CHAPTER 1

THE USE OF MODELS IN MARKETING

Introduction

The personal computer has affected the entire world of model building. Just a few years ago we were visiting with a mid-level marketing manager marketing models. He told us of his experience with marketing models, then added, "Models are used only by the megabuck consumer-products companies that can afford the high risk and costs of model development, data collection, and analysis."

His was a widely-held view in those days before personal computers, and was supported in the history of mainframe-based marketing models. What are the reasons why?

First, in some cases model development costs have been outrageous, with no payback whatever. In the early 1970's, LaButz of Canada spent over $50,000 to develop a model which would determine the effects of advertising and sales force expenses on market share, but it was never used. A few years earlier, Amstutz's Sprinter Mod II market simulation had cost over $200,000 to develop and it was never used. Today, many large corporations contract for large research projects that do not answer the right question, and are not even directed at the right problem.

Second, the development time for a model was often too long for it to be useful. Many mainframe models have taken years of development to simulate market conditions which had changed long before the model was finished.

Third, costs of collecting data to be used as input to a model have frequently been huge. In 1999, IBM will spend nearly $2,000,000 to model world wide sales potential on a country by country basis.

And finally but probably most important, many of those mainframe models were almost incomprehensible to anyone but their developer. Amstutz's model was staggeringly complex, having more than 500 equations. Another well-publicized 1960's model was so large and complex that running it took more than $75,000 in computer time.

The personal computer changed all this. It has encouraged the development of models that are smaller, cheaper to develop and cheaper to run. These models are more understandable, appeal to a larger audience and, if not always "friendly," are at least more companionable than the impersonal, and even intimidating, mainframe models of a decade ago.

Model development software, such as Excel, is also becoming plentiful. These programs, or executable versions may ultimately standardize much model building. This standardization will help non-model-builders feel more comfortable interacting directly with the model themselves, instead of relying on their operations research people to do it for them. Undoubtedly, this interaction will increase management's comfort zone for models. This is likely to increase the use, confidence in, and understanding of the models.

A model is simply a representation. A map, which represents the countryside, is a model, and so is a graph which represents a company's sales over time. In this sense, all our perceptions of reality are "models" since they are only perceptions and not reality itself. Our language is a model, since words are only representations of objects or ideas. Similarly, our numbering system is a model by which quantities of objects or ideas can be represented. These are trivial, but real, distinctions between models and reality.

Models characterize either what currently exists in fact, or what might exist in the future. Marketing models might depict such operations as an existing product distribution system; a consumer's value structure, consumer preference modeling for product choices, or the effects of advertising on consumer awareness, knowledge, attitudes, or intention to purchase.

The purpose of a model is typically to provide the manager with a guide for evaluating the effect of a set of input variables. For example, a design engineer for General Motors might model the aerodynamics of a different body design features to determine their impact on airflow and fuel economy. Or a civil or hydraulics engineer designing a dam to hold back a reservoir 200 feet deep might use a model to determine the concrete density and materials necessary to withstand the water pressure at each depth level.

Model builders in the applied or physical sciences often use readily measurable and quantifiable input data. But marketing model builders typically often have only very loose and imprecise inputs. These inputs are often estimates of such unknowns as new product demand, advertising effectiveness, advertising efficiency, optimal sales force allocation, results of a product positioning option, or estimates of pricing strategy.

Modeling is simply a means of depicting, examining, specifying, or operationalizing relationships. This suggests, then, that management decision making is inherently a modeling process, in which the manager models the environment and competitive strategy.

Companies, in their evolution toward becoming "model-building" firms, will usually include four processes or stages in marketing decision making. Most are stuck in the first or second stage. These processes are:

1. Making decisions through managerial experience, where the years of integrative executive experience guides managers to the proper course of action.

2. Using market data or other facts derived from marketing research along with the above, where marketing research may describe the attitudes, perceptions, preferences, product usage by market segments, or product use scenarios. Often data is collected to examine relationships, such as the relationship between use situation and usage rate, but the underlying assumptions of a consumer behavior model are not readily recognized.

3. Developing accepted organizational standards for decision making, where the organization explicitly recognizes acceptable alternatives in planning and implementing marketing strategy.

4. Conducting explicit model building, where model building objectives, constructs (variables), and relationships between constructs are specified in detail. The relationships between the constructs are specified in detail using theory and management judgement, and the evaluation of the model is based on a conceptual framework.

An obvious question to be asked is, "Why don't more managers do explicit model building?" MIT professor John Little helped answer this by identifying several reasons why management science models are not used more often in marketing decision making. The reasons why models are not used in marketing include:

2. Managers do not have time or desire to spend the effort to become knowledgeable in the area

To overcome these problems, Little gave thought to what characterizes "good" models -- models that will be used by management. These characteristics provide a litmus test for evaluating the "goodness", or managerial acceptability of all model building efforts within the social sciences. These are:

2. It must be robust -- its results must not vary wildly with small changes to the input data,

It is easy to appreciate the complexity of the marketing decision making process when the difficulties of creating and parameterizing the models are understood.

Total market effort is the mixture of activities the firm undertakes to meet its objectives (sales, profitability, donations, or whatever). Marketing activities include, of course, marketing mix variables such as the types and levels of pricing, promotional activities, distribution strategies, and product development strategies. Interactions are known to exist between these functions, but the duration, relevant range, and extent is unknown.

In practice, the mathematical, and even directional, relationships between the marketing mix variables and market response are not usually known. Consider the "Be a Pepper" advertisements aired by Dr. Pepper. Their upbeat music and image resulted in high advertising awareness, but Dr. Pepper market share steadily declined. Although it is generally thought that the campaign was unsuccessful because it provided no product benefit focus, little is actually known as to why (or even "whether") the campaign was unsuccessful in selling the product.

Contributing to this complexity are many other factors that can influence the implementation or interpretation of the marketing effort. Specifically, these variables may include:

Market segments which differ in their response to the marketing mix. Segments may be defined according to geographic areas, attitudes held, benefits sought, characteristic life styles, demographic characteristics and so on.

Multiple products which exist or are offered to the consumer. The allocation of resources, such as sales effort, to one product may influence sales of other products within the line.

Conflicting objectives which present difficulties when setting goals and developing implementations for product offerings.

Functional Areas Interaction. For example, inventory management objectives of limited stock may conflict with sales management objectives of no waiting time and abundant stock. Pricing and advertising strategies may influence inventory, employee scheduling, cash flows, production scheduling, and financial management.

Competitive Effects. The nature and impact of competitive actions is typically unknown.

Delayed Response To Marketing Efforts. The effects of advertising campaigns or selling efforts may carry over substantial periods of time and even last for years.

In spite of the complexities of model building, models offer significant benefits. Perhaps most important is the sensitization process they help develop. Managers must gather and enter the input data, then evaluate the model's outputs. They re-evaluate the input data, modify it, and watch the effect on the model's outputs. By using models in this way, managers become sensitized to recognize and evaluate the elements that are important to making an appropriate decision.

Modeling activities also force both the manager and the researcher to be critical (and parsimonious) in evaluating the impact of variables that could explain the process. They begin to question assumptions about supposedly influential variables, and begin to discover important new variables.

Finally, the manager is forced, as part of the variable selection process, to consider the relationships between variables. He or she begins to be aware of interactions between them, and they may begin to recognize that a symbiotic, synergistic, relationship exists between many marketing activities and effects.

Models are usually developed to increase our understanding, prediction, and control of real world events. They can further be described as being either descriptive, predictive, or normative. This distinction between model types is important since it delineates the purpose or use of the model in management decision making.

Descriptive Models. Some models are intended merely to describe a real-world process. These descriptive models provide a characterization of the nature and workings of the modeled process. For example, a simple accounting model of,

describes a large number of events, including sales, purchases of materials, expenditures for labor and overhead, and so on. In marketing, descriptive models show market information in a way that identifies the components of a market or system. In consumer research, this system might be the consumer information search process that occurs within the overall process of consumer behavior. Descriptive models make statements that certain phenomena are produced by other factors. For example, we might predict sales with the factors,

Factor 1 = size of the customer's business
Factor 2 = number of salespersons
.
.
.
Factor k = Overall health of the economy

Descriptive models are often used to depict large systems, because the large number of variables and interactions make other types of models impractical. The Engel, Blackwell, and Kollat model provides one example (Figure 1-1). In addition to providing graphic detail, descriptive models also serve to develop descriptive structural hypothesis that specify the causal relationships between variables. This specification may provide for better measurement of variable relationships. Descriptive models provide a structured basis for discussion, analysis and understanding of a problem.

Predictive Models. Predictive models are usually more complex than descriptive ones. Besides describing objectives and events, they are designed to predict future events. A sales forecasting model, for example, is designed to predict the result of the purchase decisions by a firm's customers.

As another example, we might develop a time-series regression model to predict the impact of advertising reach and frequency on advertising effectiveness. Predictive models make inferences about the underlying structure of variables that govern a phenomenon. Understanding the predictive relationships between sets of variables is beneficial because these relationships can be used to forecast future events, making them invaluable in planning and control. Predictive models can be useful in validating descriptive models and in determining sensitivity of predictions to variables in the model.

Extending our sales example from above, predictive models provide a more precise explanation by specifying how factors 1...k interact in manner x1...xk to predict sales.

Normative Models. Normative (or, control) models are the most difficult models to construct since these models not only describe and predict, but provide direction about the proper course of action. If our sales forecasting model includes the prices we can charge for our products, we might be able to make a decision about which price we should charge in the future. The normative models tell us what should be done. They assess the implications of decisions and provide solutions to problems.

Once again using our sales example, normative models extend to provide the reasons for the predictive relationship, and may be expressed as: "Factors 1...k interact in manner x1...xk for reasons w1...wk to predict sales."

Iconic and Symbolic Models. Management science models may assume a variety of forms, as shown by Zaltman, Pinson and Anglemar (1970), who identify two basic types of models: iconic and symbolic.

Iconic (which means "image") models are like reality in the sense that they look like reality. Photographs, maps, architectural miniatures, and rough layouts of advertisements are all iconic models.

In contrast to iconic models, symbolic models do not look like reality, but emulate reality in other ways. They include either (a) verbal, (b) schematic, or (c) mathematical forms that describe a specific process. For example, consider several simple models of these types which describe the relationship between consumer attitudes and intention to purchase a brand:

Verbal: To discover consumers' purchase intention for a brand, measure their evaluation of each of its attributes and then add these evaluations together.

Purchase Intention = Evaluation of Attitude 1 + Evaluation of Attitude 2 + Evaluation of Attitude 3

Marketing models, like all management science models, are developed through either inductive or deductive logic which leads to generalization about market behavior. From these generalizations, sets of premises or theories are developed. These lead to sets of relationships which constitute marketing models. This process is depicted graphically in Figure 1-2.

1. Induction moves from the particular to the general--using particular observations to build a general model. Deduction moves from the general to the particular--using inference to establish the particulars from a general model.

Since models are intended to represent reality, a fundamental issue is the convergence between the model and the reality it is designed to represent. We might hope that a model would confidently represent reality on all significant issues.

Model builders should measure the quality of their models against the criteria of validity and utility. Validity refers to the accuracy of the model in describing and predicting reality. A sales forecasting model which does not forecast sales with reasonable accuracy is probably worse than no sales forecasting model at all.

Yet, a major obstacle to the adoption of early marketing models was not caused by their being incomplete, but because they were too complete. Their developers, in trying achieve validity, were led to include so many variables (with correspondingly difficult data collection problems) that the basic structure of the model was buried, input data costs were escalated, and confidence in them was lost. The models may have been reasonably valid, but they had little utility because they slowed down the decision making and increased its cost.

The completeness and validity required in a model depends on the accuracy required in the results. Model users should not expect a model to make their decisions for them. The output from a model should typically be taken as one additional piece of information to help the managers make their own decisions.

Given this perspective, models can be excused from not representing reality perfectly and, in fact, will probably benefit from it if they are simple enough for the managers to understand and deal with. Clearly, though, models used to help make hundred-million-dollar decisions should be more complete than those used to make hundred-dollar decisions.

Too, the model to be used depends on the model's purpose. A simplified model does not preclude its user from considering other factors not included in it. We measure the value of a model on the basis of its efficiency in helping us arrive at a decision. If we arrive at better decisions more easily without the model, then the model is inefficient. In fact, models should be used only if they can help us arrive at results faster, with less expense, or with more validity.

The building blocks for models are concepts, constructs, variables, operational definitions, and propositions. Let us take a brief look at each of these.

Concepts and Constructs. A concept is an abstraction formed by generalization about particulars. "Mass", "strength", and "love" are all concepts, as well as "advertising effectiveness", "consumer attitude", and "price elasticity." Constructs are also concepts, but they are the conscious inventions of researchers to be used for a special research purpose. When we refer to "consumer attitude" as a construct, we are suggesting not only that it exists as a concept, but that it can be observed and measured, and is related to other constructs.

Variables. Model builders loosely call the constructs they study "variables." Variables are constructs that can be measured and quantified. A variable takes on different values (a variable varies). Treated as a variable, "consumer attitudes" suggests a some form of measurement which has produced data that represents consumer attitudes.

Cause and Effect. Relationships between variables usually involve cause and effect. For example, if we turn up the heat under a pan of water, the water will boil. We conclude that the heat caused the water to boil. Or if we increase our advertising expenditures we might see our sales increase. We can conclude that the advertising caused the sales to increase.

Or can we? A model builder would argue that we can not. To establish a cause-effect relationship, three conditions must be met:

1. Concommitant variation is necessary. If variable X has an effect upon variable Y, movement of the two variables must be associated with each other. Increasing X should increase (or decrease, or otherwise change) Y.

2. Proper time order of effects. If we want to believe that variable X has a causal effect on variable Y, then X should precede Y in time. If an increase in advertising expenditures is causing a sales increase, then the advertising increase should precede the sales increase. (Have you ever noticed ... the more firefighters, the larger the fire?)

3. Absence of competing explanations. To be convinced that X is causing Y, we must be convinced that other variables are not responsible for the change in Y. If these are not controlled, we may ultimately discover that X is be causing Z, and Z is causing Y, or that both X and Y are caused by Z, etc. For example, our increase in advertising might have coincided with a price increase in our competitor's product and it was this price change that increased our sales.

Using the above three points to establish that the advertising increase caused the sales increase, we might argue that movement of those two variables is associated, and even that they have the proper time order of events. But we might be hard pressed to establish that no other variables are accountable for the sales increase. Those variables must be controlled, or at least monitored, before we could comfortably draw such a conclusion.

Operational Definitions. We can talk about "consumer attitudes" as if we know what it means, but the term makes no sense at all until we define it in a specific, measurable way. An operational definition assigns meaning to a variable by specifying how it is to be measured. It is a set of instructions about how we are going to treat a variable. For example, the variable "height" could be operationally defined in a number of different ways:

--as measured, in inches, with a precision ruler, with the person wearing shoes,

As another example, suppose we were interested in "purchase intentions" for Brand X window cleaner. We might operationally define the variable as the answer to the following question:

Please indicate your intention to purchase Brand X window cleaner the next time you purchase a window cleaning product:

We could have chosen to operationally define "purchase intention" in other ways. For example, we could have used the concepts of "attitudes" and "beliefs," which have been shown to predict purchase intention, and used a simple mathematical model:

Propositions. A proposition is a statement of the relationships between variables. Propositions require an explicit statement of the relationship between variables, including both the variables influencing the relationship and the form of the relationship. It is not enough to simply state that the concept "sales" is a function of the concept "advertising." More appropriately, any intervening variables must be specified, along with the relevant ranges for the effect, including saturation and threshold effects, and the symbolic form of the relationship.

A proposition is quite close, you can see, to a model. It is produced by linking propositions together in a way that gives us a meaningful explanation for a system or process.

As we discussed above, the modeling process is helpful to managers because it sensitizes them to variables that are important in explaining a process. Modeling forces both managers and researchers to scrutinize and select appropriate variables, and to consider the relationships between them.

A checklist to serve as a guide in evaluating this model-building process can be helpful. Some important questions to be asked (and answered) are:

[ ] Is there concensus as to which concepts are relevant in explaining the problem?

[ ] Is the concept specific enough to be operationalized reliably and with validity?

If the model does not meet the relevant criteria, it should probably be revised. Concept definitions may be made more precise, variables may be added or deleted, operational definitions may be tested for validity, mathematical forms may be revised, and assumptions may be strengthened or weakened.

In this chapter we have reviewed some of the fundamental concepts behind models. In the next few chapters we will get much more focused and help you with an understanding of specific building blocks to be able to construct your own models.

4. What extraneous variables may iconic implementation or interpretation of marketing models?

7. Symbolic models include verbal, schematic, or mathematical terms that describe a specific process. Identify the verbal, schematic, and mathematical forms of two different processes.

9. Define completeness and validity as they apply to model building and identify the relationship between the two criteria for measuring quality of models.

11. Identify the relationship between a variable, the operational definition of a variable, and a proposition.

Preface	Preface
Chapter 1	The Use of Models in Marketing
Chapter 2	An Excel Spreadsheet Primer
Chapter 3	Advanced Commands: Graphics and Database
Appendix 1 Assignment 1	Functions for Finance, Logic, Statistics Assignment1: Mathematical Functions
Chapter 4	Modeling Marketing Phenomenon
Chapter 5	Segmentation Concepts and Models
Case 5-1	Cougar Visa: Developing a Means-End Chain
Chapter 6	Product Planning Models
Technical Notes Case 6-1 Case 6-2 Case 6-3 Case 6-4	Product Planning Technical Notes AMF, Inc.: New Product Trial MooSoda I: Trial-Repurchase Air Jordan: Purchase - Repeat Quite Write, Inc.: Product Portfollio Analysis
Chapter 7	Sales Management Models
Case 7-1	SALTFLATS, INC., Sales Force Allocation Model
Chapter 8	Distribution and Production Models
Case 8-1 Case 8-2 Case 8-3	RAW Manufacturing, EOQ Model THE AZTEC COPY CENTER, EOQ Problem Set Acme Filter Company, EOQ Problem Set
Chapter 9	Advertising Models
Case 9-1 Case 9-2 Case 9-3 Appendix 2	Rivergrove Out-Patient Clinic: Media Planning Guthrie Gourmet Foods: Media Planning MooSoda II: Advertising Budgeting ADBUDG: Advertising and Budgeting Model