Predicting Temporal and Spatial Patterns of Aggregate Consumer Demand

Donald H. Granbois, Indiana University
ABSTRACT - Two papers, each presenting results of tests of predictive models of aggregate consumer demand, are discussed and compared. Each paper contributes to our evaluation of the predictiveness of such models, but important questions remain about the "true" nature of the consumer behavior such models attempt to predict.
[ to cite ]:
Donald H. Granbois (1984) ,"Predicting Temporal and Spatial Patterns of Aggregate Consumer Demand", in NA - Advances in Consumer Research Volume 11, eds. Thomas C. Kinnear, Provo, UT : Association for Consumer Research, Pages: 396-399.

Advances in Consumer Research Volume 11, 1984      Pages 396-399

PREDICTING TEMPORAL AND SPATIAL PATTERNS OF AGGREGATE CONSUMER DEMAND

Donald H. Granbois, Indiana University

ABSTRACT -

Two papers, each presenting results of tests of predictive models of aggregate consumer demand, are discussed and compared. Each paper contributes to our evaluation of the predictiveness of such models, but important questions remain about the "true" nature of the consumer behavior such models attempt to predict.

INTRODUCTION

A noticeable pattern in recent consumer research has been the increasing tendency of researchers to study micro-phenomena. The two preceding papers provide a refreshing change of pace in that both represent attempts to improve our ability to predict patterns of behavior among large aggregates of consumers. These papers report on recent work in two disciplines (behavioral economic geography and macroeconomics) infrequently represented in the consumer research literature. Both streams of research (predicting patronage among urban customers and predicting period-to-period fluctuations in aggregate auto demand) touch on circumstances in which situational variables are important determinants of behavior Along with the contribution these papers make by calling our attention to the role of situational variables, each paper raises fundamental conceptual issues such as how best to advance our understanding of the real meaning of variables in formal models whose founders stopped short of full conceptual development, problems in variable specification, and questions of the appropriate range of products and/or consumer segments to which such models should appropriately be applied. These issues are hardly confined to the two streams of research represented here.

COMMON ASPECTS OF THE TWO MODELS

Both the spatial interaction model first proposed by Huff (1963) and the psychological economics model of Katona (1951) apply to "high involvement" consumer behavior, where extended decision making and deliberation over alternatives would be predicted. As originally proposed, the Huff model has been considered best suited for predicting the allocation of patronage among shopping centers by customers seeking shopping goods (especially clothing). Katona's theory concerns durable goods and housing demand. It is therefore remarkable that both models arc quite simple (each includes only two major variables) and that early empirical tests of the predictive value of the two were sufficiently promising to encourage substantial research over an extended period. Perhaps the early predictive success claimed for these models inhibited early attention to the means underlying the operational measures devised to permit empirical testing.

Unlike currently popular models from psychology, which encourage research with somewhat obscure policy implications, the early promise of empirical findings with significant policy implications undoubtedly spurred on tests of both the Huff and Katona models. estimating potential demand for proposed shopping center developments and urban planning and traffic analysis are problems well-addressed by models guiding research into patronage behavior. The volatility of durable goods demand presents a vexing problem for both public and private policy makers and research helping to predict turning points in demand would provide a welcomed input.

Neither the Huff nor the Katona model specifics variables susceptible to researcher manipulation, nor can either be tested under conditions of experimental control in either laboratory or field settings. About all one can do is to make predictions about consumers' likely response to a policy change affecting the consumer behavior in question, implement the policy (build a new shopping center or take action to stimulate auto demand, for example), and attempt to assess results. In such settings, it is usually difficult to determine what would have happened in the absence of policy implementation and to attribute results to the several uncontrollable factors that may have contributed to the behavioral resPOnse.

Perhaps because of the constraints they face, researchers in these fields have tended to resort to model-fitting research, in which formal models have been tested for their ability to conform to measures gathered from the field representing the dependent variable of interest. A temptation in such research (which neither spatial interaction model testers nor macroeconomic model builders have successfully resisted) is to manipulate and vary either aspects of model structure or variables included in the test so as to more closely fit the model's "predictions" to the data Without much consideration of the conceptual meaning such manipulations represent.

TESTING THE SPATIAL INTERACTION MODEL

The test reported by Ellinger and Lindquist (1983) is not really empirical in the usual sense, but rather was intended to explore the shifting patronage patterns between two hypothetical competing shopping centers (of specified size and location) as exponents attached to the distance and attractiveness components of the Huff model were mathematically varied. The resulting patterns of patronage therefore could provide a basis for evaluating empirical data on actual patronage behavior, which might bc gathered in a real setting served by two centers having similar size and distance relationships. The purpose of the manipulation, though, appears to have been more conceptual in nature -- that is, to see if the predicted patterns of patronage appeared to conform with what would be predicted if "distance" reflected a classification of goods explanation and "attractiveness" signified a multiple goods' effect.

The two-exponent version of the model, as implemented by Ellinger and Lindquist, is attributed by them to Gerry R. Wilson, although the source where the model was first presented is not known. Despite the fact that economic geographers have apparently tested this version of the model with unsatisfactory results (Shepherd and Thomas 1980, p. 25), two thorough reviews of the literature (Shepherd and Thomas 1980, and Kivell and Shaw 1980) fail to discuss the meaning of the exponent applied to the attraction variable beyond the notion that the larger shopping center offers greater "possibilities of comparison shopping and multi-purpose shopping trips" (Shepherd and Thomas 1980. p. 25).

The conceptual advancement in attributing meaning to the model's two variables seems modest in that discussion of patronage all along has emphasized the notion that convenience goods are quite likely to be purchased at the nearest suitable location and that larger assortments are sought when consumers seek (1) to cluster the purchase of unlike goods on a single shopping trip to achieve transactional efficiency or (2) to compare and evaluate a large number of alternative offerings when neither what exactly was preferred nor what exactly was available could be learned before undertaking the shopping trip. The authors' contribution has been to put these rather intuitive notions into a precise framework.

The framework does little, however, to quell doubts about the conceptual soundness or completeness of the scheme. Recent attempts to develop goods classifications have focused on concepts such as involvement, which perhaps could serve to update and expand the early classification schemes used in the paper. More importantly, research on deliberation and shopping patterns for major purchases, at least, indicate far more variability in behavior than simple goods classifications would predict. It has been behavioral geographers, not marketing and consumer researchers who have extended application of spatial interaction models from major purchases ("shopping goods") to all goods, and one suspects the careful study of shopping behavior research would introduce strong reservations about a single model applied so generally to all products. Concern about calibrating the variables (whether to use square footage of shopping goods space or some other measure of attractiveness, and whether to use mileage or driving time to measure distance has obscured the more fundamental issues about the meaning of these concepts --issues that are appropriately raised by Ellinger and Lindquist. What seems to be needed, though, is some further rather fundamental conceptual thinking about factors determining patronage behavior (and the purchasing and shopping behavior it supports).

Several specific aspects of shopping probably should be considered in this conceptual development:

(1) While specific environmental situational variables are undoubtedly of major importance (and spatial interaction models are far more influential than typical consumer behavior models in providing this emphasis) consumer variables are also significant. Individual differences measures may usefully be explored.

(2) Many shopping trips do not originate from the home (as the model implies), nor is shopping always the primary purpose of trips on which shopping occurs.

(3) Multi-center trips are not uncommon. and goods reflecting two or more traditional classifications may well be purchased on a single trip.

(4) Shopping is not always entirely (or even largely) instrumental behavior. Besides the basic need to complete transactions and to gather information, shoppers may receive additional gratifications that may importantly influence patronage behavior.

(5) Transactional efficiency among customers is fostered by assortment width or variety (up to some optimum point that may vary both among consumers and with specific combinations of unlike items clustered together). However, when prior preferences and knowledge of available options are incomplete before shopping, selection of a satisfactory (or perhaps optimum) good is enhanced by assortment depth, offered either in a single large store or by a cluster of similar stores. These quite different effects of width and depth should be explicitly recognized.

None of these recognized points are either original or recently discovered, yet none has yet been satisfactorily addressed. At this point, it seems unlikely that a single two variable model can be expected to deal with all of these issues. Here are some brief observations on directions for devising goods specific models, using as a starting point the traditional convenience specialty-shopping goods scheme:

(1) Only customers who literally have no existing preference structure before shopping are likely to respond to assortment depth as the spatial interaction model predicts. The information needed to form such preferences is available from many alternative sources. The proliferation of direct (mail order) sources of both information and the goods themselves introduces a major modifier. There are limits, too, on how many alternative options can be evaluated, and the number of places to shop that will be considered in an "evoked set" is probably more constrained than the model predicts.

(2) A true specialty good is one for which strong preference exists. It is likely such preference is accompanied by good knowledge of available sources. Buying at the closest known source, regardless of that source's assortment and distance characteristics. may be the predominant decision rule.

(3) For convenience goods, circumstances determine the relative attractiveness of low effort versus the transactional efficiency offered by a wide assortment. Habit and loyalty patterns based on familiarity and store environmental preferences (including store personnel) may dominate behavior for multi-item trips even in the common situation where two or more comparable places to shop are nearly equal in spatial accessibility. When one or 8 small number of convenience goods are purchased, a "nearest store" model may predict patronage well. However, because many of these trips do not originate at home, "nearest store" is difficult to determine in the absence of knowledge of circumstances.

TESTING THE PSYCHOLOGICAL ECONOMICS MODEL

Kumar and colleagues' test of Katona's psychological economics model regresses several variables posited to reflect consumers' "willingness to buy" and "ability to buy" against monthly unit auto sales data for the period September 1977 through March 1982. Both the extent and (especially) the timing of consumers' spending for durables are somewhat discretionary and sensitive to economic forces affecting income. prices, interest rates and consumers' expectations about future trends in these forces. Over the years, many researchers have experimented with various series thought to serve as surrogates for consumer "willingness" and "ability" to buy with remarkably little success in fitting models to sales data for specific products, although patterns of change in total durable goods sales and more aggregated measures of economic activity have been more successfully captured by such models.

Kumar et al., have succeeded in finding a simple model that fits reasonably well the fluctuations in unit demand for autos during the period studied. The conceptual linkage between their chosen independent variables and Katona's models appears to be so weak, though, that little explanation is provided for the extremely volatile demand pattern. Additionally, the several rather unusual situational aspects of the period chosen raise questions about any possible generalizations about the relationship presented should prediction of future unit auto sales be attempted.

By choosing unit auto sales as the dependent variable, the researchers avoided the need to deal directly either with the dramatic inflation in auto prices over the period or with the important shifts in the mix of models selected (e.g., the shift from domestic to foreign manufacturers and changes in the relative importance of various size classes). It should be remembered that these changes characterized the period, though, so that the results of the study may reflect the particular circumstances that prevailed.

Neither "willingness" nor "ability" to buy durable goods has been given definitive meaning by macroeconomists, and the consumer research literature on the household level is seldom influential in macro studies, so researchers are free to select from the time series available those which will stand as surrogates for the two measures. In the present study, Kumar et al., followed substantial earlier precedent in using the SRC Index of Consumer Sentiment to stand for consumer "willingness" to purchase a new auto. Of the five variables entering equally into the index only one ("is now a good or bad time to buy major household durables?") has anything to do with specific purchase behavior, and all five components in essence and consumers about indicators of general financial and economic conditions. Fabian Linden concludes that consumers' responses to such questions are predictive of aggregate economic fluctuations because consumers are "quick to detect changes in the economic pulse at an early stage, apparently before they are sufficiently pronounced to have a significant effect on the Department of Commerce's aggregate statistical series." (Linden, 1982, p. 355). Consumers are literally serving as shortrun forecasters of changes in economic conditions in their role as respondents to the SRC survey. The conceptual connection between this role and their role as reporters of their personal "willingness" to buy a new car is clearly pretty remote.

"Ability to buy" in the Kumar study is not measured by household variables at all, but rather by cost, inventory and advertising series that could be classified as environmental pressures, constraints and influences to which "ability to buy" could conceivably respond.

Cost is the Suggested Retail Price index multiplied by the interest rate. Several aspects of SRP as a component of Cost could cloud its appropriateness as an indicator of month- to-month changes in consumers' ability to buy. During the time period of the study, auto manufacturers changed from a policy of annual price increases (coinciding with new model introductions) to intermittent smaller increases spaced irregularly throughout the year, both as a means of reducing the visibility of price increases and as a result of inability to forecast the need for such increases. As was the case with sharply increased fuel prices, consumers gradually accommodated rapidly escalating car prices through the period through downsizing (and shifting to cheaper imports), lengthened purchase cycles, and increased purchases of used cars. Consumers were not fooled by the policy of spreading out price increases over the model year. In fact, Katona found that in 1977 and 1978, at least one third of all new cars bought were motivated by fear of further inflation rather than immediate needs and wants, (Katona, 1979, p. 16). This finding is an example of the tendency for price expectations to reverse the expected inverse relationship between unit auto demand and price. The irregular and unexpected spacing of these price increases represented a new situation for consumers that could have exaggerated the uncertainty over rising prices in the future

The relationship between SRP and actual transaction prices has traditionally varied seasonally, with deeper discounts from list growing in winter months (after the excitement over new models has worn off), tapering off again as demand rises in the spring, and deepening again with old model clean-up. This pattern may have changed somewhat during the period under study, as imported cars (where discounts have been far less important) have increased year--around pressure on transaction prices for domestic models.

Overall, price has undoubtedly played a complex and changing role as an influence on consumers perceived "ability" to buy new cars. Exploring the hypothesized dimensions of this relationship would be an interesting study in itself. The regression model has disguised many of the interesting aspects of price that would further our understanding of bow consumers' perceived "ability" to buy responds to this complex determinant.

Inventory is included in the model since dealer inventory levels are suspected as an important determinant of manufacturers' rebate and special interest rate promotions. It is unfortunate that a more direct and precise indicator of these programs couldn't have been found, since they undoubtedly had a major impact on monthly variations in auto demand. One suspects that the inability of consumers to predict the onset of such special promotions contributed considerably to instability of demand during the period. Like the "new" pattern of price increases during the model year, rebate and interest rate programs signify a different and unusual time period in which to test a predictive model of auto demand. Unfortunately, the present research does not illuminate the impact of these "new" influences on consumers' perceived "ability" to buy.

Advertising as an indicator of monthly variations in auto demand is a concept directly contrary to much of what appears in contemporary writings on the role of advertising in the marketing programs for autos. In a now famous article, Gail Smith (1965) revealed the elaborate multi--wave research system established by General Motors to assess advertising effects for new cars, the purpose of which was to provide a realistic alternative to simple direct measures of purchases as measures of response to advertising. Much car advertising is intended to influence awareness, knowledge and attitudes toward specific makes and models, and its spacing throughout the year is undoubtedly geared more to the achievement of these specific goals rather than to the achievement of monthly sales quotas. The series used includes only newspaper and magazine spending, which for 1982 included only 33 - 43% of total advertising spending for the domestic Big Three manufacturers. (Ad Ale, 1983, p. 8). There is evidence too that advertising spending levels are often deliberately out of phase with sales. In 1982, Ford, General Motors and Volkswagen of America all increased ad budgets despite decreased sales for all three companies, while Chrysler had increased sales but cut its budget 10.45 (Ad Arc, 1983, p. 156). There seems to be Little reason to believe that the aggregate effect of newspaper and magazine advertising could have a direct impact on monthly variations in total unit demand.

If taken individually, one would be inclined to conclude that the very weak conceptual link between each of the independent variables as surrogates for "willingness" and "ability" to buy. the irregular and changing nature of these variables throughout the time period, and the often questionable nature of their predicted functional relationships with unit auto sales raises serious doubts that any of the variables could contribute much toward the explanation of monthly variations in auto demand. On the other hand, the tests of the various models' fit as indicated by adjusted 12's in the .58 range indicate that conceptual problems are not all that relevant if model fit is the criterion. However, this totally pragmatic view seems inconsistent with such interpretive statements as that offered by the authors in speculating why the models with inventory lagged one period produced higher R2's than the one where inventory was not lagged. "It may be that automobile manufacturers, upon observing high inventory levels in one month, may have a greater emphasis on price promotions (rebates, etc.) during the nest".

One is almost compelled to conclude that although the models (especially Model IV) fit the data fairly well, we don't really know why. At the same time, in puzzling over the conceptual questions raised by the kind of regression studies represented by the Kumar et al. study, we are stimulated to think of how situational variables such as those rather crudely represented by Cost, Inventory and Advertising do affect consumers' perceptions of their "willingness" and "ability" to buy durables. One wonders, too, what other possible variables within the household might be studied with monthly surveys that might more directly assess changes in consumers' inclinations to initiate, accelerate, postpone or abandon plans to expand, restore or upgrade household durable goods assortments.

REFERENCES

Advertising Age (1983), "100 Leading National Advertisers." September 8.

Ellinger, Robert S. and Lindquist, Jay D. (1984), "The Gravity Model: A Study of Retail Goods Classification and Multiple Goods Shopping Effect," in Thomas Kinnear (ed.), Advances in Consumer Research, Vol. XI.

Huff, David L. (1963), "A Probabilistic Analysis of Shopping Center Trade Areas." Land Economics, 39: 81-90.

Katona, George A. (1979), "Consumer Expectations as Guide to the Economy," Public Opinion, (Jan./Feb.), 15-19.

Katona, George A. (1951), Psychological Economic Behavior, New York: McGraw-Hill.

Kivell, P. S. and Shaw, G. (1980), "The Study of Retail Location" in John A. Dawson (ed.), Retail Geography, New York: John Wiley & Sons, 96-155.

Kumar, V., Leone, Robert P. and Srivastava, Rajendra K. (1984), "A Model of U.S. Automobile Demand," in Thomas Kinnear (ed.), Advances in Consumer Research, Vol. X1.

Linden, Fabian (1982), "The Consumer as Forecaster," Public Opinion Quarterly, 46: 353-360.

Shepherd, I. D. and Thomas, C. J. (1980), "Urban Consumer Behavior," in John A. Dawson (ed.), Retail Geography, New York: John Wiley & Sons, 18-94.

Smith, Gail (1965), "How GM Measures Ad Effectiveness," Printers' Ink. (May 14).

----------------------------------------