2.1 Politicians’ ability

In addition to ideological preferences, voters may care about differences in politicians’ abilities (Stokes, Reference Stokes1963; Dal Bó and Finan, Reference Dal Bó and Finan2018). In fact, multiple surveys and experiments among voters have shown that perceptions of a politician's abilities play an important role when voters choose who to vote for (McGraw, Reference McGraw2011; McAllister, Reference McAllister2016; Laustsen and Bor, Reference Laustsen and Bor2017). But what, exactly, is ability for a politician? Studies have frequently used politicians’ formal education as a proxy for their ability (Ferraz and Finan, Reference Ferraz and Finan2009; Besley and Reynal-Querol, Reference Besley and Reynal-Querol2011; Galasso and Nannicini, Reference Galasso and Nannicini2011; Kotakorpi and Poutvaara, Reference Kotakorpi and Poutvaara2011; Fisman et al., Reference Fisman, Harmon, Kamenica and Munk2015). However, results are not unequivocal. While Besley et al. (Reference Besley, Montalvo and Reynal-Querol2011) find a link between the educational level of politicians and their performance, Carnes and Lupu (Reference Carnes and Lupu2016) fail to find any stable link between education and performance. Although there is scholarly disagreement on the link between politicians’ education and performance in office, Danish voters perceive candidates with a university education as more competent than candidates without such an education and candidates with an unknown educational level (Pedersen et al., Reference Pedersen, Dahlgaard and Citi2019).

One caveat with using formal education is that it may be correlated with class background and unobservable ability, which may also make some people more likely to have political success or run for office regardless of their education. As an alternative strategy to measuring ability, recent studies, inspired by labor economics, have used Mincer earning regressions to estimate the income residual from an elaborate regression on income (Galasso and Nannicini, Reference Galasso and Nannicini2011; Besley et al., Reference Besley, Folke, Persson and Rickne2017; Dal Bó et al., Reference Dal Bó, Finan, Folke, Persson and Rickne2017; Bhusal et al., Reference Bhusal, Callen, Gulzar, Pande, Prillaman and Singhania2020). If the labor market value abilities not accounted for by the variables in the regression, we would expect people who are high in unobservable ability to have positive residuals as they would earn more than what we predict based on observable characteristics.

We can illustrate this with an example: say one needs to sell a home. They might not want just any real-estate agent, but the best real-estate agent. If so, they should be willing to pay a premium for a better real-estate agent. The better real-estate agent might be comparable to their competitors in the market on observable characteristics, but have a good, unobservable reputation because they have higher ability. This will materialize in higher income compared to their colleagues who are similar to them in terms of observable characteristics, which will give them a positive residual in an earnings regression.

Obviously, politicians’ ability is a multifaceted concept and it would be naive to think that formal education and residuals from income regressions on even a rich set of covariates such as ours captures all aspects of ability that voters may care about. However, Besley et al. (Reference Besley, Folke, Persson and Rickne2017) show that for Swedish politicians both measures correlate with political success, cognitive ability, and leadership tests from the military, as well as policy success. As we will see below, in our case, the results are also functionally the same regardless of whether we use ability measured as years of formal education or by the income measure. Still, we emphasize that these two measures are at best proxies for some aspect of ability that voters may care about.

2.2 Political representation

Voters can have preferences for candidates that are ideological congruent or responsive to their views (Hager and Hilbig, Reference Hager and Hilbig2020; Wolkenstein and Wratil, Reference Wolkenstein and Wratil2021), but they can also think of their politicians as representative along several other dimensions (Mansbridge, Reference Mansbridge2003; Rehfeld, Reference Rehfeld2009; Wolkenstein and Wratil, Reference Wolkenstein and Wratil2021). For our purpose here, we follow Dal Bó et al. (Reference Dal Bó, Finan, Folke, Persson and Rickne2017) and consider only descriptive representation with respect to economic background. That is the extent to which voters are represented by politicians who grew up in economic conditions similar to their own. While they are beyond the scope of our paper, other measures of descriptive representation, e.g., gender and ethnicity, may be equally important to the measure we use here. For now, we bracket the discussion of the strengths and limitations of using economically descriptive representation and revisit this question in the discussion.

3.1 Danish national elections

Danish national elections do not follow a fixed term. Instead they are called by the Prime Minister no more than four years after the latest election. Voter turnout is usually around 85 percent. In our time period, a total of eight elections were held in 1990, 1994, 1998, 2001, 2005, 2007, 2011, and 2015. The elections are proportional elections where candidates run in one of ten large districts, but the overall distributions of seats take place at a national level (Elklit, Reference Elklit2005). In the elections, 175 are elected as members of Parliament in Denmark while four additional members are elected in Greenland and the Faroe Islands. We focus on only those who run for election in Denmark and compare them with only the population in Denmark. The number of candidates varies across the elections from 799 to 1274. To run for national office, one must be 18 years old on election day, not be under guardianship, and be a Danish citizen with a permanent residence in Denmark. While it is possible to run as an independent candidate, the vast majority of political candidates run for a political party. The parties can choose to use open or semi-open lists. On open lists, the election of candidates from within a party are determined solely by the number of personal votes, while list position is also taken into consideration when candidates run on semi-open lists. Most Danish parties choose to use open lists and approximately half of the voters choose to vote for a specific candidate, while the rest casts a non-preferential vote for a party (Statistics Denmark, 2023). Here, Denmark differs from Sweden, where candidates are listed on semi-open lists. Accordingly, a candidate in Sweden is only elected ahead of their higher-ranked co-partisans if they receive at least 5 percent of the personal votes cast for her party, and the share of personal votes in Swedish national election is approximately half of that in Denmark (Buisseret et al., Reference Buisseret, Folke, Prato and Rickne2022; Sveriges Kommuner och Regioner, 2023).

3.2 Danish municipality elections

Denmark is divided into 98 municipalities. The municipalities are responsible for a broad portfolio of issues including some of the main service areas in the Danish welfare state: childcare, elder care, and public schools. The municipalities are also responsible for setting their own tax rate (with some national restrictions). Elections for the local councils are fixed term, held every four years. In our data, we have all the candidates in each election from 1993 to 2013. The elections are competitive and mostly dominated by the national parties, sometimes in competition with local party lists. They are also salient with the voters and the turnout rate usually fluctuates around 70 percent (Bhatti et al., Reference Bhatti, Dahlgaard, Hansen and Hansen2014).

To run for office in the local elections, one must be 18 years old on election day, not be under guardianship, and have a permanent residence in Denmark (rules regarding residency differ somewhat for citizens from Nordic countries, EU-countries, and other countries). In addition, a candidate is required to have permanent residence in the municipality where one wants to run. While Danish citizenship is not a requirement in local elections, we limit our sample to Danish citizens to ensure comparability across local and national elections. The municipalities vary in population size from less than 2000 to more than 500,000. The council boards have between 9 and 55 members. Regardless of the board size, most of the council members are part-time politicians, and their remuneration is not equivalent to a full-time salary. The notable exemption in all municipalities is the mayor who earns a salary well above the Danish median income. Similar to the national elections, the parties can choose to use open or semi-open lists at the municipal elections. Most parties also choose to use open lists at this level, and approximately 75 percent of the voters typically choose to vote for a specific candidate (Statistics Denmark, 2023). Again, Denmark differs on this point from Sweden, where semi-open lists are also used at the local level, and only 25–35 percent of the voters vote for a specific candidate.

3.3 The structural reform

In 2007, a major reform changed the local political landscape of Denmark. Denmark had 271 municipalities up to this point, the reform meant that 239 of these municipalities were merged into 66, while the remaining 32 municipalities continued to exist as previously.Footnote ³ The reform was first suggested in 2002 shortly after the center-right government had won the 2001 election. In 2002, a commission on the administrative structure was appointed. The commission published its report in January 2004, in which it suggested a grand administrative reform. The proposal found parliamentary support and it was soon made into law. As part of the reform municipalities were to be merged into larger units with populations of at least 30,000 people. In effect, that meant that smaller municipalities were asked to find other municipalities to merge with before January 1, 2005. Some small municipalities, mainly small, isolated islands were exempted from the population criteria. The reform meant that the average municipality size increased dramatically (Lassen and Serritzlew, Reference Lassen and Serritzlew2011). The reform was implemented politically in 2005 when the first elections took place in the coming 98 municipalities. In that election, voters in continuing municipalities elected members for the local council as usual. In the to-be-amalgamated municipalities, voters elected members for the board that would be responsible for the amalgamation and become the new local council when the reform was administratively implemented in 2007. In addition to increasing the average size of the municipalities, the reform also included a transfer of responsibilities to the municipalities, primarily from the county level (Lassen and Serritzlew, Reference Lassen and Serritzlew2011). This transfer of assignments was uniform across municipalities and thus while it may have increased the complexity of being a local government politician, it should have affected politicians in continuing and amalgamated municipalities equally.

While neither the reform nor the decision of which municipalities to merge with was random, we posit that it makes a good case for a difference-in-differences (DiD) comparison between the continuing and amalgamated municipalities. As we describe in further detail in Section 5, a number of previous studies have used a similar design to look at different impacts of the reform (Lassen and Serritzlew, Reference Lassen and Serritzlew2011; Blom-Hansen et al., Reference Blom-Hansen, Houlberg, Serritzlew and Treisman2016; Bhatti and Hansen, Reference Bhatti and Hansen2019).

The reform had a dramatic effect on the number of candidates as well as the number of elected politicians. In the three elections for which we have data before the reform, between 16,914 and 17,636 candidates ran for office. In 2005 when the reform was politically implemented, the number was 11,407, and in 2009 and 2013 when it had been administratively implemented, it had settled on 9049 and 9089, respectively. The number of elected candidates varied from 4647 to 4690 in the years before the reform and dropped to between 2444 and 2522 in the years following the reform.Footnote ⁴ Since the sizes of the underlying populations were unaffected by the reform, the decrease in the number of seats meant an increased competition measured in population per seat.

As a consequence of the reform, municipal politicians in Denmark are a more select group than municipal politicians in Sweden. While Sweden has a population which is approximately 75 percent larger than the Danish population, the number of municipalities is three times as large (290 versus 98), and the number of municipal politicians is approximately five times as large as in Denmark (Sveriges Kommuner och Regioner, 2023).

3.4 Danish administrative data

Everyone with a permanent residence in Denmark is assigned a civil registration number, and the central authority on statistics, Statistics Denmark, routinely collects and stores a comprehensive set of data for all Danes linked to this number. The information in the administrative data includes, but is not limited to, age, sex, complete residential history including current address, ethnicity, personal income, education, employment sector, and deidentified civil registration numbers of one's parents. Crucially, this information is not self-reported, which means that it is highly reliable and not subject to non-response bias. For our purpose here, we were granted access to information on everyone born after December 12, 1915.Footnote ⁵

We are able to link these data with our election data. Prior to the elections, the parties and candidates running as independents have to submit information to the election authorities, including each candidate's civil registration number. For the elections, Statistics Denmark have stored the list of running candidates including information on whether they were elected or not and for what party they ran. Linking the election data to the administrative data provides us with a comprehensive dataset on Danes born in 1915 or later. The population is cut off at people who had turned 75 in our data's first election in 1990. Some senior candidates still run for office at that age. However, considering the detailed and highly reliable dataset, we consider the age limitation on our data to be an insignificant limitation, and we point out that due to natural replacement of the electorate, the results approach full generalizability over time.

We use the deidentified personal registration number of parents to identify siblings for both the population and candidates.Footnote ⁶ Combining the data sources, we construct a comprehensive dataset where we observe the entire population of voting age in each election year, their parents and siblings, candidacy and election in elections, and a wide range of additional information including income, education, and parents’ income.

3.5 Measuring ability

To describe selection of politicians based on merit, we follow two different approaches to measuring ability. One simple approach is to proxy ability by years of formal education, which is often done in the literature (Baltrunaite et al., Reference Baltrunaite, Bello, Casarico and Profeta2014; De Paola and Scoppa, Reference De Paola and Scoppa2015; Fisman et al., Reference Fisman, Harmon, Kamenica and Munk2015). For this measure, we use educational categories. Second, we follow recent papers in the literature and use Mincer earning regressions to measure ability by estimating the income residual from an elaborate regression on income (Gulzar, Reference Gulzar2021). To estimate the earning score, we run a regression of total earnings on a set of predictors in each year for which we have data. Specifically, we estimate:

$$Y_{i, t} = \ f( age_{i, t},\; education_{i, t},\; sector_{i, t}) + \alpha_{\,j, t} + \epsilon_{i, t}$$

where Y _i,t is the income in year, t, for person, i.Footnote ⁷ To avoid bias from politicians who already receive a high income, we omit in each election politicians who were already elected in a previous election. We model age by a set of indicator variables for five-year age groups.Footnote ⁸ Education is measured in four main groups: primary, secondary, tertiary, or vocational education. For employment sector, we use a set of 19 sector categories from 2000 and later.Footnote ⁹ To maximize model flexibility, we include indicators for missing values on each of the variables and fully saturate the model with interactions between age group, education, and sector. To account for regional variability, we also include fixed effects, α _j,t, for municipalities.Footnote ¹⁰ Finally, we take into account that women, people with differing age, and non-native workers may experience different labor market outcomes. We therefore estimate the model individually for each of the 16 configurations of two binary variables, gender and being a native Dane,Footnote ¹¹ and four age groups: 18–32 years, 33–47 years, 47–61 years, and 62 years or older. For each year we fit the model without anyone who ever runs for office, before we estimate the individuals’ earning scores as the average income residual in the four years of the latest electoral cycle to diminish the impact of yearly fluctuations in income. In the online Appendix (Section 1.1), we present descriptive statistics. To be clear, this measure of ability does not assume that a member of a high-income group, e.g., a lawyer, has a higher ability level than a member of a group with lower income, e.g., a primary school teacher. Rather, the measure is based on the assumption that within each group, as defined by the variables above, higher earning indicates higher abilities. In other words, a school teacher with an income above median earnings in their own group will have a higher score on this measure than a lawyer with an income below median earning in their group, regardless of the fact that the teacher may still have a significantly lower income than the lawyer.

5.1 Does increased competition lead to increased ability?

To study the reform effect on politicians’ ability, we run a DiD model for two elections, the last before the reform and the first after. The model is of the form:

$$Y_{i, t} = \eta T_{\,j[ i] } + \gamma t_{i, t} + \tau_{\rm DiD}T_{\,j[ i] } t_{i, t} + \alpha_{\,j[ i] } + \epsilon_{i, t}$$

where Y _i,t is the outcome of interest for politician i, in election, t, T _j[i] is the treatment, i.e., being in a merged municipality, t _i,t is a post-reform indicator, and α _j[i] are municipality fixed effects for the post-reform structure. Finally, $\epsilon _{i, t}$ is an error term, which we cluster at the municipality level. τ _DiD is the quantity of interest as it measures the treatment effect of the reform.Footnote ¹³ The election in 2001 took place in the old municipalities. Elections in 2005 were for the future 98 municipalities, although the reform was not implemented before 2007. We skip the 2005 election and compare 2009 outcomes to 2001. As we saw above, the number of candidates did not fully settle to a new level in 2005. By moving to 2009, we make sure that the effect of the reform on the number of candidates was fully felt.

Before estimating the effect, we plot in Figure 7 the average of the income-based ability measure with sibling fixed effects for candidates and election winners in amalgamated and continuing municipalities for each election except for 2005 compared to 2001. The black points are for the municipalities affected by the reform, whereas the gray points are for the continuing municipalities. We include sibling fixed effects, because without these the parallel trends assumption does not seem to hold. In the online Appendix, we present a plot without sibling fixed effects. Looking at the pre-reform years for selection compared to siblings, we see no clear violation of the parallel trends assumption.

Figure 7. Selection on ability (mincer measure) with sibling fixed effects. The figure shows the average difference relative to 2001 on the ability measure based on income for candidates for city council and election winners for every election year except for 2005 in the data with 95 percent CIs. We omit the 2005 election, because it was the election that took place when the reform was known but not implemented. The plot is split by whether the municipality was affected by the reform (amalgamated) or not (continuing). The population average for fathers is zero, so values above zero indicate that there is some selection on ability. The dashed vertical line signifies when the reform was known.

Table 1 presents the DiD between 2001 and 2009 for candidates running for local offices and the subset that wins. For each outcome we estimate a model with and without sibling fixed effects. As noted, we prefer the specifications with sibling fixed effects as we doubt the parallel trends assumption holds without. In our preferred specification, we have fewer politicians, because we can only include politicians for whom we could identify a sibling.

Table 1. Effect of the reform on ability for candidates running for office

Notes: The table includes the DiD estimates obtained when comparing the difference for selection based on the ability measured using income residuals between candidates/election winners in 2009 post-reform amalgamated municipalities to post-reform continuing municipalities vis-à-vis the differences in 2001 pre-reform municipalities. Models (2) and (4) include sibling fixed effects. All models include post-reform municipality fixed effects. Standard errors clustered by the municipality in parenthesis.

The point estimates for the candidate pool running for office is 0.04 in our preferred specification with a 95 percent confidence interval (CI) of [−0.01; 0.09]. Since the outcome variable is normalized, the estimate is the change in standard deviations. For the elected candidates, the point estimates are generally larger at 0.15 standard deviations in our preferred specification with a 95 percent CI of [−0.06; 0.36]. While the point estimates are positive, both of the CIs include zero. In our alternative specification without sibling fixed effects, we also have positive point estimates both of which marginally exclude zero. However, we remind the reader that the parallel trends assumption was potentially violated in this case, and this was in a direction that could bias the DiD estimates upward and leave them entirely spurious. In an unpublished manuscript, Barfort et al. (Reference Barfort, Harmon, Lassen and Serritzlew2015) do find an effect of the reform on selection on ability using a design that resembles ours. However, when using our preferred specification, we find that the effect is not statistically significant.

5.2 Does increased competition lead to decreased representation?

Although we characterized the Danish system as inclusive, we did see some selection on the income of fathers prior to the studied time period. This was especially the case among the most elite politicians, members of parliament, where selection on ability is strongest and competition is fiercest. When competition increases at the local level, it may reduce the economic diversity of the candidates and electees. To address this question, we run a DiD model similar to the one, we saw above for selection on ability.

To see if the reform affected the economic diversity of the candidates/election winners, we look at the average position of fathers in the income distribution in 1985–1987. Specifically, we take as our dependent variable the income percentile, which the fathers belonged to when compared to all fathers living in the same municipality at that time and born in the same year. In Figure 8, we show the average income percentile for fathers of both candidates simply running for election and candidates winning election. The gray points are for the municipalities continuing after the reform, whereas the black points are for the amalgamated municipalities. An average of 0.5 implies that the average candidates or elected councilor is representative. Values above or below 0.5 mean that candidates or elected councilors are disproportionately from high- or low-income fathers. The average is always above 0.5, demonstrating that there is at least some consistent over-representation of better off fathers.

Figure 8. Average income percentile for fathers in 1985–1987. The figure shows the average income percentile that fathers belonged to in 1985–1987 relative to their municipality and birth year for candidates for city council and election winners for every election year in the data with 95 percent CIs. We average over three years to limit potential influence from fluctuations in one year. The plot is split by whether the municipality was affected by the reform (amalgamated) or not (continuing). The population average for fathers is 0.5, so values above 0.5 indicate that there is some selection on father's background. The dashed vertical line signifies when the reform was known.

From Figure 8 we also see that in every year, both candidates and city councilors in the continuing municipalities are less representative than in the municipalities affected by the reform. In Table 2, we focus on the DiD estimate for 2009 compared to 2001 to establish if the increased competition could have affected representation. We consider the amalgamation to be the treatment, and the average on the fathers’ income percentile is always above 0.5. Accordingly, a positive coefficient implies that amalgamated municipalities become less representative.

Table 2. DiD result on selection on the income of fathers (1985–1987) in 2009 compared to 2001

Notes: The table includes the DiD estimates obtained when comparing the difference for selection based on the income of fathers in 1985–1987 between candidates/election winners in 2009 post-reform amalgamated municipalities to post-reform continuing municipalities vis-à-vis the differences in 2001 pre-reform municipalities. Standard errors clustered by the municipality in parenthesis.

We do not find any evidence that the reform decreased the representation based on economic background of the candidates that voters could choose from: the estimated effect is 0.014 with a 95 percent CI of [−0.017; 0.044]. As the outcome variable is the fathers’ position in the income distribution, the point estimate means the fathers of the candidates were placed 1.3 percentiles higher in the distribution in the merged municipalities following the reform. However, the CI is wide relative to the estimate. For becoming elected, we estimate the effect to be negative. The estimate is −0.033 with a 95 percent CI of [−0.099; 0.033]. As there are considerably fewer councilors than candidates, we estimate the effect on councilors with much less precision.

5.3 Does increased competition change the ability–background trade-off?

Winning a seat includes multiple steps. We can identify at least four here: (i) One has to be willing to stand for office. (ii) The party has to be willing to put one on the list. (iii) One should ideally get a good position on the list. Although the Danish system is mostly open list with a minority of parties running candidates on semi-open lists, the parties do prioritize the candidates, and those at the top of the list are far more likely to win a seat. (iv) One should win enough personal votes. Voters assign personal votes, which finally determines the faith of the parties and candidates. Our results so far suggest that the reform did not necessarily make the output of steps (i) and (ii) less inclusive nor more selective on ability, and neither did steps (iii) and (iv). Our best estimate is that the system did not become any less economically inclusive or selective on ability as a consequence of the reform.

One question still needs to be addressed: did the reform affect the trade-off between selection on ability and background? We revisit the two indices that we created for Figure 6 where we looked at the ability–background trade-off. This is to determine whether the slope of this trade-off changed as a result of the reform. If the reform induced an ability–background trade-off, we should see that the slope becomes more positive. To explore this, we estimate the DiD of the slopes before and after the reform for the continuing and amalgamated municipalities. Specifically, for municipalities, indexed by m, in 2001 and 2009 we run a model of the form:

$$\eqalign{Y_{m} & = \alpha + \beta T_{m} + \gamma t_{m} + \delta rep_{m} + \zeta T_{m} t_{m} + \eta T_{m} rep_{m} + \theta t_{m} rep_{m} \cr & \quad + \tau_{\rm DiD} T_{m} t_{m} rep_{m} + \epsilon_{m}}$$

where Y _m is the outcome of interest, T _m is the treatment, i.e., being in a merged municipality, t _m is a post-reform indicator, rep _m is the representation index, and $\epsilon _{m}$ is an error term. τ _DiD is the quantity of interest as it measures how much the slope for rep _m changes post-reform in merged municipalities compared to the post-reform change in slope in continuing municipalities.

In Table 3, we present the DiD for the change in slopes between merged and continuing municipalities from 2001 to 2009.Footnote ¹⁴ We show the coefficients for both candidates and those elected for office. The coefficients for both groups are close to zero, both at 0.008 with 95 percent CIs of [−0.054; 0.069] and [−0.068; 0.085]. We also include the standardized coefficients, which may be easier to interpret. Our best estimate is that the reform meant the ability–background trade-off increased such that a 1σ increase in the representation index was associated with an additional 0.030σ ability–background trade-off in merged municipalities among candidates. Among councilors, our best estimate is that the reform meant the ability–background trade-off increased such that a 1σ increase in the representation index was associated with an additional 0.028σ ability–background trade-off in merged municipalities among candidates. Regardless of which of the two scales we interpret the coefficients on, the conclusion is the same: the reform did not substantially alter the ability–background trade-off.

Table 3. DiD for ability–background trade-off in 2009 versus 2001

Notes: The table includes the DiD estimates for the slopes of the ability–background trade-off obtained when comparing the slope for the candidates/election winners in 2009 post-reform amalgamated municipalities to the slope of post-reform continuing municipalities vis-à-vis the differences in slopes in the 2001 pre-reform municipalities. τ _DiD,beta is the standardized coefficient. Standard errors are in parenthesis.