ArticlePDF Available

Abstract and Figures

Any election may result in six possible situations. The incumbent or challenger may win according to the official results. If the incumbent wins, he may remain in power, or a standoff or coalition may ensue. In contrast, if the challenger wins, he may become the new incumbent, or a standoff or coalition may ensue. Using a database of all presidential and legislative elections in Africa over the period 1960–2010, we found the following distribution of election outcomes: the incumbent wins with no contestation 63.9 per cent, coalition 6.4 per cent, and standoff 1.2 per cent. The incumbent loses and accepts defeat 15.9 per cent, coalition 12.3 per cent, and standoff 0.3 per cent. We have then tested empirically 22 hypotheses on the determinants of election outcomes in Africa using a discrete-choice multinomial logit model. We study the impact of the shape of the economy, the provision of public goods, education, social diversity, number of years in power of the incumbent, whether the incumbent is a military official or not, the strength of the opposition, natural resource endowment, colonial origins of the country, and whether the election is presidential or legislative.
Content may be subject to copyright.
Determinants of Election Outcomes: New Evidence from Africa
Kjell Hausken and Mthuli Ncube*
Abstract:Any election may result in six possible situations. The incumbent or challenger may win according to the ofcial
results. If the incumbent wins, he may remain in power, or a standoff or coalition may ensue. In contrast, if the challenger wins,
he may become the new incumbent, or a standoff or coalition may ensue. Using a database of all presidential and legislative
elections in Africa over the period 19602010, we found the following distribution of election outcomes: the incumbent wins
with no contestation 63.9 per cent, coalition 6.4 per cent, and standoff 1.2 per cent. The incumbent loses and accepts defeat 15.9
per cent, coalition 12.3 per cent, and standoff 0.3 per cent. We have then tested empirically 22 hypotheses on the determinants of
election outcomes in Africa using a discrete-choice multinomial logit model. We study the impact of the shape of the economy,
the provision of public goods, education, social diversity, number of years in power of the incumbent, whether the incumbent is a
military ofcial or not, the strength of the opposition, natural resource endowment, colonial origins of the country, and whether
the election is presidential or legislative.
1. Introduction
Democratization is aimed at strengthening governance institutions, improving economic performance, reducing general political
repression, and improving service delivery. Elections are widely considered as the best test of democracy. When they are fair,
free and peaceful, the country is considered to be democratic. On the other hand, when the elections are violent, or followed by
erce claims of vote rigging and contestations, or power conscation by the incumbent despite a defeat, the country is considered
to be autocratic.
The transition to democracy, especially in Africa, has been epitomized by election outcomes that have not been seen to be free
and fair. Sometimes the election process is manipulated and violence is used to achieve victory. Some work by Schedler (2007),
Schmitter (1978, 994) and De Mesquita et al. (2003) goes into certain aspects of this phenomenon. Dercon and Gutierrez-
Romero (2012) analyse the triggers and characteristics of violence following the Kenyan elections of 2007. Additional work by
Fearon and Laitin (2003) brings out the ineffectiveness of democratic elections in improving governance and economic
performance. Apart from reliance on use of violence, there are other factors that determine election outcomes. The election
outcomes are either accepted, disputed and may even result in power-sharing arrangements.
The history of elections for the majority of African countries starts from their independence in the late 1950s and early 1960s.
Since then, African countries have been especially slow in establishing democratic institutions, and various democratic processes
have been reversed. Many African election outcomes have been considered as unfair and not free. Also, various elections have been
violent, and even well-established democracies such as Kenya have experienced violence during the election process.
However, it is not easy to classify elections as autocratic or democratic. If there are no visible contestations or violence after an
election, it does not necessarily mean that the elections were democratic. It could be because the opposition is silenced by the
incumbent or votes have been bought by the incumbent or there is a lack of political awareness in the country. In addition, vote
rigging could be so sophisticated that it is difcult to prove that elections were rigged or not and the true winner of an election is
difcult to observe as in the 2010 elections in the Ivory Coast. This makes it difcult to study election outcomes particularly in
Africa.
*Professor Kjell Hausken (corresponding author), Faculty of Social Sciences, University of Stavanger, 4036 Stavanger, Norway; e-mail: kjell.
hausken@uis.no. Mthuli Ncube, Senior Research Fellow, Blavatnik School of Government, University of Oxford, Oxford OX1 4JJ, e-mail: mthuli.
ncube@bsg.ox.ac.uk. We thank Finn Tarp, Zuzana Brixiova, three anonymous referees of this journal and the editor for useful comments. We
acknowledge the excellent research assistance from Siliadin Yaovi Gassesse and Letsara Nirina of the African Development Bank Group.
African Development Review, Vol. 26, No. 4, 2014, 610630
© 2014 The Authors. African Development Review © 2014 African Development Bank. Published by Blackwell Publishing Ltd, 9600 Garsington Road,
Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 610
The objectives of this study are to analyse the factors driving election outcomes in Africa. Some researchers argue about
economic drivers, such as Lewis-Beck and Steigmaier (2000), Vergne (2009) and Litschig and Morrison (2010). Other
researchers have focused on social, political or historical drivers (see, for example, Nie et al., 1996; Hillygus, 2005; Howell and
Brent, 2013). Agreement rarely exists among researchers about the impact of the various factors. Surprisingly, in the literature
not a lot of analysis has been done on Africa in this regard despite the continents struggle for democracy.
This paper asks two main questions. The rst is: what is the distribution of election outcomes in Africa? Such a distribution
provides the answer to some interesting questions like: what is the most frequent outcome occur in Africa? How often do
contestations occur in Africa? How often do incumbent presidents lose and refuse to concede defeat? The second question is:
why does an election result in one situation or the other? More specically, why do some incumbent African regimes concede
defeat when they lose elections while others refuse to concede defeat and instead cling to power? Why do some incumbent
African regimes win elections without contestations while others face erce and violent contestations if they win? This paper
attempts to ll this gap in the literature by studying the determinants of election outcomes in Africa, and answering the questions
above.
The rest of the paper is organized as follows. Section 2 presents the literature review. Section 3 presents the data and
descriptive statistics. Section 4 presents the methodology and empirical analysis, Section 5 concludes.
2. Literature Review and Thirteen Informally Formulated Hypotheses
Drivers of election outcomes fall most importantly into economic, social, and political factors, natural resources, and former
colonizer. Literature abounds on how such factors impact on election outcomes.
2.1 Economic Factors
Hypothesis A. Good economic conditions boost the popularity of the incumbent and increase his chances of winning
the election.
Lewis-Beck and Stegmaier (2000) have reviewed several studies that have successfully tested this relationship. They
found strong support for it in elections in the US and European countries. So far, no such study has been conducted on African
elections.
Hypothesis B. If public investment increases the year prior to the election, the incumbent is more likely to cling to
power if he loses.
This is because he wants to collect the benets of the investment.
Hypothesis C. If public consumption increases the year prior to the election, the incumbent is more likely to win the
election without contestation.
Many studies have provided empirical evidence of opportunistic manipulation of budgets by governments. Barreira and Baleiras
(2004) showed that central and local governments in the European Union tend to increase public consumption relatively to public
investment in the year prior to elections. Vergne (2009) also found similar results in developing countries and showed that
electoral budget cycles may persist even as the country gets more experience in democracy. Drazen and Eslava (2010) suggested
a theoretical model where incumbent governments try to inuence voters by changing the budget composition in the years prior
to election. Using data on municipal elections in Colombia, they provided empirical evidence that voters do positively respond to
this manipulation by the government. Using data on Brazilian elections, Litschig and Morrison (2010) showed that an increase in
public spending the year prior to election increases the chance of re-election of the incumbent. Another example of opportunistic
behaviour is found by Robinson and Torvik (2005) who show that investment projects with negative social surplus (so-called
white elephants) may be preferred to socially efcient projects if the political benets outweigh the surplus of efcient projects.
In the empirical analysis we test for the importance of investment.
Determinants of Election Outcomes 611
© 2014 The Authors. African Development Review © 2014 African Development Bank
2.2 Social Factors
Hypothesis D. The more educated the population, the more likely citizens are to contest the election results if they
think that the results are awed.
The strong positive linkage between education and political awareness is well established in political behaviour research (see, for
instance, Hillygus, 2005; Berinsky and Lenz, 2011; Persson, 2013; Rosenstone and Hansen, 1993; Nie et al., 1996; Herrnstein
and Murray, 1994). Hillygus (2005) found that education is the strongest predictor of political participation even after controlling
for other socio-economic factors. He argued that the most plausible reason for this is that education expands the capacity of
citizens to engage in self-rule by teaching citizens the behaviors and knowledge necessary for identifying political preferences,
understanding politics, and pursuing political interests.
Hypothesis E. The higher ethnic and religious fractionalization in the country, the more the incumbent has the
incentive to cling to power.
Rabushka and Shepsle (1972) have argued that it is not possible to resolve intense but conicting preferences in plural societies
in a democratic framework. This suggests that democratic outcomes are less likely in countries with high ethnic and religious
fractionalization. Horowitz (1985) argued that in societies with high ethnic fractionalization, power is an end unto itself because
it demonstrates the value of the ruling ethnic group and ensures the ruling groups survival. The implication is that incumbent
presidents in ethnically divided countries are more likely to cling to power than incumbent presidents in less ethnically divided
countries. However, recent works have argued that there is no direct causality running from social cleavage to democracy.
Chandra (2005) contended that explanations of Rabushka and Shepsle (1972) and Horowitz (1985) were based on the awed
assumptions that ethnic identities are xed, one-dimensional and exogenous to politics. He claimed that ethnic identities evolve
with time, are pluridimensional and endogenous to politics. He also argued that the impact of social pluralism on democracy
depends on the institutions in place. Institutions that encourage social pluralism foster democracy, while institutions that restrict
ethnicity to one dimension destroy democracy. Therefore, according to Chandra (2005) social pluralism by its intrinsic nature
does not threaten democracy. This implies that it is not necessarily true that higher ethnic and religious fractionalization would
increase the odds that the incumbent clings to power if he loses.
2.3 Political Factors
Hypothesis F. The appetite for power increases with number of years in power.
This view is supported by Howell and Brent (2013) who argued that there is indeed an appetite for power and that this appetite is
built into the presidential seat itself. They claimed that it is not the appetite for power that motivates individuals to run for
presidency the rst time but it is only when they are in power that they begin realizing what presidential power actually is. As a
result, the more time incumbents spend in power, the more they learn about the intrinsic value of power and the higher their
appetite for it. Military incumbents are more likely to cling to power than civil incumbents.
Hypothesis G. If the opposition is strong, it is more likely to contest the re-election of the incumbent and the
incumbent is less likely to cling to power when losing.
A strong opposition is one that is organized and is not easily silenced by the incumbent regime. We suspect that another reason
why most African presidents are re-elected without contestation is that the opposition is silenced or is so disorganized that it
cannot conduct a large-scale contestation. If this suspicion is true, then, contestations of election results should be more common
in countries with a strong opposition.
Hypothesis H. Single party political systems favour the incumbent to win the election
© 2014 The Authors. African Development Review © 2014 African Development Bank
612 K. Hausken and M. Ncube
Hypothesis I. Military incumbents are more likely to cling to power than civil incumbents.
This hypothesis follows from the fact that military incumbents generally have the support of the army and are more likely to rule
by force than civil incumbents. Clinging to power is therefore easier to achieve for a military incumbent. If it is easier, then the
temptation and the chances that it happens should also be higher. However, the implication is not so straightforward according to
Marinov and Goemans (2014). On the one hand they argue that professional militaries are more likely to cling to power without
free and fair election. But on the other hand they argue that sometimes, in anticipation of future divisions in the leadership of the
army, military incumbents may choose to hold free and fair elections. This dilemma weighing conict in the present against
conict in the future is well known in economic theory.
1
Hypothesis J. Whether the country is coastal or not has limited impact on the election outcome.
2.4 Natural Resources
Hypothesis K. Incumbents are more likely to cling to power in natural resource rich countries.
This hypothesis is widespread in political science. The main argument presented to support this negative correlation between
natural resource wealth and democracy is that of the rentier state. According to Jensen and Wantchekon (2004) the main
characteristic of rentier states is a high dependence on rents produced by few economic agents. Thus rentier states are generally
autonomous and less accountable to voters. As a result, they feel less pressure for democratization. Moreover, rentier states
can use the rents to buy votes, corrupt key political actors and pursue non-democratic agendas. Jensen and Wantchekon
(2004) concluded that this lack of transparency and accountability is the reason why resource rich countries are more likely
to be authoritarian. Related to this hypothesis, Ross (1999) analyses political economy related to the resource curse,
Mehlum et al. (2006) analyse institutions and the resource curse, and Robinson et al. (2006) consider the political foundations
of the resource curse. Ross (2001) explores the breakdown of the institutions that govern natural resource exports in developing
states.
2.5 Former Colonizer
Hypothesis L. Former colonizer matters.
The main argument behind this hypothesis is the work of Acemoglu and Robinson (2006) who argued that colonial origins of a
country affect the type of institutions that subsequently develop in the country, in the post-colonization era.
Other factors that impact on election outcomes, not considered in this study, include electoral rules (Cox, 1997; Persson and
Tabellini, 2002), information asymmetry (Ellman and Wantchekon, 2000) and election process (Collier, 2009).
2.6 Type of Elections (Presidential or Legislative Elections)
Hypothesis M: Presidential elections are more associated with the incumbent winning outright without contestation,
than legislative elections.
For presidential elections we expect the incumbent not to share power at the presidential level, but instead include a deputy from
the opposition. However, for legislative elections, we expect that the incumbent can accept being in a coalition with the
opposition. Therefore, we expect that presidential elections are more associated with the incumbent winning outright without
contestation, than in legislative elections.
This study tests these 13 hypotheses, further rened to 22 hypotheses, and relate to the various arguments in the literature,
using data from elections conducted in Africa.
© 2014 The Authors. African Development Review © 2014 African Development Bank
Determinants of Election Outcomes 613
3. Data
3.1 Classication of Elections
The database used in the study includes all 653 elections in Africa from 1960 to 2010, of which 299 (46 per cent) are legislative
and 354 (54 per cent) are presidential. It covers all African countries except Libya, São Tomé and Príncipe, Eritrea, Somaliland
(not internationally recognized) and South Sudan, where no elections were held. This data set, constructed for this paper by the
African Development Bank, is new in both the period covered as well as in the classication of election outcomes. Data on
elections is generally publicly available but not pooled together in a format suitable for econometric analysis. The African
Development Bank has extensively relied on data from africanelections.tripod.com and governmentsofcial sources to
construct this new database. The classication of election outcomes is performed by the authors based on detailed information
from africanelections.tripod.com and African governmentsofcial sources. Data on ethnic fractionalization, religious
fractionalization and language fractionalization is also provided by the African Development Bank which has retrieved it mainly
from Encyclopedia Britannica. Ethnic fractionalization is measured using an index that captures the degree of fractionalization
as captured by the number of ethnic groups within the population. The indices for ethnic, religious and language fractionalization
lie between 0 and 1, and are continuous within that range (see Alesina et al., 2003). Data on economic variables, education and
the Gini index is from the African Development Indicators Database. The dataset also uniquely covers both legislative and
presidential elections. Of the 299 legislative elections, the incumbent won 210 without the challengers contestation. Only
ve legislative elections where the incumbent won were contested. These are the 2010 Guinea election where the contestation
resulted in a coalition, and the 2009 Malawi election and the 1991 and 2007 Benin elections where the contestations resulted in
standoff. In 52 elections, the incumbent lost and accepted defeat while in 32 elections the incumbent lost, contested the election
results and negotiated a coalition with the challenger.
The study in this paper classies the six election outcomes as follows: First, the incumbent wins and remains in power (WP).
In this outcome, an election supervisory body declares the incumbent leader or party, the winner, and this is accepted by the
Table 1: An example: classication of recent presidential election outcomes in
Africa 20062011 þEritrea 1993
Incumbent loses Incumbent wins
LC: Coalition LS: Standoff
LP: Challenger
becomes new
incumbent WC: Coalition WS: Standoff
WP: Incumbent remains in
power
Guinea-Bissau 2009,
Kenya 2007,
Lesotho 2007,
Zimbabwe 2008
Sierra Leone 2007,
Niger 2011,
São Tomé 2011,
Morocco 2007,
Mauritania 2009,
Mauritius 2010,
Ghana 2008,
Guinea 2010,
Comoros 2010,
Liberia 2005,
Côte dIvoire 2010
**
Gabon 2009,
Mozambique 2009,
Madagascar 2006
***
Central Africa
Republic 2011,
Chad 2011,
Djibouti 2011,
Equatorial Guinea
2009, Gambia 2006,
Mali 2007, Nigeria
2011, Republic of
Congo 2009, Senegal
2007, Somalia 2009,
Sudan 2010.
Algeria 2009, Angola 2008,
Benin 2011, Botswana
2009, Burkina Faso 2010,
Burundi 2010, Cameroon
2007, Cape Verde 2006,
Democratic Republic of
Congo 2007, Eritrea 1993,
Egypt 2010
*
, Ethiopia
2010, Namibia 2009,
Rwanda 2010, Seychelles
2011, South Africa 2009,
Tanzania 2010, Togo 2010,
Tunisia 2009
*
, Uganda
2011, Zambia 2008,
Malawi 2009
***
*
Incumbents were later toppled in the revolutionary uprising of the Arab Spring in 2011.
**
Challenger, who was the winner, only took over after a violent standoff of bloody conict and some foreign military intervention.
***
In Malawi Bingu wa Mutharika won with a new party and having abandoned his old party. In Madagascar, Ravalomanana won as the incumbent but created a
coalition government.
© 2014 The Authors. African Development Review © 2014 African Development Bank
614 K. Hausken and M. Ncube
opposition, and the incumbent thus stays in power. Second, the incumbent wins and a standoff ensues (WS). In this outcome, the
incumbent declares themselves the winner or the electoral supervisory body declares them the winner, but the opposition
contrasts this outcome as not being free and fair. Then standoff ensues, and some of the standoffs could even be violent or involve
legal recourse. Third, the incumbent wins and a coalition ensues (WC). This outcome results in the incumbent winning the
election but not sufciently as to be able to govern effectively on their own, and need a coalition with the challenger, to do so. A
coalition government is then formed with the challenger, and the incumbent as the leader. Fourth, the challenger wins and
becomes the new incumbent (new leader) (LP). In this outcome the challenger wins outright and is declared so by the electoral
supervisory body, and the incumbent accepts the outcome. The challenger then takes over as the new leader. Fifth, the challenger
wins and a standoff ensues (LS). In this outcome the challenger is declared the winner by the electoral body but the incumbent
does not accept this, and a standoff ensues. Such a standoff could be violent or involve legal recourse. Sixth, the challenger wins
and a coalition ensues (LC). In this outcome, the challenger is declared the winner but not sufciently so as to be able to govern
effectively on his own. The challenger then invites the current incumbent into a coalition arrangement. Due to tractability we do
not consider further outcomes such as that the incumbent does not contest (he is not eligible to contest due to various reasons
including the fact that it has covered its legally required terms). Table 1 shows recent examples of election outcomes classied
under the six outcomes, based on the available data.
In Table 2 we summarize the data for all election events in the study, showing the frequency of each election outcome.
Of the 354 presidential elections, the incumbent president won 207 without contestation. On the other polar opposite, in 52
elections, the incumbent lost and conceded defeat. In 95 elections, the election results were contested by the loser. The incumbent
lost, rejected the results and formed a coalition in 48 elections. The challenger lost, contested the results and formed a coalition
with the incumbent in 40 elections. Seven elections resulted in a standoff.
Overall, during 19602010, 80 per cent of the presidential and legislative election results were accepted, 18 per cent resulted
in a coalition and 2 per cent resulted in a standoff, see Table 2. Incumbent regimes tend to win elections they organize with a 71
per cent probability. When the incumbent loses, he tends to reject the results (79 per cent of the time). The challenger tends not to
contest the results (contestation occurs in only 7 per cent of the elections). However, the challengers contestation rate is higher
for presidential elections (12 per cent) than for legislative elections (2 per cent).
Table 2: Classication of African election outcomes (frequency (%)): 19602010
Outcomes Legislative Presidential Total
Incumbent loses, accepts defeat 52 (17%) 52 (15%) 104 (15.9%)
Incumbent loses, contestation, coalition 32 (11%) 48 (14%) 80 (12.3%)
Incumbent loses, contestation, standoff 0 (0%) 2 (0.6%) 2 (0.3%)
Incumbent wins, contestation, coalition 2 (0.7%) 40 (11%) 42 (6.4%)
Incumbent wins, contestation, standoff 3 (1%) 5 (1.4%) 8 (1.2%)
Incumbent wins, no contestation 210 (70%) 207 (58%) 417 (63.9%)
Total 299 (100%) 354 (100%) 653 (100%)
Table 3: Real per capita GDP
Outcomes
Growth Lagged growth
Mean SD Mean SD
LP 0.7268605 4.720912 1.417979 4.940245
LCS 0.1974261 4.851998 0.2480947 4.498186
WCS 3.298996 13.83949 2.463653 8.60462
WP 1.721023 7.380211 1.336681 5.455691
© 2014 The Authors. African Development Review © 2014 African Development Bank
Determinants of Election Outcomes 615
3.2 Descriptive Statistics and Correlation Analysis
The nal outcome of an election depends onseveral factors including the economic performance of the incumbent, the provision of
public goods, institutional factors, social factors, the incumbent characteristics, the challenger characteristics, the electoral system,
historical and geographical factors and initial conditions, as we also stated above. In this section we discuss the correlation between
these different variables and the election outcomes. It is important to do so in order to assess the relevance of the selected variables
for the study. If there is no interdependence between these variables and the election outcomes, there is no need to proceed to
regression analysis. Correlation analysis also allows us to form preliminary expectations about the regression results. If the
regression results differ from these expectations, either there is an explanation for this or the regression model is awed.
In this study, we cannot use standard correlation measures such as the Pearson correlation, because the dependent variable in
our study is not a quantitative variable. Fortunately, statistics is rich enough to provide us with tools usable to assess the
interdependence between two qualitative variables or between one qualitative variable and a quantitative variable. In the event of
two qualitative variables, one can resort to the Chi-square independence test and categorical analysis. In the event where one
variable is qualitative and the other is quantitative, to the best of our knowledge there is no straightforward interdependence test.
The existing methods are descriptive rather than inferential. What is usually done is to compare the distribution of the
quantitative variable across the different modalities of the qualitative variable. In this latter event, it is generally sufcient to
consider the rst and second moments of the quantitative variable, or to compare the probabilities of the quantitative variable
being negative across the different modalities of the qualitative variable.
It is also important to assess the correlation between the different explanatory variables in the study to check whether
multicolinearity is a problem for regression analysis. Many of these variables are quantitative so it is possible to use standard
correlation analysis with signicance testing. This has been done but the results are not presented here. Typically, these
correlations are very low and pose no multicolinearity problem.
The data shows that election standoffs are few. Because standoff outcomes are few we have reduced the number of outcomes
from six to four by merging outcomes WS and WC to outcome WCS, and merging outcomes LC and LS to LCS. That is, we do
not make a distinction between a coalition and standoff, in the econometric analysis. The reason for this is not purely statistical.
Some of the coalitions are formed after a certain period of standoff. And standoff may result from a broken coalition, or
the coalition may be imposed by the international community while the political situation is a real standoff as in Zimbabwe. The
outcomes WP and LP are as before. The codication of the LP, LCS, WCS, WP election outcomes, was then performed on the
data. The 1993 Eritrea election, whether included or excluded, does not change the pattern of the results below. In the analysis
Table 4: Public investment as a share of GDP (assuming normal distribution)
Growth Lagged growth
Mean SD Prob(<0) Mean SD Prob(<0)
LP 3.538136 32.7774 0.457019908 0.6714323 31.1541 0.491402673
LCS 10.572 44.96217 0.407053529 8.66837 39.48807 0.413123005
WCS 5.798454 23.58011 0.597121901 1.770407 28.18244 0.474955116
WP 2.282698 33.3301 0.472698752 11.34125 61.55915 0.426915189
Table 5: Public consumption as a share of GDP
Growth Lagged growth
Mean SD Prob (<0) Mean SD Prob(<0)
LP 1.042618 8.215567 0.44950676 0.7193151 13.32434 0.47847354
LCS 0.3466147 8.117501 0.51702953 0.4373123 6.389229 0.47271561
WCS 1.054017 5.908831 0.4292123 0.5772024 6.35427 0.53618892
WP 1.1817 9.773311 0.4518808 1.033623 7.786003 0.44719397
*assuming normal distribution.
© 2014 The Authors. African Development Review © 2014 African Development Bank
616 K. Hausken and M. Ncube
we have pooled the legislative and presidential elections together because the behaviour of the election parties is similar whether
the elections are parliamentary or presidential.
The economic performance of the incumbent is measured by the real per capita GDP growth. For elections that took place at
the beginning of a calendar year, citizens would judge the incumbents economic achievement by the lagged real per capita GDP
growth. As shown in Table 3, on average, the incumbent will lose the election if the economic performance is poor. Interestingly,
LCS elections where the incumbent loses and clings to power are those where the country recorded on average the worst
economic performance. Highest economic performance is on average recorded for the WCS elections where the incumbents
victory is contested, though this is also the outcome with the largest standard deviation. The lagged variable is trying to capture
the actions of the incumbent prior to the election, which may be different from actions taken during the election. GDP per capita
captures the economic impact of these actions, and hence we have utilized both in the equation.
Public investment as a share of GDP grew by 0.67 per cent and 3.54 per cent on average respectively, a year before and the
year of LP elections (see Table 4). For LCS elections, the average increase was higher: 8.66 per cent the year before and 10.57 per
cent the year of the elections.
The share of public investment in GDP declined by 5.8 per cent on average the year of WCS elections. Public investment
growth as share of GDP was substantially high (11.34 per cent) on average the year before WP elections. Taking into
consideration the high standard deviation for each outcome, we see that WCS elections corresponded to the greatest probability
of having a decline of the share of public investment in the GDP. This probability is lowest for LCS elections.
In the year of LP, WCS, and WP elections, public consumption as a share of GDP, on average, increased. The year before LP,
WCS, and WP elections, public consumption also increased on average (see Table 5).
On average, for LCS elections, public consumption as a share of GDP increased the year before the elections and decreased the
year of the elections. For WCS elections the situation is opposite. On average, the share of public consumption in GDP decreased
the year before the election and increased the year of the election. LCS elections corresponded to the lowest average enrolment
rate in tertiary education and lowest average language fractionalization (see Table 6). On the other hand, high average enrolment
rate in tertiary education corresponded to outcomes where the re-election of the incumbent is contested.
Countries where WCS elections happened had on average the lowest Gini index, a measure of degree of inequality. The lowest
average ethnic fractionalization was recorded in countries with WP elections. LP elections corresponded to the lowest average
Table 6: Social factors
Outcome Enrolment in tertiary education (%) Gini index Ethnic fractionalization Religious fractionalization
LP 3.18572 45.4516 68.8641 45.6417
LCS 2.60313 46.86885 67.9773 46.2417
WCS 4.68302 42.90359 66.8176 49.313
WP 3.87274 46.13543 62.3375 50.6454
Table 7: Association of other variables with election outcomes
Other variables Pr (Chi
2
) Associations
Did incumbent come to power through a coup? 0.000 Yes with WPStrong opposition
0.000 No with WP, Yes with LP
Freedom of press 0.000 Semi-freedom with LCS and WCS, No freedom with LP and WP
Number of rounds 0.000 One round with WP, Two rounds with LCS
Party system (multiparty or not) 0.001 Multiparty with LCS
Incumbent from military? 0.085 Yes with LCS and WP, No with LP and WCS
Main religion 0.000 Traditional with WCS, Islam with WP, Christianity with LCS
Country proximity to coast 0.538
Natural resource endowment 0.000 Abundant with LCS
Election type 0.000 Presidential associated with LCS and WCS, Legislative associated with
LP and WP
Former colonizer 0.000 France associated with LCS
© 2014 The Authors. African Development Review © 2014 African Development Bank
Determinants of Election Outcomes 617
religious fractionalization. The Chi-square tests show statistically signicant association between electoral outcomes and
election types, number of rounds of the election, the political system (multi-party, single party or non-partisan) political coups,
opposition strength, freedom of press, main religion in the country and the countrys natural resource endowment (see Table 7).
The moderate and abundant natural resource endowment variables are categorical in nature. Whether the incumbent president is
from the military or not, and whether the country is coastal or not, matter to some degree with LCS and WP outcomes. Using
categorical analysis we have found that the WP elections are associated with political coups, weak opposition, no press freedom,
one-round elections, military incumbent, Muslim majority in country, and legislative elections. WCS elections are related to
semi-freedom of press, civil incumbent, traditional religions and presidential elections. LP elections are associated with strong
opposition, no press freedom, civil incumbent and legislative elections. LCS elections are related to semi-freedom of the press,
two round elections, incumbent from military, Christianity, abundant natural resources, presidential elections and France as a
former colonizer. These associations are mere statistical associations similar to correlations in the event of quantitative variables.
4. Methodology and Empirical Analysis
4.1 Multinomial Logit Regression
The study utilizes robust multinomial logit regression analysis to test which factors determine electoral outcomes. Standard OLS
regression would not work because the dependent variable is categorical. We also cannot use simple probit or logit regressions
because our dependent variable has more than two modalities. In fact, it has four modalities. Technically, if the outcomes could
be ordered in an increasing or decreasing order, one could try the OLS regression or use the more appropriate ordered logit
regression or the ordered probit regression. These methods are not usable here because the outcomes of elections cannot be
ordered. An outcome is not necessarily superior to the others. For instance, one cannot say that the LP outcome is preferable to
the WP outcome. Neither could one clearly determine which one of the LCS and the WCS outcome is preferable to the other. The
appropriate regression tool for our research is the multinomial logit model which allows for a categorical dependent variable with
more than two modalities.
Suppose the dependent variable has Koutcomes, J¼1,2, ,K. Then multinomial logit model is given by
ln PðYi¼jÞ
PðYi¼KÞ

¼bjXiþajZiþuji;j¼1K1ð1Þ
where Y
i
is the outcome for election i,X
i
is the set of regressors, Z
i
is the set of control variables, and u
ji
is the error term. The
outcome variable Yhas Kpossible outcomes. If the dependent variable has K outcomes, the multinomial logit model consists of
K1 regression equations. Each of these equations, say equation j, is an OLS regression model where the left-hand side of the
regression equation is not the dependent variable itself, but rather the relative likelihood of outcome j of the dependent variable
with respect to outcome Kwhich is called the pivotal outcome. Technically, the pivotal outcome need not be the outcome K.It
could be any of the outcomes. It sufces to change the order of the outcomes to make any particular outcome the lastone.
Common wisdom suggests using the most frequent outcome as the pivotal outcome. Accordingly, in our analysis we chose WP
as the pivotal outcome.
An important assumption of the multinomial logit regression model is the independence of the irrelevant alternatives (IIA)
assumption. The IIA assumption asserts that if an additional outcome is included or one is excluded, this does not change the
relative probabilities of the remaining outcomes. The IIA assumption is relevant when analysing rational choices of individuals
because individuals often face more or fewer alternatives. For our research, there are always four possible outcomes to an election
according to our classication. It is not possible to remove an outcome or add another outcome. Therefore, whether or not the IIA
assumption is veried is irrelevant. For instance, the IIA assumption implies that the relative likelihood of LP versus WP does not
depend on whether LCS is a probable outcome or not. In other words, asking whether this implication of the IIA assumption is
veried or not is equivalent to asking the following question. If itwere not possible for the incumbent to cling to power if he loses the
election, would the LP outcome become more likely relative to the WP outcome? The answer to that question is not easy but for our
research, the question itself is irrelevant because it is always possible for the incumbent to cling to power.
The data is not treated as a panel because every election tends to be a unique event in time. Electoral conditions even within the
same country vary a lot from time to time. However, dummies for various decade-periods and country specic variables are
© 2014 The Authors. African Development Review © 2014 African Development Bank
618 K. Hausken and M. Ncube
included as control variables. Categorical regressors are included as a set of dummies. Regressors that perfectly predict failure or
success in some outcomes were excluded, otherwise estimation is rendered impossible. A robust estimation technique is used to
control for heteroscedasticity and possible outliers.
4.2 Relative Likelihood
Tables 8a, 8b and 8c report the relative likelihood of having respectively LP vs WP, LCS vs WP and WCS vs WP for a unit
increase in the regressor value (continuous regressors) or a category switch (categorical regressors), which means switch of
outcome. This relative likelihood (also referred to as risk) is calculated as the ratio of the probability of one outcome to the base
outcome, for a one unit change in the regressor. However, for a categorical regressor (qualitative variable) the relative likelihood
is the ratio of the probability of one outcome to the base outcome, when the regressor switches from the base outcome to any of
the remaining outcomes. The interpretation of the relative likelihood is made relatively to the base outcome where it is equal to
one. For instance, what does a relative likelihood of LP vs WP of 0.7 after a unit increase in a regressor mean? In the base
outcome the relative likelihood is one, which means that LP and WP are equally probable. But after a unit increase in the
regressor, LP becomes 0.7 times less likely than WP. More generally, a relative likelihood of LP vs WP less than one, means that
a unit increase in the regressor of interest shifts the odds in favour of WP relative to LP. Likewise, a relative likelihood of LP vs
WP greater than one means that a unit increase in the regressor of interest shifts the odds in favour of LP relative to WP.
LP vs WP
At a 5 per cent level, only the number of years spent in power or the religious fractionalization index can signicantly turn the
odds in favour of WP instead of LP (see Table 8a). For an additional year spent in power by the incumbent, there is a relative
likelihood of 0.08 that an election results in LP vs WP. Economic performance and social factors do not matter much for this
Table 8a: Robust multinomial logit regression results for the LP election outcome
Relative likelihood zp>z
Presidential elections
*
6.546043 1.79 0.073
Economic factors
Real per capita GDP growth 1.179175 1.45 0.148
Real per capita GDP growth (1) 1.093019 0.71 0.478
Public investment as a share of GDP growth 1.01468 0.76 0.445
Public investment as a share of GDP growth (1) 0.9743023 0.73 0.468
Public consumption as a share of GDP growth 0.9291368 0.3 0.760
Public consumption as a share of GDP growth (1) 0.8109139 1.45 0.148
Social factors
Enrolment in tertiary education 0.9625418 0.14 0.889
Inequality 0.9126661 0.93 0.352
Ethnic fractionalization 1.044613 0.94 0.348
Religious fractionalization
**
0.952821 2.03 0.042
Political factors
Number of years incumbent has been in power
***
0.085349 3.04 0.002
Strong opposition
*
19.64509 1.66 0.097
Single party political system
*
19.4085 1.65 0.099
Incumbent from military 0.4485358 0.41 0.683
Coastal country 1.263784 0.09 0.925
Natural resources
Moderate natural resources 0.2945416 0.8 0.423
Abundant natural resources 5.568181 0.66 0.508
*** signicant at the 1% level; ** signicant at the 5% level; * signicant at the 10% level.
© 2014 The Authors. African Development Review © 2014 African Development Bank
Determinants of Election Outcomes 619
likelihood. A one percentage increase in the religious fractionalization index puts the relative odds of LP vs WP at 0.95.
Economic performance and social factors matter to a low degree for this relative likelihood.
LCS vs WP
If growth in the share of public investment in GDP the year before the election increases by one percentage point, the likelihood
that the election results in WP or LCS is about the same (see Table 8b). Switching from multiparty elections to single party
elections signicantly turns the odds in favour of WP with a relative likelihood of 0.02. If the incumbent is from the military, the
probability that the election results are LCS-type becomes nine times the probability that it results in WP. If the country has a
moderate natural resource endowment instead of no or few natural resources then the odds are higher that the outcome is WP than
it is LCS. Social factors do not matter at a 5 per cent level.
WCS vs WP
Since WCS and WP outcomes are the two outcomes where the incumbent wins, the relative likelihood of WCS vs WP could be
interpreted as the risk of contestation when the incumbent wins (see Table 8c). As expected, the likelihood of contestation is
signicantly higher when the opposition is stronger and economic performance is poorer.
A one percentage point increase in the growth of public consumption as a share of GDP the year of the election will put the
relative likelihood of WCS vs WP at 0.81. A one percentage point increase in the growth of public investment as a share of GDP
the year of the election will put the relative likelihood of WCS vs WP at 0.96. A lagged increase in public investment and public
consumption do not signicantly change the likelihood of contestation in the event of incumbent victory. Social factors such as
tertiary education and ethnic and religious fractionalization also matter. Contestation is likely to occur when voters have more
Table 8b: Robust multinomial logit regression results for the LCS election outcome
Relative likelihood zp>z
Presidential elections 3.464222 1.35 0.177
Economic factors
Real per capita GDP growth
*
0.7994088 1.67 0.094
Real per capita GDP growth(-1) 0.7992969 1.29 0.196
Public investment as a share of GDP growth 1.01516 1.62 0.105
Public investment as a share of GDP growth (1)
**
1.02257 2.12 0.034
Public consumption as a share of GDP growth 0.9687683 0.35 0.730
Public consumption as a share of GDP growth (1) 0.9350947 0.69 0.489
Social factors
Enrollment in tertiary education 0.8453182 1.39 0.164
Inequality 1.163716 0.54 0.592
Ethnic fractionalization 1.074441 1.28 0.202
Religious fractionalization
*
0.912011 1.66 0.097
Political factors
Number of years incumbent has been in power 0.8988326 1.62 0.105
Strong opposition 0.0419288 1 0.320
Single party political system
**
0.0204111 2.14 0.032
Incumbent from military
**
7.49151 2.07 0.038
Coastal country 0.7866554 0.18 0.854
Natural resources
Moderate natural resources
***
0.0041545 3.97 0.000
Abundant natural resources 13.89547 1.17 0.242
*** signicant at the 1% level; ** signicant at the 5% level; * signicant at the 10% level.
© 2014 The Authors. African Development Review © 2014 African Development Bank
620 K. Hausken and M. Ncube
tertiary education. The likelihood of contestation is also signicantly higher when ethnic fractionalization and religious
fractionalization increase.
4.3 Testing the Hypotheses by Considering Marginal Effects
The relative likelihood analysis in the previous subsection shows how the relative likelihood of an outcome behaves relatively to
the base outcome when regressors vary. However, this analysis does not test the hypotheses succinctly. To do so we must
calculate the marginal effects of the regressors and test their signicance and their signs. The marginal effects of each of the
regressors in the multinomial logit regression are reported in Table 9, where each row corresponds to a designated hypothesis.
We test the following succinctly formulated hypotheses:
Economic Factors
Testing Hypothesis A
Hypothesis A1: When the real per capita GDP growth increases in the year of the election, the probability of the
incumbent losing, LP or LCS, decreases.
Hypothesis A2: When the real per capita GDP growth increases in the year preceding the election, the probability of
the incumbent losing, LP or LCS, decreases.
The difference between Hypotheses A1 and A2 is that the former tests GDP growth in the year of the election, and the latter tests
per capita GDP growth in the year preceding the election, expressed with (1)in Table 9, which eliminates the impact of the
Table 8c: Robust multinomial logit regression results for the WCS election outcome
Relative likelihood zp>z
Presidential elections
***
107.4596 3.35 0.001
Economic factors
Real per capita GDP growth 0.9436423 1.03 0.304
Real per capita GDP growth (1) 0.9478922 0.9 0.367
Public investment as a share of GDP growth
***
0.9634736 2.66 0.008
Public investment as a share of GDP growth (1)
*
1.016657 1.88 0.060
Public consumption as a share of GDP growth
**
0.8198925 2.26 0.024
Public consumption as a share of GDP growth (1) 0.9163532 1.35 0.178
Social factors
Enrolment in tertiary education
***
1.308708 3.36 0.001
Inequality 0.9932135 0.1 0.919
Ethnic fractionalization
***
1.072039 3.24 0.001
Religious fractionalization
***
1.055098 3.09 0.002
Political factors
Number of years incumbent has been in power
***
0.9248327 0.98 0.328
Strong opposition
***
43.93362 4.42 0.000
Single party political system 1.263435 0.2 0.840
Incumbent from military 1.537599 0.29 0.775
Coastal country
***
14.08584 3.12 0.002
Natural resources
Moderate natural resources
***
0.0151371 4.3 0.000
Abundant natural resources
***
0.0047783 3.77 0.000
*** signicant at the 1% level; ** signicant at the 5% level; * signicant at the 10% level.
© 2014 The Authors. African Development Review © 2014 African Development Bank
Determinants of Election Outcomes 621
election on GDP growth. Real per capita GDP growth has a signicant effect on the likelihood of the incumbent losing the
election. Overall, for Hypothesis A1 (GDP growth in the year of the election) the probability that the incumbent loses (LPþLCS)
decreases by 0.0053 ¼0.0084 0.0137 as per capita GDP growth increases by one percentage point. Thus Africa is not an
exception to the theory of economic determinants of elections. Good economic conditions do boost the popularity of the
incumbent as suggested by the work of Lewis-Beck and Stegmaier (2000). When distinguishing between the LP and the LCS
outcomes, the picture becomes puzzling. The impact of growth on the probability of the LCS outcome is as expected. If real per
capita GDP growth increases by 1 per cent, the probability of the LCS outcome decreases by 0.0137. However, the impact of
growth on the probability on the LP outcome seems counterintuitive. If real per capita GDP growth increases by 1 per cent, the
probability of the LP outcome increases by 0.0084. This puzzling impact could reect the fact that real per capita GDP growth
captures the impact of other unobserved factors such as the quality of institutions and the Acemoglu and Robinson (2006)
effect. On the one hand, sound democratic institutions are known to enhance growth and Acemoglu and Robinson (2006)
suggested that colonial origins of a country determine the type of institutions that arise in the country. On the other hand,
sound democratic institutions increase the chances the incumbent concedes defeat if he loses, which in turn increases the
likelihood of the LP outcome. Thus, the positive relationship between the probability of the LP outcome and growth should be
interpreted as a correlation rather than a causal relationship. The Acemoglu and Robinson (2006) effect is further discussed later
in the paper
Testing Hypothesis B
Hypothesis B1: When the public investment as a share of GDP growth increases, in the year of the election, the
probability of the incumbent losing, LP or LCS, decreases.
Hypothesis B2: When the public investment as a share of GDP growth, in the year preceding the election, the
probability of the incumbent losing, LP or LCS, decreases.
Table 9: Multinomial logit regression: marginal effects
Variable LP LCS WCS WP
M Presidential elections 0.0216 0.0161 0.2278 0.2278
***
Economic factors
A1 Real per capita GDP growth 0.0084
**
0.0137
**
0.0020 0.0073
A2 Real per capita GDP growth (1) 0.0056 0.0131 0.0012 0.0087
B1 Public investment as a share of GDP growth 0.0006 0.0010
*
0.0020
**
0.0003
B2 Public investment as a share of GDP growth (1) 0.0013 0.0013
***
0.0007
*
0.0009
C1 Public consumption as a share of GDP growth 0.0009 0.0007 0.0089
**
0.0091
C2 Public consumption as a share of GDP growth (1) 0.0063 0.0011 0.0020 0.0095
*
Social factors
D1 Enrolment in tertiary education 0.00186 0.0118 0.0151
***
0.0014
D2 Inequality 0.0046
*
0.0095 0.0011 0.0039
E1 Ethnic fractionalization 0.0409 0.3067 0.2414
**
0.5891
**
E2 Religious fractionalization 0.1539 0.3918
***
0.3634
***
0.1823
Political factors
F1 Number of years incumbent has been in power 0.0872
***
0.0168
**
0.0154
***
0.0549
***
G1 Strong opposition 0.0834 0.1438
***
0.2019
***
0.1415
**
H1 Single party political system 0.1282
***
0.2478
***
0.0150 0.1046
I1 Incumbent from military 0.0454 0.1204
**
0.0069 0.0819
J1 Coastal country 0.0025 0.0369 0.1082
***
0.0689
Natural resources
K1 Moderate natural resources 0.0264 0.1447
***
0.2204
***
0.3387
***
K2 Abundant natural resources 0.0632 0.2332
***
0.3123
***
0.0159
*** signicant at the 1% level; ** signicant at the 5% level; * signicant at the 10% level.
© 2014 The Authors. African Development Review © 2014 African Development Bank
622 K. Hausken and M. Ncube
An increase by one percentage point in public investment as a share of GDP the year before the election (Hypothesis B2)
increases the likelihood of a LCS outcome by 0.0013 while it has no signicant effect on the probability of winning the election,
but increases the probability of the WCS outcome by 0.0007. The impact on the WP outcome is not signicant. This pertains to
Hypothesis B2. An increase in public investment during the year of the election has no signicant impact on the chances of
winning without contestation, but has a weakly signicant positive impact on the chances of clinging to power when losing.
These results are in line with our Hypothesis B1 that an increase in public investment at the eve of the election or during the
election does not boost the popularity of the incumbent because the returns on the investment will be mostly felt only after the
election. However, when losing, the incumbent would not want to concede defeat because he wants to rip the benets of his
investment.
Testing Hypothesis C
Hypothesis C1: When the public consumption as a share of GDP growth increases, in the year of the election, the
probability of the incumbent losing, LP or LCS, decreases.
Hypothesis C2: When the public consumption as a share of GDP growth increases in the year preceding the election,
the probability of the incumbent losing, LP or LCS, decreases.
Our results also provide weak evidence (at the 10 per cent level) that an increase in public consumption in the year of election
increases the chances of victory without contestation, WP, in line with Hypothesis C1. For Hypothesis C2, an increase in public
consumption in the preceding year signicantly increases the probability of winning but with a collation outcome, WCS. These
ndings are similar to that of Litschig and Morrison (2010) with the qualication that their dataset consists of municipal elections
in Brazil and their evidence is stronger than ours. An increase in public spending the year of the election, whether it is through
higher public consumption or investment, reduces the probability of contestation when winning. Such a nding is consistent with
the work of Drazen and Eslava (2010) that provided a theoretical model and an empirical test that shows that voters do positively
respond to the electoral budget cycle. For poorer countries, the share of the public goods to GDP ratio is high. This implies that
even a small public project can have a huge impact, compared to richer countries. The differential impact of public goods is
recognized but not analysed in the results.
Social Factors
Testing Hypothesis D
Hypothesis D1: When enrolment in tertiary education increases, the probability of the incumbent contesting the
election when winning, WCS, increases.
The Hypothesis D1 that education increases political awareness in Africa is supported by the results. If the enrolment rate in
tertiary education increases by one percentage point, the probability of the WCS outcome increases by 0.0151. Thus there are
more contestations of the incumbents victory in countries with more educated populations since the population obtains a better
understanding of the political, social and economic situation of their country. One puzzling observation in Africa is the low level
of incumbents victory contestation despite the reputation of the continent for vote rigging. Only 7 per cent of election results are
contested. The fact that contestations are more likely when the population is more educated (see, for instance, Hillygus, 2005),
suggests an explanation for the puzzle. There is a lack of political awareness in Africa. Since political awareness increases with
education, more education leads to more contestation of the incumbents re-election.
Hypothesis D2: When the inequality expressed as by the Gini coefcient increases, the probability of the incumbent
losing power LP, increases.
The Hypothesis D2, that an increase in inequality, as measured by the Gini coefcient, increases the probability of the incumbent
losing power, LP, is supported by the ndings. Increases in inequality are viewed negatively by the population that may be
largely poor, and therefore would not be supportive of the incumbent.
© 2014 The Authors. African Development Review © 2014 African Development Bank
Determinants of Election Outcomes 623
Testing Hypothesis E
Hypothesis E1: When ethnic fractionalization increases, the probability of the incumbent clinging to power when
losing, LCS, is not affected signicantly.
Ethnic fractionalization has no signicant impact on the probability of clinging to power when losing. Since losing and clinging
to power is clearly an autocratic outcome, this result does not support the theories of Rabushka and Shepsle (1972) and Horowitz
(1985) that social divisions threaten democracies. It does support the claim of Chandra (2005) that social divisions per se do not
negatively affect democracies and that it is rather the institutional framework within which the plural society evolves that shapes
the political regime. We do not formally test this hypothesis of Chandra (2005) because we do not observe whether the
institutions in place in a given country are pro or con social plurality.
Hypothesis E2: When the religious fractionalization increases, the probability of the incumbent clinging to power
when losing, LCS, decreases.
Unexpectedly, higher religious fractionalization reduces the probability of clinging to power when losing. This is clearly
additional evidence against the outbidding theory of Rabushka and Shepsle (1972) and Horowitz (1985). On the other hand it
tends to support the view of Chandra (2005) that social pluralism can encourage democracy. However, one question arises: why
does ethnic fractionalization not impact the chances of clinging to power when losing, while religious fractionalization does? Is it
because institutions in Africa serve well enough religious pluralism but hamper ethnic pluralism? Or is it because the costs of
clinging to power for the incumbents are higher in the presence of religious fractionalization than in the presence of ethnic
fractionalization? This latter explanation could be true if religious contestations are more violent and ercer than ethnic
contestations. More research is needed to elucidate this unexpected observation. Finally, both ethnic fractionalization and
religious fractionalization increase the likelihood of a contested victory of the incumbent. This result seems to give some support
to the theory of Rabushka and Shepsle (1972) and Horowitz (1985) since it shows that more fractionalization leads to more
contestation when the incumbent wins. However, if the elections are rigged in the rst place, a contested victory is actually a
democratic outcome. Hence to the extent that one believes that the majority of elections in Africa where the incumbent is re-
elected are rigged, this nding rather contradicts the theory of Rabushka and Shepsle (1972) and Horowitz (1985). Instead the
nding suggests that religious and ethnic pluralism could foster democracy.
Political Factors
Testing Hypothesis F
Hypothesis F1: When the number of years the incumbent has been in power increases, the probability of the
incumbent losing, LP, decreases, and the probabilities of the other three outcomes, LCS, WCS, and WP, increase.
An additional 5-year mandate in power signicantly decreases the probability of the LP outcome by 0.43. However, it increases
the probability of the WP outcome by 0.2745, the WCS outcome by 0.077 and the LCS outcome by 0.084. This is in line with our
expectation that the appetite for power increases with the time spent in power and the theory of Howell and Brent (2013).
Testing Hypothesis G
Hypothesis G1: When the opposition is strong, the probabilities of the incumbent clinging to power, WCS, or win
without contestation, WP, decrease.
If the opposition is strong, the incumbent is less likely to cling to power or win without contestation as expected since the costs of
electoral fraud and chances of rejection of results, are high. As expected, challengers have more freedom to campaign and
contestation is more likely to occur. The marginal effects of switching from a weak to a strong opposition are 0.141, 0.201,
0.141 for LCS, WCS and WP, respectively.
© 2014 The Authors. African Development Review © 2014 African Development Bank
624 K. Hausken and M. Ncube
Testing Hypothesis H
Hypothesis H1: Single party political systems favor the incumbent to win the election outright without contestation,
WP.
Changing the political system from multiparty to single party decreases the probability of LCS by 0.24 while it increases the
probability of LP by 0.12. Overall, as expected, the probability to lose signicantly decreases by 0.12. Accepting defeat is not a
signicant issue in a single party election since the power remains under the party control no matter the election outcome. For the
same reason there is no need to cling to power when losing.
Testing Hypothesis I
Hypothesis I1: Military incumbents are more likely to cling to power, WCS, than civil incumbents
If the incumbent is from the military, the probability of the LCS outcome increases by 0.12. This result conrms our hypothesis
that military incumbents tend to govern by force causing voter-discontent in the long run. To legitimize their power and
demonstrate their popularity to the international community, military incumbents often organize democratic elections. Voters
are likely to express their discontent through the ballot and the incumbent is likely to lose. Because the incumbent did not
expect to lose and wants to hold power, he will not concede defeat. Regarding the work of Marinov and Goemans (2014), these
results suggest that, in the African context, possible factionalism among military ofcials does not discourage them to cling to
power.
Testing Hypothesis J
Hypothesis J1: Whether the country is coastal or not has limited impact on election outcomes.
Being landlocked, as opposed to coastal countries, often implies that a country has many borders with other countries and that
there is a stronger neighbourhood effect on elections. However, this does not seem to matter in our ndings.
Natural Resources
Testing Hypothesis K
Hypothesis K1: In countries with moderate natural resources the incumbent is more likely to win the election, WP or
experience contested election results LCS or WCS.
Hypothesis K2: In countries with abundant natural resources the incumbent is more likely to experience
contestation of election results, LCS or WCS.
Resource rich countries are more likely to experience the LCS or WCS outcomes, in line with Hypothesis K2. The marginal
effect of switching from a country with moderate natural resources to abundant natural resources is 0.23. On the other hand, for
countries with moderate natural resource endowment, natural resources are less likely to experience contested elections
outcomes, in line with Hypothesis K1. These countries are more likely to experience a WP outcome. The marginal effects of
switching from a country with moderate natural resources to abundant natural resources is 0.33, 0.14 and 0.22 for the WP, LCS
and WCS respectively. These results conrm the rentier state hypothesis of Jensen and Wantchekon (2004). Jensen and
Wantchekon (2004) also concluded that lack of transparency and accountability explains why resource rich countries are more
likely to be authoritarian. For research related to the resource curse see Ross (1999), Mehlum et al. (2006) and Robinson et al.
(2006), and Ross (2001) analyses institutional breakdown compromising the governing of natural resource exports in developing
states. Hypothesis K2 is widespread in political science. The main argument presented to support this negative correlation
between natural resource wealth and democracy is that of the rentier state. According to Jensen and Wantchekon (2004), the
© 2014 The Authors. African Development Review © 2014 African Development Bank
Determinants of Election Outcomes 625
main characteristic of rentier states is a high dependence on rents produced by few economic agents. Thus rentier states are
generally autonomous and less accountable to voters. As a result, they feel less pressure for democratization.
Former Colonizer
Testing Hypothesis L
In this section we want to test the importance of former colonizer in predicting electoral outcomes. The work of Acemoglu and
Robinson (2006) asserts that the colonial origins of a country impact the institutional framework that develops subsequently.
Different types of institutions were developed by different colonizers in their colonies and indeed the same colonizers developed
different institutions in their respective colonies. For example, Jamaica and United States developed different institutions,
despite sharing the same colonizer, Britain. Due to the limited number of African countries, in a statistical sense, we only
consider the differences in colonizers as signalling, broadly, the type of institutions that developed, and we then tested the
signicance of the different colonizers.
In the general model, the variable for the former colonizer cannot be included in the list of regressors because of the
multicolinearity it could generate. It has been widely demonstrated in the literature that former colonizer is a determinant of
growth, educational level and quality of institutions in African countries. It is also widely accepted that geographic situations,
natural resource endowment and religious factors are correlated with the type of colonizer. Therefore, an indicator of the former
colonizer is highly correlated with the other regressors.
In our methodology, we used the average of the regressors by former colonizer and then nd the mean predicted probability of
having each type of outcome at these average values, for each former colonizer. Then we test the signicance of these mean
predicted probabilities. Categorical variables cannot be averaged. They are therefore excluded. However, this poses less of a
problem because the categorical variables in our regression analysis are related to but are not determined by the former colonizer.
The results are presented in Table 10.
Hypothesis L1: Regardless of the former colonizer, the most likely outcome is winning without contestation, WP, and
the least likely outcome is losing and relinquishing power, LP.
Hypothesis L2: Winning without contestation, WP, is more likely in former Belgium colonies than in former British,
French and Portuguese colonies.
Hypothesis L3: Winning with contestation, WCS, is more likely in former British colonies than in former Belgium,
French and Portuguese colonies.
Hypothesis L4: Losing and clinging to power, LCS, is more likely in former French and Portuguese colonies than in
Belgium and British colonies.
From the results in Table 10, regardless of the former colonizer, the most likely outcome is WP and the least likely outcome is LP,
thus conrming Hypothesis L1. What depends on the former colonizer is the second most likely outcome which is LCS for
French and Portuguese former colonies, and WCS for former British and Belgian colonies, thus conrming Hypothesis L4. The
highest probability for WP is recorded by former Belgian colonies, thus conrming Hypothesis L2. WCS outcomes are more
Table 10: Mean expected probabilities by former colonizer
Former colonizer LP LCS WCS WP
France 0.0002021 0.2436243
***
0.1577884
***
0.5983852
***
Britain 7.15E-08 0.1228762
***
0.1877097
***
0.689414
***
Belgium 0.0134881 0.0750294 0.0608593
***
0.8506231
***
Portugal 0.0000175 0.2248267
***
0.100267
***
0.6748888
***
***
signicant at the 1% level.
© 2014 The Authors. African Development Review © 2014 African Development Bank
626 K. Hausken and M. Ncube
likely to occur in British colonies, thus conrming Hypothesis L3, while LCS is prevalent in French colonies, thus conrming
Hypothesis L2. These results provide new empirical support for the Acemoglu and Robinson (2006) theory.
Presidential vs Legislative Elections
Testing Hypothesis M
Hypothesis M: Presidential elections are more associated with the incumbent winning outright without contestation,
than legislative elections.
From Table 9, observe that the type of election is important for whether the incumbent wins outright or not. The coefcient for
WP is signicant. Incumbents are more likely to cling to power in natural resource rich countries.
5. Conclusion
We have classied African elections into six outcomes based on ofcial election results and the observed political situation in the
country after the election. Our database includes all presidential and legislative elections in Africa over the period 19602010.
We have then tested empirically for the main determinants of election outcomes in Africa using a discrete-choice multinomial
logit model.
Our rst main contribution is the distribution of elections according to the following classication. The incumbent wins with
no contestation 63.9 per cent, coalition 6.4 per cent, and standoff 1.2 per cent. The incumbent loses and accepts defeat 15.9 per
cent, coalition 12.3 per cent, and standoff 0.3 per cent. That is, the most frequent outcome in Africa is the incumbent winning
without contestation of the results. Out of the 653 elections held, 417 were won by the incumbent without contestation, of which
210 were legislative and 207 were presidential.
Second, the multinomial logit analysis provides new empirical evidence on some theories of determinants of election
outcomes, using economic, political, social, natural resource endowment factors, and as to the type of election. Overall, 22
hypotheses were tested in the analysis above. On economic factors, we found that an increase in both the growth in per capita
GDP in the year of the elections, and the year preceding the election, the probability of the incumbent losing, referred to as
LP or LCS, decreases. The well-known hypothesis that good economic performance increases the chances that the
incumbent wins the election (Lewis-Beck and Stegmaier, 2000) is conrmed. We also nd evidence that voters respond
positively to the electoral budget cycle as theorized by Drazen and Eslava (2010). That is, increased public investment and
public consumption as a percentage of GDP decreases the probability of the incumbent losing. A policy implication of this is
that we could have cases of political-budgetary cycles where budgets typically balloon just prior to the election, and GDP
growth is not sustained beyond the election. This has the potential to present challenges in terms of scal sustainability, and
calls for medium-term budget frameworks to be instituted in order to counter this potentially damaging political budgetary
cycle.
On social factors we tested the impact of education, inequality, ethnic fractionalization, and religious fractionalization. On
education, consistent with the work of Hillygus (2005) contestation of ofcial election results when the incumbent is declared
winner is more likely to occur in countries where the population is more educated. We do not nd evidence that social pluralism
threatens democracy. This is contrary to the outbidding theory of Rabushka and Shepsle (1972) and Horowitz (1985) but it is in
line with recent work by Chandra (2005). One policy implication of these ndings is that more investment needs to go into
education to raise literacy levels, and thus create a more democratic environment.
Regarding political factors, we tested the impact of number of years in power, strength of the opposition, the presence of a
single party system, whether the incumbent has a military background, and whether the country is coastal or not. The hypothesis
that appetite for power increases with years in power as in Howell and Brent (2013) is conrmed. As one could expect, in
countries with a strong and well-organized opposition, the incumbent is less likely to cling to power if he loses. In contrast,
military incumbents tend to cling to power when they are defeated. This nding suggests that in the African context, the
argument of Marinov and Goemans (2014) that anticipated factionalism among the military ofcials could force them to
organize democratic elections and abide by the results is not veried. One policy implication of these ndings is that term-limits
© 2014 The Authors. African Development Review © 2014 African Development Bank
Determinants of Election Outcomes 627
should be imposed in national constitutions, in order to prevent the incumbent from staying too long in power and undermining
the process of creating democratic, effective and inclusive institutions.
We also tested for the impact of natural resource endowment on election outcomes. We found evidence that incumbents are
more likely to cling to power in resource rich countries as Jensen and Wantchekon (2004) and others, have argued. With regards
to policy design, this calls for setting up transparent and effective mechanisms for the managing of natural resource revenues, and
the subsequent investment of the revenues in long-term investments that not only benet the current generation but also future
generations.
In our analysis we also tested whether the type of former colonizer is signicant in the outcome of elections as this may point to
the type of institutions that have emerged through colonization. Our ndings support the argument of Acemoglu and Robinson
(2006) that colonial origins of a country impact the subsequent institutions that emerge in the country. Winning without
contestation is more likely in former Belgium colonies. Winning with contestation is more likely in former British colonies,
while losing and clinging to power is more likely in former French and Portuguese colonies. This nding points towards the need
for building inclusive and effective institutions that allow more democratic expression.
Finally, we tested whether the type of election matters, that is, presidential versus legislative elections. Presidential elections
are more associated with the incumbent winning outright without contestation, than in legislative elections. Again, this nding
points to the need for strengthening institutions for democratic expression not only at legislative levels but also at the presidential
level.
Future research could also focus on analysing the decision by an autocrat to hold elections in the rst place by an autocrat.
Follow-up research could focus on analysing the incidence of revolutions where public goods are for the benet of a few, and
voice and accountability are suppressed.
. Note
1. The literature on whether accounting for the future hardens or softens the playersstance is contested. Folk theorem arguments
(Fudenberg and Maskin, 1986) exemplied by repeating the prisoners dilemma (Axelrod, 1984) are often taken to imply
cooperation in long-term relationships. This result is often applied uncritically out of context. Skaperdas and Syropoulos (1996)
equip each agent with a resource which can be allocated into production versus arms. They show that increased importance of the
future may harm cooperation. Hausken (2005) analyses a repeated battle of the sexes where player 1 values the future and player
2 is myopic, and shows that player 1 is more inclined through conictful behaviour to risk a conict in the present when the future
is important and/or there are many periods left in the game.
References
Acemoglu, D, and J. A. Robinson (2006), Economic Origins of Dictatorship and Democracy, Cambridge University Press,
Cambridge.
Alesina, A., A. Devleeschauwer, W. Easterly, S. Kurlat and R. Wacziarg (2003), Fractionalization,Journal of Economic
Growth, Vol. 18, No. 2, pp. 15594.
Axelrod, R. (1984), The Evolution of Cooperation, Basic Books, New York.
Barreira, A. P. and R. N. Baleiras (2004), Elections and the Public Expenditure Mix, Working Paper, Faculdade de Economia,
Campus de Gambels, Universidade do Algarve, Portugal.
Berinsky, A., and G. Lenz (2011), Education and Political Participation: Exploring the Causal Link,Political Behavior,
Vol. 33, No. 3, pp. 35773. doi: 10.1007/s11109-010-9134-9.
Chandra, K. (2005), Ethnic Parties and Democratic Stability,Perspectives on Politics, Vol. 3, No. 2, pp. 23552.
Collier, P. (2009), Wars, Guns, and Votes: Democracy in Dangerous Places, Harper Collins, New York.
Cox, G. W. (1997), Making Votes Count: Strategic Coordination in the Worlds Electoral Systems, Cambridge University Press,
Cambridge.
© 2014 The Authors. African Development Review © 2014 African Development Bank
628 K. Hausken and M. Ncube
De Mesquita, B. B., A. Smith, R. M. Siverson and J. B. Morrow (2003), The Logic of Political Survival, MIT Press, Cambridge,
MA.
Dercon, S. and R. Gutierrez-Romero (2012), Triggers and Characteristics of the 2007 Kenyan Electoral Violence,World
Development, Vol. 40, No. 4, pp. 73144.
Drazen, A. and M. Eslava (2010), Electoral Manipulation via Voter-friendly Spending: Theory and Evidence,Journal of
Development Economics, Vol. 92, No. 1, pp. 3952.
Ellman, M. and L. Wantchekon (2000), Electoral Competition under the Threat of Political Unrest,Quarterly Journal of
Economics, Vol. 115, No. 2, pp. 499531.
Fearon, J. D. and D. D. Laitin (2003), Ethnicity, Insurgency and Civil War,American Political Science Review, Vol. 1,
pp. 7590.
Fudenberg, D. M. and E. Maskin (1986), The Folk Theorem in Repeated Games with Discounting or with Incomplete
Information,Econometrica, Vol. 54, pp. 53354.
Hausken, K. (2005), The Battle of the Sexes When the Future is Important,Economics Letters, Vol. 87, No. 1, pp. 8993.
Herrnstein, R. J. and C. A. Murray (1994), The Bell Curve: Intelligence and Class Structure in American Life, Free Press,
New York.
Hillygus, D. S. (2005), Campaign Effects and the Dynamics of Turnout Intention in Election 2000,The Journal of Politics,
Vol. 67, No. 1, pp. 5068.
Horowitz, D. L. (1985), Ethnic Groups in Conict, University of California Press, Berkeley, CA.
Howell, W. G. and D. M. Brent (2013), Thinking about the Presidency: The Primacy of Power, Princeton University Press,
Princeton, NJ.
Jensen, N. and L. Wantchekon (2004), Resource Wealth and Political Regimes in Africa,Comparative Political Studies,
Vol. 37, No. 7, pp. 81641.
Lewis-Beck, M. S. and M. Stegmaier (2000), Economic Determinants of Electoral Outcomes,Annual Review of Political
Science, Vol. 3, No. 1, pp. 183219. doi: 10.1146/annurev.polisci.3.1.183.
Litschig, S. and K. Morrison (2010), Government Spending and Re-election: Quasi Experimentation from Brazilian
Municipalities, Department dEconomia, Universitt Pompeu Fabra.
Marinov, N. and H. Goemans (2014), Coups and Democracy,British Journal of Political Science, Vol. 44, No. 4, pp. 799825.
Mehlum, H., R. Torvik and K. O. Moene (2006), Institutions and the Resource Curse,The Economic Journal, Vol. 116,
No. 508, pp. 120.
Nie, N. H., J. Junn and K. Stehlik-Barry (1996), Education and Democratic Citizenship in America, University of Chicago Press,
Chicago.
Persson, M. (2013), Testing the Relationship between Education and Political Participation Using the 1970 British Cohort
Study,Political Behavior,121. doi: 10.1007/s11109-013-9254-0.
Persson, T. and G. E. Tabellini (2002), Political Economics: Explaining Economic Policy, MIT Press, Cambridge MA.
Rabushka, A. and K. A. Shepsle (1972), Politics in Plural Societies: A Theory of Democratic Instability, Macmillan, London.
Robinson, J. A. and R. Torvik (2005), White Elephants,Journal of Public Economics, Vol. 89: pp. 197210.
Robinson, J. A., R. Torvik and T. Verdier (2006), Political Foundations of the Resource Curse,Journal of Development
Economics, Vol. 79, pp. 44768.
Rosenstone, S. J. and J. M. Hansen (2009), Mobilization, Participation and Democracy in America, Longman, London.
Ross, M. L. (1999), The Political Economy of the Resource Curse,World Politics, Vol. 51, pp. 297322.
© 2014 The Authors. African Development Review © 2014 African Development Bank
Determinants of Election Outcomes 629
Ross, M. L. (2001), Timber Booms and Institutional Breakdown in Southeast Asia, Cambridge University Press, New York.
Schedler, A. (2007), The Mexican Standoff: The Mobilization on Distrust,Journal of Democracy, Vol. 18, No. 1, pp. 88102.
Schmitter, P. C. (1978), Elections without Choice, Macmillan, London.
Schmitter, P. C. (1994), Danger and Dilemmas of Democracy,Journal of Democracy, Vol. 5, No. 2, pp. 5774.
Skaperdas, S. and C. Syropoulos (1996), Can the Shadow of the Future Harm Cooperation?Journal of Economic Behavior and
Organization, Vol. 29, pp. 35572.
Vergne, C. (2009), Democracy, Elections and Allocation of Public Expenditures in Developing Countries,European Journal
of Political Economy, Vol. 25, No. 1, pp. 6377.
© 2014 The Authors. African Development Review © 2014 African Development Bank
630 K. Hausken and M. Ncube
Article
This paper analyses the impact of household food insecurity on electoral participation in 30 sub‐Sahara African countries with the aid of micro‐level data drawn from the sixth round of the Afrobarometer survey. Estimates from logistic regression indicate that being food insecure reduces the likelihood of electoral participation by 7%. Notably, results from the endogenous binary‐variable regression, which controlled for potential reverse causality, confirm that household food insecurity is a crucial driver of voter turnout in sub‐Saharan Africa. Further analysis reveals that voting behaviour was much higher and statistically significant amongst voters who were intermittently food insecure than those that were always food insecure. Finally, it appears that turnout at national elections depends mostly on the severity of food insecurity. Therefore, it can be argued that the implementation of policies aimed at stemming household food insecurity could play an essential role in increasing voter turnout.
Article
In sub-Saharan Africa (SSA), competitive elections can cause considerable violence and widespread destruction of property, most of which is ethnically motivated. Recent literature shows that ethnic identification is more prominent during competitive election periods in comparison to other identifying categories such as gender, religion, and class/occupation. This paper utilizes data from 12 SSA countries and over 40,000 respondents taken from the Afrobarometer. It asks if individual perceived living conditions changes in the run-up to competitive elections. Strong evidence shows that a perceived living condition does change. It is positively related to the proximity to an election and this proximity effect depends on the competitiveness of the election. The paper further investigates the background mechanisms behind this positive relationship, that is, to what extent does living conditions of the individual change if the party that the individual supports wins the election and is there a change in living conditions of the individual before and after the election? In addition, this paper documents that ethnic identification also has a positive impact on individual perceived living conditions after controlling for electoral cycle variables.
Book
Full-text available
All American presidents, past and present, have cared deeply about power--acquiring, protecting, and expanding it. While individual presidents obviously have other concerns, such as shaping policy or building a legacy, the primacy of power considerations--exacerbated by expectations of the presidency and the inadequacy of explicit powers in the Constitution--sets presidents apart from other political actors. Thinking about the Presidency explores presidents' preoccupation with power. Distinguished presidential scholar William Howell looks at the key aspects of executive power--political and constitutional origins, philosophical underpinnings, manifestations in contemporary political life, implications for political reform, and looming influences over the standards to which we hold those individuals elected to America's highest office. Howell shows that an appetite for power may not inform the original motivations of those who seek to become president. Rather, this need is built into the office of the presidency itself--and quickly takes hold of whoever bears the title of Chief Executive. In order to understand the modern presidency, and the degrees to which a president succeeds or fails, the acquisition, protection, and expansion of power in a president's political life must be recognized--in policy tools and legislative strategies, the posture taken before the American public, and the disregard shown to those who would counsel modesty and deference within the White House. Thinking about the Presidency assesses how the search for and defense of presidential powers informs nearly every decision made by the leader of the nation. In a new preface, Howell reflects on presidential power during the presidency of Barack Obama.
Article
Full-text available
According to conventional wisdom in political behavior research, education has a direct causal effect on political participation. However, a number of recent studies have questioned this established view by arguing that education is not a direct cause of political participation but only a proxy for other factors that are not directly related to the educational experience. This paper engages in a current debate regarding the application of matching techniques to assess whether there is a direct causal effect of education on political participation. It uses data from a British cohort study that follows everyone born during 1 week in the UK in 1970. The data includes a rich set of variables measuring factors through childhood and adolescence such as cognitive ability and family socioeconomic status. This data provides the opportunity to match on a number of important variables that are not included in the US datasets used by previous studies in the field. Results show that after matching there are no significant effects of education on political participation.
Book
This book is about differences in intellectual capacity among people and groups and what those differences mean for America's future.(preface) The major purpose of this book] is to reveal the dramatic transformation that is currently in process in American society---a process that has created a new kind of class structure led by a "cognitive elite," itself a result of concentration and self-selection in those social pools well endowed with cognitive abilities. Herrnstein and Murray explore] the ways that low intelligence, independent of social, economic, or ethnic background, lies at the root of many of our social problems. The authors also demonstrate the truth of another taboo fact: that intelligence levels differ among ethnic groups. (PsycINFO Database Record (c) 2012 APA, all rights reserved)(jacket)
Article
We use new data on coup d’états and elections to document a striking development: whereas the vast majority of successful coups before 1991 installed durable rules, the majority of coups after that have been followed by competitive elections. We argue that after the Cold War international pressure influenced the consequences of coups. In the post-Cold War era those countries that are most dependent on Western aid have been the first to embrace competitive elections after the coup. Our theory also sheds light on the pronounced decline in the number of coups since 1991. While the coup d’état has been and still is the single most important factor leading to the downfall of democratic government, our findings indicate that the new generation of coups has been far less harmful for democracy than their historical predecessors.