asked people to set aside their fears and spend money. In his first fireside chat,
March 12, 1933,8 he appealed to morality, asking people not to withdraw more
money than they needed when the banks reopened. He was spinning a narrative
of what could happen if unreasoning people with little social consciousness
destroyed the economy. We can speculate that President Roosevelt’s request
worked because it was based on a moral standard; his chats roughly coincided
with upturns in the US economy. However, we do not have a way of quantifying
exactly how salient the narratives of the time really were. We would know more,
perhaps, if economists had collected better data and conducted more analysis on
what people were saying in 1933. If they had, we might now have a better
understanding about how to frame such moral-appeal narratives in the future.
A problem in using narratives to forecast economic variables is that human
judgment and discourse about narratives tend to be politicized and emotion-
ridden. It has been difficult for scholars to research popular narratives, focusing
on the core elements that make them contagious, without being accused of
taking sides in political, or sometimes religious, controversies. Because many
professional economists try to remain nonpartisan, they tend to rely on
quantitative, rather than qualitative, observations. However, with modern
information technology, economists can now collect data on economic narratives
themselves, on their essential elements of meaning, without being overly focused
just on words, and they can model the transmission of narratives. If we maintain
quantitative rigor, we can make narrative epidemics a part of economic science.
Some may doubt that it is possible to have nonpartisan discussion of
economic narratives. However, if we are careful and polite, it should be possible
to speak in a nonpartisan way about epidemics of economic narratives. Most
people have some instinct about how to speak in a nonpartisan way, and they do
so when the occasion demands it. We do not have to go so far in our efforts to be
nonpartisan that we exclude study of some ideas and emotions that drive
economic changes.
Economic research is already on its way to finding better quantitative
methods to understand narratives’ impact on the economy. Textual search is a
small but expanding area. A search of the NBER working paper database finds
fewer than one hundred papers with the phrase textual analysis. Economists have
used textual analysis to document changes in party affiliation (Kuziemko and
Washington, 2015), political polarization (Gentzkow et al., 2016), and news and
speculative price movements (Roll, 1988; Boudoukh et al., 2013). Much more

could be done. For example, economists could carry the historical analysis
further into databases of personal diaries, sermons, personal letters, psychiatrists’
patient notes, and social media.

Collecting Better Information about Changing Narratives
Should Start Now
Economists must make more serious efforts to collect time-series data on
narratives, going beyond the passive collection of others’ words, toward
experiments that reveal meaning and purpose. Such great quantities of digitized
data are now available that it boggles the imagination. Even so, this vast dataset
is minuscule compared to the even vaster universe of human communications
that go on every day, most of which are not adequately sampled, described, or
understood.
It is important that such data collection be maintained on a consistent basis
through decades, so that we can make intertemporal comparisons of major
influencing public narratives in the future. There has been relatively little
incentive to undertake such a project, because there is little immediate payoff to
doing so. Instead, most narrative data collection focuses on immediate interests,
such as marketing specific products or predicting upcoming elections.
It is also important to apply creative energies toward such consistent long-
term data collection. Understanding people, their behavior, and their thinking
may even require the help of psychoanalysts and philosophers.
It will be difficult to combine these two needs, consistency through time and
creativity. But we must do so if we are to make real progress in narrative
economics.
The first step requires improving existing search engines so that they can
better measure the time-varying incidence of narratives. The search engines do
not tell us exactly how they determine the estimated total number of hits. Rather,
they are designed primarily to help users find articles or information they are
looking for. Thus some anomalies pop up when researchers attempt to count the
number of references. For example, Google’s search engine instructions say that
a search for a phrase should enclose the phrase in quotation marks so that the
search is confined to exactly those words in exactly that order. But sometimes
including the phrase in quotation marks results in more hits than the phrase
without quotation marks. A Google spokesman says that the greater number of
hits for the phrase in quotation marks may happen because quotation marks
cause Google to “dig deeper” into the database.9 We need to see evidence that
such deeper digging is not compromising the accuracy of counts. Google