|
Article Excerpt
Models of advertising response implicitly assume that the entire advertising budget is spent on disseminating one message. In practice, managers use different themes of advertising (for example, price advertisements versus product advertisements) and within each theme they employ different versions of an advertisement. In this study, we evaluate the dynamic effects of different themes of advertising that have been employed in a campaign. We develop a model that jointly considers the effects of wearout as well as that of forgetting in the context of an advertising campaign that employs five different advertising themes. We quantify the differential wearout effects across the different themes of advertising and examine the interaction effects between the different themes using a Bayesian dynamic linear model (DLM). Such a response model can help managers decide on the optimal allocation of resources across the portfolio of ads as well as better manage their scheduling. We develop a model to show how our response model parameters can be used to improve the effectiveness of advertising budget allocation across different themes. We find that a reallocation of resources across different themes according to our model results in a significant improvement in demand.
Key words: Bayesian dynamic linear models; Gibbs sampling aggregate advertising models; wearout effects; forgetting effects; copy effects; scheduling of ad copy
History: This paper was received June 30, 2005, and was with the authors 4 months for 1 revision; processed by Greg Allenby.
1. Introduction
Managers, policy makers, and researchers are interested in understanding the effect of advertising on demand. A large number of response models have been proposed in the literature linking advertising expenditures to sales or market shares. These studies have focused on the shape of the response function (linear or S-shaped), the dynamic effects of advertising (carry-over effects or wearout), and interaction effects with other marketing mix variables, consistent with the desirable properties of advertising response models in Little (1979).
In these response models, advertising expenditures are aggregated and there is an implicit assumption that advertising expenditure is spent on propagating one message or theme. In practice, firms concurrently run several themes in their advertising campaigns (for example, price and product advertisements). Each theme may have multiple versions, or executions, which are rotated over time. In effect, there is a portfolio of ads that is run in each week. Managers need to understand the individual effects as well as the interaction effects among the different versions of advertising on overall demand. Such an understanding then would allow managers to allocate their budgets over multiple themes more effectively. Most of the published research on the impact of advertising themes is categorized under copy research and these studies have employed experiments. They have focused on the effects of different copy on individual consumer attributes such as brand awareness and attitudes toward the brand. The literature that links the copy effects to sales or market share is limited to experimental studies and does not consider wearout effects (Aaker and Carmen 1982, Eastlack and Rao 1989).
The substantial research issues that we address in this study focus on understanding the effects of different themes of advertising on demand and on how to allocate a firm's resources across different themes to improve sales performance. The specific research questions pertinent to the study of the effectiveness of multiple themes in advertising are: How different are the wearout effects for the different themes of advertising? How can researchers accurately assess the magnitude of such wearout effects? What is the nature of interaction between the different themes of ads? For a given advertising budget, what is a more effective way to allocate resources across different advertising themes?
Our response model extends the model by Naik et al. (1998) to multiple themes of advertising. They estimate a dynamic model of the effect of advertising on consumer awareness, rather than on demand, using Kalman filtering methods. Their model captured decay effects in the presence and absence of advertising and found that a gap in advertising has the effect of restoring ad quality. They also separate out the effects of two different types of wearout: copy wearout and repetition wearout. However, in their paper, they study the effectiveness of a single ad campaign with a single message and do not consider multiple themes of advertising which may be scheduled by the manager for the same product category. Our study generalizes their advertising model to account for multiple thematic executions of advertising. In addition, there are a few other notable differences between the two models. Naik et al. (1998) use brand awareness as the dependent variable while we use a measure of demand which is more managerially useful. We employ a different methodology, Bayesian estimation of a dynamic linear model (DLM) as in West and Harrison (1997).
Our paper makes three contributions. First, we extend earlier research on advertising by developing a generalized model of multiple themes of advertising which are aired concurrently in a given week. Such a model can allow managers to assess the wearout effects of different themes of advertising as well as consider the interaction effects between these different themes. Second, we capture the dynamic effects of different advertising themes as a function of wearout and forgetting. In other words, the parameters of the model are time varying and we are able to capture the dynamic path of these parameters. To do this, we employ Gibbs sampling in estimating the state space model of West and Harrison (1997). Two recent papers in marketing have also employed Gibbs sampling in a DLM framework (Neelamegham and Chintagunta 2004, Van Heerde et al. 2004). Leichty et al. (2005) employed Gibbs sampling to study the dynamic development of consumer preferences in a conjoint application. Finally, we develop a model to help managers conduct a what-if scenario analysis so that they can allocate their resources over different advertising themes more effectively.
Our data are obtained from a major telecom company that is a monopolist in its category. We have data on demand in terms of calling time, prices per minute of calling, and advertising spending for each of the themes that are defined by the company. A unique aspect of our data is that advertising is measured in gross rating points (GRPs) and not in dollars. One GRP represents advertising exposure to one percent of the population that owns television sets as defined by the ACNielsen TV ratings. There are two advantages of using GRPs instead of advertising dollars. First, GRPs provide a more accurate picture of advertising input than advertising expenditures since it is not clear how much advertising exposure can be purchased for a given budget. Second, most media buying is done in terms of GRPs and managers evaluate the effectiveness of their campaigns in terms of demand generated per GRP. The company classifies its advertising into five advertising themes: call stimulation ads, product offer ads, price offer ads, reconnect ads, and reassurance ads. We restrict our attention to these five themes.
The remainder of the paper is organized as follows. In [section] 2, we provide a brief review of the relevant literature on estimation of advertising response with particular emphasis on wearout and dynamic effects. In [section] 3, we present the details of our econometric model. In [section] 4, we present the data and discuss the results of our estimation. Finally, we conclude with an overview of findings, the managerial implications, and the limitations of the study.
2. Literature Review
We present a brief overview of response models and review the literature on wearout effects and dynamic effects in greater detail. We also present arguments for studying the differences in wearout effects across multiple themes.
2.1. Response Models
In a seminal paper, Little (1979) stated that aggregate advertising response models should have the following desirable characteristics: (a) the effect of advertising should be nonlinear, (b) the models should capture the dynamic effects of wearout and forgetting, (c) models should consider the effect of competitive advertising, and (d) the ad effects could change over time due to changes in media and copy. Most of the models in the literature have been developed consistent with some, if not all, of these principles.
Early aggregate advertising response models linked advertising expenditures to sales or market share (Bass and Clarke 1972, Blattberg and Jeuland 1981, Hanssens et al. 1990) and considered the carryover effects of advertising (Bass and Leone 1983, Broadbent 1984, Clarke 1976, Srinivasan and Weir 1988). These models used distributed lag models to capture the carry-over effect. Clarke (1976) showed that the magnitude of the effect of advertising and the duration of carry-over effects depended on the data interval used. The response models were then used to appropriate advertising dollars to maximize profits in both monopoly and oligopoly contexts (Simon 1965, Nerlove and Arrow 1962, Telser 1964, Palda 1964). These models used aggregate advertising expenditures and did not consider the effect of multiple themes. A good recent review of the advertising literature is in Vakratsas and Ambler (1999).
Another stream of research discussed the shape of the advertising response function: whether it is concave or S-shaped. Wittink (1977), Rao and Miller (1975), and recently Vakratsas et al. (2005) found evidence of an S-shaped function while Simon (1969) found no evidence for the S-shape. This was an important question because the theoretical models showed that the phenomenon of pulsing in advertising was related to the S-shape of the response function (Simon 1982, Mahajan and Muller 1986, Feinberg 1992). In a recent paper, Naik et al. (1998) show that pulsing can occur due to ad copy wearout, while Bronnenberg (1998) shows that pulsing can also occur in the context of a monopolist facing a Markovian sales response function.
2.2. Wearin and Wearout
Wearin refers to the positive effect on consumers who are exposed to an ad (Pechmann and Stewart 1990). The term wearout refers to the decay in advertising quality of an ad over time (Grass and Wallace 1969, Strong 1972, Calder and Sternthal 1980, Simon 1982). An ad is worn out if it either does not have any significant effect on consumers or has a negative effect. Both wearin and wearout effects depend on factors such as whether the ad was based on an emotional appeal or a rational appeal, whether the persuasion in the message was strong or weak, whether consumers were motivated or not to process the ad, and whether the level of competitive ads was high or low (Pechmann and Stewart 1990). The wearout effects may also depend on the change in ad copy. This is based on research that suggests that variations in copy improve the effectiveness or, specifically, recall of ads (Grass and Wallace 1969).
In a series of experiments to study of the effects of repetition of ads, Ray and Sawyer (1971a, b) found that the response functions for repetition varied across different measures (e.g., recall or intention), segments, brands, and type of advertising. They also studied the effect of different messages on repetition functions. Relevant to our study, they found that "grabber" ads were less effective over repetitions (i.e., had higher wearout) than "nongrabber" ads. MacInnis et al. (2002) find evidence of a significant positive relationship between ad repetition and sales when emotional ads are employed, but not for rational ads. They argue that one possible explanation of the above finding is that emotional ads have less rapid wearout.
Naik et al. (1998) model two sources of wearout--repetition wearout and copy wearout. When a customer is exposed to ads repeatedly, she can become bored, irritated, or simply lose interest as the benefits of processing the ad are perceived to be worthless (Berlyne 1970, Greyser 1973, Weilbacher 1970). This...
|
