The Selection of Top Politicians: A Natural Experiment in UK Parliament

Transcription

1 The Selection of Top Politicians: A Natural Experiment in UK Parliament Matt Lowe First Version: May 1, 2015 Current Version: December 10, 2017 Abstract Senior national politicians hold wide-ranging powers to direct policy-making and legislation, yet little is known about the nature of their selection: is the selection process efficient? Is there path-dependence in political careers after early successes? I exploit a natural randomized experiment in the UK Parliament to shed light on these questions. Each year, hundreds of Members of Parliament (MPs) enter a lottery for the opportunity to legislate. Using archival data from 1950 to 1990 I find that high-ranked winners are 34% (8 p.p.) more likely to ever become ministers and hold 28% (0.4) more political offices over their careers. Three pieces of evidence suggest that the key mechanism is exposure, as opposed to learning-by-doing or political survival. First, the effect of winning is larger for women, an under-represented group for which priors are likely to be more diffuse. Second, the effect is smaller if there are randomly more winners from the same party in the same year, dividing the attention of senior party members. Third, the effect is smaller when the MP has won before, consistent with diminishing returns to signals. These results suggest that early exposure can have long-run career effects even in information-rich political settings. Department of Economics, MIT. mlowe@mit.edu. Many thanks to Daron Acemoglu, Abhijit Banerjee, Esther Duflo, John Firth, Donghee Jo, Horacio Larreguy, John Marshall, Rachael Meager, Ben Olken, Arianna Ornaghi, Arthur Spirling, Tavneet Suri, and participants at the MIT and Harvard Political Economy Lunches for helpful comments and suggestions. Thanks especially to Paul Lester (House of Commons Library), Richard Ward, Sally Rushton (Parliamentary Assistant to George Osborne MP), and Jean Fessey (HC Information Office) for extraordinary assistance with the Parliamentary Archives, Prof. Michael Rush for sharing the MP biography archive he compiled, and Oset Babur and Micah Villarreal for research assistance. All errors are my own. 1

2 1 Introduction In any representative democracy, many of the most powerful political positions are filled not by a vote of the general population, but rather by internal party processes. An initial process of selection, for example a general election, determines which citizens become politicians. A second process of selection determines which politicians become senior decision-makers party leaders, government ministers, and Prime Ministers. In the United Kingdom (UK), and elsewhere, there are fundamental reasons why the two layers of selection may differ. In the former, the general population vote. Each voter has a minuscule chance of affecting the outcome, little knowledge on average of the inner workings of politics, and strong party preferences. In contrast, the selection of top politicians is characterized by direct influence over the outcome (particularly in the case of party leaders directly selecting ministers), personal relationships with those being selected, and the choice is largely within-party, negating the role of party preferences. Both layers of selection are important to understand, but it is the second that has been most understudied. Given that it is senior politicians that wield the most power to affect society, it is important to explore the nature of their selection. I exploit a natural randomized experiment in the UK House of Commons (the lower house) to explore whether early successes have a causal effect on long-run careers. Each year junior politicians, those without positions in government, can enter a ballot for bills the Private Members Bill ballot. Several hundred enter every year, with 20 names randomly chosen as winners, and ranked in random order. Winners are given the opportunity to table legislation with a reasonable probability of success. I digitized archival data on the winners and losers during 1950 to 1990 to explore issues of political selection. My analysis can be divided into three parts. In the first part I answer the narrow question: do winners experience long-term career gains? I find that high-ranked winners are 34% (8 p.p.) more likely than losers to ever become government ministers, and they hold 28% more political offices over their careers. Low-ranked winners in contrast experience null, or even negative effects, relative to losers. Second, to understand the heterogeneous effects by winning rank, I collated data on the characteristics of each Private Members Bill: the number of co-sponsors, whether the bill was successfully passed, and the number of total citations. Higher-ranked winners have more co-sponsors for their bills, are more likely to have their bills passed, and receive more citations to their bills. Since the ballot rank is itself random, these differences are not a consequence of selection. Instead, they reflect institutional features of UK Parliament: higher-ranked winners can table their bills earlier, increasing the probability of their passage, and ultimately increasing the attention their bills receive. 2

3 In the third part of the paper, I propose several possible mechanisms for the long-term career effects of winning the ballot with a high rank. First, winning the ballot may affect political survival winners can signal to their constituents that they are representing well their interests by proposing a bill specific to constituents concerns. Constituents then may reward this in elections, ensuring that the ballot winner remains in Parliament long enough to advance to more senior positions. Second, there may be learning-by-doing politicians develop skills through the on-the-job training of legislating that they would not have developed in the absence of the legislative opportunity. Third, there may be an exposure channel. Winning the ballot provides a means for politicians to make their names known throughout the party, increasing their chances of future promotion. More formally, winning provides an opportunity for senior party members to learn about the winner s underlying quality, such that their posterior beliefs are more precise. Assuming that party leaders incur losses from risk, they should promote those for which they have more precise beliefs, other things equal. These three mechanisms have different efficiency implications. The exposure channel suggests that there is a lack of information in Parliament, with party leaders uncertain about the underlying quality of party members. In this case, party leaders can make mistakes by promoting members that are not of the highest quality. In contrast, the other two channels are more consistent with efficient political selection, with the highest quality politicians promoted. In particular, the political survival channel would suggest that winners are more likely to be promoted merely because they are still in Parliament among the pool of potential candidates. The learning-by-doing channel would suggest that winners are more likely to be promoted because their quality is endogenously higher because of the legislative opportunity. Additional empirical tests suggest that the political survival and learning-by-doing channels are unlikely to explain the career effects. First, though high-ranked winners remain in Parliament for 1.7 more years, they are no more likely to have a higher vote share in the election immediately after winning. This result is counter to the claim that politicians use the ballot to connect with constituents. Second, there is some evidence that low-ranked winners experience negative career effects, inconsistent with the learning-by-doing channel, which should imply weakly positive effects for all winners. In contrast, three pieces of evidence support the exposure channel. First, the career effects of winning the ballot are significantly higher for women than for men, consistent with a learning model in which party leaders have more diffuse priors about under-represented groups. Second, career effects are larger when an MP wins in a year when there are fewer winners (randomly) from the same party, giving the MP greater attention from senior party members. Third, the effect of 3

4 winning the ballot is smaller when the MP has already won before, consistent with diminishing returns to signals. Taken together, these findings show that early successes can have long-term career effects, providing evidence for path-dependence in careers. The primary channel for these effects appears to be exposure. It follows that for the case of senior politicians, the selection process may be inefficient, with high-quality candidates overlooked as a consequence of imperfect information. A priori, it is not obvious that this mechanism should be important in a place like UK Parliament we might expect that the repeated daily interaction of politicians at work should ensure that the information environment is sufficiently rich for additional learning to have small effects. The results in this paper suggest that exposure and learning may be important channels of long-term career success even in supposedly information-rich environments. This paper relates to several strands of existing work. Theoretically, our starting point is that individuals are limited in their ability to commit credibly to certain policy positions (Besley and Coate (1997), Osborne and Slivinski (1996), Chattopadhyay and Duflo (2004)) politicians are personalities with at least partially fixed ideological positions, giving the processes of political selection substantive implications. On this basis, Besley (2005) argued for the importance of work on political selection. Since then, there have been several examples of careful empirical work on political selection. Ferraz and Finan (2008) use random variation in municipal audits in Brazil to explore the question of whether voters hold officials accountable for corrupt practices, and Ferraz and Finan (2011) use non-linear salary caps on wages of local legislators to explore the effect of salary on selection. 1 This work has largely focused on the selection of politicians from among citizens rather than the selection of senior decision-makers from the pool of existing politicians. The latter may be especially important given evidence that country leaders themselves matter for policies and welfare (Jones and Olken (2005), Besley et al. (2011)). Existing work on the selection of senior politicians is more descriptive, with weaker claims at good identification. In structural work, Diermeier et al. (2005) model the careers of members of Congress in the US, focusing more on the returns to a political career than the determinants of career success itself. Closer to this paper s focus, Hollibaugh et al. (2014) study the role of campaign experience and connections in presidential appointments, Jia et al. (2014) and Shih et al. (2012) study political selection in China, Iyer and Mani (2012) explore the control of bureaucrat progression (through transfers) by politicians, and Miquel 1 Though not on political selection, Dal Bó et al. (2013) explore the selection effects of wages for civil servants finding that higher wages improve selection with no adverse effects on motivation. Similarly, Ashraf et al. (2014) find that career incentives are more effective than social incentives at attracting qualified applicants for health worker positions in Zambia. 4

5 and Snyder (2006) emphasize the importance of learning-by-doing in legislative careers in North Carolina. Even closer to my work, Rossi and Tommasi (2012) focus on the relationship between legislative effort and career success in the Argentine House of Representatives. However, unlike in this paper, the opportunity to legislate is not randomly determined. Most specific to my context, Dewan and Myatt (2010) have studied theoretically the process of selecting and firing ministers, recognizing the problem of the talent pool being drained over time due to ministerial turnover. The focus in my paper is instead more on the informational element: efficiency losses are from the lack of knowledge about which MPs are most talented, as opposed to the need to hold ministers accountable leading to a shortage of competence. Finally, Loewen et al. (2014) and Horiuchi and John (2017) leverage similar natural experiments (in Canada and the UK respectively) to explore the short-term effects of random legislative opportunities. In contrast, by studying long-term career effects I focus on the different conceptual questions of path-dependency in careers and inefficiency in political selection. 2 Background UK Parliament is formed of two houses the upper house, the House of Lords, and the lower, elected house, the House of Commons. 650 Members of Parliament are elected to sit in the House of Commons from single-member constituencies in first-past-the-post elections every five years. 2 The dominant parties are the Labour and Conservative parties, though in 2010, no party could form a majority alone, leading to a coalition government of the Conservative and Liberal Democrat parties (the first coalition government since 1945). My focus is on the period 1950 to In this period there were no coalition governments, and power was held only by the Labour and Conservative parties. While the vast majority of successful legislation is proposed and passed by government ministers, there exist a number of opportunities for non-government ministers (either those in opposition, or those in the governing party without positions) to table Private Members Bills. These bills must go through the same stages as other Public Bills, but there is less time allocated to them. These bills are on the whole unlikely to be successful, but even if not passed, MPs may be able to raise publicity for a particular issue by proposing related legislation. There are four types of Private Members Bills in the House of Commons: 3 Ten Minute Rule 2 Elections could be called earlier historically (and during the period I focus on), though this changed with the Fixed-Term Parliaments Act of General information here: Private Members. 5

6 Bills, Presentation Bills, Lords Bills, and the focus of this paper, Ballot Bills. Ten Minute Rule Bills are often used by Members to raise the profile of a particular issue, and are very unlikely to become legislation. Presentation Bills are similarly unlikely to become legislation. Lords Bills are ones introduced in the House of Lords which then get picked up in the House of Commons but given that Commons Bills receive priority, their success is also hindered. Ballot Bills are bills proposed by winners of the annual Ballot for Bills. These are by far the most likely to become legislation (Figure 1). The graph shows the number of bills successful in each category from 1948 to 2009, with vertical lines for the start and end of the period under study. During 1950 to 1990, roughly 7 ballot bills (of, in most cases, 20) are successful on average per year. The Ballot for Bills has been a feature of Parliamentary process since at least Each year, near the start of a Parliamentary session, MPs without position in government are eligible to put their name down for the lottery over two days. Hundreds of MPs take this opportunity unsurprisingly given that it affords the extremely rare opportunity for junior politicians to legislate. Figure 2 shows that the number of entrants generally rose from around 300 to 400 from 1950 to A day or two after MPs have entered the ballot, names are randomly picked out of a hat, and ranked in the order that they are chosen. 6 With a few exceptions, 20 winners are chosen, and ranked from 1 to 20. The ranking is meaningful the highest ranked MPs get priority floor-time for their bills. A certain number of days are set aside for Private Members business in the Parliamentary calendar, limiting the time available to get through the requisite stages of a bill in order to pass it before the session is over. The high-ranked winners can then maximize their chances of success by picking earlier weeks in the calendar for the second reading of their bill. 7 MPs do not need to know the bill that they would propose in advance of entering the ballot they may decide in the days following the ballot (though in time for the first reading). Indeed this means that ballot winners have a short period of power immediately after winning when they are contacted by lobbyists and ministers hoping that they might take on a bill they wish to pass. The ballot seems a peculiar institution to many, and this is not lost on MPs. In 1979, Kevin McNamara MP commented at the second reading of his ballot bill (Social Security (Maternal Grant) (Amendment) Bill): 8 4 Newspaper references to ballot winners exist from The actual origins of the ballot itself are obscure (and not known even at the History of Parliament Trust). 5 The dotted vertical lines signify years in which the ballot is held just prior to a general election. There is weak regression evidence that the number of entrants is higher in these years (available on request). 6 See Ballot Winners 2014 Video. 7 The first reading is a formality, with no actual debate. 8 Link: Hansard Nov

7 As a Back-Bench Member I am in somewhat of a difficulty when constituents ask why, after 30 years, I am for the first time introducing a Private Member s Bill on a Friday. It is difficult to explain to them that the privilege of standing here on a Friday is the result of getting the fifth prize in a raffle. The prize is to have a full debate, possibly to get the Bill into Committee, and for it eventually to receive the Royal Assent. To Members of other legislatures and other democratic systems it seems nonsense, and perhaps in some ways it is, because the measures that are introduced in this way depend not on any scale of values social, economic, political or philosophical but on the luck of the draw and the Member s attitudes. Fortunately, many of the measures if not all of them introduced by this method have been of social or administrative importance, but it is still a rum way of doing things. The ballot is then peculiar also to politicians, but nevertheless, winning is highly valued. Andrew George MP, winning in 2014, claimed the ballot to be the MPs equivalent of winning the jackpot in the National Lottery. 9 Perhaps most interestingly, Margaret Thatcher, Prime Minister of the UK from 1979 to 1990, came second in the ballot on her second try in She describes the experience at length in her autobiography, emphasizing both the experience and exposure it provided (two mechanisms that I investigate empirically): As it happens, I had very little opportunity during my first few months as an MP for the relaxed acquisition of experience of the House. With 310 other Members I had entered the Commons ballot for the introduction of Private Members Bills. Never previously having so much as won a raffle, I was greatly astonished to find myself drawn second. Only the first few Private Members Bills have any chance of becoming legislation, and even 9 Link: Andrew George. 7

8 then the Government s attitude towards them is crucial......in the end, however, there is no substitute for one s own efforts. I wanted to get as many MPs as possible to the House on a Friday (when most MPs have returned to their constituency) for the Bill s Second Reading this was the great hurdle. I have always believed in the impact of a personal handwritten letter even from someone you barely know. So just before Second Reading I wrote 250 letters to Government backbenchers asking them to attend and vote for my measure......i was delighted that nearly 200 Members voted, and we won handsomely... It was clear from the press next day that the speech had been a success and that I was for the present at least a celebrity. 3 Data and Specification Ballot Entrants and Winners. Though the ballot has been in use for over 100 years, its use was suspended during the Second World War, and it was reinstated in I was able to compile data on entrants and winners (with ranks) of the ballot for the period 1950 to 1990 by using scanned lists from Parliamentary Archives 10 and information on winners from the House of Commons Journals. 11 It is crucial for identification, of course, to have data on entrants since the set of winners are only random draws from the pool of entrants, not the overall pool of politicians in Parliament. Unfortunately the list of entrants from Parliamentary Archives is missing or incomplete for some of the years. 12 I am left with 33 years with complete data on entrants and winners. I restrict the analysis to these years throughout. Restricting to these years, there are 1797 unique entrants of the ballot, entering around six times 10 The lists of entrants and winners are included in the Vote Bundle a set of papers with information on the Business of the House for each day. 11 Scans available here: HCJ. 12 Specifically: , , , 1974 (Nov) (there were two ballots held in this year), , , , , and

9 on average. In a given year the probability of winning is around 5%. While 71% of entrants in my sample never win, 23% win exactly once and 6% win more than once (Table 1). MP Career Histories. Since my primary interest is in the long-run career effects of ballot wins, it is important to have high quality MP-level data on positions held. For this I draw on a linked database on the careers and backgrounds of MPs from 1832 to 2001 developed by Prof. Michael Rush. 13 Most of this data has been updated as of 2001, with a subset updated as of My main set of outcomes will be various codings of the highest office ever achieved, with the hierarchy used shown in Figure 3. Here there are 10 rungs on the political ladder: beginning at Backbencher, and ending at Cabinet Minister 15 (roughly 20 MPs are Cabinet Ministers at any one time). The breakdown is given here for the 1797 entrants in the sample around 50% never make it past the bottom rung, while the highest office is Parliamentary Private Secretary (PPS) for around 14%. Roughly 17% eventually become a non-cabinet or a Cabinet Minister. The career outcome variables I use are: a dummy variable for whether the highest office is at least Junior Whip 16 (Whip+), a dummy variable for whether the highest office is at least Minister of State (Minister+), a dummy variable for whether the highest office is at least Cabinet Minister (Cabinet+), the rung number of the highest office from 1 to 10 (O f f icescore[1 10]) and the number of offices held during an MP s entire career (NoO f f icesheld). It is worth noting too that, for the most part, to hold any of these positions an individual has to be in the party that is currently in government. This introduces a timing problem winners may not benefit if their party doesn t win a general election for a long time. This issue reduces my statistical power to detect effects. 17 In addition, winners from minor parties will have no chance of career gains detectable in the data. For this reason, I restrict the sample only to Labour, Labour & Co-operative, 18 and Conservative MPs, according to party affiliation upon entry into Parliament. From the career data I also look at total years served (not necessarily unbroken) in Parliament (Y rsservice) and I put together baseline covariates (detailed below) to use to increase the precision of estimates. 13 More details here: Biography Data. 14 The lack of more recent updates limits the end-point of the period I study i.e. I opt to not go beyond 1990 given that the careers of MPs entering after then are more likely to be incompletely coded in the data. 15 The category of Prime Minister is not included here, given that only a small number ever hold this political office. 16 A Whip plays an important role in party discipline for example, ensuring Members attend important votes. 17 A solution would be to use data on members of the Shadow Cabinet (the Cabinet of the opposition). In future work I plan to collate this data. 18 These MPs stand on behalf of both the Labour Party and the Co-operative Party, and are still sometimes chosen as Ministers in Labour governments. Ed Balls MP (Shadow Chancellor at the time of writing) is one prominent example. 9

10 Election Data. I web-scraped election data for general elections from 1951 to and cleaned Excel spreadsheets for general elections from 1966 to I use this data to estimate the effects of winning on the vote share in the nearest next general election (VoteShare, equal to zero if doesn t run), whether the candidate runs again in the next general election (Run), whether the MP wins the next election (WinNext, equal to zero if doesn t run) and whether the MP wins conditional on running (WinI f Run, missing if doesn t run). Other Data. I also use data on the characteristics of bills proposed through the ballot. These variables will be detailed in Section 4. Stacked Specification. If there was only one year of the ballot, the specification would be straightforward, and similar to that used to analyze the results from randomized controlled trials: y i = α 1 + β 1 Won i + γ 1 X i + ε i where the sample is restricted only to entrants, since winning is only random conditional on entering. y i is some long-run, time-invariant, career outcome for MP i, Won i is a dummy variable equal to one if i wins the ballot, and X i are a vector of predetermined covariates, included for precision, not for identification. Since there are in fact many rounds of the ballot, I stack the data from all years, restricting only to entrants each time. By adding lottery fixed effects, dummy variables for each of the 33 ballot years, I ensure that winning is conditionally random. This specification is: y i = α 2 + β 2 Won it + γ 2 X it 1 + θ t + ν it where Won it = 1 if MP i wins in year t, θ t are lottery fixed effects, and the covariates may vary with time, provided they are predetermined. For example, one covariate I use is HeldO f f ice which is a dummy variable equal to one if the MP has already held any office prior to that year of the ballot. Since MPs can enter the ballot in multiple years, they can also enter multiple times into the stacked specification, with the same y i each time, given that it is time invariant. To deal with this, I cluster standard errors at the MP-level. In practice, there is heterogeneity by rank in the effects of winning, so I will primarily use the 19 Source: Richard Kimber Political Science Resources. Link: Election Data. 20 Source: Iain Outlaw s data compiled from various sources (mainly The Daily Telegraph). Link: Election Data. 10

11 following specification: 21 y i = α 3 + β 3 Won1to3 it + β 4 Won4to10 it + β 5 Won11plus it + γ 2 X it 1 + θ t + ν it where Won1to3 = 1 if ranked 1 to 3 in the ballot, Won4to10 = 1 if ranked 4 to 10, and Won11plus = 1 if ranked The full set of predetermined covariates I use are HeldO f f ice, as previously described, a dummy variable equal to one if the party of entry is currently in government (InGov), dummy for female (Fem), set of previous occupation (prior to entering Parliament) and education dummy variables, Conservative and Labour & Co-operative party (of entry) dummy variables (Labour party is the excluded category), age when entered Parliament (EntryAge), year of birth (BirthYear), a dummy variable equal to one if ever educated at Oxford or Cambridge (Oxbridge), a dummy variable equal to one if ever educated at Balliol/Christ Church/Magdalen/New College, Oxford or Trinity College, Cambridge (the five most common Oxbridge Colleges among MPs), a dummy variable equal to one if have some local government connection e.g. if served on local council prior to becoming an MP (LocalGov), a dummy variable equal to one if the descendant of another MP (Descendanto f mp). The only covariates to vary with time are HeldO f f ice and InGov. Balance Checks. Given that the randomization was done by officials in Parliament and not by the researcher, I look for balance on observables as a check that the lottery was in fact implemented randomly. In Table C1 I use the stacked specification to check for balance on seven predetermined variables covering education, age, local government connections, political party, and whether ever held office before. These regressions, as with others that follow, restrict only to Labour, Labour & Cooperative, and Conservative MPs. Encouragingly, none of the coefficients on Won1to3, Won4to10, or Won11plus are statistically significant at conventional levels. Most importantly, this is true of HeldO f f ice which is highly predictive of the career outcomes I will look at, and so balance on this variable is especially important. This said, I include the full set of covariates throughout in order to increase precision, and to mitigate concerns of any slight imbalances. 21 An alternative specification that does not involve stacking is: y i = α + β 1 Wins1to3 i + β 2 Wins4to10 i + β 3 Wins11plus i + Σ t γ t Enter it + u it where the key variables are now the number of wins ever (in each category) and I include a set of entry fixed effects (equal to one if i entered the ballot in year t). With this specification, we instrument for the three Wins variables with a large set of dummy variables for whether an individual won the ballot in a given category in a given year (33 years times 3 categories = 99 instruments). This specification is similar in spirit to that used in Angrist et al. (2010). Preliminary results with this specification are similar, and available on request. 22 Usually the maximum rank is 20, but in the year with the most winners, this is

12 In Table C2 I test for balance specifically for female winners, given that I explore this as a key dimension of heterogeneity. In this case all the level effects of Won are statistically insignificant, but two of the interactions with Fem are negative and statistically significant. The interaction is also large and positive (though insignificant) for the Graduate regression (i.e. female winners tend to be more likely to be graduates than female losers). This being said, the balance check for HeldO f f ice is again encouraging, and most important given how predictive HeldO f f ice is of career outcomes. In general, there is a greater likelihood of problematic imbalance when treating women separately given the small number of female winners during this period. 23 This is alleviated by including baseline covariates, and in any case, the main results are similar with and without such baseline controls. 4 Results Long-Run Career Impacts In Table 2 I explore the effect of winning the ballot on the political careers of MPs. In each of the five regressions I include a full set of controls, 24 together with the lottery fixed effects, with 1,542 unique MPs entering the sample in total. The first three columns give the results for effects on three binary variables: whether the highest office was at least Junior Whip, Minister of State, or Cabinet Minister. For all columns I include the Control Mean in the table defined as the mean of the outcome across all entrants that never won. 35% of non-winning entrants became at least Junior Whip in their career, while 25% became at least Minister of State, and only 10% become a Cabinet Minister. For high-ranked winners (1 to 3), the effect on each of the three is positive, and statistically significant at the 10% level for Whip+ and Minister+ (p = 0.13 for Cabinet+). More notably, the point estimates are large in magnitude, at 25% of the control mean for Whip+ (8.7 p.p.), 34% for Minister+ (8.4 p.p.), and 52% for Cabinet+ (5.2 p.p.). The winning rank matters there are no such positive effects for Won4to10 or Won11plus. If anything, there is some evidence that low-ranked (11+) winning has negative impacts. In columns 4 and 5 I consider effects on more omnibus measures of political careers: a score between 1 and 10 for the highest office achieved (1 for backbencher, 10 for Cabinet Minister) and 23 Twenty-two women win a total of 24 times during 1950 to 1990, when restricting only to Labour, Labour Cooperative, and Conservative winners. The numbers are 24 and 28 respectively when not restricting by party. 24 In Table C3 I present the same regressions without controls. The point estimates are nearly identical, though there is some loss of precision. 12

13 the total number of offices held during a career. The control mean for the former is 3.2, and 1.3 for the latter. Here the effect of high-ranked winning is 0.7 for O f f icescore, significant at the 5% level high-ranked winners are on average promoted almost one more time over their careers. The effect on number of offices held is 0.36 (28% of the control mean), significant at the 10% level. To summarize: high-ranked winners of the ballot for bills are more likely to become Ministers and hold more offices in their political careers than losers. The rank matters there are no positive effects for lower-ranked winners. It remains to explain what forces of political selection explain these impacts. Before addressing this directly, I first shed light on the role of ranking by using data on bill characteristics. Bill Characteristics Higher-ranked ballot winners are known to have a greater likelihood of legislative success. Thatcher herself writes in her autobiography: Only the first few Private Members Bills have any chance of becoming legislation, which as we will see, is somewhat of an exaggeration, but correct in spirit. In this section I test this claim empirically, and more generally explore how the type of bill varies with the rank of the winner. For these empirics I have data on whether each ballot bill was successful (in receiving Royal Assent, and thus becoming legislation) from a Parliamentary factsheet. 25 In addition I digitized the number of co-sponsors (maximum 11) of each of these bills from the House of Commons Journals. The number of co-sponsors is a signal of the effort the winner made in order to get support for the bill, as well as underlying support for the bill, and its prominence. Finally, I also code the number of mentions of each bill in Parliament from digitized Hansard records. 26 These mentions may occur before or after the winner proposes the bill (since bills may be proposed that have been proposed, but defeated, before). I consider this variable to be a proxy for the general importance or ambition of the bill for example David Steel s Private Member s Bill to legalize Abortion (Medical Termination of Pregnancy Bill, successful in ) has 77 mentions (the median is 11) whereas Mr Dodds-Parker s Exportation of Horses (Amendment) Bill (of , which was unsuccessful) has only one mention in Hansard. To look at these outcomes (Success, dummy equal to one if successful, Sponsors, number of cosponsors, MaxSponsors, dummy equal to one if maximum number of co-sponsors, and Mentions, 25 Source: Success of PMBs. 26 Source: Hansard Bills (beginning with A). 13

14 number of mentions) I use the following bill-level specification: y b = α + Σ 19 i=1 β i1[rank b = i] + γ t + ε b where y b is some outcome y of bill b, β i is the coefficient on a dummy variable equal to one if the rank of the bill is i, with bills with a rank above 19 in the excluded category. γ t are lottery fixed effects. Since for any given lottery year the ranking of a winner is random, β i is the causal effect of ranking on equilibrium bill outcomes. There are of course intermediate mechanisms (e.g. the effort that the MP puts in as a function of the rank, the earlier date the MP chooses for the second reading etc.) which cannot be completely separated. First I look at how the ranking affects success. The control mean for bills ranked above 19 is 27.7% Thatcher was not quite right to say that lower-ranked bills are never successful, since over a quarter are. However, she was right that higher-ranked bills are more likely to be successful. I plot the full set of coefficients (with 95% confidence intervals) from a regression of Success on rank dummies in Figure 4. The overall pattern is quite clear bills are more likely to be successful if higher-ranked. The coefficient for rank equal to one is 0.22 (significant at 5%), and 0.12, 0.08, 0.12 and 0.05 for ranks 2 to 5 respectively (though all insignificant). The higher-ranked winners have roughly a 40 to 50% chance of having their bill become legislation. Next I look at co-sponsors. The control mean is 6.8. The relationship with ranking is plotted in Figure 5. The top-5 winners have around 2 more co-sponsors on average than the excluded category. The top-ranked winner in particular is far more likely to gather the maximum number of co-sponsors 35.7 p.p. more likely, which is over double the control mean of 32.5%. High-ranked bills then get more support (and more exposure) in terms of co-sponsors, though this is likely a combination of these winners exerting more effort to amass co-sponsors (e.g. consider the related anecdote of Thatcher writing hundreds of handwritten letters to get support for her bill) and these winners being more prominent, attracting more co-sponsors without effort. Though not easy to test directly, the patterns in the data are consistent with the marginal returns to effort being higher for higher-ranked bills. Ranking lower means that the bill is less likely to pass, and therefore effort on ensuring its passing may be wasted. Finally I look at mentions. Given the right-skewness of the data (the median is 11, mean is 56, maximum is 4221) I drop outlier bills with more than 500 mentions (dropping only 2% of bills). 27 The coefficients are plotted in Figure 6. The top-3 ranked bills have around 20 to 40 more mentions (relative to a control median of 8), and the relationship between ranking and mentions is fairly 27 Results are similar if I drop fewer outliers, or if I use 1[Mentions b > median] (where median = 11). 14

15 monotonic. The effect of ranking on mentions does not work through success high-ranked bills are around 20 p.p. more likely to be successful, but successful bills actually have fewer mentions than unsuccessful bills. The effect seems instead to be that the highest ranked winners propose more ambitious legislation either important bills that have been proposed frequently in the past but failed (having many past mentions) or new bills that are likely to be important for years to come (having many future mentions). A contemporary example that fits this theory is the EU Referendum Bill, a bill that would commit to a referendum before the end of 2017 on the UK s membership of the EU. James Wharton MP topped the ballot in 2013 and chose to advance the referendum issue. 28 It is much less likely that an MP would propose such an ambitious and controversial bill if ranked lower in the ballot. Mechanisms The preceding analysis is informative of why only the high-ranked winners experience career gains higher ranked winners are more likely to be successful with their bills, they get more co-sponsors, and choose bills that are, or become, more prominent. However, these results on ranking still leave open the question of why exactly winners experience career gains. To address this, in this section I consider three specific mechanisms: 1. Political Survival: it is possible that high-ranked ballot winners experience immediate electoral gains for example by choosing a bill that could potentially benefit constituents, 29 and then being rewarded for this in the next general election. This mechanism might suggest that the returns to winning are higher immediately before elections (and there is some weak evidence for this the number of entrants are higher in election years). By increasing the probability of winning elections, ballot winners remain in Parliament longer, enhancing their chances of eventually being promoted to higher office. 2. Learning-by-doing: by getting the opportunity to legislate, a ballot winner gets the chance to become more competent by learning more about debate and the legislative process. Party leaders seek to promote the most competent MPs to ministerial roles, and they are more likely to find competence among winners than losers. 28 See Northern Echo and EU Referendum Bill. This particular attempt was ultimately unsuccessful as the legislation was blocked in the House of Lords. The Bill was later introduced by Secretary of State Philip Hammond, and successfully passed in 2015, ultimately leading to the UK s exit from the European Union. 29 There is some anecdotal evidence for this, e.g. Andrew George topped the ballot in 2014 and said Over the next week I will pause and reflect and produce a shortlist of three or four proposals which I will then put to my constituents for final consultation and response and encourage local people to help me decide. Source: Andrew George Site. 15

16 3. Exposure: high-ranked ballot winners become more prominent and well known by party leaders those who will make promotion decisions in the future. For some, this may be a bad thing, since they reveal to party leaders that they are a liability, and therefore unsuited to office, when previously this was unknown. For others it is a good thing, since they reveal that they are competent, capable in debate, and capable of gathering support for legislation. This mechanism is one in which party leaders lack information about potential candidates for office within their party. They learn about candidates over time when these candidates get opportunities to showcase their talents. Ballot winners are more likely to be promoted on average since they are known quantities, whereas there remains a degree of uncertainty about the quality of losers. These three mechanisms have different efficiency implications: 30 the exposure channel suggests that there is a lack of information in Parliament (party leaders are uncertain about the quality of party members), and the presence of the ballot permits learning. In the presence of imperfect information, party leaders can make mistakes promoting members that may not be of the highest quality. The other two mechanisms on the other hand could be consistent with fully efficient selection of top politicians. It could be that those best suited to be ministers are those with the most experience, and winning the ballot increases longevity, so winners are more likely to be promoted (political survival channel). On the other hand, if learning-by-doing is important, it could be that ballot winners are far more competent than losers, and thus it is efficient to promote them. In what follows, I will provide more direct evidence on each channel. Political Survival. To test for the survival channel, I look at the effects of winning on total years in Parliament (Y rsservice), as well as subsequent election outcomes. High-ranked winners remain in Parliament for 1.7 more years (significant at 5%), an increase of 10% from the control mean (column 1, Table 3). On the other hand, there is no effect on outcomes in the election immediately after winning: the effects on vote share, whether the politician runs again, winning, and winning conditional on running are all small in magnitude and insignificant. This suggests that the positive effect on the length of service in Parliament is less likely to be an intermediate mechanism of the career impact, and more likely to be an outcome. For example, politicians may desire to remain in the job longer if they are higher in the hierarchy. If winning the ballot helped political careers by improving the chances of re-election, it is likely that the largest effect on re-election would be 30 Consider selection efficiency to be perfect when the highest quality politicians are ministers, and imperfect when there exist non-ministers who are higher quality than ministers. This definition is sensible if we assume that party leaders want to maximize the quality of ministers, but may be able to do so only imperfectly if they have imperfect information. 16

17 in the first election following the win, rather than later on. Even if this is not the case, it seems that an effect on total service of 1.7 years could only reasonably explain a small proportion of the 8 percentage point impact on Minister+. In particular, if I regress Minister+ on Y rsservice with the full set of controls in the same stacked specification, while Y rsservice is highly predictive of Minister+, the coefficient is only around The effect of 1.7 years might then roughly predict an effect on Minister+ of less than 25% of the effect we actually see. 31 Overall, there is little evidence for a political survival channel. Learning-by-doing. According to the learning-by-doing channel, politician quality is not fixed, but endogenous, and the ballot gives politicians a chance to improve their quality, making them better suited to future ministerial office. This mechanism has one specific implication for which I find contrary evidence: if the key mechanism is learning-by-doing, all ballot winners should have weakly positive career impacts i.e. even if ranked low, with a lower probability of legislative success (say 25%), ballot winners should gain some experience from the process, increasing their quality. The exposure mechanism does not have the same prediction. In this case winners of low true quality but high expected quality (e.g. groups that are overestimated) can actually lose out by winning the ballot revealing their true quality to party leaders that might have expected them to be high performers otherwise. In addition, if party leaders learn in a somewhat non-rational way, even random sets of winners could lose out by winning the ballot. For example, low-ranked winners could experience negative effects if party leaders update about their ability by observing their bill success rate, without fully conditioning on their low rank. Since the bill success rate is low for low-ranked winners, party leaders could then erroneously update their beliefs downwards about low-ranked winners. In this sense, there is some evidence against the learning-by-doing mechanism. In Table 2 the effects for low-ranked winners are negative, though only significant for NoO f f icesheld, at the 5% level. In addition, in Table 3 we see that low-ranked winners have a shorter political career by one year (significant at 5%) than losers. These negative effects are largely inconsistent with a learning-by-doing story. One counterargument is the following: low-ranked winners spend effort trying to get their bill passed, learn a little along the way, but ultimately get sidetracked from constituency/electoral concerns, and are then more likely to lose the next election (resulting in shorter political careers, and 31 This result comes with the caveat that the identification is not clean when we regress Minister+ on Y rsservice (Y rsservice may be correlated with many MP-level unobservables). 17

18 fewer offices held). This distraction story might reconcile the negative results with a learningby-doing story. However, in Table 3 it is also the case that there is no effect of low-ranked winning on outcomes in the upcoming election. It seems unlikely then that this particular reconciliation has promise. In summary, the possibility of negative effects goes against the learning-by-doing channel. Exposure. Next I test the exposure channel. A model in which a Bayesian party leader learns about other party members has implications that do not necessarily follow from other models. Given the general implications of models of this type, I propose three separate tests of the exposure channel: 1. Larger effect for under-estimated (incorrect prior) group the returns to exposure should be higher for those individuals that have underlying quality that is higher than the priors of party leaders, for example, those individuals that are discriminated against. 2. Smaller effect if the party pays less attention the returns to exposure should be smaller if the signal sent is less precise. More practically, the returns to exposure should be smaller if the attention of party leaders is spread over more people. 3. Diminishing returns to signals the returns to exposure should fall over time as the posterior distribution of the party leader becomes tighter. For the first test, I look for heterogeneity by gender. Women have been historically under-represented in Parliament, and indeed of my sample of 1797 entrants, only 90 are female. It is plausible that women have higher returns to exposure, particularly if party leaders have tended to underestimate their quality. I test this in Table 4. Given the smaller number of women, I lack power to break the results down by rank, and so I instead regress career outcomes on Won and Won Fem. The results are stark given that for the entire sample we only see effects for high-ranked winners. The interaction with female is positive throughout, and statistically significant for W hip+, Minister+, and NoO f f icesheld. Focusing on Minister+, the effect of winning in general is 16.9 p.p. higher for women than for men a large increase over the control mean of 25%. The results here suggest in general that the career gains of a woman winning the ballot with any rank are perhaps even larger than the career gains for men winning the ballot with a rank of 1 to 3. To complement the result on gender heterogenity, I move to the second test. One feature of the ballot is that there is random variation in the party composition of winners each year. If the exposure channel is important, it should be the case that it is better to win the ballot (even conditional on rank) in a year in which there are fewer winners from the same party. In these years, the winner will 18

19 garner more interest from their own party leaders who have a stake in what legislation is proposed by their party members. To test this, I create a new variable: Partysignal pt = #Winners f rom party pinlotteryt #Winners in lotteryt and regress career outcomes on Won, together with the interaction Won Partysignal pt. In addition, I control for Won InGov, since when an MP s party is in government, there may be more winners from that party given that they have more Members in general, or less winners given that more Members from that party are ineligible for the ballot, Won Rank to control for general rank effects, though Rank should be orthogonal to Partysignal pt, and Won Tory, where Tory is a dummy variable for the Conservative party. The results are in Table 5. The coefficient on the key interaction term is negative throughout, and statistically significant for the two summary outcomes O f f icescore and NoO f f icesheld (at 10% and 1% respectively). This is consistent with the exposure story even conditional on the rank in the ballot, the career impact is larger if an MP wins in a year when there are fewer winners from the same party. The magnitude is sizable too considering that the difference between the 25th and 75th percentile of Partysignal pt is 0.21, a move from the 25th to the 75th percentile of same party winners erodes roughly half the career gain from winning. For the third test I look at whether the effect of winning the ballot is smaller when the MP has already won before. Given that there is some evidence of negative effects of low-ranked winning, and to trade this factor off with the loss of statistical power due to relatively few people winning more than once, I run a regression splitting Won into Won1to10 and Won11plus. I construct a variable WonAlready equal to one if the MP has won (with any rank) the ballot in a previous year. I then include the dummy variable WonAlready as well as the interactions Won1to10 WonAlready and Won11plus WonAlready. Given that high-ranked winning has positive effects (at least for ranks 1 to 3), the prediction here is that the coefficient on Won1to10 will be positive, but the coefficient on Won1to10 WonAlready negative (smaller effects of winning if already won). Given that low-ranked winning has negative effects, the prediction is the opposite: the coefficient on Won11plus should be negative, and that on the interaction positive due to smaller negative effects of winning a second time. The evidence in Table 6 is broadly consistent with these predictions. The coefficients on Won1to10 WonAlready are all negative with 3 out of 5 significant. That said, they are larger than theory would predict, always more than offsetting the coefficient on Won1to10. This result may just be due to noise, given the smaller number of repeat winners identifying the effects. Al- 19

20 ternatively, it may be due to a different kind of heterogeneous treatment effects. Repeat winners could differ in other ways, for example, they are likely to enter the ballot more frequently, and less likely to be promoted (since MPs are ineligible to enter if in government office). The coefficients on Won11plus are all negative, and those on Won11plus WonAlready mostly positive, though not significant. These three tests taken together give evidence that exposure is the crucial channel. The ballot can lead to career gains for winners by increasing their prominence within the party, increasing their likelihood of future promotion if their underlying competence is high. With this evidence in mind, the results on ranking and bill characteristics may best be interpreted as consistent with the exposure channel higher ranked bills and their proposers become more prominent through their greater success, ambition, and number of co-sponsors. 5 Conclusion Despite growing evidence on the mechanisms driving the selection of politicians from the populus, there is a lack of rigorous evidence on the process of selection of top politicians from among the political pool. In this paper I explore this question in the context of UK Parliament where junior politicians are given the annual opportunity to legislate by random lottery. High-ranked winners experience significant career gains: they are around 8 p.p. more likely to eventually become a Minister and hold around 0.4 more offices throughout their career. I argue that the key mechanism driving the political selection of ballot winners is an exposure channel: high-ranked winners become more prominent, revealing their underlying competence to party leaders, who are then more likely to promote them in future if their underlying quality is high. Consistent with this hypothesis I find that women (an under-represented group in Parliament) have higher returns to winning, the returns to winning are lower if more members from the same party win in the same year, and the returns to winning are lower if the MP has won before. More broadly, these findings suggest that political careers can be path-dependent, and that exposure and learning can be important mechanisms for career progression even in information-rich settings. 20

21 References Angrist, J., Lavy, V., and Schlosser, A. (2010). Multiple experiments for the causal link between the quantity and quality of children. Journal of Labor Economics, 28(4): Ashraf, N., Bandiera, O., and Lee, S. S. (2014). Do-gooders and go-getters: career incentives, selection, and performance in public service delivery. Technical report. Besley, T. (2005). Political selection. The Journal of Economic Perspectives, 19(3): Besley, T. and Coate, S. (1997). An economic model of representative democracy. Quarterly Journal of Economics, pages Besley, T., Montalvo, J. G., and Reynal-Querol, M. (2011). Do Educated Leaders Matter? Economic Journal, 121(554): The Chattopadhyay, R. and Duflo, E. (2004). Women as policy makers: Evidence from a randomized policy experiment in India. Econometrica, 72(5): Dal Bó, E., Finan, F., and Rossi, M. A. (2013). Strengthening state capabilities: The role of financial incentives in the call to public service*. Quarterly Journal of Economics, 128(3): Dewan, T. and Myatt, D. P. (2010). The declining talent pool of government. American Journal of Political Science, 54(2): Diermeier, D., Keane, M., and Merlo, A. (2005). A Political Economy Model of Congressional Careers. American Economic Review, 95(1): Ferraz, C. and Finan, F. (2008). Exposing Corrupt Politicians: The Effects of Brazil s Publicly Released Audits on Electoral Outcomes. Quarterly Journal of Economics, 123(2): Ferraz, C. and Finan, F. (2011). Motivating Politicians: The Impacts of Monetary Incentives on Quality and Performance. Hollibaugh, G. E., Horton, G., and Lewis, D. E. (2014). Presidents and patronage. American Journal of Political Science, 58(4): Horiuchi, Y. and John, P. (2017). Opportunities in Parliament and Political Careers: A Natural Experiment in the United Kingdom. SSRN Scholarly Paper ID , Social Science Research Network, Rochester, NY. 21

22 Iyer, L. and Mani, A. (2012). Traveling agents: political change and bureaucratic turnover in India. Review of Economics and Statistics, 94(3): Jia, R., Kudamatsu, M., and Seim, D. (2014). Political Selection in China: the Complementary Roles of Connections and Performance. Jones, B. F. and Olken, B. A. (2005). Do Leaders Matter? National Leadership and Growth Since World War II. Quarterly Journal of Economics, 120(3): Loewen, P. J., Koop, R., Settle, J., and Fowler, J. H. (2014). A natural experiment in proposal power and electoral success. American Journal of Political Science, 58(1): Miquel, G. P. I. and Snyder, J. M. (2006). Legislative effectiveness and legislative careers. Legislative Studies Quarterly, 31(3): Osborne, M. J. and Slivinski, A. (1996). A model of political competition with citizen-candidates. Quarterly Journal of Economics, pages Rossi, M. and Tommasi, M. (2012). Legislative Effort and Career Paths in the Argentine Congress. Shih, V., Adolph, C., and Liu, M. (2012). Getting ahead in the communist party: explaining the advancement of central committee members in China. American Political Science Review, 106(01):

23 A Figures Figure 1: Private Members Bill Success Figure 2: Ballot Entrants 23