Beruflich Dokumente
Kultur Dokumente
The present study addresses how to measure three constructs commonly used in adver-
tising research, namely attitude towards the ad, brand attitude and brand purchase
intention. The study replicates and extends Bergkvist and Rossiter's (2007) finding that
single-item measures are equally predictively valid as multiple-item measures of basic
(doubly concrete - see Rossiter's 2002 C-OAR-SE procedure) constructs in marketing,
namely A^^ and \^^^¿. One extension is that the finding holds for free-standing, tailor-
made single-item measures, whereas the previous study establishes this result only for
single-item measures extracted from multiple-item measures. Another extension is that
single-item equivalence of predictive validity further holds for another widely employed
dependent variable construct, Plß^^^j. The present study goes beyond Bergkvist and
Rossiter's study in that it shows that items commonly used in multiple-item measures of
A^j and Ag^^^^ vary in their predictive validity and that, in some cases, the differences are
substantial. The main finding is the further empirical proof that multiple-item scales are
unnecessaiy for validly measuring basic constructs.
Introduction
Multiple-item measures of all constructs have been the norm in academic
marketing research, including research on advertising, for the last 30
years. Recent research challenges this norm on theoretical grounds for not
being applicable to all types of constructs of interest in the social sciences
(Rossiter 2002) as well as on empirical grounds (Drolet & Morrison 2001;
Bergkvist & Rossiter 2007). Specifically, Rossiter (2002) argues that single-
item measures provide valid measurement of 'doubly concrete' constructs
- that is, constructs for which both the object of measurement and the
attribute of measurement are clear and unambiguous for those rating
the object on the attribute. Bergkvist and Rossiter (2007) show that this
argument holds empirically by demonstrating that the predictive validity
of single-item measures of the doubly concrete constructs attitude towards
the ad (A^^) and brand attitude {\^^„J is equal to the predictive validity
of multiple-item measures of the same constructs. Moreover, Drolet and
Morrison (2001) demonstrate mathematically that increasing the number
of items in a measure of a doubly concrete construct (in their study, A^^)
will actually decrease its validity compared with a measure with one,
or at most two, good items, although they did not test this relationship
empirically. The multiple items were attempted synonyms of the attribute
descriptor, which is the typical way of generating multiple-item measures
of doubly concrete constructs.
In their study of A^^ and \^^^¿, Bergkvist and Rossiter (2007) extract
one of the items from the multiple-item battery to compare the predictive
validity of single-item and multiple-item measures - that is, they do not
test free-standing single-item measures. Potentially, extracted single-item
measures could be 'contaminated' by being rated in a battery of similar
items. The purpose of the present study is to compare the predictive
validity of single-item measures, extracted from the multiple-item meas-
ures as well as free-standing, of A^^ and A^^^^^, with the predictive validity
of traditional multiple-item measures of the same constructs. In addition
to replicating the findings of Bergkvist and Rossiter (2007), and thereby
strengthening confidence in their conclusions (Evanschitzky et al. 2007),
the study adds to the literature by suggesting and validating tailor-made
single-item measures of A^^, \^^^¿ and purchase intention (PIQ,^^¿). In
particular, the study makes a contribution to the literature as the valida-
tion of the tailor-made measures is made not only vis-à-vis multiple-item
measures, as in the study by Bergkvist and Rossiter (2007), but also vis-
à-vis four extracted single-item measures, thereby making it possible to
evaluate which attribute descriptor (e.g. 'like-dislike' or 'good-bad') is the
more valid of those tested.
608
TAILOR-MADE SINGLE-ITEM MEASURES OF DOUBLY CONCRETE CONSTRUCTS
attributes that nearly all the raters (e.g. consumers in a survey) describe
identically. Rossiter (2002) argues that multiple-item measures are not
necessary' for doubly concrete constructs since the raters agree on the defi-
nition of both the object and attribute of the construct. The agreement
between raters means that these constructs are not 'elusive, intangible phe-
nomena' that require multiple-item measures (cf. DeVellis 1991, p. 7) and
that they do not have different 'facets' to be tapped into (cf. Baumgartner
& Homburg 1996). In the C-OAR-SE procedure for scale development,
the measurement requirements are different for constructs with an
'abstract' object (e.g. materialism or capitalism) or an 'abstract' attribute
(e.g. service quality or extraversion). Both these cases require multiple-
item measures. However, these types of construct are not evaluated in the
present study and will not be discussed further here (see Rossiter 2002 for
a discussion of the measurement of constructs with an abstract object, an
abstract attribute, or both). Furthermore, given that single-item measures
are sufficient for doubly concrete constructs, one of the main arguments
for the use of multiple-item measures - that they are more reliable in
the sense of internal consistency than single-item measures in that they
allow for the computation of coefficient alpha (e.g. Churchill 1979) - is
irrelevant. Rossiter (2002) argues that the only meaningful use of the term
'reliability' is the precision of a score obtained from a measure in a specific
application. The only case in which the number of items affects reliability
is the case of an 'eliciting' attribute, which is one type of abstract attribute,
and which most attributes in marketing constructs are not. For an eliciting
attribute (e.g. the personality trait of extraversion), the items are a sample
and larger sample size means greater precision. See Rossiter (2002, p. 328)
for a thorough discussion of reliability.
609
INTERNATIONAL JOURNAL OE ADVERTISING, 2009, 28(4)
Common-methods bias
Bergkvist and Rossiter (2007) investigate the additional issue of whether
common-methods bias (see, e.g., Williams et al. 1989) spuriously inflates
the correlation between multiple-item measures. Inflated correlations
could occur if two multiple-item measures are made up of identical-format
items - for example, 'semantic differential' items - or if two single-item
measures use the same adjectives to describe the attribute of interest (e.g.
'like-dislike' in both or 'good-bad' in both). Bergkvist and Rossiter (2007)
find no support for common-methods bias in any of these forms, but they
caution that their results with respect to multiple-item measures may not
hold for multiple-item measures with more than the three items, which
was the number of multiple items they tested in their study. The present
study investigates multiple-item measures made up of four items.
610
TAILOR-MADE SINGLE-ITEM MEASURES OE DOUBLY CONCRETE CONSTRUCTS
Sample of ads
The questionnaire contained four different ads with brands from four
different product categories: beer, breakfast cereal, wine and painkillers.
To maximise the effect of A^^ on Ag^^^^^, the ads were taken from foreign
magazines and the brands were not available in the local market where the
survey was conducted, which means that the only possible causes of Ag^^^^
would be the information in the ad and the participants' attitude towards
the ad itself (cf. Campbell & Keller 2003; Bergkvist & Rossiter 2007). The
order of the ads was rotated to average out possible order effects.
611
INTERNATIONAL JOURNAL OF ADVERTISING, 2009, 28(4)
Operationali-
sation and
Construct variable name Question Answer scale
Ajj Multiple-item, Below you will find four pairs of Dislike : : : : : : : : Like
A adjectives. Indicate how well one Good : : : : : : : : Bad
"Ad(Multiple)
or the other adjective in each pair Pleasant : : : ; : : : : Unpleasant
describes how you perceived the ad Uninformative : : : : : : : : Informative
for/BRAND/.
ABrand Multiple-item, Below you will find four pairs of Bad : : : : : : : : Good
adjectives. Indicate how well one Like : : : : : : : : Dislike
or the other adjective in each pair Pleasant : : : : : : : : Unpleasant
describes how you feel about Useful : : : : : : : : Useless
/BRAND/.
P'Brand Single-item, If you were going to buy /PRODUCT Cl Certain or practically certain
CATEGORY/, how likely would you Cl Very probable
be to try /BRAND/? Cl Probable
n Fairly good possibility
n Some possibility
Û Slight possibility
O No chance or almost no chance
PiBrand Multiple-item, Below you will find four pairs of Unlikely : : : : : : : : Likely
°'Brand(Mulliple)
adjectives. Indicate how well one Probable : : : : : : : : Improbable
or the other adjective in each pair Uncertain : : : : : : : : Certain
describes the likelihood that you Impossible : : : : : ; : : Possible
would try /BRAND/ if you were to
buy /PRODUCT CATEGORY/.
Notes: Single-item measures coded 1 to 7, where 7 is the 'positive' response. Items in multiple-item measures coded - 3 to +3,
where +3 is the 'positive' response. In the questionnaire, the semantic differential items were labelled at the top of the set
using the adverb 'neither' in the middle and the adverbs 'slightly,' 'quite,' and 'extremely' on either side.
612
TAILOR-MADE SINGLE-ITEM MEASURES OF DOUBLY GONCRETE GONSTRUGTS
Results
613
INTERNATIONAL JOURNAL OF ADVERTISING, 2009, 28(4)
Ad as predictors of A
In the first analysis, A^^ is used to predict A^^^^^. The results of the analy-
sis show that the single-item measure, A^^^gj ,^^, performs equally as well
as the multiple-item measure, A^j^j^^|^¡ j^^, irrespective of which opera-
tionalisation of the dependent variable, Ag^^^^, is used (see Table 2, first
two rows). In only two instances was the correlation between AAd(Single) ( g )
and the dependent variable significantly different from the correlation li
between A^j^j^^,^¡p|^^ and the dependent variable; the correlation is signifi-
cantly lower (p < 0.05) for A^^^gj^g,^^ in the model for the cereal ad with
^Branci(Muitipie) ^^ ^^^ dependent variable, and the correlation is significantly
higher for A^^^gj^g,^, {p < 0.05) in the model for the beer ad with \,,^^(^,^,,^
as the dependent variable. The results also show that the validity coef-
ficients, r, are the same no matter whether the dependent variable,
is measured with multiple items, \,,,^(umpw o^ ^ «ingle item, ^J^
In sum, the results do not indicate at all that the multiple-item measure
of A^j has higher predictive validity than the single-item measure when
^Brand '« ^^^ dependent variable. The most important overall result is
that the single-item 'doubles' - that is, the validity coefficients between
a single-item measure independent variable and a single-item measure
dependent variable, which is the worst approach according to advo-
Table 2: Validity coefficients (r) for multiple-item and single-item measures of A^j
as predictors of multiple-item and single-item Ag^^^^
A
"Brand(Multiple) AB and(Single)
614
TAILOR-MADE SINGLE-ITEM MEASURES OE DOUBLY GONGRETE GONSTRUGTS
PI PI
^ 'Brand(Multiple) •^'BrandlSingle)
Beer Cereal Wine Painkillers Beer Cereal Wine Painkillers
615
INTERNATIONAL JOURNAL OF ADVERTISING, 2009, 28(4)
616
TAILOR-MADE SINGLE-ITEM MEASURES OE DOUBLY CONCRETE CONSTRUCTS
shows that Ag^^^^ measures with a 'good-bad' response scale have predic-
tive vahdity that is better than or equal to measures with the other three
commonly used response scales in the comparison. Thus, answer scales
with 'good-bad' are most likely to be the best available choice for meas-
uring Ag_.^^j. Also, again, an important finding is that a good single-item
measure, this time of A^^^^^, is impervious to being placed among similar
adjacent items.
Table 4: Validityr coefficients (r) for multiple-item and single-item measures of A^^
as predictors o extracted measures of single-item Ag
AB rand(Like) As rand(Good)
^AdiMultiple)
0.61 0.67 0.43 0.66 0.58 0.61 0.44 0.61
A
"Ad(Single)
0.54 0.64 0.53 0.53 0.65 0.63 0.57 0.49
A
"Ad(Uke)
0.65 0.56 0.47 0.61 0.64 0.67 0.53 0.58
A 0.56
"Ad(Good)
0.59 0.44 0.56 0.49 0.45 " 0.36 ^ 0.48
"Ad(Pleasant)
0.50 0.53 0.39 0.59 0.38 0.41 0.31 0.53
A Í 0.27 0.44
"Ad(lnformative)
0.01 0.38 0.37 0.40 0.10 0.38
617
INTERNATIONAL JOURNAL OF ADVERTISING, 2009, 28(4)
Conclusions
This study replicates and extends the study by Bergkvist and Rossiter
(2007). The present results show that tailor-made single-item measures of
^Ad' "^Brand ^^'^ ^^Brand ^^^ ^^ predictively Valid as traditional multiple-item
measures of these doubly concrete constructs. The results extend those of
Bergkvist and Rossiter (2007) for A^^ and Ag^^^^ to another doubly concrete
construct widely used in marketing studies, Plßrand" "^^^ present study also
extends Bergkvist and Rossiter's (2007) study by comparing the predictive
validity of different single-item measures. The results show that items
commonly used in multiple-item measures of A^^ and Ag^^^^ vary in their
predictive validity and that, in some cases, the differences are substantial.
Moreover, the results show that 'extracted' single-item measures, as long
as the extracted item is a valid one, are not affected by other adjacent
items that appear in a multiple-item battery, and that they perform the
same as stand-alone, tailor-made single-item measures. Finally, the results
also show that common-methods bias is not a problem for either the
multiple-item measures or the single-item measures.
The present study strengthens the conclusion in Bergkvist and Rossiter
(2007) that carefully crafted single-item measures - of doubly concrete
constructs - are at least as valid as multiple-item measures of the same
constructs, and that the use of multiple items to measure them is unnec-
essary. The recommendation for advertising researchers measuring A^^,
^Brand '^^ ^^Brand '^ '"'-* ^^^ ^^^ tailor-made single-item measures validated
in the present study. The recommendation for journal editors and review-
ers is not to reject submitted manuscripts on the grounds that they used
single-item measures to measure doubly concrete constructs.
The large differences in the predictive validity of the different single-
item measures of A^^ that were evaluated in the study emphasise the
importance of carefully selecting the adjective descriptor in measures of
618
TAILOR-MADE SINGLE-ITEM MEASURES OF DOUBLY CONCRETE CONSTRUCTS
References
Aaker, D.A., Kumar, V., Day, G.S. & Lawley, M. {Z^^S) Marketing Research, The Pacific
Rim Edition. Milton, Queensland: Wiley.
619
INTERNATIONAL JOURNAL OF ADVERTISING, 2009, 28(4)
620
TAILOR-MADE SINGLE-ITEM MEASURES OF DOUBLY CONCRETE CONSTRUCTS
621