A Comparison of the University of Michigan and Conference Board Indices of Consumer Economic Attitudes

John C. Mowen, Oklahoma State University
Clifford E. Young, Oklahoma State University
Patriya Silpakit, Oklahoma State University
ABSTRACT - The indices to assess consumer economic attitudes developed by the University of Michigan and the Conference Board were compared over the time span 1964-1983. The analysis investigated the variables predictive of the two indices. Stepwise regression analysis indicated that greater variance (R2 = .84) could be accounted for in the University of Michigan Index of Consumer Sentiment than in the Conference Board Index (R2 = .60). The indices were also analyzed via canonical correlation, and two roots were extracted. The independent variables explained 86 percent of the variance in the first root and 34 percent of the variance in the second root. Results were discussed in terms of the composition of the two indices of consumer economic attitudes.
[ to cite ]:
John C. Mowen, Clifford E. Young, and Patriya Silpakit (1985) ,"A Comparison of the University of Michigan and Conference Board Indices of Consumer Economic Attitudes", in NA - Advances in Consumer Research Volume 12, eds. Elizabeth C. Hirschman and Moris B. Holbrook, Provo, UT : Association for Consumer Research, Pages: 532-537.

Advances in Consumer Research Volume 12, 1985      Pages 532-537

A COMPARISON OF THE UNIVERSITY OF MICHIGAN AND CONFERENCE BOARD INDICES OF CONSUMER ECONOMIC ATTITUDES

John C. Mowen, Oklahoma State University

Clifford E. Young, Oklahoma State University

Patriya Silpakit, Oklahoma State University

ABSTRACT -

The indices to assess consumer economic attitudes developed by the University of Michigan and the Conference Board were compared over the time span 1964-1983. The analysis investigated the variables predictive of the two indices. Stepwise regression analysis indicated that greater variance (R2 = .84) could be accounted for in the University of Michigan Index of Consumer Sentiment than in the Conference Board Index (R2 = .60). The indices were also analyzed via canonical correlation, and two roots were extracted. The independent variables explained 86 percent of the variance in the first root and 34 percent of the variance in the second root. Results were discussed in terms of the composition of the two indices of consumer economic attitudes.

INTRODUCTION

A rarely asked question concerns the issue of how much variance in consumer buying behavior is controlled by the variables commonly studied by consumer behaviorists. The authors suspect that if a way of calculating how much of buying behavior is controlled by information processing variables, such as cognitive responses, the answer would be very little. This focus on "micro" variables has recently been criticized (Sheth 1979 ). A more "macro" approach to consumer behavior could yield dividends, if it resulted in the ability to explain the combined actions of the millions of consumers making up consumer markets.

The study of psychological economics (Katona 1974) has been proposed as one approach for predicting the actions of large numbers of consumers within an economy (Didow, Perreault, and Williamson 1983; Young, Mowen, and Silpakit 1984). Founded by George Katona, the field of psychological economics is based upon the importance of the consumer in modern western economies. Post World War II consumers have had discretionary income to spend as they wish after fulfilling their basic consumption needs. Whether aggregates of consumers choose to spend or save this discretionary income can have major effects on the economy.

Katona (1974) proposed that discretionary spending was based on a subjective factor, called consumer sentiment, in addition to the traditional economic variables such as household income, interest rates, employment levels and so forth. When consumer sentiment rises, consumers tend to loosen their purse strings; conversely, when sentiment becomes more pessimistic, discretionary spending slows. The outcome of reduced spending is that a falloff in demand occurs, retail inventories grow, orders to manufacturers slow, layoffs occur, and recession is born. When consumer sentiment rises, the economic optimism results in greater discretionary spending. As a consequence, inventories fall, retailers order, manufacturers recall workers, and the economy begins to expand.

In order to assess the changes in consumer sentiment, Katona developed an index which was obtained from periodic surveys of consumers. Three basic areas of consumer economic confidence were assessed in the surveys C consumer perceptions of their personal financial position, their business outlook, and their evaluation of market conditions. These surveys of consumer sentiment have been conducted since 1946 at the Survey Research Center, University of Michigan.

Competing measures of consumer economic confidence have appeared since Katona developed his measure. In particular, the Conference Board has created an Index of Consumer Confidence and run surveys on a quarterly basis since 1964. The purpose of the Conference Board Index is similar to that of the Michigan index, however, both the questions and the survey research approach are different. The Index of Consumer Confidence is based upon five attitudinal questions concerning the consumer's view of business conditions and employment, both now and six months into the future, and the consumer's view of his or her own income six months in the future.

The published research on consumer economic confidence has been devoted entirely to the Michigan Index of Consumer Sentiment. The early work investigated either the ability of the index to forecast the economy or the economic variables which were predictive of the index. The early studies revealed that when economic variables were used to predict the index, as much as 80 percent of the variance of the index could be accounted for (Hymans 1970). The index was also found to improve the predictive ability of economic variables to forecast the economy (Adams and Duggal 1974; Hymans 1970).

Recent work investigating the indices of consumer economic confidence has been sparse. Didow, Perreault, and Williamson (1983), using a cross-sectional optimal scaling analysis, raised some questions about the procedures used to measure and report the Michigan Index. For example, the researcher found that identical values on the index could be obtained despite quite different response distributions to the questions due to the method of scoring the responses. Despite these findings, the index does appear to have pragmatic validity. It is also the most frequently used of the indices and is even reported in the U.S. Commerce Departments Business Conditions Digest.

Young, Mowen, and Silpakit (1984) replicated and extended the earlier work by Hymans (1970) in order to identify the economic variables predictive of the Michigan index. Their results showed that the lag of average stock prices, the lag of the consumer price index, and the lag of the 60-day commercial paper rate accounted for 79 percent of the variance of the index. When the lag of the index was included in the equation, as Hymans had done, the variance accounted for rose to 87 percent. These researchers also found that the 60-day commercial paper rate, which Hymans (1970) had not investigated, became predictive of consumer sentiment only after 1974 when economic conditions changed due to government policies and world economic conditions.

THE STUDY'S PURPOSE

The purpose of the present study is to compare and contrast the variables predictive of Michigan and Conference Board indices. In the first phase, the research extends the work of Young, Mowen, and Silpakit (1984) on the Index of Consumer Sentiment to the Conference Board's Index. In addition, variables previously analyzed by Adams and Green (1965) are added to the analyses. Specifically, phase one of the study explores the economic variables predictive of the two indices through stepwise regression analysis.

In the second phase of the study, a canonical correlation analysis is performed in order to assess the combined properties of the scales. The idea is that the two indices can be viewed as independent measures of consumer economic confidence and could be used jointly in macro consumer analyses. The canonical correlation will provide information on how the independent variables are related to the linear combination of the two indicators of consumer economic confidence.

METHODOLOGY

The Data

The University of Michigan's Index of Consumer Sentiment and the Conference Board's Index of Consumer Confidence were the dependent variables in the study. The indices and the various predictor variables were collected from the first quarter of 1964 to the first quarter of 1983. The total number of data points actually used were 64, due to the necessity to develop lags in a number of independent variables. Where quarterly data were not reported 9 monthly data were averaged over three consecutive months. Where discrepancies in data values from different sources were found, the most recently published values were chosen.

Operationalization of the Predictor Variables.

In the first phase of the study, stepwise regressions were developed for each of the consumer economic attitude indices. The variables chosen for inclusion in the exploratory study were those originally identified by Hymans (1970), Adams and Green (1965), and Young, Mowen, and Silpakit (1984). Table 1 identifies the independent variables investigated.

TABLE 1

LIST OF VARIABLES USED IN THE ANALYSES

A number of the variables require explanation. Disposable income was the ratio of real disposable income (net of transfer payments) lagged one quarter to a lagged eight-quarter average of real disposable income. According to Hymans (1970), if the eight quarter average represents the level of income to which consumers are well adjusted, the ratio measures any deviation from that average. Thus, the measure assesses how the immediate tone of business activity diverges from baseline conditions.

Stock prices from the Standard and Poor's index were operationalized as the lagged rate of increase from the preceding quarter. Furthermore, the rate of increase of a lagged four-quarter average of common stock prices was also used as another independent variable. This measure was intended to assess the underlying market trend, while the former measure assessed more transitory movements in the stock market.

The consumer price index was included to represent inflation, because persistent inflation could be expected to influence consumer attitudes (Katona 1974). It was operationalized as the lagged ratio of current consumer prices to an eight-quarter average of consumer prices. The use of the ratio picks up how current prices diverge from baseline changes.

The accession-layoff variable was included to obtain information on factors related to the employment picture faced by consumers. Other variables investigated included change in net orders, new housing units, and the length of work week. All of these variables were lagged one quarter.

Johansson, Bagozzi, and Sheth (1982) argued that straight redundancy analysis or simultaneous equations systems are more appropriate than canonical analysis. For two reasons, however, the authors chose to use canonical correlation followed by a redundancy analysis, rather than a simultaneous equations model. First, the authors could state no a priori model of the causal relations among the variables. The study was exploratory, making a structural equation approach inappropriate. Second, straight redundancy analysis tends to be used when the goal is to obtain a single overall measure of a construct. In the present study, the authors were as interested in finding the differences in what the indices measure as in their similarities. Based on this study, future researchers can develop more explicit forms of the relationships among the variables.

RESULTS

A stepwise regression analysis was performed using the Index of Consumer Sentiment as the dependent variable, followed by the same type of analysis using the Conference Board's Index of Consumer Confidence. Final equations are presented in Tables 2 and 3 and include independent variables reaching the .05 level of significance.

Inspection of the results reveals that the consumer price index entered both equations first, indicating its superior predictive ability for the Michigan index and for the Conference Board index. The remaining two independent variables reaching significance in both equations were average stock price ratio and work week. Work week entered the equation before average stock price ratio with the Conference Board index as the dependent variable, while the reverse was true with the Michigan index as the dependent variable.

A total of six economic variables reached the .05 level of significance in predicting the Michigan index, with an associated R2 of .84. Only three of the potential variables reached significance, however, in predicting the Conference Board index with an associated R2 of .58. Clearly, more of the Michigan index variance was predicted with the set of economic variables used in this study.

TABLE 2

STEPWISE REGRESSION RESULTS ON THE MICHIGAN INDEX

TABLE 3

STEPWISE REGRESSION RESULTS ON CONFERENCE BOARD INDEX

Inspection of the coefficients of the independent variables in the equations for predicting the Michigan index reveals some potentially disturbing results. Two of the independent variables have signs associated with their coefficients that are counter intuitlve. The signs for average disposable income and accession-layoff are negative. A priori, one would expect increased income and greater net hiring to be indicative of improving economic conditions.

Quite likely this counter intuitive result occurred because of high multicollinearity in the data, attributable both to the variables themselves and to the fact that the data are time series data with underlying common cycles. Simple bivariate correlations of the variables are presented in Table 4. Inspection of the the correlations of disposable income and accession-layoff reveal that they are positively related with the sentiment variables as expected. However, the high multicollinearity of the variables, particularly with the consumer price index, probably caused the regression coefficients to turn negative in the equation.

To analyze the overall relationship between the set of economic indicator variables and the consumer economic attitudes, a canonical analysis was performed. Results are presented in Table 5. Both potential canonical roots were highly significant and were retained for subsequent analysis. Inspection of the loadings of the variables on the variates reveals high loadings for both the Michigan index and the Conference Board index on the first variate, whereas the Conference Board index is the dominant loading dependent variable on the second variate.

TABLE 4

BIVARIATE CORRELATTONS OF VARIABLES

Independent variables loading more highly on the first variate included the consumer price index, commercial paper rate, stock price ratio, and average stock price ratio. Variables loading more highly on the second variate included average disposable income, work week, and accession-layoff rate. Change in net orders and housing starts did not load highly on either variate.

In conjunction with the canonical analysis, redundant variance calculations were performed and are presented in Table 6. Inspection of the results reveals that the Michigan index has particularly high redundant variance (.849) with the first variate whereas the Conference Board index exhibits only about half (.436) the amount. Independent variables having substantial redundant variance with the first variate include consumer price index, commercial paper rate, work week, accession-layoff rate, average stock price ratio and average disposable income.

TABLE 5

RESULTS OF CANONICAL ANALYSIS ON THE MICHIGAN INDEX AND CONFERENCE BOARD INDEX

TABLE 6

REDUNDANT VARIANCE FOR EACH VARIABLE FOR EACH CANONICAL VARIATE

For the second variate, effectively all (.170) of what redundant variance exists for the dependent variables comes from the Conference Board index. Independent variables having substantial redundant variance with the second variate include average disposable income, accession-layoff, and work week.

As a final aid in interpretation, the pattern matrices were rotated using varimax criterion on the dependent variables. Results of the rotation are present in Table 7.

TABLE 7

RESULTS AFTER VARIMAX ROTATION ON DEPENDENT VARIABLES

DISCUSSION

Interpreting canonical analysis is difficult at best. The tentative findings of this study were that the University of Michigan Index of Consumer Sentiment and the Conference Board Index of Consumer Confidence measure different concepts. The redundancy analysis revealed that the Michigan index exhibits its shared variance almost exclusively with the first root. The Conference Board index also shares variance with the first root, but in addition shares some variance with the second root.

Thus, it appears that the Index of Consumer Sentiment is capturing consumer reactions to what is happening with financial related factors (e.g., prices and interest rates) and to a lesser extent, such variables as the work week, the stock market, and the employment picture. In contrast, the Index of Consumer Confidence is picking up some additional information on employment related variables, such as the length of the work weeks the accession-layoff rate, and disposable income.

Based on which variables loaded most highly on the two roots, the authors gave them names. The name chosen for the first root was "value of assets in the future." The name selected for the second root was "future employment optimism." The indices did reveal a moderate amount of overlap, as shown by their bivariate correlation (r-.79). However, their differences were captured substantially by the two roots extracted by the canonical analysis.

Thus, it would be appropriate to describe Michigan's Sentiment index as measuring predominantly consumer perceptions of their future financial assets and how they will be affected by interest rates, inflation, the stock market, the length of the work week, and so forth. In contrast, the Conference Board Index appears to be measuring predominantly consumer perceptions of their likely employment prospects and how they will be affected by the length of the work week and layoffs and hires.

Why do the indices in part reflect different constructs? A likely explanation lies in the types of questions which each ask. The Michigan index asks five basic questions (Curtin 1982). Two questions focus on the past and expected changes in personal finances, two questions target the short and long-term business outlook, and one concerns whether buying conditions are appropriate to purchase large consumer durables.

The Conference Board index also has five basic questions. Two concern current business conditions and what they will be six months into the future. Two concern current employment and employment six months hence. One asks about expected income six months from now. Quite clearly, the Conference Board index emphasizes the employment situation substantially more than does the Michigan index. In two of its questions it asks whether Jobs are plentiful, not so plentiful, or hard to get. The Michigan index never mentions employment at all, other than asking whether over the next five years could there be widespread unemployment or depression.

The finding that the two indices are in part measuring different constructs has substantial implications. The media tents to view the indices as identical and uses them interchangeably. However, the Conference Board index appears to include an employment component that is independent of what is contained in the Michigan index. The Michigan index appears to focus primarily on financial asset maintenance. Importantly, employment opportunities and financial asset value may not always covary.

Fabian Linden (1982), the director of the Conference Board index, noted that during the late 1970s the indices diverged with Michigan's drifting down and the Conference Board's fluctuating. One reason could be that the high inflation of the late 1970s could have caused the downward trend of the Michigan index, while the Conference Board's index responded more to consumer perceptions of the employment picture, which was generally improving.

A second finding of the study was that the economic variables captured substantially more variance in the Michigan index than in the Conference Board index (R2 of .84 versus .58). Two factors could account for the divergence. First, it may be that the Conference Board Index is picking up more of a "psychological" component of economic attitudes than is the Michigan index. If such a component exists, and accounts for some proportion of future spending separate from purely economic factors, the Conference Board index could be of greater utility in forecasting future consumer spending.

The other possibility is that the lower explained variance resulted from the Conference Board index having a greater error component than the Michigan index. It remains for future research to identify the reasons for the differences in the indices.

In order to accomplish this task, a theoretical model must be developed. The model should describe the relationships among the economic and psychological variables related to consumer optimism and to consumer spending patterns. A possible approach to theory development involves linking Katona's work more tightly to traditional economic analyses of consumption. For example, Friedman (1957) developed the permanent income hypothesis. This model takes into account a household's expected future income as a predictor of permanent consumption. Friedman also distinguished the concepts of permanent income and transitory income as well as permanent consumption and transitory consumption.

These ideas are generally consistent with those of Katona in which consumers adjust their spending based upon an economic forecast. Such economic optimism may be based wholly or in part on expected changes in permanent or expected income. It may be possible to bridge the ideas of Katona and Friedman or other theorists, eg., Modigliani et. al's (1966) life cycle theory, in order to add theoretical flesh and bones to Katona's work.

Future Research

The study delineated a number of future research needs. The next step in the study of the indices of consumer economic attitudes involves developing a model which depicts the structural relationships among the economic variables, the sentiment indices, and consumer buying patterns. A second area of research, not unrelated to the first, concerns finding variables to account for the remaining unexplained variance in the measures. The economic variables accounted for 84 percent of the variance in the Michigan index. The authors suspect that a part of the remaining variance may consist of a general mood factor. That is, how does the public feel about themselves, the United States, its political system, and its place in the world? As noted in the introduction, the Index of Consumer Sentiment adds to the predictive ability of econometric models of consumer spending. Thus, the index seems to be picking up something, in addition to that accounted for by economic variables. The general mood factor could account for that missing variance.

A third research need involves performing analyses to prefilter and prewhiten the data (Catalano, Dooley, and Jackson 1983). Time series data have severe auto-regressive problems which future work needs to begin dealing with. For example, in the present research the signs of the coefficients for disposable income and accession-layoff are negative in the stepwise regression on the Michigan index, although their bivariate correlations are positive. It is quite likely that high multicollinearity due to underlying auto-correlation problems with the time series data is the reason for this result.

A final research need involves a further comparison of the Michigan and Conference Board indices. In particular, how does their ability to predict consumer spending compare? Given that the indices do measure different components of consumer economic attitudes, it may be necessary to use both to forecast consumer spending.

REFERENCES

Adams, F. Gerard and Vijaya G. Duggal (1974), "Anticipation Variables in an Econometric Model,' International Economic Review, 15 (June), 267-84.

Adams, F. Gerard and Edward W. Green (1965), "Explaining and Predicting Aggregate Consumer Attitudes," International Economic Review, 6 (September), 275-92.

Catalano, Ralph A., David Dooley, and Robert Jackson (1983), "Selecting a Time-Series Strategy," Psychological Bulletin, 94, 506-23.

Curtin, Richard T. (1982), "Indicators of Consumer Behavior: The University of Michigan Surveys of Consumers," Public Opinion Quarterly, 46, 340-52.

Didow, Nicholas N., William D. Perreault, and Nicholas T. Williamson (1983), "A Cross-Sectional, Optimal Scaling Analysis of the Index of Consumer Sentiment," Journal of Consumer Research, 10, (December), 339-347.

Dornbusch, Rudiger and Stanley Fischer (1981), Macro- Economics, 2nd ed., New York: McGraw-Hill, Inc.

Friedman, Milton (1957), A Theory of the Consumption Function, Princeton, NJ: Princeton University Press.

Hymans, Saul H. (1970), "Consumer Durable Spending: Explanation and Prediction," Brookings Papers on Economic Activity, 1, 173-99.

Johansson, J. K., R. P. Bagozzi, and Jagdish N. Sheth (1982), "Canonical Correlation: Is it ever useful?" in An Assessment of Marketing Thought & Practice, Bruce J. Walker, et al, eds., Chicago: American Marketing Association, 386-89.

Katona, George (1974), "Psychology and Consumer Economics," Journal of Consumer Research, 1 (June), 1-8.

Linden, Fabian (1982), "The Consumer as Forecaster," Public Opinion Quarterly, 46, 353-60.

Modigliani, Franco (1966), 'The Life Cycle Hypothesis of Saving, the Demand for Wealth and he Supply of Capital," Social Research, 33 (2).

Sheth, Jagdish N. (1979), "The Surpluses and Shortages in Consumer Behavior Theory and Research," Journal of the Academy of Marketing Science, 7 (Fall) 414-27.

Young, Clifford E., John C. Mowen, and Patriya Silpakit (1984), "Predicting Consumer Confidence and Sentiment: A Replication and Extension," in 1984 AMA Educators' Conference Proceedings, Russell Belk, et al, eds., Chicago: American Marketing Association.

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