1.1. Coalition and policy compromise at the turn of the century

Applied to the adoption of progressive taxation between 1870 and 1913, simplifying political battles over redistribution to this two-dimensional model works as follows.

Elites in all countries prefer low levels of distribution but may have varying positions on trade. Some may favor free trade, others protection, and elites within countries may disagree. The degree of variation in elite trade preferences differs between countries.

Labor groups prefer high levels of redistribution and are in direct conflict with elites on this issue. Where elites agree on trade, labor is politically isolated. Where there is elite disagreement on trade, though, labor may be able to provide support for one side of that dispute. An elite group who shares trade policy preferences with labor may prefer to incorporate labor into a coalition supporting their trade preference than to remain “allied” to other elites, compromising on trade to block redistribution.

Figure 1 provides a spatial illustration.Footnote ³ The figure illustrates that multidimensionality matters, comparing the political space when a pro-redistribution group shares trade interests with a high-income group to that where it does not.

Figure 1. Economic groups in the two-dimensional policy space, with (panel b) and without (panel a) elite division on trade policy. e1 and e2 indicate elite groups (political insiders). By construction they have the same preference for low levels of progressivity. r indicates the pro-redistribution group. • indicates elite-compromise policy outcome; ° indicates outcome of e2−r compromise, assuming bargains negotiated with equal bargaining power. While e1−e2 “compromise” is clearly closer to e2's position where elites agree on trade, with elite division (in panel b) a compromise on progressivity with r is equally close to e2's position as a compromise on trade with e1.

In the left panel, advocates of redistribution (r) are isolated on both tariff and progressivity preferences. Political conflict is unidimensional along the diagonal line. On the right, one of the elite groups (e2) has trade interests similar to r. Although they still differ in their preferred level of redistribution, this creates the possibility of an e2−r coalition pitting free traders against the high-tariff elite (e1). With two dimensions, e2 may choose to prioritize trade over redistributive goals.

Labor will not necessarily win concessions on redistribution. e2 may still prefer the e2−e1 compromise. The model simplifies the bargaining environment considerably and the story is a complicated one overall.Footnote ⁴ But panel (b) allows for the possibility of redistributive reform to a much greater degree than (a), even though all groups' positions on redistribution are unchanged.

1.2. When does trade disagreement matter?

We can also make predictions about when and where redistributive compromise is more likely. Maintaining the spatial intuition, polarization (a greater distance between positions) on either dimension makes compromise on that dimension less appealing.

For the redistributive dimension, this means that inequality matters. Higher inequality gives low-income voters more to gain from redistributive policies, but high-income groups more to lose. In contrast, if trade differences between elite groups are relatively more pronounced, the axis of political competition—and the likelihood of progressive outcomes—shifts in the opposite direction. If elite trade interests are more polarized, more is lost in any trade compromise required to limit progressivity.

Again, this is most easily illustrated graphically. Higher inequality makes the redistributive demands of the poor more extreme (moving them away from the elite groups in the horizontal space, in Figure 2(a)). Where in Figure 1(b) they were equidistant, the elite compromise outcome is now closer to e2's position than compromise with r.

Figure 2. Groups of factor owners in the two-dimensional preference space, with high inequality (panel a) or large differences in trade preferences (panel b). e1 and e2 indicate elite groups (political insiders), r indicates the pro-redistribution group. • indicates e2−e1 compromise policy outcome; ° indicates indicates e2−r compromise policy outcome, assuming bargains negotiated with equal bargaining power. As inequality increases, r's redistributive demands get more extreme (panel a), and e1−e2 compromise becomes preferable for e2. Panel b shows more extreme protectionism from e1, which has the opposite effect, making compromise with r relatively more beneficial to e2.

Figure 2(b) shows more pronounced divergence of elite trade interests. Now the compromise between e2 and r is closer to e2's position than the elite bargain. A coalition based on redistributive concessions is more probable, and progressive tax policy more likely. Greater inequality between high- and low-income groups, and smaller differences in trade interests within elite groups, will limit the extent of progressive tax policy.

Finally, the effects of trade distance and redistributive distance interact. At higher levels of inequality, any divergence between the trade preferences, the elite groups must be larger to incentivize cross-class compromise. Compare Figures 2(a) and 1(b). Trade differences need to be twice as large in the former for the same incentive for e2 to compromise with r. Not only should trade divisions between elites make progressive reform more likely, but this should be more pronounced where inequality is low.

2.1. Data and measurement

The three core variables of interest in theoretical terms are progressivity, inequality, and labor's potential advantage in forming coalitions on the basis of trade.

2.1.3. Trade preferences and the labor trade advantage

I use the Stolper–Samuelson model of trade preferences, to characterize gains from trade according to ownership of factors of production: land, labor, and capital. Owners of different factors gain (lose) differentially from protection, and the scale of these gains (losses) shapes their support for tariffs. Gains from trade depend on the relative abundance of the factors. Owners of relatively abundant factors benefit from free trade; owners of scarce factors are harmed. Where labor is abundant, and land and capital scarce, we are in the situation of labor isolation as highlighted in Figure 1(a). If land is scarce but both capital and labor abundant, this mirrors Figure 1(b). Relative factor abundance thus drives trade policy preferences. These structural characteristics have the advantage that they are relatively exogenous to trade and tax policy choices.

Why focus on factor ownership? The alternative is that preferences depend on the industry a producer is in. However, mobility of factors equalizes their return across industries, so trade affects all owners of a factor of production similarly, and the period 1970–1914 was one of high inter-industry factor mobility (Hiscox, Reference Hiscox2002).

To measure factor abundance, I use the multi-country extension of Edward Leamer's “world abundance” (Leamer, Reference Leamer1980). Abundance is defined relative to a second factor, in proportion to the over-endowment of the home country compared to the rest of the trading world. That is, the abundance of factor i relative to factor j is given by

$$ a_{ij} = m_i m^w_j - m^w_i m_j,$$

where m _i indicates the endowment of factor i within the country relative to the average global endowment of that factor, $m^w_i$ indicates the endowment of all other countries, similarly scaled by the global mean. Labor's trade-ally advantage (LTA) is the degree to which its trade preferences are more similar to (the closer of) land or capital than capital and land's preferences are similar to each other. Thus the LTA is given by

$$ {\rm LTA} = \vert a_{kl} - a_{al} \vert - \min \big( \vert a_{ka} - a_{la} \vert, \vert a_{ak} - a_{lk} \vert \big),$$

where a _ij is the relative abundance of i relative to j, for a, land; k, capital, and l, labor.

The data needed here are endowments of labor, land, and capital worldwide: relative abundance depends not only on the situation in the western European countries under study, but in all of their trading partners. I prioritize coverage of global trade over sophisticated measures. I use the length of the railway routes open at the end of each year as the absolute measure of capital endowment. This has the disadvantage that railways may have been owned publicly or by foreigners, but the advantage that railways were a relatively universal industry across countries. Moreover, the issue of foreign ownership is common across types of capital, and the railway route measure is widely used in the literature on economic development (Bairoch, Reference Bairoch1982) and trade politics of the period (Rogowski, Reference Rogowski1989).

Labor abundance is measured by population, and land by geographic area in 1913. I do not adjust for land usable for agriculture, nor for the share of the population actually in the workforce. This is not ideal, but coverage of all trading nations is not possible for the more subtle measures. In contrast, the data used here are available for 74 countries worldwide in the Cross-country Adoption of Technology Database Comin and Hobijn (Reference Comin and Hobijn2009).

Despite these caveats, the abundance measures have good face validity, corresponding to the “stylized facts” of the period. The UK (especially), France, and Germany have the highest levels of capital abundance, but this over-endowment decreases over time. Belgium and Italy see declines from lower initial levels, while the Scandinavian countries' relative endowment of capital is low but increasing through the period. All the countries in the sample outside Scandinavia are relatively abundant in labor compared to land.

Finally we must move from the preferences of each group to a single measure of labor's advantage as a trade ally. Labor is advantaged to the degree that its preferences are closer to a potential partner than that partner's alternative option. I take the distance between the two elite groups, land and capital, and subtract the distance between labor and its closest elite group. If labor is closer to one of the two elite groups than they are to one another, this “labor trade advantage” (LTA) measure will be positive; if the smallest distance from labor to another group on trade is greater than the intra-elite distance, the LTA is negative. Bigger dissimilarities are reflected in larger absolute values.

2.2. Model specifications

In principle, taxes are on the agenda annually in budget legislation. The country-year is the unit at which variation is possible, and taken as the unit of analysis. This time-series cross-sectional structure creates particular modeling issues (Beck and Katz, Reference Beck and Katz1995). The supporting material includes results from alternative specifications (including dynamic models considering long and short run effects). Since the main inferences are robust across these specifications, I present the simplest here: ignoring the dynamics but including country and year fixed effects to account for unobserved heterogeneity.

I incorporate various control variables suggested by the literature. Gross domestic product per capita (GDP p.c.) is included to measure the level of development, and create a dummy variable to indicate whether an explicit link between (direct) taxation and voting rights exists (Mares and Queralt, Reference Mares and Queralt2015).Footnote ⁹

There are some tricky questions about the inclusion of three further controls: trade openness, the breadth of the franchise, and the size of government. Trade as a share of the economy is shaped in part by the political processes examined here, so is post-treatment. On the other hand, trade is also affected by exogenous characteristics which vary across time and space, and existing levels affect the revenue and economic effects of tariffs. Openness may be an important omitted variable.

The extent of the franchise and the size of government pose structurally similar questions in terms of omitted variable versus post-treatment bias. Franchise extensions should increase progressivity (Acemoglu et al., Reference Acemoglu, Naidu, Restrepo, Robinson, Atkinson and Bourguignon2015), but the logic of coalition and inclusion implies that democratization might be a consequence of elites seeking to bolster their side of the trade conflict (Ansell and Samuels, Reference Ansell and Samuels2014). Similarly, expanding the public sector could be a concession to secure support from lower income groups, but studies of tax progressivity today argue that the structure of taxation is determined by its overall level (Ganghof, Reference Ganghof2006; Cusack and Beramendi, Reference Cusack and Beramendi2006).

On balance, in the period in question, I think that taxes as a share of GDP are more importantly post-treatment than pre-treatment, while the other two variables are more important as omitted variables. For taxes/GDP, the primary mechanisms for the endogeneity of structure to size work through the existence of high overall levels of taxation which do not apply in the pre-war period. Thus, I prefer specifications which exclude taxes/GDP and avoid the post-treatment bias. Nevertheless, I present results from the minimal and maximal models alongside my preferred specification in Table 1. Finally, the supplementary materials present analyses including a variety of indicators of state capacity. These leave the results substantively unchanged.

Table 1. The determinants of revenue shares in total taxation, 1870–1913

***p < 0.01; **p < 0.05; *p < 0.1.

2.4. Top tax rates

Table 2 shows the results for top tax rates. Data on the top rate of income and inheritance taxation, respectively, come from Scheve and Stasavage (Reference Scheve and Stasavage2016).Footnote ¹⁰

Table 2. The determinants of top direct tax rates, 1870–1913

***p < 0.01; **p < 0.05; *p < 0.1.

Figure 4 shows the marginal effects of labor's trade advantage at different levels of inequality.

Figure 4. The marginal effects of labor's trade advantage on (a) top income tax rates and (b) top inheritance tax rates at varying levels of inequality. Estimates from Table 2.

These results are more mixed. For income tax rates, inequality itself is negatively associated with top tax rates (in models without the interaction, both inequality and trade advantage are statistically significant, and signed as predicted by the theory), but the estimated interaction effect is “wrongly” signed (though not statistically significant). Figure 4(a) clarifies that only at high levels of inequality is a labor trade advantage associated with higher top rates, at odds with both the theory and the previous empirical results.

On the other hand, the results for top inheritance tax rates mirror those for the revenue shares (Figure 4(b)). At low levels of inequality, an improvement in labor's strategic position on trade increases the top inheritance tax rate. At the 25th percentile on inequality, a one-point change in trade advantage is associated with inheritance tax rates 2.65 points higher—which is almost exactly one standard deviation of that outcome.

Overall, the data from ten countries between 1870 and 1913 are broadly consistent with the idea that trade politics in the period shaped the politics of progressive taxation. The one exception is top rates of income taxation, where the critical interaction term is wrongly signed. Explaining this discrepant outcome is beyond the scope of this paper, but highlights the importance of considering outcomes beyond income taxation (Barnes, Reference Barnes2018).

3.1. Case selection

What kind of evidence does the British case provide? The idea is not to make claims of representativeness and generalizability from the case study, but to check that the proposed mechanism can be found somewhere. This drives case selection on the basis that “one should first examine cases where we expect to see [the proposed mechanism] in action” (Goertz, Reference Goertz2017, p. 13).

Britain by 1903 is a most-likely case for the coalition mechanisms for two reasons. Economic inequality was relatively low, so is in the area of the “inequality space” where trade preferences should matter. By 1903, only Norway is measured as more equal using the rural inequality measure from the comparative analyses. Second, the issue of trade politics became highly divisive among elite politicians once Joseph Chamberlain launched his campaign for Tariff Reform.

Useful variation in local interests and behavior also recommends the British case. Political constituencies were deliberately varied in their economic and social characteristics, thanks to the 1885 Redistribution Act (Pelling, Reference Pelling1967, p. 3), creating identifying variation on trade interests. The decentralization of the Edwardian Liberal party, with national candidates selected by constituency associations, makes variation in local preferences relevant in national votes. Finally, there is recorded variation in two kinds of outcome: the ease of labor's run in the 1906 election, and voting on a progressive piece of tax legislation.

3.2. Trade, alliances, and redistribution in Edwardian Britain

The national-level evolution of trade, alliances, and progressive tax policy in Britain provides prima facie evidence of the mechanisms in question, and gives some context for the sub-national comparisons which follow. It is important to note that despite Britain's free trade history, a return to protection was a real possibility in the 1900s. Free trade was preserved, but not without “immanent threat” at the turn of the century (Howe, Reference Howe1997, p. 231).

Britain entered the 20th century with a Conservative–Unionist government with a large majority. But the government was fragmented, incorporating all factions of British politics opposed to Irish Home Rule. In 1903, Joseph Chamberlain, Secretary of State for the Colonies, introduced a proposal for Tariff Reform. Chamberlain advocated tariffs on imported goods to protect industry, raise revenue, and allow preferential terms for British colonies. The campaign re-mobilized the trade cleavage.

The possibility of protection was real. Britain is often seen as a bastion of free trade, but was not immune to the return of protectionist sentiment across Europe during the agricultural depression, nor in the renewed conflict after 1903. “Chamberlain's dramatic political initiative [Tariff Reform]… put Free Trade firmly ‘in the dock’ ” (Gomes, Reference Gomes1990, p. 74), and propelled tariffs to the top of the agenda. Disagreement within the cabinet in 1903 raised the possibility of an election, inspiring panic among the free traders lest protection be endorsed at the polls. The threat was in fact averted, but was real. Compromise on progressivity may not have been needed to secure free trade in the 1900s, but in the face of these new pressures, the need for concessions to maintain it is highly plausible.

Liberal and Labour leaders agreed the Gladstone–Macdonald electoral pact in 1903 to ensure that Liberal and Labour Representation Committee (LRC) candidates would not split the anti-Conservative vote. When the election came in 1906, this meant the deliberate withdrawal of Liberals in certain constituencies, to give Labour a free run. The LRC ran 50 candidates, 32 of whom ran with limited Liberal opposition (Bealey and Pelling, Reference Bealey and Pelling1958, p. 258). The removal of Liberals from electoral contests seems to have been motivated by fear of protectionist Conservative victory—as the theory implies. But the nationally-negotiated Gladstone–Macdonald pact was not uniformly easily implemented. Some local Liberal associations baulked at allowing redistributive candidates a free run. In the subsequent analysis, I examine whether trade interests at the local level explain where Liberals were more willing to make this concession.

Once in office, the Liberals also moved policy toward redistribution. The People's Budget was introduced to the parliament in April 1909 to “wage implacable warfare on poverty and squalidness” (quoted in Murray, Reference Murray1973, p. 172). Its major changes were to the taxation of income. Top rates were increased by the differentiation of earned and unearned income, by increased graduation within those schedules, and by the imposition of a super-tax. Those with incomes of over $\pounds 5000$, previously taxed at the rate of 1 s, would become liable to a rate of 1s2d on their entire income, plus a supertax of 6d on income in excess of $\pounds 3000$. The budget also incorporated provisions for the taxation of land, although its subsequent implementation failed (Daunton, Reference Daunton2001, Table 11.1, p. 361). It was a deliberately progressive reform, and understood to be so. Members of the parliament knew that they were voting for, or against, progressivity. Moreover, the People's Budget did not include any directly trade-relevant changes. It incorporated changes to the taxation of alcohol, increasing domestic excises, but nothing that would lead us to expect greater support from free trade supporters.

Thus we can take two questions to the British data. First, did constituency trade orientations affect the way in which the Gladstone–Macdonald pact was implemented locally? Second, were free trade constituency MPs more likely to vote for the People's Budget?

3.3. Operationalization: constituency trade interests

The main challenge in this analysis is the measurement of trade preferences (and other variables) at the local level. Generating constituency-level measures directly from census data is impossible: parliamentary boundaries in Britain do not overlap with the census boundaries.Footnote ¹¹. I proceed instead by quantifying secondary material. I rely on the well-developed historical study of British electoral geography and create a dataset from the canonical treatment of the period, Henry Pelling's Social Geography of British Elections, 1885–1910 (Pelling, Reference Pelling1967).Footnote ¹²

Social Geography was published in 1967, and provides a constituency-by-constituency account of late Victorian and Edwardian Britain. Pelling's focus is the geography of parties' electoral success. He links electoral outcomes to constituency characteristics drawn from primary (census) and secondary (county gazette, geographic books, and essays) sources, and uses parliamentary papers and national and local press to identify major political issues within each constituency.

I create an index of trade interests from these constituency descriptions. It has three levels: − 1 if the area is in favor of protection; 0 if trade is not mentioned, or the area is characterized as neutral; and 1 if Pelling reports support for free trade. For example, the description “[t]he other objection to the Liberals was that they were always hostile to Tariff Reform, which was a continuously live issue in Sheffield throughout this period” (p. 232) yields a protectionist coding. In contrast, Manchester North-West earns a free trade coding on the basis that “a predominantly Conservative middle class had been forced to change its allegiance [to the Liberals] by the Tariff Reform question” (p. 243).

I also use Pelling to code three other variables: the level of industrialization; socioeconomic level; and an indicator of military interests which captures the presence of barracks, navy shipyards, or industries reliant on military contracts. The first two factors track local payoffs to redistribution itself. I measure military presence given the link between Tariff Reform and Imperial (and thereby military) interests (Daunton, Reference Daunton2001, p. 314).

3.4. Analysis: coalition

The first set of outcomes to consider is the local implementation of strategic coordination in the 1906 election. I measure the outcome variable using data from Kollman et al. (Reference Kollman, Hicken, Caramani, Backer and Lublin2016). It takes three levels: no labor candidate; labor candidates running, but opposed by Liberals; and finally the easiest situation for those with redistributive preferences, where labor candidates run unopposed. I count as “labor” LRC candidates and “Lib-Labs”: trade unionist Liberals.

The three-level outcome calls for an ordered specification: I implement logit models. Along with the trade interest variable, I include the constituency level of industrialization and socioeconomic status, distinguish boroughs from counties, and include an indicator for double-member districts. Ceding one seat to labor was more likely where Liberals could nevertheless contest the constituency. Since the electoral pact did not apply north of the border, I include an indicator for Scotland.

Finally, I capture the political status quo at the previous election.Footnote ¹³ I differentiate seats contested by both Liberals and Conservatives in 1900 from those in which one or both did not run. For contested constituencies, I include the Conservative vote share and its square. This is because where the Liberals were already weak, ceding “unwinnable” constituencies implies no real concession.

Figure 5 shows the results of this analysis in terms of marginal effects (coefficient estimates are in Table 3). The positive effect of the free trade indicator for unopposed labor candidates, compared to the (omitted) protectionist category, indicates that free trade interests translated to easier contests for labor candidates. The (approximately) two-point change on the linear predictor for free trade is slightly smaller than the effect of working-class constituencies in the figure, or a similar difference once the industrialization of the constituency is accounted for (model 5). This is a substantial effect.

Figure 5. Marginal effects on each level of how easy the Liberals made Labour's electoral situation in 1906. Estimates based on model 4, Table 3.

Table 3. Ease of labor election in 1906. Model 4 provides estimates for Figure 5

Variables beginning “Election” indicate constituencies contested by only one, or neither, of the major parties—the omitted category contains Lib-Con contested constituencies.

***p < 0.01; **p < 0.05; *p < 0.1.

3.5. Analysis: policy

As described above, the “People's Budget” 1909 Finance Bill was a major progressive tax reform. I examine the final vote (“division”) on the bill in the House of Commons. In British parliamentary proceedings, bills are passed in the Commons upon their third reading. This had been proposed by the Chancellor of the Exchequer, but faced an amendment to delay for three months. The division was taken on rejecting this amendment.Footnote ¹⁴ I use binary logistic models of each MP's vote. Excluding Irish and university constituencies and the City of London (which Pelling's discussion omits) yields 500 observations.Footnote ¹⁵

The estimated impact of trade interests (and other factors) on the Finance Bill votes, and their associated uncertainty, are displayed in Table 4. The data show that trade interests indeed mattered. The top two rows show the coefficients associated with neutral or free trade seats. Both types are statistically significantly more likely to support the budget measure than the (omitted) protectionist category.

Table 4. Determinants of MPs' votes on the Third Reading of the 1909 Finance Bill

***p < 0.01; **p < 0.05; *p < 0.1.

Standard errors clustered by constituency type.

Omitted categories are: protectionist trade, upper class socio-eonomic, non-industrial.

“Split delegation” models include only MPs from parties and regions which included both “aye” and “no” votes.

Model 1, with no additional covariates, shows that the predicted pattern holds in the raw data. But the theoretical claim is that trade interests should matter for progressive taxation independent of other sources of support for redistribution. Models 2–5 incorporate the controls for direct interests in redistribution through the industrialization, county/borough, and socioeconomic status measures; and the control for military presence. To these I also add controls for region (model 3) or party (model 4) or both (model 5). Models 6 and 7 replicate models 3 and 5 using only those MPs in regions and parties where there is within-region, within-party variation in the outcome.

What is striking about Table 4 is not that the impact of trade interests is statistically significant, but that the effects are large. The conditional difference between a free trade and tariff reform constituency is of roughly equal size to the difference between a predominantly working class and predominantly middle class constituency.Footnote ¹⁶

Models 4 and 5 warrant further discussion. These models include each MP's party as a covariate—model 4 dropping the region variable, and 5 including both. Political parties were becoming increasingly disciplined by the time of this vote, and the budget was a Liberal proposal. Given the importance of trade as an electoral issue, we might be concerned that trade interests are translated completely into partisan differences, making their effects impossible to differentiate from any broader differences between parties. Figure 6 indicates that while free trade constituencies were more likely to be held by Liberals, a good number of MPs in 1909 represented areas whose trade interests diverge from their party's position. These include 13 Liberal constituencies with protectionist positions (as many as there are protectionist Conservative constituencies), and 21 free trade Conservative constituencies.

Figure 6. The distribution of constituencies by constituency-type trade interests and party, 1909. Source: Author's calculations based on Pelling (Reference Pelling1967).

Second, it is unclear whether we should be interested in the effect of trade positions independent of party. Again this is an issue of post-treatment bias. While an MP's party should affect his vote, we are trying to isolate the difference between free trade, neutral, and protectionist constituencies. The party affiliation of the MP selected for a given seat is causally subsequent to this difference: Liberal MPs are elected because an area favors trade. Thus while I include these models with the party for completeness, I think models 3 and 6 are more useful estimates of the impact of trade interests.

To interpret the size of these effects, Figure 7 shows the expected probabilities of an “aye” vote for a protectionist constituency, and for a free trade seat (from model 3). For a predominantly middle class, non-industrial borough in the central region, with no notable military presence, an MP from a protectionist constituency has only a 28 percent chance of supporting the bill; if his constituency is in favor of free trade, this probability is 63 percent.

Figure 7. Expected probability of MPs voting “aye” on the Third Reading of the 1909 Finance Bill. Generated from model 3, Table 4. Other constituency covariates set to: central region; predominantly upper/middle class; non-industrial; borough.

At the local level, greater elite concern about free trade is associated with redistributive concessions, both policy and political. The data from Britain are consistent with the theoretical mechanism of coalition and compromise. This same logic underpins the interaction between inequality and trade proximity in the cross-national data. The empirical record indicates an important role for trade politics in the emergence of progressive taxation in Europe.