The Policy Feedback Effects of State Assistance: Did Small Business Bailouts Increase Support? (Journal of Experimental Political Science, 2026)

Full text

Journal of Experimental Political Science (2026), 1–10
doi:10.1017/XPS.2026.10031

RESEARCH ARTICLE

The Policy Feedback Effects of State Assistance:
Did Small Business Bailouts Increase Support
for Public Aid Programs?
Neil Malhotra and Saikun Shi
Stanford University, USA
Corresponding author: Neil Malhotra; Email: neilm@stanford.edu

Abstract
We examine whether exposure to government assistance can generate positive policy
feedback effects among constituencies not traditionally supportive of the welfare state.
Focusing on the U.S. Paycheck Protection Program (PPP), a small business bailout enacted
during the COVID-19 pandemic, we study how making government assistance salient
affects attitudes toward social assistance programs among a typically Republican-leaning
and relatively affluent group. Leveraging a bespoke survey of verified program recipients
and an embedded experimental manipulation, we find that reminding PPP recipients of
program participation increases support for government spending on healthcare, nutri-
tional assistance, and unemployment benefits by an average of 6.9 percentage points—
equivalent to roughly 16 percent of the partisan divide on these issues. The findings
provide novel causal evidence that making the receipt of government assistance salient
increases support for anti-poverty programs among well-off people, even when those
programs do not directly benefit them.

Keywords: redistribution; policy feedback; government aid; survey experiment; small business owners

Introduction
A prominent explanation for variation in public support for government assistance
programs centers on path dependence and the feedback effects of government
programs (Korpi and Palme, 1998; Pierson, 2001). Once put in place, goes the
argument, policies influence public support through two channels: they shape the
resources people possess as well as their interpretative schema, including who is
deserving of assistance (Béland et al., 2022).
In line with the resource-based account, research shows that higher-income
individuals tend to be less supportive of public spending on assistance programs to

This article has earned badges for transparent research practices: Open data and Open materials.
For details see the Data Availability Statement.

© The Author(s), 2026. Published by Cambridge University Press on behalf of American Political Science Association. This
is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (https://
creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction, provided the
original article is properly cited.

Downloaded from https://www.cambridge.org/core. IP address: 73.252.170.234, on 04 Oct 2026 at 19:13:43, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/XPS.2026.10031
2 Neil Malhotra and Saikun Shi

the needy than lower-income people. This is consistent with the fact that better-off
individuals are less likely to benefit from or rely on such programs in the future.
Explanations related to the second channel hold that when policies are instituted,
they also affect people’s interpretive and informational environment. Through the
implementation of the policy, people are exposed to pertinent information—about
how the government operates, the efficiency or impact of the programs (or lack
thereof), and empathetic understanding of social risk—that subsequently feeds their
assessment of the policy and of related matters.
Evaluating the feedback effects of a policy is difficult because exposure to it is
typically endogenous: those who utilize the program are often a self-selected group
that differs from others, making it hard to tease out the attitudinal effects of the
policy from other factors related to the characteristics of the group exposed to the
policy. Given this problem, this study leverages a unique setting in which people
who typically do not rely on direct government assistance were in need of sub-
stantial help due to exogenous factors (i.e., a global pandemic). Specifically, we
examine the effect of the Paycheck Protection Program (hereafter PPP)—one highly
salient policy—on a constituency of small business owners in the aftermath of a
major economic shock (the COVID-19 pandemic). Hence, our study focuses on the
second channel described above: whether and how government assistance affects the
interpretive schema of well-off individuals. We use an experimental design to mimic
a heightening of the salience of government assistance to a population not tradi-
tionally used to government support and assess the downstream effects of this
experience on policy preferences.1
Substantively, small business owners are an important, yet understudied,
population in American politics. In the U.S., there are over 33 million small
businesses, defined as independent businesses that have fewer than 500 employees.
These small businesses employ almost half (46%) of all American workers and are
the source of about two-thirds of the growth in employment over the past quarter
century.2 Moreover, they tend to be better off economically than the average
American, making them an excellent case study for examining the effects of social
assistance on a group that is typically not dependent on the welfare state (Malhotra
et al., 2025).
Leveraging a bespoke sampling frame of PPP recipients, we conduct a survey in
which we embed an experimental manipulation where half of respondents were
randomly assigned to receive a salience prime. They were reminded of their par-
ticipation in the PPP program and that the government had bailed out small
business owners whose livelihoods were in jeopardy during the pandemic, when
many forms of economic activity ground to a halt. Unlike other forms of financial
assistance that come in the form of subsidies or tax breaks, the PPP provided direct

1
There are several potential mechanisms underlying interpretive schema, including perceptions of the
efficacy of programs, the competence of government, and norms of fairness and reciprocity. Although our
empirical design is not intended to disentangle these mechanisms, they all point to a positive treatment
effect of salience.
2
See U.S. Chamber of Commerce report: https://www.uschamber.com/small-business/state-of-small-busi
ness-now.

transfers of money as forgivable loans to small business owners that helped them
make payroll and keep their businesses afloat during a stressful time. On the margin,
this manipulation made the program salient to PPP recipients for whom it may not
have been top of mind. As such, from respondents who were affected by the
treatment, we can learn about the way the program as a whole changed attitudes by
exposing people to a visible case of them benefiting from social spending on a
government assistance program.
This study adds to prior efforts to document the feedback effects of government
assistance. Gilens (1999), for example, shows that means-tested welfare in the U.S.
depresses political support among both recipients and non-recipients by reinforcing
racialized deservingness beliefs. Demonstrating positive feedback effects, Campbell
(2012) exploits individual-level survey data and program eligibility rules to show
that Social Security and Medicare beneficiaries are highly supportive of both
programs and are accordingly more likely to vote and contact public officials than
are demographically similar non-beneficiaries. Mettler (2011) finds that recipients
of visible programs (e.g., Social Security) are substantially more likely to express
pro-government attitudes than recipients of “submerged” tax expenditures, even at
similar income levels. Our study adds to this work by showing that similar patterns
hold also in the context of economically better-off individuals who, for exogenous
reasons, happen to need government assistance. Further, whereas extant research
has shown policy feedback effects on support for politicians who implemented the
program, participation, trust, and beliefs of deservingness, we uniquely show that
government support policies can have spillover effects on attitudes toward anti-
poverty programs that do not directly benefit well-off recipients.
Specifically, we find that making PPP benefits salient caused small business
owners to increase their support for various government programs designed to assist
those in need: nutritional assistance, healthcare, and income assistance for the
unemployed. Overall, we find that the experimental prime increased support for
antipoverty programs by about 6.9 percentage points, which represents about 16%
of the partisan divide in support.
Taken together, this paper provides novel evidence of the policy feedback process
and advances the literature in several ways. First, the unique features of our case—a
group not typically reliant on social assistance programs needing help for exogenous
reasons—make it particularly useful for evaluating the feedback effects of public aid.
Second, our sampling strategy and survey design allow us to focus on an important
population for which we have proof from administrative records that they actually
received government support. Finally, our experimental evidence on the influence of
program salience is causal in nature, supplementing observational research on the
topic that correlates people’s awareness of and experience with government pro-
grams and their attitudes toward the welfare state.

Empirical design
Bespoke survey
As there is no national repository of small business owners, we constructed our
survey sampling frame based on the list of individuals who applied for the Paycheck

Protection Program (PPP) as of September 30, 2023.3 According to the US Small
Business Administration (SBA), PPP was part of a COVID-19 relief program that
aimed to help qualifying business entities keep their workforce employed through
uncollateralized, low-interest, and forgivable loans. PPP loans were available to any
US-based small business, sole proprietor, independent contractor, self-employed
person, 501(c)(3) nonprofit organization, 501(c)(19) veterans organization, or tribal
business that was in operation on February 15, 2020, and satisfied at least one of the
following conditions: (1) had 500 or fewer employees; (2) met the industry size
standards set by the SBA (i.e., had a tangible net worth not exceeding $15 million on
March 27, 2020, and an average net income not exceeding $5 million for the two full
fiscal years prior).
By the time of its conclusion in 2021, PPP provided $793 billion in loans to close
to 1 million organizations. We use PPP as the foundation of our survey sample for
two reasons. First, instead of relying on respondents’ self-reporting of benefit
receipt, using a government-verified list of small business owners helps us more
accurately target the population of interest. Second, unlike many other government
grants for small business entities, PPP is not set up to target any specific subsample
of businesses. In order to avoid double counting the same entity applying for
multiple rounds of loans, we rely on the first-draw loan list, which contains in total
557,859 entities who applied for PPP.
In order to make sure our sample is balanced and representative of the distri-
bution of PPP loans, we took the following steps: (1) a random sample of entities is
selected from the foundational PPP pool; (2) business owners of these entities as
well as their contact information are identified through various sources, including
utilizing the PPP application records, Internet searches, and LinkedIn data; (3)
among the matched list, a stratified recruitment sample that is balanced on entity
characteristics is invited to take the survey through a link. Each of these business
owners received at least one text and one email.
Respondents were first screened and disqualified from the survey if they self-
identified as non-small business owners or said that they never owned a small
business in the past. We also excluded responses based on various data quality
checks (see Online Appendix A for additional details). To reach our target sample
size while still keeping the final sample balanced, we launched the survey in four
waves, repeating the steps as detailed above. Respondents were provided an
incentive of $20 for participating. Overall, our matched business owner sample
containing personal contact information consists of 262,347 individuals. Of these,
221,142 individuals were invited to our survey. The final sample size is 701, which
makes the final qualified response rate to be 0.32%.4 Although this seems low, it is
comparable to survey efforts using probability samples that attempt to reach
individuals using digital methods. For example, the SMS portion of the Cornell

3
This experiment was part of a larger research project in which we also asked a series of questions about
SBOs’ political attitudes and business experiences. The descriptive profile of SBOs is reported in Malhotra
et al. (2025).
4
There is minimal within-survey attrition. Among the 701 respondents, 1 person did not reach the
portion of the survey with the experiment, leaving the main outcomes used for this paper unanswered;
3 additional people did not answer all the questions related to the main outcomes.

Midterm Election Survey has a final response rate of 0.8%. Recent phone surveys
conducted by the New York Times have response rates of 0.4%. Table 1 presents
characteristics of the sample as compared to the sampling frame. As shown in the
table, despite the low response rate, the sample is fairly representative of the PPP
recipients who were invited to take the survey as well as the larger PPP dataset.
Nonetheless, to address potential issues with non-response, we apply post-
stratification weights to the data. Because we employ methods of probability
sampling, we can reweight the data to make it representative based on the plethora
of sampling frame characteristics to which we have access. The final sample is
weighted based on mostly business entity characteristics along with some business
owner characteristics5 to match the PPP distribution list. The specific list of vari-
ables that are used to construct the weights is reported in Online Appendix A. The
survey process, including the construction of the sampling frame, was administered
by Verasight.
Table OA2 shows descriptive statistics of our sample of PPP recipients. We
compare this group against a national sample of the U.S. adult population con-
ducted by Prolific.6 Overall, we observe that the small business owner population is
older, less white, and wealthier than the U.S. population at large.

Experimental manipulation
For a randomly selected half of our sample, we ask respondents the following
question: “During the COVID-19 crisis, the US Government’s PPP provided
government-backed loans to businesses to help them keep their workforce
employed. Which of the following best describes your engagement with the
Paycheck Protection Program?” The response options were “I have been a recipient
of the program’s loans,” “I have applied for the program but did not receive a loan,”
“I have heard about the program but did not apply,” and “I have not heard about the
program.”
Importantly, the purpose of this question is not to elicit individual program
participation information—we know from administrative data that all respondents
have applied and received loans from PPP—but to remind respondents of a gov-
ernment program from which they have gotten financial support. The “Prime”
variable is an indicator for whether this question is asked before the measurement of
the outcome variables.7 Randomization was successful; balance tables are presented
in Online Appendix C.

5
Recall that PPP data is at the business-entity level and focuses more on the business entity rather than
the individual owner.
6
More details on the baseline Prolific sample can be found in the Online Appendix B.
7
Despite the fact that all people in the sampling frame were known to have received PPP funding, only
61% of respondents acknowledged receiving it. Why is this the case? Some business owners may not be
directly involved with the application for, or management of, PPP loan funds. Other business owners may
recall receiving PPP loans but be reluctant to disclose that on a survey. There have been a number of high-
profile prosecutions of PPP loan fraud. Even if respondents themselves did not commit fraud, they may be
wary about disclosing their participation in the program. Table OA1 presents descriptive characteristics of
respondents who acknowledged receiving PPP funding vs. those who did not. Although we cannot draw
strong inferences from these comparisons, it seems that less-sophisticated businesses (i.e., non-corporations,

Table 1. Descriptive statistics of paycheck protection program (PPP) recipients

Sample Sample
PPP (full) PPP (invited) (unweighted) (weighted)
N = 557,859 N = 221,142 N = 701 N = 701

Loan status
Charged off (%) 5.3 7.2 3.9 5.2
Exemption 4 2.8 3.8 2.0 2.9
Paid in full 91.9 89.0 94.2 91.9
Rural or urban
R 22.2 24.0 25.7 22.2
U 77.8 76.0 74.3 77.8
Business hub zone?
N 70.8 67.8 74.9 70.8
Y 29.2 32.2 25.1 29.2
LMI indicator
N 71.7 69.5 73.9 70.9
Y 28.3 30.5 26.1 29.1
Business type
Corporation 18.6 10.5 9.4 18.6
Limited liability company (LLC) 20.0 8.7 8.7 19.5
Sole proprietorship 32.2 48.9 52.2 32.2
Other 29.2 32.0 29.7 29.7
Business age
Existing or more than 2 years old 85.7 90.7 89.3 85.7
New business or 2 years or less 8.1 4.6 6.0 8.0
Startup, loan funds will open business 0.0 0.0 0.3 0.2
Unanswered 6.1 4.6 4.4 6.2
Loan term
0–24 50.9 33.1 34.2 50.9
25–48 4.6 6.8 3.3 4.3
49–60 44.5 60.1 62.3 44.4
Over 60 0.0 0.0 0.1 0.4
Initial approval amount
$0–10K 28.1 31.8 44.5 33.6
$10,001–25K 41.9 50.0 40.5 36.3
$25,001–100K 18.7 10.4 9.8 18.9
$100K+ 11.3 7.8 5.1 11.2
(Continued)

Table 1. (Continued )

Sample Sample
PPP (full) PPP (invited) (unweighted) (weighted)
N = 557,859 N = 221,142 N = 701 N = 701
Current approval amount
$0–10K 28.1 31.8 44.8 34.1
$10,001–25K 41.9 50.1 40.5 36.4
$25,001–100K 18.7 10.4 9.4 18.0
$100K+ 11.3 7.7 5.3 11.5
Forgiveness amount
$0–10K 25.5 28.6 41.5 30.1
$10,001–25K 36.5 42.0 37.7 33.7
$25,001–100K 18.3 10.2 8.8 16.9
$100K+ 11.0 7.7 5.3 11.5

Missing 8.6 11.5 6.7 7.8
Notes: PPP data as of September 30, 2023. Data was obtained through FOIA request.

Although the experimental manipulation presented in this paper was part of a
larger study on this sample, all respondents received the PPP experiment in the
same position in the survey.8

Outcome measures
We employ three measures to assess support for government redistributive policies.9
While asking these questions, we remind the respondents that higher government
spending would come from higher taxes, consistent with best practices to remind
respondents of tradeoffs inherent in public policies (Margalit and Raviv, 2024). For
ease of interpretation, we recode all survey measures to lie between 0 and 1.

those in low- and middle-income areas, those in distressed federal business zones, and those with longer
loan terms) were less likely to acknowledge receiving PPP.
8
There was one additional experimental manipulation that was not directly relevant to the context of this
paper, intended as a robustness check to assess whether asking about business experiences primed responses
to political attitudes. Half of respondents were asked about their small business owner experience before
questions about their political attitudes, and the other half were asked about their small business owner
experience after those questions. The PPP experiment, however, had a fixed position, which was after all the
aforementioned blocks. Any effect of the position of the experiment within the larger survey would be the
same for treatment and control groups and not produce any bias. We also tested for treatment spillover
effects using the method designed by Transue et al. (2009) and found none (i.e., the PPP treatment effect was
not conditioned by assignment in the prior manipulation, see Online Appendix Table OA7).
9
Our interest is in the impact of personal exposure to government assistance programs on welfare
attitudes more generally, not just about help for SBOs. We therefore ask participants for their attitudes on a
range of welfare programs that cover different domains—healthcare, nutritional assistance, and unem-
ployment assistance. This allows us to assess how broad and systematic the attitudinal effects of the
treatment are.

Healthcare. We ask whether the respondent would like to see more or less
government spending on healthcare for the poor: “Would you like to see more or
less government spending on health care for the poor? Remember that increasing
such spending might require higher taxes to pay for it.” The response options were
“spend much less,” “spend less,” “keep at current level,” “spend more,” and “spend
much more.”
Nutritional Assistance. We ask whether the respondent would like to see more or
less government spending on nutritional assistance (such as food stamps) for the
poor: “Would you like to see more or less government spending on nutritional
assistance for the poor (food stamps)? Remember that increasing such spending
might require higher taxes to pay for it.” The response options are the same as
Healthcare.
Unemployment Assistance. We ask whether the respondent would like to see
more or less government spending on income assistance for workers who lose their
jobs: “Would you like to see more or less government spending to provide income
assistance for workers who lose their jobs? Remember that increasing such spending
might require higher taxes to pay for it.” The response options are the same as
Healthcare.
Index. We build a combined scale based on the three aforementioned measures
via a simple average.

Results
As shown in Figure 1, the experimental prime increased support for all three
antipoverty programs. For the table version of this figure, see Online Appendix C
Table OA5. The figure presents estimates from simple OLS regression models
regressing the outcome variables against a dummy variable representing the
experimental prime.10 The strongest effect was for healthcare for the poor—making
PPP salient increased recipients’ support by 9.4 percentage points (p = .001). This
effect size is substantively large, representing about 22% of the association between
partisanship and the outcome variable. The prime increased support for nutritional
assistance by 7.5 percentage points (p = .009) and unemployment assistance by 4.0
percentage points (p = .120). Averaging across the three outcome variables, the
treatment shifted support for antipoverty programs by 6.9 percentage points overall
(p = .006). In other words, the treatment shifted respondents from being slightly
against the programs to somewhat in favor (the constant terms representing the
control group in Table OA5 are below 0.5). Although we can only speculate as to
why the unemployment outcome exhibited the weakest treatment effects, one
possibility is that the PPP was intended to reduce the need for unemployment
benefits by keeping workers on payroll. Alternatively, programs such as Medicaid
and SNAP are means-tested and targeted toward the poor, whereas unemployment
insurance is paid into by all workers and therefore might be perceived differently by
respondents.

10
As a robustness check, we also estimated an ordered logit model predicting the three policy outcomes,
which is reported in Online Appendix C. Marginal effects are plotted in Figure OA1. The regression results
are reported in Table OA6. These robustness checks return results that are similar to the main OLS results.

Figure 1. SBOs exhibit stronger support for antipoverty programs when primed about PPP. Notes: 95%
confidence intervals. Outcome is a 5-point scale on [0,1].

Conclusion
The role and size of the state have become a central topic of public discussion
following President Trump’s decision to form a new Department of Government
Efficiency (DOGE). Led by Elon Musk, this department had set its sights on
radically transforming the operation of government by shrinking the public sector,
eliminating prominent functions, and cutting an array of public programs. An open
question is what impact these changes will have on citizens’ views of the state as
some of its key operations come to the fore.
This paper offers insight on this question by investigating a case where a gov-
ernment assistance program is made salient to its beneficiaries. Leveraging a unique
sample of small business owners, we take advantage of a historic bailout program
that made the government’s role in mitigating social risk salient to a typically higher
income and disproportionately right-leaning constituency (Malhotra et al., 2025).
Our experimental approach allows us to study this research question causally. Our
findings reveal support for the idea that welfare policies have feedback effects that
extend beyond the policy itself. When the government implements a social assis-
tance program, it influences the way beneficiaries evaluate government intervention
on a broader set of causes and needs. Making salient government assistance does
appear to be effective in building support for the welfare state.
Supplementary material. The supplementary material for this article can be found at https://doi.org/10.
1017/XPS.2026.10031.

Data availability. The data, code, and any additional materials required to replicate all analyses in this
article are available at the Journal of Experimental Political Science Dataverse within the Harvard Dataverse
Network, at: https://doi.org/10.7910/DVN/WUQ8B5 (Malhotra and Shi 2026).

Competing interests. The authors report no conflicts of interest.

Ethics statement. This research was approved by the Stanford University Institutional Review Board (IRB
Protocol 71035). The authors affirm that the research adheres to APSA’s Principles and Guidance for
Human Subjects Research. Participants were shown an information screen with study information prior to
beginning the survey.

References
Béland, Daniel, Andrea Louise Campbell, and R. Kent Weaver. 2022. Policy Feedback: How Policies Shape
Politics. New York: Cambridge University Press.
Campbell, Andrea Louise. 2012. How Policies Make Citizens: Senior Political Activism and the American
Welfare State. Princeton, NJ: Princeton University Press.
Gilens, Martin. 1999. Why Americans Hate Welfare: Race, Media, and the Politics of Antipoverty Policy.
Chicago, IL: University of Chicago Press.
Korpi, Walter, and Joakim Palme. 1998. “The Paradox of Redistribution and Strategies of Equality: Welfare
State Institutions, Inequality, and Poverty in the Western Countries.” American Sociological Review 63(5):
661–87.
Malhotra, Neil, and Saikun Shi. 2026. “Replication Data for: The Policy Feedback Effects of State
Assistance: Did Small Business Bailouts Increase Support for Public Aid Programs?” Harvard Dataverse
V3. https://doi.org/10.7910/DVN/WUQ8B5.
Malhotra, Neil, Yotam Margalit, and Saikun Shi. 2025. “The Politics of Small Business Owners.” British
Journal of Political Science 55: e94.
Margalit, Yotam, and Shir Raviv. 2024. “Does Support for Redistribution Mean What We Think It
Means?” Political Science Research and Methods 12(4): 870–8.
Mettler, Suzanne. 2011. The Submerged State: How Invisible Government Policies Undermine American
Democracy. Chicago, IL: University of Chicago Press.
Pierson, Paul. ed. 2001. The New Politics of the Welfare State. Oxford, UK: Oxford University Press.
Transue, John E, Daniel J. Lee, and John H. Aldrich. 2009. “Treatment Spillover Effects Across Survey
Experiments.” Political Analysis 17(2): 143–61.

Cite this article: Malhotra N and Shi S (2026). The Policy Feedback Effects of State Assistance: Did Small
Business Bailouts Increase Support for Public Aid Programs? Journal of Experimental Political Science.
https://doi.org/10.1017/XPS.2026.10031

File and source

File: malhotra-2026-the-policy-feedback-effects-of-state-assistance-did-small-business-bailouts.pdf
Size: 245,947 bytes
SHA-256: fd373c9eed16b19155d397fd429db8bab061315c492fd1c24b5b35090b72ab6a
Our copy: malhotra-2026-the-policy-feedback-effects-of-state-assistance-did-small-business-bailouts.pdf
Original: No public link identified.