Public Preferences Over Changes to the Composition of Government Tax Revenue

Abstract

How governments raise tax revenue is at the core of domestic political conflict. Public opinion towards taxation is measured generally and qualitatively by many surveys, but previous research has not closely linked public preferences to the budget problem faced by governments regarding how best to raise or cut a marginal quantity of revenue. We present results from a novel tax preference experiment in which UK respondents are given choices over different tax ‘levers’ that are expected to raise or cut equal revenue. We find that while different tax levers vary substantially in their popularity, there is a ‘hidden consensus’ regarding different tax levers across income levels and partisanship of the respondents.

Collecting taxes is one of the most fundamental actions of government, and decisions about how to raise revenue have important consequences for distribution and growth. However, we know relatively little about how citizens would prefer government revenues to be raised: which taxes are popular (or less unpopular) and with whom. The burgeoning experimental literature on public tax policy preferences has largely neglected questions about the tax mix, while scholarship on the tax mix has sometimes overlooked public opinion.

Inattention to public preferences over how tax revenue is raised is surprising in light of canonical political economy models highlighting the optimization problem that balances political satisfaction and revenue goals (Hettich and Winer 1984). From a policy perspective, political science has produced little direct evidence regarding the ‘dissatisfaction prices’ of different revenue sources, a critical question in a time of high public deficits and rising future spending pressures. We study preferences over revenue-equivalent tax changes in the UK. We propose marginal changes to actually-existing taxes to a nationally representative sample of voters. Our survey experiment presents a choice between randomly paired possible changes to two different taxes at a time, specifying the quantitative change needed for each tax to generate the same revenue change. We model respondents' choices following a Bradley-Terry framework (Bradley and Terry 1952) to estimate the relative popularity of different revenue-equivalent changes to the tax structure.

This empirical exercise makes three important contributions. First, we provide a comprehensive description of preferences over the balance of all the major taxes in the UK system, providing rare empirical evidence on public opinion over the tax mix. The differences in popularity between the relatively preferred versus disliked taxes suggest that there is space in the UK tax system for majority-popular reforms. Second, we are able to separate preferences over the composition of taxation from preferences over its level. This reveals a hidden consensus among voters over where revenue should be raised. While partisanship and material interest may generate disagreement over the appropriate level of taxation, there is widespread agreement on its composition.

Finally, our approach contributes to the emerging experimental literature on preferences over taxation (Ballard-Rosa, Martin, and Scheve 2017; Kneafsey and Regan 2022), expanding its scope to consider the composition of revenue collection across a wide range of taxes. Understanding public tax attitudes through this cross-tax lens is an important complement to these studies which often focus on explaining the unpopularity of certain taxes – especially those with redistributive benefits (Elkjær et al. 2023; Scheve and Stasavage 2022) – but which do not allow for the even lower popularity of raising revenue through less progressive channels.

Tax Composition and Public Preferences

Our theoretical inspiration comes primarily from an old public choice approach that sets the political resistance generated by different taxes against the revenues generated from each tax base (Hettich and Winer 1984). In the original model, the marginal pain of a pound paid in tax is assumed to be equal across taxes, increasing non-linearly in the rate. Additional political costs arise from (different) administrative burdens across tax bases. Balancing revenue gains with political costs implies a diversified tax base due to the increasing marginal costs, with higher relative reliance on easily administered taxes. However, to our knowledge, there have been no empirical calibrations of these popularity costs.¹

Citizens may also dislike some taxes more than others for reasons beyond financial and administrative burdens, as highlighted in existing research. Particular attention has been given to visibility (Wilensky 2002), fairness (Scheve and Stasavage 2022), and progressivity (Prasad 2006). However, the generality of these categories, and the potential for slippage between tax design and voter perception, mean that they do not provide strong expectations about attitudes towards specific taxes.

On visibility, we follow Martin and Harper's (2021) view that attributions of visibility are typically based on untested assumptions, and sometimes on circular reasoning, where opposition to a tax is cited as an indication of its visibility, and visibility is given as the reason for opposition. Where more specific predictions are made, visibility arguments often derive from idiosyncratic features of the United States tax system, which has received the most scholarly attention (Campbell 2018).

Equally, the perceived fairness of a tax seems intuitively likely to affect its popularity but what fairness consists of is indeterminate. Some accounts point to ‘equal treatment’ (Scheve and Stasavage 2022), but countervailing evidence points to fairness as the ‘ability to pay’ (Daunton 2002), inherently requiring unequal treatment. Similarly, misperceptions of how taxes actually work can lead to slippage from what voters might think is fair under full information (Kuziemko et al. 2015). This makes it difficult to hypothesize in advance which taxes should elicit greater support on fairness grounds.

The one exception here, perhaps, is to expect progressive taxes to be relatively popular. A large body of work finds widespread support for the principle of progressivity (Barnes 2014; Limberg 2020) and majority support for progressive changes from the status quo (Ballard-Rosa, Martin, and Scheve 2017).

However, studies of support for progressivity have focused more on variations between people than comparisons with other taxes. Progressivity preferences have been shown to be highly structured by income (Beramendi and Rehm 2016) but this has not been cleanly empirically separated from this tax-level effect since progressivity is typically presented as higher taxes on the rich, not lower taxes on the poor.

Meanwhile, in the literature on the tax mix, considering public opinion over types of taxes directly is rare. The central explanations of variations across countries (and over time) are located in political institutions and the relative power they give to groups with different interests (Kemmerling and Truchlewski 2021). The preferences are inferred from the material positions of these groups. Those with lower incomes ‘should favor a more progressive tax system, whereas richer voters should reject tax progressivity’ (Haffert 2021, 99). Since they consume a larger share of their incomes, the less well off should be less supportive of taxes on consumption. Symmetrically, (progressive) taxes on income and capital fall more heavily on the better off (Timmons 2005). These materialist building blocks underpin the taxes that different parties and organized interests endorse; constituents' preferences are assumed rather than investigated. The prediction of variation in tax-mix preferences across income and partisan groups motivates our empirical verification.

Empirical Approach

We examine preferences over tax composition at the margin of current UK tax policy and consider variations in preferences by income and party vote in a novel survey. Our design directly tracks the quantities we want to estimate. Our interest in tax composition means we want to consider preferences over budget-equivalent propositions. Second, we want to make sure that the comparisons we analyse are quantitatively informed. Otherwise, people may overestimate the feasibility of raising revenues from certain taxes (Johnson 2023). Third, we want to elicit preferences over a comprehensive set of tax levers rather than (only) those most salient to researchers. Taken together, these three considerations point to asking respondents their opinions on revenue equivalent increases (or decreases) to as many existing taxes as possible.

We are able to do this in the UK thanks to the annual publication (by HMRC, the central tax authority) of the revenue effects of indicative changes to major national taxes: Income Tax, Corporation Tax; Capital Gains Tax; Inheritance Tax; National Insurance contributions; Stamp Duty Land Tax;² duties on alcohol, tobacco and fuel; and VAT rates. Where possible, the revenue estimates incorporate estimates of taxpayers' behavioural responses (HMRC, 2021). The data cover major thresholds as well as rates. We used the figures from June 2021 to calculate the changes to twenty-three tax levers implied by the same (£1 billion) revenue change from the status quo.³ This incremental approach is similar to how tax policy tends to be made through small adjustments to existing revenue levers (Rose and Karran 1987).

We presented 9,713 respondents with one pairwise choice between tax changes.⁴ Our survey was fielded by YouGov to a nationally representative sample of UK adults between the 4^th and the 14^th of October 2021. Each response is a choice between two reforms relative to the pre-existing baseline, and each proposal includes the headline change, an account of how the relevant tax works and the size of the change required to raise or cut the required revenue. Figure 1 shows an example choice, as delivered to the respondents.

Figure 1. Survey Experiment Prompt Example. The direction of the change (increase/decrease) and the two taxes proposed are randomized across the choices. The size of the change to the tax is determined by the change necessary to change the revenue yield by £1 billion. See the supplementary material for the texts of all randomized statements.

Our presentations are different to the way citizens typically encounter tax proposals. In public debate, there is usually no counterfactual budget-equivalent option to change to another tax instead. Tax reform proposals also typically provide less practical explanations and more overt normative framing. It is not our concern here to ascertain the effects of framing on tax popularity (it matters, McCaffery and Baron 2004). Rather, we try to elicit any views the public may have on the underlying budget problem where revenue equivalencies are critical. Budget-equivalent alternative proposals reflect an important feature of political reality if one less commonly presented to the public.⁵

Basic Response Statistics and Task Complexity

Of 9,713 responses to our experiment, 2,565 endorse proposal A and 2,528 endorse proposal B. Neutral responses amounted to 4,620, of which 2,911 expressed ‘I think both of these changes are equally good or bad’ while 1,709 ‘Don't know’.⁶ The latter may include respondents who failed to engage with the task; in real politics, individuals equally fail to engage with the task. We retain both neutral responses rather than dropping respondents to maintain representativeness. Higher rates of neutral responses for particular taxes simply make these less likely to be estimated as especially popular or unpopular.

The extent of the neutral responses is understandable given that the random pairwise comparisons yield many comparisons that even well-informed individuals might not have strong views about.⁷ We see some evidence of variation in neutral response rates by the complexity of the choice.⁸ However, some real tax changes would be complex and it is of substantive interest if that yields neutrality. What we ask of respondents is still less complicated than many applications in the literature (for an example on the spending side, see Bonica 2015).

Models for Tax Preference Choices

We build a series of models to summarize the data. Using Y_i to denote respondent i ‘s choice, we code responses as follows:

• • Y_i = 1 if respondent prefers A

• • Y_i = 0.5 if respondent gives a neutral response

• • Y_i = 0 if respondent prefers B.

This allows us to interpret differences on the scale of proportions of respondents preferring one tax option to another while retaining the neutral responses.

Following a generalized Bradley-Terry model framework, we model the expected value of Y _i as a function of the competing ‘popularities’ π_j of different tax change proposals j. With proposals j ∈ A, B, this can be written as:

$$E[ {Y_i} ] = \alpha + \pi _{iA}- \pi _{iB}.$$

α is the expected value of Y_i when the two proposals are equally popular, that is if π_iA = π_iB.⁹ Note that the popularities in this model are only identified relative to one another: pairwise comparison data only yields information about relative, not absolute, popularity of options. Full identification and estimation details for our baseline and variant models are in the supplementary material.

Results: Preferences Over Tax Levers

Figure 2 shows estimates of the relative preferences for each tax lever (averaging over all comparisons in the experiment).¹⁰ The differences are substantial. Increasing (or not decreasing) the corporation tax rate is preferred to increasing (or not decreasing) Council Tax by 25 percentage points. With a representative level of neutral responses, this corresponds to a population-level response distribution where 37.5 per cent of respondents prefer the corporate tax rate increase, and only 12.5 per cent prefer the council tax increase. The remaining 50 per cent are indifferent or don't know. From the perspective of political efficiency, the differences across taxes imply that popular reforms to the composition of tax revenues are available.

Figure 2. Relative public preference for tax levers, in units of probability of supporting taxation via a given lever versus others.

Second, the taxes that are most popular are generally progressive: those on higher earners and on capital or corporate incomes. This is consistent with previous research asking about general preferences but replicates with reference to concrete policy levers. Moreover, while support for these taxes may be economically naive, our design decreases naivety as much as possible. We provided estimates which try to include the behavioural responses to tax changes and the scale of the required changes to rates reflects the narrow bases of these taxes.¹¹

The Hidden Consensus on Taxation

We also examine differences in the popularity of tax levers between types of respondents, characterized by income and partisanship. We discover very little variation by income and only slightly more by party in the taxes that British citizens prefer. This consensus may be hidden by divergent views on the overall level of taxation which contaminate simpler designs' estimates of the popularity of particular taxes.

Figure 3 shows estimates for respondents with household incomes above and below £45,000¹² and those who did not give an income response. Figure 4 shows estimates for Conservative and Labour voters. In both figures, the overall orderings of the taxes are similar across groups, and there are few levers (indicated with solid points on the figures) where there are statistically significant differences in the popularities of individual taxes between groups.

Figure 3. Relative public preference for tax levers for respondents with household incomes above £45,000 (blue squares), below £45,000 (red circles), and those who did not answer the income item (grey triangles) in units of probability of supporting taxation via a given lever versus others. Solid points indicate tax levers where the 95 per cent interval for the difference between those below £45,000 and the respective other group excludes zero.

Figure 4. Relative public preference for tax levers for 2019 Conservative (blue squares) versus 2019 Labour (red circles) voters, in units of probability of supporting taxation via a given lever versus others. Solid points indicate tax levers where the 95 per cent interval for the party difference excludes zero.

Only the corporation tax rate and council tax have statistically differentiable levels of popularity by income. Those with incomes over £45,000 view both of these taxes more favourably than those with incomes below £45,000. For corporation tax, this reinforces support for a very popular tax, while the council tax is less unpopular with high-income respondents. There are no significant differences by income for the two higher rates of personal income taxation (the higher and the top rates) nor for the threshold at which the higher rate kicks in. Higher-income respondents also endorse raising revenue through other progressive taxes (capital gains tax rates, stamp duty, and inheritance taxation) just as strongly as lower-income respondents. Overall, the correlation between the preference estimates for those with incomes under versus over £45,000 is 0.96.

There are more taxes where partisan differences can be found but, again, the headline picture is of consensus. Labour voters are more supportive than the Conservatives of higher rates of personal income tax on the highest earners and of raising revenue through inheritance and fuel taxation. Conservative voters are more supportive of three of the eight possible changes to social insurance contributions.

These social insurance differences deserve some comment. The UK Conservative government just announced changes to this tax when the experiment was fielded.¹³ These comprised small cuts to revenue via adjustments to tax-free allowances. Meanwhile, substantial increases in rates for employees and the self-employed increased revenue. In our data, one of these three rates (the main rate for employees) and two of the thresholds are more popular among Conservatives. While Conservative voters do not quite endorse the precise enacted changes, it seems plausible that the partisan patterns could reflect short-term effects rather than durable preference cleavages.

Even with this immediate pre-experiment shock to attitudes, partisan differences are not very large when considered across all levers. The correlation between the preference estimates for Labour vs Conservative voters is 0.82. This consensus is surprising in light of the comparative literature on the tax mix which grounds partisan differences in the divergent interests of different parties' constituents.¹⁴

An alternative interpretation of these patterns in the data is not consensus but incomprehension, or a lack of engagement. That is, sceptics may argue that our survey respondents are not really giving us meaningful responses to the choices we give them. We disagree with dismissing the consensus we see here on this basis of three reasons. First, we do see meaningful differences, in the aggregate, in the relative popularities of the taxes. Second, there are consistent differences between taxes within partisan groups; that is, it is hard to explain away the fact that (for example) the popularity advantage of the corporation tax over taxes on property transactions is the same among Conservative and Labour supporters. Finally, equally complex surveys on other topics – such as the allocation of government spending – do reveal strong partisan divisions (Barnes, Blumenau, and Lauderdale 2022). Overall, the fact that there are differences in the popularity of different taxes but that the patterns of variation differ little across types of respondents points more to consensus than to a lack of substantive engagement with the task.

Robustness

Our results are robust to a number of other experimental variations (reported in the supplementary material). First, we model choices over increases separately from decreases to gauge the appropriateness of our underlying idea of general popularity driving choices on both kinds of choices. Second, we consider much larger changes – £10 billion instead of £1 billion – for the ‘big five’ taxes with which it is plausible to raise that much revenue. Finally, we consider choices made when we provide additional arguments for or against both options as a check on the sensitivity of our results to differences in presentation. For all three of these variations, there is little evidence of any substantial difference from our main results.

Generalizability

How idiosyncratic is the result that there are popular, revenue-neutral tax reforms available, relative to the politically efficient tax mix? There may be some theoretical reasons to expect low responsiveness of policy to public opinion in Britain (Hobolt and Klemmensen 2008), making the gap between preferences and the status quo tax system that we discover unusual. But more recent data show little variation across countries, with the UK even among the more responsive (Rasmussen, Mäder, and Reher 2018). Taking taxation more specifically, politicians setting tax policy in Britain have relatively high levels of insulation (Steinmo 1993), but this cuts two ways: it limits direct public influence, but politicians (compared to tax experts or civil servants) are the policy actors most likely to be sensitive to public preferences.

On the popularity ranking of taxes, we cannot draw conclusions about whether the source of (relative) popularity lies in specific features of Britain's implementation of particular taxes or in the broader characteristics shared by these taxes across countries. However, with the possible exceptions of property taxes (Council Tax and Stamp Duty), most UK taxes are not particularly unusual in comparative perspective. Moreover, while our experiment makes this limitation obvious, it is not unique to our design. In broader cross-national studies, or more general question wordings, we also do not know if respondents are reacting to their experience of country-specific particularities.

The obvious extension to fill these gaps is to field appropriately domesticated equivalent surveys in other countries, yielding cross-national evidence on preferences over concrete policies. Researchers could then consider which underlying theoretical characteristics (progressivity, visibility) are associated with support for different tax mixes as a useful complement to asking respondents their views on these characteristics directly.

A more consequential limitation of the generality of our methodology is that the design is difficult to extend beyond existing taxes. This precludes the examination of, for example, a well-designed wealth tax or a flat tax on income. However, there are offsetting gains in terms of the practicability of the proposed reforms (and thus the policy utility of our results), as well as the relative familiarity and credibility of the proposals to respondents.

Conclusion

We use experimental control to identify preferences over specific tax parameters in isolation from accompanying revenue changes, which would otherwise make the measurement of preferences about tax composition difficult. We rely on the respondents' ability to make comparisons between concrete proposals – such that they need not articulate a full preference ordering, nor the details of what they like or dislike about specific taxes – which is a more feasible task in a highly technical area. The revenue-equivalent changes bring the policy choice much closer to politicians' (or Treasury civil servants') tax policy problem.

We thereby identify the levers that might be involved in politically viable tax reform in the UK, minimizing public dissatisfaction with taxation for a given revenue level, and show that the existing composition of UK taxation is far from optimizing the revenue-discontent trade off. Specifically, increasing taxes on corporations, higher-income taxpayers, capital gains, and alcohol and tobacco are likely to be less politically painful than other increases. To the extent that tax cuts can be found, they will be most popular if broadly distributed and targeted to the lower end of the income tax. Equally, two of the UK taxes widely regarded as dysfunctional by policy experts and economists, Council Tax and National Insurance, are also disliked by the general population. Communicated with appropriate reference to the real revenue trade-offs, their reform should be politically feasible. Given the partisan (and socio-demographic) consensus over the tax mix, these aggregate patterns do not mask major electoral cleavages blocking this kind of reform.

Our approach in this paper uses the actually-existing tax system as its starting point, asking questions (only) about concrete potential modifications. A far more challenging problem would be to attempt to characterize public attitudes away from the current margin. The concrete details required also make any implementation of measuring such tax mix preferences parochial: our measurement tool could be ‘domesticated’ to other tax systems, but we would only be able to learn that VAT in Germany is more (or less) relatively popular compared to the actually-existing German income tax system, not about whether Germans or Brits are more predisposed to favour sales taxes in the abstract. Nevertheless, replicating the comprehensive approach to attitudes to a broad universe of tax levers in different countries would vastly increase our understanding of attitudes towards taxation by taking preferences over tax composition seriously.