Direct methods of demand estimation are techniques that do not involve regres- sion analysis. After reading about some of these direct methods of estimation, you may get the impression that direct estimation techniques are quite simple and straightforward. This is far from correct. Many of the techniques used in mak- ing direct estimates of demand are quite sophisticated and require a great deal of experience and expertise to estimate demand accurately. This section is designed only to give an overview of some of the methods that can be used and is not meant to teach you how to make these types of estimates. Such instruction is left to more advanced marketing courses.
Consumer Interviews
Since consumers themselves should be the most knowledgeable about their indi- vidual demand functions for particular commodities, the most straightforward
method of demand estimation would be simply to ask potential buyers how much of the commodity they would buy at different prices with alternative values for the determinants of demand (i.e., the price of substitute commodities, the price of complementary commodities, and so on). At the simplest level, this might be accomplished by stopping shoppers and asking them how much of the prod- uct they would buy at various prices. At a more sophisticated level, this proce- dure would entail administering detailed questionnaires to a selected sample of the population by professional interviewers. While this procedure appears very simple, there are several substantial problems. Among these problems are (1) the selection of a representative sample, (2) response bias, and (3) the inability of the respondent to answer accurately. Let’s look at each of these problems briefly.
When selecting a sample of the population for a survey, the resulting demand estimation is reliable only if the survey uses a representative sample. A representative sample has the same characteristics as the population as a whole.
A representative sample is typically obtained by randomly selecting members for the sample from the general population. For example, if 52 percent of the popula- tion is female, and if 35 percent have annual incomes over $65,000, then a repre- sentative sample should have approximately 52 percent females and 35 percent persons with incomes over $65,000. In actuality, it is very difficult to obtain a truly representative sample.
A classic illustration of what can happen if the sample is not random occurred during the presidential campaign of 1948. A survey was performed that predicted an overwhelming victory for Thomas Dewey. In fact, Harry Truman won the elec- tion. The problem with the survey was that the sample was drawn from the sub- scription list of a particular magazine. The subscribers were not representative of the entire population of the United States; they were instead a subgroup of the voting population and had some important characteristics in common. Thus, the biased sample led to biased results. In 1936, in a similar but less celebrated election forecast error, a popular magazine predicted Franklin Roosevelt would lose the election, but it was wrong because the pollsters used a telephone survey and only wealthy people were able to afford phones at that time. Today, election forecasting has become so accurate—in large part due to the advanced sampling techniques now employed by pollsters—that television networks are not allowed to project winners until the polls are all closed on election day.
Another example of a biased sample yielding misleading results occurred at a home-building convention, during which Owens-Corning Fiberglass Corpora- tion commissioned a survey to determine the industry’s outlook for future sales.
The results were startling. The survey indicated that builders were planning to increase housing starts by an amazing 30 percent. When asked to interpret the bullish forecast, Michael Sumichrast, chief economist for the National Association of Home Builders, replied that “it shows when you ask stupid questions, you get stupid answers.” Apparently, the survey did not use a representative sample. As it turns out, the survey was taken only among the builders who attended the con- vention, and these builders tend to be the larger and more aggressive companies that would naturally be more bullish in their outlook.
representative sample A sample, usually drawn randomly, that has characteristics that accurately reflect the population as a whole.
A response bias can result simply from the fact that those interviewed are giving hypothetical answers to hypothetical questions. The answers do not necessarily reflect what the individual will do; rather, they may reflect intentions or desires. More importantly, however, the responses may be biased by the manner in which the question is asked. In many cases, the questions may be such that the respondents give what they view as a more socially acceptable response, rather than reveal their true preferences.
One example of response bias is found in a survey by an automobile manufacturer taken many years ago—during a time of cheap gasoline. Potential consumers were asked if they would be interested in buying small, economical cars (i.e., fuel-efficient cars) that were not flashy, fast, or showy. A large number of people said they would indeed buy such a car. On the basis of this survey, the manufacturer introduced a small, fuel-efficient car—with disastrous results.
Perhaps had the respondents—who indicated that they wanted economy cars—
been asked whether their neighbors would buy such cars, they might have provided more valid responses. It’s easier to say that your neighbor wants a flashy car than to admit that you do. The point is that the wrong question was asked. The way the question was asked induced a response bias.
Past surveys by food manufacturers have yielded bad results because of response bias. The food industry has a lot riding on the claims that people make about what they eat. Food companies have, in the past, conducted their market research by asking people what they eat. On the basis of the results of these surveys, the food manufacturers would develop new products. But, as noted in The Wall Street Journal, there is one big problem: “People don’t always tell the truth.”1 As Harry Balzer, the vice president of a market research firm, said:
“Nobody likes to admit he likes junk food.” In other words, a response bias exists in such surveys. Instead of answering truthfully, a consumer is likely to give a socially acceptable answer. Asking a sweets-eater how many Twinkies he eats “is like asking an alcoholic if he drinks much.”
Finally, it is quite possible that the respondent is simply unable to answer accurately the question posed. Conceptually, the firm performing the survey may want to know about the elasticity of demand for its products. Thus, the firm is interested in the response of consumers to incremental changes in price and some other variable. For example, the firm needs to know how the consumers would react to such things as a 1, 2, or 3 percent increase (or decrease) in price or a 5 percent increase (decrease) in advertising expenditures. Obviously, most people interviewed are not able to answer such questions precisely.
Although the survey technique is plagued with these inherent difficulties, it can still be an extremely valuable tool for a manager to use in quantifying demand.
The trick in doing a survey is to avoid the pitfalls, and, as the following discussion indicates, that can be done.
response bias The difference between the response given by an individual to a hypothetical question and the action the individual takes when the situation actually occurs.
1Betsy Morris, “Study to Detect True Eating Habits Finds Junk-Food Fans in Health-Food Ranks,”
The Wall Street Journal, February 3, 1984.
Market Studies and Experiments
A somewhat more expensive and difficult technique for estimating demand and de- mand elasticity is the controlled market study or experiment. The analyst attempts to hold everything constant during the study except for the price of the good.
Those carrying out such market studies normally display the products in several different stores, generally in areas with different characteristics, over a period of time. They make certain that there are always sufficient amounts available in every store at each price to satisfy demand. In this way the effect of changes in supply is removed. There is generally no advertising. During the ex- periment period, price is changed in relatively small increments over a range, and sales are recorded at each price. In this way, many of the effects of changes in other things can be removed, and a reasonable approximation of the actual de- mand curve can be estimated.
An example of such an approach is a study conducted by M&M/Mars using 150 stores over a 12-month period to determine the optimal weights for its candy bars.2 Instead of altering the price from store to store, the company kept price constant and altered the size of the product. As the director of sales development reported, in stores where the size was increased, “sales went up 20 percent to 30 percent almost overnight.” As a result, M&M/Mars decided to change much of its product line.
A relatively new technique for estimating demand is the use of experiments performed in a laboratory or in the field. Such experiments are a compromise between market studies and surveys. In some types of laboratory experiments, volunteers are paid to simulate actual buying conditions without going through real markets. Volunteer consumers are given money to go on simulated market trips. The experimenter changes relative prices between trips. After many shopping trips by many consumers an approximation of demand is obtained. The volunteers have the incentive to act as though they are really shopping because there is a probability that they may keep their purchases.
Going a step further, some economists have conducted experiments about consumer behavior—with the help of psychologists—in mental institutions and in drug centers, by setting up token economies (which incidentally are supposed to have therapeutic value). Patients receive tokens for jobs performed. They can exchange these tokens for goods and services. The experimenters can change prices and incomes and thus generate demand curves, the properties of which are compared with the theoretical properties of such curves.
The experimental approach to estimating the demand for products has rapidly moved out of the laboratories to the real-world applications more of interest to Wall Street and Main Street. The rapid growth of microcomputers and cable television systems has made possible market experiments that could only have been dreamed of a decade ago.
Now try Technical Problem 1.
2See John Koten, “Why Do Hot Dogs Come in Packs of 10 and Buns in 8s or 12s?” The Wall Street Journal, September 21, 1984.