Experimental Studies of Consumer Demand Behavior: Towards a Technology of Making the Slutsky-Hicks Theory Technologically Applicable to Individual Behavior

Raymond C. Battalio, Texas A&M University
John H. Kagel, Texas A&M University
[ to cite ]:
Raymond C. Battalio and John H. Kagel (1975) ,"Experimental Studies of Consumer Demand Behavior: Towards a Technology of Making the Slutsky-Hicks Theory Technologically Applicable to Individual Behavior", in NA - Advances in Consumer Research Volume 02, eds. Mary Jane Schlinger, Ann Abor, MI : Association for Consumer Research, Pages: 657-670.

Advances in Consumer Research Volume 2, 1975      Pages 657-670

EXPERIMENTAL STUDIES OF CONSUMER DEMAND BEHAVIOR: TOWARDS A TECHNOLOGY OF MAKING THE SLUTSKY-HICKS THEORY TECHNOLOGICALLY APPLICABLE TO INDIVIDUAL BEHAVIOR

Raymond C. Battalio, Texas A&M University

John H. Kagel, Texas A&M University

[This research was partially supported by NSF Grant #GS32057 "Interpretation Systems for Empirical Economic Theories with Application to Theoreis of Factor and Consumer Demand" (Principal Investigators R. L. Basmann, R. C. Battalio and J. H. Kagel).]

The methods and results of experimental studies of individual consumer demand behavior with human subjects in controlled economic environments and with laboratory animals in operant conditioning chambers are reviewed. The review focuses on some of the technical problems encountered in conducting the experiments reported. The implications of these problems and their solutions for the efficient design of economic experiments and empirical consumer demand research in general are discussed.

During the past several years our research in consumer demand behavior has included experimental studies with human subjects in controlled economic environments and with laboratory animals in operant conditioning chambers. The goal of these studies has been to begin to provide an observational interpretation system for the Slutsky-Hicks theory of consumer demand (Hicks, 1946; Uzawa, 1960) that is adequate for the technological application of this theory to the behavior of individual decision making units. [For a more detailed discussion of the concept of an interpreted economic theory and the importance of focusing more research attention on the empirical interpretation system see Basmann, 1974; Brunner, 1969a,b; Morgenstern, 1963;and Sidman, 1960.]

The purpose of this paper is two-fold. First, this paper provides an opportunity to review our experimental procedures and some of the empirical results obtained from this research. The review of the empirical results, while focusing on tests of the consistency of the behavior of individual consumers with the Slutsky-Hicks theory, also includes, to indicate the breadth of problems which can be investigated using these experimental methods, a brief review of the results of investigations of several empirical hypotheses which have been of interest to consumer demand researchers using data from the national economy. Second, our review will focus on some of the technical considerations involved in the actual design and execution of controlled economic experiments. More specifically, we present a detailed discussion of some of the technical problems we encountered in the recording and processing of the data from our experiments and the implications of these problems on the efficient design of economic experiments and on empirical research in general. Initial solutions for dealing with the problems in question are also presented.

EXPERIMENTAL STUDIES OF ECONOMIC BEHAVIOR: HUMAN SUBJECTS

The experimental studies reported on below were conducted in a controlled economic environment. Controlled economic environments are organized systems in which individuals live for continuous extended time periods and receive tokens or points for work performed which are in turn exchangeable for present or future consumption goods. [See Kagel (1972) for a general discussion of token economies and token economy research. The differences between token economy experiments and previous experimental studies of consumer demand behavior are discussed in Battalio, Kagel, Winkler, et al., (1974) and will not be repeated here.] From our perspective controlled economic environments closely correspond to the economists' concept of a closed economic system where tokens are money, token payments for work performed are wages and the exchange rates in tokens for consumers' goods are the prices of these goods.

The token economy in which we conducted our initial experiments was established in a female ward for chronic psychotics at Central Islip State Hospital in New York. [A detailed description of this economy, the experiment performed and the results can be found in the papers referenced, and is not reported here. The same is the case for the other experiments referred to in this paper.] The experimental design used was, in part, selected to allow us to investigate whether any member of a Slutsky-Hicks system of demand functions was consistent with the observations of individual's purchases. That is, suppose we observe n distinct goods bundles, xi, actually purchased at prices pi, i = 1, ..., n. It can be shown that a necessary condition for the observed price and quantity data to satisfy a system of Slutsky-Hicks demand functions is that: [See for example Theorem 6, p. 144 ff in Uzawa (Uzawa, 1960).]

For any positive integer m<n, if xi = xj, for all i, j<m, and (1) poxo > pox1 & p1x1 > p1x2 & ... pm-1xm-1 > pm-1xm then ~(pmxm > pmxo) (where the symbols "~" and "&" denote negation and inclusive conjunction respectively). In cases where condition (1) is not satisfied the observed price and quantity data do not satisfy any system of Slutsky-Hicks demand functions; i.e., the collection of purchases observed is inconsistent with the Slutsky-Hicks theory. By virtue of the logical equivalence of the Slutsky-Hicks theory and the theory of revealed preference, a test of condition (1) is also a test of the consistency of consumer purchases with the axioms of revealed Preference theory.

In conducting this experiment, for each occasion the store was open a separate record was made of the items purchased (exchange point record) for each individual. These records were made by trained patient observers who sat next to the store "salesman" and recorded each purchase as it was made on specially prepared tabulating sheets. The observers were supervised by the ward psychologist. Store expenditures were later calculated from these records by multiplying quantities of goods purchased by their respective prices. A second, independent record of total store expenditures was obtained by collecting each token handed in at the store, stamping it "token store" and storing it by week spent (token record). Since each token had the consumer's name on it and was not transferable to other patients, counting these tokens provides an independent measure of the value of total store purchases by individual patients. Comparison of these two independent measures of total store expenditures reveals that there were errors in making and reporting of observations of these purchases. In most cases where the two measures were not the same, the exchange point record totals were less than the token record totals. The difference between the two measures was less than or equal to 5 tokens, or less than or equal to 10% of the larger measure, for about 80% of all individual observations. The sources of these differences were errors made in the counting and classifying of physical objects and their subsequent recording; e.g., misclassification of commodities, misclassification of tokens by ownership, denomination and week spent, and mistakes in counting, including the failure to record store purchases at the time they occurred. Independent checks of non-store purchases, i.e., room rent, breakfast, honor card rental, etc., showed infrequent errors to have been made in recording these expenditures.

Although the sources of the token store errors are not unlike those reported in consumer purchase panel reports (Neter, 1970; Sudman, 1964, Sudman and Bradburn, 1973), there are several important differences between the measures of the observational error in the consumer purchase data reported here and the measures of observational error available for consumer panel reports. [Consumer panel reports have also been used to examine condition (1) (Koo, 1963; Koo and Hasenkamp, 1972; Mossin, 1972). For a discussion of the problems in evaluating the results of these studies see Battalio, Kagel, Winkler, et al., 1973.] First, our measure of error is based on two independent measures of the same concept which, if carried out correctly, would give the same value without adjustments to either figure. Measures of error in consumer panel reports rely on indirect measures to check for observational errors. Consequently, the two measures will not yield the same value without adjustment of the indirect measure even if both measurement procedures were carried out correctly; e.g., the indirect measurement of the error in consumer panel reports through means of adjusted commodity shipment data. Second, our error assessment procedure provides a measure of error for each subject, for each observation period, whereas no such effort has been reported for consumer panel data. Thus, a measure of the error for each individual observation can be used in assessing our test results. A comparison of our measures of errors with those reported for consumer panel data indicates that our data are at least as accurate as data reported from consumer panel groups. For example, Sudman's (Sudman, 1964) assessment of error bias in panel records of food purchases as assessed against adjusted shipment data shows that for over half of the 55 items studied there is a net undercounting of purchases by panel members of 20% or greater.

In assessing the outcome of our test of condition (1) a measure of the observational error was incorporated into the empirical analysis. While this analysis showed that the behavior of 95 percent of the individuals was in agreement with the test condition within the measured accuracy of the data, for about one-half of the individuals failure to account for observational inaccuracies would have led us to conclude, incorrectly, that behavior was inconsistent with condition (1). The data were then investigated to determine the sensitivity of the empirical tests actually conducted to measurement errors. For the two subjects whose purchase patterns were found to be inconsistent with condition (1), the measurement of the error was zero for both of the inconsistent weeks for one subject, while for the other subject measurement of the error was zero and one half token respectively for the two inconsistent weeks. Further, in both cases an error of 10% or more in total expenditures would have been sufficient to be unable to distinguish between observation errors and actual contradictions of the theory. Of the 15 subjects who were judged inconsistent with condition (1) prior to accounting for inaccuracies in observations, for 10 of them errors of 3 tokens or less in measuring total token expenditures, which is less than 10% of average per capita expenditures during the test period, were responsible for our inability to distinguish between observation errors and contradictions of the theory. Further, for at least 5 of these subjects, an error of 1 token, the smallest unit of currency, would have been sufficiently large to prevent such a determination.

From a pragmatic point of view, the cases for which the observational errors were too large to allow us to distinguish between observational errors and actual contradictions of condition (1) represent a loss of information which is potentially quite important with respect to designing further tests and in suggesting modifications to the theory. In view of the large experimental changes in relative prices and the relatively high quality of the data set used (compared to measures of observational error available for consumer purchase panel reports), it is clear that further direct examination of condition (1) requires an improved observational technology. We have made some progress toward developing such a technology (Winkler, Kagel, Battalio, 1974). This technology, employed in a subsequent experiment, relied on a system of internal consistency checks and multiple, independently recorded, measures of the same economic activity. If there were discrepancies in the regular daily check of the records the relevant subject was required to stay back at the end of the day until his records balanced. Using this system the unreconciled discrepancy remaining in the records for a three week period, during which approximately 135,000 transactions of earning and spending took place with an average daily income level of about 500-600 tokens, was only 23 tokens.

Recent advances in technology allow for further improvements in the collection, recording and processing of data. A technology is now available whereby coded information, attached to each commodity prior to placing the commodity in the purchase area, can be read with an optical scanner at the time of purchase. At the time of purchase other information such as date, time or subject number can also be automatically recorded. In addition to reducing the errors in counting, recording and transcribing the data, this system would substantially shorten the time lag between the collection of the primary data and a complete analysis of the results of the experiment.

The importance of collecting data in a form that is suitable for continuous and relatively detailed analysis cannot be overemphasized in conducting economic experiments in controlled environments. In our experiments, where 3,000 to 4,000 observations per day were recorded and the day-to-day problems of managing the environment are quite time consuming, it is often months after the completion of the experiment before these data have been in a form suitable for computer analysis and the analysis been completed. Our experience has shown that a much greater range of empirical data could have been obtained from our experiments at little additional cost if the information obtained from a continuous and detailed data analysis had been available and used to modify the design and procedures of succeeding phases of a study. The empirical outcomes of our experiments have never been (nearly) completely foreseen. The more quickly these unforeseen outcomes are incorporated into the experimental design and procedures, the more efficient the utilization of the scarce resources.

It is clear from the above discussion that specific decisions about the allocation of research resources toward improving the accuracy of the data are made in conjunction with the choice of the experimental environment to be used and the particular hypotheses to be examined. Since any proposed test of one hypothesis against an alternative requires some minimal level of data accuracy to allow for observationally distinguishing between the hypotheses, many of these same considerations are relevant in using non-experimental data. That is, the choice of hypotheses which can be examined using any data set is constrained by the accuracy of the data set used. Although these constraints are seldom mentioned when using non-experimental data in economics, most of which is collected for purposes other than its use in testing scientific economic hypotheses, the failure to do so makes it extremely difficult to gauge the empirical content of the published results. In view of our direct experience (which indicates that simple counting and classification errors often exceed 5 percent) and the other independent studies in this area (e.S., Sudman, 1964; Morgenstern, 1963), we conjecture that in using data where no information is given about the level of accuracy an estimate of an average absolute error of 5-10 percent for the combined conceptual, observational, truncation and rounding errors is an optimistic estimate. In this context, when such data are used to examine empirical hypotheses which have a derived demand for accuracy of greater than this level it should be the researcher's obligation to provide the evidence required to enable the reader to gauge the empirical content of the results presented.

In addition to providing an initial test of condition (1) and direct empirical evidence on the importance of an explicit recognition of the observational technology and a measure of the observational errors, this experimental study also provided empirical data important to modifying the Slutsky-Hicks theory.

An analysis of the individual consumer data indicated that many of the individuals who changed their composition of consumption when relative prices changed, buying more of the less expensive commodity and less of the more expensive commodity, did not return to the original consumption pattern when prices were returned to the initial baseline value (Battalio, Kagel, Winkler, et al., 1974). In these cases, compared to the initial baseline consumption, the subjects purchased a greater percentage of the commodity which had previously been relatively less expensive. This response was symmetric across commodities during the experiment. These results were quite similar to the results of an earlier, independent study by Ayllon and Azrin (1968a,b) where baseline consumption patterns were displaced by offering subjects small quantities of a free sample between baseline periods. Ayllon and Azrin found that purchases of the sampled commodities were consistently greater following the free sample period as compared to before it. The range of commodities they investigated included soda, trips to a fair, listening to music, and popcorn. The residual effects of increased consumption of the sampled commodity were found both for patients who had consumed the goods in the first (baseline) phase of the experimental period as well as for those who had not. Ayllon and Azrin's results suggest that the changes in consumption associated with the relative price changes in our experiment may account for the residual effects noted above. From these studies it remains indeterminant whether these sampling effects represent temporary, transitional effects following price changes and sampling periods or whether they represent permanent changes in consumption patterns.

While our primary research aim is to characterize the behavior of individual consumers with a specific class of systems of demand functions, thereby removing much of the uncertainty now present with respect to the effect of a price or income change on an individual's purchases, the relatively detailed micro-data set obtained also provides the basis for a relatively inexpensive examination of some empirical hypotheses which are of interest to consumer demand researchers using data from the national economy. For example, using the above data we have examined the following oft-mentioned empirical hypothesis:

If the individuals in an economy can be characterized by a Slutsky-Hicks system of demand equations, then the aggregate per-capita composite demand functions are also characterizable by a Slutsky-Hicks system of demand functions, providing the relative prices of each composite commodity remain exactly proportional.

As the data in Figure 1 illustrate, it is possible to organize these data quite well with the system of demand equations

xti = AiMtP-1ti,   i=1, 2, 3; t=1, ..., 7.

A formal test of this system of demand equations has shown that, modifying the system to account for the sampling effects of price changes, the discrepancy between the predicted data points and the actual data points is less than the measured errors of observation. Consequently, the introduction of more complicated systems of demand equations at this point, i.e., systems for which Oxi/Opj + 0, is unnecessary (Basmann, Battalio, Kagel, 1974).

Since the change in relative prices also resulted in changes in real wages for some of the consumers, this data set could also be used to examine the shape of the individual's labor supply schedule. Performing this analysis showed that the low income earners responded to an increase in real wages primarily by increasing real income and working about the same while the higher income earners responded with small increases in real income and substantial decreases in work (Kagel, Battalio, Winkler, et.al., 1974). The contrasting behavior of the upper and lower income earners has been accounted for in terms of the initial set of consumer's goods attained relative to available set of consumer's goods. That is, the low income earners had considerably more range, in terms of consuming new goods and increasing consumption of the goods already being consumed, to increase consumption than did the upper income group given the restricted range of consumer's goods available in the economy. (In a subsequent study where the subjects could transfer unspent tokens to currency at the end of a 70 day experiment the high income earners responded to an increase in money wages with little or no decrease in work.) While others have explained labor supply behavior on a similar basis this explanation has heretofore received little direct empirical examination. In this context it is of interest to note that the labor supply response obtained for the low income earners is consistent with the results obtained in the recent income maintenance experiments where the subjects were selected from the working poor (U.S. Dept. H.E.W.. 1973).

FIGURE 1

AGGREGATE PER CAPITA INCOME AND EXPENDITURE DATA FOR THE CENTRAL ISLIP TOKEN ECONOMY

The preceding discussion has primarily drawn on data from one study in a therapeutic token economy. Controlled economic environments, using volunteer subjects from the national economy who have agreed to reside continuously in these environments for extended periods of time (up to three months), have been developed and successfully applied to the explicit study of particular socio-economic problems (Bigelow and Emurian, 1974; Miles, et al., 1972). While such controlled environments are expensive and require a research team to manage, the relative costs, when weighed against the benefits, make these environments ideally suited to examining certain aspects of economic behavior.

EXPERIMENTAL STUDIES OF ECONOMIC BEHAVIOR: LABORATORY ANIMALS

Ever since Darwin it has been widely recognized that behavior as well as structure varies continuously across species and that the applicability of behavioral principles does not stop suddenly at the boundary separating humans from other animals. The principles of economic behavior would be virtually unique among behavioral principles if they did not apply, with some variation of course, to the behavior of non-humans. In comparison to the experimental studies with human subjects reported above, experimental studies of economic behavior using laboratory animals have several advantages. Since the subjects are housed in an experimental chamber and automatic programming equipment is used to carry out and record the transactions, the errors associated with making the transaction and recording the price and quantity observations are greatly reduced at a fraction of the cost of a comparable technology for human subjects. Working with laboratory animals also allows us to conduct experiments and change economic parameters in ways that would be unacceptable on ethical and/or legal grounds with human subjects. For example, although it would be technically feasible to reduce the number of goods in a human consumer's budget set in a token economy, thus simplifying the problems of characterizing a commodity, we do not consider this a practical strategy in view of the empirical evidence that suggests such a simplification may be associated with extensive physiological and psychological changes, many with adverse therapeutic effects. Further, in working with laboratory animals we can, ethically and legally, control for a substantially greater number of environmental factors affecting the organism's behavior than we can with human subjects. This control, e.g., placing an animal in a sound attenuated test chamber, allows a reduction in experimental "noise" to a negligible magnitude compared to studying consumer behavior in controlled economic environments or in national economic systems.

In what follows we summarize the results of two series of experimental studies of consumer demand behavior we have conducted using white male albino rats as subjects (Kagel, Battalio, et. al., 1974). In one experiment the commodities over which the budget set was defined consisted of root beer and Tom Collins mix with food and water available at all times in unlimited amounts. In the other experiment the commodities in the budget set consisted of food and water with no other consumption commodities available outside the budget set. In conducting the experiments our primary goal has been to determine whether laboratory animals would change consumption patterns in response to changes in the budget set and whether the behavior observed could be characterized by members of the class of Slutsky-Hicks demand functions.

In the root beer-Collins mix experiment the subjects were individually housed in a completely enclosed experimental chamber continuously, 7 days a week, except for a period each morning when the cage was cleaned, commodities were replenished and the data were recorded. Two metal levers, projecting into the chamber and located on the back panel of the chamber wall, operated separate dipper feeder mechanisms located behind the panel on which the levers were mounted. A single response on the left lever resulted in the presentation of a dipper cup containing root beer at floor level directly below the lever. A single response on the right lever produced Collins mix from the other dipper feeder. Each subject had a limited sum of lever presses available for operating the dipper mechanisms each day which could be distributed in any way chosen between the two fluids. Under baseline experimental conditions each subject was allotted 300 lever presses with a .05 ml. dipper feeder cup on both dipper feeders giving the budget set represented by the area on or below the straight line A in Figure 2; i.e., the subject could choose any combination of root beer and Collins mix represented by a point on or below line A. Changing the total number of lever presses allotted while maintaining constant dipper cup sizes and lever pressing requirements, would result in shifting out or in of the boundary of the budget set parallel to itself. This gives the same effect that an income change would have on a human consumer's budget set. Thus, the total lever presses allotted to the subject was operationally equivalent to his income level and will be referred to as such. Increasing (decreasing) the size of a dipper feeder cup (and, proportionately, the amount of time the cup remains available) while maintaining a constant lever pressing requirement per dipper presentation, would have the same effect on the rat's budget set as a price decrease (increase) would have on a human consumer's budget set. Thus, the number of lever presses required to obtain 1 ml. of either fluid, maintaining constant lever pressing requirements per dipper presentation, was operationally equivalent to the Price of the commodity and will be referred to as such.

Each experimental condition (constant set of prices and income) was maintained for a minimum of 14 days. In addition, a stability criteria of 10 consecutive days, during which no trend was observed in either consumption patterns or time taken to exhaust income, had to be met before conditions were changed.

Mean daily consumption, or purchases, for one of the subjects under baseline experimental conditions (budget line A in Figure 2) is designated by the point X in Figure 2, and the brackets indicate + one standard deviation about the mean. [Data are reported for all but the first three days of an experimental condition.] At the income level reported, with prices of both commodities the same, this subject consumed about four times as much root beer as Collins mix. Doubling the price of root beer and halving the price of Collins mix, while simultaneously adjusting income so that the subject could continue to consume the same commodity bundle as before (budget line B in Figure 2), resulted in the subject consuming about half as much root beer as Collins mix (point B in Figure 2 with the brackets again indicating + one standard deviation about the mean). While the return of prices and income to baseline values resulted in an immediate and sharp reversal in the magnitude of consumption towards the original values (point A in Figure 2), this reversal was incomplete, similar to the sampling effect reported above for human consumers. A further halving of the price of root beer and doubling of the price of Collins mix, while again adjusting income so that the same commodity bundle could be consumed as under baseline conditions (budget line C in Figure 2), resulted in a substantial increase in root beer consumption and reduction of Collins mix intake (point F in Figure 2).

FIGURE 1

While the above data for mean purchases are consistent with the Slutsky-Hicks theory of consumer demand, the virtually complete control over observational errors maintained in these experiments has allowed us to demonstrate that under a constant set of prices and income the individual consumers do not consume the same commodity bundle from one observation period to another. We view this behavior as resulting from uncontrolled variations in the subject's environment. since even under the most rigorous control of experimental conditions the subjects are exposed to certain changes in environmental conditions, e.g., changes in temperature, humidity, the composition of commodities, sexual cycles, etc. (Sidman, 1960). Whether the degree of control of experimental conditions maintained in any particular study is acceptable can, of course, only be determined with respect to the hypothesis being examined and the magnitude of the changes in the recorded behavior associated with the changes in the experimental conditions. In the root beer-Collins mix studies the daily variation in purchase patterns was small relative to the differences in the purchase patterns between experimental conditions which enabled us to unambiguously determine that the subjects did change mean consumption patterns in response to changes in the budget set and that this behavior could be characterized by members of the class of Slutsky-Hicks demand functions (Kagel, Battalio, et al., 1974).

At this stage in our research we have eschewed as premature the hypothesis that the daily variations in purchase patterns under a constant set of prices and income are, in any part, attributable to any inherent randomness on the part of the individual consumers. Transparently, the empirical outcomes of our experiments cannot be affected either by adopting or avoiding of such a definition. Notwithstanding this decision, in finding definite members of the class of systems of Slutsky-Hicks demand equations to characterize the data we have adopted the methodological convention of considering purchases under a constant set of prices and income as random variables, i.e., variables derived from a probability distribution. The economy gained, at this stage of research, by considering the effects on consumer purchases of uncontrolled factors in the subjects' environment within the framework of probability concepts rather than explicitly incorporating hypotheses about these factors into the demand equations is more or less unproblematical.

Irrespective of the profferred explanation of the failure of the subjects to consume the same commodity bundle under a constant set of prices and income, this behavior has important implications for the design of further experiments aimed at testing and modifying consumer demand theory. For example, if the design of a proposed test of consumer demand theory calls for "small" changes in prices and/or income then: (1) enough observations must be collected under each experimental condition to distinguish the effects of the systematic changes in the subject's environment from the effects of the unsystematic factors and/or (2) the amount of unexplained variability in purchase patterns under a constant set of prices and income must be sharply reduced. Of course, considerations such as these only become relevant when observational inaccuracies have been reduced to the point where we can distinguish changes in actual behavior.

The experiments in which the budget set was defined over food and water, with no other commodities available for consumption, followed procedures similar to those reported in the root beer-Collins mix studies. In these studies price changes also resulted in changes in the composition of consumption away from the higher priced commodities and in favor of the lower priced goods, although the responses to the price changes were substantially less than those reported in the root beer-Collins mix studies. More importantly, large rotations in the budget set brought about by large increases in the relative price of food (apparent real income held constant) also resulted in severe disruptions of the rat's behavior with the subjects failing to spend all of their income in the allotted 24 hours and steadily losing weight (notwithstanding the fact that as a result of the income constraint they were at only some 80% of normal body weight at the time the suppressed responding began). Since the experimentally induced changes in the budget set in the root beer-Collins mix experiment and the smaller changes in the budget set in the food-water experiment resulted in changes in consumption patterns quite similar to those reported for individual human consumers (Battalio, Kagel, Winkler, et al., 1973, 1974), the question arises as to whether certain changes in the budget sets of human consumers may also be associated with disruptive behavior. Although we do not have any direct empirical data at present, there is a substantial body of literature in psychology and sociology relating severe disruptions of human behavior, i.e., suicides and mental disorders, to socio-economic factors such as poverty and unemployment. [Two of the most important studies in this area are the books by Brenner (1973) and Henry and Short (1954). This research is, of course, not without its critics (see the references cited in Kagel, Battalio, et al., 1974).] While it is transparent that the disruptions in the rat's behavior (suppressed responding and weight losses) are not the same thing as suicides and mental disorders in humans, this literature does indicate that severe disruptions in human behavior are associated with economic factors. This suggests that in experimental studies of consumer behavior using human consumers the experimenter must guard against changes in the budget set that may result in severe disruptions of the consumers' behavior and be prepared to alter the experimental design accordingly. Further, experimental studies designed to determine the precise factors underlying the disruptions in behavior reported here must be limited to laboratory animals on both ethical and legal grounds. The generality of this behavioral process to human consumers can then be determined by applying the information gained from these studies to the design of permissible experiments with humans.

CONCLUSIONS

The research described in this paper represents our first steps in the process of achieving our goal of providing an observational interpretation system for the Slutsky-Hicks theory of consumer demand. While we view the actual outcomes of the experiments reported above as important, when appraised against the aim of our research - making the theory technologically applicable it is clear that they represent merely the beginnings of a task whose completion will require the work of many researchers for a considerable period of time. Our research aim also makes it clear that we do not entertain the concept of "a crucial test" of an economic theory. Rather, we view the theory, and its subsequent modifications, as simply the starting point for designing our experiments, the results of which will be used to modify and augment the observational concepts, the experimental design and the existing theoretical structure. Since every empirical study, whether experimental or not, is unique, the generalizability of a given empirical behavioral relationship can only be determined by continued empirical investigations. In this context, we view the future role of experimental research as an important complement to non-experimental studies. The particular research method, or combination of methods, used for any given problem will, of course, depend on the nature of the question being examined.

REFERENCES

Ayllon, T., and Azrin, N. H. The token economy: A motivational system for therapy and rehabilitation. New York: Appleton-Century-Crofts, 1968(a)

Ayllon, T., and Azrin, N. H. Reinforcer sampling: A technique for increasing the behavior of mental patients. Journal of Applied Behavioral Analysis, 1968, 1, 13-20 (b)

Basmann, R. L. Modern logic and the suppositions weakness of the empirical foundations of economic science. Schweizerische Zeitschrift fur Volkswirtschaft und Statistik, 1975, in press.

Basmann, R. L., Battalio, R. C., and Kagel, J. H. Aggregate per capita systems of demand functions: A test of a frequently mentioned empirical hypothesis. NSF Project #GS32057, Technical Report #17, Texas A&M University, (mimeographed), 1974.

Battalio, R. C., Kagel, J. H., Winkler, R. C., Fisher, E. B., Jr., Basmann, R. L. and Krasner, L. A test of consumer demand theory using observations of individual consumer purchases. Western Economic Journal, 1973, 11, 411-28.

Battalio, R. C., Kagel, J. H., Winkler, R. C., Fisher, E. B., Jr., Basmann, R. L. and Krasner, L. An experimental investigation of consumer behavior in a controlled environment. Journal of Consumer Research, 1974, 1, 52-60

Bigelow, G. and Emurian, H. A programmed environment for the experimental analysis of individual and small group behavior. In C. G. Miles (Ed.), Experimentation in controlled environments and its implications for economic behavior and social policy making. Toronto, Canada: Addiction Research Foundation, 1975, in press.

Brenner, M. A. Mental illness and the economy. Cambridge, Mass: Harvard University Press, 1973.

Brunner, K. Some reflections on the state of econometric practice. In Essays in honor of Richard Buchner. Zurich: Verleg Schutlhess & Co., A. G., 1969.

Brunner, K. Assumptions and the cognitive quality of theories. Synthese, 1969, 20. 501-525.

Henry, A. F. and Short, J. F. Suicide and homicide. Glencoe, Ill: Free Press. 1954.

Hicks, J. R. Value and capital. (2nd ed.) Oxford, 1946.

Kagel, J. H. Token economies and experimental economics. Journal of Political EconomY, 1972, 80, 779-785.

Kagel, J. H. Battalio, R. C., Rachlin, H., Green, L., Basmann, R. L., and Klemm, R. R. Experimental studies of consumer demand behavior using laboratory animals. Economic Inquiry, 1974, in press.

Kagel, J. H., Battalio, R. C., Winkler, R. C., Fisher, E. B., Jr., Basmann, R. L., and Krasner, L. An experimental analysis of the effect of changes in real wages on work behavior in a token economy. NSF Project #GS32057, Technical Report #3 (mimeographed), Texas A&M University, March 1974.

Koo, A.Y.C. An empirical test of revealed preference theory. Econometrica, 1963, 31, 646-664.

Koo, A.Y.C., and Hasenkamp, G. Structure of revealed preference: Some preliminary evidence. Journal of Political Economy 1972, 8Q, 724-744.

Miles, C. G., Congreve, G.R.S., Gibbons, R. J., Marsham, J., Devenyi, R., and Hicks, R. C. An experimental pilot study of the effects of daily cannabis smoking on socio-economic behavior. Paper presented at the 30th International Congress on Alcoholism and Drug Dependence, Amsterdam, September, 1972.

Morgenstern, O. On the accuracy of economic observations. (2nd ed.) Princeton: Princeton University Press. 1963.

Mossin, A. A mean demand function and individual demand functions confronted with the weak and the strong axioms of revealed preference: An empirical test. Econometrica, 1972, 40, 177-192.

Neter, J. Measurement errors in reports of consumer expenditures. Journal of Marketing Research, 1970, 7, 11-25.

Sidman, M. Tactics of scientific research. New York: Basic Books, 1960.

Sudman, S. On the accuracy of recording consumer panels: I. Journal of Marketing Research, 1964, 1, 14-20.

Sudman, S. and Bradburn, N. J. Effects of time and memory factors on response in surveys. Journal of the American Statistical Association, 1973, 68, 805-815.

U.S. Department of Health, Education, and Welfare. Summary report: New Jersey graduate work incentive experiment. Washington, D. C.: Department of Health, Education, and Welfare, 1973.

Uzawa, H. Preference and rational choice in the theory of consumption In K. Arrow, S. Karlin and P. Suppes (Eds.), Mathematical methods in the social sciences, 1959. Stanford: Stanford University Press, 1960.

Winkler, R. C., Kagel, J. H., and Battalio, R. C. Assessment of error in concurrent records of income and expenditure in token economies. NSF Grant GS32057, Technical Report #15, Texas A&M University (mimeographed), 1974.

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