‘I acknowledge that you can’t keep an organization together without patronage. Men ain’t in politics for nothin’. They want to get somethin’ out of it.’
W. L. Riordon, Plunkitt of Tammany Hall, p.47.Footnote 1
Why do party members contribute to their party’s success? Under what conditions do party members adversely affect their party’s electoral strategies? Many theories assume that the prize of electoral victory motivates party members, and that the pursuit of non-electoral objectives risks electoral loss.Footnote 2
Yet, these theories cannot explain why many individuals contribute to the success of an important type of political party, the political machine. Political machines are prevalent throughout developing democracies, and they rely upon a large number of local intermediaries, commonly known as brokers, to distribute benefits directly to voters.Footnote 3 Brokers allow parties to pursue clientelist strategies, and make voters who receive resources more likely to support the brokers’ party.Footnote 4 They achieve this by responding to particular problems that arise in a voter’s life with appropriate resources,Footnote 5 identifying voters who will be responsive if given party resources, and ensuring that these voters provide political support in exchange for these resources.Footnote 6
By undertaking extensive effort brokers help political machines win elections, but they would undersupply this effort if party leaders only motivated them with electoral victory. The ongoing activities that brokers must undertake to be effective require substantial time investments, which limit the number of voters that each broker can organize. So machines must extract effort from many brokers, all of whom organize small groups of voters relative to the size of the electorate. This creates a collective action problem for brokers who work for electoral victory; their individual marginal efforts do not significantly impact electoral outcomes, and they are inclined to free-ride on the efforts of other brokers. How then do leaders extract effort from brokers?
To solve the collective action problem, parties offer brokers private rewards, such as career advancement and resources in return for the voters that each broker produces. This causes a broker to focus on expanding her local following, and effectively solves the collective action problem. Although a broker is unlikely to affect an electoral outcome, she can expand her local following through her own marginal effort. Moreover, she cannot free-ride on the effort of others, since she is only rewarded for the voters she produces.
But this result indicates that the brokers who work hardest for the party will have distinct objectives from the party leader who employs them; party leaders seek to win elections, while brokers work to advance their careers by maximizing their local followings. As brokers develop loyal followings of voters, they can use the threat of taking their voters to another party to bargain for additional resources, or they may abandon their party altogether if a competing party can offer a better deal. This forces party leaders to over-invest resources in maintaining their current brokers, rather than expanding the reach of their machine. Moreover, seemingly dominant political machines can collapse with shifts in access to resources.
In this article, I contribute to research on political parties, machines, and clientelism. I show why electoral victory provides insufficient motivation for brokers, and I identify how this can make political machines vulnerable. Importantly, my theory identifies conditions that contribute to a machine’s dominance, while also identifying conditions that encourage party leaders to circumvent brokers and pursue alternative forms of distributive politics. This argument is developed formally and tested with unique survey data from Argentina, an important developing democracy.
There are other multi-actor theories of political machines, which focus upon principal–agent problems between party leaders and brokers. Stokes et al., Camp, Dixit and Stokes, and Keefer and Vlaicu identify principal–agent problems by recognizing that brokers have superior information regarding voters.Footnote 7 Using this informational advantage, brokers distort a party’s expenditure by personally consuming resources and giving resources to ‘core’ voters. Stokes et al. show how these informational problems can encourage parties to shift away from clientelism.Footnote 8 Camp et al. identify situations when the costs of brokers encourage party leaders to turn away from them.Footnote 9
This article differs from those in two important ways. First, I explicitly consider the counterfactual of brokers who are motivated by their party’s electoral victory, and show that electoral victory provides insufficient motivation when brokers organize small groups. This raises doubts that brokers can be sufficiently motivated by a bonus contingent upon victory as assumed in Stokes et al. and Camp et al.Footnote 10 More importantly, it reveals that party leaders derive an electoral advantage from relying on brokers whose objectives are distinct from theirs.
Secondly, instead of focusing on a principal–agent problem, I focus on an intra-party bargaining problem. Certainly, information asymmetries are important for political machines, but this is not the whole story. By focusing on these bargaining problems, this article identifies an additional set of inefficiencies associated with clientelist exchange. Unlike principal–agent problems, these problems cannot be solved with better monitoring technology. Moreover, this model also indicates that the most successful brokers impose the greatest bargaining costs on their party. Finally, I go further than the other models, by showing that the costs of brokers depend inversely upon the resource advantage of the brokers’ party. Thus, this article depicts an important dilemma for party leaders. These leaders need effective brokers, but effective brokers can harm a party when they gain better career opportunities elsewhere.
In the following sections, I develop my argument formally, and then test the theoretical model with survey data from Argentina. The theoretical model developed in the next section, focuses first on how party leaders motivate brokers to undertake effort for their party. The model then develops a bargaining problem that can arise between party leaders and brokers who work in politics to expand their own local following. The model is evaluated using a unique survey of political brokers from four provinces in Argentina.
The Model
How do political machines best extract effort from brokers? The initial setup of my model depicts a common empirical situation: a single machine distributes clientelist benefits.Footnote 11 A team of brokers must individually choose the level of effort that each is willing to contribute toward organizing voters. This effort consists of building relationships with voters to monitor their vote choice, helping them access state resources, mobilizing voters, building neighborhood party organizations, and other electoral and social activities. A broker maximizes his utility by allocating time between achieving a political objective and pursuing other interests, which could include: familial obligations, other forms of employment, and recreation. The model first shows how different incentives affect the level of effort that a broker contributes to achieving political objectives.
I make two different assumptions regarding the incentives of brokers. In one case, brokers and leaders share the same political objective of maximizing their party’s vote share. In the second case, brokers no longer care about their party’s vote share, and instead, each broker pursues a separate political objective of maximizing her own local following. After providing this stark contrast between the most extreme types of brokers, I consider brokers who prioritize both their local following and their party’s victory. Finally, I consider the implications that arise when brokers have valuable exit options from their party.
Party Activists: Brokers Prioritize Their Party’s Victory
Consider a party comprised of n brokers. Each of these brokers produces votes from a distinct group of voters for her party by expending effort. If a broker does not contribute effort, then voters in the broker’s group will either not vote or vote for the opposing party. Each broker i has a vote production function of:
$$F_{i} (L_{i} )={p \over n}\theta L_{i} ,$$
where L i ∈[0,1] is the effort expended by the broker, θ∈[0,1] measures brokers’ skill level, and p∈[0,1] is the proportion of the total electorate that could be organized by a party’s current brokers. Normalizing the total labor that a broker can expend to 1, makes the product of broker’s effort and skill level equal to the share of her constituency that she produces for her party. For example if L i =0.5 and θ=0.5, then the broker will turn out 25 percent of her constituency to vote for her party.
The total vote share that is produced by a machine’s brokers is the sum of the vote shares that each broker earns from her constituency. Formally, this is:
$$\pi (L)=\mathop{\sum}\nolimits_{i=1}^n {F_{i} (L_{i} } ).$$
From Equation 2 we see, for example, that if p=0.7, then the brokers’ party would earn a vote share of 70 percent if all of the brokers contributed maximum effort and had a maximum skill level.
First, consider the case when brokers are party activists and only care about their party’s victory and their own time. The utility function for brokers is then defined as a Cobb–Douglas function:
$$v_{i} (\pi (L),\,L_{i} )=\pi (L)^{\alpha } (1{\minus}L_{i} )^{{1{\minus}\alpha }} ,$$
where α∈[0,1] measures the weight that a broker places on her political goals relative to other interests.
Each broker independently chooses a level of effort that maximizes her utility. This level of effort is defined by the maximization problem:
$$max_{{L_{i} \in [0,1]}} v_{i} (\pi (L),L_{i} ).$$
Using the first order condition and using the fact that each broker contributes the same amount of labor, we can solve for L i in terms of α and n:Footnote 12
$$L_{i}^{{\asterisk}} ={\alpha \over {\alpha {\plus}n(1{\minus}\alpha )}}.$$
The machine’s vote share is:
$$\pi (L^{\asterisk})={{p\theta \alpha } \over {1{\plus}(1{\minus}\alpha )(n{\minus}1)}}.$$
When brokers participate in politics only to help their party win elections, they face a collective action problem. Equation 5 indicates that each broker’s effort declines as the number of brokers increases. Equation 6 shows that a machine suffers when these brokers organize constituencies that are small relative to the size of the electorate. It shows that a machine’s vote share decreases as the number of brokers required to organize a fixed share of the electorate increases. The basic problem facing these brokers is that every broker’s effort enhances a party’s victory, and so each activist broker is inclined to free-ride on the efforts of others.
The model indicates that this collective action problem becomes especially pressing when brokers organize small groups of voters relative to the total number of voters in the electorate, which forces a party to depend upon many brokers to win elections. This is problematic for machines since the activities that brokers undertake to make clientelist exchange work are extremely time-intensive and prevent brokers from organizing large groups.
Careerists: Brokers Prioritize Their Own Power
Now consider a broker who simply wants to advance her own power and career by expanding her local following. A broker’s vote production function is still defined by Equation 1 and the machine’s vote share is defined by Equation 2. Yet now the broker is only concerned with the voters that she can produce from her own constituency. Thus, her utility function is:
$$v_{i} (L_{i} )=\left( {{p \over n}\theta L_{i} } \right)^{\alpha } (1{\minus}L_{i} )^{{(1{\minus}\alpha )}} .$$
Solving for the first order condition of this maximization problem defines broker i’s labor contribution as:
$$L_{i}^{{\asterisk}} =\alpha .$$
Now a broker’s effort is entirely determined by α, which measures how much a broker values her political power and career relative to leisure or non-political objectives. Unlike the previous case, a broker’s effort is not affected by the number of brokers that the machine employs. When brokers prioritize their own power, they do not suffer from a collective action problem.
The machine’s vote share is now defined as:
$$\pi (L^{{\asterisk}} )=p\theta \alpha .$$
Comparing Equations 6 and 9, we can see that generally a machine produces fewer votes when it is comprised of brokers who prioritize the machine’s vote share rather than their own power. These two machines win the same number of votes only if one of four conditions holds: brokers derive no utility from non-political activities, so α=1; brokers derive no utility from political activities, so α=0; the machine only consists of one broker, so n=1; or the machine does not organize any voters, so p=0. Moreover, the difference between Equations 6 and 9 increases as n increases.Footnote 13 This indicates that a machine comprised of party activists would dramatically underperform one comprised of careerists when brokers organize small groups relative to the electorate, a situation commonly documented by scholars, and supported in the empirical portion of this article.
Figure 1 illustrates the advantages that parties derive by relying on careerist brokers. This figure allows us to consider the counterfactual of what happens to a party’s vote share when the party must rely on brokers who organize small groups of voters. The figure shows a party’s vote share as a function of the number of brokers that are required to organize 75 percent of the electorate.Footnote 14 If brokers are limited by their economies of scale and, thus, end up organizing small groups of voters, then a party will be forced to rely on a large number of brokers to organize a fixed share of the electorate. A party’s vote share declines dramatically with an increase in the number of brokers, when it is comprised of activists. When a party is comprised of careerists, its vote share is unaffected. Empirically, brokers organize small groups of voters, which means that political machines are often comprised of thousands of intermediaries. Under these circumstances, this figure indicates that the collective action problem facing activists could potentially decimate the machine’s vote share.
Fig. 1 A machine’s vote share as a function of the number of brokers required to organize 75 percent of the electorate
Mixed Motivations
By making extreme assumptions about a broker’s motivation, I showed that brokers who only pursue their party’s victory would experience a collective action problem, and that this collective action problem is completely resolved when brokers pursue careerist objectives. Yet, brokers may value both their party’s victory, and their own power and career. After all, a broker’s power and career might be partially contingent upon her party’s victory. For example, a broker could experience a cost associated with abandoning her party when it loses an election. Even if this is the case, a slightly more complicated model still shows that the individual pursuit of one’s power and career softens any collective action problem that brokers might face. Thus, the model anticipates that party leaders will utilize individual career motivations to extract effort from brokers.Footnote 15
Broker Exit and Bargaining Strength
To win elections, political machines often need to reach voters who are not directly organized by brokers who work for the machine. Stokes et al. present extensive evidence that shows a minority of voters receive goods or help from brokers.Footnote 16 These authors find that as little as 5 percent of voters in Argentina receive goods from party operatives, and that party operatives distribute goods in 30 percent of the voters’ neighborhoods. They also find that 20 percent of voters in eighteen sub-Saharan African countries and 7 percent of voters in seventeen Latin American countries reported that they were offered a gift or were pressured by someone to vote a certain way. Using a list experiment that was designed to elicit honest responses, Gonzalez-Ocantos et al. find that 24 percent of registered voters in Nicaragua were offered a gift or service in exchange for their vote.Footnote 17 Together these results indicate that resources distributed through brokers play a significant role in many electoral democracies, but that parties must also reach voters who do not have relationships with their current brokers.
If party leaders have surplus resources, they may want to use the resources to reach voters who are not organized by the machine’s brokers. Party leaders could use surplus resources in two ways to attract these voters. First, they could begin building party organizations in districts where the party does not currently have brokers, which could consist of finding neighborhood leaders and training them to organize voters. Alternatively, they could dedicate the resources to media and alternative outreach that attracts voters through non-clientelist strategies, such as emphasizing policy positions. This latter strategy may become increasingly important as voters become wealthier and are less persuaded by resources that address basic needs.Footnote 18
To capture the idea that party leaders want to use surplus resources to expand the reach of their machine, I redefine the party’s vote share so that it accounts for voters who are organized by brokers and voters who are not. The party’s vote share and the party leader’s utility function can be written as:
$$\prod(r_{{pl}} ,\,L)=(1{\minus}p)\pi _{{pl}} (r_{{pl}} ){\plus}p\pi _{b} (L);$$
where p∈[0,1] is still the percentage of voters in the electorate who are organized by the machine’s brokers. The share of resources that a party leader uses to reach voters who are not organized by the machine’s brokers is r
pl
∈[0,1]. The party’s vote share that it earns from constituencies that are currently not organized by the machine’s brokers is π
pl
(r
pl
)=σr
pl
; the party’s vote share that it earns from constituencies that are organized by the machine’s brokers is
$$\pi _{b} (L)=\mathop{\sum}\nolimits_{i=1}^n {{1 \over n}\theta L_{i} } $$
. A measure of the inefficiencies associated with using resources to capture votes from voters who are not organized by the machine is provided by σ∈[0,1]. The value of σ would likely be a function of the effectiveness of media strategies or the ease of building party organizations in neighborhoods that lack them.
The problem facing political machines is that brokers who focus on developing their local following as a means to develop their own power and career have an incentive to maximize the resources that they can procure from a party. Brokers who secure large amounts of resources for their followers gain reputations of effectively looking after their voters.Footnote 19 Careerist brokers may also consume these resources privately, or hoard them to expand their following further in future periods. Below, I provide substantial empirical evidence that many brokers prioritize the resources that they can procure from their party. In sharp contrast to careerist brokers, I assume that party activist brokers want party leaders to have full discretion over resources as long as the party leaders use the resources to maximize the party’s vote share.
Now a careerist broker has a utility function of:
$$v_{i} =\left( {{p \over n}\theta L_{i} {\plus}r_{i} } \right)^{\alpha } \left( {1{\minus}L_{i} } \right)^{{\left( {1{\minus}\alpha } \right)}} ,$$
where r
i
is the value of the resources that broker i receives. I normalize the total amount of surplus resources as equal to 1, and so
$r_{{pl}} {\plus}\mathop{\sum}\nolimits_{i=1}^n {r_{i} =1} $
.
To capture the potential costs of careerist brokers, I provide brokers with an exit option and add a second stage to the model. In the first stage of this framework, brokers still choose their effort level to maximize their utility, and they undertake this effort to build their local following. In the background, I am assuming that the party’s current brokers already have sufficient resources to organize their own constituency.
But now, during the second stage, each broker and the party leader bargain over surplus resources that could be used to attract voters who are not currently organized by brokers. A party leader bargains to retain full discretion over the resources, which she will use to expand the reach of the machine.Footnote 20 In contrast, careerist brokers want the resources to develop their own careers and, therefore, bargain for as many resources as possible. I employ a subgame perfect Nash equilibrium concept; I assume that brokers recognize that the resources each will secure are endogenous to the second-stage bargaining process and choose their labor accordingly.
If the party leader and each broker can come to an agreement over the distribution of resources, then each broker delivers her voters to the party. Therefore, the party leader gets
$r_{{pl}}^{{\asterisk}} $
resources and all of the votes that the brokers produce, and each broker receives
$r_{i}^{{\asterisk}} $
resources. In contrast, if a party leader cannot come to an agreement with a broker, then the party leader receives all of her share of the resources
$r_{{pl}}^{{\asterisk}} $
, and the resources that would have been distributed to the broker in equilibrium,
$r_{i}^{{\asterisk}} $
. But the broker withholds her votes, and so the party leader only earns a vote share of
$\prod(r_{{pl}}^{{\asterisk}} {\plus}r_{i}^{{\asterisk}} ,L){\minus}{p \over n}\theta L_{i} $
.
If the broker and party leader cannot come to an agreement, I assume that the broker delivers her voters to a competing party. For tractability, I assume that the competing party simply provides the broker with resources that value
$$\rho {p \over n}\theta L_{i}$$
. Thus, the resources that a broker could earn from an outside option are endogenous to the votes that the broker can produce multiplied by ρ, which measures the resources available to the opposing party.
There is an intuitive and substantive interpretation for ρ. For example, a broker who works for the governor’s party in the Argentine province of San Luis would have a low-valued ρ. The Saá family has dominated San Luis’s provincial government since Argentina democratized in 1983.Footnote 21 So if a broker abandons the governor’s party in this province she will be hard-pressed to leverage her local following to build a political career and advance her power. In situations where parties are more competitive, like in the Argentine province of Córdoba or in late nineteenth-century Britain, ρ is likely to be higher. In these cases, a broker can leave the incumbent party, and still use her local following to build a career. Finally, a broker working for an opposition party in a province like San Luis should have an extremely high-valued ρ. This broker’s local following might be more valuable to her if she abandons her current party and works for the governor’s party.
Given a set of parameter values, an equilibrium in this model constitutes a vector of
$L_{i}^{{\asterisk}} $
for each broker i, a vector of
$r_{i}^{{\asterisk}} $
for each broker i, and
$r_{{pl}}^{{\asterisk}} $
, which solve the following maximization problems. Since all of the brokers are engaged in the same bargaining and maximization problems, the equilibrium vectors can be solved using a system of equations that consists of three first order equations for the bargaining problem and one first order equation for the effort maximization problem.
$$\mathop{{{\rm max}}}\limits_{{L_{i} }} \left {\left( {{p \over n}\theta L_{i} {\plus}r_{i} (L_{i} )} \right)^{\alpha } (1{\minus}L_{i} )^{{(1{\minus}\alpha )}} ,} \right.$$
$$\eqalignno{ & \mathop{{\max }}\limits_{{r_{{pl}} ,r_{i} }} \left( {\prod {\left( {r_{{pl}} ,L} \right)} {\minus}\left( {\prod {\left( {r_{{pl}}^{{\asterisk}} {\plus}r_{i}^{{\asterisk}} ,L} \right){\minus}{p \over n}\theta L_{i} } } \right)} \right)^{\beta } \left( {r_{i} {\minus}\rho {p \over n}\theta L_{i} } \right)^{{(1{\minus}\beta )}} \cr & \quad\quad\quad\quad s.t.1=r_{{pl}} {\plus}\mathop{\sum}\limits_{i=1}^n {r_{i} }. $$
Equation 13 represents the solution to the Nash bargaining problem that occurs in the second stage, where β measures the bargaining ability of the party leader relative to each broker.Footnote 22
Solving the system of equations yields the following interior equilibrium outcomes.
$$L_{i}^{{\rm {\asterisk}}} =\alpha ,$$
$$r_{i}^{{\rm {\asterisk}}} ={{p\theta \alpha ((1{\minus}\beta ){\plus}(1{\minus}p)\sigma \rho \beta )} \over {n(1{\minus}p)\sigma }},$$
$$r_{{pi}}^{{\rm {\asterisk}}} =1{\minus}{{p\theta \alpha ((1{\minus}\beta ){\plus}(1{\minus}p)\sigma \rho \beta )} \over {(1{\minus}p)\sigma }}.$$
These results capture a fundamental dilemma for party leaders. Party leaders need effective brokers to make clientelism work, and one way that party leaders cultivate effective brokers is by rewarding those brokers who build large local followings with career opportunities and political power. Yet, careerist brokers who build large local followings can then use the threat of leaving the party with their local following to extract additional resources from the party leader. This forces a party leader to over-invest in conserving her organization and the voters organized through it, rather than using resources to attract new voters and expand the reach of her machine.Footnote 23
This dilemma is captured by the model, and the model indicates that parties with relatively less resources will be even more adversely affected by this dilemma. Party leaders earn more votes through their machine as brokers contribute more effort, an increase in α,Footnote 24 or as brokers become more skilled, an increase in θ. Yet, a broker earns more resources and a party leader earns less resources with an increase in α or an increase in θ. This dilemma becomes even more severe when competing parties can offer brokers valuable exit options, since a broker also earns more resources and a party leader earns less resources with an increase in ρ.
Party leaders benefit when brokers undertake more effort by earning a higher vote share from the constituencies organized by brokers, but are harmed by earning a lower vote share from the constituencies not organized by brokers. In the bargaining environment, an increase in α or in θ increases a party leader’s total vote share when
$\rho \,\lt {1 \over {(1{\minus}p)\sigma }}$
. But when
$$\rho \,\gt {1 \over {(1{\minus}p)\sigma }},$$
an increase in α or in θ decreases the party’s total vote share. When brokers have valuable exit options, effective-careerist brokers harm the party. Also effective brokers are more likely to harm the party when the party’s machine does not organize many voters, p is small, or when strategies of reaching voters outside of the machine are effective, σ is large. Moreover, the party leader’s utility declines at a faster rate with a decline in ρ when the value of α or θ is larger. Finally, if ρ is sufficiently large brokers cannot benefit by bargaining with their party leaders, and they will exit the party en masse.
Generally, then the model indicates that opposition parties incur higher costs from bargaining than competing incumbent parties, which can offer brokers valuable exit options. These opposition parties may have incentives to devise alternative mobilizing strategies that rely on fewer brokers, who are activist rather than careerist brokers. Thus, the model also provides insights into why ideological parties that rely less on clientelist exchange or even organize against this form of distributive politics may rise up against political machines. For example, the progressive movement in the early twentieth century in the United States, the PAN and PRD in Mexico, and third parties in Argentina, all relied heavily on non-clientelist electoral strategies, especially when they experienced resource disadvantages.Footnote 25
The Data
The quantitative data used to test the theory come from a unique survey that I and a team of scholars conducted in Argentina. We surveyed 800 subjects, and this survey is the first to draw upon a probability sample of brokers in any country. The sampling frame consists of city council members and their brokers in four provinces: Buenos Aires, Córdoba, Misiones, and San Luis. Although these provinces were not randomly selected, they have important distinctions that characterize much of Argentina. Buenos Aires and Córdoba have large urban populations and more competitive party systems. Misiones and San Luis are more rural and dominated by a single party. Individuals were randomly selected using a clustered random sampling design and are representative of city councilors and brokers in each province.Footnote 26 To my knowledge, this article along with that of Stokes et al.,Footnote 27 who use the same data, provide two of the most systematic analyses of political machines’ internal organization of any research that has been conducted.
Evaluating the Model
In this section, I test three implications and evaluate one core assumption from the model. First, I use regression analysis to test if careerist brokers will undertake more effort and produce more votes than party activist brokers. Next, I provide evidence that many brokers exit their parties, which is a central assumption of the model and compliments the rest of the analysis. Then, I use a survey experiment to test whether an effective broker can procure more resources than a less effective broker, even when doing so harms the broker’s party. Finally, I use a pairing of questions to test if brokers believe that a party over-invests in core neighborhoods and under-invests in swing neighborhoods.
Extracting Broker Effort
The survey indicates that brokers organize small groups relative to the electorate, which would make the collective action problem a challenge for activist brokers. The average number of brokers, who worked for a city councilor, ranged from about twenty-three brokers per councilor in Buenos Aires to fourteen brokers per councilor in Córdoba and San Luis.Footnote 28 A majority of brokers in every province reported that they can individually transport less than 200 voters to a rally. A broker’s ability to turn people out for political rallies is one of the primary metrics that party leaders use when evaluating a broker’s capacity and effort.Footnote 29 Municipal electoral data from seven provinces and 1,154 municipalities in Argentina show that the average number of voters who participated in a municipal election is 11,088.Footnote 30 Even a councilor who employs a team of brokers and controls 300 voters would control only 2.71 percent of the votes in an average municipality, and would control much less in the larger municipalities.
Since party leaders must depend upon many brokers who organize small groups relative to the electorate, the model predicts that careerist brokers will outperform activist brokers. To test this prediction, I test whether careerist brokers report a larger capacity to mobilize voters for political rallies. In addition to providing a measure of a broker’s capacity, mobilizing for rallies also contributes to a party’s electoral success. Over 95 percent of the brokers believed that at least half of all the voters who turned out to political rally would vote for the broker’s candidate. Over 70 percent of the brokers believed that at least 70 percent of all the voters who turned out to political rally would vote for the broker’s candidate.
The brokers’ survey provides a measure of a broker’s mobilizing capacity: the number of voters a broker can turn out for a rally. The survey asks respondents: Speaking of political rallies, approximately how many people would you say you can transport: fewer than 50 people, between 50 and 100 people, between 100 and 200 people, between 200 and 300 people or more than 300 people? The variable, termed Rally Capacity, is an interval variable; its values range from 1 to 5.
Although this measure might suffer from a social desirability bias as brokers overestimate their capacity, careerist brokers should not be any more likely to overestimate their capacity than party activist brokers. For careerist brokers, their capacity to mobilize voters for a rally enhances their career and power. For party activist brokers, their capacity to mobilize voters enhances their contribution to their party’s victory. Moreover, this measure is one of the most systematic measures of brokers’ mobilizational capacities ever collected for such a large sample.
The simplest way to distinguish between careerist and activist brokers would be to directly ask the brokers whether they work in politics to empower themselves or to help their party win elections. Unfortunately, such a direct question would not be likely to elicit honest responses. Careerist brokers may believe that expressing strong support for their current party is a good strategy for advancing their career. The survey provides three less direct measures that indicate a broker is prioritizing her own power and career.
The first measure is a variable entitled Career Ambition that distinguishes brokers who express an ambition to advance their own careers within a party. This dummy variable is coded as 1 if a respondent expressed a preference to run for a powerful office even if she was not certain of winning. The variable is coded as 0 if the respondent expressed a preference to run for a less powerful office that she was certain of winning. By assigning more risk to the more powerful office, this question creates a cost for the higher office and prevents all brokers from taking the default answer of preferring the more powerful office. Thus, Career Ambition distinguishes the brokers who place a stronger priority on advancing their own power and political career, and are even willing to pay the cost of more risk to advance their careers.
A variable entitled Career Reward distinguishes brokers who are paid for their political activities. This is a dummy variable that is coded as 1 if the broker reported that she is paid for her political work. It is coded as 0 if the broker reports that she has never been paid for her political work. This measure identifies brokers who have achieved private gain from their political career.
Finally, a variable entitled Competitive distinguishes between brokers who view competition from brokers within their own party as presenting more risk than brokers who work for opposing parties. In the survey, brokers were asked to imagine two brokers compete for support from voters who live in the same neighborhood. This variable is a dummy variable that is coded as 1 if brokers reported that competition from a broker who works for the same party presents a risker situation than a broker from a competing party. The variable is coded as 0 if the broker responded that a broker from a competing party presents more risk. Careerist brokers should view brokers within their own party as the primary source of risk since they are competing with these individuals for career opportunities, resources, and power. Party activists should view brokers from the opposing party as the primary source of risk, since brokers from the opposing party can cause an electoral loss for the respondent’s party.
I evaluate the model with six linear regression models, three of which contain control variables that are defined in the appendix. All three independent variables should have a positive and significant effect on Rally Capacity. A positive and significant relationship effect for each of these variables would indicate that brokers who are seeking or have achieved private career objectives have a greater mobilizing capacity.
Table 1 reports the results of the linear regression models. Brokers who seek or have achieved private career objectives have a larger mobilizing capacity. Competitive has the weakest effect on Rally Capacity, but in the model with controls the coefficient increases in magnitude and becomes significant. These results indicate that party leaders can use the career objectives of brokers to extract effort.
Table 1 Broker Motivation and Rally Turnout
Notes: Significance levels: †10%, *5%, **1%, ††Includes controls. The standard errors are adjusted to account for the design effects of the survey sampling method. Controls include: Provinces, Municipalities, Parties, Age, Councilor Dummy, Gender, Income, Education, Survey Version.
I now test two counterarguments to the model. The most direct counterargument is that brokers who are particularly loyal to a party will exert more effort than other brokers to help their party win elections. A less obvious counterargument is that brokers who reside in swing neighborhoods will not experience a collective action problem and will contribute more effort since the brokers believe their effort will have a larger effect on their party’s victory. Objectively, brokers who live in swing neighborhoods will not have a more meaningful effect on their party’s probability of victory, since municipal elections utilize closed-list proportional representation systems, which do not correspond to neighborhoods. But inter-party competition should be more visceral for these brokers, and they may believe that their efforts are more likely to have an effect on an electoral outcome. Importantly, finding that brokers undertake more effort when they live in competitive neighborhoods would indicate that inter-party competition can motivate brokers, which challenges my theory.
To identify brokers who are loyal to their party, I develop a variable titled No Exit. This dummy variable assumes the value of 1 if the broker has never abandoned her party in her career, and assumes the value of 0 if the broker has abandoned her party at least once in her career. Brokers who have not abandoned their party should on average be more committed to their party’s victory.Footnote 31
To measure the inter-party political competition that a broker faces, I develop a variable entitled Core. In the survey, each respondent was asked how many brokers work in her neighborhood. The respondent was then asked how many brokers from her party work in her neighborhood. Core is defined as the absolute value of:
$$0.5{\minus}{{Number\,of\,Respondent \rsquo s\,Co{\hbox{-}}Partisans\,in\,Neighborhood} \over {Total\,Number\,of\,Brokers\,in\,Neighborhood}}.$$
If more than half of the brokers in a neighborhood share the respondent’s party, then Core increases with an increase in the percentage of the respondent’s co-partisans. If less than half of the brokers share the respondent’s party, then Core increases with a decrease in the percentage of the respondent’s co-partisans. If Core has a negative and significant effect on Rally Capacity, this would indicate that brokers contribute more effort when their neighborhood is highly contested with inter-party competition.
Table 2 reports the results of four linear regression models, in which I regress No Exit and Core on Rally Capacity. The results indicate that party activist brokers do not have a greater mobilizing capacity than other brokers. Together Tables 1 and 2 indicate that private motivations can be utilized to extract effort from political brokers, but a stronger commitment to a political party or high levels of inter-party competition does not necessarily cause a broker to undertake more effort.
Table 2 Broker Partisan Loyalty and Rally Turnout
Notes: None of these values are significant. ††Includes controls. The standard errors are adjusted to account for the design effects of the survey sampling method. Controls include: Province, Municipality, Party, Age, Councilor Dummy, Gender, Income, Education.
Broker Exit and Withdrawal of Support
Although careerist brokers exert substantial effort for their parties, they also make parties vulnerable to bargaining problems and broker exit. Unlike all other theories that model political machines as multi-actor organizations, this theory emphasizes the effect of a broker’s exit options by assuming that a broker can withdraw a proportion of the voters in her constituency from her party. In doing so, this approach identifies a key organizational vulnerability in political machines, and a reason why party leaders may pursue alternative forms of distributive politics and electoral strategies.
The survey data provide substantial evidence of broker exit. Across all four provinces 23.65 percent of the brokers abandoned a party at least once in their career.Footnote 32 In the province with the highest rate of broker exit, Missions, 48.41 percent of the brokers abandoned at least one party. In the province with the lowest rate, San Luis, 19.74 percent of the brokers abandoned at least one party.Footnote 33 Moreover, 57.75 percent of the brokers indicated that a broker would either stop organizing, or abandon his current party boss or party, if his party boss transferred resources from him and awarded the resources to another party broker to increase vote share.Footnote 34 Clearly the threat of exit is credible, and the high rates of exit indicate that brokers are not entirely bound by partisan attachment. Moreover, scholars have observed similar rates of exit by party actors in other settings where clientelism is pervasive such as Brazil.Footnote 35 Still, the model also indicates that the value of exit options will affect a broker’s bargaining power, and that the threat of broker exit will force parties to over-invest in voters who are already captured by the machine’s brokers.
Overinvestment in Effective Brokers
The theory predicts that effective brokers who mobilize many voters can use the threat of exit to encourage party leaders to over-invest in their constituencies. This prediction identifies an important vulnerability in political machines that has been overlooked by most of the formal literature on clientelism. Most theories assume parties can distribute resources in any manner that best serves the party’s electoral goal. Even those models that consider multi-actor parties do not model intra-party bargaining dynamics and do not predict that effective brokers will capture resources that could be better invested elsewhere.Footnote 36
The respondents were asked to:
Imagine that a political boss thinks he can increase the number of votes by taking resources from one broker and giving the resources to another broker. The broker who loses resources has many (few) voters and all (very few) of the voters are loyal to him. How difficult would it be for the party leader to do this? Very Easy, Easy, Neither Easy nor Difficult, Difficult, or Very Difficult.
The experimental design of this question allows me to test whether effective brokers are more able to procure resources, even if doing so adversely affects their party. Half of the respondents were randomly assigned a treatment that asked them to consider a broker who had many voters and the other half were randomly assigned a treatment that asked them to consider a broker who had few voters. An effective broker should have many voters opposed to few voters. In terms of the model, the size of broker’s local following and the broker’s share of resources increases with effort, α, and skill, θ. If the respondents say that it is just as easy to take resources away from a broker with a large number of voters as it is from a broker with few voters, then the prediction from my model is incorrect.
My theory further anticipates that a broker with many voters all of whom are loyal should have more bargaining strength than brokers who have few voters. Yet my theory does not anticipate that a broker with many voters of whom very few are loyal will have more bargaining strength than a broker who has few voters. In this case, a broker may have many voters, but she would be less able to withdraw them to punish a party leader. In the survey, half of the respondents were asked to consider a broker with voters who are all loyal and half were asked to consider a broker who has only a few loyal voters. Below I use the interaction between loyalty and the number of voters that a broker has to test whether a broker should cultivate many voters, who are also loyal, to procure more resources.
To test the predictions, I perform three difference-in-means tests. The dependent variable consists of the five answers very easy, …,very difficult measured on an interval scale of 1 to 5. In the first test, I subtract the mean response of those who were asked to consider a broker who has few voters from the mean response from those who were asked to consider a broker who has many voters. In the subsequent tests, I make the same calculation, but I divide the respondents into those who were asked to consider a broker all of whose voters were loyal, and those who were asked to consider a broker who had only a few loyal voters. Table 3 summarizes the Sample Average Treatment Effects (SATE) by reporting the 95 percent and 90 percent confidence intervals for the difference in mean responses between the various treatments.
Table 3 Confidence Intervals for the Difference in Means of Difficulty between Treatments
The tests support the prediction that an effective broker is able to procure more resources, even when doing so adversely affects her party’s vote share. Brokers who have many voters are able to procure more resources than brokers who have few voters. Moreover, it is more difficult to take resources from a broker who has many voters than from a broker who has few voters only if these voters are loyal to the broker. When voters are not loyal to a broker, the effect of the broker’s group size is no longer significant.
During my fieldwork, many brokers articulated dynamics that support this result. For example, one broker argued:
It’s relative. There are people who have a lot of influence … If they go, they take everyone with them. Everyone goes with a strong broker. But with many brokers, the same individuals would return to the party and not follow the broker … It depends on the reputation of the broker, the influence he has with his people.Footnote 37
Overinvestment in Captured Neighborhoods
If parties are forced to award resources to brokers who have large followings to prevent defection, then parties will be forced to overinvest in constituencies that already provide strong support for the party. Effective brokers earn more resources from a party by making voters more responsive to resources, but they also have more bargaining power, which causes party leaders to give them more resources than they would without intra-party bargaining. To test this prediction, I present evidence that shows a considerable discrepancy between the distributive strategies that party members think parties should use and the strategies that are actually used. The respondents were given the following scenario:
Imagine that your party is in a very competitive electoral campaign. One broker works in a neighborhood that has historically voted for your candidates. There is another broker who works in a neighborhood where half of the voters are undecided. How would you prefer the boss to distribute resources? Give more resources to the broker whose neighborhood has historically voted for the broker’s candidate or prioritize the broker whose neighborhood has many undecided voters?
The follow-up question asked respondents ‘Which of the brokers do you think will be more successful in obtaining resources for their neighborhood?’Footnote 38
The model predicts that effective brokers force parties to over-invest in neighborhoods that are already captured by the party. If this prediction is correct, then more respondents should report that brokers in captured neighborhoods will be more successful in obtaining resources, than those who say it would be ideal to invest in captured neighborhoods. To test this prediction I perform a difference-in-means test. The respondent’s answer to the first question is coded as 1 if the respondent prefers that her party distributes resources to the captured neighborhood, and 0 otherwise. The respondent’s second answer is coded as 1 if the respondent believes that the broker working in the captured neighborhood is more likely to be successful in obtaining resources, and 0 otherwise. I then calculate a difference between the mean response for each question by subtracting the mean response of the second question, the respondent’s belief about their party’s actual distribution, from the mean response to the first question, the respondent’s ideal distribution.
Table 4 shows that the difference in means is significantly less than 0, which indicates that a larger number of brokers believe that brokers working in captured neighborhoods will be successful in obtaining resources than the number of brokers who would prefer that these brokers actually receive the resources. While an estimated 54.1 percent of brokers believed a broker in a captured neighborhood will be more successful in obtaining resources, only 25.2 percent of brokers preferred that their party prioritize the captured neighborhoods. The finding indicates substantial overinvestment in neighborhoods that already support the party.
Table 4 Confidence Intervals for the Difference in Means of Difficulty between Ideal and Actual Distribution
Notes: The confidence intervals use Linearized Standard Errors to account for the sampling method.
This finding stands in contrast to other multi-actor theories of political machines. For example, Stokes et al. predict that parties target swing groups, while brokers target core voters.Footnote 39 In particular, Stokes et al. predict that parties will target ‘ideologically “mobile” groups,’ that have many voters who could be won over by a small increase in resources distributed to this group.Footnote 40 Without explicitly attributing bargaining power to brokers, these authors predict that parties, seeking to maximize the probability of victory, will distribute resources to constituencies with maximum efficiency. Although brokers seem to share these authors’ views about the efficient distributive strategy of a party, brokers depart from this theory’s prediction when considering how a party actually distributes resources.
This finding is also inconsistent with a theory of party building that predicts that parties should invest in captured neighborhoods to take advantage of the efficient vote-getting operations in these neighborhoods. In contrast to a theory of party building, a large majority of brokers agree with theories that predict parties should target swing neighborhoods as the most efficient strategy. Yet in spite of preferences regarding an ideal distribution, brokers believe parties end up investing in captured neighborhoods, which is consistent with a theory of intra-party bargaining.
Conclusion
Leaders of political machines face a dilemma. To make clientelist exchange work, they need effective brokers who undertake substantial effort over long periods of time. By relying on careerist motivations, party leaders can use the promise of resources, public employment positions, and political offices to extract effort from brokers. Yet, careerist motivations create organizational vulnerabilities that can hinder a machine’s ability to win elections. As others have argued, careerist brokers create principal–agent problems that waste party resources. In addition to these principal–agent problems, I have shown that brokers also create bargaining problems. The very brokers who effectively cultivate constituencies are the same ones who impose the most severe costs when they gain valuable exit options.
Using a unique survey that draws upon a probability sample of brokers, this article has developed three general findings. First, brokers organize small groups relative to the electorate as a whole and have very little chance of affecting an electoral outcome. Second, careerist brokers, operationalized in a variety of ways, mobilize more individuals for electoral rallies, which is a crucial measure of broker capacity. Party activist brokers do not have such a large capacity for rallies. Third, brokers with large and loyal followings are able to extract more resources from their party even when it harms their party’s vote share. Also parties over-invest in neighborhoods where they already have strong support.
On one hand, these findings indicate that political machines, even seemingly dominant political machines, have deep vulnerabilities. Scandals, economic volatility, economic reform, electoral losses, and even the death of a party leader can all have dramatic effects on machines. These organizations succeed by allowing a large number of individuals to build careers and procure power by contributing to the electoral dominance of a machine. But, if these individuals suddenly gain better career opportunities elsewhere, the motivation to advance their career might trump any partisan attachment that they have with their current party. The survey provides substantial evidence of broker exit. Both Desposato and Novaes have documented similar types of exit in Brazil.Footnote 41 Mayhew documents the collapse of Republican machines and the mass defection of precinct captains to the Democratic party in Philadelphia and Pittsburg.Footnote 42 Johnston argues that under the threat of defection the New Haven Democratic Town Committee chair forced the party to give too many summer jobs to the dwindling Italian community, when it should have prioritized the black community.Footnote 43 These results provide reasons why elected candidates may seek to circumvent their party brokers.
On the other hand, these findings suggest that machine politics can become deeply entrenched within a state. If machines succeed by providing career opportunities and power to large numbers of individuals, then these individuals have an important stake in retaining clientelist forms of distribution. Moreover, with the ability to exit, these brokers are not necessarily bound to losing parties and have the ability to push any party in power to give them resources and implement clientelist exchange. This problem is compounded by the fact that although voters may oppose clientelist exchange on a general level, they may also derive important benefits if they are organized by an effective broker. So even when a clientelist party loses an election, uprooting a clientelist network may remain a much more daunting task.
Supplementary Material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S0007123415000411
APPENDIX
Solving for L i * in Equation 5
Solving for the first order condition of Equation 4, we find that broker i’s labor contribution is defined by the following equation:
$$L_{i} ={{\alpha {p \over n}\theta {\minus}(1{\minus}\alpha )\mathop{\sum}\nolimits_{{\minus}i=1}^{n{\minus}1} {{p \over n}\theta L_{{{\minus}i}} } } \over {{p \over n}\theta }},$$
where L −i defines the labor and constituency size of a broker who is not broker i.
Since we know that each actor is solving the same maximization problem, and we have assumed that each broker’s constituency is the same size, Equation 17 can be rewritten as:
$$L_{i}^{{\rm {\asterisk}}} ={{\alpha {p \over n}\theta {\minus}(1{\minus}\alpha )(n{\minus}1){p \over n}\theta L_{i}^{{\rm {\asterisk}}} } \over {{p \over n}\theta }}.$$
Simplifying terms yields Equation 5.
Proving that the Difference between Equations 6 and 9 Increases as n Increases
Denote the difference in the party’s vote share when brokers prioritize their own power minus the party’s vote share when brokers prioritize their party’s victory as Δπ(L *). By simplifying the difference between Equations 6 and 9 the difference Δπ(L *) can defined by the following equation:
$${\rm \Delta }\pi (L^{{\rm {\asterisk}}} )={{p\theta \alpha (1{\minus}\alpha )(n{\minus}1)} \over {1{\plus}(1{\minus}\alpha )(n{\minus}1)}}.$$
Equation 19 indicates that if n>1 and 0<α<1, a party’s vote share will be greater when brokers prioritize their own power rather than their own career.
Taking the derivative of Δπ(L *) with respect to n, we find that this difference increases with an increase in n. This derivative is defined as:
$${{\partial {\rm \Delta }\pi (L^{{\rm {\asterisk}}} )} \over {\partial n}}={{p\theta \alpha (1{\minus}\alpha )} \over {(\alpha {\plus}n(1{\minus}\alpha ))^{2} }}.$$
Since the denominator is a squared term it must be positive. Moreover, if 0<α<1, then the numerator is also strictly positive. Thus, Δπ(L*) increases with an increase in n.
Mixed Motivations
Consider a broker who derives utility from her own empowerment and her party’s victory in the same framework of this article. Then the broker’s utility function is defined as:
$$v_{i} (\pi (L),L_{i} )=(p\theta L_{i} )^{{(1{\minus}\alpha {\minus}\omega )}} \pi (L)^{\alpha } (1{\minus}L_{i} )^{\omega } .$$
With this utility function the equilibrium contribution of labor is:
$$L_{i}^{{\rm {\asterisk}}} ={{n(1{\minus}\alpha {\minus}\omega ){\plus}\alpha } \over {n(1{\minus}\alpha ){\plus}\alpha }}.$$
From Equation 22, we can see that brokers with mixed motivations will experience a collective action problem, but that this collective action problem grows less severe as the broker places a greater priority on her own local following. When α>0 and
$\omega \,\gt\, 0,\,{{\partial L_{i}^{{\rm {\asterisk}}} } \over {\partial n}}\lt 0$
. Thus, if a broker derives utility from the party’s victory and non-political activities, then the broker’s equilibrium effort will decrease with an increase in the number of broker in the party. If ω>0, then
${{\partial L_{i}^{{\rm {\asterisk}}} } \over {\partial \alpha }}\lt 0$
, which means that if the broker derives utility from non-political activities, then the broker’s equilibrium effort will decrease as the broker derives less utility from building a local following and more utility from the party’s victory.
Deriving Equations 14, 15, and 16
To solve this maximization problem we must first define the resources allocated to broker i as a function of L i . This requires solving Equation 13. To do this, first simplify the terms in Equation 13 and define the Lagrangian function:
$${\rm \Phi }(r_{i} ,r_{{pl}} ,\lambda )=\left( {{\rm \Pi }(r_{{pl}} ,L){\minus}\left( {{\rm \Pi }(r_{{pl}}^{{\rm {\asterisk}}} {\plus}r_{i}^{{\rm {\asterisk}}} ,L){\minus}{p \over n}\theta L_{i} } \right)} \right)^{\beta } \left( {r_{i} {\minus}\rho {p \over n}\theta L_{i} } \right)^{{(1{\minus}\beta )}} {\minus}\lambda \left( {1{\minus}r_{{pl}} {\minus}\mathop{\sum}\limits_{i=1}^n {r_{i} } } \right).$$
We can take advantage of the fact that each broker is facing the same bargaining problem, which means that r
i
will be equal for each broker. Without loss of generality, we only need to solve for
$r_{j}^{{\rm {\asterisk}}} ,\,r_{{pl}}^{{\rm {\asterisk}}} $
and λ, where j is an arbitrary broker. With some simplification we can write the first order conditions as:
$${{\partial {\rm \Phi }} \over {\partial r_{{pl}} }}={{\beta (1{\minus}p)\sigma } \over {{\rm \Pi }(r_{{pl}}^{{\rm {\asterisk}}} ,L){\minus}\left( {{\rm \Pi }(r_{{pl}}^{{\rm {\asterisk}}} {\plus}r_{j}^{{\rm {\asterisk}}} ,L){\minus}{p \over n}\theta L_{j} } \right)}}{\plus}\lambda =0,$$
$${{\partial {\rm \Phi }} \over {\partial r_{j} }}={{1{\minus}\beta } \over {r_{j}^{{\rm {\asterisk}}} {\minus}\rho {p \over n}\theta L_{j} }}{\plus}\lambda =0,\,and$$
$${{\partial {\rm \Phi }} \over {\partial \lambda }}=R{\minus}r_{{pl}} {\minus}\mathop{\sum}\nolimits_{i=1}^n {r_{i} } {\rm =0}{\rm .}$$
From these equations we get:
$$r_{j} (L_{j} )=L_{j} {{p\theta ((1{\minus}\beta ){\plus}(1{\minus}p)\rho \sigma \beta )} \over {n((1{\minus}p)\sigma )}}.$$
Equation 27 allows us to solve broker j’s maximization problem, which can now be defined as:
$$\mathop{{{\rm max}}}\limits_{{l_{j} }} {\left( {{p \over n}\theta L_{j} {\plus}L_{j} {{p\theta ((1{\minus}\beta ){\plus}(1{\minus}p)\rho \sigma \beta )} \over {n((1{\minus}p)\sigma )}}} \right)} \right^{\alpha } (1{\minus}L_{i} )^{{(1{\minus}\alpha )}} .$$
Taking the first order condition yields:
$${\alpha \over {L_{j} }}{\minus}{{1{\minus}\alpha } \over {1{\minus}L_{j} }}=0.$$
Equations 29 and 27 yield Equations 14, 15, and 16.
We can then define a party leader’s total vote share as a function of θ, α, and ρ:
$${\rm \Pi }(\theta ,\alpha ,\rho )=p\theta \alpha {\plus}1{\minus}{{p\theta \alpha ((1{\minus}\beta ){\plus}(1{\minus}p)\sigma \rho \beta )} \over {(1{\minus}p)\sigma }}.$$
Taking the partial derivative with respect to α or θ yields:
$${{\partial {\rm \Pi }(\theta ,\alpha ,\rho )} \over {\partial \alpha }}=p\theta (1{\minus}(1{\minus}\beta {\plus}(1{\minus}p)\beta \rho \sigma )),\,and$$
$${{\partial {\rm \Pi }(\theta ,\alpha ,\rho )} \over {\partial \theta }}=p\alpha (1{\minus}(1{\minus}\beta {\plus}(1{\minus}p)\beta \rho \sigma )).$$
From Equations 31 and 32, we can see
${{\partial {\rm \Pi }(\theta ,\alpha ,\rho )} \over {\partial \alpha }}\gt 0$
and
${{\partial {\rm \Pi }(\theta ,\alpha ,\rho )} \over {\partial \theta }}\gt 0$
when
$\rho \lt {1 \over {(1{\minus}p)\sigma }}$
.
Survey Design
The sample consists of 800 city council members and their brokers. The surveys were conducted from October 2009 through 2012. The sample was selected from four Argentine provinces: Buenos Aires, Córdoba, Misiones, and San Luis. The provinces were non-randomly selected. Within each province we used a random-clustered design. We first randomly selected municipalities, then randomly selected half of the city council members in each municipality, and finally we randomly selected one-third or one-fifthFootnote 44 of the brokers who worked for each selected city council member.Footnote 45
In some of the provinces, we restricted the municipalities that were included in the sampling frame. In Córdoba and Misiones all of the city council members and their brokers are in the sampling frame. In Buenos Aires we restricted the sampling frame to the Conurbano Bonaerense, which is comprised of twenty-four municipalities that surround Buenos Aires. We did this to focus on the municipalities that are known to have large party machines. Numerous studies note that many politicians use large territorial organizations to mobilize voters in the Conurbano.Footnote 46 In San Luis we excluded small villages (‘comunas’).
There were four versions of the survey that were systematically assigned to respondents. While most of the questions were uniform throughout the different versions of the survey, some questions were worded differently in each version of the survey. Each surveyor started with Version 1 of the survey and proceeded in numerical order with each subsequent interview. This systematic assignment should make the attributes of the respondents statistically independent of the survey version that they were assigned.Footnote 47
Descriptive Statistics and Control Variables
This subsection contains a description of the control variables and the descriptive statistics for the analysis. Table 5 contains the descriptive statistics for all of the variables used in the analysis in the section dealing with the evaluation of the model.
Table 5 Summary Statistics for Survey Data
