The Government of India Act of 1919

The British Empire in India lasted almost 200 years, beginning with the annexation of Bengal by the East India Company in 1757. Following a massive uprising of Indian soldiers against their British officers (the “Sepoy Mutiny” of 1857), the British crown took over the administration of the colony in 1858, and very gradual reforms were undertaken to include more representation of Indians in policymaking (see Online Appendix 2 for details of these early developments). Direct elections to provincial councils were first provided by the Government of India Act of 1919, also called the Montagu-Chelmsford reforms. This Act demarcated specific policy areas (such as education and health) to be under the control of these elected provincial councils, while other areas such as land revenue, finance, and law and order remained under the control of the (appointed) Governors.

Suffrage under the 1919 Act was limited to men above a certain level of income or property.Footnote ¹⁰ Our data indicate that only 2.5 percent of the population were eligible to vote in the first direct elections of 1920 (Table 1, Panel A). All citizens who were eligible to vote were also eligible to contest the election as candidates. An earlier 1909 policy of separate electorates and separate representation on the basis of religion was retained. This policy meant that there were separate Muslim (and Sikh) electoral constituencies, and these religious groups could only become candidates and vote in these reserved constituencies (Online Appendix 2 provides further details).

Table 1 SUMMARY STATISTICS OF KEY VARIABLES

Notes: Data for 1951 and 1957 exclude districts that became part of Pakistan after 1947. Voter turnout is undefined for uncontested constituencies. “Age-specific population” refers to population aged 20 and above for years prior to 1947 and to population aged 21 and above for 1951 and 1957 obtained by extrapolation from a 10 percent single-age sample.

Source: Authors’ compilation; see text for details of data sources and variable creation.

Our analysis will focus on eight major provinces of British India where provincial councils were set up by the reforms of 1919: Assam, Bengal, Bihar & Orissa, Bombay, Central Provinces & Berar, Madras, Punjab, and the United Provinces. Provincial council elections were held in 1920, 1923, 1926, and 1930. However, several of these were affected by political parties’ boycotts. The Indian National Congress declined to participate in the 1920 elections, since Mohandas Gandhi had launched a non-cooperation movement in August of that year. In 1923, after a fierce internal dispute over whether to boycott or not, many Congressmen participated under the banner of the Swaraj Party, with the aim of undermining the working of the ministries from within. The Swarajists did win a considerable number of seats on the provincial councils and continued to participate in the 1926 elections despite experiencing some internal splits. In March 1930, Gandhi launched a second Civil Disobedience movement, and Congress boycotted the 1930 elections, leading to low voter turnout and a high fraction of uncontested seats.Footnote ¹¹ In light of this, most of our analysis of the 1935 enfranchisement reform will compare the 1926 election to that of 1937, omitting the 1930 election. We show that results are similar when comparing the 1937 election to 1930.

The Government of India Act of 1935: Partial Franchise Extension

Through the 1920s, many political parties and prominent Indian individuals provided their own reports and views about constitutional changes, and two Round Table Conferences were held by the Viceroy to consult with Indian representatives. After much negotiation, the principle of federation was agreed upon, as well as continued separate communal representation for Muslims and Sikhs. Representation for lower caste Hindus was to be ensured by setting aside seats for them, but without any provision for separate electorates. All of these provisions were finally codified in the Government of India Act of 1935.

The Act of 1935 conferred full policy autonomy on provincial councils, in contrast to prior reforms that had reserved some subjects to the sole control of the Governor. The Governor was now obliged to act on the advice of the ministers, except in matters of “grave menace to peace or tranquillity” or “safeguarding the interests of minorities.” The provincial legislative councils were expanded (and renamed Legislative Assemblies), and the bigger provinces were provided bicameral legislatures.

The franchise was considerably expanded in several ways. First, in most provinces, the property thresholds were lowered considerably (see Table A1 in Online Appendix 1 for details of suffrage requirements). For instance, the minimum thresholds for voter eligibility in Bengal under the 1919 Act included at least Rs 1-8-0 in municipal taxesFootnote ¹² and fees, Rs 1 in public works cess, Rs 2 in chaukidari tax, or occupying a house valued at Rs 150. These thresholds were reduced considerably in the 1935 Act to 8 annas, 8 annas, 6 annas, and Rs 42, respectively. Given that there was little net inflation over this period, this amounted to reductions of more than 50 percent in the asset thresholds required for voter eligibility. Second, suffrage was also extended in some provinces to educated persons or literate women. Third, women who were wives or widows of qualified male voters (with higher property thresholds than required for male voting) were also allowed to vote. The share of women increased from 4 percent of the electorate in the 1930 election to 14 percent in 1937. The 1935 reform therefore changed the electorate towards a poorer and somewhat more feminine one; we expect that these societal groups are overall less likely to be educated than those previously enfranchised.

As a result of all these changes, the nationwide fraction of enfranchised electors increased to 11.7 percent in the provincial elections of 1937, though the figures varied considerably across provinces and across districts within the same province (see the map in Figure A1 of Online Appendix 1). Such variation is driven both by differences in the voting requirements across provinces and by differences in the distribution of assets, incomes, and education levels across districts.

Post-1935 Political Developments

There was widespread participation by voters and political parties in the provincial elections of 1936–37. The Indian National Congress formed governments in 8 out of 11 provinces. However, all the Congress ministries resigned in October 1939, in protest against Viceroy Linlithgow’s announcement of India’s entry into WWII without any consultation from Indian representatives. This extremely short tenure of the representatives elected after franchise extension makes it difficult for us to examine the policy consequences of the 1935 reform. However, there is evidence that the short-lived Congress ministries formed after the 1936–37 elections made concerted efforts to implement their policy agendas. Education stood out as one of the main areas where the efforts of the ministries were concentrated. In Bombay, the government passed a bill that made provisions for the “better management and control of primary schools,” instituted a board of education to deal with the problem of adult illiteracy in the province and provided special educational grants to disadvantaged groups. In Bihar, a mass literacy movement was initiated that “made good progress with the help of about 14,000 volunteers drawn from the intelligentsia of the province.” The government in the United Provinces financed the construction of “a network of 960 adult schools, 760 circulating libraries and 3000 reading rooms.” The Orissa government provided funds for a literacy campaign and library movement across all villages and also abolished fees in public primary schools (Indian National Congress 1939). We will examine education spending as a key policy outcome.

The demands for Indian independence continued, with the Congress launching the “Quit India” movement in 1942; many Congress leaders were jailed for their participation in this event. After the end of WWII, it was clear to many that India would not remain a British colony for much longer. Elections to provincial and central legislatures were held in December 1945–January 1946, with all major parties participating.

Party Politics in the Colonial Period

Prior to the 1935 Act, electoral candidates were predominantly social and economic elites who used their influence to obtain votes for themselves as opposed to subscribing to a political party’s platform (Jaffrelot Reference Jaffrelot1998). Political parties were not well organized and the largest one, the Indian National Congress, had boycotted some elections in the 1920s. Narrative accounts emphasize that candidates nominated themselves and personally ran their campaigns rather than relying on (barely existing) party machinery. The main issues emphasized were “ …of personality, community and local influence [rather] than of party or programme” (Indian Statutory Commission 1930, p. 199). Other accounts concur that voting was on the basis of individual personality and status rather than policy issues.Footnote ¹³ The relative weakness of political parties means that we need to think of political competition as being among individuals rather than across parties. We therefore use the number of candidates contesting as well as whether the incumbent manages to retain their seat as measures of political competition in this period.

By the 1937 elections, political parties, especially the Congress, were much better organized for political activity. The parties set up provincial committees and provincial parliamentary boards to recruit candidates, helped candidates with filing nominations, provided (some) campaign funding, and coordinated campaign messages. The issues emphasized were national in nature rather than focused on local interests, including the question of independence from colonial rule; the rising support for the Muslim League also brought religious divisions to the forefront. However, the role of parties was not fully institutionalized. For instance, nomination forms in the United Provinces did not ask candidates to specify their party affiliation (Reeves, Graham, and Goodman Reference Reeves, Graham and Goodman1975, p. li). By 1945, the Muslim League had emerged as a dominant political force in Muslim-majority areas. The issue of Pakistan and the partition of the subcontinent took center stage during the 1945 election campaign, with the Congress strongly opposing partition and the Muslim League championing the cause of a separate Pakistan (Azad Reference Azad1988; Zaidi Reference Zaidi1970).

It is important to note that even though parties became important vehicles of political mobilization in the post-1935 era, this did not mean that political competition between different individual candidates went away. For instance, individuals could choose whether to become candidates of the dominant Congress party, or those representing other sectional or regional interests such as the Unionist and National Agriculturist Parties (who represented landowners in Punjab and the United Provinces, respectively) or Dr. Ambedkar’s Independent Labour Party in Bombay (which championed the cause of lower castes). Inter-candidate political competition persisted within the Congress party, as the party changed its criteria for choosing candidates over time (Chiriyankandath Reference Chiriyankandath1992, pp. 55–56). Our main measure of political competition—change in the fraction of incumbents who get re-elected—is well placed to track both these types of inter-candidate electoral competition both before and after the 1935 Act. We also compute the fraction of winners from the Congress party as a partial measure of cross-party competition. This is possible only for years 1937 and later.

The Indian Constitution of 1950: Universal Adult Suffrage

In 1947, India ceased to be a British colony and was partitioned into the two new nations of India and Pakistan. Partition resulted in one of the largest, most rapid, and most violent migrations in human history (Bharadwaj, Khwaja, and Mian Reference Bharadwaj, Khwaja and Mian2008). The first provincial elections in independent India were held in 1951, following the adoption of a new constitution in 1950.Footnote ¹⁴ This constitution established India as a sovereign democratic republic, and universal suffrage for all citizens aged 21 or older was secured under Article 326. Since the 1950 reform had no property, income, education, or gender restrictions, it greatly enfranchised poorer, less educated, and female citizens, similar to the 1935 reform. Consequently, the ratio of registered voters to total population increased dramatically to 48.2 percent in the provincial elections of 1951 and 1952 (Table 1, Panel A). The share of women in the electorate increased from 14 percent in the 1937 elections to 47.5 percent in 1957. Reserved seats were retained for members of the Scheduled Castes (communities that were at the bottom of the Hindu caste hierarchy) and Scheduled Tribes (communities traditionally outside the Hindu caste system), but there were no more reservations for women under the new constitution. There were also no provisions for separate electorates on any basis. In 1989, the minimum age to vote was reduced to 18 years by the 61st Amendment to the Indian constitution.

Emerging from colonial rule, the Congress Party transformed itself from a broad-based nationalist movement into the dominant political party of the nation, winning a national legislative majority in the 1951 elections and also becoming the largest party in every state legislature. In most states, Congress remained the largest party for the first three decades after independence.Footnote ¹⁵ Given this dominance in the pre- and post-Independence periods beginning with the 1937 elections, we track two measures of party-level political competition in the post-1950 period, namely the share of winners from Congress and the number of parties that nominate candidates for each seat. We continue to track the fraction of incumbents who get re-elected as a measure of inter-candidate political competition, and we are also able to track the fraction of incumbents who run for re-election.

Data Sources

We collected and digitized archival data on enfranchisement and election outcomes from the “Returns Showing the Results of Elections in India” published after each election during the colonial era. (India Office 1921, 1924, 1927, 1931, 1937; Government of India 1948). These reports cover the elections to provincial assemblies in the colonial period, namely those of 1920, 1923, 1926, 1930, 1937, and 1945. Data for the post-colonial state assembly elections of 1951, 1955, and 1957 were obtained from the official election reports of the Election Commission of India for elections in the 1950s.

All the election reports provide the number of registered voters, the number of votes cast, and the number of candidates in each constituency. However, colonial-era reports often do not report names of non-winning candidates or party affiliations, and data on winner names is also incomplete. We used several supplementary sources of information to create a full panel of election winners’ names over time, including the “Who’s Who” publication from the Times of India Yearbook, and region-specific accounts (Online Appendix 3 provides a detailed list of sources). Vote shares and vote margins are consistently available only in the post-colonial data. Only 10–12 percent of candidate names from the colonial period mention education or occupation and the fraction varies considerably across provinces and years. The extent of missing data precludes systematic analysis on the characteristics of candidates. Even such partial data is not available in the post-colonial election reports.

Data Aggregation to District Level

To track political outcomes over time, we face the issue of constructing geographically stable units over time. Electoral constituency boundaries changed over time, and both the enfranchisement reforms we examine greatly expanded the number of elected representatives. As a consequence, we created district-level aggregate variables, since these administrative district boundaries remained relatively stable over time. In the few cases in which new districts were created, we aggregate them back to their previous boundaries. We consistently perform our analysis at the level of the 1921 district boundaries. Our district-level data is available for replication purposes (Cassan, Iyer, and Mirza 2024).

Measures of Enfranchisement

Our main explanatory variable is the enfranchisement rate, which is defined as the number of registered voters in the district divided by its total population. Data on the number of registered voters is available even if the election is uncontested, since these figures are compiled prior to the election taking place. The district-level number of registered voters is obtained by aggregating constituency-level figures, using the procedures described in Online Appendix 3. For district-level population data, we use the censuses of 1921, 1931, and 1951 (the 1941 census quality and coverage were compromised by the constraints of wartime), assigning the previous census-year population to each election year. Using this measure, we find that only 3 percent of a district population was eligible to vote in 1926, and this fraction increased to 11.7 percent in the 1937 election (Table 1, Panel A). The introduction of universal adult franchise in 1950 had a huge impact, raising the average fraction of enfranchised population in a district to 48 percent in the 1951 election.

We examine robustness to using two alternative population estimates. The first is an interpolated population figure in the denominator of the enfranchisement measure. For districts that experienced abnormal increases or decreases in population due to the 1947 partition, such extrapolated population figures for 1957 can be erroneous. The second is an age-adjusted population figure, which suffers from measurement constraints.Footnote ¹⁶ Despite these, we find that both of these alternative measures show similar four-fold increases in enfranchisement rates after both the 1935 and 1950 reforms (Table 1, Panel A).

A potential source of mismeasurement in our data arises from the possibility that not all eligible voters may be actually registered to vote. An electoral roll was prepared for every constituency on which the names of all persons entitled to be registered as voters were to be entered. Once prepared, the roll was to be published in the constituency together with a notice specifying the mode and time period within which any claims by individuals who felt that they should be included in the roll were to be submitted to the revising authorities. The responsibilities for the preparation of the roll, the timing of its publication, the procedure for addressing claims regarding the electoral roll, and the constitution of the revising authorities were all left to district administration officials. These officials were required to be non-partisan, being mostly career bureaucrats from the Indian Civil Service (ICS) and related provincial civil services. ICS officers were also rotated across districts quite frequently (Potter Reference Potter1996); we expect this to limit both the district officials’ ability and their incentives to skew electoral rolls in favor of specific candidates in a given area. Such mismeasurement can lead to bias in our estimates if district-level registration rates are correlated with other unobservable district-specific trends that also drive the outcome. This could be the case, for instance, if districts that experienced lower (or more) political competition for any reason also promoted greater rates of registration and therefore recorded higher enfranchisement figures. We therefore control explicitly for such pre-reform trends as a robustness check in our estimation.

Measures of Political Participation

We track citizens’ political participation as voters and as candidates. Our main measure of voter participation is the number of citizens who voted divided by the total population. Note that if none of the newly enfranchised voters exercised their right to vote, this measure would not change with enfranchisement. On the other hand, if all of them chose to vote, then this measure would increase exactly as much as the enfranchisement measure. We face some measurement challenges in computing this variable: we do not observe the population of each constituency (but that of the more aggregated district), and we do not know the number of voters in a constituency when the election is unopposed in that constituency. As a consequence, we cannot distinguish if an increase in the ratio of voters to population is due to a decrease in the number of uncontested seats in that district or to a genuine increase in the number of voters, holding the number of uncontested seats constant. We address this concern by tracking these two supplementary measures, namely the fraction of uncontested seats in the constituency and the voter turnout, defined as the ratio of total votes cast in a specific election to the total number of registered voters (this explicitly excludes the uncontested seats).

We see that the population share of voters did increase after each reform, rising to 5.5 percent in 1937 from 0.5 percent in 1930 and from 4.5 percent to 22.2 percent between 1945 and 1951 (Table 1, Panel B). Some of this increase could potentially be driven by the steep decline in the share of uncontested elections in both 1937 and 1951 (Table 1, Panel C). For 1930, voter share of the population could be further affected by the calls for widespread boycott. Tracking voter turnout as a share of registered voters, we see that this rises from 51 percent in 1926 to 57 percent in 1937 (with a big drop in 1930). In contrast, voter turnout declined slightly from 49 percent in 1945 to 46 percent in 1951, the first election after the institution of universal adult suffrage in 1950.

We measure the candidate participation of citizens as the number of candidates per 1,000 registered voters in the district. Since the eligibility criteria were the same for voting and for candidacy, this measures what fraction of the potential candidate pool actually becomes candidates. As with voters, the newly eligible candidate pool may not translate into actual candidacy if there are significant informational, financial, or societal barriers to becoming candidates. We also track a supplementary measure, namely the candidate-population ratio.

Both measures of candidate participation declined considerably between 1926 and 1930, reflecting the impact of the Congress-led boycott (Table 1, Panels B and C). In contrast, after the franchise extension of 1935, the candidate share of the population almost tripled between 1930 and 1937, reflecting the expansion of the potential candidate pool, but the share of registered voters who became candidates actually fell from 0.27 per 1,000 registered voters in 1926 to 0.14 in 1937. This suggests that the expansion of the candidate pool did not translate to a proportionate expansion in the number of actual candidates. In a similar manner, the number of candidates per 100,000 population increased more than three-fold after the franchise extension of 1950, but the share of registered voters that became candidates fell from 0.091 per 1,000 voters in 1945 to 0.077 in 1951.

Measures of Political Competition

We compute three main measures of political competition: the number of candidates per seat in the district (which reflects the extent of opposition faced by those standing for election), the incumbent re-election rate, and the fraction of seats won by Congress (as a measure of party competition). Data limitations prevent us from computing other measures of political competition, such as the effective number of parties or the winning margin. The incumbent re-election rate is computed as the fraction of incumbent politicians who get re-elected in the next election. So “fraction of incumbents re-elected 1923” refers to the fraction of incumbent politicians (i.e., those who got elected to the provincial legislature from that district in 1920) that win re-election in 1923. Tracking specific politicians across election years is conducted via fuzzy matching combined with manual checking (see Online Appendix 3 for details). For the elections of 1937 and later, which provide data on all candidates’ names (not just the winners), we can compute a supplementary measure of incumbency advantage as the percentage of incumbents who run for re-election. We are also able to track the fraction of uncontested elections over time.

We see that the average candidate-seat ratio increased after both the franchise extensions, from 2.27 in 1926 to 2.60 in 1937 (as expected, the boycott in 1930 resulted in an unusually low candidate-seat ratio of 1.83), and from 2.22 in 1945 to 4.51 in 1951 (Table 1, Panel B). The fraction of incumbents who got re-elected fell from 35 percent in 1926 to 13 percent in 1937, and from 25 percent in 1945 to 16 percent in 1951.Footnote ¹⁷ Some of the decline can be attributed to incumbents not running for re-election: only 22 percent of previous winners ran for re-election in 1951 (after franchise extension), compared to 31 percent in 1945 (Panel C). The share of uncontested elections also decreased drastically after each reform, from 15 percent in 1926 to 6 percent in 1937, and from 26 percent in 1945 to 0.2 percent in 1951 (Panel C). Turning to party-level competition, we find that the fraction of Congress winners increased from 64 percent in 1945 to 74 percent in 1951, consistent with the narrative of Congress establishing a nationwide dominance after Indian independence. This dominance is also reflected in the fact that the number of parties that nominate candidates for a given seat barely increased after the 1950 enfranchisement.

First-differenced Regressions

We examine whether districts with larger enfranchisement increases experience larger changes in measures of political participation and competition compared to districts with smaller increases in enfranchisement. Our main regression is a first-differenced specification:

(1) $$\Delta {Y_d}{\rm{ }} = {\rm{ }}\alpha {\rm{ + }}\gamma \,\Delta Enfranchisemen{t_d}{\rm{ }} + {\rm{ }}{X_d}'\delta {\rm{ }} + {\rm{ }}{e_d}$$

where ∆Y _d is the measured change in political participation or competition in district d, ∆Enfranchisement _d is a measure of the change in enfranchisement in district d, and X _d is a vector of district-level covariates such as total population, population growth between censuses, urbanization rates, gender ratios, literacy rates, and religious mix; e _d is an error term. Due to concerns about electoral boycotts, we will compute changes between 1937 and 1926 in order to measure the impact of the 1935 reform and show the change between 1930 and 1937 as a robustness check. For the 1950 reform, we compute the change between 1945 and 1951, and we also include the fraction of refugees in the population as an additional control in order to account for the direct effects of partition and displacement on political outcomes.

When using data from two periods, the coefficients from the first-differenced Specification (1) are identical to those from a difference-in-difference specification with district fixed effects and a time dummy. Since we are comparing each district to itself before and after enfranchisement, we are in effect controlling for any time-invariant characteristics of districts such as geography, prior history, length of colonial rule, land tenure systems, or other institutional characteristics. Any changes induced by the reforms that are common to all districts, such as national changes in the political environment or increases in the total number of representatives, are captured by the constant term α.

The key identification assumption in any difference-in-difference analysis is that areas that were more or less affected by enfranchisement would have experienced similar changes in the outcome variable before and after the reform in the absence of this enfranchisement reform (the “parallel trends” assumption). While this assumption is not directly testable (we cannot observe what affected areas would look like in the absence of reform), this would be violated if changes in enfranchisement rates happen to be correlated with other characteristics of the district (time-varying or time-invariant) that lead to divergent growth in the outcome variables over time, or if we observed more vs. less affected areas to be trending differently even before the reform took place (differential pre-trends). Not observing differential pre-trends would suggest that this assumption is reasonable.

We examine and control for these possibilities in several ways. First, we include several district-level observable characteristics X _d in our regression. Controlling for these characteristics in a first-differenced specification means that we are controlling for differential trends based on these characteristics. Second, we directly control for pre-reform trends in the outcome variable. Third, we run an event-study specification to check for the existence of differential pre-trends in the data. Finally, we run a difference-in-difference specification with district-specific trends to control for any sources of linear trends that are not captured by observable characteristics.

Event Study Approach and Difference-in-Difference Regression

The event study specification compares changes in the outcome variables across different time periods and across high vs. low enfranchisement areas:

(2) $${Y_{dt}} = {\lambda _d} + {\Sigma _t}{\beta _t} + {\Sigma _t}{\gamma _t}*\Delta Enfranchisemen{t_d} + {\rm{ }}{X_{dt}}'\delta {\rm{ }} + {\rm{ }}{u_{dt}}$$

where Y _dt is the outcome for district d in year t, λ _d is a dummy (fixed effect) for district d, β _t is a series of time dummies for each election year t, ∆Enfranchisement _d is as defined earlier (measuring whether a district is more or less affected by the reform), and γ _t is the election-year-specific impact of the change in enfranchisement. For the 1935 reforms, the election years included are 1920, 1923, 1926, 1930, 1937, and 1945. The omitted category will be 1920 so that the γ _t coefficients represent the impact of 1935-reform enfranchisement on the outcome in year t. The corresponding specification for the 1950 reform would include election years 1937, 1945, 1951, and 1957, with the year 1937 being the omitted category.

This specification directly allows us to assess the possibility of differential pre-trends: if our identification strategy is sound, the increase in enfranchisement should affect outcomes only after it is implemented and not before. This also provides a partial check for the possibility of reverse causality: if franchise changes are endogenous to the outcome variables, we may see a differential change in outcomes across more vs. less affected areas prior to the reform. Further, the event study approach allows us to see if the impact of the reform persists over time. Note, however, that the event study specification is more demanding in terms of statistical degrees of freedom, since additional coefficients need to be estimated. It is also less straightforward to interpret since it provides several coefficients for the impact of the reform.

Our results (Figures A2 and A3 of Online Appendix 1) indicate that there are no differential pre-trends prior to the reform and that the reform appears to have a one-time level effect on the outcomes. Based on this observation, we can run a more parsimonious difference-in-difference (DID) regression as follows, grouping all pre-reform years and post-reform years into two categories:

(3) $${Y_{dt}} = {\lambda _d} + {\sum _t}{\beta _t} + \gamma *Pos{t_t}*\Delta Enfranchisemen{t_d} + {\rm{ }}{X_{dt}}'\delta {\rm{ }} + {\rm{ }}{u_{dt}}$$

The variables are the same as in Equation (2), and Post _t is an indicator for post-reform years. For the 1935 reform, election years 1937 and 1945 would take on values of one for the Post dummy, while election years 1920–1930 would have a zero value. For the 1950 reform, election years 1951 and 1957 would have Post = 1, while years 1937 and 1945 would have Post = 0. The advantage of this specification is that it is less demanding econometrically and easier to interpret, because it produces only one coefficient of interest. Another advantage of this specification is that we can include district-specific time trends in the set of control variables X _dt , so that we are controlling for any linearly evolving district unobservables. It comes at the cost of assuming a homogeneous treatment effect over time. To adjust for outcomes being correlated over time within the same district, we cluster our standard errors at the district level in Specifications (2) and (3).

Political Participation

We find that increasing enfranchisement does not lead to equally large increases in citizen participation as voters or candidates. Table 2 shows the results of running Specification (1) for the 1935 reforms, comparing the 1937 elections to that of 1926 (as the 1930 election was affected by boycotts). Our preferred specification is in Column (3), where we include the district demographic controls mentioned earlier and exclude four districts that are outliers in terms of the enfranchisement variable.Footnote ¹⁸ We find that increasing the population share of enfranchised citizens by 10 percentage points increases the voter share of the population by 4.1 percentage points. If we make the (somewhat extreme) assumption that all of the previously enfranchised voters continue to vote, this would mean that only 41 percent (i.e., less than half) of the newly enfranchised voters are exercising their right to vote. Similarly, a 10 percentage point increase in enfranchisement results in a statistically significant decline of 0.14 candidates per 1,000 registered voters (Table 2, Panel B, Column (3)). Note that such reduced citizen participation is not an obvious consequence of poorer voters becoming enfranchised, since poor citizens’ voter turnout is higher in many post-independence Indian elections (Kumar Reference Kumar2009). Therefore, the less-than-proportional increase in turnout may be caused either by the newly enfranchised population voting less or by the previously privileged population now being less politically engaged (as documented for a different reform by Suryanarayan Reference Suryanarayan2019). Unfortunately, our data does not enable us to distinguish between these two mechanisms.

Table 2 IMPACT OF 1935 REFORMS ON POLITICAL PARTICIPATION AND COMPETITION

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes: Robust standard errors in parentheses. Controls include district population, literacy, urbanization, gender ratio, fraction of Hindus and Christians, fraction of population employed in agriculture in 1921, and population growth rates between 1921 and 1931. Difference-in-difference (DID) estimates based on Specification (3) in the paper.

Source: Authors’ compilation; see text for details of data sources and variable creation.

We verify that these patterns are not driven by pre-existing trends in these political outcome variables that just happen to coincide with high versus low enfranchisement (Table 2, Panels A and B, Columns (4) and (5)). We show that the estimated coefficients remain similar in size and statistical significance when we directly control for pre-reform changes in the outcomes (i.e., the change between 1923 and 1926), and when we use a DID specification with district-specific linear time trends as in Specification (3).

We examine the medium-term effects of the 1935 reforms by examining the changes between 1937 and 1945. In particular, if we think that the relatively muted effects on citizen participation result from a lack of familiarity with the electoral system, we would expect the participation outcomes to show increases over time as citizens acquire political knowledge and experience. We find, however, that this is not the case: places with greater increases in enfranchisement due to the 1935 reform actually exhibit a statistically smaller growth in the voter share of population between 1937 and 1945. While these places do show a larger increase in the candidates per 1,000 voters, the coefficient is roughly one-third as large as the original negative coefficient, suggesting that the subsequent increase in candidacy is not enough to offset the original decline (Table 2, Panels A and B, Column (6)).

A similar pattern can be observed in the election-year-specific effects of enfranchisement, obtained from running Specification (2). Figure A2 of Online Appendix 1 shows that the voter share of the population did not increase prior to the reform and increases sharply thereafter (Figure A). Figure B shows a slight decreasing trend for candidates per 1,000 voters prior to 1935, but a much bigger decline in the 1937 elections with some recovery in the 1945 election. Note that the points on this graph represent how much outcomes change with respect to 1920 as the omitted year so that the main effects in Table 2 (Column (3)) would be equivalent to taking the difference between the point estimate for 1937 and the point estimate for 1926 on this graph. Similarly, the medium-term effect would be the difference between the point estimate for 1945 and that for 1937.

Political Competition

Turning to measures of political competition, we find that enfranchisement does not have any statistically significant impact. A 10 percentage point increase in enfranchisement results in 0.21 additional candidates per seat and a 1.5 percentage point decline in the fraction of incumbents getting re-elected (Table 2, Panels C and D, Column (3)). These remain statistically insignificant when controlling for the pre-reform change in outcomes or controlling for district-specific trends (Columns (4) and (5)). Interestingly, we see decreases in political competition in the medium term: places that experienced greater increases in enfranchisement show a bigger decline in the candidate-seat ratio between 1937 and 1945 and a larger increase in the fraction of incumbents getting re-elected (Column (6)). The event-study graph (Online Appendix 1, Figure A2, part D) shows that this is mainly because the point estimate for 1937 is negative (in line with our main estimate of Table 2, Column (3)) and there is an increase in incumbency advantage in 1945, bringing it back to 1923 levels.

Robustness Checks

We conduct several robustness checks for our results, shown in Online Appendix 1, Table A3. First, we show that our results remain similar in size and significance when we use the interpolated census population to calculate enfranchisement rates rather than the previous census population (Column (1)), and when we use the age-specific population as the denominator to calculate enfranchisement rates (Column (2)). In fact, with this latter measure of enfranchisement, we also see a marginally significant increase in the number of candidates per seat. The magnitudes of these coefficients are similar to our main measure: a one standard deviation change in our original enfranchisement measure results in a 1.27 percentage point increase in the voter share of the population and a decline of 0.044 candidates per 1,000 registered voters; the corresponding magnitudes for a one standard deviation change in the age-specific enfranchisement measure are 1.07 and 0.046. Similarly, our results remain similar in size and significance when controlling for the change in the number of seats in each district (Column (3)), when we examine changes between 1930 and 1937 instead of the changes between 1926 and 1937 (Column (4)), and when we drop districts that later became part of Pakistan (Column (5)).

An alternative way to control for potential omitted variables bias is to use an instrumental variable (IV), namely something that changes district-level enfranchisement but is uncorrelated with district-specific characteristics. In our setting, one key source of variation in enfranchisement rates across districts is the wealth threshold and other rules for enfranchisement chosen by the provincial government, which were usually uniform over the whole province with only a few exceptions. While this may not be fully uncorrelated with district characteristics, it does provide a somewhat exogenous reason for enfranchisement rates to vary across districts. Accordingly, we reran our empirical specification using province dummies as instruments for district-level enfranchisement changes.Footnote ¹⁹ Again, we find the same results as before: a less-than-proportional increase in the voter share of the population, a significant decline in the number of candidates as a share of registered voters, and no significant effect on candidate-seat ratios or the fraction of incumbents getting re-elected (Column (6)).

Supplementary Outcomes

Our main conclusions regarding the effect of enfranchisement on political participation and competition do not change when we examine alternative outcome measures in Online Appendix 1, Table A4. We see that a 10 percentage point increase in enfranchisement results in a statistically significant 7.32 percentage point decline in voter turnout, measured as the share of registered voters who turn out to vote (Panel A, Column (3)). This is consistent with the result that enfranchisement results in a less-than-proportional increase in the voter share of the population. We also see a significant increase in the candidate share of the total population, but the magnitude of this is fairly small: a 10 percentage point increase in enfranchisement (i.e., in the potential candidate pool) results in 0.33 more candidates per 100,000 population (Panel B, Column (3)). As before, the results are robust to controlling for the pre-reform change in outcomes as well as to the inclusion of district-specific time trends (Panels A and B, Columns (4) and (5)), and show no significant increase in the medium term either (Column (6)). Our supplementary measure of political competition—the fraction of uncontested seats—shows no signification relationship with our measure of enfranchisement (Panel C). This is consistent with the insignificant results on political competition in Table 2.

Main Results

We examine the impact on political participation and competition of the adoption of the 1950 constitution that granted suffrage to all adult citizens of India, greatly increasing the fraction of enfranchised population in a district by 36.3 percentage points on average, compared to the last pre-independence election of 1945 (Table 1, Panel A). In theory, the impact of such a large extension can be quite different from the earlier expansion of 8.8 percentage points, both because of the different characteristics of those enfranchised by each reform and because of the different political contexts in which these franchise extensions took place. Our main regression specification is still based on Equation (1), but with changes now being measured between the elections of 1945 and 1951.

The impact of the 1950 reform on political outcomes is strikingly similar to that of the 1935 reform. On citizen participation, we find that the enactment of universal adult franchise results in a statistically significant but less-than-proportional increase in voting and a statistically significant decrease in candidates per 1,000 voters. A 10 percentage point increase in enfranchisement results in only a 3 percentage point increase in the voter share of the population, according to our preferred estimates (Table 3, Panel A, Column (3)). A similar calculation suggests that a 10 percentage point increase in enfranchisement results in 0.011 fewer candidates per 1,000 registered voters (Panel B, Column (3)).

Table 3 IMPACT OF 1950 REFORMS ON POLITICAL PARTICIPATION AND COMPETITION

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes: Robust standard errors in parentheses. Controls include district population, literacy, urbanization, gender ratio, fraction of Hindus and Christians, fraction of population employed in agriculture in 1931, population growth rates between 1931 and 1951, and the fraction of refugees in 1951. Difference-in-difference (DID) estimates based on Specification (3) in the paper.

Source: Authors’ compilation; see text for details of data sources and variable creation.

These estimates remain similar in size and significance when controlling for the pre-reform change in outcomes from 1937 to 1945 (Panels A and B, Column (4)), and when controlling for district-specific time trends, though the estimate for the candidate-voter ratio loses statistical significance when we control for district-specific time trends using the DID specification (Panels A and B, Column (5)). The medium-term effects, namely the change from 1951 to 1957, are not larger than the immediate effects, showing that citizen participation does not pick up over time in heavily enfranchised districts; in fact, the trends in voter participation are somewhat worsened (Table 3, Panels A and B, Column (6)). The event-study graphs in Figure A3 of Online Appendix 1 (parts A and B) confirm that our estimates are not confounded by any pre-existing differences in outcomes between areas with higher and lower enfranchisement.

Similar to the effects of the 1935 reform, we find that increased enfranchisement does not result in increased political competition. A 10 percentage point increase in enfranchisement increases the candidate-seat ratio by a statistically insignificant 0.17, increases the fraction of incumbents getting re-elected by 6 percentage points (statistically significant at the 10 percent level), and reduces the fraction of Congress winners by an insignificant 1.5 percentage points (Table 3, Panels C–E, Column (3)). There are also no significant effects of enfranchisement on measures of political competition in the medium term (Column (6)). Our event study graphs in Figure A3 of Online Appendix 1 (parts C–E) also show no effects of enfranchisement reform on political competition.

Robustness Tests and Supplementary Outcomes

We conduct a number of robustness checks for the relationships documented in Table 3: using alternative measures of enfranchisement (using interpolated census population or estimated age-specific population as the denominator rather than the previous census population), controlling for the increase in the number of seats, and dropping the provinces of Punjab and Bengal to avoid any confounding effects of partition-related deaths or displacement.Footnote ²⁰ The results are shown in Online Appendix 1, Table A5. Our results are mostly robust to these changes in our specifications, with the exception of the change in the candidate-seat ratio, which becomes positive and statistically significant when controlling for the change in the number of seats or when we drop Punjab and Bengal. Note that using age-specific population to compute enfranchisement produces estimates that are very similar in magnitude: a one standard deviation increase in our main enfranchisement measure results in a 2.10 percentage point increase in the voter share of the population and a decline of 0.0079 candidates per 1,000 registered voters; the corresponding estimates for a one standard deviation increase in the age-specific measure are 2.11 and 0.0091.

We show results for supplementary political outcomes in Online Appendix 1, Table A6. Areas with higher enfranchisement increases show a bigger decline in voter turnout, which is measured as the fraction of registered voters who actually voted (preferred estimates in Panel A, Column (3)). This is in line with our less-than-proportional increase documented earlier for the voter share of the population. Our supplementary variable of candidate participation, namely the number of candidates per 100,000 population shows a significant increase of about 0.75 for every 10 percentage point increase in enfranchisement (Panel B, Column (3)). We should note that this amounts to only 68 percent of the mean value in 1945, while enfranchisement increased fourfold between 1945 and 1951; it is thus consistent with our finding in Table 3 that candidates as a share of registered voters declined significantly.

Turning to supplementary measures of political competition, we see no impact of enfranchisement on the fraction of uncontested seats, similar to the 1935 reform. However, places with bigger increases in enfranchisement have a significant increase in the fraction of incumbents who run for re-election and a significant decline in the average number of parties contesting each seat. A 10 percentage point increase in enfranchisement results in a 7 percentage point increase in the fraction of incumbents that run for re-election (Panel D, Column (3)), which is 23 percent of the 1945 mean value. Table 3 showed that the fraction of incumbents winning increased by 5.92 percentage points for the same 10 percentage point increase in enfranchisement, indicating that about 84 percent of these re-running incumbents managed to retain their seats. Similarly, a 10 percentage point increase in enfranchisement decreases the number of parties per seat by 0.16 (Panel E, Column (3)), which is 22 percent of the 1945 mean value. Overall, we find no evidence that enfranchisement increases political competition and some evidence that it may increase incumbency advantage both at the individual and the party level.

Education and Health Spending after the 1935 Reform

Delineating the policy effects of the 1935 enfranchisement reform is hampered due to historical events and data limitations. The Congress-led ministries that were elected in 1937 resigned in 1939 in protest against Viceroy Linlithgow’s unilateral announcement of India’s entry into WWII, and policy decisions after 1939 may be confounded by the effects of wartime constraints. This gives us a relatively short time frame to assess the impact of the 1935 reform. Education spending is a particularly interesting outcome, given that primary education will disproportionately benefit the poorer strata of the population, from which the newly enfranchised population is drawn, whereas middle and secondary education will benefit the relatively richer, already enfranchised population. We also examine data on health spending, where we expect a lower preference differentiation between rich and poor voters, since large-scale public health measures (vaccination, sanitation, etc.) to combat the widespread infectious diseases would be equally available to rich and poor voters. While provincial governments also enacted other policies (e.g., Bombay Trade Disputes Act of 1938, the United Provinces Tenancy Bill, 1939), we do not have systematic data on how these measures affected each district.

We obtained district-level data on per capita education and health spending from the annual issues of the “Report on the Working of District Boards” for the provinces of Assam, Bihar, Central Provinces, and the United Provinces over the period 1931–1940 (with some missing years for each provincial series). Using the period 1931–1934 as “pre-enfranchisement” and 1937–1940 as “post-enfranchisement” years, we calculate the difference in the per capita spending on education and regress it on district-level enfranchisement increases, as in Equation (1). We find that districts that experienced larger increases in enfranchisement also experienced larger increases in per capita education spending, though the effects are somewhat imprecisely estimated given the limited nature of the data. In particular, we find that districts with a 10 percentage point increase in enfranchisement had 0.01 rupees per capita higher education spending, which is 5 percent of the pre-1935 mean.Footnote ²¹ This estimate is computed after controlling for demographics and removing outliers and is statistically significant at the 10 percent level of significance (Table 4, Panel A, Column (3)). Most of this increase is attributable to the increase in primary school spending rather than middle school spending: while these coefficients are not statistically significant, we see that the primary school spending coefficient is almost the same size as the overall coefficient while the coefficient on middle school spending is much smaller and negatively signed (Online Appendix 1, Table A7, Panels A and B, Column (3)). These results are consistent with the earlier narrative evidence of elected provincial councils prioritizing education as a policy area. Note that our finding of greater expenditure is net of any revenue declines that may have occurred in anticipation of the 1935 reform, as Suryanarayan (Reference Suryanarayan2021) has documented for the initial enfranchisement of 1919. In contrast, we find no significant increase in health spending directed towards areas with greater enfranchisement increases (Table 4, Panel B), consistent with the idea that rich and poor voters do not have divergent preferences over public health spending.

Table 4 IMPACT OF THE 1935 AND 1950 REFORMS ON HUMAN CAPITAL SPENDING AND ACCESS

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes: Robust standard errors in parentheses. Controls include district population, literacy, urbanization, gender ratio, fraction of Hindus and Christians, fraction of population employed in agriculture, population growth rates, and fraction of refugees (for the 1950 reform). Dependent variable for the 1935 reform is the change in per capita spending between pre-1937 (1931–1934) and post-1937 (1937–1940) years. Figures for spending and enfranchisement were calculated using interpolated population figures for each year.

Source: Authors’ compilation; see text for details of data sources and variable creation

Access to Education and Health Facilities after the 1950 Reform

District-level expenditure data is not available for the post-independence period. To track the progress of education provision, we therefore track the fraction of villages that had at least one primary school, obtaining data from the 1951 and 1961 district census handbooks. We are able to obtain these data for the provinces of Bengal, Bombay, Madras, Punjab, and the United Provinces. We also track similarly the fraction of villages that had a hospital or dispensary; data for this variable is available for fewer districts than for schools.

We regress the change in the fraction of villages with primary schools or health facilities (between 1951 and 1961) on the change in enfranchisement induced by the 1950 reform. We find a positive and statistically significant relationship between these two variables for education facilities (Table 4, Panel C, Column (3)), showing that increasing enfranchisement does result in better education provision, even though the reform does not increase measures of political competition. Note that this is not simply a continuation of pre-independence trends, since the areas with greater enfranchisement increases after 1950 are the ones that had lower increases in enfranchisement after the 1935 reform. Our post-independence results thus show a shifting of spending priorities across areas based on enfranchisement changes.

As for the colonial period, we find only a small and insignificant relationship with enfranchisement increases for health facilities (Table 4, Panel D). This is consistent with the idea that public health provision is unlikely to be differential across less and more enfranchised districts, since rich and poor alike benefit from programs such as malaria eradication or vaccination campaigns, and because private provision of such programs is much less feasible than private provision of education. Consistent with this low association with health facilities, we find that while there was an overall reduction in crude death rates between 1951 and 1961, this decline is not differential across districts with greater or lesser enfranchisement increases (Appendix Table A7, Panel C, Column (3)). A similar insignificant effect is obtained for the decline in death rates due to infectious diseases (Panel D), which accounted for more than half of the overall decline in death rates during this period.Footnote ²²