Reported Effects vs. Revealed-Preference Estimates: Propensity to Spend Tax Rebates (NBER Working Paper 23920)

Summary

NBER Working Paper No. 23920, Reported Effects vs. Revealed-Preference Estimates: Evidence from the Propensity to Spend Tax Rebates, by Jonathan A. Parker and Nicholas S. Souleles, September 2017, revised March 2019. The abstract states that the paper evaluates the consistency of two methods for estimating the effect of an economic policy, asking people how the policy changed their behavior and inferring the change from data on behavior. Using federal stimulus payments disbursed quasi-randomly in 2008, it finds larger revealed-preference estimates of spending propensities for households who report greater spending responses, and similar average propensities from the two methods. It states that evidence is mixed on the relationship between spending propensities and liquidity.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Issuer: Source documents
Document type: Reported Effects Versus Revealed Preference Estimates Evidence From The Propensity To Spen
Case: Reported Effects Versus Revealed Preference Estimates Evidence From The Propensity To Spend Tax Rebates

Cite this document

“Reported Effects vs. Revealed-Preference Estimates: Propensity to Spend Tax Rebates (NBER Working Paper 23920).” Original at www.nber.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/reported-effects-versus-revealed-preference-estimates-evidence-from-the-propensity-to-spend-tax-rebates/, SHA-256 05d3873f192db131017cdc912d6c39e00a4d7c3721e4bbce9cbe08f24b502b50.

Full text

NBER WORKING PAPER SERIES

REPORTED EFFECTS VS. REVEALED-PREFERENCE ESTIMATES:
EVIDENCE FROM THE PROPENSITY TO SPEND TAX REBATES

Jonathan A. Parker
Nicholas S. Souleles

Working Paper 23920
http://www.nber.org/papers/w23920

NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
September 2017, Revised March 2019

This paper first circulated as a working paper as “Reported Preference vs. Revealed Preference:
Evidence ...” We thank Michael Boutros, Christian Broda, Jeff Campbell, Zvi Hercowitz, Erik
Hurst, David S. Johnson, Arik Levenson, Maarten Meeuwis, Robert McLelland, Scott Nelson,
Claudia Sahm, and Matthew D. Shapiro, as well as participants at seminars at the Bank of
Canada, the Board of Governors of the Federal Reserve System, Brandeis, Columbia, Chicago,
Georgetown, MIT, UT Austin, and the 2017 ASSA Meetings. Colin Gray, Yijun Liu, and
Fangzhou Lu provided excellent research assistance. We also thank the staff of the Division
of Consumer Expenditure Surveys at the Bureau of Labor Statistics and Ed Grove, Matt Knain,
and Molly Hagen at Nielsen for their work implementing our survey questions and their careful
explanation of the data. Parker thanks the MIT Sloan School of Management, the
Kellogg School of Management at Northwestern University, the Initiative for Global
Markets at the University of Chicago, and the Zell Center at the Kellogg School of Management
for funding. Some of our results are calculated based on data from The Nielsen Company (US),
LLC and marketing databases provided by the Kilts Center for Marketing Data Center at The
University of Chicago Booth School of Business. The conclusions drawn from the Nielsen data
are those of the researchers and do not reflect the views of Nielsen. Nielsen is not responsible
for, had no role in, and was not involved in analyzing and preparing the results reported herein.
The views expressed herein are those of the authors and do not necessarily reflect the views of the
National Bureau of Economic Research.

At least one co-author has disclosed a financial relationship of potential relevance for this
research. Further information is available online at http://www.nber.org/papers/w23920.ack

NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.

© 2017 by Jonathan A. Parker and Nicholas S. Souleles. All rights reserved. Short sections of
text, not to exceed two paragraphs, may be quoted without explicit permission provided that full
credit, including © notice, is given to the source.
Reported Effects vs. Revealed-Preference Estimates: Evidence from the propensity to spend
tax rebates
Jonathan A. Parker and Nicholas S. Souleles
NBER Working Paper No. 23920
September 2017, Revised March 2019
JEL No. B40,C42,D14,E21,E62,H31

ABSTRACT

We evaluate the consistency of two methods for estimating the effect of an economic policy: i)
asking people how the policy caused them to change their behavior (reported effects); ii) inferring
this change using data on behavior and differences in treatment across people (revealed-
preference estimates). Both methods are widely used to measure spending caused by increases in
liquidity. Using Federal stimulus payments disbursed quasi-randomly in 2008, we find larger
revealed-preference estimates of spending propensities for households who report greater
spending responses, and the methods produce similar average propensities. But evidence is mixed
on the relationship between spending propensities and liquidity.

Jonathan A. Parker
MIT Sloan School of Management
100 Main Street, E62-642
Cambridge, MA 02142-1347
and NBER
JAParker@MIT.edu

Nicholas S. Souleles
Finance Department
The Wharton School
2300 SH-DH
University of Pennsylvania
Philadelphia, PA 19104-6367
and NBER
souleles@wharton.upenn.edu

A data appendix is available at http://www.nber.org/data-appendix/w23920
The foundation of testing and estimation in economics is the revealed-preference ap-
proach in which inferences about causal eﬀects are drawn from people’s actions in diﬀer-
ent situations. The alternative to the revealed-preference approach is to rely on people
to make the causal inferences themselves by asking them to report their choices in diﬀer-
ent hypothetical situations. This alternative is prevalent in other fields, such as history
and psychology, where people’s reports of what they would have done in counterfactual
situations are commonly used as evidence, and it has always had a presence in economic
theory. It is also becoming more widespread in empirical economics: major survey datasets
contain estimates of preference parameters derived from this approach, structural models
are fitted to reported changes in behavior, and researchers are using reported responses
to evaluate policy.1 Most closely related to our analysis, there has been a recent boom
in research on consumption and saving behavior using reported causal eﬀects, with the
dual goals of distinguishing among models of household saving behavior and better un-
derstanding stabilization policy.2
Traditional revealed-preference estimation requires the economist to observe diﬀer-
ences in circumstances that are either plausibly exogenous or well-enough understood
as to be used in structural estimation. Surveying people to directly report causal ef-
fects appears comparatively simple. Further, the survey approach can elicit behavioral
responses to idealized ceteris paribus experiments through the careful design of survey
instruments. The pivotal question is whether the elicited behavior corresponds to the ac-
tual behavior that would occur were the counterfactual of interest not just hypothetical.
Do people fully understand the scenario(s) in the question, do they accurately determine
what their own behavior would have been, and do they respond honestly? Friedman and
Wallis (1942, p. 179-80), discussing an early application of this methodology in Thurstone
(1931), famously wrote “Questionnaires or other devices based on conjectural responses
to hypothetical stimuli . . . are valueless because the subject cannot know how he would
react.”
This paper evaluates the use of surveys to directly elicit causal eﬀects by analyzing
whether the spending that people report as caused by lump-sum tax rebates — specifically
the Economic Stimulus Payments of 2008 — matches the spending that the same people
are estimated to have done in response to the same rebates using revealed-preference
methods. We estimate the revealed-preference propensity to spend following Johnson,
Parker and Souleles (2006), Parker et al. (2013) and Broda and Parker (2014), identifying
the spending propensity from the quasi-randomly-timed disbursement of the rebates across
groups of households. We measure reported spending eﬀects using both the Shapiro and
Slemrod (1995, 2003a, 2003b, 2009) survey instrument and our own instrument designed

1
See for example Barsky et al. (1997), Ameriks et al. (2016), and D’Acunto, Hoang and Weber (2016)
respectively.
2
Questions asking respondents to report spending propensities have recently been added to surveys
run by the Federal Reserve Bank of New York, the Bank of England, and the Bank of Italy. While large
sample survey studies of reported preferences go back to at least Juster and Shay (1964), in addition to
papers discussed later, the boom in research is exemplified by Smeeding, Phillips and O’Connor (2000),
Coronado, Lupton and Sheiner (2005), Leigh (2012), Crump et al. (2015), Graziani, van der Klaauw and
Zafar (2016), Auclert (2017), Bunn et al. (2017), Kan, Peng and Wang (2017), Ameriks et al. (2018), and
Fuster, Kaplan and Zafar (2018).

2
Figure 1: Percent Increase in Spending by Reported Main Use

to elicit the dollar spending caused by the rebates. We worked with survey administrators
to design and field survey instruments in both the Consumer Expenditure (CE) Survey
and the Nielsen Consumer Panel (NCP) during 2008, and we perform our comparison
separately within each dataset.
While the Friedman and Wallis (1942) quote above refers to purely hypothetical choices
(or ‘forced choice’ situations), our analysis instead involves the elicitation of causal eﬀects
in past choices, a situation in which reported eﬀects are likely to be more accurate (Manski,
1990). Respondents need only infer how their spending would have been diﬀerent in the
hypothetical counterfactual situation in which a payment was not received, a situation
they typically encounter.
For our specific case involving survey elicitation methods applied to past spending, we
establish three main results.
First, reported spending is highly informative about revealed-preference propensities
to spend. As shown in Figure 1, we find large revealed-preference estimates of spending by
households that report that they mostly spent their stimulus payments. These revealed-
preference spending responses are economically much larger than the spending responses
of households that report mostly saving their payments or mostly using them to pay
down debt. This is true in both the CE survey sample and the NCP sample. While we
find statistically and economically significant revealed-preference estimates of spending by
households that report that they mostly used their payments to save or pay down debt,
these estimates of spending are not inconsistent with the quantification of the reported-

3
Figure 2: Reported and Revealed Estimated Spending Propensities in NCP Dataset

Panel A: Spending Propensities by Income Panel B: Spending Propensities by Liquidity

80
80

S pe nding P rope nsitie s S pe nding P rope nsitie s

40 60 40 60

20 20

Low Mi ddle High Low Middle H igh Insufficient Suf ficie nt Insuffi cient Sufficient
Re ported Prefer ence Revea led P referenc e Reporte d Prefere nce Reve aled Prefer ence

Note: Based on Tables 9, 10, and 11, see notes to these Tables. Whis kers display one standard error bands.

preference answers in previous research. In sum, reported spending captures economically
large diﬀerences in spending.
Second, in each dataset, the average propensities to spend estimated from the two
methods are similar. The one diﬀerence of note seems likely related to the use of debt
for the purchase of durable goods, and the resulting diﬀerence between expenditures
(measured by our revealed-preference propensities) and out of pocket expenses (likely
measured by survey reports of spending eﬀects). These first two findings lend credence to
the use of reported causal eﬀects to measure the eﬀect of economic policy or to estimate
model parameters.
Third, we find that, in both datasets, reported propensities to spend show almost no
variation across households with diﬀerent levels of income or across those with diﬀerent
levels of liquidity. This is inconsistent with our reading of the large literature on con-
sumption smoothing (e.g. Fuchs-Schündeln and Hassan, 2015), which is that households
with little liquidity have higher propensities to spend. And this is also inconsistent with
our revealed-preference estimates for NCP goods as shows in Figure 2. And we also find
greater spending by less liquid households on broadly-defined nondurable goods in the
CE but the diﬀerence by liquidity is not statistically significant. So one might conclude

4
that the survey methodology for inferring causation fails to capture an important dif-
ference in the population. However, this would be premature. We find no diﬀerence in
revealed-preference propensities to spend on total goods in the CE, consistent with what
we find with our survey reports. Thus, it is possible that less liquid households do not
spend at greater rates. We discuss possible interpretations of these results in Section VIII,
including that these findings could be driven by shortcomings in either measure or be an
artifact of statistical imprecision.
Two caveats apply to all of our findings. First, current survey questions about spend-
ing are not clear about the time horizon over which additional spending is to be reported.
Our evaluation of these measures uses the horizons typically assumed by other researchers
and by policymakers. Second, our revealed-preference propensities only capture spending
caused by the arrival of the payment (e.g. omitting any spending that occurred as the
policy was developed and when the payments were announced) while reported propen-
sities (presumably) include spending unrelated to the timing of the payment’s arrival.
Nevertheless, as we discuss in Section V, both theory and empirical evidence suggests
that these diﬀerences are minor, at least relative to the statistical uncertainty that exists
around the quantitative estimate of the propensity to spend from each method.
Our paper’s closest predecessors are Hainmueller, Hangartner and Yamamoto (2014)
and Karlan, Osman and Zinman (2016). Karlan, Osman and Zinman (2016) finds that
reported spending is biased relative to revealed-preference estimates and interprets this
finding as due to strategic responses that are specific to the incentives in the micro-credit
setting of the study. Our evaluation of reported behavior also relates to the analysis of
reported probabilities (as in Manski, 2004, 2017), to the use of reported willingness to pay
in environmental economics (see Diamond and Hausman, 1994; List and Shogren, 1998;
Harrison and Rutström, 2008), and to conjoint analysis in marketing (see Juster, 1966,
and Rao, 2014). These last two literatures conclude that survey-based intentions data are
informative for prediction but do not provide precise quantitative responses.

I The economic stimulus payments of 2008
The economic stimulus payments of 2008 were distributed to about 130 million eligible
taxpayers (about 85% of tax units) in the spring and summer of 2008. Conditional on a
taxpayer meeting a minimum earnings test for 2007, the stimulus payment consisted of
$600 ($1,200) for individual (joint) filers plus $300 per child that qualified for the child
tax credit. The payment was phased out for high income households. Most importantly,
the distribution of the payments involved some quasi-random variation over time. For
recipients that had provided the IRS with their pfersonal bank routing number (i.e., for
direct deposit of tax refunds), the stimulus payments were disbursed electronically over
three one-week periods ranging from late April to mid-May. For recipients that did not
provide a personal bank routing number, the payments were mailed (using paper checks)
in one of nine one-week periods ranging from mid-May to mid-July. Within each of these
two groups, the timing of the payment was determined by the last two digits of the
recipient’s Social Security number, digits which are eﬀectively randomly assigned. The
on-line Appendix contains further details.

5
II The two datasets
A strength of our study is that we conduct our analysis separately on two diﬀerent datasets
— the CE survey and the NCP — that have diﬀerent sampling time-frames, survey methods,
and recall periods.
We use the 2007 and 2008 waves of the CE interview survey data which contains
detailed measures of the expenditures of a stratified random sample of U.S. households.
We worked with the Bureau of Labor Statistics to add questions about the 2008 stimulus
payments to the survey for interviews conducted between June 2008 and March 2009. A
first module of questions asked households whether they received any “economic stimulus
payments. . . also called a tax rebate” since the beginning of the reference period for the
interview and, if so, the amount of each payment, the date it was received, and whether it
was received by check or direct deposit. A second module, described in the next section,
asked households to report spending from the payments.
In our analysis, we focus on two measures of spending: ) nondurable spending (and
some services) which includes CE categories like food, utilities, household operations,
gas, personal care, and tobacco as well as semi-durable categories like apparel, health, and
reading materials (average spending $5,400 over three months), and ) total CE spending
which adds durable expenditures such as home furnishings, entertainment equipment, and
auto purchases (average $10,410 over three months).
Our second dataset is the NCP, available through the Kilts Center for Marketing at
the University of Chicago Booth School of Business. The NCP is a panel survey of U.S.
households in 52 metropolitan areas that tracks spending mainly on household goods
with Universal Product Codes (UPCs, or “barcodes”). At the conclusion of every shop-
ping trip for household items, participants use barcode scanners to manually input the
total amount spent and scan the items they purchase.3 These procedures imply that
spending is reported accurately for the goods (and trips) that are reported. On the other
hand, the NCP only measures a small share of consumption spending (average spending
$601 per month). NCP spending primarily covers purchases at grocery, drugstore, and
mass-merchandise stores, and primarily goods such as food and drug products, small ap-
pliances and electronic goods, and mass-merchandise products largely excluding apparel.
We aggregate total spending reported for each trip to the household-week level for the
year 2008.
We merge these NCP data with data from a two-part, multi-wave survey that we
designed and was administered by Nielsen while the payments were being distributed.
Part I asked questions pertaining to the household’s liquid assets and typical behavior.
Part II described the program of economic stimulus payments, asked the household to
report the details of any stimulus payment received, and then asked the household to
report how much of their payments they spent.
Table 1 presents summary statistics for the two datasets, and the on-line Appendix
contains our supplemental surveys.

3
Participants get newsletters and personalized reminders via email and/or mail to upload spending
information and to answer occasional surveys. Households that participate regularly are entered in prize
drawings or receive Nielsen points that can be used to purchase items from a catalogue.

6
Table 1: Sample Summary Statistics

CE Data, Three-month periods NCP Data, Weekly
Standard Standard
Mean Median Deviation Mean Median Deviation
Panel A: Observations
Number of Observations 10,353 995,748

Spending type: CE nondurables & services NCP household goods
Spending (dollars) 5,436 4,867 3,017 150.25 99.51 186.24
Spending | Spending  0 5,436 4,867 3,017 180.13 126.74 190.26
Obs. w/ Spending  0 10,353 845,487
Spending type: CE total
Spending (dollars) 10,410 8,646 7,195

ESP (dollars) 259 0 498 70 0 282
ESP | ESP  0 951 900 504 910 600 521

Panel B: Households
Number of Households 4,296 19,149

Household Size 2.65 2 1.48
Number of Adults 1.96 2 0.82
Number of Children 0.67 0 1.06

Income (dollars) 58,707 48,800 41,611
Indicators: Income ≤ $20K 0.16 0 0.36 0.15 0 0.35
$20K  Income ≤ $50K 0.36 0 0.48 0.37 0 0.48
$50K  Income ≤ $100K 0.33 0 0.47 0.36 0 0.48
Income  $100K 0.16 0 0.36 0.13 0 0.33
Households w/ income data 3,427 15,449

Liquidity (dollars) 9,172 2,100 19,347
Indicator: Liquidity  $2,000 0.50 1 0.50
"Yes" to "Enough Liquidity"? 0.57 1 0.49
Households w/ liquidity data 1,819 19,149
Note: For the CE, statistics are based on the first-diﬀerenced dataset and so drop the
first observation per household in levels. We calculate means, medians, and standard de-
viations using CE household weight and the NCP projection factor for 2008. In the NCP,
income and liquidity are categorical variables. The researchers own analyses calculated
(or derived) based in part on data from The Nielsen Company (US), LLC and marketing
databases provided through the Nielsen Datasets at the Kilts Center for Marketing Data
at The University of Chicago Booth School of Business.

7
III Survey methodology for reported spending
To measure reported spending, we use two diﬀerent survey instruments.
The first is that used by Shapiro and Slemrod (2003b) in the Michigan Survey of
Consumers, which is the basis for most existing research. In the CE this question is:

Earlier, you or someone in your CU [consumer unit] reported receiving a one-
time tax rebate that was part of the Federal government’s economic stimulus
package. Did the tax rebate lead you or someone in your CU [consumer unit]
mostly to increase spending, mostly to increase savings, or mostly to pay oﬀ
debt?

As with many CE questions, the respondent can respond that they do not know.
In the NCP, this question is:

Thinking about your household’s financial situation this year, is the tax rebate
leading you mostly to increase spending, mostly to increase savings, or mostly
to pay oﬀ debt?

The respondent can then choose one of the same three responses or ‘Not sure/don’t
know.’ Consistent with previous use, our question is preceded by a lead-in to remind the
respondent about earlier questions about the stimulus payment (the CE) or by questions
about the stimulus payment (the NCP).
Our second survey instrument, asked only in the NCP, asks respondents to provide a
continuous estimate of the dollar amount that the payment caused them to increase their
spending:4

For the following questions, please think about the extra amount you are
spending because of this rebate. How much (in dollars rounded to the nearest
dollar) are you spending on each of the following:

The first category is ‘Food, health & beauty aids, and household products’ which is
designed to capture spending on household items reported as spending in the NCP. There
are four additional categories: ‘entertainment and services,’ ‘durable goods,’ ‘clothing’,
and ‘all other types of purchases.’
A critical question is: over what time horizon is spending reported? The present-tense
wording of the question, the lag between payment receipt and survey, and the typical use
of language all suggest that people report additional spending caused by the payment
over a few weeks or months following the payment, and possibly also preceding it. This
is exactly how the answers have been used in previous research. And we evaluate this

4
There is no standard phrasing for quantitative questions. For example, the 2010 Italian Survey
of Household Income and Wealth (SHIW) asks about hypothetical receipt of future income with the
phrasing: “Imagine you unexpectedly receive a reimbursement equal to the amount your household earns
in a month. How much of it would you save and how much would you spend? Please give the percentage
you would save and the percentage you would spend” (as translated in Jappelli and Pistaferri, 2014).

8
Table 2: Reported-Preference Spending Propensities

CE Data NCP Data
Standard Standard
Reported Use of Payment Mean Median Deviation Mean Median Deviation
Mostly to Spend 0.32 0 0.47 0.19 0 0.39
Mostly to Save 0.17 0 0.38 0.27 0 0.44
Mostly to Pay Debt 0.51 1 0.50 0.54 1 0.50
Households w/ Response 4,076 19,149

Reported Spending ($)
All Spending (A) 452 300 627
Household Items (B) 62 0 182
Non-Household Items 390 200 564
Households w/ Responses 19,059

Stimulus Payment Amount ($) (C) 951 900 504 910 600 521
Spending Propensity (A/C) 50
Spending Ratio: All Items
to Household Items (A/B) 7.32
Note: For the CE, statistics are calculated using household weights. For the NCP,
statistics are calculated using the NCP projection factor for 2008. The last two rows
are calculated as ratios of means. See notes to previous table.

interpretation of the answers by contrasting them to revealed-preference spending over
one to three months following payment receipt.
Closely related, our questions are only asked of households that have reported receiving
a payment. Previous research suggests that ex post questions capture more spending and
that they are more accurate than ex ante questions.5
Table 2 presents the responses to the reported-spending questions in each of our
datasets. We find that roughly a fifth of households report that they mostly spent their
payments in the NCP and roughly a third report that they mostly spent in the CE. In
terms of the quantitative spending amounts that people report, NCP households report
spending $452 on average, of which $62 was on goods covered in the NCP and $390 was
on other goods and services. Given an average stimulus payment amount of $910, the
implied average propensity to spend is 50 percent. Both our discrete and continuous
quantitative responses lie within the range found in the literature (e.g. Bureau of Labor
Statistics, U.S. Department of Labor, 2009, and Jappelli and Pistaferri, 2014).
Why do we find a significantly higher share of households reporting that they will

5
First, there is a tendency for less spending to be reported in response to questions before tax re-
ductions or payments as compared to ex post questions (Sahm, Shapiro and Slemrod, 2012). Second, ex
ante expected behavior correlates significantly but imperfectly with ex post reported behavior (Manski,
1990). Shapiro and Slemrod (2003b) find spending responses of the same households before and after the
stimulus payments have a correlation of 044.

9
mostly spend the payment in the CE than in the NCP? One possibility is that this is due
to the tendency for reported spending to rise with the amount of time between receipt
and survey.6 In the CE, we survey households up to four months after receipt while in the
NCP we survey them at most 7 weeks after receipt. But as we argue in on-line Appendix
A, existing evidence suggests that this eﬀect is likely too small to explain the diﬀerence
we observe.
The more likely possibility is that the reported spending propensities diﬀer between
the surveys because the samples in the surveys diﬀer. This diﬀerence in samples does not
impede our main analysis of whether, in each dataset, the two diﬀerent methodologies
estimate the same propensity to spend.

IV Revealed-preference methodology
We use the diﬀerent timing of the disbursement of payments across households to estimate
the causal eﬀect of the receipt of a payment on spending, following Parker et al. (2013)
for the CE and Broda and Parker (2014) for the NCP.
Letting  be the response to the discrete spending question, we estimate an equation
of the form:

⎧ ⎫
Δ ⎨ () 1[ =  ]+ ⎬
(1) or =   () 1[ =  ]+ + θ0 X +   + 
 ⎩  ⎭
  () 1[ =    ]

where  is the log or the dollar amount of spending,  is an indicator of payment receipt
or the dollar amount received by household  at time , the X are control variables that
include  and otherwise diﬀer by survey,   is a period-specific intercept. Because it is
possible that households with large payments diﬀer from those with small payments, we
instrument the distributed lags of the dollar payment with distributed lags of the indicator
of payment receipt. The lag polynomials, ()   () and   (), measure the changes
in spending caused by receipt over time (relative to before receipt). When studying those
reporting most spending vs. those not, we impose   () =   ().
We use a (slightly) diﬀerent method in each dataset because of diﬀerences in the fre-
quency of the data, diﬀerences in the available household-level information, and diﬀerences
in the previous literature. In the CE, spending data is observed over three-month periods
and we do not include individual eﬀects but control for observable household character-
istics: age, change in the number of children, and change in the number of adults. In
the NCP, the frequency is weekly and, because the NCP lacks any covariates that change
(during a year) and we estimate in levels, we include individual eﬀects.
As we discuss in the next section, these measures omit any spending responses that are
unrelated to the timing of receipt, such as response to aggregate news about the stimulus
payment program or responses that occur at a common calendar time such as during an

6
Sahm, Shapiro and Slemrod (2010) Table 3 shows that 36% of those who say that they will mostly
spend say their spending rises “within a few weeks,” 50% report “within 1-3 months”, and 14% “more
than 3 months.”

10
August vacation.
We first estimate variants of equation (1) without the interaction with  . We find in
the CE that spending on nondurable goods in the three-month period of receipt rises by
$298, 47% of spending, or 31% of the payment. In the NCP, we estimate that spending on
NCP goods in the month following receipt rises by $40, 69% of spending, or 43% of the
payment, with some continued spending in the next month or two.7 Complete results are
detailed in on-line Appendix 6. The estimates imply a slightly lower rate of spending
in the NCP than in the CE even after adjusting for less spending measured in the NCP.
Although this diﬀerence between surveys is not statistically strong, it is consistent with
the diﬀerence in reported propensities across surveys presented in Table 2.

V Diﬀerences in spending concepts in reported and
revealed spending methodologies
First, as noted, the revealed-preference measure only captures changes in spending that are
related to the timing of the arrival of the payment, while the reported spending measure
may also capture anticipatory spending unrelated to this timing. Based on theory and
evidence, this diﬀerence is likely to be a very small part of the total spending response.
In theory, households that respond to news about the stimulus program in advance of
receipt are those that optimally smooth consumption over time. Such households should
have small spending responses to the news because it represents a very small change in
lifetime wealth. In practice, revealed-preference studies find small anticipatory spending
eﬀects (e.g. see Table 6, Broda and Parker, 2014).
Second, purchases that use credit may cause a significant diﬀerence between our mea-
sures. Revealed-preference spending propensities are based on measured expenditures.
The CE measures total expenditures, and spending is the price of a new car not just
the down payment if it is financed. Similarly, the NCP measures spending using receipt
totals, regardless of whether the household paid with a debit card or credit card.
In contrast, the concept of spending measured by the survey-based self-reports is less
clear. It seems likely that for larger purchases, such as a financed purchase of a car, the
out-of-pocket expense (down payment) is likely to determine the response. To some extent
this is not a concern for the discrete response questions, since a large financed purchase
with a substantial down payment will lead to an answer of “mostly spend” even if the
household responds only based on a down payment. But it is not at all clear whether
the reported additional dollar spending is based on total purchase price or out-of-pocket
payment.
The final issue is whether the two methods both measure the spending caused by the
payment over a similar time horizon. While there is no clear answer, the two measures
are used by the previous literature and by policymakers to quantify the same concept of
interest: additional spending around the time of payment disbursement and not spending
done years later in response to the payment. Additionally, for both measures, the eﬀect

7
These estimates are generally consistent with those of the earlier literature, but are diﬀerent due to
the fact that we omit households that do not respond to reported spending questions.

11
of horizon on spending propensity seems to be quantitatively relatively minor for the
variation we observe and are interested in (see the discussion on time horizon in Section
III and on-line Appendix A).

VI Revealed spending and reported use of payment
Our first main finding is that self-reports of ‘mostly spending’ are highly informative
about the revealed-preference propensity to spend.
The first column of Table 3 shows that households reporting that they mostly spent
their payments did indeed spend at large rates. For the CE, during the three month
period of receipt, spending on non-durable goods rose by $366 (Panel A) and on total
goods rose by $1,167 (Panel B) which is more than the average payment amount of $910.
For the NCP, Panels C and D show substantial spending by households reporting that
they mostly spent — 13% higher spending in spending over the month following receipt
and 5.7% higher spending on average over the three months following receipt.
While not always statistically significant, for all specifications, these spending re-
sponses are quantitatively much larger than those of households reporting that they
mostly saved their payments or used them to pay down debt (comparing the first and
second columns of Table 3). Except for Panel A, spending responses for households re-
porting that they mostly spent their stimulus payments are nearly double the spending
responses of other households. The third column shows that these diﬀerences are mostly
statistically significant.
Note that, while not always statistically significant, there is economically significant
spending by households reporting either that they mostly saved or mostly paid down debt
(Table 3, column 2). The final two columns of Table 3 break down this diﬀerence between
households that report mostly saving and those that report paying down debt.
In sum, the people’s discrete reports of their propensity to spend are highly predictive
of their behavior. People who report mostly spending are econometrically estimated
to have marginal propensities to consume roughly double those of households reporting
mostly saving or paying down debt.

VII Average spending propensities
We quantify spending associated with each discrete reported spending response (mostly
spend, save, or pay debt) in three diﬀerent ways.
First, we follow the literature on reported spending and Shapiro and Slemrod (2003b)
in particular. This approach assumes that the distribution of spending propensities is
triangular and calibrates the parameters from the data and by assumption. Given the
share of households that report ‘mostly spend,’ this method implies a propensity to spend
of 23 for these households. While this method is somewhat arbitrary, it is important
to evaluate this method because it is the one used for inferring the (partial-equilibrium)
aggregate eﬀect of the tax policies on the economy in existing work.
Our second calibration instead assumes a higher upper bound on the distribution
of spending propensities to account for the possibility that some households purchase

12
Table 3: Revealed Spending by Qualitative Reported Spending

Report: Mostly Mostly Save or Equality Test Mostly Mostly Pay
Spend Pay Down Debt p-value Save Down Debt
Panel A:CE Survey, Nondurable Spending During Three-Month Period of Receipt
Spending in dollars 366 267 0.29 221 255
(120) (111) (133) (103)
Log-percent increase 7.02 3.71 0.02 2.79 3.87
(1.89) (1.72) (2.08) (1.61)
Percent of payment spent 39.0 27.5 0.24 22.5 25.9
(13.5) (12.4) (13.9) (10.9)

Panel B: CE Survey, Total Spending During Three-Month Period of Receipt
Spending in dollars 1167 539 0.03 645 357
(400) (350) (393) (327)
Log-percent increase 8.19 2.52 0.00 2.72 2.51
(2.52) (2.17) (2.63) (2.01)
Percent of payment spent 122.4 52.9 0.03 63.9 32.9
(45.1) (39.4) (41.5) (34.7)

Panel C: NCP, Spending on Household Goods in Month Following Receipt
Spending in dollars 76.46 31.33 0.00 27.95 32.84
(10.80) (5.85) (7.83) (6.76)
Percent increase 13.30 5.39 0.00 3.79 6.15
(2.34) (1.29) (1.80) (1.43)
Percent of payment spent 8.16 3.39 0.00 2.99 3.58
(1.19) (0.64) (0.85) (0.75)

Panel D: NCP, Spending on Household Goods Over Three Months Following Receipt
Spending in dollars 89.34 44.65 0.03 72.85 30.89
(25.79) (20.03) (24.18) (20.81)
Percent increase 5.70 3.08 0.08 9.86 8.84
(1.99) (1.51) (5.62) (4.61)
Percent of payment spent 8.67 4.23 0.06 6.93 2.90
(2.74) (2.08) (2.48) (2.20)
Note: In Panels A and B, each row comes from a regression of a distributed lag of an
indicator of payment receipt, or the amount instrumented with the indicator, as well as
month eﬀects, age, change in the number of children, and change in the number of adults
on dollar change, or log change, in spending. In Panels C and D, each row comes from a
regression of a complete distributed lag of an indicator of payment receipt or the indicator
used as an instrument for the amount, as well as month and household eﬀects on dollar
spending or that normalized by average monthly spending during 2008Q1. The final two
columns are from a diﬀerent regression than the first three. Parentheses contain standard
errors that are robust to arbitrary heteroskedasticity and within-household correlation.
See notes to Table 1.

13
durable goods and so spend more than the payment amount.8 In this calibration, the
average propensity to spend among households reporting ‘mostly spend’ is roughly 80%
and the average propensity of the rest of the population is 25% (by assumption).
Our third quantification is based on the dollar amounts that people report spending
in the NCP. For households reporting ‘mostly spend,’ we divide average total spending
by the average payment amount. We do the same for households reporting ‘mostly save’
and ‘mostly pay down debt.’ This gives a self-reported average propensity to spend for
each reported spending response that we can use in each survey.
In order to calculate revealed-preference spending propensities for all expenditures
from CE nondurable goods and from NCP goods, we scale up the propensities by the
ratios of goods covered. For example, in the CE, we scale up the propensities to spend on
non-durable goods by the ratio of average total spending to average non-durable spending
(1.94).
Our main finding is that the two methods — reported and revealed preference — deliver
broadly similar estimates of the sample-average propensity to spend (Table 4, final col-
umn). The average propensity to spend based on reported spending range from 40 to 58
percent in the CE Survey and from 27 to 50 percent in the NCP. The revealed-preference
propensities range from 57 to 67 percent in the CE, and from 33 to 48 percent in the
NCP.
Second, while the two methods deliver broadly similar estimates, the revealed-preference
estimates are slightly larger than those derived from household reports of spending in the
CE, but not in the NCP. This diﬀerence is consistent with diﬀerences between methods
in the measurement of debt-financed purchases. If people report only out of pocket pay-
ment as the amount of the payment that they spent, since the CE instead records total
expenditures, then we would expect the CE revealed-preference measure to exceed the
CE reported measure. This is what Panel A of Table 4 shows: total revealed spending
and reported spending are quite diﬀerent, while non-durable scaled revealed spending
and reported spending are more similar (and mainly for the columns besides ‘mostly pay
down debt’). Since the NCP does not cover large durable goods, we would not expect as
much of a diﬀerence in the NCP, consistent with what we find.

VIII Spending propensities by liquidity and income
In this section, we test the ability of the methodologies to produce mutually consistent
estimates of the propensity to spend for subsamples defined by income and liquidity, char-
acteristics that are associated with diﬀerent spending responses both in leading theories
and, at least in our reading, much previous empirical research.
We measure income as family pre-tax income over the previous 12 months (in the CE)
and during the previous calendar year in the NCP. In each dataset, we divide the sample
roughly into thirds. We measure liquidity as the sum of balances in checking and saving
accounts prior to the first interview in the CE. In the NCP, we measure liquidity from

8
Further, to maintain a reasonable distribution, we also assume that the spending propensity is uni-
formly distributed for households that report not mostly spending. See On-line Appendix 7 for details.

14
Table 4: Reported and Revealed Spending Propensities (percent of payment)

Qualitative Reported Spending: Mostly Mostly Mostly Weighted
Spend Save Pay Debt Average

Panel A: Consumer Expenditure Survey
Fraction of Sample (Table 1) 0.32 0.17 0.51

Reported Spending Propensities
Shapiro-Slemrod Calibration 67 27 27 40
Alternative Calibration 80 25 25 42
Reported Spending in NCP 98 25 44 58

Revealed Spending Propensities, Three Months of Arrival
Total Scaled from Nondurable (1.94) 76 44 50 57
Total Spending 122 64 33 67

Panel B: Nielsen Consumer Panel
Fraction of Sample (Table 1) 0.19 0.27 0.54

Reported Spending Propensities
Shapiro-Slemrod Calibration 67 18 18 27
Alternative Calibration 80 25 25 36
Reported Spending in NCP 98 25 44 50

Revealed Spending Propensities, Month After Arrival
Scaled up by CE Revealed by Category (9.4) 77 28 34 40
Scaled up by NCP Reported (9.9, 6.5, 6.2) 81 20 22 33
Revealed Spending Propensities, Three Months After Arrival
Scaled up by CE Revealed by Category (9.4) 81 65 27 48
Scaled up from NCP Reported (9.9, 6.5, 6.2) 86 45 18 38
Note: Scale factor for CE nondurable goods and some services is the ratio of CE total
spending to nondurable spending (See on-line Appendix Section A.7 and Table A.2). The
first scale factor for NCP data is the ratio of the revealed propensity to spend on all goods
in the CE relative to the spending on NCP-type goods in the CE (method 3 in Table 5 of
Broda Parker, 2014, which is the intermediate scale factor of the three considered). The
second scale factor is the average reported payment spent on all goods and services in
the NCP divided by the average reported payment spent on NCP goods for each discrete
reported spending response. Further details and the calibrations are described in the
on-line appendix. See notes to previous tables.

15
our survey question “In case of an unexpected decline in income or increase in expenses,
do you have at least two months of income available in cash, bank accounts, or easily
accessible funds?”
First, the first two columns of Table 5 show that the discrete reported spending re-
sponses are unrelated to income (Panel A) and liquidity (Panel B). The share of house-
holds reporting ‘mostly spend’ does not diﬀer by income or liquidity in the CE or NCP.
The only diﬀerence of note is in the CE, where we find evidence inconsistent with most
revealed-preference studies: 37 percent of households with high liquidity report that they
mostly spent their payments compared to only 29 percent of households with low liquidity.
Second, column 3 of Table 5 shows some evidence of somewhat higher quantitative
reported spending by the lowest income group (below $35,000): an average propensity of
59 percent relative to 48 or 45 percent for middle and high income groups.9 But panel
B shows that reported spending propensities are very similar between levels of liquidity.
Thus, as with the discrete responses, we find little relationship between reported spending
propensities and liquidity.
Canonical theory and our reading of the literature on consumption smoothing would
lead us to expect larger spending propensities by households with low income or low liq-
uidity. Although statistically weak, we find this pattern in revealed-preference spending
propensities on nondurable goods in the CE (the fourth column of Table 5). Similarly,
in the NCP (last column of Table 5), the spending propensities estimated by revealed-
preference imply significantly larger spending responses by lower income or liquidity
groups (by factors of nearly 3) than do the propensities reported by households, or than
do the reports of ‘mostly spend’ (which do not diﬀer across income groups).
However, there is no evidence that spending propensities on total spending as mea-
sured in the CE are higher for households with low income or liquidity (column 5). Total
spending propensities reported by households and propensities estimated from behav-
ior are both uncorrelated with income and liquidity, although these revealed-preference
estimates have very low statistical power.
We see three reasonable explanations for this pattern. The revealed-preference spend-
ing propensities for NCP goods and for nondurable goods in the CE could be misleading
because they measure only spending on a subset of expenditure categories, missing services
and vehicle purchases for example. If tax rebates cause households with significant liquid-
ity to spend disproportionately more on goods and services outside of these measures, then
the observed diﬀerences in revealed-preference non-durable spending propensities across
liquidity or income would be overstated. If this bias were strong enough, propensities to
spend from the payment could be uncorrelated with income or liquidity and the reported
propensities could be correct. Alternatively, it may be that reported causal estimates
are biased. One possibility is that high-income or high-liquidity households report pur-
chases that they would have made anyway as spending caused by the payment. A final
possibility is that people may report spending or saving relative to their usual behavior.
People who are persistently constrained and so spend money as it arrives may perceive
and report ‘mostly saving’ when they spend more slowly than they typically do; similarly

9
This diﬀerence is driven by lower spending by households who report that they mostly saved or paid
down debt.

16
Table 5: Propensities to Spend by Income and Liquidity
Reported Spending Revealed-Preference Propensities
Share Reporting Reported Three Months One Month
Mostly Spend Propensity Nondurable Total NCP Goods
CE NCP NCP CE CE NCP
Panel A: By Household Income
Income  $35K 0.32 0.20 59 52.6 70.0 6.9
(23.5) (45.4) (1.3)
$35K ≤ Inc.  $70K 0.31 0.19 48 -11.2 -25.4 4.5
(21.5) (88.5) (1.1)
$70K ≤ Income 0.33 0.20 45 17.0 84.1 2.4
(20.1) (82.0) (1.1)

Low less middle 0.01 0.01 11 63.8 95.4 2.4
(31.9) (99.5) (1.7)
Low less high -0.01 0.00 13 35.6 -14.1 4.5
(30.9) (93.7) (1.7)

Panel B: By Household Liquidity
Liquid Assets  $2K or 0.29 0.17 48 48.3 44.6 7.5
Insuﬃcient Liquidity (26.5) (66.5) (0.9)
Liquid Assets ≥ $2K or 0.37 0.21 51 19.1 48.3 2.1
Suﬃcient Liquidity (21.7) (86.0) (0.7)

Low liquidity less high -0.08 -0.04 -3 29.2 -3.7 5.4
(34.3) (108.7) (1.1)

Note: The first three columns report the share of households in each income or liquidity
group that reports ‘mostly spend.’ Each income group contains roughly a third of obser-
vations and each liquidity group roughly half. The last three columns are from estimation
of equation (1) imposing   () =   () separately on each liquidity or income group.
See notes to other tables.

17
people who typically save may perceive and report ‘mostly spend’ when they spend some
of their payment more rapidly than usual.
Does other research also find that reported propensities are unrelated to income or
liquidity? The lack of diﬀerence in reported spending responses across households is con-
sistent with the pre-existing evidence on discrete reported propensities. Sahm, Shapiro
and Slemrod (2010) analyzes reported spending responses to these same 2008 stimulus
payments and concludes that less-well-oﬀ households were not more likely than rich house-
holds to spend the tax payment. Similarly, the Shapiro and Slemrod (2003a) study of the
2001 tax rebates finds that the rate of reporting ‘mostly spend’ actually increases with
household income and with households stock wealth.
In contrast, Jappelli and Pistaferri (2014) finds that the continuous quantitative re-
ported spending propensities rise with cash-on-hand in the 2010 Italian SHIW. There
are (at least) three possible reasons for this diﬀerent finding. First, households in the
NCP and SHIW may have diﬀerent survey response rates and sample selection, so that
they cover diﬀerent types of households (in addition to drawing from the populations of
diﬀerent nations). Second, the questions in the SHIW are purely hypothetical, referring
to an abstract increase in income, and to an amount significantly larger than the typ-
ical stimulus payment we study (relative to income). Finally, the questions are asked
diﬀerently. The SHIW asks the questions as propensities, bounded between 0 and 100
percent. Our questions in NCP asked about spending in dollars in five diﬀerent categories
of expenditures, with spending bounded by 0 and $2,999 in each category.

IX Discussion
Our analysis finds that reported spending propensities are highly informative. The people
who report ‘mostly spending’ their payments spent at roughly twice the rate as people
reporting that they used their payments to save or pay down debt. Further, on average,
people report spending roughly the same fraction of their 2008 tax payments as their
behavior reveals. Thus, estimates of the spending caused by the 2008 economic stimulus
payments based on survey reports of spending propensities appear quite similar to those
based on traditional revealed-preference measures.
Our analysis of whether reported propensities accurately capture diﬀerences in spend-
ing across households is less conclusive. In both surveys, we find that people do not report
diﬀerent spending propensities at diﬀerent levels of liquidity. This finding is consistent
with the (noisy) estimated spending propensities on total spending in the CE, which
like the reported propensities are uncorrelated with liquidity. But this pattern is incon-
sistent with estimated spending propensities on NCP goods and on CE broadly-defined
non-durable goods, both of which are lower at higher levels of liquidity and income. Pre-
vious research on reported propensities is also mixed in that the share of people reporting
‘mostly spending’ payments has generally been found to be uncorrelated with liquidity,
while reported continuous propensities to spend from hypothetical income increases have
generally been found to be much higher among households with low levels of liquidity.
Our surveys and policy experiment represent an almost ideal situation for the approach
of asking people to infer and report the causal eﬀect of a given policy. We employed ex

18
post questions about a policy that the households had recently experienced, so that recall
was relatively straightforward. People also had experience with the counterfactual of no
payment (most of the rest of their lives). Thus, respondents were familiar with their
behavior in both the baseline event and the hypothetical alternative.
We conclude by suggesting several avenues for improvement in the design and collection
of self-reported data on spending propensities. First, reported spending questions could
be improved by being explicit about the time period that they are asking about. Second,
reported spending questions could clarify whether spending refers to purchase price or
out-of-pocket payment, and so be more explicit about the purchase of durable items and
the treatment of debt. Finally, the accuracy of reported spending might be improved by
incentivizing unbiasedness or accuracy. But given idiosyncratic diﬀerences in propensities
over time, providing such incentives is far from simple.

19
References
Ameriks, John, Gábor Kézdi, Minjoon Lee, and Matthew D. Shapiro. 2018.
“Heterogeneity in Expectations, Risk Tolerance, and Household Stock Shares: The
Attenuation Puzzle.” NBER Working Paper No. 25269.
Ameriks, John, Joseph Briggs, Andrew Caplin, and Matthew D. Shapiro.
2016. “Late-in-Life Risks and the Under-Insurance Puzzle.” NBER Working Paper No.
22726.
Auclert, Adrien. 2017. “Monetary Policy and the Redistribution Channel.” NBER
Working Paper No. 23451.
Barsky, Robert, Thomas Juster, Miles Kimball, and Matthew Shapiro. 1997.
“Preference Parameters and Behavioral Heterogeneity: an Experimental Approach in
the Health and Retirement Survey.” Quarterly Journal of Economics, 112(2): 537—80.
Broda, Christian, and Jonathan A. Parker. 2014. “The Economic Stimulus Pay-
ments of 2008 and the Aggregate Demand for Consumption.” Journal of Monetary
Economics, 68(S20-36).
Bunn, Philip, Jeanne Le Roux, Kate Reinold, and Paolo Surico. 2017. “The
consumption response to positive and negative income changes.” Bank of England Bank
of England working papers 645.
Bureau of Labor Statistics, U.S. Department of Labor. 2009. “Consumer Ex-
penditure Survey Results on the 2008 Economic Stimulus Payments (Tax Rebates).”
http://www.bls.gov/cex/taxrebate.htm.
Coronado, Julia Lynn, Joseph P. Lupton, and Louise Sheiner. 2005. “The house-
hold spending response to the 2003 tax cut: evidence from survey data.” Board of
Governors of the Federal Reserve System (U.S.) Finance and Economics Discussion
Series 2005-32.
Crump, Richard K., Stefano Eusepi, Andrea Tambalotti, and Giorgio Topa.
2015. “Subjective intertemporal substitution.” Federal Reserve Bank of New York Staﬀ
Reports 734.
D’Acunto, Francesco, Daniel Hoang, and Michael Weber. 2016. “The Eﬀect of
Unconventional Fiscal Policy on Consumption Expenditure.” NBER Working Paper
No. 22563.
Diamond, Peter A., and Jerry A. Hausman. 1994. “Contingent Valuation: Is Some
Number Better than No Number?” Journal of Economic Perspectives, 8(4): 45—64.
Friedman, Milton, and W. Allen Wallis. 1942. “The Empirical Derivation of Indif-
ference Functions.” 177—183. Chicago University Press.
Fuchs-Schündeln, Nicola, and Tarek Hassan. 2015. “Natural Experiments in Macro-
economics.” In The Handbook of Macroeconomics. Elsevier.

20
Fuster, Andreas, Greg Kaplan, and Basit Zafar. 2018. “What woould you do with
$500? Spending Responses to Gains, Losses, News and Loans.” NBER Working Paper
24386.

Graziani, Grant, Wilbert van der Klaauw, and Basit Zafar. 2016. “Workers’
Spending Response to the 2011 Payroll Tax Cuts.” American Economic Journal: Eco-
nomic Policy, 8(4): 124—59.

Hainmueller, Jens, Dominik Hangartner, and Teppei Yamamoto. 2014. “Validat-
ing vignette and conjoint survey experiments against real-world behavior.” Proceedings
of the National Academy of Sciences, 112(8): 2395—2400.

Harrison, Glenn W., and E. Elisabet Rutström. 2008. “Experimental Evidence on
the Existence of Hypothetical Bias in Value Elicitation Methods.” Vol. 1 of Handbook
of Experimental Economics Results. 1 ed., 752—767. Elsevier.

Internal Revenue Service U. S. Department of the Treasury. 2008. “IRS An-
nounces Economic Stimulus Payment Schedules, Provides Online Payment Calcu-
lator.” https://www.irs.gov/uac/irs-announces-economic-stimulus-payment-schedules-
provides-online-payment-calculator.

Jappelli, Tullio, and Luigi Pistaferri. 2014. “Fiscal Policy and MPC Heterogeneity.”
American Economic Journal: Macroeconomics, 6(4): 107—36.

Johnson, David S., Jonathan A. Parker, and Nicholas S. Souleles. 2006. “House-
hold Expenditure and the Income Tax Rebates of 2001.” American Economic Review,
96(5): 1589—1610.

Juster, Thomas F. 1966. “Consumer Buying Intentions and Purchase Probability:
An Experiment in Survey Design.” Journal of the American Statistical Association,
61(315): 658—696.

Juster, Thomas F., and Robert P. Shay. 1964. Consumer Sensitivity to Finance
Rates: An Empirical and Analytical Investigation. National Bureau of Economic Re-
search, Inc.

Kan, Kamhon, Shin-Kun Peng, and Ping Wang. 2017. “Understanding Consump-
tion Behavior: Evidence from Consumers’ Reaction to Shopping Vouchers.” American
Economic Journal: Economic Policy, 9(1): 137—53.

Karlan, Dean, Adam Osman, and Jonathan Zinman. 2016. “Follow the money not
the cash: Comparing methods for identifying consumption and investment responses
to a liquidity shock.” Journal of Development Economics, 121(C): 11—23.

Leigh, Andrew. 2012. “How Much Did the 2009 Australian Fiscal Stimulus Boost De-
mand? Evidence from Household-Reported Spending Eﬀects.” The B.E. Journal of
Macroeconomics, 12(1): 1—24.

21
List, John, and Jason Shogren. 1998. “Calibration of the diﬀerence between actual
and hypothetical valuations in a field experiment.” Journal of Economic Behavior and
Organization, 37: 193—205.

Manski, Charles F. 1990. “The Use of Intentions Data to Predict Behavior: A Best-
Case Analysis.” Journal of the American Statistical Association, 85(412): 934—940.

Manski, Charles F. 2004. “Measuring Expectations.” Econometrica, 72(5): 1329—1376.

Manski, Charles F. 2017. “Survey Measurement Of Probabilistic Economic Expecta-
tions:Progress And Promise.” Vol. 32, 411—471. Chicago University Press.

Parker, Jonathan A., Nicholas S. Souleles, David S. Johnson, and Robert
McClelland. 2013. “Consumer Spending and the Economic Stimulus Payments of
2008.” American Economic Review, 103(6): 2530—53.

Rao, Vithala R. 2014. Applied Conjoint Analysis. New York:Springer.

Sahm, Claudia R., Matthew D. Shapiro, and Joel Slemrod. 2010. “Household
Response to the 2008 Tax Rebate: Survey Evidence and Aggregate Implications.” In
Tax Policy and the Economy, Volume 24. NBER Chapters, 69—110. National Bureau of
Economic Research, Inc.

Sahm, Claudia R., Matthew D. Shapiro, and Joel Slemrod. 2012. “Check in the
Mail or More in the Paycheck: Does the Eﬀectiveness of Fiscal Stimulus Depend on
How It Is Delivered?” American Economic Journal: Economic Policy, 4(3): 216—50.

Shapiro, Matthew D., and Joel Slemrod. 1995. “Consumer Response to the Timing
of Income: Evidence from a Change in Tax Withholding.” American Economic Review,
85: 274—83.

Shapiro, Matthew D., and Joel Slemrod. 2003a. “Consumer Response to Tax Re-
bates.” American Economic Review, 93(1): 381—396.

Shapiro, Matthew D., and Joel Slemrod. 2003b. “Did the 2001 Tax Rebate Stimulate
Spending? Evidence from Taxpayer Surveys.” Tax Policy and the Economy, 17: 83—109.

Shapiro, Matthew D., and Joel Slemrod. 2009. “Did the 2008 Tax Rebates Stimulate
Spending?” American Economic Review, 99(2): 374—79.

Smeeding, Timothy M., Katherin Ross Phillips, and Michael O’Connor. 2000.
“The EITC: Expectation, Knowledge, Use, and Economic and Social Mobility.” Na-
tional Tax Journal, 53(4): 1187—1210.

Thurstone, L. L. 1931. “The Indiﬀerence Function.” The Journal of Social Psychology,
2(2): 139—167.

File and source

File: reported-effects-versus-revealed-preference-estimates-evidence-from-the-propensity-to-spend-tax-rebates.pdf
Size: 334,401 bytes
SHA-256: 05d3873f192db131017cdc912d6c39e00a4d7c3721e4bbce9cbe08f24b502b50
Our copy: reported-effects-versus-revealed-preference-estimates-evidence-from-the-propensity-to-spend-tax-rebates.pdf
Original: www.nber.org