THE MIDDLE-CLASS U.S. CONSUMPTION FUNCTION:A HYPOTHETICAL-QUESTION STUDY OF EXPECTED
By JULIAN L. SIMON and CARL B. BARNES*
In the main, economic theory contents itself with qualitative assertions, e.g.,whether the oligopoly price is determinate, or whether a rise in the quantity ofmoney affects the interest rate. But consumption theory goes further; it makesassertions about the shape of the function and its parameters. Two of the mostimportant quantitative assertions are that the function goes through the origin(Friedman 1957, 1963; and Modigliani and his co-workers, e. g. Modigliani andBrumberg, 1954; Ando and Modigliani, 1963, and Modigliani, 1966, hereaftercalled M. et al.), and that the proportion of present and future wealth consumedthis period equals the ratio of the length of this period to the expected length oflife (M. et al.).
There is good reason to try to estimate the parameters, and as accurately aspossible, because the national economy of a country such as the United States orGreat Britain is very sensitive to aggregate consumption. Important nationaldecisions are therefore affected by our understanding and estimates of con-sumption-function parameters.
This paper explores a new way of studying the consumption function. Werequested people to assume various sets of facts about their economic and lifestatus, and we then asked them how much they would spend and save in variousimagined situations. This approach has little in common with attitude studiessuch as are commonly conducted in other social sciences, in that we do not askpeople what they think but rather what they would do. And this approach is alsodifferent from studies of intentions, because it asks people how they wouldbehave in hypothetical situations rather than how they really expect to behave,given their real economic situations. That is, it obtains conditional counter-factual intentions, rather than unconditional intentions. The pioneer studyusing this hypothetical-question method in economics was by Gilboy (1931);other such studies have been done by Shaffer(1959) and Simon (1965). Themethod also has an intellectual link to such laboratory simulations as oligopolygames, which also try to learn from people's responses to hypothetical situations.
The weaknesses of our approach are many, as the reader will immediately see.Perhaps the greatest problem is that many people may not be able to projecttheir minds into a situation so very different than their real-life situation. Thatis, they may not be able to imagine accurately how they would behave if theirlives were different. And there are other important defects. But despite itsweaknesses (which will be discussed in more detail later), we believe that both
* The authors are grateful for useful comments made by Thomas Mayer and by the referee.73
a priori and in the light of the results this approach offers a ceteris paribusstudy of the consumption function that is not even conceptually possible withother methods: we can hold constant everything except the single variable thatwe change from question to question. Not only this one, but all approaches tothe study of the consumption function have defects, larger or smaller, and webelieve that understanding is best advanced by comparing many different sortsof studies that have different weaknesses and yield different results.
THE METHODThe questionnaire, whose questions are shown alongwith the datain the table,
is the heart of our method. It allows us to make two sorts of inquiries, onestronger than the other. The stronger inquiry is about the absolute amount thatpeople will spend (or save) in given economic situations, e.g., with assets of$10,000 and yearly income of $10,000 at age 30. This sort of inquiry estimatesthe proportions that they would spend, both the central value for the group andthe dispersion. The weaker inquiry concerns relative consumption, e.g., how muchmore people would spend if they receive a windfall than if they did not, or whethera windfall and already-saved assets affect consumption in similar ways.
1 2 3 4 5 6 7Sample Theoretical Mean Standard Theoretical Observed Standard
g?oup and level of observed deviation of difference difference deviationits size consumption consumption data in between between of datei in
from column 3 benchmark benchmark column 6equation 2 questions for questions for
particular particularage (Ql & age (Ql 6Q9) and Q9) and
this question this question'You are 30 years old, earn $10,000 a year after taxes and have total savings of $1,000
(except for your car, home furnishings, etc.). This year, how much will you spend and how muchwill you save?' (Reader Note: This is the benchmark question to which other questions beloware compared)60 Students $8240 $104352 Adults $8559 $1050
'You are 30 years old, earn $10,000 a year after taxes and have total savings of $10,000.This year, how much will you spend and how much will you save?' (N.B.: Saving [s] is higherthan in question 1 by $9,000)60 Students 8000 8635 977 + 200 + 395 95452 Adults 8636 1043 + 77 506
'You are 30 years old, earn $20,000 after taxes and have total savings of $1,000. Thisyear, how much will you spend and how much will you save?' (N.B.: Income Ev:1 is higherthan in question 1 by $10,000)60 Students 15,577 14,320 3400 +7777 +6079 279952 Adults 14,578 3389 +6019 2990
'This morning you received an unexpected inheritance of $10,000. Before this yourtotal savings were $1,000. Your income for this year is $10,000 after taxes and you are 30years old. This year, how much will you spend and how much will you save?' (N.B.: A $10,000windfall [w] has been added to the conditions in Q1).60 Students 8022 10,956 2899 -F 222 +2715 259052 Adults 10,355 3208 +1796 2768
'As above, you are 30, have savings of $1,000, and earn $10,000 after taxes. A trustfund has been established on your behalf and you know that in two years you will receivean inheritance of $10,000. This year, how much will you spend and how much will you save?'(N.B.: Same as in Q4 except here the inheritance is anticipated in two years. The theoreticaleffect on lifetime wealth is the same as in question 4).60 Students 8022 8745 921 -f- 222 + 504 63852 Adults 8955 1557 + 396 1129
MIDDLE-CLASS U.S. CONSUMPTION FUNCTION 75
/ 2 3 J 5 6 7Sample Theoretical Mean Standard Theoretical Observed Standard
group and level of obsevved deviation of difference difference deviationits size consumption consumption data in between between of data in
from colman 3 benchmark benchmark column 6equation 2 questions for questions for
particular particularage (Ql &Q9) age (Ql &Q9)
and this and thisquestion question
6. 'Currently, and for the next two years, you are in a training position at your job. Yourincome now is $10,000 after taxes and you have $1,000 savings. You know that as soon as yourtraining is over your salary will be increased to $20,000 after taxes. You are 30. This year,how much will you spend and how much will you save?' (N.B.: Beginning in 2 years y is higerthan in question 1 by $10,000).59 Students 15,133 8638 898 +7333 427 71352 Adults 8805 1215 + 246 698
7. 'You are 30 years old and have total savings of $1,000. You will not be paid until theend of this year, when there will be a 50 per cent chance that your income will be $20,000after taxes and a 50 per cent chance that your income will be $0. This year, how much willyou spend?' (N.B.: The expected value is identical to number 1; risk is the only difference).44 Students 7800 7152 2136 0 1062 176147 Adults 7797 2796 - 757 2557
8. 'You are 30, have $1,000 savings, and make $10,000 after taxes. You are going to workthe next 9 Saturdays on a special project for your employer that will not be done again. Forthis extra work you will receive an extra $10,000 (making your actual income this year $20,000).You will not have this opportunity in future years. This year, how much will you spend andhow much will you save?' (N.B.: A one shot $10,000 income is added to the conditions inquestion 1).46 Students 8022 10,950 3031 + 222 +2668 260952 Adults 11,875 3568 + 3315 3151
9. 'You are 60 years old, earn $10,000 after taxes, and have $1,000 savings. Remember,you have no retirement programme. This year, how much will you spend and how much willyou save?' (N.B.: All conditions are identical with those of number 1 except age 60 is assumedand therefore N T = 15 instead of 45).60 Students 3400 5540 1443 4400 2700 133451 Adults 5460 1489 2090 1268
10. 'Your salary is $10,000 and you have $10,000 savings. You are 60. This year, howmuch will you spend and how much will you save?' (N.B.: Saving s is higher than in question9 by $9,000).59 Students 4000 6740 1535 + 600 +1276 147552 Adults 7067 1639 + 568 1000
11. 'You earn $20,000, have $1,000 savings and are 60. This year, much will how youspend and how much will you save?' (N.B.: y is higher than in Q9 by $10,000).60 Students 6733 9631 3229 +3333 +4091 239552 Adults 10,461 3941 +4029 3312
Notes on the Table:Sample size varies because respondents occasionally neglected to answer a question.
This was especially true for question 7 where risk was involved; some people had troubleunderstanding the situation and did not answer.Question 1 is used as the benchmark for comparison with questions 2 through 9; allassume age 30. Questions 10 and 11 are compared to question 9; all assume age 60.
If people's imaginations were perfect, and if Consumption were a function ofonly absolute income and wealth and unaffected by tastes associated withincome, then the nature of the sample of respondents would not matter. But toassume such perfect imaginations would be foolhardy at this point. Therefore itis relevant that our respondents were of two sorts, (a) 60 students at the Univer-sity of Illinois; and (b) 52 middle-class white adults (26 in Champaign-Urbana,Illinois and 26 members of a central Illinois Rotary Club), ranging in age from
22 to 74. The adult group had a median reported income after taxes, of $10,000,the range being from $2,500 to $28,000.
At first we thought that people's answers would be strongly affected by thefacts of their own lives. Some reassurance that the method is not overly sensitiveto the respondent's own situation is, however, provided by what we consider tobe considerable similarity in the replies of the two groups. (The reader maymake his own judgment about this by referring to the data, both the groupmeans and the standard deviations.)
The absence of low-income people from the sample may not be critical,because low-income people consume a relatively small proportion of GNP andare not as relevant as are middle-income and upper-income people to the typeof policy decisions toward which this paper is aimed A study of this type aimedat low-income units would be interesting, however, and relevant to other typesof policy decisions.
Perhaps the most important remediable flaw in our method is that we madeno attempt to handle the savings or consumption aspect of durables purchases.We presume that respondents considered durable purchases as consumption,which leads to an understatement of savings estimates. Hopefully, futurestudies will remedy this defect.
The introduction to the questionnaire described the general life situationwhich was to apply to all questions. It read:
'As you answer the following questions try to forget how you actually liveand assume the following circumstances are the ones under which you live:
You are married, have two young children, and are finished with schooland military service.You rent a home and own a Ford that is one year old.You have no debts.You work in a local office of the Federal Government but you have noretirement programme, not even social security.For your type of job, raises in salary are given only to meet increases inthe cost of living; that is your income will not increase or decrease inpurchasing power throughout your life.You feel you will probably live to 75 years of age (you will have to retireat age 65).The purchase of life insurance, retirement programmes, investments, etc.,should be considered as saving.'
The situation that we asked respondents to assume was as close as we couldmake it to the assumptions of the Modigliani-Brumberg model, in order that wecould compare our results with the predictions derived from that model. Notice,however, that we did not specify that there would be no inheritance. Nor did wespecify anything about the future course of the national economy. Our specifica-tions about the individual's income expectations are built into his job descriptiona Federal Government worker.
Three different forms of the questionnaire were tried as we attempted toperfect the instrument. The questions shown in the table constitute the finalform, and that was the one on which all of the data shown were collected.
MIDDLE-CLASS U.S. CONSUMPTION FUNCTION 77
The theoretical expectations were obtained by inserting the appropriate datainto the following equation, and then solving it:
(1) CToATo + + TO
where CT=consumption in year T; the index T runs from 0, the present yearYT=income in year TA==savings as of the present\'T=windfa11s in year T
n=75 minus assumed present age, i.e., number of future years in which therewill be consumption
m=65 minus assumed present age, i.e., number of future earning years beforeretirement.
RESULTSThe table displays the questions and the outcomes.
From Question 1: Observed consumption is somewhat above the M. et cil.theoretical expectations, and closer to the American national average propensityto consume. This discrepancy might be a failure either of our method or of theM. et al. theory as stated above. The observed outcome would be more consistentwith an M. et al. theory suitably modified to allow for the discounting of incomeand consumption in future years. The reason for reporting the result here is as acalibration for the rest of the results.
From Questions 1 and 2: The amount of one's assets apparently affectsone's consumption, and in the theoretically expected direction. The differencebetween savings of $1,000 and $10,000 at age 30, with an income of $10,000, isassociated with mean differences of $395 and $77 in the two samples. Thetheoretically expected difference between saving in Questions 1 and 2 is $200,which suggests to us that the theory yields a better prediction than we would belikely to have arrived at in an ad hoc way. This we take to be a successful test ofthe correspondence of the M. et al. theory and our findings.
Here a word about the standard deviations in column 7 may be appropriateFor the comparison of questions 1 and 2, the standard deviations of the differ-ences are much larger for both groups than are the mean differences. The readermight therefore wish to disregard the comparison of the observed results fromquestions 1 and 2 with the theoretical prediction. Please note, however, that themeans of the two groups bracket the theoretical expectation. And if the twogroups were homogeneous and it were therefore reasonable to compute a singlemean and a single standard deviation, the latter would be considerably smallerand more impressive than at present.
The remarks in the foregoing paragraph apply a fortiori to the results forother questions, where the standard deviations of the differences are mostly ofrespectable size when compared to the mean differences, even for the twoindividual samples taken separately. We continue to stress, however, the simil-arity of the group means for the two disparate groups as our basis for confidencein the meaningfulness of our results.
From Questions 9, 10, 11, 1, 2, 3: Clearly, a strong life-cycle effect showsin these three pairs of questions. When told to assume that they are 60 yearsold, respondents said they would save considerably more than when they weretold to assume they were 30. However, the indicated absolute consumption atage 60 is far above what the theory predicts. The reason for this discrepancymight be a failure of our method, i.e., it may be impossible for younger people toaccurately imagine the appropriate situational conditions of being 55 or 60. Butwe believe it more likely that these estimates of consumption are as high as theyare because respondents did not or could not take seriously the condition thatthey would have no income from other sources after retirement.
We believe that the M. et al. theory is (also) wrong here, however. That is,we believe that real people act more like the responses to our questions than likethe theoretical predictions. (It would be interesting to get people's responses tothese kinds of questions without specifying a life expectancy and no othersources of income. We believe that such responses would be even further awayfrom the M. et al. model predictions).
Another plausible reason why consumption responses for age 60 are higherthan the theoretical expectations is that people expect and choose to consumeless in their later years, rather than spreading consumption equally over earlierand later years (Thurow, 1968). If this effect is large in the aggregate, it wouldmake the M. et al. model less accurate.
From Questions 1, 2 and 4: Question 2 specifies savings of $10,000 andQuestion 4 specifies savings of $1,000 plus an inheritance (perhaps the purestsort of windfall) of $10,000. The theoretical expectations at age 30 are almostidentical, being $200 and $222 more than in Question 1. But indicated consump-tion is much much larger for the windfall (Question 4) than for the savings(Question 2). This can be interpreted in either of two ways (or by combinationof the two). It may be that the mere fact that the assets have been handledin the past, and presumably already converted into financial instruments ofsome sort, may lower the propensity of consume from them. The other explana-tion is that windfalls are simply perceived differently than are sums that oneworks for or acquires other than fortuitously, i.e., 'Easy come, easy go'. Nomatter what the true explanation, the results are 'irrational' from the point ofthe M. et al. theory, and may seriously compromise that theory. (The extent ofthe damage may depend partly on how important windfalls are in the economy).
This does not imply that such behaviour, or answers to that effect in thisquestionnaire, would really be irrational. The psychic and perhaps transactionalcosts of establishing a new lifetime plan and choosing a new portfolio may besufficiently great so that after a plan has been made, it makes sense to leave itundisturbed and treat the windfall as outside the lifetime plan.
The easy-come-easy-go notion is weakened somewhat by the evidence ofQuestion 8, however, where respondents earn $10,000 in one-time overtimework. The propensity to consume out of that sum is high, i.e., total consumptionis much higher with $1,000 of assets and $10,000 in overtime pay than with$10,000 in savings (Questions 6), both with a $10,000 salaried income. And acomparison of Question 8, in which the 'extra' money comes from a higher
MIDDLE-CLASS U.S. CONSUMPTION FUNCTION 79
salary this year (and this year only) with Question 4, in which the extra moneycomes from a windfall inheritance, also does not show a great deal higher con-sumption out of the inheritance than out of the special income source.
Taken together, this evidence seems to suggest that savings affect consump-tion in ways that are theoretically expected, but one-time increments of moneyare spent at a considerably higher rate than the theory predicts. (It may also bethat a better theory would incorporate a discounting of future utility.)
The comparison of the results from windfall and savings questions is particu-larly interesting in the context of this method because most or all of the sourcesof bias affect both questions equally (e.g. the problem of purchases of durables).All life conditions are the same, and all economic conditions are the same. Theonly difference between questions is the difference in labels on the funds availableto the person. Therefore it seems reasonable that the difference in behaviourshould be attributed to the difference in labels.
From Questions 1, 4, 5 and 6: Expected future income seems to be dis-counted sharply. The comparison of Questions 4, 5 and 1 shows that even thoughtheoretical consumption is the same for Questions 4 and 5, the people who wouldreceive a windfall in two years (Question 5) behave more like the people whoexpect to receive no windfall (in Question 1) than the people who receive awindfall this year (Question 4), This is indeed a sharp rate of discount of futureincome.
One might also interpret this result as a reluctance to go into debt in orderto realize the optimum lifetime consumption plan. But in the context of thelife cycle hypothesis, such behaviour is not rational.
If one compares question 5 with question 2, however, the outcome is muchthe same, as is theoretically expected. So there are two possible interpretationsfor the outcome of question 5: (a) the future inheritance is thought of as awindfall, but sharply discounted; or (b), for some reason the future inheritanceis thought of in the same way as are present savings.
A sharp rate of discount is seen in the comparison of Questions 1, 3, and 6.In question 6, a big rate of increase in income is expected in two years, whichtheoretically should cause a very large increase in consumption, close to that ofQuestion 3. But in fact the respondents consume in the present little if atall more than they consume in Question 1.
From Questions 7 and 1: The responses to Question 7, in which there isa 50-50 chance of receiving double the income in Question 1, or of receiving noincome at all, reveal an aversion to risk and uncertainty in the income stream,i.e., indicated consumption is far below the theoretical expectation. The expectedvalue of situation 7 is the same as that of question 1, and the theoretical level ofconsumption is therefore the same. However, the consumption responses are farlower in the risky situation.
CONCLUSIONS1. Once more the evidence confirms the permanent-income and life-cycle
theories: Expected income beyond the current year clearly does affectconsumption.
Assets affect consumption in accordance with Pigou, Wicksell, Patinkin,and the M. et al. model.
The propensity to consume out of one-time income in the present periodis apparently higher than the propensity to consume out of savings from pastrevenue.
Age affects consumption in the direction predicted by the M. et al.model. But at later ages the propensity to consume in the present is much higherthan the M. et al. model predicts. This suggests that people (a) discount futureutility, or (b) choose to consume less after retirement, or (c) do not respond toour questions in accordance with the assumption that all income will cease atretirement.
Future revenue is apparently discounted sufficiently sharply to make thesimple Modigliani-Brumberg model unreliable for predicting the effect of changesin future income upon individuals.
Risk and uncertainty in future income sharply reduce the propensity toconsume.
The most important conclusion is that this method produces results thataccord with the combination of the M. et al. theory and common sense, and hencethe method seems to be useful for research on consumption and saving.
REFERENCESAndo, Albert and Franco Modigliani, 'The "Life Cycle" Hypothesis of
Saving: Aggregate Implications and Tests', American Economic Review, Vol. 53,March, 1963, 55-84.
Friedman, Milton, A Theory of the Consum75tion Function (Princeton:Princeton U. Press, 1957).
Friedman, Milton, 'Windfalls, the "Horizon", and Related Concepts in thePermanent-Income Hypothesis', in I'Ieasurement in Economics: Studies inMathematical Economics and Economists in Memory of Yehuda Grunfeld (Stan-ford: Stanford U. Press, 1963).
Gilboy, E. W., 'Demand Curves by Personal Estimate', Quarterly Journal ofEconomics, Vol. 46, 1931, 376-384.
Modigliani, Franco, 'The Life Cycle Hypothesis of Saving, The Demand forWealth, and the Supply of Capital', Social Research Vol. 33, Summer, 1966,160-2 17.
Modigliani, Franco and Richard Brumberg, 'Utility Analysis and the Con-sumption Function: An Interpretation of Cross-Section Data', in K. Kurihara(ed.), Post-Keynesian Economics (New Brunswick: Rutgers U. Press, 1954), 383-436.
Shaffer, J. D., 'Information About Price and Income Elasticity for FoodObtained from Survey Data', Journal of Farm Economics, Vol. 16, February,1959, 113-118.
Simon, Julian L., 'The Cause of the Newspaper Rate Differential: A Sub-jective-Demand-Curve Analysis', Journal of Political Economy, Vol. 73, October,1965, 536-539.
Thurow, Lester, 'The Optimum Lifetime Distribution of ConsumptionExpenditures', American Economic Review, Vol. 59, June, 1969, 324-330.
hebrew University, Jerusalem