At
the micro level, the serial position of each commercial in a same commercial pod was
measured. When the length of a commercialpod was controlledfor, an earlierposition for
a commercial generated better brand recall When the number of preceding ads was held
constant, a commercial in apod with fewer ads generated better brand recognition. At the
macro level, the serialposition of each commercialpod within the whole Super Bowl game
broadcast was measured. The commercial pods at earlier positions generated better
brand memory. Both findings confirmed a strongprimacy effect Managerial implications
of the findings were also discussed.
Copyright © 2009 John Wiley & Sons, Ltd.

Introduction only in the list learning, but also in any task for
which people are required to give responses to
Serial position effects have been studied jtems that are arranged in a numerical order. In
extensively in cognitive psychology research ^^j^^^. ^ ^ ^ . ^ ^^^ ^ ^^^^ g^^^p ^f ^^^^ stimuU,
(Cowan et al, 2002), When people are the ones that correspond to extreme values on
presented with a Ust of items, it is a weU- ^j^^ relevant continuum are more easUy
estabUshed finding that the items at the identified than are those closer to the average
beginning or at the end of the Ust are more ^ ^ j ^ ^ ^^ ^^^ stimulus set Qohnson, 1991).
likely to be recalled than the items in the Although serial position effects could be
middle ofthat Ust. Such phenomena are termed ^ppHed to a number of different research domains,
the primacy and recency effects (Stewart et al., ^ ^ ^ relatively less explored within marketing
2004). Psychological researchers have also communication contexts such as advertising. Only
argued that serial position effects occur not ^ ^^^ researchers have attempted to investigate
the primacy and recency effects for television
commercials, either in a naturalistic or a labo-
•Correspondence to: Cong Ii, Assistant Professor of j.atory setting (e.g., Pieters and Bijmolt, 1997;
communication, School of communication, university ^^^^^ ^ ^ Henderson, 1998;
of Miami, USA,
E-maU: congu@miami,edu Newell and Wu, 2003; Terry, 2005). The

Copyright © 2009 John Wiley & Sons, Ltd, Joumal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10,1002/cb
Long-term memory test of Super Bowl commercials 33

current study follows this research direction by Although the model has been challenged and
examining 109 commercials aired within the criticized by other researchers, it is still one of
2006 Super Bowl broadcast. the most influential works in the literature of
Most prior studies examined the serial posi- verbal learning and memory (Healy and
tion effects at the micro level. The researchers McNamara, 1996). In their study, Atkinson
treated a commercial pod as a "list" (analogous and Shiffrin proposed three distinct memory
to a word list applied in cognitive psychology stores: sensory registers, short-term store
research), and each commercial aired within (primary memory), and long-term store (sec-
that pod as a "word" (analogous to a word in a ondary memory). According to their model,
word list). However, an undeniable fact is that the primacy effect was more related to the
most television commercials are aired within long-term store, and the recency effect could
certain shows. They are not isolated, but strongly be attributed to the short-term store.
or loosely related to each other. In other words, Atkinson and Shiffrin (1968) argued that
not only are commercials within a same com- each item in a list entered into a fixed-capacity
mercial pod competing ^ t h each other for rehearsal buffer and displaced a randomly
vie^vers' attention, but commercial pods within selected item already there when the capacity
the same show are as well. The current study (about four items) had been exceeded. When
takes this issue into consideration and tests the an item was in the short-term buffer, its
serial position eflects at both the micro level and information may be transferred to the long-
the macro level. Specifically, besides treating a term store. Aktinson and Shiffrin made the
commercial pod as a "list" and each commercial assumption that information could be retained
as a "word" (the micro-level test), the whole in both the short-term store and the long-term
2006 Super Bowl broadcast is considered to be a store. When people were asked to free recall
"Üst," and commercial pods within that game the items, they recalled those items still
broadcast to be "words" (the macro-level test). remaining in the short-term buffer and then
It is important to make such distinctions made a fixed number of searches in the long-
between the micro-level test and the macro- term memory store.
level test, especially for managerial implications.
Based on this rationale, items at the begin-
Such distinctions will be further elaborated in
ning of a list receive more rehearsals than items
the follow^ing sections.
in the middle or the end of the list. Therefore,
these items are more likely to be transferred to
and retained in the long-term store, which is
Literature review observed as the primacy effect. Regarding the
Researchers have created and developed recency effect, Atkinson and Shiffrin (1968)
several theoretical models to explain primacy argued that in a short-term memory test, there
and recency effects, such as Atkinson and was little delay between the item presentation
Shiffrin's (1968) modal model and Murdock's and the recall task. When people are required
(I960) distinctiveness model. The purpose of to free recall what they have viewed, the items
the current study is not to provide an overall in the end of the list are still available in the
review of all such models. However, it is short-term store and then are ready to be
necessary to review briefly how cognitive recalled immediately.
psychology researchers approached the mech- Atkinson and Shiffrin (1968) also noticed
anism vinderlying serial position effects. that some factors might affect primacy and
recency effects, such as the presence of a
Multiple memory systems distracter task, the rate of item presentation,
and list length. When a distracter task is put
Atkinson and Shiffrin's (1968) modal model is between the item presentation and the recall
one of the most significant frameworks for task, the recency effect tends to disappear
understanding primacy and recency effects. because the information in the short-term store

Copyright © 2009 John Wiley & Sons, Ltd. Journal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10.1002/cb
Cong Li

is wiped out by the time delay. However, the Other researchers argued that serial position
primacy effect stays intact since those items at effects could be due to different levels of
the beginning of the list are recalled from the attention (Lewandowsky and Murdock, 1989),
long-term store rather than the short-term known as the attentional primacy gradient
store. In contrast, presentation rate and list model. According to this model, during the
length will affect the primacy effect, but not encoding of a Ust, people will pay less and less
the recency effect. With a faster presentation attention to each additional item, thus gen-
rate, items at the beginning of the list will erating the primacy effect. The authors also
remain in the short-term buffer for a shorter suggested that people's memories for listed
time and, thus, less information about them items were based on a chaining mechanism.
will be transferred to the long-term store. Each item was associated with its predeces-
When people are presented with a longer list sors, and items later in the chain depended on
of items, the primacy effect will also be the accurate recall of the earlier items.
affected. Since people are assumed to make The order of recall was also argued to affect
a fixed number of searches for the items in both the primacy and recency effects (Ober-
the long-term store, the probability of retriev- auer, 2003). Forward serial recall usually
ing a particular item is lower when there are generates a large primacy effect and a small
more items. The recency effect will not be recency effect. In contrast, backward recall
affected by the presentation rate or the list wiU usually result in an extended recency
length because the items at the end of the list effect and a relatively small primacy effect.
are primarily recalled from the short-term
buffer rather than from the long-term memory Advertising research on the serial
store. position effects
As stated earlier, serial position effects have not
Other alternatives attracted full attention from advertising
Atkinson and Shriffin's (1968) modal model researchers. There are a limited number of
explained primacy and recency effects from a studies that have explicitly tested the primacy
multiple memory systems' perspective. How- and recency effects for television commercials.
ever, other researchers proposed a unitary Most of these studies have discovered signifi-
memory system and generated alternative cant serial position effects (e.g., Pieters and
frameworks to understand the serial position Bijmolt, 1997; Zhao, 1997; NeweU and Wu,
effects. For example, another explanation for 2003; Terry, 2005) although some have not
the primacy and recency effects was proposed (e.g., Newell and Henderson, 1998).
by Murdock (I960), and it was named the Two early studies were conducted by Zhao
distinctiveness model. Later, the model was (1997) and Pieters and Bijmolt (1997). In
further developed by other researchers (e.g., Zhao's quasi experiment, trained college
Johnson, 1991; Neath, 1993). students interviewed Super Bowl game view-
Murdock (I960) was the first to apply a ers and asked them to recall and recognize
quantitative index of distinctiveness to examine brands advertised within the game broadcast.
serial position effects. He transferred the The study results suggested a primacy effect
physical value of the items in a list into log and the author drew a conclusion that a
values. The distinctiveness of an item was commercial in a pod with fewer ads and/or in
defined as the sum of log differences for it. an earlier position within the pod would
Based on this measurement, the items at the generate better brand memory. In a similar
beginning and at the end of a list stand out due vein, Pieters and Bijmolt analyzed data col-
to their distinctiveness. Therefore, such items lected in the Netherlands. Television viewers
are easier to be recalled than the items in the were interviewed and required to free recall
middle of that list. brand names advertised within specific pro-

Copyright © 2009 John Wiley & Sons, Ltd. Joumal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10.1002/cb
long-term memory test of Super Bowl commercials 35

grams. Again, the free recall of product names pod is examined, a primacy effect will be
revealed a primacy effect. There were some observed (the micro-level test).
suggestions of a recency effect as well, in that
the final item alone was better recalled than H2: When the serial position of each
were the immediately preceding ones. commercial pod within the whole 2006
However, Newell and Henderson's (1998) Super Bowl game broadcast is examined, a
study did not produce significant serial position primacy effect will be observed (the macro-
effects. By conducting a survey for Super Bowl level test).
commercials, the authors found that the
commercials placed in the first position of a
pod were not remembered significantly more Method
than those placed in the middle or the last. Sample of commercials
However, using a similar research technique,
Newell and Wu (2003) found significant serial A Super Bowl broadcast is a good resource of
position effects. The ads placed first in a com- advertising research since Super Bowl com-
mercial pod were recalled significantly more than mercials are always considered to be the best
those in the middle and at the end, confirming a of that year (Zhao, 1997; Youn et al, 2001).
primacy effect. The authors also found that the The commercials analyzed in this study came
recall rate of ads and brands declined signifi- from the 2006 Super Bowl broadcast. Counting
cantly as the Super Bowl game progressed. from the commercial pod preceding the
kick-off of the game to the pod that followed
All of the abovementioned studies applied
the post-^ame interview, the broadcast of the
naturalistic settings to test the serial position
event carried 109 commercials in total for 67
effects for television commercials; these set-
brands.
tings lack control over confounding variables
to some degree. To solve this problem, Terry
(2005) conducted some experiments to pro-
vide more convergent validity to the natur- Telephone interview
alistic studies. In his experiments, college The data were collected via telephone inter-
students viewed lists of 15 commercials and views during the week following the 2006
recalled the brand names. In a short-term recall Super Bowl game. Undergraduate students
task, both primacy and recency effects were enrolled in research classes at a large, eastem
found. In a long-term recall task, the primacy effect university used random digit dialing method to
persisted, but the recency effect disappeared. reach local residents in a county of an eastem
The current study retests the serial position state in the United States. The person who had
effects in a naturalistic setting (the 2006 Super the next birthday at the designated telephone
Bowl broadcast). The unique contribution of number was asked to be interviewed. If a call
this study is to test the primacy and recency resulted in no answer, that number was redialed
effects at both the micro level and the macro later for three times before it was excluded. A
level, which will be further discussed in the total of 769 interviews were then completed
Method section. A long-term memory test for with a response rate of 56.2 per cent. The mean
the Super Bowl commercials is adopted in this age of the interview respondents was 48.0
study. In accordance with most prior study (SD = 15.0), and there were more females than
findings, primacy effects are hypothesized to males (females: 60.8%; males: 39.2%). Most of
occur. Two hypotheses are generated, and data the respondents were working professionals
analyses are conducted at the micro level and (66.4%), although there were some house
the macro level accordingly. makers (7.8%), students (5.1%), retired employ-
ees (14.7%), and others (6.0%). Among the 769
HI: When the serial position of each respondents, 489 persons (63.6%) claimed that
commercial within a same commercial they had watched the 2006 Super Bowl broad-

Copyright © 2009 John Wiley & Sons, Ltd. Joumal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10.1002/cb
36 Cong Li

cast, either some segments or all. The numbers Measurement of independent variables
of people who had watched thefirsthalf and the at the micro level
second half of the game were very close (first
In the micro-level test, the utiit of analysis was
half: 435; second half: 438).
the 67 brands that aired commercials in the
2006 Super Bowl game broadcast.
Brand memory In traditional serial position effects research
In the telephone interviews, respondents who conducted by cognitive psychologists, exper-
had watched the Super Bowl game were asked iment participants were often presented with a
to recall any brand that aired commercials list of words. In such experiment settings, the
during the game broadcast. The recall score for serial position of each stimulus is easy to
each brand advertised in the broadcast was identify (such as the 1st word, the 2nd word,
aggregated by dividing the number of people and the 3rd word, etc.). However, in a
who recalled the brand name by the total naturalistic setting such as the Super Bowl
number of people who watched the segments broadcast, it is more complex to conceptualize
in which the brand was advertised. the serial position for each commercial ("^^ord")
After the free recall question, each inter- within a same commercial pod ("list").
viewee was given a list of brands which had Specifically, the whole broadcast of the
appeared during the game broadcast and asked Super Bowl game contained several commer-
if they remembered seeing a commercial for cial pods. Each pod carried different numbers
that brand. To ensure the validity of the of commercials, ranging from 1-8 in the year
recognition data, eight false alarm brands 2006. The fact that each commercial pod
which did not appear within the Super Bowl differed in length proposed a challenge to
broadcast were also included in the recog- define the serial position for each commercial.
nition test.^ Although some respondents mis- For example, for a pod with three cotnmer-
takenly recognized one or more false alarm cials, the commercial in the third position
brands, at an overall level, the recogtiition data would be both the third ad and the last ad.
were illustrated to be legitimate (excluding However, in a pod with four commercials, the
those respondents who claimed seeing more ad in the third position would not be the last
than one of the false alarm brands from data but the second to the last. Another challenge
analysis did not change the study findings).^ was that some brands were advertised more
The recognition score for each brand was then than once in the Super Bowl broadcast. In one
calculated by dividing the number of respon- pod, it was the first. In another pod, it may be
dents who had recogtiized the brand by the the second. Therefore, a simple definition of
total number of people who watched the the serial position such as the 1st, the 2nd,
segments in which the brand was advertised. and the 3rd was not suitable for the research
design in the current study.
To solve these problems, two measurements
Data analysis and results were applied to conceptualize the serial
Based on the different characteristics of the position at the micro level: the total number
micro-level test and the macro-level test, the of other ads in a pod and the total number of
independent variables were measured in preceding ads in the same pod. This method
different ways accordingly. was applied successfully in prior research
(Zhao, 1997). If the total number of other ads
'The eight false alarm brands were BMW, Ibuprofen, 7up, in a commercial pod was held constant, the
Hardees, NY Life Insurance, Verizon Wireless, Monster, serial position of a commercial could be
and Professor X.
^Some respondents recognized one or more false alarm determined by the number of its preceding
brands (one brand: 94 respondents; two brands: 23 ads in the same pod. For example, if we know
respondents; three brands: 9 respondents; four brands: the total number of other ads in a pod is four.
6 respondents).

Copyright © 2009 John Wiley & Sons, Ltd. Journal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10.1002/cb
Long-term memory test of Super Bowl commercials 37

Table 1. Univadate statistics of the independent and dependent variables at the micro level

Min Max M SD N

Total number of other ads 0 23 4,49 4,49 67


Total number of preceding ads 0 15 2,18 2,17 67
Brand recall (%) 0 55,44 2,95 7,87 67
Brand recognition (%) 10,09 87,56 33,82 16,57 67

and the total number of preceding ads of a Higher advertising frequency was expected to
commercial in the same pod is two, we shall be associated with stronger brand memory
know that this particular commercial is the 3rd (e.g., Berlyne, 1970), Furthermore, brands in
one in that pod. By adopting such measure- certain product categories might be easier to
ments, the confusion of commercials' serial memorize than others (e.g., Zhao, 1997). To
positions in different pods could be avoided. statistically control for these two confounding
Based on the fact that one brand could be variables, the advertising frequency of each
advertised multiple times during the Super brand was content analyzed and coded based
Bowl broadcast, the numbers of other ads and on the video tape of the 2006 Super Bowl
preceding ads for each brand were then broadcast. Regarding the product category, all
aggregated for data analysis. For example, if the brands advertised were categorized into six
a brand ran two commercials within the game, groups: automobiles, food and beverage,
with one (being 3rd) in a pod of five ads, and health and beauty products, services, house-
the other (being 2nd) in a pod of four ads, the hold products, and entertainment-related pro-
total number of other ads for this brand would ducts. Five dummy variables were then created
be seven (four plus three), and the total to represent the first five product categories,
number of preceding ads would be three (two and the sixth product category, entertainment-
plus one).' The univadate statistics of the related products, served as the comparison
independent variable and the dependent group. Both the advertising frequency and five
variable for the micro-level test are presented product category dummy variables were
in Table 1. entered into the multiple regression equations
as the first block (as predictors) for statistical
control. By adopting this analysis strategy, any
Confounding variables significant effects associated with the advertis-
ing frequency and the product category would
Multiple regression analyses instead of ANOVA be controlled,'^
tests were performed in this study due to its
power and flexibility for field studies (Cohen Although the overall quality of the Super
and Cohen, 1983). There are two confounding Bowl commercials may be higher than the
variables that need to be statistically controlled average of other daily commercials, the game
for in the micro-level analysis: advertising broadcast carried a large number and a broad
frequency and product category.
The advertising frequency was defined as Advertising frequency was significantly and positively
the number of commercials for a same brand correlated with the two independent variables (the total
number of other ads and the total number of preceding
aired within the 2006 Super Bowl broadcast. ads) in the micro-level analysis. However, it was entered
into the regression equations in the first block and the
'Another way of defining the numbers of other ads and two independent variables were entered into the
preceding ads for each brand is to calculate the mean regression equations in the second block. The separation
instead ofthe total. However, using the mean numbers of of blocks largely reduced the multicollinearity concerns
other ads and preceding ads for each brand in data in the data analysis because we only focused on the
analysis yielded similar results as using the total numbers incremental R2 change caused by the second block in
of other ads and preceding ads for each brand. this situation.

Copyright © 2009 John Wiley & Sons, Ltd, Joumal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10,1002/cb
38 Cong Li

range of brands and products. There was no pod. Regarding the brand recognition, the
explicit reason to doubt that the commercials significant predictor was the total nvimber of
in some specific pods were better than the other ads (/?<0.01). The regression coeffi-
others. Therefore, the quality of commercials cient revealed that when the number of
was not statistically controlled for. preceding ads was held constant, one more
other ad (it would be a subsequent ad in this
context) lowered the brand recognition by 3 5
Data analysis at the micro level per cent. In the traditional sense of serial
After the control block (advertising frequency positions, no matter how the number of
and product category) was entered into the succeeding ads changes, the position of the
multiple regression equations, the two original ad remains the same. For example, the
measurements of the serial position of each traditional way of counting the serial position
commercial within a same pod (the total for a commercial in the second position within
number of other ads and the total number of a pod of five ads and that for a commercial in
preceding ads) were then entered into the the second position within a pod of seven ads
equations as the second block (as predictors). is the same (the serial position is the 2nd in
The dependent variables in the equations were both cases). However, the regression analysis
brand recall and brand recognition. in this study showed that the total number of
Regarding the effects of the first block, the other ads could generate significant effects on
regression analysis results show^ed that the brand memory even if the traditional sense of
advertising frequency did positively affect the serial position for a commercial remained
brand recall and brand recognition. One more unchanged. Taking the two commercials
ad for a brand promoted the brand recall by 6.5 mentioned above for example again, the 2nd
per cent (p < 0.001) and the brand recognition commercial in a pod offiveads would generate
by 18.7 per cent ( p < 0.001). However, the stronger brand recognition than the 2nd
product category had no significant effect on commercial in a pod of seven ads, even though
brand memory (ps > 0.1). The two confound- both are defined as the 2nd ad in the traditional
ing variables explained 14.5 and 15.7 per cent sense. The results of the multiple regression
of the variance of brand recall and brand analysis at the micro level are presented in
recognition, respectively. Since the effects of Table 2.
these confounding variables are not the focus It should be noted that there is a potential
of this study, they are not further discussed. shortcoming associated with the brand recall
Instead, we want to examine the second block measure adopted in this study for those brands
of the equations and see whether serial that aired multiple commercials during the
position variables are significant predictors Super Bowl game. In the telephone interviews,
of the dependent variables. respondents were only requested to recall
Specifically, when serial position variables brand names, but not related commercial
were entered into the multiple regression details.' Thus, the recall data only revealed
equation as the second block, the increment of what brands interview respondents recalled,
R2 appeared to be significant for both brand but did not suggest from which ads they
recall (p<0.01) and brand recognition recalled the brands. To ensure the validity of
(p<0.01). Regarding the brand recall, the using brand recall in the micro-level analysis,
total number of preceding ads had a significant some additional analysis was performed. The
negative effect, which implied that when the
total number of other ads was held constant,
one more preceding ad decreased the brand
recall by 2.2 per cent (p<0.01). In other Super 'Due to the large number of commercials aired during the
Bowl game and the length limit of the telephone
words, for the best effect of brand recall, a interview, it is not very feasible to request interviewees to
commercial needs to be in thefirstposition of a recall both brand names and related commercial details.

Copyright © 2009 John Wiley & Sons, Ltd. Journal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10.1002/cb
Long-term memory test of Super Bowl commercials 39

Table 2. Results of multiple regression analysis at the micro level

Brand recall (%) Brand recognition (%)

Control block
Constant -2.611 17.685
Advertising frequency 6.491 (0.950)"* 18.744 (1.302)***
Automobiles -2.486 (-0.118) 4.947 (0.111)
Food and beverage .101 (0.004) 7.838 (0.163)
Health and beauty products 5.468 (0.311) 11.165(0.301)
Services -2.678 (-0.098) -1.836 (-0.032)
Household products 1.154 (0.069) 1.874 (0.053)
Serial position block
Number of other ads -.346 (-0.173) -3.486 (-0.827)**
Number of preceding ads -2.173 (-0.734)** -1.958 (-0.314)
Total R2 of control block (%) 14.5 15.7
Incremental R2 of serial position block (%) 23.7*** 24.1***

Note: Numbers in the parentheses are standard betas.


*p < 0.05; **p < 0.01; ***p < 0.001.

brands with multiple commercials during the was a significant predictor of the brand
game (24 brands) and an extreme outlier (one recognition (/><0.05). When the number of
brand) were excluded. This exclusion reduced preceding ads was held constant, one more
the sample size from 67 brands to 42 brands. other ad decreased the brand recognition by
Since all these 42 brands only aired one 5.13 per cent. Both regression analysis results
commercial during the game, we can assure on brand recall and recognition with an
that the recall data were not contaminated. exclusion of the brands that aired multiple
Using the same regression analysis method as commercials are presented in Table 3. The
described earlier (brand recall as the depen- convergence of regression analysis results from
dent variable, total numbers of other ads and both the full-sized sample (67 brands) and the
preceding ads as the predictors, and product reduced-sized sample (42 brands) confirmed
category in the first block for statistical the validity of using brand recall and recog-
control),^ the result was similar to what was nition in the data analysis.
discovered earUer. The total number of other
ads (instead of the total number of preceding
ads) was found to be a significant predictor of Measurement of independent and
the brand recall (p < 0.05). Specifically, when dependent variables at the macro level
the number of preceding ads was held Different from the micro-level test, the unit of
constant, one more other ad lowered the analysis in the macro-level test was not the 67
brand recall by 0.82 per cent. brands. Instead, it was the commercial pods
Moreover, an additional analysis on brand within the whole 2006 Super Bowl game
recognition w^as also performed with an exclu- broadcast. In this context, the traditional
sion of the brands that advertised multiple method to conceptualize the serial position
times. The result was consistent with that of each commercial pod (such as the 1st, the
without brand exclusion. Specifically, it was 2nd, and the 3rd, etc.) was applied. Based on
discovered that the total number of other ads the video tape of the broadcast, there were 28
commercial pods from the first quarter to the
fourth quarter of the game. Therefore, their
^In this case, we do not need to statistically control for serial positions w^ere coded as from the 1st to
advertising frequency since aU the brands (42 brands) in
the analysis only advertised once. the 28th.

Copyright © 2009 John Wiley & Sons, Ltd. Joumal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10.1002/cb
40 Cong Ii

Table 3. Results of multiple regression analysis at the micro level (Excluding Brands with Multiple Ads)

Brand recall (%) Brand recognition (

Control block
Constant -0.321 33.431
Automobiles 1.359 (-0.242) 9.350 (0.274)
Food and beverage 2.952 (0.376) 19.029 (0.397)
Health and beauty products 3.314 (0.632) 9.656 (0.303)
Services 3.535 (0.395) 5.725 (0.105)
Household products 0.711 (1.664) 3.901 (O.I3I)
Serial position block
Number of other ads 0.819 (-0.452)* -5.130 (-0.464)*
Number of preceding ads 0.674 (0.285) 1.684(0.117)
Total R2 of control block (%) 15.4 7.6
Incremental R2 of serial position block (%) 14.1* 16.8*

Note\ Numbers in the parentheses are standard betas.


*p<0.05; **p<0.01; ***p< 0.001.

Since each pod contained several commer- of ads in each commercial pod was entered
cials, the dependent measurements (brand into the multiple regression equations as the
recall and brand recognition) within each first block (as a predictor) for statistical
commercial pod were averaged. For example, control.
for a pod with four ads (promoting brand A, B,
C, and D respectively), the recall and recog- Data analysis at the macro level
nition scores for brand A, B, C, and D were
averaged to generate the memory measure- After the first block (control block) was
ments for that pod. The univariate statistics of entered into the multiple regression equations,
the dependent variables for the macro-level the serial position of each commercial pod w^as
test are presented in Table 4. then entered into the equations as the second
block (as a predictor). The dependent variables
in the equations were brand recall and brand
Confounding variable recognition for each pod.
There was one confounding variable that Regarding the effects of the first block, the
needed to be statistically controlled for in regression analysis results showed that the
the macro-level analysis: number of ads in each number of ads in each commercial pod nega-
commercial pod. A pod with a smaller number tively affected the brand recognition (p < 0.001)
of commercials might be better memorized but not the brand recall (p = 0.89), suggesting
due to less information load. Thus, the number that one more ad in each commercial pod
decreased the brand recognition by 5.9 per cent.
It explained 28.4 per cent variance of the brand
recognition. Similar to what was stated in the
Table 4. Univariate statistics of tbe dependent variables micro-level data analysis, the effect of the
at the macro level confounding variable is not the focus of this
study. Thus, it is not further discussed. Instead,
Min Max M SD N
we want to examine the second block of the
regression equations and see whether the
Brand recaU (%) 0 28.54 6.05 7.11 28
Brand recognition (%) 19.35 65.23 42.62 12.04 28
serial position variable is a significant predictor
of the dependent variables.

Copyrigbt © 2009 Jobn Wiley & Sons, Ltd. Joumal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10.1002/cb
Long-term memory test of Super Bowl commercials 41

When the serial position variable was advertising frequency, it suggested that the
entered into the multiple regression equations, function of forwarding a commercial by three
the increment of . ^ appeared to be significant positions in a pod was almost equal to running
for the brand recognition (/> < 0.05) and one more commercial for the same brand
marginally sigtiificant for the brand recall (about 6.5%) in terms of the brand recall.
(p < 0.10). One unit increase of serial position Considering the high expense of running
lowered the brand recognition by 0.58 per commercials within the Super Bowl game
cent and the brand recall by 0.29 per cent. That broadcast (CNNMoney, 2006), this finding is
was to say, a commercial pod generated better significant. Regarding the brand recognition,
brand memories than its succeeding pods. The the data analysis restilts illustrated that a
results of multiple regression analysis at the commercial in a less "crowded" pod increased
macro level are presented in Table 5. the brand recognition (about 3-5% by reducing
one ad in a pod) if the number of its preceding
ads was held constant.
Discussion and managerial Due to the shortcoming associated with the
suggestions brand recall measure in the current study
Discussion of the findings setting, some additional analyses were per-
formed with an exclusion of the brands that
This study examined the serial position effects aired multiple commercials. Similar results
for television commercials in a naturalistic were observed as opposed to no brand
research setting at both the micro level and the exclusion. Specifically, with the number of
macro level. Potential confotmding variables preceding ads being constant, one more ad in a
(advertising frequency, product category, and pod decreased brand recall (about 0.82%) and
number of ads in a pod) were statistically recognition (about 5.13%) significantly. It
controlled for in the data analysis. The study should be noted that these results are in
results suggested strong primacy effects at support of primary effect, too. For example,
both the micro level and the macro level. the 3rd ad in a commercial pod of three ads is
At the micro level, a commercial pod is con- also the last ad. However, if one more ad is
sidered to be a "Ust" and each commercial in added to the end of this pod, the 3rd ad is still
that pod to be a "word." When the length of a the 3rd but it is not the last anymore. Instead, it
commercial pod was held constant, a com- becomes the second to the last. Based on the
mercial in an earlier position generated a better regression analysis results presented in this
brand recall (about 2.2%) than the one study, this ad will generate stronger brand
following it. In comparison with the effect of memories (0.82% more on brand recall and

Table 5. Results of multiple regression analysis at the macro level

Brand recall (%) Brand recognition (%)

Control block
Constant 10.681 66.936
Number of ads in each commercial pod -0.185 (-0.026) -5.936 (-0.485)***
Serial position block
Serial position of each commercial pod -0.285 (-0.330) -0.581 (-0.397)*
Total R2 of control block (%) 0.004 28.4*"
Incremental R2 of serial position block (%) 0.112 43.9*
Note: Numbers in the parentheses are standard betas.
* 5 ; **p<0.01; ***p<0.001.

Copyright © 2009 John Wiley & Sons, Ltd. Journal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10.1002/cb
42 Cong Li

5.13% more on brand recognition) compared The findings in this study are also consistent
to the newly added one, thus confirming the with many of the psychological models
primary effect. explaining serial position effects. In fact, the
At the macro level, the whole 2006 Super modal model (Atkinson and Shiffrin, 1968) and
Bowl game broadcast was treated as a "list" the attentional primacy gradient model
and each commercial pod in the broadcast as a (Lewandowsky and Murdock, 1989) could be
"word" in that "list." The regression analysis applied to explain the findings here. First,
results suggested that an earlier position of a since the telephone interviews for this study
commercial pod generated stronger brand were conducted at least 1 day after the Super
memories (with one position earlier, there Bowl game broadcast, it was a long-term
were about 0.29% increase in brand recall and memory test. Based on the modal model,
0.58% increase in brand recognition). information in viewers' short-term stores was
Suchfindingsare consistent w^ith most prior already wiped out in such a context. This
research, confirming the primacy effect of explains why only the primacy effect, but not
television commercials' serial positions. The the recency effect, was detected. Alternatively,
unique contribution of the current study is to the results ofthe study could also be explained
extend the primacy effect from the micro level from the perspective of the attentional pri-
to the macro level. This study indeed brings up macy gradient model. At the micro level,
a new conceptualization method of serial viewers' attention could be gradually decreas-
positions. Most previous studies in advertising ing when they were exposed to a commercial
and marketing (e.g., Pieters and Bijmolt 1997; pod. Therefore, the commercials in earlier
Zhao, 1997) have illustrated that an earlier positions of a pod were better memorized, and
position in a commercial pod is more effective a pod with fewer commercials also generated
than a later position (which is replicated in this better brand memories. At the macro level,
study). However, no known study has adopted viewers' attention could also be gradually
a macro-level method to define serial positions decreasing as the Super Bowl game went on
(e.g., regarding a whole commercial pod as a (especially when the competition of the two
"big commercial"). Such a conceptualization teams was not very tight) due to an increase of
method is important and should not be tedium, hence resulting in decreased brand
neglected. Based on the findings in this study, memories.
the current media planning doctrine of "first
position is more effective than later positions" Managerial implications
still holds true, but how to define "first
position" becomes an intriguing theoretical Placing commercials on TV programs is one of
question. The "first" ad in the last commercial major functions of media buyers. The commer-
pod within a TV show does not seem to be so cial time is generally purchased in three primary
"first" if we examine its serial position at a ways: long-term contract, short-term contract,
macro level. It is shown in this study that and last-minute buys (Sissors and Baron, 2002;
positions of commercials in commercial pods Katz, 2007). However, none of the three ways
matter, and so do positions of commercial will guarantee the position of a commercial
pods in TV programs. Although the data in within a show unless a premium rate is paid. In
this study do not provide specific answers general, TV stations rotate commercials' pos-
to questions like "whether the 2nd ad in itions on a random basis if no advertisers pay
the 1st pod will generate better memory additional cost. There seems to he no standard
than the 1st ad in the 2nd pod," the study or consensus among TV stations and advertisers
results do offer enough reasons for scholars for how much extra need to be paid to
and practitioners to conceptualize serial pos- guarantee a "better" position.
itions from both micro-level and macro-level The current study provides some new
perspectives. insights on media buying industry. Taking

Copyright © 2009 John Wiley & Sons, Ltd, Joumal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10,1002/cb
Long-term memory test of Super Bowl commercials 43

the findings of both micro-level and macro- game. Alcoholic consumption is expected to
level tests together, the "best" position for a affect people's memory in a negative way,
commercial w^ithin a TV show^ would be the 1st which might explain why no recency effect
one in the 1st pod. An explicit managerial was observed in this study. To rule out this
implication for advertisers and media buyers alternative explanation, some lab experiments
(especially for those who intend to advertise with strict control over the independent
within the Super Bowl games) is to try every variables need to be conducted.
effort to position their commercials in an early Finally, the findings of this study are based
position (as early as possible) in an early on a single sports event broadcast (Super Bowl
commercial pod (as early as possible), given game broadcast). To obtain more robust and
that brand memories (recall and recognition) generalizable conclusions, more data across
are the criteria for judging advertising effec- events are needed.
tiveness.

Biographical note
Limitations and future research Dr. Cong Li received his Ph.D. from the Univer-
There are some limitations of this study that sity of North Carolina at Chapel Hill in the
need to be addressed. First, as described in the United States. He currently is an Assistant Pro-
data analysis section, the telephone interview fessor in the School of Communication at the
respondents in this study were only requested University of Miami. He is interested in applying
to recall brand names but not to relate psychological frameworks to the study of adver-
commercial details due to the length limit of tising and marketing messages. He has published
the interviews. Without respondents providing in journals such as Cyber Psychology & Beha-
ad details, it is hard to tell whether the brand vior, Journal of Relationship Marketing,
recall was truly generated by the commercial Prism, among others.
messages people viewed in the Super Bowl
broadcast. Given the fact that several brands
aired multiple commercials within the game, it
is also difficult to differentiate which commer- References
cial generated brand memory and which did Atkinson RC, Shiffrin RM. 1968. Human memory: A
not. This shortcoming was partly compensated proposed system and its control processes. In
by additional data analyses with an exclusion of The Psychology of Learning and Motivation:
the brands with multiple ads, but future Advances in Research and Theory, Spence
research (especially laboratory experiments) KW, Spence JT (eds). Academic Press: New
is definitely needed to better control the York; 89-195.
confounding effects caused by advertising Berlyne DE. 1970. Novelty complexity and hedonic
repetition on brand recall. value. Perception and Psychophysics 8(5): 279-
286.
Second, a naturalistic research design such CNNMoney. 2006. Super Bowl XL's extra-large ad
as the current study has certain advantages on sales. Available at http://money.cnn.com/2006/
external validity, but it lacks control on 01/03/news/companies/superbowlads/ [accessed
confounding background factors. Although on 11 December 2008].
some confounding variables were statistically Cohen J, Cohen P. 1983. Applied Multiple
controlled for in the data analyses, there were Regression/Correlation Analysis for the Beha-
certain alternative explanations that could not vioral Science, 2nd edn. Lawrence Erlbaum:
be completely ruled out. For example, a Super Hillsdale, NJ.
Bowl game is often viewed as a happy event Cowan N, Saults JS, Elliott EM, Moreno MV. 2002.
and a "party." It would not be surprising if Deconfounding serial recall. Journal of Memory
people consumed alcohol while watching the and Language A6(iy. 153-177.

Copyright © 2009 John Wiley & Sons, Ltd. Journal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10.1002/cb
44 Cong Li

Healy AF, McNamara DS. 1996. Verbal learning and Oberauer K. 2003. Understanding serial position
memory: does the modal model still work. curves in short-term recognition and recall./owr-
Annual Review of Psychology 47(1): 143-172. nal of Memory and Language 49(4): 469-
Johnson GJ. 1991. A distinctiveness model of serial 483.
learning. Psychological Review 98(2): 204-217. Pieters RGM, Bijmolt THA. 1997. Consumer mem-
Katz H. 2007. The Media Handbook: A Complete ory for television advertising: a field study of
Guide to Advertising Media Selection, Planning, duration, serial position, and competition effects.
Research, and Buying, 3rd edn. Lawrence Erl- Joumal of Consumer Research 23(4): 362-
baum: Mahwah, NJ. 372.
Lewandowsky S, Murdock BB Jr. 1989. Memory Sissors JZ, Baron RB. 2002. Advertising Media
for serial order. Psychological Review 96(1): 25- Planning, 6th edn. McGraw-Hill: New York, NY.
57. Stewart DD, Stewart CB, Tyson C, Vinci G, Fioti T.
Murdock BB Jr. I960. The distinctiveness of 2004. Serial position effects and the picture-
stimuU. Psychological Review 67(1): 16-31. superiority effect in the group recall of unshared
Neath 1. 1993. Distinctiveness and serial position information. Group Dynamics: Theory,
effects in recognition. Memory and Cognition Research, and Practice 8(3): 166-181.
21(5): 689-698. Terry SW. 2005. Serial position effects in recall of
Newell SJ, Henderson KV. 1998. Super Bowl adver- television commercials./owm«/ of General Psy-
tising: field testing the importance of advertise- chology 132(2): 151-164.
ment frequency, length and placement on recall. Youn S, Sun T, WeUs WD, Zhao X. 2001. Commer-
Joumal of Marketing Communications 4(1): cial liking and memory: moderating effects of
237-248. product categories. Joumal of Advertising
Newell S, Wu B. 2003. Evaluating the significance of Research 41(3): 7-13.
placement on recall of advertisements during the Zhao X. 1997. Clutter and serial order redefined and
Super Bowl. Joumal of Current Issues & retested./owrwa/ ofAdvertising Research 37(5):
Research in Advertising 25(2): 57-67. 57-73.

Copyright © 2009 John Wiley & Sons, Ltd. Joumal of Consumer Behaviour, Jan.-Feb. 2010
DOI: 10.1002/cb
Copyright of Journal of Consumer Behaviour is the property of John Wiley & Sons, Inc. / Business and its
content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for individual use.

You might also like