Introduction
Over the past two decades, Western democracies have seen notable increases in ethnic minority legislators (Erzeel and Rashkova Reference Erzeel and Rashkova2023). This has prompted extensive research into these legislators’ roles and impacts, particularly regarding the relationship between descriptive and substantive representation following seminal works by Phillips (Reference Phillips1998) and Pitkin (Reference Pitkin1967). Studies consistently show minority legislators are more likely than majority counterparts to provide constituency services to minorities (Lowande et al Reference Lowande, Ritchie and Lauterbach2019), raise awareness of minority policy priorities (Saalfeld Reference Saalfeld2014), and support legislation advancing minority interests (Preuhs Reference Preuhs2006) and rights (Grose Reference Grose2005). However, as noted by Broockman (Reference Broockman2013), Crisp et al (Reference Crisp, Demirkaya, Schwindt-Bayer and Millian2018), and others, whether this positive association primarily stems from intrinsic motivation or electoral incentives remains unclear.
Increasing descriptive representation of minorities is considered key for reducing social and economic inequalities between dominant majorities and marginalized groups (Griffin Reference Griffin2014). Following Mansbridge (Reference Mansbridge1999), we define descriptive representation as the extent to which representatives resemble their constituents in social or demographic identities. By intrinsic motivation, we mean representatives’ genuine commitment to advancing the interests of groups with whom they share identity-based characteristics, independent of electoral calculations – contrasting with strategic motivation to secure minority votes. Theory suggests minority politicians are more likely to promote minorities’ interests – that is, substantively represent them – partly because shared experiences and social bonds create group solidarity (Williams Reference Williams2000). Such intrinsic motivation helps explain why minority representatives more often promote minority group welfare (Whitby Reference Whitby2000).
However, the positive correlation between descriptive and substantive representation may instead reflect strategic electoral motivations. Since minority politicians often represent constituencies with large minority populations, their focus on minority interests may be a rational electoral strategy. Consistent with this, Saalfeld and Bischof (Reference Saalfeld and Bischof2013) found all British MPs respond to minority constituents where these populations are concentrated. Similarly, Crisp et al (Reference Crisp, Demirkaya, Schwindt-Bayer and Millian2018) found that in New Zealand, Māori MPs are no more likely to represent their group than non-Māori MPs when institutional incentives for such behavior are absent. Bailer et al (Reference Bailer, Breunig, Giger and Wüst2022) provide further evidence of electoral motivation, showing that as minority MPs gain seniority, their focus on group-specific representation declines, shifting towards broader national policy areas.Footnote 1
If minority MPs’ actions reflect solely a rational response to electoral incentives, then perhaps descriptive representation is unnecessary for substantive representation (though it may still be important for symbolic reasons [Vishwanath, Reference Vishwanath2025]). Given the importance of establishing whether intrinsic motivations exist alongside electoral incentives for both theory and policy, it is no surprise that studies have attempted to demonstrate the existence of intrinsic motivations. Identifying such intrinsic motivations, however, is methodologically challenging because both explanations are generally observationally equivalent (Broockman Reference Broockman2013, 523–24). As we discuss below, past studies have struggled to overcome the problem of controlling for compositional effects of two types: the composition of places that elect minority representatives and the composition of characteristics of minority representatives.
Existing studies typically use regression frameworks that control for constituency characteristics when examining minority MP effects on substantive representation (for example, Canon Reference Canon2020; Grose Reference Grose2005; Juenke and Preuhs Reference Juenke and Preuhs2012). While Saalfeld (Reference Saalfeld2011) improves this approach by adding an interaction between a minority representative indicator and the minority population share, the fundamental identification challenge remains: these methods assume no unobservable confounding between minority descriptive representation and constituency demands for minority representation (for example, unobserved MP political ambition or constituency mobilization capacity), despite substantial observable differences between constituency types. Matching estimators (for example, Kopkin Reference Kopkin2017) also cannot address unobservable confounding.
An alternative and innovative approach by Crisp et al (Reference Crisp, Demirkaya, Schwindt-Bayer and Millian2018) exploits variation in New Zealand’s electoral systems, where Māori MPs face different levels of dependence on minority voter support. Yet this strategy cannot rule out that candidate selection and sorting across systems may correlate with underlying representational motivations.
Audit experiments address observational challenges by manipulating constituent identity, thereby controlling for constituent needs (for example, Lowande et al Reference Lowande, Ritchie and Lauterbach2019). Broockman (Reference Broockman2013) innovates by also varying constituency status, providing evidence for intrinsic motivation by showing black representatives are more responsive to black citizens even outside their constituencies. While valuable, audit experiments suffer from two limitations: first, they may not distinguish intrinsic motivation from statistical discrimination and in-group bias (see, for example, Costa Reference Costa2017 on generalized legislator in-group favoritism), and second, constituent service findings may not generalize to legislative behavior where MPs face different electoral and party pressures (see also Grossman and Slough Reference Grossman and Slough2022 on the limited scalability of constituent services).Footnote 2
We complement the existing literature with a regression discontinuity design (RDD) of close elections between white and minority candidates in the 2010–24 UK general elections. By comparing minority MPs who narrowly win against white candidates to white MPs who narrowly win against minority candidates at the victory cut-off, we ensure both groups face constituencies with similar characteristics, overcoming compositional effects and controlling for electoral incentives to identify intrinsic motivations.Footnote 3
The UK provides a valuable context to explore minority representation. As a parliamentary democracy with substantial immigration, the UK’s ethnic minority population rose from 14 per cent to 18.3 per cent between 2011 and 2021, prompting parties to field more minority candidates (Sobolewska Reference Sobolewska2013). From 2010 to 2024, 1,616 ethnic minority candidates contested elections across the UK (Appendix A). Minority MPs increased from 4 per cent (22 MPs) in 2010 to 14 per cent (90 MPs) in 2024. Combined with comprehensive parliamentary records and competitive elections, this enables RDD analysis examining substantive representation.
Our analysis of 600,000+ speeches and 700,000+ parliamentary questions classified using large language models (LLMs) shows narrowly elected minority MPs are more likely than white MPs to discuss issues important to ethnic minorities. Specifically, minority MPs allocate 6–7 percentage points (pp) more of their speeches to ethnic minority problems and rights (a 230–360 per cent increase), and are 17–20 pp more likely to take pro-immigration positions when discussing immigration (a near-tripling). These effects are robust and statistically significant at conventional levels. Analyses of parliamentary questions show consistent positive patterns, though not statistically significant. Additional tests confirm both groups face similar electoral incentives, constituency demographics, MP backgrounds, and overall parliamentary participation. Crucially, descriptive evidence suggests minority MPs face re-election penalties for representing minority issues – meaning they engage in this representation despite electoral costs rather than because of electoral benefits – providing additional support for an intrinsic motivation interpretation.
This article makes two key contributions. While previous RDD studies examined minorities’ representational outcomes (for example, Bhalotra et al Reference Bhalotra, Clots-Figueras, Cassan and Iyer2014; Hopkins and McCabe Reference Hopkins and McCabe2012; Kogan et al Reference Kogan, Lavertu and Peskowitz2021), ours is the first to systematically describe the inferential challenges in existing studies as arising from compositional effects – both of places that elect minorities (Saalfeld Reference Saalfeld2011) and candidate sorting (Crisp et al Reference Crisp, Demirkaya, Schwindt-Bayer and Millian2018) – and to use RDD to address these effects.
Second, we systematically measure minorities’ substantive representation through comprehensive parliamentary questions and speeches, going beyond roll-call votes (Geese et al Reference Geese, Goldbach and Saalfeld2015; Vishwanath, Reference Vishwanath2025) that may not reflect constituent concerns due to strong UK party discipline (Slapin et al Reference Slapin, Kirkland, Lazzaro, Leslie and O’Grady2018; Slapin and Kirkland Reference Slapin and Kirkland2020). Unlike audit studies focused on individual constituent services (Broockman Reference Broockman2013), our outcomes measure group-level representation. Additionally, our LLM classification of both speeches and questions provides more comprehensive measurement than studies using questions alone (Saalfeld and Bischof Reference Saalfeld and Bischof2013; Saalfeld Reference Saalfeld2011).
Data and Methods
Election Results, Candidate Ethnic Background, and Sample Selection
We assess representation in five UK Parliament sessions following the 2010, 2015, 2017, 2019, and 2024 general elections. Our sample includes constituencies where at least one minority candidate stands against a white candidate, yielding 1,053 MP Parliament instances from 2010 to 2024, of which 264 (25 per cent) are represented by minority MPs. To construct this sample and compute the RDD independent and running variables (detailed further in section Estimation Method), we use data on UK general election results from the Electoral CommissionFootnote 4 and data on 2010–2019 candidates’ ethnic backgrounds collected by Zonszein and Grossman (Reference Zonszein and Grossman2024). We classify 2024 candidates’ ethnic groups based on predicted nationalities from their names using OnoGraph.
Parliamentary Activities: Speeches and Questions for Written Answer
Parliamentary speeches and questions for written answer (PQs) effectively capture substantive representation (Bird Reference Bird2005; Kroeber Reference Kroeber2018; Saalfeld and Bischof Reference Saalfeld and Bischof2013). These activities influence policy at scale by affecting constituents and other MPs (Blumenau Reference Blumenau2021; Dietrich et al Reference Dietrich, Hayes and O’Brien2019), and better reveal MPs’ commitment to group-specific interests than divisions – roll-call votes – as they face fewer party constraints (Proksch and Slapin Reference Proksch and Slapin2012).
However, speeches and PQs differ in their value as signals of intrinsic motivation. PQ submission is nearly unlimited, so MPs may submit PQs without spending much political capital. Existing studies analyze only PQs (for example, Bailer et al Reference Bailer, Breunig, Giger and Wüst2022; Saalfeld and Bischof Reference Saalfeld and Bischof2013; Saalfeld Reference Saalfeld2011). Therefore, while we include PQs, we center our analysis on parliamentary speeches. Speaking opportunities are restricted and allocated by the Speaker (Slapin et al Reference Slapin, Kirkland, Lazzaro, Leslie and O’Grady2018), meaning MPs must compete for limited time. This scarcity makes speech choices particularly revealing of representational priorities, helping distinguish intrinsic motivation from credit-claiming.
We web-scraped MP speeches and PQs from TheyWorkForYou (Appendix D). Our analysis focuses on the first 22 months following each election – the maximum period between 2017 and 2019, ensuring equal weight across sessions. This yields 142,750 speeches and 148,092 PQs from 1,053 MP Parliament instances (Appendix B).
Parliamentary Activities: Defining Minority Interests
Following existing literature, we focus on two issues relevant to ethnic minority voters: the problems and rights of ethnic minorities in the UK (including discrimination, equal opportunities, and bias in policing) and immigration. We examine both issue salience and positions, breaking immigration into inclusionary and exclusionary positions to extend existing salience-focused studies (for example, Saalfeld and Bischof Reference Saalfeld and Bischof2013). Appendix C presents examples.
These are the two issues covered by Saalfeld and Bischof (Reference Saalfeld and Bischof2013). Since then, ethnic minority population increased from 14 to 18.3 per cent and minority MPs more than quadrupled, making it valuable to revisit substantive representation. Moreover, ethnic minority voters prioritize anti-discrimination policy (Heath et al Reference Heath, Fisher, Rosenblatt, Sanders and Sobolewska2013) and are more supportive of immigration (Martin and Blinder Reference Martin and Blinder2021), underscoring these domains’ centrality.
Parliamentary Activities: Classification and Outcome Measures
To classify speeches and PQs into issues, we used OpenAI’s GPT-3.5 after creating a gold standard from almost 1,500 manually coded excerpts. Our team classified each excerpt into relevant issues and determined constituency references. We prompted GPT-3.5 with classification instructions and refined prompts to achieve high accuracy and F1 scores against hand-coded benchmarks. Immigration speeches were further classified into anti-immigration, pro-immigration, and neutral stances using similar methods. Appendix E provides further details on this classification process and performance metrics. This automated classification enabled large-scale analysis that would have been prohibitively costly with human coding alone and outperformed dictionary approaches and trained text classification models (see, for example, Mellon et al Reference Mellon, Bailey, Scott, Breckwoldt, Miori and Schmedeman2024; Rathje et al Reference Rathje, Mirea, Sucholutsky, Marjieh, Robertson and Van Bavel2024). LLM classification captures nuanced discussion that keyword searches would miss while maintaining consistency across over 1 million parliamentary actions.
For each MP Parliament instance, we measure speech on relevant issues as the proportion of an MP’s total speeches discussing each issue. This scaling accounts for differences in parliamentary activity due to expertise, seniority, or institutional positions (see for example, Blumenau and Damiani Reference Blumenau, Damiani, Back, Debus and Fernandes2021). Similarly, we measure PQs as the count of issue-relevant questions divided by months not serving as government minister (since ministers cannot submit PQs). We extract ministerial data from Wikipedia infoboxes, which complements Cagé and Dewitte (Reference Cagé and Dewitte2025)’s data. Following existing studies (for example, Saalfeld Reference Saalfeld2011), we do not scale PQs by total questions submitted since PQ submission is unlimited.
We acknowledge all MP actions can be strategic. However, different actions face different constraints: while votes face strong party whipping, speeches enjoy greater autonomy and capture individual MP positions even when rebelling against party lines (Slapin and Kirkland Reference Slapin and Kirkland2020). Speeches are also more costly signals than PQs due to restricted speaking opportunities. Though imperfect, speeches represent the most feasible observable measure of intrinsic motivation in parliamentary settings.
To validate our RD design, we measure MPs’ overall parliamentary activity: speeches, written questions, oral questions, petitions, and division participation. These variables help rule out compositional effects by testing whether white and minority MPs differ in overall parliamentary productivity. This addresses both the general concern from observational studies that minority MPs may be systematically different, and Marshall (Reference Marshall2024)’s specific RDD concern about compensating differentials – where marginally winning minority MPs might be superior in other characteristics. The section Compositional Effects finds no evidence of such differentials.
Estimation Method
Our challenge is distinguishing intrinsic motivation from electoral incentives. Minority MPs typically represent constituencies with larger minority populations, confounding MP ethnicity with electoral pressures. Using a sharp RDD addresses this by comparing minority MPs who narrowly won against white candidates to white MPs who narrowly won against minority candidates at the victory cut-off. Close elections quasi-randomly assign MP ethnicity while holding constant constituency characteristics. Narrow margins are crucial because, on average, constituencies in landslide elections differ substantially, while those in competitive elections are remarkably similar (Appendix F.2). By holding constant constituency factors, the RDD isolates MP ethnicity – revealing intrinsic motivations.
We estimate the following weighted linear regression:
$${Y_{it}} = \alpha + {\beta _1}{\rm{VictoryMargin}}{_{it}} + \tau {\rm{EthnicMinorityVictory}}{_{it}} \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\\ + {\beta _2}({\rm{VictoryMargin}}{_{it}} \times {\rm{EthnicMinorityVictory}}{_{it}}) + {\beta^\prime} {X_{it}} + {\epsilon_{it}}$$
where Y it is MP i’s parliamentary action in Parliament t. VictoryMargin it is the running variable measuring the ethnic minority candidate’s winning margin. Following existing multiparty studies (for example, Klašnja and Titiunik Reference Klašnja and Titiunik2017), we define this as the vote share of the strongest ethnic minority candidate minus the strongest white opponent’s vote share. This excludes constituencies with only white or minority candidates (sample characteristics in Appendix F.3). EthnicMinorityVictory it is a binary treatment indicating minority candidate victory (I(VictoryMargin it > 0) = 1). We include predetermined covariates X it (detailed in Appendix F.1) to reduce residual variation and standard errors. All tables report estimates with and without controls using robust bias-corrected inferences.
We estimate this equation using local polynomial methods via the rdrobust package, which fits linear regressions above and below the victory cut-off within an optimal bandwidth. This optimal bandwidth defines ‘narrow elections’ and is selected with a data-driven algorithm minimizing the mean squared error (MSE) of the RD estimate (Calonico et al Reference Calonico, Cattaneo and Titiunik2014). Observations are weighted by their distance from the cut-off using a triangular kernel, with observations outside the optimal bandwidth receiving zero weight. As a result, the effective number of observations used for estimation varies across outcomes depending on the bandwidth selected, ranging from 136 to 258 MP Parliament instances in our main specifications.
Assuming continuity of potential outcomes at the cut-off, τ̂ captures the average treatment effect of descriptive representation on substantive representation of ethnic minorities at the cut-off. The section Compositional Effects and Appendix J.2 show compensating differentials are not a major concern. Table 1 presents summary statistics for outcomes, treatment, and controls.
Summary statistics
Table 1. Long description
A table summarizing statistics on various political and demographic metrics. The table has 19 rows and 6 columns. The columns are labeled Mean, SD, Min, Max, and N. The rows are labeled with different metrics such as % Immigration speech, % Discrimination speech, % Pro-immigration speech, % Neutral/anti-immigration speech, Num. immigration PQs, Num. discrimination PQs, Minority victory margin, Minority win, MP: incumbent, MP: Labour, MP: Conservative, MP: Labour minority, MP: Conservative minority, % Ethnic minority, Pop. density, % Employed, % High econ deprivation, and % Social housing. Each row provides the mean, standard deviation, minimum, maximum, and number of observations for each metric. For example, the % Immigration speech has a mean of 0.08, a standard deviation of 0.09, a minimum of 0.00, a maximum of 0.61, and 1,053 observations. The % Discrimination speech has a mean of 0.07, a standard deviation of 0.09, a minimum of 0.00, a maximum of 0.83, and 1,053 observations. The % Pro-immigration speech has a mean of 0.24, a standard deviation of 0.27, a minimum of 0.00, a maximum of 1.00, and 901 observations. The % Neutral/anti-immigration speech has a mean of 0.75, a standard deviation of 0.27, a minimum of 0.00, a maximum of 1.00, and 901 observations. The Num. immigration PQs has a mean of 0.41, a standard deviation of 1.06, a minimum of 0.00, a maximum of 17.21, and 1,053 observations. The Num. discrimination PQs has a mean of 0.21, a standard deviation of 0.48, a minimum of 0.00, a maximum of 5.83, and 1,053 observations. The Minority victory margin has a mean of -19.86, a standard deviation of 33.04, a minimum of -82.05, a maximum of 80.64, and 1,053 observations. The Minority win has a mean of 0.25, a standard deviation of 0.43, a minimum of 0.00, a maximum of 1.00, and 1,053 observations. The MP: incumbent has a mean of 0.66, a standard deviation of 0.47, a minimum of 0.00, a maximum of 1.00, and 1,053 observations. The MP: Labour has a mean of 0.52, a standard deviation of 0.50, a minimum of 0.00, a maximum of 1.00, and 1,053 observations. The MP: Conservative has a mean of 0.34, a standard deviation of 0.47, a minimum of 0.00, a maximum of 1.00, and 1,053 observations. The MP: Labour minority has a mean of 0.14, a standard deviation of 0.35, a minimum of 0.00, a maximum of 1.00, and 1,053 observations. The MP: Conservative minority has a mean of 0.08, a standard deviation of 0.28, a minimum of 0.00, a maximum of 1.00, and 1,053 observations. The % Ethnic minority has a mean of 22.01, a standard deviation of 19.63, a minimum of 1.00, a maximum of 79.90, and 1,053 observations. The Pop. density has a mean of 31.70, a standard deviation of 33.08, a minimum of 0.11, a maximum of 197.58, and 1,053 observations. The % Employed has a mean of 59.93, a standard deviation of 6.11, a minimum of 40.90, a maximum of 74.60, and 1,053 observations. The % High econ deprivation has a mean of 5.08, a standard deviation of 2.33, a minimum of 1.20, a maximum of 14.16, and 1,053 observations. The % Social housing has a mean of 19.14, a standard deviation of 8.40, a minimum of 5.30, a maximum of 50.60, and 1,053 observations.
Note: N = MP Parliament instances. Speech measures are share of MP total speeches; PQs are share of non-ministerial months.
Results: Substantive Representation
Figure 1 presents regression discontinuity estimates of electing an ethnic minority MP. The top panels examine issue salience (proportion of speeches devoted to specific topics), while the bottom panels examine positions on immigration. As a constituency moves from barely represented by a white MP to barely represented by an ethnic minority MP, we observe increases at the victory cut-off in attention to minority issues. In each panel, the two lines represent local linear regressions fitted separately on each side of the victory cut-off; the RD effect estimates reported in Table 2 measure the vertical distance between the points where these two lines meet the cut-off.
Substantive representation in parliamentary speeches: issue salience and positions.
Note: lines represent the average outcome (with 95 per cent confidence intervals) from local linear regression fitted to the sample of constituencies with minority victory margins within the MSE-optimal symmetric bandwidth around the victory cut-off. Points are the average outcome for equally spaced mimicking-variance bins.
Substantive representation: parliamentary speeches
Table 2. Long description
The table is divided into two main panels: Panel A and Panel B. Panel A focuses on issue salience, while Panel B focuses on issue positions. Each panel is further divided into columns representing different aspects of the issues. Panel A has four columns: Problems and rights of ethnic minorities in the UK, Immigration, Pro-immigration, and Neutral/Anti-immigration. Panel B also has four columns with the same headings. Each column contains several rows with specific data points. The rows include indicators such as VictoryMargin, Mean control, Percentage change, R squared, Effective N (within bandwidth), N, MSE-opt bw, and Controls. The values in the table provide insights into the substantive representation of these issues in parliamentary speeches. For example, in Panel A, the first column under Problems and rights of ethnic minorities in the UK shows a VictoryMargin value of 0.073 with a confidence interval of [0.024, 0.145]. The table provides a detailed comparison of how different issues are addressed in parliamentary speeches, highlighting the salience and positions on ethnic minority rights and immigration.
Note: dependent variable is speeches on issue divided by total MP speeches (Panel A), or pro/anti-immigration speeches divided by total immigration speeches (Panel B). I(VictoryMargin it > 0) = 1 is treatment effect at cut-off using local linear regression with triangular kernel and MSE-optimal bandwidth. Brackets show robust bias-corrected 95% CI. Standard errors in parentheses. Mean control: average for narrowly elected white MPs. Percentage change: effect as percentage of white MP mean. Eff. N (within bandwidth): observations within bandwidth; N: total observations. MSE-opt bw: optimal bandwidth size. Controls: constituency ethnic minority share, employment, deprivation, social housing, density; MP incumbency and party. Panel B excludes MPs with no immigration speeches. * * * p < 0.01; **p < 0.05; *p < 0.1.
Table 2 quantifies these effects: narrowly elected minority MPs allocate 5.7–7.3 pp more of their speeches to ethnic minority problems and rights (columns 1–2, Panel A) – a 229–363 per cent increase, statistically significant at the 1 per cent level. Immigration speeches show positive (2.7–3.3 pp) but not statistically significant effects (columns 3–4, Panel A).
White and minority MPs, however, differ in positions taken on immigration issues – consistent with minority voters’ pro-immigration preferences (Martin and Blinder Reference Martin and Blinder2021) (Appendix G). Minority MPs are 16.8–20 pp more likely to take pro-immigration positions (Panel B, columns 1–2). Given white MPs’ 11 per cent rate, minority MPs are nearly three times as likely to voice inclusionary positions. Appendix H shows consistent positive effects for parliamentary questions though not statistically significant, and Appendix I examines other issues.
Validity Tests and Robustness Checks
RD estimates require that candidates cannot manipulate election results around the cut-off. We conduct density tests (showing similar observation counts above and below the cut-off) and covariate tests (showing null RD effects on predetermined variables). Additional robustness checks include sensitivity tests for bandwidth choice and polynomial order, placebo cut-offs, and RD estimates with Armstrong and Kolesár (Reference Armstrong and Kolesár2020)’s bias-aware inference, and also under local randomization. Appendix J presents these tests, all of which support the validity of our RD design.
Our substantive representation results are not driven by party differences for three reasons. First, our model controls for MP party. Second, there is no discontinuity in MPs’ parties at the cut-off (Appendix Figure J.2). Third, RD estimates are similar across Conservatives and left-leaning parties (Appendix K).Footnote 5 While constrained by limited statistical power, this finding suggests minority MP effects reflect more than partisan differences.
Discussion: Minority MPs’ Intrinsic Motivation
The section Results: Substantive Representation shows increased ethnic minority descriptive representation has increased substantive representation. To interpret this as intrinsic motivation, we must rule out: (1) positive selection whereby minority MPs are more productive overall, and (2) differences in electoral incentives. Under intrinsic motivation, we expect null results – no differences in productivity or electoral incentives at the cut-off.
Compositional Effects
We test for positive selection by comparing parliamentary engagement across five metrics. If narrowly elected minority MPs were systematically more productive – a compensating differential (Marshall Reference Marshall2024) – we would observe discontinuities. Instead, we expect null effects. Table 3 shows narrowly elected minority and white MPs have similar parliamentary engagement – estimates are substantively small and statistically insignificant. Appendix J.2 confirms narrowly elected minority and white MPs are similar in partisan affiliation, incumbency, education, political background, and demographic characteristics, and represent constituencies with similar socio-demographic profiles. These results suggest our RDD mitigates compositional effects and avoids bias from compensating differentials (Marshall Reference Marshall2024). However, MPs could still differ on unobserved characteristics, making RD estimates the bundled effect of ethnicity and these correlated characteristics.
MP parliamentary participation
Table 3. Long description
The table compares parliamentary engagement across five metrics for narrowly elected minority and white MPs. It includes six columns labeled Petitions, Divisions, Speeches, Oral questions, and Written questions, with sub-columns for different conditions. The rows are labeled with indicators, mean control values, effective sample sizes, total sample sizes, bandwidths, and controls. Panel A focuses on Petitions, Divisions, and Speeches, while Panel B focuses on Oral questions and Written questions. Each cell contains numerical values representing the engagement metrics. The table shows detailed statistical data, including confidence intervals and mean values, for each metric under different conditions. The data is presented with specific values for each category, providing a comprehensive comparison of parliamentary engagement.
Note: dependent variables are MP activities per parliamentary day (petitions, speeches, oral questions), division attendance rate, and written questions per month not serving as minister. I(VictoryMargin it > 0) = 1 is treatment effect at cut-off using local linear regression with triangular kernel and MSE-optimal bandwidth. Brackets show robust bias-corrected 95% CI. Standard errors in parentheses. Mean control: average for narrowly elected white MPs. Eff. N (within bandwidth): observations within bandwidth; N total observations. MSE-opt bw: optimal bandwidth size. Controls: constituency ethnic minority share, employment, deprivation, social housing, density; MP incumbency and party. ***p < 0.01; **p < 0.05; *p < 0.1.
Electoral Incentives
We test whether narrowly elected MPs face different electoral incentives using three measures: constituency ethnic minority population share (2011, 2021 Census), minority turnout, and minority vote share contribution (British Election Study).Footnote 6 Under intrinsic motivation, we expect null effects; differences would suggest strategic responses.
Table 4 shows no discontinuities across all measures (point estimates are substantively small and statistically insignificant), confirming narrowly elected MPs face similar electoral incentives. This is consistent with an intrinsic motivation interpretation of our substantive representation findings.
Electoral incentives: ethnic composition and voting behavior
Table 4. Long description
A table comparing ethnic composition and voting behavior. The table has six columns and seven rows, including headers. The columns are labeled as follows: I(VictoryMargin > 0) = 1, Minority population, Minority turnout, Minority contribution, Mean control, and E ff. N (within bandwidth). The rows are labeled with values and confidence intervals. Row 1: I(VictoryMargin > 0) = 1, 6.973, -1.048, 0.037, 0.034, -0.002, -0.022. Row 2: Mean control, 16.62, 17.869, 0.744, 0.745, 0.044, 0.078. Row 3: R & 00302; 2, 0.014, 0.551, 0.019, 0.038, 0.03, 0.674. Row 4: E ff. N (within bandwidth), 163, 274, 163, 165, 153, 236. Row 5: N, 1053, 1053, 796, 796, 791, 791. Row 6: MSE-opt bw, 15.295, 23.19, 19.304, 19.486, 18.662, 27.16. Row 7: Controls, N, Y, N, Y, N, Y. The table presents data on the relationship between victory margin, minority population, minority turnout, minority contribution, and control variables. Each row provides specific values and confidence intervals for these variables.
Note: dependent variables are constituency ethnic minority population share (columns 1–2), minority turnout (columns 3–4), and minority vote contribution to winner (columns 5–6). I(VictoryMargin it > 0) = 1 is treatment effect at cut-off using local linear regression with triangular kernel and MSE-optimal bandwidth. Brackets show robust bias-corrected 95% CI. Standard errors in parentheses. Mean control: average for narrowly elected white MPs. Eff. N (within bandwidth): observations within bandwidth; N total observations. MSE-opt bw: optimal bandwidth size. Controls: constituency ethnic minority share, employment, deprivation, social housing, density; MP incumbency and party. ***p < 0.01; **p < 0.05; *p < 0.1.
Minority MPs might still strategically prioritize minority voters if this ensures re-election (Broockman Reference Broockman2013). However, within the optimal bandwidth, ethnic minorities comprise 23 per cent of the population while winners need 45–47 per cent of votes – making minority support insufficient for victory even with bloc voting. This differs from US majority–minority districts. Conversely, narrowly elected white MPs also need minority voters, as their co-partisan white electorate (∼38 per cent) is insufficient. Therefore, differences may reflect, in part, minority MPs’ intrinsic motivations.
Appendix L provides descriptive evidence consistent with this interpretation. Among narrowly elected MPs, minority MPs who speak more frequently about ethnic minority rights or are pro-immigration have substantially lower re-election rates (38 vs. 70 per cent for minority rights; 39 vs. 58 per cent for pro-immigration), while white MPs face no such penalties for discussing these issues (26 vs. 30 per cent; 27 vs. 30 per cent). This asymmetric pattern suggests minority MPs prioritize minority representation despite – not because of – electoral consequences, further supporting that their behavior is not purely strategic vote-seeking.
Conclusion
This study provides causal evidence that ethnic minorities’ accession to parliament increases substantive representation. Ethnic minority MPs focus disproportionately on issues minority voters prioritize: discrimination, equal opportunities, and inclusionary immigration policy.
Additional evidence shows our RD design controls for electoral incentives while minority MPs face electoral penalties, suggesting intrinsic motivation is a driver of their behavior. This complements attitudinal research showing minority candidates feel greater responsibility to represent ethnic minorities (Sobolewska et al Reference Sobolewska, McKee and Campbell2018), using behavioral evidence from a causal design.
This study has limitations. First, sample size constraints limit statistical power for heterogeneity analyses by ethnic group, party, or region, as only a subset of minority candidates compete in races close enough for RDD analysis. Second, our measures rely on observable parliamentary actions (speeches and questions), which inevitably contain strategic elements alongside intrinsic motivations. While speeches are more costly signals than questions due to restricted speaking opportunities, no measure perfectly captures unobservable motivations. Third, RDD identifies local average treatment effects at competitive election thresholds. Finally, as a single-country study, findings likely generalize best to democracies with similar electoral systems (first-past-the-post) and immigration histories, though further research across diverse contexts would strengthen these conclusions.
Our findings suggest descriptive representation delivers substantive democratic benefits beyond symbolic value. As democracies diversify, understanding these dynamics becomes crucial for institutional design promoting both inclusion and effective governance.
Supplementary material
Supplementary material for this article can be found at https://doi.org/10.1017/S0007123426101690.
Data availability statement
Replication data and code for this paper can be found at https://doi.org/10.7910/DVN/ZCMRZM.
Acknowledgements
We thank August Crane and Arda Enfiyeci for excellent research assistance, and Edgard Dewitte and Julia Cagé for kindly sharing the dataset on descriptive characteristics of the candidates in the UK general elections. We also thank Erik Bleich, Irma Clots-Figueras, Olle Folke, and Maria Sobolewska, and audience members at the NICEP 2024, the EBRD-KCL Migration Workshop 2024, the CES 2024, and the CP/IR Workshop of the USC Center for International Studies for helpful comments.
Financial support
This research received no specific grant from any funding agency, commercial or not-for-profit sectors.
Competing interests
None to disclose.
Open access
