Measuring and Framing Support for Universal Basic Income

We execute an original survey experiment to examine the extent and determinants of support for a nascent policy issue, universal basic income (UBI), in the American public. We explore the effects of how UBI is framed (either in the context of values or the context of policy), ﬁnding three key results. First, UBI is primarily a Democratic and liberal policy. Second, negative arguments against UBI move support for UBI more than positive arguments. Third, and surprisingly, respondents are equally affected by both policy-driven and value-driven arguments about UBI. In conclusion, an increase in messaging about UBI is likely to widen existing partisan differences in UBI support. These differences are unlikely to be won over by policy or values arguments.


Introduction
Policy attention to a relatively new issue, Universal Basic Income (UBI), has exploded in recent years. Broadly speaking, UBI promises an unconditional cash transfer from government to its citizens, as (proponents argue) automation displaces individuals from employment with no practical replacement. Small-scale trials of the policy (in Hudson, New York, Stockton, California, Finland, and Kenya, among others (Ward, 2020)), combined with direct stimulus payments through coronavirus relief funds (Jammot, 2020), have brought an unprecedented amount of attention to UBI in the American context. From a basic social science standpoint, however, there is a lack of data regarding American public opinion towards UBI. This gap is critical: as the conversation around UBI accelerates, there's little corresponding evidence on how receptive the American public would be to a large-scale UBI policy.
We seek to fill this gap. In so doing, we focus our attention on the policy implications and values arguments inherent in UBI. These two dimensions -policy and values -are especially important, as they encapsulate the narrative of how UBI is framed in popular discourse. Winfree (2019) is a typical example: in that news story, UBI is simultaneously referred to as a policy espoused by individuals "concerned that future workplaces will replace humans with robots" (a policy solution) that also reflects "the idea that everyone is entitled to an income" (a value proposition). However, "the sustainability of any basic income program depends on whether a democratic system could support and afford it over a period of time" (a policy cost), and it violates the principle that "people who can work are expected to pay their own way" (a value violation). These four considerations are simultaneously operating in a single story; however, given the limitations of current research, we currently have no way of discriminating which of these considerations is most important in driving support or opposition to UBI. While some research on UBI exists in the European context, there is little to no work on the determinants of American opinions towards the policy.
Classic work on American public opinion suggests a puzzle, rather than an answer. On one hand, UBI has the potential to cut across traditional partisan or ideological divisions on the basis of values, given that many sorts of individuals are vulnerable to impending automation. This unique economic consequence would implicate two American core values: economic individualism and equality of opportunity (Feldman, 1988). It could also be the case that UBI maps to existing policy cleavages on social welfare. Campbell and Gaddis (2017), examining social welfare policy broadly, find respondents overwhelmingly volunteered (unprompted) opinions that denounce cash transfers, the exact policy of UBI. One respondent summarized the tension nicely: "I don't agree with giving cash -too easy to abuse the privilege. Food stamps and free medical benefits would be more appropriate" (Campbell andGaddis, 2017, 1365).
Using a survey experiment on an American MTurk sample, we demonstrate how these implicit arguments around UBI affect support for the program when they are made more or less salient to respondents. We find that, in line with prior research on negativity bias, negative frames decrease support for UBI, while positive frames have little effect on support for UBI. The effects of negative frames are concentrated among conservatives, who are already likely to be receptive to such arguments. Liberals, who are inclined to favor UBI, are not affected by negative frames. This suggests that UBI is an "easy issue" (Hill, Jordan, and Hurley, 2015) for many ideological respondents, with negative framing of UBI reminding them of preexisting beliefs. In the next section, we review the origin of these frames and apply them specifically to the question of UBI.

Framing UBI: Policy, Values, and the Negative/Positive Difference
While little research exists on support for UBI in the American context, very recent research on UBI focuses on public support in Europe. These studies find UBI support is higher in countries where social spending is low (Parolin and Sioland, 2020) and among left-leaning individuals (Schwander and Vlandas, 2020), but may be context-dependent, given the much stronger European support for redistribution policies generally. In contrast, redistribution policy and attitudes in the United States are more conservative, making American public opinion about UBI more uncertain. Even the designers of small-scale UBI trials note this uncertainty, with one stating that her UBI trial results "could support arguments both for and against basic income" (Henley, 2020; Eurobarometer Survey 471, 2017).
Given that UBI involves competing considerations, we need to understand how the arguments presented for or against UBI might shape support for the policy, especially since these frames can intentionally push respondents to support one position or another (Tourangeau, Rips, and Rasinski, 2000). We focus on two particular types of frames: frames about the policy effects versus values inherent in UBI, and frames that positively or negatively frame these considerations.
Questions with policy frames ask respondents about potential government actions in terms of policy effects. These include increased or decreased government spending (Jacoby, 2000), as well as tax rates, economic productivity, and political participation (Sniderman and Theriault, 2004). Questions with values frames, in contrast, ask respondents to evaluate political decisions in terms of their own values and whether the potential decision supports or opposes them (Johnson, 2012). With values, the respondent does not need to know anything about how a political decision might relate to other policies or political phenomena. This last point is especially important, as values might help respondents form opinions on relatively new issues (like UBI), and discussing issues in the context of values might shortcut a respondent's evaluation of policy implications.
Whether a question is framed negatively or positively (also known as valence frames; see Bizer, Larsen, and Petty (2011)) can also affect the opinions that survey respondents give, even on questions that only vary in the way probability is expressed (Kahneman and Tversky, 1984). This extends to survey experiments, where framing attitudes negatively can cause respondents to hold them more strongly (Bizer, Larsen, and Petty, 2011). This effect can even persist if followed by a positive frame, providing additional evidence for the strength of negativity bias (Boydstun, Ledgerwood, and Sparks, 2019). The conflicts between policy versus values frames, and positive versus negative frames, drive our expectations on support for UBI.
Regarding the first distinction -policy versus values -we expect that values will outweigh policy, as public opinion on policy that is easily understood by average citizens, and fits into existing knowledge, is more affected by values frames. Past work indicates values are more general than specific attitudes about policy (Feldman, 2003, 480) and conceivably may have a stronger effect on how the public evaluates political phenomena. Feldman (2003, 490) notes that "easy" issues, including those related to "moralistic or economic values" (Pollock, Lilie, and Vittes, 1993, 30), are more connected with values. Views on UBI may be related to values of working hard and the proper role of government involvement in the economy. Thus, we expect that public opinion on UBI is more strongly affected by values framing than policy framing.
Regarding the second distinction -positive versus negative -we expect that negative frames will swamp the effects of positive frames. On their own, positive frames might raise support for UBI, either by framing UBI as a positive solution to a policy problem or a positive reinforcement of a value proposition (like reducing inequality). However, as soon as these considerations are presented alongside corresponding negative arguments with respect to policy or values, the effect of the negative frames will dominate the positive frames. Like many other issues related to the size and scope of government, UBI involves potentially costly tradeoffs between taxes and spending which easily can be framed in negative terms. We formalize these expectations with two hypotheses: H 1 : Frames about the values inherent in UBI will have a larger effect on support for UBI more than frames about its policy implications.
H 2 : Negative frames will have a larger (negative) effect on support for UBI than the (positive) effect of positive frames.
Although UBI is relatively new on the policy agenda, and it could activate cross-cutting partisan or ideological sentiments, UBI does map somewhat to existing partisan and ideological cleavages in social welfare policy more broadly. We know that contradictory considerations to previously held beliefs are more likely to be dismissed (Zaller, 1992), so we expect question framing should be conditional on partisan or ideological predispositions. Due to negativity bias, however, we expect these framing effects to be stronger when acting as a negative reinforcement of beliefs.

Experimental Research Design and Data
We test our hypotheses using a 3 x 3 between-groups factorial design. The experiment was conducted November 1-2, 2019, and a sample of 3,600 Mechanical Turk (MTurk) users participated. 1 Our dependent variable of interest is support or opposition to UBI. To measure this support, we first prompted respondents with a bare minimum of information about UBI: Now we'd like to get your opinion on new policy proposal. There is a new policy being tested around the world called Universal Basic Income (UBI). This policy guarantees a payment of money to individuals from their government, regardless of need, qualification, or work.
Next came our experimental manipulation. There are two treatments given to respondents: an argument based on policy, followed by an argument based on values. In each treatment, there are three potential conditions: a positive condition (that UBI could solve policy or value considerations), a negative condition (that UBI exacerbates policy or value considerations), and a control (no argument for or against policy or values). Each respondent received one of the three conditions for each treatment. The experiment structure is more full outlined in the Supporting Information Appendix.
The policy treatment's negative condition is that UBI would be a more expensive social welfare policy than any before it ("Some argue that it would be much too expensive to implement on a broad scale"). The policy treatment's positive condition is economic stimulus and lower effects of unemployment ("Some argue that it would help solve unemployment problems caused by increasing automation across a variety of industries"). The value treatment's positive condition is that, because of increasing automation, it is only fair that the government institute policies that reduce income inequality and spread the gains of technology ("Others argue that it would help make society more fair by reducing income inequality caused by automation across a variety of industries"). The value treatment's negative condition is that UBI might discourage hard work and unfairly redistribute resources ("Others argue that it would encourage laziness by rewarding someone for not working"). After being exposed to these manipulations, respondents were asked for their support for UBI on a five-point scale.
We designed these frames to reflect popular discourse around UBI. 2 Additional validity for our treatments is reflected in the latent way the respondents themselves think about UBI. At the end of the survey, we asked respondents "When you hear about a policy called Universal Basic Income (UBI), which guarantees a payment of money to individuals from their government, regardless of need, qualification, or work, what are your own personal thoughts?" Respondents gave us open-ended responses such as those seen in Table 1. These responses all come from the control condition, so they are not simply a reflection of the treatments. Note that they implicate both policy and value considerations.
In modeling support for UBI, we account for standard political and demographic information: party identification, ideology, income, past voting behavior, and education. In particular, ideology is measured with the standard American National Election Study (ANES) self-placement on the scale of liberal to conservative, common in American political behavior (for instance Jordan and Ferguson 2016; Lupton et al. 2020). Partisanship is similarly measured with the standard ANES branching question. We also include indicators to control for a respondent's a priori level of concern about the economy: economic retrospections, economic prospections, concern about the deficit, and a dummy for employment status. Full treatment wordings, questions wordings, and a more in-depth explanation of the experimental structure are available in the Supporting Information Appendix.
While MTurk samples are not nationally representative, research has shown that MTurk samples are more diverse than typical convenience samples (Berinsky et al. 2012). MTurk samples also regularly replicate experimental Note: all open-ended comments drawn from pure control condition.
results from nationally representative surveys (Berinsky et al. 2012;Mullinix et al. 2015;Weinberg et al. 2014) and find reliable treatment effect estimates (Kennedy et al., 2020). Roughly half of our sample identifies as female and the average respondent has a college education. Most of the respondents are employed, with an average salary range of $40,000-$49,000. Politically, a majority (76%) of the respondents voted in the 2016 election and tend to lean liberal on our ideology scale. This is also reflected in the party identification of the respondents, where we see more Democrats than Republicans. These demographic characteristics are well balanced across our main experimental treatments. 3 A full breakdown of demographic characteristics is available in the Supporting Information Appendix.

Experimental Results
To test H 1 and H 2 , we regress support for UBI on our main independent variables of interest: the experimental treatments (policy and value arguments) and conditions (positive and negative arguments). To reflect H 3 and H 4 , we include an interaction of these treatments with party identification and ideology. 4 Before examining the interactions, we note some basic political dynamics. Democrats, even in the control conditions, are statistically more likely to support UBI than Republicans (β = 0.395, p < 0.05), but this is not true among ideologues. Interestingly, among moderate Independents, both of the negative treatments (negative value: β = -0.106, p < 0.05; negative policy: β = -0.165, p < 0.05) decrease support for UBI, but the positive treatments have no similar effect (early support for H 2 ). However, the magnitude of the policy treatment is greater than the magnitude of the value treatment, contradicting H 1 . 5 The interactions, however, limit our ability to draw inferences from the raw coefficients. Accordingly, Figure  1 shows the effects of the four framing treatments, ideology, and party in more detail. Each point estimate shows the predicted level of UBI support for the appropriate combination of ideology, party identification, and treatment. The panel grid is such that party identification is displayed in rows and ideology in columns: each combination, then, is the predicted support for UBI for that particular group (for instance, the top left panel shows UBI support among conservative Republicans). Within each panel, the value treatment (negative, positive, or control conditions) is shown grouped along the x-axis, and the policy treatment (positive, negative, or control conditions) is shown by character type. The estimates are presented with 95% confidence intervals. Calculating the effect of the policy treatment, holding the value treatment constant, can be done by comparing different character types within the same grouping To aid in interpretation, we first fully explicate a single panel before drawing attention to the broader results. Start with the results for conservative Democrats, shown in the top-right panel of Figure 1. The first group of points (the circle, triangle, and square) represent the estimated support for UBI for conservative Democrats who all received the negative value treatment but got the negative (circle), control/none (triangle), or positive (square) policy treatment. Conservative Democrats who received the negative value and negative policy treatment (the circle) only support UBI at about a 2.25, below the midpoint of the five-point scale. But the effect of receiving no policy treatment, relative to the negative policy treatment, is to increase this support from 2.25 to 3.0 (the triangle). If we wanted to calculate the effect of receiving no value treatment, compared to the negative value treatment, we would compare the circle in the center group to the circle in the left group. We can see that, holding the policy treatment at negative, replacing the negative value treatment with no value treatment also raises support for UBI from about 2.25 to 2.9. However, if we compared replacing no value treatment with a positive value treatment, holding policy treatment at negative (comparing the center circle to the right circle), there is no substantive change in support amount conservative Democrats (both circles are about 2.9).
Looking across the rest of the panels of Figure 1, within the policy treatment, negative frames affect support for UBI much more than positive frames do, consistent with H 2 . The negative policy condition strongly decreases support for UBI among all conservative groups, consistent with H 4 . Receiving a negative policy frame instead of a neutral frame has large effects, ranging from a decrease in UBI support of .4 for conservative Republicans to almost .6 for conservative Independents to about .7 for conservative Democrats. The substantive effects decrease conservative respondents' support for UBI from almost half a category on the five-point scale to about three-quarters of a category. In contrast, the positive policy condition does not affect anyone.
The effects of the values treatment exhibit a similar contrast. The negative values condition has measurable effects on conservative Independents and conservative Democrats. Receiving a negative condition instead of a neutral condition has marginally significant effects, 6 ranging from a decrease of about .4 for conservative Independents to a decrease of about .6 for conservative Democrats. The substantive effects decrease conservative Independents' and Democrats' support for UBI by about half a category on a five-point scale. The positive values condition, like the positive policy condition, does not affect anyone (more support for H 2 ).
The results in Figure 1 show strong support for H 2 . Negative frames have significant effects on UBI support among some groups, while positive frames have no significant effects. Looking at the right side of each of the nine boxes in Figure 1, there is no significant difference between receiving a neutral frame (the effects in the middle) and a positive values frame (the effects on the right). The confidence intervals for those estimates overlap for every partisan and ideological group. The confidence intervals for the triangle estimates (the neutral, no-policy-frame question) and square estimates (the positive policy frame question) overlap in every box as well. Interestingly, our results also allow us to tentatively reject H 1 . In contrast to previous research emphasizing the strength of values frames, values frames have a weaker effect on UBI support than policy frames, though not demonstrably so.
Among core groups already predisposed to support (or reject) UBI, the treatments have no effect. Conflicting frames do not affect UBI support among ideological respondents. Positive frames do not increase UBI support among conservatives (Republicans). Liberals (Democrats) resist negative frames about UBI. We therefore have strong support for H 3 and H 4 . Interestingly, though, negative conditions do further depress conservative (Republican) support for UBI, but we see no corresponding reaction in enhanced support among liberals (Democrats) who received positive conditions. 7

Discussion and Conclusion
Empirical research on public opinion about UBI in the American context has not caught up to popular discourse about the policy, and we remedy this by investigating the underpinnings of support for UBI. Overall, our survey experiment confirms negativity bias, bolstering a long line of behavioral research arguing that negative information has stronger effects on public preferences than positive information. Positive framing of UBI does not affect public support, while negative framing does. This negativity bias in public opinion is related to the loss aversion aspects of prospect theory (Kahneman and Tversky, 1984), and in the American context, has been shown to strongly affect individual attitudes about Congress (Munis and Benjamin Ashton, 2021). Our results provide further evidence that negativity bias affects a broad range of public attitudes. Future research should investigate this phenomenon more generally.
Despite UBI's novelty, American attitudes already appear to be hardened. Liberals are somewhat supportive, conservatives are somewhat opposed, and moderates are somewhere in the middle. To the extent that public attitudes on UBI are malleable, they mostly appear to be malleable for conservatives, whose opposition to UBI can be increased by negative question framing. This finding suggests that an increase in messaging about UBI from media, and Democratic and Republican politicians, is likely to widen existing partisan differences in UBI support. These differences are unlikely to be won over by argument, as liberal support for UBI is not affected by negative question framing, and conservative opposition to UBI is not affected by positive question framing. This resilience to framing challenging ideological predispositions comports with expectations and suggests UBI is what Carmines and Stimson (1980) called an "easy" issue.
UBI satisfies Carmines and Stimson's (1980)  For the public, UBI may be just one more facet of the ongoing conflict between the Democratic and Republican parties over the size and scope of government involvement in the economy. This is an important distinction, as proponents of UBI often present UBI as an issue with the potential to achieve a lot of bipartisan support. 8 Our preliminary analysis suggests this is unlikely. In fact, the factors that explain support for UBI appear similar to those that explain support for a government-guaranteed job and standard of living, another economic policy that is polarizing in American politics, as evidenced by our placebo test (see the Supporting Information Appendix).
The most prominent group of voters that face competing considerations over whether to support or oppose UBI appears to be conservative Democrats, an important part of the Democratic Party coalition. In response to UBI messages with negative policy and values frames, they are likely to become more opposed to UBI. Our initial results therefore show some potential for UBI to become a wedge issue (Hillygus and Shields, 2009) with this group.
For most others, UBI is an easy issue, and already somewhat divides liberals and conservatives. As political elites provide cuing information to their co-partisans in the electorate (Hill et al., 2015), UBI may simply be mapped onto the longstanding divide on economic issues between the two parties. UBI may become one more divisive issue in the gradual "conflict extension" (Layman and Carsey, 2002) of American political polarization to encompass more and more policies.
the open-ended responses through descriptive word clouds, a discussion of the experimental structure, tabular results, detailed coding summaries, and a validation of our sample and a discussion of our attention checks are all shown in the Supporting Information Appendix. 4 While the two are traditionally highly correlated, the unique intersection of values and policy inherent in UBI, coupled with growing ideological divisions within each party (Groenendyk, Sances, and Zhirkov, 2020), justifies our attention to the interactions separately. Partisans whose policy preferences do not match those of their party behave differently than those whose preferences do (Green, 2020). 5 Tabular results are available in the Supplemental Appendix. In addition, we initially focus all respondents (following Berinsky, Margolis, and Sances (2014)), but in the Supplemental Appendix, we show analysis for only survey respondents who passed an attention check and did not speed through the survey. The results are substantively identical. 6 These effects are significant at p < 0.1 7 We also execute a robustness check through a placebo test outlined in the Supplemental Appendix. We test whether our frames affected support for a similar policy: government guaranteed jobs. We find no effect, even though the two attitudes are correlated at r = 0.56. 8 For one example, see Breland (2017).