1 No taxation without representation?

Bargaining over taxation between rulers and taxpayers lies at the heart of theories of consent and representation in early modern Europe (Bates and Lien, Reference Bates and Lien1985; Levi, Reference Levi1989; North and Weingast, Reference North and Weingast1989) and democratization transitions at the turn of the 19th century (Boix, Reference Boix2003; Acemoglu and Robinson, Reference Acemoglu and Robinson2006). Despite the prominence of the “fiscal contract hypothesis” in modern-day political science, empirical evidence of the taxation–representation connection remains largely mixed. Although some studies show a positive relationship between taxation and representation in cross-country analysis (Prichard et al., Reference Prichard, Salardi and Segal2014; Wiens et al., Reference Wiens, Poast and Clark2014), others show that results are sensitive to measurement and model specifications (Ross, Reference Ross2004; Prichard, Reference Prichard2015; Garcia and Haldenwang, Reference Garcia and Haldenwang2016).Footnote ¹

Attempts to overcome the empirical conundrum with individual-level experimental data have led to similarly inconclusive findings. Some researchers find supportive evidence that taxation induces political participation and pressure for political accountability (Paler, Reference Paler2013; Weigel, Reference Weigel2020), but others show that taxation does not lead to greater demand for accountability relative to non-tax revenue (de la Cuesta et al., Reference de la Cuesta, Milner, Nielson and Knack2019). The absence of consistent evidence of the taxation–representation connection in contemporary settings is concerning because the fiscal contract hypothesis is central to theories of democratization and (re)distributive politics.

Understanding the taxation–representation connection requires analytical clarity about key aspects of the fiscal contract negotiation. Who are the relevant players; what are their preferences, bargaining power, and enforcement mechanisms; and in which venue does bargaining take place? We take a first stab at this endeavor by focusing on nondemocratic contexts and mapping preferences over hypothetical tax reforms that may or may not include advances in political representation and by taking into account the elite status (or lack thereof) of the individual.

We contend that in nondemocratic regimes business elites possess stronger preferences than nonelites for tax policies that offer greater political representation. Our argument is built on seminal theoretical contributions in the fiscal contract literature. Ansell and Samuels (Reference Ansell and Samuels2010) and Bates and Lien (Reference Bates and Lien1985), for instance, maintain that merchants (in early modern Europe) and industrialists (in the Industrial Revolution)—not ordinary people—were at the forefront of demands for representation in return for taxation. Today, business elites in industrializing autocracies are the closest equivalent to merchants and industrialists in the past. First, business owners are disproportionally exposed to higher tax burdens, even expropriation because state predation targets large income streams rather than atomized individuals (Gehlbach, Reference Gehlbach2008). Second, business elites have a good understanding of the tax burden and tax incidence, activating the “ownership effect of taxation” that connects taxation with political accountability (de la Cuesta et al., Reference de la Cuesta, Martin, Milner and Nielson2022). Third, business elites have a comparative advantage in using various resources to advance their interest if channels of political influence are opened by the state (Grossman and Helpman, Reference Grossman and Helpman2001).

Taken together, these three reasons led us to the expectation that the taxation–representation connection is stronger among business elites than nonelites and particularly so in autocratic settings, where incumbents face substantial credibility issues respecting private wealth. We test the hypothesis in modern-day China, a paramount autocratic regime that has experienced rapid economic growth over the last four decades. Our design entails two major departures from previous studies. The first is the sampling choice: We recruited separate business elite and nonelite samples. The elite sample draws from a dedicated panel of business owners and managers; and the nonelite sample, from a standard panel of ordinary citizens. Second, we solicit preferences over political influence by running a conjoint experiment in an autocratic setting, where government credibility issues are severe. With this design we can retrieve a preference order over various aspects of a hypothetical tax reform by elite status in a politically sensitive environment. In particular, the conjoint experiment enables us to estimate whether business elites attach greater weight to political influence than nonelites in a hypothetical world in which everyone is allowed to choose among predefined tax policy menus.Footnote ² The results generate important implications about which players would be more likely to demand gains in political influence associated with tax reform if that opportunity arose.

Among nondemocratic regimes, China is perhaps one of the least likely cases in which business elites embody strong preference for political representation. Scholars have argued that business elites in China are allies of the regime and exploit the existing political system to their own advantage (Chen and Dickson, Reference Chen and Dickson2010; Truex, Reference Truex2014; Hou, Reference Hou2019). If these business elites are satisfied with the existing institutional configuration or fear expressing their political preferences, we should observe null results in the conjoint experiment and auxiliary questions.

Nonetheless, our conjoint experiments reveal compelling evidence that the taxation–representation connection is actually stronger for business elites than nonelites. We evaluate the representation aspect by examining preferences for four political outcomes—electoral accountability, policy responsiveness, fiscal transparency, and guarantees of property rights protection—which we collectively refer to as Institutional Political Influence (IPI). In contrast to business elites, we find that nonelites have stronger preferences for public goods and services than IPI. The elite–nonelite differences that we observe in the data are robust to a number of considerations, including preference falsification, sensitivity to sample selection, and satisfaction with existing public goods provision.

To shed light on the business elites’ preferences and their differences from those of nonelites, we explore a number of mechanisms and find suggestive evidence for three of them. First, business elites who do not trust the government express stronger attachment to IPI gains at time of tax reform, consistent with the credibility issues of rulers in Levi (Reference Levi1989) and North and Weingast (Reference North and Weingast1989). Second, elite–nonelite differences in preference for IPI are largest for tax reform involving income taxation. Whereas all our business elites qualify for paying individual income taxes, no respondent in the nonelite sample does because of their low income. The weight of difference in IPI between the groups, which also differ in their exposure to income taxation, is consistent with the so-called “ownership effect” of public funds that others have found to activate demands of political accountability in sub-Saharan Africa (de la Cuesta et al., Reference de la Cuesta, Martin, Milner and Nielson2022). Third, our analysis shows that the elite–nonelite gap in preference for IPI narrows as the tax rate increases, namely when taxation becomes a salient issue. Altogether, our findings suggest that credibility issues, tax ownership, and tax incidence are important factors explaining preference formation for IPI in modern-day autocracies.

The incentives of rulers to selectively co-opt individuals and organizations to acquire private information, appease demands for change, and consolidate power has been examined in the literature on autocratic politics (Gandhi, Reference Gandhi2008; Svolik, Reference Svolik2012). Our research, however, shifts attention to the demand side of that equation by studying societal preferences for representation in the context of a hypothetical tax reform. Our findings resonate with recent work showing the critical role of business elites in shaping (and gaming) political institutions to their own advantage and against that of competing elites and the populace (Grzymala-Busse, Reference Grzymala-B2002; Haggard and Kaufman, Reference Haggard and Kaufman2006; Gehlbach and Keefer, Reference Gehlbach and Keefer2012; Albertus and Menaldo, Reference Albertus and Menaldo2018).

The rest of the paper is organized as follows. In Section 2 we delineate the scope conditions of the taxation–representation connection based on our reading of canonical work on this topic. In Section 3 we outline the empirical strategy and report the main empirical results. Section 4 contains an evaluation of a number of potential mechanisms connecting taxation and representation for business elites and—to a lesser extent—nonelites. We conclude in Section 5 by discussing practical implications of our analysis for the study of political effects of taxation in already democratic settings.

2 Taxation, elite status, and political influence

Economic elites play a central role in theories of both limited government and democratization. The former emphasizes credibility issues in fiscal policy faced by monarchs when executive constraints were absent in early modern Europe (Bates and Lien, Reference Bates and Lien1985; Levi, Reference Levi1989; North and Weingast, Reference North and Weingast1989). Monarchs overcame credibility issues by sharing power over fiscal policy with powerful merchants, leading to the rise of representation and consent—the original fiscal contract hypothesis. Theories of democratization for the Industrial Revolution emphasize a different but related credible commitment: Traditional land-based elites in power could not credibly refrain from expropriating the wealth of the new industrial elites nor redistributing the tax burden toward the modern industrial sector (Ansell and Samuels, Reference Ansell and Samuels2015; Mares and Queralt, Reference Mares and Queralt2015, Reference Mares and Queralt2020). To secure an even distribution of tax incidence across sectors and protect the property rights of industrial investors, the new industrial elites demanded franchise extensions and further political representation (Lizzeri and Persico, Reference Lizzeri and Persico2004; Llavador and Oxoby, Reference Llavador and Oxoby2005).

We contend that economic elites in modern-day autocracies resemble those in the past once scope conditions are adjusted for present times. To start, owners and managers of middle- and large-sized firms in modern-day industrializing autocracies resemble merchant elites in theories of limited government and democratization. These individuals accumulate considerable wealth but differ from highly connected plutocrats because they are not at the pinnacle of the economic and political system. Lacking institutionalized means to protect their wealth, business elites are disproportionally exposed to the threat of state predation in the form of confiscation and disproportional tax incidence (Dickson, Reference Dickson2008; Gans-Morse, Reference Gans-Morse2012; Markus, Reference Markus2012).

Business elites do not necessarily oppose taxation. They benefit from public spending that strengthens public infrastructure and improves the quality of the labor force (Beramendi et al., Reference Beramendi, Dincecco and Rogers2019; Hollenbach, Reference Hollenbach2019). Nonetheless, the new elites have major concerns about whether the government will spend their tax money wisely or refrain from imposing the incidence of taxation onto them—a covert form of property rights erosion. In order to mitigate predatory behavior by the state, business elites may prefer tax reforms that include some gains in Institutionalized Political Influence (IPI), such as fiscal transparency and electoral accountability. Those gains can be mutually beneficial for autocrats and business elites: Autocrats often agree to some power-sharing institutions to accommodate sectoral interests and preempt political contestation (Gandhi, Reference Gandhi2008; Svolik, Reference Svolik2012), whereas business elites seek institutionalized political influence to advance their preference in policymaking, establish political connections to grow profit, and reduce risks of expropriation (Fisman, Reference Fisman2001; Truex, Reference Truex2014; Earle and Gehlbach, Reference Earle and Gehlbach2015; Kung and Ma, Reference Kung and Ma2018; Hou, Reference Hou2019). As a corollary, that gains in IPI do not necessarily equate to democratization should now be evident.Footnote ³

In the event of a fiscal reform, we expect business elites to hold stronger preference for IPI gains than nonelites for three reasons: First, state predation tends to prioritize high-yield economic assets owned or managed by business elites. Second, business elites have better information about the tax burden because they have to deal with tax codes on a regular basis to make profit-maximizing decisions, avoid taxation, or both. Awareness of the tax burden also makes business elites more likely to experience the “ownership effect,” a psychological mechanism that makes people perceive government monies as their own, spurring preferences for political influence (Prichard, Reference Prichard2015; Weigel, Reference Weigel2020; de la Cuesta et al., Reference de la Cuesta, Martin, Milner and Nielson2022). Third, business elites tend to be more politically active in advancing their interests. In democratic contexts business elites make their voices heard,Footnote ⁴ raise and donate more money,Footnote ⁵ vote more often,Footnote ⁶ and run for office more often than ordinary citizens.Footnote ⁷ We contend that, relative to ordinary citizens, business elites have a comparative advantage in exploiting opportunities created by IPI to advance their interests. Accordingly, we expect business elites to hold particularly high valuation of IPI gains in the autocratic context.

Conversely, we expect nonelites to attach less weight to IPI gains in the context of tax reform. First, their modest incomes make them less vulnerable to state predation, and even then ordinary citizens often demand tax breaks, not necessarily gains in IPI (Tilly, Reference Tilly1993; Bianco, Reference Bianco2001; Bernstein and Lü, Reference Bernstein and Lü2003). Second, nonelites tend to underestimate the tax burden, particularly for low-salience taxes, such as tariffs and consumption taxes.Footnote ⁸ If taxpayers underestimate the tax burden, expecting tax reform to stimulate strong preference for political influence is unrealistic. Third, even when ordinary citizens are informed about the tax burden, they may lack key resources to advance their preferences into the policy-making arena: capital, time, and expertise (Schlozman et al., Reference Schlozman, Brady and Verba2018).

Based on the three reasons, we expect ordinary citizens’ preferences for tax policy to follow a simpler “tax-for-services” (Ross, Reference Ross2004) or “exchange of services for support” logic (Lake, Reference Lake2016, p.17) when they think about changes in the tax code. Consistently, Beramendi and Rueda (Reference Beramendi and Rueda2007) and Timmons (Reference Timmons2005) show that ordinary citizens in wealthy and developing democracies pay taxes in expectation of public-funded goods and services, not political rights. Moreover, studies of patronage and clientelism extensively document citizens’ willingness to relinquish their “paper stones” (Przeworski and Sprague, Reference Przeworski and Sprague1986) in exchange for public or private goods (Golden and Min, Reference Golden and Min2013). The tax-for-services calculus is also present in nondemocratic regimes. Public spending is shown to be a popular tool to garner political support in authoritarian regimes (Svolik, Reference Svolik2012; Albertus et al., Reference Albertus, Fenner and Slater2018). Certainly, an autocrat's promises to offer government services to ordinary citizens are not exempted from credibility issues. These likely exist, and strategic autocrats will do their best to keep citizens’ expectations to a bearable minimum (Gottlieb, Reference Gottlieb2016).

We can summarize the preceding discussion in a single hypothesis:

3 Empirical design

To test our hypothesis, we examine the taxation–representation connection in China with separate samples of business elites and nonelites. In the next sections, we elaborate our case selection, experimental design, and sample recruitment strategies.

3.2 Experimental design

To evaluate the connection between political influence and tax policy, we implement a conjoint experiment. This is a survey method in that respondents compare and choose between pairs of hypothetical tax policies that include randomized combinations of IPI, government services, tax types, and tax rates.

Motivation

We prefer a conjoint experimental design over a typical survey experiment design with question prompts for two reasons: relative preferences and social desirability bias. First, respondents may have a wide range of preferences in return for taxation, and IPI is a multifaceted phenomenon. The conjoint experiment allows us to evaluate the relative weight attached by elites and nonelites to different representation aspects attached to a hypothetical tax policy reform: elections, policy responsiveness, transparency, and property rights protection.

Second, a direct experimental manipulation involving how taxation may induce demand for representation suffers a severe social desirability bias because of the current political environment in China. A growing number of studies show that conjoint experiments potentially mitigate social desirability issues because sensitive items are made part of a choice bundle (Hainmueller et al., Reference Hainmueller, Hopkins and Yamamoto2014; Horiuchi et al., Reference Horiuchi, Markovich and Yamamoto2020). This feature is particularly important to solicit preferences for IPI in China, where we anticipate social desirability bias if preferences over IPI were directly requested. Although conjoint experiments cannot completely eliminate the social desirability bias, we also chose this design because it minimizes any negative repercussion aimed at our respondents under an unfavorable political climate.

Conjoint Experiment Attributes

Each tax reform comprises four dimensions or attributes, the values of which are randomly drawn from a list of plausible values reported in Figure 1(a). The values of the IPI dimension are based on several considerations, key among them our determination to resort to familiar and understandable concepts to our respondents. We carefully choose values that match the conceptualization of political participation, accountability, and representation commonly used in the existing literature while maintaining consistency with local context and language.Footnote ¹¹ For example, the submission of citizens’ opinions online or via telephone as well as through public hearing is a typical form of political participation in China (Shi, Reference Shi2015; Chen et al., Reference Chen, Pan and Xu2016; Distelhorst and Hou, Reference Distelhorst and Hou2017). Meanwhile, fiscal transparency is often part of the open information campaign in many countries that aims at enhancing government accountability.

Fig. 1. Attributes and values in the conjoint experiment: Panel (a) shows all attributes and values, and Panel (b) shows a randomly generated paired comparison.

One may argue that the first two IPI values are not necessarily strong enough to ensure representation and accountability because the government may choose not to respond to citizens’ demands. We thus include a third value: citizens’ ability to elect the district government executive, which is expected to enable responsiveness, representation, and accountability. Last but not least, under IPI we list property rights protection, the primary concern among elites in the standard taxation–representation models (Bates and Lien, Reference Bates and Lien1985; Ansell and Samuels, Reference Ansell and Samuels2015). Although one could treat property rights protection as a government service, we group it with the rest of IPI values to emphasize the limits on state predation imposed by private property safeguards.

Because our theoretical framework highlights the importance of tax-financed government services to ordinary citizens, we consider four types of public good in our conjoint experiment: education, health care, and pensions; national defense; environmental policy; and infrastructure. These are all tax-funded, common-interest public goods and services in China.

The third and fourth attributes account for the implementation details of a hypothetical tax reform. By randomizing the tax type and tax rate we intend to factor out any preconceived notion of what a tax reform is and how tax progressivity might influence the taxation–representation connection, an important consideration that we seek to analyze in future work. We consider two hypothetical tax types—the income tax and VAT—with five possible rates, starting at a negligible 1 percent and up to 20 percent. We purposively disregard a 0 percent rate to avoid impossible combinations with other values that imply government spending.

Figure 1(b) shows an example of a paired comparison in our conjoint experiment. Every respondent is asked to complete six rounds of paired comparisons. In each round respondents are assigned a different pair of randomly generated tax reform proposals and are requested to choose which is most preferred.Footnote ¹² We then estimate the average marginal component-specific effect (AMCE), namely the unconditional marginal effect of an attribute value averaged over all possible values of the other attributes.Footnote ¹³ Altogether, the conjoint coefficients offer a stylized map of the weights that respondents attach to different aspects of a hypothetical tax reform.

4 Main findings

In this section we examine the conjoint estimates for both samples and assess whether they are different at conventional levels of statistical significance. We then complement the main analysis with a crude yet informative question in which respondents chose between two conceptualizations of the tax contract: tax for services versus tax for political influence.

4.1 Conjoint results

Figure 2 reports the baseline conjoint coefficients for the screened elite and nonelite samples. Conveniently, these estimates are directly interpretable in percentage points and comparable within and across attributes. Consistent with our expectation, business elites show stronger preferences for every value of IPI than nonelites. Any given tax reform proposal accompanied by new opportunities for citizens to elect the district government executive (i.e., Election) increases support for that tax reform among business elites by almost 10 percentage points; by contrast, the point estimate of Election for nonelites is not statistically different from zero. In addition, a tax reform proposal that entails Fiscal Transparency increases support by 18.2 percentage points for business elites, but only 6.9 percentage points for nonelites. Finally, Citizen Input and Property Rights increase business elites’ support by 15.5 percentage points and 13.7 percentage points, respectively. For nonelites, the increase is only 6.9 and 7.2 percentage points, respectively.

Fig. 2. China conjoint experiment by elite status. This plot shows estimates of the effects of randomly assigned attributes for tax reform dimensions on the probability of supporting a tax reform policy. Estimates are drawn from the screened samples. The model in regression format (also including socioeconomic controls) can be found in Appendix E. The bars indicate 95 percent confidence intervals.

Meanwhile, both elites and nonelites reveal similar preference for government services. Education, Health Care, and Pensions receive the highest score of all in both groups, followed by Infrastructure, Environmental Policy, and National Defense. Substantively, a tax reform proposal that increases spending on Education, Health Care, and Pensions would increase support by 28.9 and 28.7 percentage points for elites and nonelites, respectively. This result is consistent with the tax-for-services rationale for tax compliance and seemingly applies to elites and nonelites.

Crucially, results reported in Figure 2 do not qualitatively change when we work with the original sample (i.e., a more liberal definition of elite) or when we focus only on owners and managers of top 5 percent and top 1 percent firms based on employment size (i.e., a more conservative definition). In Appendix F we also confirm that individuals in the original nonelite sample who earn high income show slightly smaller estimates for IPI attributes compared to our elites. Altogether, these robustness tests show that the more individuals approach our definition of a business elite (i.e., managerial position plus high income), the more their preferences meet our expectations for that group; and the more individuals approach our definition of nonelite status (i.e., salaried or inactive individual with low income), the more their preferences meet our expectations for ordinary citizens.

Elite–nonelite difference

Figure 2 suggests that Chinese citizens reveal stronger preference for government services compared to IPI. This set of results is unsurprising in light of the Chinese government's strategies to use public spending to bolster regime support (Lü, Reference Lü2014; Dickson et al., Reference Dickson, Landry, Shen and Yan2016). However, key to our argument is the elite–nonelite difference in the preference for IPI in our hypothetical, multifaceted tax reform. To evaluate this point in further detail, next we estimate between-group differences with the following interaction model:

(1)$$\matrix{\rm{Preferred\, tax\, reform} & = \beta_0 + \beta_{1j} \left(\displaystyle\sum_{\,j = 1}^4 IPI_{\,j}\right) + \beta_{2j} \left(\displaystyle\sum_{\,j = 1}^4 Govt Service_{\,j}\right)\cr & \quad + \beta_{3j} \left(\displaystyle\sum_{\,j = 1}^4 Tax Rate_{\,j}\right) + \beta_{4j} VAT_{\,j} + \beta_{5} Group_{g = elite} \cr & \quad + \beta_{1jg} \left(\displaystyle\sum_{\,j = 1}^4 IPI_{\,j} \times Group_{g = elite}\right)\cr & \quad + \beta_{2jg} \left(\displaystyle\sum_{\,j = 1}^4 Govt Service_{\,j} \times Group_{g = elite}\right)\cr & \quad + \beta_{3jg} \left(\displaystyle\sum_{\,j = 1}^4 Tax Rate_{\,j} \times Group_{g = elite}\right)\cr & \quad + \beta_{4jg} ( VAT_{\,j} \times Group_{g = elite}) + \epsilon_{\,jg} }$$

where i ∈ {1, 2, 3, 4} denotes attributes, j values for each attribute, and group $g \in \{ \rm {business\, elites,\; nonelites}\}$. Figure 3 shows the estimated between-group differences for each value in the conjoint experiment. Results confirm that elites and nonelites value government services to the same extent—differences overlap with the zero line. By contrast, business elites attach stronger preference to IPI than nonelites. In particular, support for a hypothetical tax reform is 10+ points higher for business elites than nonelites if accompanied by advances in citizens’ input, fiscal transparency, and local elections, holding everything else constant. The coefficient for property rights protection, although half the size, remains statistically different from zero with 90 percent confidence (p = 0.078, two-tailed). Taken together, Figures 2 and 3 suggest that both elites and nonelites care about IPI and government services, but elites care disproportionally more about IPI than nonelites.

Fig. 3. Differences in conjoint estimates between business elites and nonelites. This plot shows the differences in AMCE between business elite respondents (Business Elites =1) and nonelite respondents (Business Elites = 0) as defined in Equation (1). Estimates are drawn from the screened samples. The bars indicate 95 percent CI. We report regression results in Appendix E.

Other dimensions of our hypothetical tax reform in the conjoint experiment—the tax rate and type—offer somewhat unexpected results. First, nonelites dislike higher tax rates more intensely than business elites perhaps because the latter are less averse to higher taxation if they anticipate a similar increase in government services than nonelites plus advances in IPI. In other words, for any specific tax rate business elites may anticipate a larger “bang for the buck” than ordinary citizens. Second, nonelites do not penalize policy reform involving consumption taxes or VAT, often considered regressive. The mechanism section elaborates on the role of tax types and tax rates in the taxation–representation connection.

4.2 Direct question

The conjoint experiment elicits respondents’ preferences for IPI through a bundle of policy attributes. This technique is particularly advantageous in dealing with sensitive topics, but we also made respondents face a tradeoff between two ways of conceptualizing the connection between taxes and policy-making. The first option emphasizes gains in government services following taxation, explicitly excluding advancement in institutionalized political influence, whereas the second option explicitly connects political influence in policymaking with tax increases. Although institutionalized political influence may be a means toward better government services, our wording is meant to retrieve what respondents prioritize if faced with a stark choice. The key contrast between these two choices is the preference for representation in policy-making or lack thereof. The exact wording of the question in English reads as follows:

1. (a) As long as government spends my tax money on high-quality and generous public goods, I do not care about participating in policy-making.

2. (b) In return for paying taxes, I would like to have some say in policy-making.

In Figure 4 we plot the proportion of respondents choosing option (b) in each group. This figure shows that business elites would be more inclined to prioritize gains in political representation than nonelites at the time of tax reform. The elite–nonelite difference spans 17 points and is statistically significant at 95 percent. Combined, Figures 3 and 4 indicate that elites and nonelites in China differ in their preference over IPI gains in a hypothetical tax policy proposal. In Appendix G.4 we examine whether differences between groups may be driven by the preference falsification of nonelites, finding no supportive evidence.

Fig. 4. Direct request about preferences for IPI and government services by elite status in China. The vertical axis indicates the proportion of respondents who prioritize IPI over government services when they are directly requested. Estimates are drawn from the screened samples. The bars indicate 95 percent CI.

5 Mechanisms

This section addresses two related questions: First, why do business elites value IPI? We argue that the autocrat's credibility problem in refraining from expropriating private assets is particularly salient for business elites because of their accumulated wealth. In addition to motive, business elites have better information and means (i.e., comparative advantage) to pursuetheir interest if they are granted IPI, hence their stronger valuation of the latter.

Second, why do preferences for IPI differ between business elites and ordinary citizens in China? We consider a series of nonmutually exclusive mechanisms: (i) tax ownership, (ii) tax salience, (iii) tax awareness, and (iv) use of public services. We find supporting evidence of the first two mechanisms in our data and discuss mixed evidence on other mechanisms, particularly the one we expected to confirm before analyzing the data: VAT awareness.

5.1 Credible commitment

We expect business elites to have grave concerns about state predation, particularly those who do not find government's promises credible. Measuring the latter perception is challenging. Here we follow the lead of Levi (Reference Levi, Braithwaite and Levi1998, p. 85–6), who offers an extensive discussion on the relationship between credible commitment and trust in government:

Despite that trust and commitment are different mechanisms for resolving uncertainty, commitment is one of the means to create trust. [Levi adds,] credible commitments […] reduce the citizen's need to make a personal investment in monitoring and enforcing government and thus enhance citizen trust of government.

Building on this reasoning, we split the respondents in the business sample into two groups—those who trust the government and those who do notFootnote ¹⁸—and then investigate differences in their preferences for government services and IPI using the survey instrument employed in Figure 4. Results, reported in Figure 5, suggest that preference for political say over government services in exchange for taxation is 18 points higher (and statistically significant at 95%) among those who do not trust the government, consistent with the idea that IPI alleviates credibility issues of fiscal policy within the group that is most exposed to state predation.

Fig. 5. Trust in government and preference for IPI among business elites. This figure reports the proportion of business elites trusting and not trusting the government and who prioritize IPI over government services. Estimates drawn from the screened samples: N = 64 for Elite + No Trust, and N = 208 for Elite + Trust. The bars indicate 95 percent CI.

The credibility of government promises may correlate with the ability to escape taxation. Next we consider two factors that alleviate exposure to state predation—asset mobility and state ownership—and examine whether they help explain within-elite variation in IPI preference.

5.1.1 Foreign versus domestic firms

Asset mobility enhances the bargaining power of taxpayers because they can credibly threaten to flee to other jurisdictions if the ruler deviates from the agreed upon political compromise (Bates and Lien, Reference Bates and Lien1985; Boix, Reference Boix2003). If this rationale applies to China, business elites with mobile assets may attach higher valuation to gains in IPI than business elites without mobile capital, everything else constant.

China's Central Bank imposes tight capital control, hence business elites cannot easily move their assets out of the country at will. In the Chinese context, firms could alleviate state predation through joint ventures with foreign firms. Wang (Reference Wang2015) argues that foreign corporations in China have stronger property rights protection because they can credibly reallocate to other jurisdictions. Wang shows that joint ventures between local and foreign corporations extend property rights protection to the domestic partners of the foreign firms.

In the spirit of Wang's (Reference Wang2015) argument, we compare conjoint results of business elite respondents working at domestic private firms to those working at foreign firms based in China. Unfortunately, the size of latter group is small (N=83), so results must be interpreted with caution. In our test (reported in Appendix G.3) the only statistically significant difference between groups is for holding local elections, possibly the strongest expression of IPI. This result is consistent with the logic of exit threats enforcing fiscal contracts. That is, mobile tax payers may attach higher valuation to elections because their mobility confers upon them some ability to make autocrats abide by the results of a hypothetical electoral contest.Footnote ¹⁹

5.1.2 Private versus state ownership

SOE managers and party members may be considered agents of the state and be already sheltered from predation (Hou, Reference Hou2019). If this were the case, the inclusion of CCP members or SOE managers in the elite sample would attenuate differences between elites and nonelites. In our elite sample, 87 (31.98 percent) and 51 (18.75 percent) respondents are members of the CCP or working in an SOE, respectively. Appendixes G.1 and G.2 show no substantive change in the weight of IPI values when we exclude all CCP and SOE respondents from the sample. Those findings reassure us of the nature of our sample: Although the SSI business panelists are winners of economic modernization, they are still exposed to state predation and hence may value advances in IPI (if only for reasons of self-interest).

5.2 Ownership effect

The connection between taxation and IPI might be a function of psychological ownership over public moneys. De la Cuesta et al. (2022) show that citizens who feel that the budget belongs to them are more willing to demand more accountability from leaders. In a related work, de la Cuesta et al. (Reference de la Cuesta, Martin, Milner and Nelson2020) find that the ownership effect on demands for political accountability is stronger for direct taxes because of their high salience. In our sample, none of the nonelites qualify for the income tax because their low income exempts them. By contrast, all business elites in our sample qualify for income taxation.Footnote ²⁰ The disproportional incidence of the income tax (a high salient tax) on business elites might accentuate their tax ownership effect, strengthening preferences for political influence associated with tax reform.

We examine the ownership effect indirectly by studying separately responses to tax reforms involving an income tax and the VAT. Specifically, Figures 6a and 6b report AMCE once we fix the tax type in our hypothetical tax reform to income tax and VAT, respectively. Whereas both sets of tax reforms induce preference for government services, results for IPI between groups are starkly different. Business elites attach positive weights to both IPI and government services regardless of the nature of the tax reform. Nonelites, by contrast, show positive weights for IPI for reforms involving the VAT but zero weight for reforms involving an income tax. This difference between elites and nonelites is consistent with the ownership effect: Because everybody pays VAT, changes in this tax are followed by stronger preference for both IPI and goods and services among elites and nonelites. By contrast, the relatively little experience of nonelites with income taxation made this group less inclined to prefer gains in political influence in the hypothetical tax reform, everything else constant.

Fig. 6. Conjoint estimates of business elites and ordinary citizens by tax reforms involving an income tax (Panel a) and a VAT (Panel b). ACME estimates are drawn from the screened samples. The bars indicate 95 percent CI. All between-group differences in IPI values in the left panel are statistically different from zero at 95 percent.

5.3 Tax salience

The income tax is more salient than consumption taxes, but only high-income earners effectively pay income tax in China. Another way to examine the role of tax salience in shaping preferences for IPI is by separating the analysis for different tax rates. Arguably, when rates increase substantially, ordinary citizens might pay more attention to tax reform, potentially activating the taxation–representation connection (Prichard, Reference Prichard2015). We evaluate responses to different levels of tax rate in our conjoint estimation by dividing tax reforms into two groups: Those including a tax rate of 15 or 20 percentage points are listed under the high tax rate group; those including a tax rate of 1, 5, and 10 percent are listed under the low tax rate group.Footnote ²¹

We report the results of this exercise in Figures 7(a) and 7(b). An interesting pattern emerges: The elites–nonelites differences in preferences for IPI are more pronounced for lower tax rates. When the hypothetical tax rate is mild, business elites show stronger preference for Fiscal Transparency and Elections than ordinary citizens. The elite–nonelite differences grow smaller and lose statistical significance in Figure 7(b) when we focus on policy bundles that include high tax rates only. These findings suggest that ordinary citizens may become more scrupulous in evaluating a tax reform when the proposed tax rates grow salient, everything else constant.

Fig. 7. Differences in conjoint estimates between business elites and ordinary citizens for tax reforms with low and high tax rates, Panels (a) and (b), respectively. In the low tax rate group, we pool all tax reforms that include 1, 5, and 10 percent tax rates, and in the high tax rate group we pool all tax reforms that include 15 and 20 percent tax rates. Estimates are drawn from the screened sample. 95 percent CI reported.