INTRODUCTION
A large body of research on politics in developing countries suggests that politicians tend to distribute material benefits to potential supporters through clientele networks (Stokes Reference Stokes2005; Nichter Reference Nichter2008; Stokes & Dunning Reference Stokes and Dunning2008; Gans-Morse et al. Reference Gans-Morse, Mazzuca and Nichter2009; Diaz-Cayeros et al. Reference Diaz-Cayeros, Estévez and Magaloni2012; Stokes et al. Reference Stokes, Dunning, Nazareno and Brusco2013). By controlling the levers of public spending, politicians can use their discretion, especially in contexts of weak formal institutions, to channel valued public resources to certain voters and maximise their chances of re-election. Thus when much of Africa transitioned to democracy in the early 1990s, some scholars argued that if political competition grows, clientelism and other forms of discretionary or non-programmatic distribution of public resource will dominate politics (van de Walle Reference van de Walle, Kitschelt and Wilkinson2007).
In the last decade or more, multiparty elections in a growing number of Africa's new democracies have become more competitive. At the end of 2010, more than 60% of countries in the region had conducted three or more successive elections without interruptions and about 30% had experienced executive and/or legislative turnovers (Lindberg Reference Lindberg2009; Weghorst & Lindberg Reference Weghorst and Lindberg2013). Yet over this period, more and more governments in the region and the developing world in general, have initiated programmes to transfer cash and provide complementary public services directly to citizens using mainly economic and technical criteria. In fact more than 120 of these programmes were implemented in sub-Saharan Africa between 2000 and 2009 (Garcia & Moore Reference Garcia and Moore2012). Studies on a number of these programmes show that partisan political criteria play no role in predicting who does or does not receive benefits (Fried Reference Fried2012; De La O Reference De La O2013).Footnote 1 Some scholars speculate that the rising popularity among politicians of rule-based public transfers to citizens in the developing world is indicative of a potential decline of clientelism or non-programmatic distributive politics in those countries (Stokes et al. Reference Stokes, Dunning, Nazareno and Brusco2013).
The last 15 or more years have also seen the introduction of Constituency Development Funds (CDFs) in many developing countries with increasing involvement of Members of Parliament (MPs) in grassroots development. CDFs are monies drawn from national revenues and allocated to MPs to enable them to undertake development projects in their districts. Cross-national analysis of these programmes reveals significant variations in the degree of control and mechanisms of distribution of benefits by individual MPs (Tshangana Reference Tshangana2012). Whereas in some countries the CDF law gives MPs maximum control over the management and allocation of CDFs, in others MPs have approved or amended these laws to reduce or eliminate their own influence on distribution. For instance, Kenya's CDF legislation has been revised twice since its introduction in 2003. The original legislation, enacted in 2003, gave MPs near absolute control over the management and distribution of CDF benefits to their constituents. In the most recent (2013) revision, MPs approved a new law that significantly constrains their own ability to influence the actual allocation of CDF benefits.Footnote 2
Non-discretionary or rule-based distribution of benefits to voters by politicians, especially in the context of growing political competition seems inconsistent with predictions of standard models of distributive politics. Standard theories expect politicians to use their power, sometimes including manipulating or overriding formal rules, to channel resources to the ‘right voters’ and maximise their chances of re-election (Golden & Min Reference Golden and Min2013). Why would politicians adopt resource allocation rules and procedures that limit their own power and influence on distribution? I address this question in this paper. I propose and test a theory about the sources of non-discretionary distributive politics in Africa's new democracies using Ghana as a case study.
This question is particularly important in Africa's new democracies where formal institutions are not very strong and economic and social conditions create political incentives that are more compatible with discretionary distribution of public resources.
VOTER BEHAVIOUR
Voters may develop strong attachment to, and vote consistently for certain parties for many reasons including ideology, social identity or other connections that enable them to enjoy material and/or psychic benefits (Ferree Reference Ferree2006; Chandra Reference Chandra2007; Fridy Reference Fridy2007; Habyarimana et al. Reference Habyarimana, Humphreys, Posner and Weinstein2007; Carlson Reference Carlson2010). In a multiparty democratic system, voters may be classified into three broad groups: strongly attached voters, weakly attached voters and unattached voters. Much of the literature on voter behaviour often characterises unattached and weakly attached voters as ‘persuadable’ or ‘swing’ (Lindbeck & Weibull Reference Lindbeck and Weibull1987; Campbell Reference Campbell2008; Lindberg & Morrison Reference Lindberg and Morrison2008). The extent of voter attachment to political parties features prominently in the variety of approaches used to measure and evaluate the effects of ‘swing’ or ‘core’ voting on the behaviour of politicians as well as various policy outcomes. Some of the popular measures include self-reported ambivalence or lack of party affiliation (Lindbeck & Weibull Reference Lindbeck and Weibull1987; Hoffman et al. Reference Hoffman, Gibson, Ferree and Long2009), past voting behaviour such as voting straight ticket or split ticket (Lindberg & Morrison Reference Lindberg and Morrison2008; Keefer & Khemani Reference Keefer and Khemani2009), or some combination of these measures (Weghorst & Lindberg Reference Weghorst and Lindberg2013). Though there is no consensus on the best approach, unattached and weakly attached voters are conceptually distinguishable from strongly attached voters in that the voting behaviour of the first two groups is presumed to be subject to greater shocks than that of the third. In other words, unattached and weakly attached voters should be more likely to switch their votes between different parties or politicians over time than strongly attached voters.Footnote 3
The impact of the hypothesised behaviour of unattached or weakly attached voters on elections, the behaviour of politicians and policy outcomes has received considerable attention in the literature (Cox Reference Cox1987; Dixit & Londregan Reference Dixit and Londregan1996; Leech Reference Leech2003; Lindberg & Morrison Reference Lindberg and Morrison2005; Stokes Reference Stokes2005; Mayer Reference Mayer2008; Keefer & Khemani Reference Keefer and Khemani2009; Fridy Reference Fridy2012), but the existing scholarship tends to emphasise the influential role of incumbent politicians in public spending. Because they have control over public spending, incumbent politicians could disproportionately favour their own loyal voters in the distribution of public resources (Cox & McCubbins Reference Cox and McCubbins1986) or channel benefits to unattached or weakly attached voters to persuade them and thus broaden their electoral support base (Dixit & Londregan Reference Dixit and Londregan1996; Stokes Reference Stokes2005; Magaloni et al. Reference Magaloni, Diaz-Cayeros, Estévez, Kitschelt and Wilkinson2007). They could also employ a combination of both methods as well as decide the type of goods – public or private – to provide to each group in order to maximise the number of votes they receive (Magaloni et al. Reference Magaloni, Diaz-Cayeros, Estévez, Kitschelt and Wilkinson2007). And where possible, they can punish opposition voters by withholding services. The standard approach in the literature implicitly assumes that when faced with strong electoral contests, incumbent politicians would be more likely to maximise control over the distribution of benefits to voters, especially in developing democracies where formal procedures are not well institutionalised. However, this is not always the case as evidenced by the rising popularity of targeted, rule-based government transfers to citizens in many new democracies and the instances where Members of Parliament in some of these countries relinquish control over the distribution of valued benefits to their constituents.
Much research has been done in other contexts on how competitive configurations and economic empowerment create incentives for politicians to acquiesce in reforms that reduce or eliminate political discretion over electorally useful resources (Geddes Reference Geddes1991; Lehoucq & Molina Reference Lehoucq and Molina2002; Finkel Reference Finkel2008) or opt out of clientelism and other forms of discretionary distribution of resources (Weitz-Shapiro Reference Weitz-Shapiro2012). To my knowledge, this is the first paper to examine how voter attachments to political parties affect politicians’ decisions about how to distribute benefits to voters in a new democracy.
EXPLAINING RULE-BASED DISTRIBUTION OF PUBLIC RESOURCES IN NEW DEMOCRACIES
When would politicians willingly constrain their own ability to influence the distribution of valued benefits to voters? I argue that the answer lies in the distribution of voter attachments to political parties. Non-discretionary distributive rules enable politicians to extend benefits to voters outside their circle of loyal voters, particularly unattached or weakly attached voters, and potentially broaden their electoral support without alienating their loyal voters who may otherwise feel betrayed if they fail to receive favours. Unattached and/or weakly attached voters who gain access to valued benefits may reward the incumbent at the polls because of the benefits. However, if incumbents were to use their own power and influence to channel resources to these ‘outsiders’ in order to broaden electoral support for the current election, they risk alienating some current loyalists in future elections (Diaz-Cayeros et al. Reference Diaz-Cayeros, Estévez and Magaloni2012). Sticking to broad-based, impersonal rules helps to minimise this risk. Current incumbent loyalists would be less likely to abandon their favourite party if allocations are governed by impersonal rules and the actual beneficiaries are people who are worse off than they are, compared with the situation where the party or politician disproportionately favours unattached or weakly attached voters who may be similar to them or perhaps even better off than they are. Thus non-discretionary distributive rules create a convenient opportunity for politicians to shift blame for the broader outcome of the allocations and in this way circumvent their loyal voters in a relatively less offensive manner.Footnote 4
Second, non-discretionary distributive rules have a signalling effect: they allow incumbent politicians to credibly signal to voters that they are committed to unbiased and efficient use of public resources. Fox (Reference Fox2007) argues that when legislators are concerned about their chances of re-election, they would be more likely to select policies that would lead the public to believe that they are unbiased. This signal helps to further broaden incumbents’ electoral support among unattached or weakly attached voters. Recent research shows that voting behaviour in Africa's new democracies is increasingly shaped by voter evaluations of the performance of incumbent politicians. Voters are more likely to cast their ballots for politicians whom they believe have performed well in areas such as overall economic management, provision of collective and developmental goods, and/or other dimensions of public service delivery that they care most about (Youde Reference Youde2005; Lindberg & Morrison Reference Lindberg and Morrison2008; Fridy Reference Fridy2007; Bratton Reference Bratton2013). However, this performance-based voting is concentrated among voters with weak or no partisan commitments. For instance, Weghorst & Lindberg (Reference Weghorst and Lindberg2013) argue and show that MP performance, as measured by the provision of collective or broad-based goods to constituents, attracts swing voters in Ghana. Individual-level survey data from the Afrobarometer reveal a similar pattern. Analysis of the 2008 Afrobarometer survey data for Ghana shows that respondents who approve of the public service delivery performance of their local government are more likely to say they would vote for the party in government in future elections than those who disapprove. But this effect is significant only among the subset of voters who report no affiliation to a political party (0·75 vs. 0·62, p = 0·000).Footnote 5 That is, unaffiliated voters who approve of the service delivery performance of local governments are significantly more likely to cast their ballots for the incumbent party in future elections than those who disapprove.
The preceding argument leads to the following expectation: the likelihood of rule-based or non-discretionary distribution of benefits to voters by politicians would be higher in places with large numbers of unattached or weakly attached voters. I use survey data on allocations of the Members of Parliament development funds in 22 electoral districts in the Brong Ahafo region of Ghana to test this prediction. The results provide strong support for the hypothesis: I find that allocations of the MPs’ development funds are largely governed by broad-based, impersonal rules in districts where voters demonstrate weak attachment to political parties.
This paper makes three contributions. First, this is one of the first projects to explore the political and policy implications of voter attachment to political parties in the context of a developing democracy. Keefer & Khemani (Reference Keefer and Khemani2009) argue and show that strong voter attachment to political parties reduces legislator incentives to provide constituency service in India. The paper contributes and extends this literature by focusing on Africa, where this question is relatively underexplored, highlighting the important role of unattached voters for political accountability. Second, the paper adds to recent research that shows that performance-based voting has increased across Africa (Fridy Reference Fridy2007; Bratton Reference Bratton2013; Weghorst & Lindberg Reference Weghorst and Lindberg2013), a region where voting in elections is often presumed to be along the lines of ethnicity, clientelism and/or patronage. Third, it sheds light on the growing political support for targeted, rules-based distribution of public resources among politicians in developing democracies (Stokes et al. Reference Stokes, Dunning, Nazareno and Brusco2013) and thus advances understanding of the conditions under which politicians would implement or acquiesce in policy and administrative reforms that reduce or eliminate political control over electorally useful resources (Geddes Reference Geddes1991; Lehoucq & Molina Reference Lehoucq and Molina2002; Magaloni et al. Reference Magaloni, Diaz-Cayeros, Estévez, Kitschelt and Wilkinson2007; Finkel Reference Finkel2008; Weitz-Shapiro Reference Weitz-Shapiro2012).
The rest of the paper proceeds as follows. In the next section, I describe the empirical setting, including the distribution of legislative seats in the study region. This is followed by a brief overview of the Members of Parliament development fund in Ghana and a detailed description of the survey and data collection. Next, I describe the variables and measurement and specify the empirical model. Finally, I present and discuss the results of the analysis and conclude with some implications of the findings and suggestions for future research.
EMPIRICAL SETTING
Ghana is one of the most electorally competitive new democracies in Africa. Since 2000, national elections have generally been free, fair and peaceful. Two parties dominate Ghanaian politics: the current governing National Democratic Congress (NDC) party and the main opposition New Patriotic Party (NPP) which was in power between 2001 and 2008. The president is directly elected in a majority-runoff system from a single national constituency, while members of parliament (MPs) are elected by first-past-the-post plurality electoral rules from single member districts known as constituencies. Parliamentary and the first round of the presidential elections are held concurrently and the president and the 275-member parliament are each elected to a four-year term. The president can only serve for two terms but there is no term limit for members of parliament.
The Brong Ahafo region, one of the most competitive regions in the country, has a total of 29 electoral districts (or constituencies) and hence 29 parliamentary seats. In the December 2012 elections, the ruling NDC party won 16 seats and the main opposition NPP party took the remaining 13 seats. The survey data used for the analysis in this paper were collected in 22 of the 29 constituencies.Footnote 6 In these constituencies, the ruling NDC party holds 10 seats and the main opposition NPP party holds 12 seats. The survey focused on allocations of development resources in the 22 districts with particular emphasis on allocations of the Members of Parliament (MPs) fund. Local government, known as District Assemblies in Ghana, is the level of government at which the national budget is allocated and spent. Ten per cent of total national revenue is shared annually among all districts based on a formula approved by parliament. For most districts, these central government transfers are the main source of funding for development and recurrent expenditures. In fact some districts depend almost entirely on the central government transfers to finance all their development activities. Each Member of Parliament (MP) is allocated up to 5% of their respective local governments’ share of the central government transfers. The next section presents a brief overview of the MPs’ development fund in Ghana.
The Brong Ahafo region is ideal for this research because the political and social context is largely typical of the broader Ghanaian population. First, elections are highly competitive and vary significantly across electoral districts. Second, though often considered a predominantly Akan-speaking area, many districts in the region are ethnically diverse. More importantly, just as in the broader Ghanaian population, the correlation between ethnicity and partisan attachment is very low. In other words the ethnic composition of a typical district in Ghana does not predict the extent of partisan attachment.Footnote 7
THE MPS' COMMON FUND IN GHANA
The Members of Parliament share of the District Assemblies Common Fund (MPDACF), known elsewhere as Constituency Development Funds (CDFs), is a significant source of funding for Ghanaian MPs. Most MPs receive between US$40,000 and US$60,000 annually from the central government transfers to local government to support the development needs of their constituents.Footnote 8 In the context of a developing country like Ghana, these amounts are substantial. The current (2015) daily minimum wage in Ghana is about US$1.7 (i.e. 7 Ghana Cedis). Thus the annual allocation to one MP can pay the gross earnings of nearly 90 minimum wage workers for 12 months. The formal guidelines for allocating these resources leave the choice of project areas and beneficiaries to the discretion of MPs. The main requirement is that MPs must choose and execute projects and programmes that are part of, or at least consistent with, the development priorities of their respective local governments, but ensuring that they do not duplicate those already completed by their local governments. However, in most cases, the projects and programmes that MPs actually choose to finance are easily captured in the books in a way that fits in the local government development agenda, even if the actual purpose or outcome diverges from that agenda. For instance handing out cash to college students from the district is frequently captured under scholarships for higher education, which is almost always on the agenda of most local governments. This means that MPs have the opportunity to control and allocate these resources in ways that would maximise their vote shares. However, the extent of control and the mechanisms used in allocating these resources vary widely across individual MPs. While some MPs have maintained full and personal control over MPDACF allocations, others have, to different degrees, introduced structures and rules to limit their own ability to influence the actual allocations.
I argue that the distribution of voter attachments to political parties in different districts explains the variation in MPDACF allocations. I expect allocations of the MPDACF to be non-discretionary or rule-based in districts with large numbers of unattached or weakly attached voters. The MPDACF is common knowledge in Ghana. Most voters know that MPs are allocated a portion of the central government transfers to local governments annually to undertake development projects in their respective districts. Beneficiaries of the MPs’ Common Fund – communities and individuals – are also often common knowledge within districts. When MPs use their Common Fund resources to finance the construction of new classroom blocks, provide safe water, or purchase corn mills for some communities, or pay health insurance premiums and education expenses for some constituents, information on such projects become public knowledge sooner or later. In many cases, projects supported by the MPDACF are visibly labelled as such and in their campaigns, MPs actively claim credit for providing these projects. Allocations of the MPDACF therefore serve as an effective tool for MPs to credibly signal their ‘type’ to constituents if or when they need to do so.
DATA AND MEASUREMENT
The survey
The survey involved interviews with local government officials responsible for disbursing the MPDACF and a random sample of voters from each district. I hired, trained and deployed a team of research assistants to collect detailed information on allocations of the MPDACF from local government administrators who handle the actual disbursements and to interview nearly 1,000 voters selected randomly from the 22 electoral districts. I use the information provided by the district administrators to measure the extent of MPs’ discretion in the actual allocation of the MPDACF and the individual survey data to measure voter attachment to political parties.
Dependent variable: MPs’ discretion in MPDACF allocations
The first survey asked the local government administrators about the process and criteria used in selecting beneficiaries – communities and individuals – of the MPDACF. The administrators responded to a small set of close-ended background questions and also wrote out in detail, how the MPDACF is actually allocated, backed with some examples from recent allocations. I then developed simple coding rules for the narratives provided by the administrators as follows: I assign a value of 1 (largely discretionary) for descriptions that suggest that allocations of the MPDACF are mainly or largely at the discretion of the MP. These were the cases where the administrators reported that all allocation decisions rest with the MP and that she or he chooses the projects and programmes to finance and also handles or oversees the actual disbursements. An example of a constituency that scored 1 is Techiman South where the administrator wrote:
The selection of beneficiary towns and people is based on requests made to the MP; the MP's own personal judgment; and political support base.
And for individual beneficiaries, the administrator wrote:
It is based on his [the MP's] own discretion.
Six of the 22 districts reported similar approaches to the MPDACF allocations: the administrators take instructions from the MP and disburse the funds accordingly.
For districts where the administrators reported a combination of objective criteria – e.g. economic and/or technical criteria – and MPs’ personal judgments or political considerations, I coded those as 2 (partly non-discretionary). These were mainly those districts where an individual or group is tasked to conduct initial screening, using need-based criteria to identify potential beneficiaries but the MP has to approve before disbursements are made. An example is Kintampo North district where the administrator reported that allocations are based on:
Gender considerations, special needs, disability and the MP's political support base. We disburse to these people once we get the go ahead from the honourable MP.
A little over 40% of the districts (nine of the 22 districts) scored a 2 on MPDACF allocations. Finally, I coded as 3 (largely non-discretionary) those cases where the description and examples suggests a limited role for the MP. Dormaa Central constituency is one such example. Here the administrator wrote:
The selection of projects for implementation is based on the Assembly's Medium Term Development Plan.Footnote 9 There is a committee that receives, vets and selects beneficiaries of the MP's share of the Common Fund.
Seven of the 22 districts scored 3 on the MPDACF allocations. I worked with two research assistants on the coding. We each coded the responses independently and then compared the results. All three correlated very highly – over 90%.Footnote 10 For the few cases where there was some disagreement, we discussed each of them and came up with one that accurately reflected the content of the narratives and the examples provided by the administrators.Footnote 11 The result of the coding of the MPDACF allocations is the outcome of interest (dependent variable) in this paper. Table I lists the coding results for all 22 districts.
Table I. List of electoral districts and code for MPDACF allocation
An interview with one of the MPs from the study districts prior to the survey corroborates the narratives provided by the administrator of his district. I successfully surveyed more than a dozen MPs during my visits to the Ghanaian parliament in 2013 prior to the surveys in the Brong Ahafo region. Among those interviewed was the MP for Dormaa West constituency, one of the districts that scored 2 (partly non-discretionary) on the MPDACF allocations. He was a Deputy Minister for Trade and Industry at the time. When asked about the management and distribution of the MPDACF, the MP indicated that he targets the extremely poor and vulnerable populations, notably children with disabilities and the elderly. The administrator for this district also mentioned these objective criteria in his narrative, in addition to orphans and widows; but added that in all cases, the MP has to give the go ahead before actual disbursements are made. In fact in response to a question on his preferred approach to handling MPDACF allocations, the MP ranked ‘personal control’ first, followed by control by ‘trusted party leaders’.Footnote 12
Main independent variable: voter attachment to political parties
The individual-level survey was face-to-face interviews with a random sample of nearly 1,000 adults selected from the 22 electoral districts. The primary sampling unit was the polling station and the target sample size was 1,200 voters. I followed the successful protocols used by the Afrobarometer surveys in Ghana and interviewed four respondents around each sampled polling station, which works out to 300 polling stations across the region (see Appendix 1 for detailed description of the sampling). Interviewers were required to interview two females and two males around each polling station since the gender ratio is approximately 50:50 in the region. I programmed the data collection tool into android devices using the Open Data Kit (ODK) platform. All responses were recorded on these devices, which were GPS enabled, allowing easy tracking of the locations of all the interviews and real time data capture.
Respondents were asked questions on access to basic public services in general and the MPDACF. They were also asked about asset ownership, contacts with key political figures in their districts, party affiliation, past voting behaviour and future voting intentions. I also collected data on respondents’ basic demographic characteristics – gender, age, ethnicity and level of education. In this paper, I focus on respondents’ voting behaviour.Footnote 13 I use respondents’ self-reported voting behaviour to measure the degree of voter attachment to political parties. The survey asked respondents which party's presidential and parliamentary candidates they voted for in the last general elections. Respondents were also asked to indicate which party's presidential and parliamentary candidates they would vote for ‘if elections were held tomorrow’. I classify those who have voted split ticket in past elections – i.e. voted for the presidential candidate of one party and the parliamentary candidate of a different party in the past elections – and/or intend to do so in future ones as unattached or weakly attached voters and those who reported voting straight ticket for the same party in the past and will vote the same way in future elections as strongly attached voters.
I focus on split ticket voting to measure voter attachment to political parties because a provision in the Ghanaian constitution incentivises parties and presidential candidates to actively discourage it. Ghana's 1992 constitution requires the president to appoint at least 50% of cabinet ministers from parliament. Since presidential and parliamentary elections are held concurrently, presidential candidates often spend a significant amount of time campaigning for their parliamentary candidates so that if elected, their party would hold the majority of seats in parliament, which is helpful for governing, and also ensures that there is a large pool of good candidates in parliament for cabinet positions. For instance prior to the most recent (2012) general election, the current president, John Mahama, is reported to have said, during a campaign tour, that:
this time I know you are going to vote for NDC, you are going to vote for me for president, but also add my MPs so that they would support me in parliament to do the work that you want me to do for you.Footnote 14
Presidential candidates of all parties often emphasise similar messages whenever they undertake campaign tours around the country. Thus voters who are strongly attached to political parties are motivated to cast their ballots for candidates of their favourite parties in both the presidential and parliamentary contests. These voters view a vote for their party's candidate for parliament as helping the president to execute his agenda, and they have little incentive to vote otherwise. In this context, those who vote split ticket, which in Ghana is often referred to as skirt-and-blouse voting, are more likely to be unattached or weakly attached voters.
For each district, I compute the proportion of unattached or weakly attached voters as the main independent variable. This measure takes values between 0 and 47·92%. The 0% occurred in one district – Sunyani East district – where all respondents reported voting straight ticket in the past and intended to vote the same way for the same parties in future ones. Higher values on this variable represent weaker partisan attachments. To check the robustness of my results, I run a separate analysis with a different measure of voter attachment to political parties. This measure uses actual election results from each district. Since the first round of the presidential elections and the legislative elections are held concurrently, polling stations are usually arranged such that voters first cast the presidential ballot and then the legislative ballot before exiting the queue. Thus turnout in both elections is usually similar. For each district, I compute the absolute value of the difference between the vote share of the elected MP and the vote share of the presidential candidate of his/her party in each of the last three elections (2004, 2008 and 2012) and take the average of these values. A large gap between the elected MP and his or her presidential candidate is indicative of split ticket voting and hence weak voter attachment to political parties. This measure takes values between 0·23% and 5·74%. The two ways of measuring voters’ partisan attachment lead to similar conclusions in the analysis.
Control variables
In the analysis below, I control for MP and district characteristics. For MP, I control for several characteristics:
Portfolio : Whether the MP holds a ministerial portfolio or has held one recently (i.e. between 2008 and 2012). MPs who are also ministers or were ministers not too long ago may have access to additional resources, which could influence how they handle allocations of the MPDACF. For instance, they may relinquish control over allocations of the MPDACF because they have access to a broader range of resources that they could channel to voters they wish to favour. Ministerial and other positions of influence in government also tend to correlate strongly with personal wealth and/or capacity to attract a significant number of votes, because MPs and other party members who make substantial contributions in these ways to the party are usually the ones who get appointed to these positions.
Margin of victory : MPs who expect very close elections would be more concerned to attract unattached voters. Hence I include a measure of the average margin of victory in each district. I compute this variable as follows: first I compute the average margin of victory (i.e. average of the difference between the vote shares of the winner and the first runner in the legislative elections) in each district over the last three elections – i.e. 2004, 2008 and 2012. I then take the inverse of this variable by subtracting each value from the maximum so that large values reflect close elections. Inverting the margin of victory makes it easier to interpret the effect of close elections on MPDACF allocations.
Distance : MPs from districts that are far from the national and regional capitals may visit their districts less frequently and may therefore be more likely to arrange for someone or group of people in their district to handle the MPDACF allocations. To prevent abuse and/or local capture, it is probably in the interest of MPs in this situation to ensure that allocations of the MPDACF are governed by impersonal rules.
Female : Women MPs may be more passionate about poverty and vulnerability, especially among women and children and may therefore opt for targeted, non-partisan allocations of the MPDACF in order to tackle those issues in their districts.
At the district level, I control for the following characteristics:
Ethnicity : One of the most important predictors of voting behaviour in Africa (e.g. Ferree Reference Ferree2004, Reference Ferree2006; Posner Reference Posner2007; Ichino & Nathan Reference Ichino and Nathan2013). I include a dichotomous variable that takes a value of 1 if the MP is from the dominant ethnic group in the district and 0 otherwise. Based on the most recent (2010) census data, the Akan ethnic group is dominant in most districts in the Brong Ahafo region. Only four of the 22 districts are dominated by other ethnic groups.Footnote 15 In each district, the census data show that the dominant groups make up more than 50% of the population. Since legislative elections are based on plurality rules, MPs who are members of the dominant group in their districts may have a strong incentive to disproportionately favour their co-ethnics in the distribution of valued benefits. In this case allocation of the MPDACF is likely to be based on political discretion.
District wealth : Extant scholarship suggests that when voters become rich, they will demand efficient use of public resources and may punish politicians who engage in clientelism and other forms of non-programmatic distribution of public resources (Weitz-Shapiro Reference Weitz-Shapiro2012). There are no independent and comparable measures of overall district wealth in Ghana so I use the proportion of the district population that is educated above high school as a proxy. A large concentration of highly educated people is likely to correlate strongly with overall wealth of the district.
METHOD
To examine the effect of voter attachment to political parties on resource allocation by MPs, I estimate the following model using ordered logistic regression analysis.
$$\eqalign{MPDACFallocation_i = \; {\rm b}_0 &+ {\rm b}_1SplitTicket_i + {\rm b}_2Avgmargin_i \cr &+ {\rm b}_3MPportfolio_i + {\rm b}_4Distance_i + {\rm b}_5MPfemale_i \cr &+ {\rm b}_6MPEthnicity_i + {\rm b}_7Wealth_i + \varepsilon _i}$$
The ordered logistic approach is appropriate in this case because the dependent variable, MPDACF allocations, takes three (ordered) values: 1, 2 and 3, representing Discretionary, Partly Non-discretionary, and Non-discretionary respectively. Table II lists the summary statistics of all the variables used in the analysis.
Table II. Summary statistics of study variables
*This variable is the alternative operationalisation of the main independent variable.
RESULTS
Table III presents the estimates using the specification above. The dependent variable is the ordered MPDACF allocation criteria. The main independent variable is the proportion of split ticket voters – uncommitted and/or weakly committed voters – in each district as described above. In each model, I cluster the standard errors by electoral district.
Table III. Effect of party attachment on MPs’ resource allocation strategy
Dependent variable: MPDACF Allocation Criteria
Robust standard errors in parentheses
*** p < 0·01, ** p < 0·05, * p < 0·1.
The results are consistent with the hypothesis above: increasing concentration of unattached or weakly attached voters significantly increases the likelihood of rule-based allocations of the MPDACF. Column 1 presents the main effect of voters’ partisan attachment without any covariates. Without controlling for other factors, the results show a significant positive correlation between MPDACF allocation criteria and the proportion of unattached or weakly attached voters. The likelihood of rule-based allocation of the MPDACF increases with the proportion of unattached or weakly attached voters.
When we control for the full range of covariates in Column 2, the effect of voters’ partisan attachment remains statistically significant and substantively large.Footnote 16 To provide a more meaningful interpretation of these results, I graph the predicted marginal probabilities of rule-based or non-discretionary distribution of the MPDACF at different levels of concentration of unattached or weakly attached voters holding all other covariates at their mean values. The results, presented in Figure 1, show that the probability of non-discretionary allocations of the MPDACF increases with increasing concentration of unattached or weakly attached voters. The effect is most pronounced between 20% and 30% concentration of unattached or weakly attached voters. In particular, the likelihood of non-discretionary allocations of the MPDACF is close to 1 (almost certain) as the proportion of less attached voters reaches the 50% mark; and it is close to 0 when the concentration of unattached or weakly attached voters in a district falls below 10%.
Figure 1. Probability of non-discretionary allocations of MPDACF by per cent of unattached/weakly attached voters
The other important predictors of non-discretionary or rule-based allocations of the MPDACF are ministerial portfolio, distance from the regional capital and MP ethnicity. MPs who hold ministerial portfolios are significantly more likely to relinquish discretion over allocations of the MPDACF. Ministers of state tend to control a broad range of valued public benefits that they could channel to voters they wish to favour. The MPDACF may therefore be a small part of the resources under their control and it would probably not be a big deal to relinquish control over allocations of these funds. Similarly, allocations of the MPDACF are also more likely to be non-discretionary or rule-based among MPs whose districts are further from the regional capital. These MPs may visit their districts less frequently and it is probably in their interest to have impersonal rules govern allocations of their development funds to prevent local capture. Meanwhile, MPs who are members of the dominant ethnic group in their district are less likely to relinquish discretion over MPDACF allocations. This result is consistent with existing scholarship on ethnic identities and voting behaviour in Africa (Chandra Reference Chandra2007; Posner Reference Posner2007). As noted earlier, in each district the dominant ethnic group makes up more than 50% of the total population. Thus MPs who are from the dominant groups could win an election with the support of only their co-ethnics and as such MPs may have a greater incentive to target their co-ethnics in the distribution of benefits.
It is important to note here that even though MPs from dominant ethnic groups may have the incentive to target their co-ethnics in the distribution of benefits, ethnicity does not (simultaneously) drive partisan attachment. There are many ethnically homogeneous districts in the region and the country at large with very weak partisan attachment. The opposite is also true. That is, there are ethnically heterogeneous districts in the region and the country with large concentrations of strongly partisan voters (Fridy Reference Fridy2007).
Finally, allocations of the MPDACF are also more likely to be non-discretionary among MPs from wealthier districts and among female MPs. However, these effects are not significant at conventional levels of statistical significance. That said, the first result is consistent with predictions of standard models of distributive politics: wealthier, more educated voters may be more likely to punish politicians who engage in clientelism or other forms of vote buying (Weitz-Shapiro Reference Weitz-Shapiro2012; Kramon Reference Kramon2013). Thus politicians in such districts may favour broad-based, non-partisan and more efficient allocations of public benefits. On the effect of gender, perhaps female MPs may be more concerned about poverty and vulnerability among children and women and they may seek to address those issues through rule-based, non-partisan allocations of their development resources.
The main independent variable in the preceding analysis is based on self-reported voting behaviour: the proportion of voters in each district who report that they have voted split ticket in past elections and/or intend to do so in future ones. In the following analysis, I replace this measure with one constructed from actual election results as described above: i.e. the gap in the vote shares between the elected MP and the presidential candidate of his or her party, which is indicative of split ticket voting. A wider gap means weaker partisan attachment. The results, presented in Table IV, are also consistent with my hypothesis. A large gap in vote share between the MP and the presidential candidate of his or her party is associated with a higher chance of non-discretionary distribution of the MPDACF.
Table IV. Effect of party attachment on MPs’ resource allocation strategy
Dependent variable: MPDACF Allocation Criteria
Robust standard errors in parentheses
*** p < 0·01, ** p < 0·05, * p < 0·1.
Figure 2 plots the predicted probability of non-discretionary allocations of the MPDACF against the average gap in the vote shares between the MP and his or her presidential candidate. In districts where this gap is more than 5 percentage points, the likelihood of non-discretionary allocations of the MPDACF is at least 60%, holding all other covariates at their mean values.
Figure 2. Average gap in vote shares and probability of non-discretionary allocations of the MPDACF
Except for the effect of distance, which is negative, small and statistically insignificant in this specification, the effects of the remaining control variables are largely consistent with the results of the preceding analysis. Allocations of the MPDACF are significantly less likely to be non-discretionary in districts where the MP is a member of the dominant ethnic group whereas in wealthier districts, allocations are more likely to be non-discretionary. The effects of ministerial portfolio and gender (female) remain positive but statistically insignificant. That of the inverse margin of victory is also positive as in the previous analysis, meaning that MPs who expect close elections are more likely to pursue non-discretionary allocations of the MPDACF; but this effect is statistically significant.
DISCUSSION AND CONCLUSIONS
This paper joins the body of research on the political and policy implications of voter attachments to political parties (Kitschelt & Wilkinson Reference Kitschelt and Wilkinson2007; Keefer & Khemani Reference Keefer and Khemani2009; Asunka Reference Asunka2016). The results show that weak voter attachment to political parties increases the likelihood of more programmatic forms of public resource allocation by politicians. Broad-based, non-discretionary allocations allow benefits to reach unattached or weakly attached voters, which might influence their voting decisions. Moreover, because unattached or weakly attached voters are more likely to base their voting decisions on the (perceived) performance of incumbent politicians (Fridy Reference Fridy2007; Weghorst & Lindberg Reference Weghorst and Lindberg2013), self-binding, broad-based distributive rules enable incumbents to broaden their electoral support among unattached voters. These findings shed light on the growing political support for targeted, rule-based public transfers to citizens in developing democracies. These transfers enable parties in government to broaden their electoral support among voters from whom they have little electoral support without alienating some of those who have been more loyal to them.
These findings are also consistent with, and important for, understanding the recent line of research that questions the popular view that elections in Africa are inherently clientelist or ethnic (Fridy Reference Fridy2007; Lindberg & Morrison Reference Lindberg and Morrison2008; Bratton Reference Bratton2013; Weghorst & Lindberg Reference Weghorst and Lindberg2013). This line of research suggests that voting behaviour in Africa is increasingly shaped by perceived performance of incumbent politicians. However, performance-based voting is concentrated among unattached or weakly attached voters, as is the case in the USA, because very loyal voters cannot be attracted by any strategy used by the other party. In other words, voters who are not strongly attached to political parties are more likely to base their voting decisions on how they evaluate the performance of incumbent politicians. This implies that in places with large concentrations of unattached or weakly attached voters, politicians would have a greater incentive to signal good performance, especially in the distribution of valued public benefits to citizens.
Finally, I offer two suggestions for future research. First, the evidence presented here suggests that voter attachment to political parties is essential when evaluating the impact of constituency development funds (CDFs) on the performance of legislators in new democracipes, notably the debate on whether or not legislators’ involvement in grassroots development through the CDFs undermines their investment in oversight and other legislative duties (van Zyl Reference van Zyl2010). Future research on this topic should consider the role of partisan attachments among voters. Future research should also interrogate how partisan attachments would impact politicians’ resources allocation decisions under proportional representation electoral rules. The results presented here are based on data from a single member plurality voting system, which may or may not hold under proportional representation.
Appendix 1: Sampling Procedure
To draw the 300 polling stations, I first computed the share of polling stations in each electoral district in the region. I then allocated the 300 polling stations such that the share of polling stations in each district in the sample is the same as the district's share of polling stations in the region. Finally, I used a random number generator to select the required number of polling stations from each district. At each polling station, research assistants used the random walk pattern to select four households, one at a time and then randomly select and interview one adult from each household.
