You are on page 1of 10

ADMOD: An Advertising Decision Model

Author(s): David A. Aaker


Source: Journal of Marketing Research, Vol. 12, No. 1 (Feb., 1975), pp. 37-45
Published by: American Marketing Association
Stable URL: http://www.jstor.org/stable/3150656 .
Accessed: 19/10/2013 10:36

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

American Marketing Association is collaborating with JSTOR to digitize, preserve and extend access to
Journal of Marketing Research.

http://www.jstor.org

This content downloaded from 142.150.190.39 on Sat, 19 Oct 2013 10:36:15 AM


All use subject to JSTOR Terms and Conditions
DAVID A. AAKER*

ADMODis an advertising decision model which is designed to address simultaneously


the budget decision, the copy decision, and the media-allocation decision. The
model, which focuses upon specific consumer decisions that advertising is attempting
to precipitate, is illustrated with an example.

ADMOD: An Advertising Decision Model

This article describes an Advertising Decision Mo- lar, the studies by Ray and his colleagues [12, 13,
del, termed ADMOD. The model is designed to make 14] on repetition and Aaker and Brown [5] on vehicle
media-allocation decisions. However, its thrust differs source effects have shown that the assumption that
from most other media models in two respects: (1) these dimensions of analysis can be considered
it includes the copy decision and the budget decision, independently of copy effects is far from reasonable.
and (2) it relies upon a different operational conception Together with the classic work of Starch [18] on
of the advertising process. advertising readership, they suggest that media-alloca-
tion models should take the copy decision into account
A MORE COMPREHENSIVE MODEL more explicitly.
There are three tactical advertising decision areas: There are, of course, good reasons why these
the media budget, the budget allocation, and the copy interactions have been largely ignored in the past.
decisions. Previous research into these decision areas First, the three areas are usually the responsibility
has been remarkably isolated. Budget-setting models of different decision makers. Firm managers set the
have generally used experimental design or statistical budget, agency media specialists allocate the budget
models, usually ignoring copy considerations or media to media alternatives, and creative people make the
placement. Media models have normally considered copy decisions. There is little pressure, therefore, to
the budget to be fixed and similarly have ignored copy integrate. Instead, a demand exists for independent
decisions, which generally involved different research efforts to aid these decision makers. Second, there
traditions. have been technical and conceptual problems in the
That interdependencies exist hardly needs demon- development of the machinery oriented toward media
stration. First, the optimal budget level will clearly allocation and copy decisions, even when they are
depend upon the creative effort. For ineffective ad- considered independently. Therefore, it has been pro-
vertising, the optimal level may be zero. One type ductive to make advances in these areas. Indeed, it
of copy approach will require a different level of is questionable whether any real effort toward an
exposure intensity than another. Second, the budget integrative approach would be feasible without the
level will depend upon the type of media used and advances that have been achieved to date in media
how effectively the target segments can be reached. models, copy testing, and related techniques. Finally,
Finally, recent research has demonstrated the interac- when media allocation or copy decisions are consid-
tions between copy and media scheduling. In particu- ered separately, the measures used need have only
relative validity, since alternative choice is the goal.
When budget decisions are introduced, a measure with
validity in an absolute sense is necessary and new
*David A. Aaker is Associate Professor of Business Administra- demands are imposed upon researchers.
tion, University of California, Berkeley. The author would like It is now appropriate to address the budget, the
to thank the W. R. Simmons Company for making data available
and Christopher Sprague and Leon Liebman of Interactive Market media, and the copy decisions in concert. The intro-
Systems Incorporated who provided convenient access to it. Thanks duction of modern information systems supported by
are also due to the Institute of Business and Economic Research on-line computers and model banks are effectively
for their research support and to Alex Chow for his able programming breaking down the isolation of the different decision
assistance and to the JMR reviewers for their helpful comments. makers. The media model technology is at the point
37

Journalof MarketingResearch
Vol. XII (February 1975), 37-45

This content downloaded from 142.150.190.39 on Sat, 19 Oct 2013 10:36:15 AM


All use subject to JSTOR Terms and Conditions
38 JOURNALOF MARKETINGRESEARCH,FEBRUARY1975

where it has achieved acceptance and demonstrated of generating the various insertion schedules to be
considerable validity. Progress has been made in evaluated.
understanding the communication process, in setting
objectives, and in testing copy against these objectives. The Objective Function
With barriers reduced, it is useful to push the field The focus of the ADMOD objective function is
in the direction of more ambitious models. not on the aggregate vehicle audience but upon sample
populations selected from the various segments. In
A DIFFERENT VIEW OF THE ADVERTISING evaluating an insertion schedule, ADMOD examines
PROCESS its likely impact upon every individual in the sample.
The impact will depend upon the net value of the
Most media models rely upon a model of advertising decision or cognition change involved (the segment
which suggests that advertising creates an advertising
effect), the number and source of exposures to the
exposure (or some similar construct) which, in turn, individual created by the schedule (advertisement
creates sales. The advertising exposure level will
exposures), and the impact of the exposures upon
dissipate over time but can be maintained or built the probability of obtaining the desired cognition
up by more advertising. The heart of the model, then, change or decision (the repetition function). Using
is an aggregate response curve that relates exposure
appropriate scaling factors, the result is projected to
levels to sales (i.e., [8, 9]), or perhaps advertising the real population, providing the total expected profit
expenditure levels to sales directly. ADMOD, in con- generated by that media schedule.
trast, has a more disaggregative view of the advertising A sample is thus required from each segment. A
process. It focuses upon specific consumer decisions, representative sample of the whole population is not
which have long-run implications for the firm, that
necessary. The proportion of people in the sample
advertising is attempting to precipitate. For example, from a certain segment need not equal the proportion
advertising may be directed toward inducing a con- of the population contained in that segment. The only
sumer to try a brand for the first time, to try a brand
in a new way, to change his brand attitude, or to requirement is that a reasonable number of people
from each segment- 100 to 200, for example-be part
become aware of the brand. Such consumer responses of the sample. Of course, the sample should be
will tend to generate a future sales or profit stream
to the firm which represents a certain present value. representative of each segment. Ideally, each member
of the segment should have an equal chance of being
For instance, an attitude change may be linked to
selected. When small audience media vehicles are
an increase in loyalty that will have an estimable effect
included, it may be useful to increase the sample size
upon purchases over time. A trial purchase will result somewhat. The sample size from segment s is termed
in a subsequent purchase sequence which can be
similarly estimated (for example, see [2]). ns. The total size of segment s is termed Ns.
Before proceeding further, it is useful to define
The concern of advertising is then upon some
several terms. A media vehicle (indexed by j; j = 1,
indicator of the probability of the consumer changing
..., J), such as Vogue, Time, NBC News, and Monitor
his cognitions in the desired way or taking the desired
action. When the objective is to obtain new customers, Radio, provides the immediate environment for the
advertisement. By specifying the vehicle, it will be
an appropriate measure might be the intention to buy.
assumed here that the position of the advertisement
Alternatively, the measure could be based upon some within the vehicle and the size of the advertisement
variable causually linked to intention to buy. For is also specified. Including such factors in the model
example, it may be that brand name awareness, or more formally would be a trivial extension. A copy
knowledge of a brand attribute, or familiarity with alternative (indexed by c; c = 1, ..., C) is a creative
a jingle may be closely linked to intentions to make
a trial purchase [6]. In any case, a construct is approach. It could be the specification of refutational
copy, fear appeals, the use of experimental evidence,
developed that may provide the dependent variable or a particular awareness campaign. An insertion
in studies used to estimate the various parameters
in the advertising decision model. This construct will option (xcj) is here defined as the insertion of a copy
alternative into a vehicle. A set of insertion options
tend to be more operational than aggregate sales, partly
is a media insertion schedule. A media insertion
because it will enable the researcher to exploit copy
schedule will be constrained to have only one copy
testing methodology more fully.
alternative; the possibility of relaxing this constraint
will be considered later in the article.
ADMOD
The goal of ADMOD is to select a budget level, The Segment Effect
a copy approach, and a media insertion schedule to The assumed objective of the advertiser is to precip-
maximize the objective function. The task of the itate a cognitive change or decision. The long-term
objective function is to assign a value to a given media value to the firm of obtaining such a result from a
insertion schedule. A selection heuristic has the task member of segment s is termed w . It will be the

This content downloaded from 142.150.190.39 on Sat, 19 Oct 2013 10:36:15 AM


All use subject to JSTOR Terms and Conditions
ADMOD: AN ADVERTISINGDECISION MODEL 39

present value of the projected purchasing behavior involves vehicle and advertisement exposure probabil-
pattern caused by the cognitive change or decision. ities, will be developed.
Model components, such as the repetition function Let bij be the probability that individual i will be
and vehicle source effects, are then developed using exposed (with exposure carefully defined) to the
operational measures that are linked to the objective. vehicle j under consideration. It is assumed that the
To fix the concept, several examples will be presented. b i,remain stable through the time period of the planned
Suppose the goal is to generate a trial purchase campaign. (For a discussion of this assumption, see
of a brand by segment members. In this context, ws [16].) Let hc. be the conditional probability of an
would be the discounted profit stream expected from exposure (again, with a suitable definition) to an
a random segment member who tries the brand for advertisement using copy alternative c in vehicle j,
the first time. The long-term value to the firm of given that an exposure to vehicle j has been obtained.
obtaining a trial purchase will reflect the gross margin The latter term will depend upon the vehicle and the
of the brand, a discount factor reflecting the cost copy. Some vehicles have a higher advertisement
of capital and the uncertainties of the market, the exposure among their audience than others, and some
segment member's product class purchasing volume, creative approaches are more capable than others of
and the degree of brand acceptance the brand is likely attracting attention. It might be useful to condition
to obtain after the use experience [4]. An estimate hl1 upon the type of individual involved. However,
of the brand acceptance can be obtained using repeat- the notation is already cumbersome and such exten-
buy statistics or by applying a formal brand choice sions should be obvious. Thus, the probability that
stochastic model to a group of new triers. Parfitt and individual i reads a particular issue of Good House-
Collins [10] and Aaker [2] describe two models keeping would be b j, and the probability that an
appropriate in this context and empirically demonstrate advertisement using copy approach c is exposed, given
their validity. Aaker [3] demonstrates a method to the vehicle is read, would be hcj
link brand acceptance to descriptors of market seg- The product of these two terms, b. hcj, reflect the
ments which can be used even when a stochastic model probability that individual i will be exposed to an
is applied to a single data base of "new triers" of advertisement using copy approach c in vehicle j.
a brand. The approach could be employed to obtain Let:
ws values for different segments.
The situation is similar when the goal is enticing Pcij= bijhcj,
a consumer to use the product in a new application where:
or to use a product class for the first time. In that
context, an important determining factor in ws will Pcij = the probability that individual i is exposed
be the consumer's acceptance of the new application to insertion option x.,
after he is induced to try it. Statistics reflecting the bij = the probability that individual i is exposed
time until the second or third use occasion or formal to vehicle j,
stochastic models of interpurchase time can be used hc = the probability that anyone exposed to
to predict acceptance levels. vehicle jwill be exposed to insertion option
The objective could be to change a brand image. Xcj.
In that case, the intent could be to maintain the loyalty
of existing customers and to make them less vulnerable Let z, be a random variable reflecting the number
to attacks from competing brands. With the existing of exposures received by individual i in a given
image, customers may have a projected purchasing insertion schedule being considered by the objective
pattern that reflects a decline in purchases of the function. The value of zi cannot be known for sure
advertised brand. It may be that a new image will but, knowing the exposure probabilities, it is a
be associated with a more stable projection of pur- straightforward task to obtain fci(zi), the probability
chasing patterns. The present value of the difference distribution for z i. The distribution can be calculated
would then be ws. Again, the best way to determine exactly by making certain independence assumptions
such projections is by observing the actual purchasing and applying simple probability theory. For example,
decisions of those who changed their perceptions of if there were only two insertions under consideration,
a brand. It may be that attitude theory will ultimately insertion option xc5 and insertion option Xc7, then
be helpful in providing such projections with less the probability of zero exposures would be (1 - P65) (1
demanding data requirements. - P67), the probability of one exposure would be
Pci5(1 - Pci7) + Pci7 (1 - Pc5 ), and the probabilityof
Determining Exposures to the Schedule two exposures would be ci5Pc17. Similar calculations
The insertion schedule will generate exposures that can be made for schedules of any size. However,
will have some impact upon the objective of the with large schedules, the calculations can become
advertising campaign. In this section, the exposure- costly, and it is reasonable to consider using the raw
generating aspect of the insertion schedule, which probabilities to estimate the parameters of a probability

This content downloaded from 142.150.190.39 on Sat, 19 Oct 2013 10:36:15 AM


All use subject to JSTOR Terms and Conditions
40 JOURNALOF MARKETINGRESEARCH,FEBRUARY1975

function such as the binomial that would then become THEADMOD REPETITION
FUNCTION
fci(Zi).
The binomial distribution is attractive because it a' (z;)
cs
is a flexible distribution defined on positive integers
and its parameters are easily estimated.' The use of
a unimodal distribution is perfectly reasonable when
only a single individual is involved, as it is in this
context. In most other media models, there is the
need to model the exposure probabilities across many amax Cs
individuals. When individuals are aggregated the re-
sulting probability distribution is inevitably multimod-
al. As a result it is nearly impossible to model well
with any tractable distribution. This fact motivates
the use of a population sample in ADMOD.
The Repetition Function and Forgetting
ADMOD assumes that an advertising campaign of
a specified duration is attempting to change cognitions
or precipitate a decision. The campaign will generate
a certain number of exposures for each individual z.
i, z,. The impact of these exposures on the probability
of the desired response occurring is termed a' (zi),
the repetition function. The subscript s indicates that
the repetition function will be different in general for functions may be estimated by separate and different
each segment, but identical for each segment member. techniques, it is quite possible to introduce inconsistent
The subscript c reflects the fact that the repetition definitions.
function will vary with the copy approach used. An One form of the repetition function is shown graphi-
awareness campaign may require only a few expo- cally as Curve A in the figure and algebraically as
sures, whereas an attitude-creation campaign may (1). It suggests that the probability of the deserved
require several additional exposures to be effective. response, a' s(zi), will start at an initial level, moving
The prime superscript indicates that all exposures are geometrically to a higher level with successive expo-
associated with the maximum vehicle source effect, sures:
Vcj. This last assumption will shortly be relaxed.
Care must be taken to ensure that a comparable (1) acs(zi) = amincs + amaxs(1 - Xc),
definition of exposure is used in hcj, the probability where:
of generating an advertisement exposure, and in zi.
Since the two terms appear in two different contexts, aminC5 = the repetition function value in the
and since exposure probabilities and the repetition absence of any exposures,
amax^ = the limiting repetition function value
as the number of exposures becomes
large,
'The parameters of the binomial can be estimated from its mean
and variance which are:
XCS = a parameter that reflects the rate at
which the repetition function moves
p. = mean= E
from amincs to amaxcs.
Pcij
In ADMOD there is no explicit conceptualization
of forgetting. However, the development of the repeti-
(2 = variance = pij(l -
pcij) tion function necessarily takes into account the length
of the campaign and therefore the length of time
The expressions are summed over all insertions in the schedule. between exposures. Curve B in the figure, for example,
Let -r and nn be the parameters of the binomial. The estimates
of these parameters, qrand nn, would be
could reflect the fact that too few exposures will suffer
from forgetting between them.
An implicit assumption in ADMOD is that the
pA 2 campaign is totally forgotten at its conclusion. The
A. - concept is that advertising is likely either to generate
Tr- the desired change in cognitions or decision in a rather
nn limited time period or not at all. Those that "with-
The estimate of nn needs to be constrained to be less than or stand" a concentrated effort for a relatively short
equal to the number of insertions in the schedule. time will probably withstand the same campaign if

This content downloaded from 142.150.190.39 on Sat, 19 Oct 2013 10:36:15 AM


All use subject to JSTOR Terms and Conditions
ADMOD: AN ADVERTISINGDECISIONMODEL 41

it is prolonged. Therefore, this forgetting assumption cognitive change. The vehicle source effect for vehi-
of the model may not be as extreme as it may first cles is termedv.j. It is conditionalon the copy approach
appear. It would be relatively easy to relax this used. Aaker and Brown [5], for example, found that
assumption. It could be assumed, for example, that the source effect of "prestige" and "expert" maga-
the average probability of a trial purchase during zines dependedupon whether an "image" or "reason
subsequent time periods would be fixed fractions of why" copy was used.
its value in the time period of immediate interest. In ADMOD the numberof exposures an individual
The objective function would then have a value receives from a given insertion schedule is known
componentfrom two or more time periods. probabilistically.However, it is not known exactly
Another assumptionof ADMOD is that the repeti- which insertion options in the schedule are creating
tion function is not sensitive to when the exposures the exposures. The solution is to compress the repeti-
are received during the campaign period. To relax tion functionby an amountdeterminedby the expected
this assumption, the timing of each insertion would value of the vehicle source effect, which is simply
have to be introduced into the decision, and the a weightedaveragewith those vehicles havinga higher
repetitionfunction would have to be conditioned on probability of creating exposures having the higher
the time pattern of exposures. Such steps would, of weight. The repetition function adjusted for vehicle
course, add considerabledetail to the model. source effects is denoted as aci(zi). Note the absence
Clearly,the determinationof the repetitionfunction of the primesuperscriptand the fact that this repetition
is at the heart of the model. One advantage of this function is unique to an individual:
formulation is that the dependent variable of the
repetition function is well defined, assuming the ad-
vertisingobjectives are well defined. It is the probabil-
(2) aci(Zi)= a's(zi) [ (Y cPi) ( p )
ity of a cognition change or a decision, such as a
decision to try a brand. A second advantage is that
this type of dependentvariablelends itself to laborato- The effect of (2) is to reduceboth amin sand amaxcs
of (1). The researcher might believe that the vehicle
ry study. The laboratorystudy, althoughhaving exter- source effect will not affect amaxcs but will affect
nal validity problems, does enable the researcher to
explore the repetition function in a highly controlled AC.Thus, inferiorvehicle settings will tend to increase
environment. Ray and Sawyer exposed respondents the numberof exposures requireduntil the individual
to from one to six repetitions of text advertisements reaches the amaxcslevel. It would be a simple matter
in the context of a futuristic shopping system [12]. to reformulate (2) to adjust XCs(to move it toward
Among the measures of interest were attitude and 1.0) instead of adjustingthe whole repetitionfunction.
purchase intentions. Such work can not only demon- However, this version of ADMOD uses the (2) for-
strate useful methodology,but can also lead to repeti- mulation.
tion function norms that can help those who must An InsertionSchedule's Value
estimate parameters subjectively. As Ray [11] has
observed, however, repetition functions obtained in The purpose of the objective function is to attach
the laboratoryshould, as a matter of procedure, be a value to a media insertion schedule. The value to
tested and refined, first with field studies and ulti- the advertiserof the campaignon individualin segment
s is the value of the desired result, ws, times the
mately with market tests. (See [14] for a discussion
of a specific field study.) probabilitythatthe advertisingwill stimulatethat result
to occur, aci(zi). If the number of exposures were
The Vehicle Source Effect known exactly, the expected value for one segment
The vehicle source effect reflects a vertical adjust- memberwould be ws aci(z ). The problem is that the
ment in the repetition function, the probability of exact number of exposures that an individual will
precipitatingthe desired response as a function of receive, given an insertion schedule, is not known.
What is known is the probabilitydistributionof the
exposures. The repetitionfunction is developed under number of exposures, fci(zi). Thus, the appropriate
the assumption that all exposures will be associated
with the optimalenvironment.The magazineor televi- procedure is to determine the expectation over zi of
sion programs creating the exposures are assumed wS aci(z), which yields the expected value from indi-
to enhance the advertisementto the maximumextent vidual i:
possible because of their audience involvement, pres-
tige, and other factors. All vehicle source effects are wSE aci (zi) fci(Zi)-
assumedto be 1.0 on a 0 to 1.0 scale. Actually, vehicle Zi=0

source effects will take on values between 0 and 1.0, The total expected value received from the sample
reflecting their relative ability to enhance or detract in market segment s is obtained by summing over
from the impact of the advertisement exposure on all individualsin the sample who belong to that market
the likelihood of generating the desired decision or segment:Ei.s. The additionof the factor Ns / ns explic-

This content downloaded from 142.150.190.39 on Sat, 19 Oct 2013 10:36:15 AM


All use subject to JSTOR Terms and Conditions
42 JOURNALOF MARKETINGRESEARCH,FEBRUARY1975

itly scales the results to the segment size. Summing complexity means that a Monte Carlo simulation model
across segments Is, and subtracting the costs of the is usually used only to evaluate a small number of
model insertion schedule, the final objective function proposed media plans. A heuristic to select among
is obtained. It is assumed that the insertion costs for plans is usually not part of the model.
vehicle j are kj for all vehicle j insertions-there are
no media discounts. This assumption is made entirely CONSTRAINTS
to avoid introducing more notation. The model could There is a wide variety of constraints that could
handle any type of cost function. be introduced. The number of insertion options con-
The value of a given insertion schedule, the objective sidered for a given vehicle may be limited by the
function, can thus be written:2 frequency with which the vehicle appears during the
time period, for example. However, here we will
V = aci (zj) fci (zj)
- formally introduce only the already mentioned con-
z-z WS 2 kCi straint that only one copy alternative should be con-
s s is Zi=0 J
sidered in any one schedule. Thus ADMOD is designed
where:
to select an insertion schedule that will maximize V,
c = index referencing the copy alternative, subject to the constraint that the selected schedule
i = index referencing the individual in the include only one copy alternative.
sample population, It would be possible to relax the constraint that
j = index referencing the vehicle, an insertion schedule can only include one copy
s = index referencing the market segment, alternative. If it could be assumed that two copy
Ns = the size of segment s, approaches operate independently, that the two
ns = the size of the sample from segment campaigns would be directed at different segments,
s, for example, extension would not be difficult. The
ws = the value to the firm of the consumer more interesting situation is when several copy ap-
action (i.e., a trial purchase) by a proaches are interspersed to the same audience. Con-
member of market segment s, sider the possibility of using an awareness advertising
zi = the number of exposures received by campaign first, followed by a reason-why campaign.
individual i, given the insertion sched- The problem is that now not only do the number
ule, of exposures need to be determined probabilistically,
aci(z,) = the probability that the desired con- but also the mix between the two copy approaches.
sumer action (i.e., a trial purchase) will Such a technical problem can be solved relatively
occur, given the fact that zi exposures easily. The difficult task is to obtain the appropriate
occurred, repetition function. An important function of such
fci(zi) = the probability that individual i will models as ADMOD is to identify new areas of produc-
receive exactly zi exposures, given the tive research. The measurement problems associated
insertion schedule, with this discussion are undoubtedly in that category.
= the cost of an insertion in vehicle j,
ki
Xcj = the insertion of copy alternative c into THE SELECTION HEURISTIC
vehicle j(xc = 0, 1). Because of the irregular nature of the objective
ADMOD Vs. Simulation function, it is necessary to use a heuristic to search
systematically through media insertion schedules to
Because a population sample is involved, it is easy select one with a high value. To implement the con-
to associate the ADMOD approach with Monte Carlo straint that any one schedule will include one copy
simulation such as Gensch's AD-ME-SIM [7]. Ac- alternative, the heuristic will be employed for each
tually, ADMOD provides a closed-form objective copy alternative. The one with the highest value will
function based upon expectations as to the number be the one selected.
of exposures and their impact upon the desired re- The heuristic search procedure should incrementally
sponse. In a Monte Carlo simulation, a generated set add insertions until a stop signal is reached. It should
of random numbers determines a specific exposure include the capability of periodically deleting insertions
pattern for a simulated individual. Such a simulation that may become redundant as other insertions are
removes inhibitions toward complexity and complete- added. Suitable heuristics as described in detail in
ness. However, simulation requires large samples
to counteract the variance introduced by the Monte [1, 9].
Carlo approach. The larger samples and the added The Budget Level
Normally a budget constraint provides a stopping
signal to the heuristic. However, in this formulation
2The objective function could easily be formulated to represent the budget level is included in the objective function.
return on the advertising investment instead of total profit. Thus, the stopping point is reached when there no

This content downloaded from 142.150.190.39 on Sat, 19 Oct 2013 10:36:15 AM


All use subject to JSTOR Terms and Conditions
ADMOD: AN ADVERTISINGDECISIONMODEL 43

longer exists any insertion option that will increase An index of an orientationof housewives to cooking
the value of the insertionschedule. Alternatively,the and the kitchenwas developed from the product-usage
heuristiccould requirethat the ratioof the value added variables.It was assumed that a "nonkitchen" orien-
by a marginal insertion to its marginal'cost be a tation would be associated with the relatively heavy
prespecified minimum.In any case, the budget level, use of fast-food carry-outestablishments,frozen din-
the cost of that final insertion schedule, is a program ners, frozen desserts, cookie mixes, instant coffee,
output instead of an input. Its value is given by the false eyelashes, and tennis balls, and the relatively
second term in the objective function- j kjx . It is light use of such products as wine. A "kitchen seg-
optimal(withinthe limitationsof the search heuristic) ment" and a "nonkitchensegment" of 200 each were
in that either an increase or a decrease in the budget selected on the basis of this index. The total size
level would reduce the net value of the advertising of the segments represented by this sample was
campaignto the firm. obtained, using the projective weight associated with
individuals in the data base. It was 10.3 million for
A MODEL APPLICATION the kitchen segment (Segment 1) and 11.5 million for
An advertisingdecision problemwas conceptualized the nonkitchensegment (Segment 2).
to illustrate the model and to provide a setting for The advertisementexposure probabilitiescondition-
conducting sensitivity tests. The product is a set of al on vehicle exposure, the h c terms, were obtained
dishes to be sold to homemakers. Two copy ap- using the 1973 Starch ad norms [17]. Specifically,
proaches are considered. (For description of such the seen-associated score which is the percent who
advertisements, see [5].) One is an image approach associated the advertiser with the advertisementwas
(c = 1), and the other is a detailed,logical, reason-why used. The relevant norms, obtained where possible
approach(c = 2). Thirteenvehicle alternatives,shown for women and food products, are shown as Data
in Table 1, were included. Set C in Table 1.
The basic data were suppliedby the W. R. Simmons The vehicle source effect terms were generated
Company. They include informationon media habits subjectively. A magazineimagestudy was drawnupon
and the usage levels of a wide variety of products (for more details see [5]), in which 30 women were
for over 7,000 individuals.The media informationwas asked to identify from a list of 18 women's magazines
in the form of whether the respondent was exposed those they would regard as the most (and least)
to zero, one, or two successive magazine issues or "prestigious"and the most (and least) "expert" with
television programs.Thus, the biiterms become 0.0, respect to cooking, foods, and kitchenware. The
0.5, and 1.0. Such an approachminimizesrespondent estimates are shown as Data Set A in Table 1.
recall and interpretationdifficulties. However, the use The repetitionfunction, based upon (1), has param-
of only two issues sacrifices precision and the use eters that were also subjectively estimated. In doing
of actual readership instead of a direct estimate of so, the research of Ray et al. [12, 13, 14] was drawn
probabilitiesadds variance to the result [15]. upon. The estimates are shown in Table 2 under Run

Table 1
BASIC VEHICLEDATA

Parameter vi hc
Data set A B C D E k
Copy approach(c) 1 2 1 2 1 2 1 2 1 2
Vehicle
1. American Home .60 .50 .53 .65 .16 .16 .21 .21 .24 .16 14,250
2. Cosmopolitan .46 .40 .53 .65 .24 .24 .21 .21 .36 .24 8,250
3. Family Circle .32 .85 .53 .65 .22 .22 .21 .21 .33 .22 26,900
4. Glamour .80 .35 .53 .65 .21 .21 .21 .21 .33 .21 7,350
5. Good Housekeeping .62 .88 .53 .65 .26 .26 .21 .21 .39 .26 22,765
6. House & Garden .60 .50 .53 .65 .20 .20 .21 .21 .30 .20 8,850
7. Ladies Home Jour-
nal .68 .96 .53 .65 .26 .26 .21 .21 .39 .26 29,030
8. McCall's .52 .74 .53 .65 .28 .28 .21 .21 .42 .28 30,750
9. Parents Magazine .40 .65 .53 .65 .20 .20 .21 .21 .30 .20 13,565
10. Redbook .42 .48 .53 .65 .24 .24 .21 .21 .36 .24 19,640
11. Women's Day .48 .78 .63 .65 .21 .21 .21 .21 .31 .21 26,825
12. Good Housekeeping
(1/2 page) .46 .68 .53 .65 .15 .15 .21 .21 .23 .15 14,190
13. Ladies Home Jour-
nal (1/2 page) .48 .66 .53 .65 .12 .12 .21 .21 .18 .12 14,350

This content downloaded from 142.150.190.39 on Sat, 19 Oct 2013 10:36:15 AM


All use subject to JSTOR Terms and Conditions
44 JOURNALOF MARKETINGRESEARCH,FEBRUARY1975

Table 2
MEDIAINSERTIONSCHEDULES

Run 1 2 3 4 5 6 7 8 9 10 11
Copy 1 2 1 2 1 2 1 2 2 2 2 2 2 2 1 2
Parameters Segment
vc - A A B B A A A A A A A A A A A A
hcj C C C C D D E E C C C C C C C C
X 1 .55 .50 .55 .50 .55 .50 .55 .50 .40 .60 .50 .50 .50 .50 .55 .50
A 2 .60 .55 .60 .55 .60 .55 .60 .55 .45 .65 .55 .55 .55 .55 .60 .55
amax 1 .14 .16 .14 .16 .14 .16 .14 .16 .16 .16 .16 .15 .16 .16 .14 .16
amax 2 .18 .12 .18 .12 .18 .12 .18 .12 .12 .12 .12 .15 .12 .12 .18 .12
ws 1 7 7 7 7 7 7 7 7 7 7 6 5 6 8 10 10
ws 2 5 5 5 5 5 5 5 5 5 5 6 7 4 6 8 8
1
2 4 4 1 2
3 1 1 3
4 4 5 1 7
5 2 4 1 4 2 4 4 4 4 4 4 4 4 6
Insertion 6
schedule 7 2 4 1 2 4 4 2 4 4 2 4 3 4
8 1 2 2 2 1 2 1 2 2 1 3
9
10
11
12 1 5
13 4 5
Contribution 1 66 207 50 111 52 134 145 207 275 118 178 150 30 277 277 426
from 2 92 131 61 75 76 86 185 131 175 75 157 287 35 182 153 315
segment: (1000's)
Profit (1000's) 25 100 17 40 27 77 120 100 155 44 97 141 6 164 153 315
Budget (1000's) 133 238 94 146 101 143 210 238 295 149 238 296 59 295 277 426

1. The term amin was assumed to be zero. For the the image copy but gave it a low rating when reason-
kitchen segment (1), amax, the asymptotic probability why copy was used. Thus, Glamour did not appear
was assumed to be slightly higher when reason-why with the reason-why copy although four insertions
advertising was used (0.16) than with image advertising were scheduled with the image copy.
(0.14). For the nonkitchen segment (2), amax was In Run 2, the vc.values are replaced by their average
assumed to be much larger for image advertising (0.18) values (Data Set B in Table 1). As a result, Cosmopoli-
than reason-why advertising (0.12). More than average tan, which had below-average v . values, does better.
repetition was assumed to be needed for the image Note that the budget and profit levels are lower than
advertising to the nonkitchen segment (X = 0.60) and in Run 1 due to the reduced v . values for Good
less than average repetition was assumed to be needed Housekeeping and Ladies Home CJournal.In Run 3,
for reason-why advertising directed to the kitchen the hc. terms were replaced by their average value
segment (X = 0.50). (Data ?et D in Table 1). The half-page options clearly
The basic w, terms used were $7.00 for the kitchen benefitted. However, the reduced advertisement read-
segment and $5.00 for the nonkitchen segment. The ership for Good Housekeeping, Ladies Home Journal,
assumption was that the kitchen segment will buy and McCall's (Vehicles 5, 7, and 8, respectively)
a better quality and more complete set of dishes. caused the profit to decline from Run 1.
Table 2 reports the results of 11 runs. In several In Run 4, the hcjterms for the image copy (c = 1)
of the runs, the results for both copy alternatives was increased 50% to reflect a possible hypothesis
are presented. For the first run, Copy 2 provides that simpler, less wordy copy will have greater expo-
greater profit and thus would be selected. The source sure. The result was the only run in which Copy 1
of the Copy 2 profit is mainly Segment 1 because generated larger profit than Copy 2.
of the repetition function parameters that reflect the Runs 5 and 6 showed the sensitivity of the budget
belief that reason-why advertising (c = 2) will be most to levels of X. In the remaining runs, the ws terms
effective with the kitchen segment (s = 1). The vehicle were altered, showing the sensitivity of the budget
source effect term favored Glamour (vehicle 4) for to the values of ws used. Run 11, with ws values

This content downloaded from 142.150.190.39 on Sat, 19 Oct 2013 10:36:15 AM


All use subject to JSTOR Terms and Conditions
ADMOD: AN ADVERTISINGDECISIONMODEL 45

of 10 and 8, generated a budget of more than $400,000. 5. and Philip K. Brown. "Evaluating Vehicle
The average cost for the runs on a CDC6400 was Source Effects," Journal of Advertising Research, 12
$6.00. Run 11 cost $16.00. (August 1972), 11-6.
6. Claycamp, Henry J. and Lucien E. Liddy. "Prediction
CONCLUSIONS of New Product Performance: An Analytical Ap-
proach," Journal of Marketing Research, 6 (November
The benefit of a model of this type is usually not 1969), 414-20.
in the machinery it makes available but, rather, in 7. Gensch, Dennis H. Advertising Planning. New York:
the vocabulary it creates, the structure it provides, Elsevier Scientific Publishing, 1973.
and its suggestions for future research. Hopefully, 8. Little, John D. C. and Leonard M. Lodish. "A Media
ADMOD will contribute by providing structure and Selection Model and Its Optimization by Dynamic Pro-
identifying productive research areas. However, the gramming," Industrial Management Review, 8 (Fall
model should not be regarded as a tool of the future 1966), 15-23.
that is impractical given today's state of the art. It 9. . "A Media Planning Calculus," Operations Re-
is true that measurement problems exist, particularly search, 17 (January-February 1969), 1-35.
10. Parfitt, J. H. and B. J. K. Collins. "The Use of Consumer
with respect to the determination of repetition func-
Panels for Brand-Share Predictions," Journal of Mar-
tions. But even in this difficult area, the work of
keting Research, 5 (May 1968), 131-45.
Ray and his colleagues and others have provided 11. Ray, Michael L. "The Present and Potential Linkages
practical methodologies and microtheories to guide Between the Microtheoretical Notions of Behavioral
researchers. Science and the Problems of Advertising: A Proposal
The fact is that enormous progress has been made for a Research System," paper presented at the Sympo-
during the past decade in mathematical models, in sium on Behavioral and Management Science in Market-
computer programs, and in measurement. Further, ing, University of Chicago, June 1969.
model users are becoming more sophisticated and 12. and Alan G. Sawyer. "Repetition in Media
interdependent. It is time to consider more compre- Models: A Laboratory Technique," Journal of Market-
hensive decision models as viable, practical aids to ing Research, 8 (February 1971), 20-9.
13. . "Behavioral Measurement for Marketing
decision making.
Models," Management Science, Part II, 18 (December
1971), 73-89.
REFERENCES 14. , and Edward C. Strong. "Frequency Effects
1. Aaker,David A. "A ProbabilisticApproachto Industrial Revisited," Journal of Advertising Research, 11 (Febru-
Media Selection," Journal of Advertising Research, 8 ary 1971), 14-20.
(September1968),46-54. 15. Schreiber, Robert J. "Letter to the Editor," Manage-
2 . "The New-Trier Stochastic Model of Brand ment Science, 14 (April 1968), B526-B527.
Choice," Management Science, 17 (April 1971), 435-60. 16. . "Instability in Media Exposure Habits," Jour-
3. . "A Measure of Brand Acceptance," Journal nal of Advertising Research, 14 (April 1974), 13-8.
of Marketing Research, 9 (May 1972), 160-7. 17. Starch Adnorms, 1973. Mamaroneck, N.Y.: Daniel
4. . "Toward a Normative Model of Promotional Starch & Starch, 1973.
Decision Making," ManagementScience, 19 (February 18. Starch, Daniel. Measuring Advertising Readership and
1973),593-603. Results. New York: McGraw-Hill, 1966.

This content downloaded from 142.150.190.39 on Sat, 19 Oct 2013 10:36:15 AM


All use subject to JSTOR Terms and Conditions

You might also like