What Makes a Good Economy? An Analysis of Survey Data

Text-only Preview

Sam Houston State University
Department of Economics and International Business
Working Paper Series

What Makes a Good Economy? An Analysis of Survey Data

Darren Grant

SHSU Economics & Intl. Business Working Paper No. 09-09
October 2009

This study analyzes nearly twenty-five years of U.S. survey data to determine the
macroeconomic conditions associated with economies the public considers “good.”
These surveys are correlated with, but distinct from, other frequently-studied,
expectations-oriented indices of consumer sentiment. The primary findings are as
follows: 1) inflation and unemployment, the variables in the Phillips curve, explain
much of the variation in the survey data; 2) consumers’ implied loss function is nearly
linear in these two variables; 3) the public is willing to trade off four percentage points
of (increased) inflation for one percentage point of (decreased) unemployment.


Darren Grant
Department of Economics and International Business
Sam Houston State University
Box 2118
Huntsville, TX 77341
[email protected]
This study analyzes nearly twenty-five years of U.S. survey data to determine the
macroeconomic conditions associated with economies the public considers “good.”
These surveys are correlated with, but distinct from, other frequently-studied,
expectations-oriented indices of consumer sentiment. The primary findings are as
follows: 1) inflation and unemployment, the variables in the Phillips curve, explain
much of the variation in the survey data; 2) consumers’ implied loss function is nearly
linear in these two variables; 3) the public is willing to trade off four percentage points
of (increased) inflation for one percentage point of (decreased) unemployment.
** This paper has several figures that are best viewed in color. **
1 Mishuk Chowdhury, Sohna Jaye, Mohammed Khan, Tino Sonora, and Mark Tuttle provided
key assistance in this project, for which I am grateful. Comments from participants at the 2008
Western Economic Association meetings are also appreciated.

1. Introduction.
What macroeconomic conditions are associated with economies the public considers “good”?
There is perhaps no more fundamental question in macroeconomics. Its answer would be useful in
several ways. It would indicate how to best assess the state of the macroeconomy–whether, for
example, to use Okun’s Misery Index as a summary measure. It would clarify the preferences of the
representative agent that populates many macroeconomic models, and thus help shape the loss
functions assumed in those models. And it would cast light on the relevance of other macroeconomic
theories in which the polity’s preferences play a key role, such as those of political business cycles,
which are feasible only if citizen satisfaction with the economy can be meaningfully manipulated over
short time frames.
Three reputable surveys spanning several business cycles have regularly asked respondents
to assess the state of the national economy. In this paper we use these surveys, which have not been
previously analyzed, to answer the question posed above. These surveys are distinct from two others
relied upon regularly by macroeconomists: the University of Michigan’s Consumer Sentiment Index
and the Conference Board’s Consumer Confidence Index. Most of this research is future-oriented,
relating consumer expectations to economic growth, while the present study focuses on assessing
economic conditions in the present. Though both the Michigan and Conference Board surveys have
questions inquiring about current conditions, neither asks for a general assessment of the national
macroeconomy, and our analysis shows that the two sets of questions are distinct statistically.
The paper proceeds as follows. The next section introduces the three surveys, shows how
they relate to each other, and describes how they differ from the Michigan and Conference Board

indices. Section 3 then shows how various macroeconomic variables influence consumers’
assessments of the state of the economy. Policy and modeling implications are then addressed in
Section 4, and Section 5 concludes.
2. Assessing the Macroeconomy.
The Data. Of the three surveys we analyze, the longest-running is a project of ABC News and
occasional partners since December 1985, which asks, “Would you describe the state of the nation's
economy these days as excellent, good, not so good, or poor?” This survey uses an overlapping
design, reporting each week the responses of about one thousand individuals who have been
interviewed over the previous four weeks. The responses to this question and two others, about
personal finances and the buying climate, are combined to form the “Consumer Comfort Index.” We
use the responses to the “good economy” question that are reported in the last week of each month,
which is basically an average of all responses obtained during that month. All empirics in this paper
use the month as the unit of observation, ending in December 2008 except as noted.
The second survey, almost as old but less regular, is conducted by CBS News, generally in
conjunction with the New York Times (NYT). This asks about the current state of the economy:
“How would you rate the condition of the national economy these days–Very Good, Fairly Good,
Fairly Bad, or Very Bad?” It also asks about the change in macroeconomic conditions: “Do you think
the economy is getting better, getting worse, or staying about the same?” Some months contain no
surveys; other months contain multiple surveys, whose responses are then averaged. Responses to
the “good economy” question are present in 147 of the 254 months between October 1987 and

December 2008; responses to the “better/worse” question are present in 131 of those months. (Each
question was occasionally asked prior to October 1987, but too infrequently to be of use here.) There
is a twenty-month gap in coverage in the “good economy” question in 1989 and 1990, and several
gaps of approximately six months in the “better/worse” question. (These will be visible in Figures
1 and 2 as smooth portions of the CBS/NYT lines illustrating the fraction of positive responses to
each of these survey questions.) Each poll contains at least one thousand respondents.
A third poll, conducted by the Gallup organization and sponsored by USA Today, also asks
two questions: one about the current state of the economy, phrased like the ABC News question but
with slightly different response options, and one about the rate of change in the economy, phrased
like the CBS/NYT question but also with different response options. Responses to each question are
reported in 111 and 109 of the 155 months between February 1997 and October 2008, respectively;
there are again roughly one thousand respondents in each of these months. This survey appears to
have been discontinued in October 2008 in favor of the USA Today/IHS Global Insight Economic
Outlook Index, which debuted several months later and is not analyzed here.
Table 1 summarizes the questions and response options in each of these three polls. These
are the only American surveys that explicitly ask for assessments of the national economy and that
span at least a decade.
The first two responses to each survey’s “good economy” question are considered “positive”
responses. Figure 1 presents the time series of each survey’s positive responses, while Figure 2
presents the time series of “better” responses to the “better/worse” questions. The level of positive
responses differs across surveys, consistent with the different phrasing of the questions and answer
choices. But the temporal variation in responses is quite similar, having a strong cyclical component

punctuated with higher frequency modulations. Because of the large samples in each monthly survey,
virtually none of this variation (about 0.1%) is attributable to sampling error, whose standard
deviation never exceeds 1.5 percentage points in any individual survey month. Thus even the high
frequency variation is similar across surveys.
The Michigan and Conference Board surveys also present indices of current conditions (as
opposed to expectations), and these too are represented in Table 1 and Figure 1. The questions asked
by these surveys are distinctly different than those listed above. None asks explicitly about the overall
macroeconomy, but about “business conditions”, “available jobs”, and durable goods purchases; three
of the four questions are about the respondent himself or his local area, instead of the country as a
whole. While the answers to these questions unquestionably have a substantial cyclical component,
the questions themselves do not have the face validity necessary to represent an general assessment
of the national macroeconomy.
Levels and Differences. Do the “good economy” questions and the “better/worse” questions measure
the same construct, one in levels, the other in differences? Over what time frame do respondents
measure, or perceive, changes in the strength of the macroeconomy? To find out, we regressed the
percentage of “better” responses to the CBS/NYT and USA Today/Gallup “better/worse” questions
on leads and lags of the percentage of positive responses “good economy” question asked by ABC
News–the only one that is reported each and every month. We used two month intervals in order to
condense the reporting of results, which are contained in Table 2.
In the more general, first and third regressions, positive coefficients near the current date
offset negative coefficients for the recent past, suggesting a differencing interpretation, which is

reinforced by support for the null that the coefficient sum is zero. The coefficients indicate that the
“better/worse” questions are essentially backward-looking, with hindsight that extends about six
months. To pin down the timing more precisely, we then conducted regressions using every feasible
two-month combination of the ABC News “good economy” measure, to determine which pairs best
explain the responses to the two “better/worse” questions. The second and fourth columns of the
table show the optimal pairs, a backward difference of six months for the shorter USA Today series
and eight months for the longer CBS /NYT series. In both cases, the R² statistic indicates there is
a minimal loss of explanatory power compared to the full set of leads and lags.
These findings indicate a reasonable degree of concordance between the level and difference
assessments of the macroeconomy. However, there is an important dimension of the “better/worse”
measures that is not captured in the “good economy” question. The R² values, ranging from 0.53 to
0.73, are not that close to one. Accordingly, the analysis below will attempt to discern the
macroeconomic determinants of each measure.
Latent Variables. It is valuable to have a time series that completely and succinctly summarizes the
complete results of each survey question for the full time period over which it is asked. This can be
done by combining two statistical techniques, an ordered probit model and transformation regression.
The former views the responses to any survey question as governed by an underlying latent variable
and a set of thresholds that distinguish an “excellent” response from a “good” response, and so on.
Random (standard normal) variation in individual circumstances (or perceived thresholds) yields the
cross-section variation in responses at any given point in time.
The latter technique expresses time as a series of splines, which can then be treated as

independent variables in the ordered probit model. Applying the estimated coefficients to the splines
yields a smoothed version of the latent variable that underlies the response percentages in each
survey. This is estimated for the full time span of the survey, “filling in” any survey-less months. The
extent of smoothing is governed by the number of splines; we use a large number (fifty-two over the
full twenty-three year period) to smooth only gently, preserving all but the highest-frequency
Figure 3 presents the results for all five questions, vertically scaled so that all latent variables
have the same mean for the periods in which they overlap. (We do not present confidence intervals,
as they are so small.) For both the “good economy” and “better/worse” questions the latent variables
are as similar as the cutoffs, driven by the differences in response options, are different. Correlations
between the “good economy” latent variables are each 0.99; the correlation between the
“better/worse” latent variables is 0.96.
For the “good economy” question the thresholds that convert values of the latent variable into
predicted discrete responses are 1-1.5 units apart, so a corresponding increase in the latent variable
implies each respondent improves her assessment of the macroeconomy by one notch. The same is
true for the CBS/NYT “better/worse” measure, but not for the corresponding USA Today/Gallup
measure, which only gives two response options (though respondents sometimes choose a third–see
Table 1). Ironically, the percentage of “better” responses to this latter question are actually a better
cardinal measure of sentiment–more closely correlated to the latent variable–because the centrally-
located threshold between the answer choices tends to be “closer to the margin.”

Dimensionality. These latent variables are, in effect, indices themselves, and can be statistically
compared to the Michigan and Conference Board indices, in order to distill out their similarities and
differences. A principal component analysis is a simple way to do this. Table 3 presents the results
of a complete analysis of all nine variables, along with a correlation matrix. To integrate the “good
economy” and “better/worse” series and address potential stationarity concerns, eight month
differences are taken of all variables except the two “better/worse” series, consistent with the findings
in the previous subsection. Analogous analyses, in levels and using subsets of series that overlap for
longer periods of time, yield similar results.
The nine series included in this analysis can be placed into four groups: “good economy”
series, Michigan and Conference Board current conditions indices, expectations indices, and
“better/worse” series. This grouping is strongly supported by both the correlation matrix, in which
intra-group correlations greatly exceed most inter-group correlations, and the principal components
analysis, in which the within-group factor loadings are almost always similar.
In fact, except for the current conditions indices, within-group correlations are approximately
0.9. Cross-group correlations range from about 0.5 to about 0.7, suggesting that there are four
distinct dimensions represented by these four groups. This conclusion is reinforced by the principal
component analysis.
This analysis yields a set of independent components that explain the overall variation of all
nine series. Because so little of these series’ variance is attributable to sampling error, most
components could be considered statistically significant, but economic significance is restricted to
the first four, which together explain 95% of the variance. These components’ factor loadings, in
Table 3, are easily interpretable. The first, dominant component is a simple almost-unweighted

average of the seven series, and reflects basic business cycle variation. The second component
distinguishes measures of expectations from everything else. The distinctive nature of the Michigan
Current Economic Conditions Index is apparent here; in this component it is (slightly) more closely
aligned with the expectations series than the “good economy” series. This reflects the predictive
orientation of this index, which is based on the change in respondents’ personal financial situations
and the environment for durable goods purchases. The third component distinguishes the
“better/worse” series, while the fourth component represents a difference between the “good
economy” series and the Michigan/Conference Board current conditions indices.
The cyclical component explains 72% of the joint variance of these series. Of the remaining
28%, about 3/7 is contributed by the distinctiveness of the expectations indices, and another 3/7 by
the distinctiveness of the two survey questions analyzed here, with the remainder essentially noise.
In summary, responses to the “good economy” questions asked in the ABC News, CBS/NYT,
and USA Today/Gallup surveys are, despite some differences in wording and response options, very
strongly related, yet distinct from expectations measures and from the current conditions indices
published by the University of Michigan and the Conference Board. Responses to the “better/worse”
questions are related, but not identical, to temporal differences of the good economy series. We now
uncover these series’ macroeconomic determinants, using the latent variables as dependent variables
in the regression analyses in the next section. Because of the close correlation of each set of the
latent variables we focus on the longest series of each type: ABC News, for the “good economy”
question, and CBS/NYT for the “better/worse” question. Both exceed twenty years in length.

Document Outline

  • wp09-09_paper_temp0.pdf
  • wp09-09_paper_temp.pdf