Since transitioning to multipartyism in 1991, Zambia has experienced three electoral turnovers. The latest incumbent defeat happened in 2021 when the ruling Patriotic Front (PF) was overwhelmingly voted out of office after 10 years in power. Amid a deep economic crisis, PF lost votes across the territory, but nowhere was the collapse as eye-catching as in the urban centres of Lusaka and Copperbelt.
Given the party's history, PF's fall from grace with urban voters is remarkable. When the party first came to power, it did so largely on a wave of urban discontent. Touting a populist message geared at the urban poor, the PF could mobilise an urban electorate with little improvement in living standards under the lengthy rule of the previous incumbent Movement for Multiparty Democracy (MMD). When the opposition United Party for National Development (UPND) outperformed the PF in both Lusaka and Copperbelt in 2021, this made for a major electoral upset (Larmer & Fraser Reference Larmer and Fraser2007; Cheeseman & Hinfelaar Reference Cheeseman and Hinfelaar2010; Resnick Reference Resnick2014; Siachiwena Reference Siachiwena2021a). While astonishing, PF's urban decline is not unique in Zambia's multiparty history. In fact, MMD had suffered a similar defeat amid dwindling urban support since 2001 and a complete collapse in national support in 2011.
In the small but growing literature on African party switching, researchers have identified important differences between urban and rural voters in their propensity to shift political allegiances (Wahman & Boone Reference Wahman and Boone2018). Urban Africa has been known as the hotbed for opposition politics (Koter Reference Koter2013; Wahman & Boone Reference Wahman and Boone2018; Harding Reference Harding2020). However, as several newly elected governments across Africa have experienced, holding on to an urban base is challenging. While rural electorates have often aligned themselves with the new ruling party, sometimes hoping that such alignments will boost access to centralised economic resources, urban electorates have generally deserted new ruling parties within one or two election cycles (Galvan Reference Galvan2001).
Previous studies on retrospective economic performance in Zambia show that voters use their experiences with service delivery when deciding whether to re-elect incumbent candidates (Hern Reference Hern, Banda, Kaaba, Hinfelaar and Ndulo2020; Seekings Reference Seekings, Banda, Kaaba, Hinfelaar and Ndulo2020, Reference Seekings2022). These studies show that voters often balance experiences with service delivery or economic performance with ethnic considerations. Hern (Reference Hern, Banda, Kaaba, Hinfelaar and Ndulo2020) further demonstrates that citizens are less likely to vote altogether if they are dissatisfied with the incumbent's performance and do not perceive the opposition as a better alternative. However, the existing studies do not address party switching specifically, nor do they consider the variation in economic evaluations of the governing party between urban and rural voters.
In this paper, we analyse the recent turnover in the Zambian 2021 general election. While the literature has suggested several plausible explanations for urban party switching in Africa, we particularly focus on the political economy basis of party switching. We analyse the economic crisis dominating the 2021 election and argue that the main manifestations of this crisis, including high unemployment, currency devaluation, and staggering inflation, affected urban voters harder than their rural counterparts. Zambia, in this sense, resembles several other African countries where restrained economic policy, structural adjustment, and a wilful prioritisation of rural economic interest have led to significant discontent among urban electorates (Bawumia Reference Bawumia1998; Bates & Block Reference Bates and Block2013; Fraser Reference Fraser2017; Harding Reference Harding2020).
We rely on a mix of qualitative and quantitative analysis. The qualitative analysis draws on several studies and reports on Zambia's economy covering 2011 to 2021. The quantitative analysis makes use of two datasets. First, a dataset of constituency-level election results from 2016 and 2021 paired with constituency-level census data. These data are used to establish that PF lost more support in urban constituencies, even in control for differences in ethnic demography. Second and more importantly, we introduce a new panel survey of Zambian voters in urban and rural PF strongholds. The panel ranges from 2019 until after the 2021 election and includes a sample of almost 1,500 respondents. Panel evidence of voter behaviour in Africa remains uncommon but is key to evaluating individual-level volatility. Aside from Seekings (Reference Seekings2022), this is one of the first studies using such data to analyse Zambian voter behaviour.Footnote 1
The survey, in combination with constituency-level voting data, confirms that urban voters were more likely to abandon the PF than their rural counterparts. We find little evidence to suggest that such differences are explained by ethnic calculus or campaign effects. Instead, we find further individual-level evidence to suggest that perceptions of economic performance can explain the increased propensity for urban voters to desert the government party. Urban voters were more likely to evaluate the government party poorly on the economy. They were also less likely than rural voters to continue supporting the PF when they deemed economic performance poor.
The paper contributes to the growing literature on party switching in Africa and to more general discussions on retrospective and economic voting in Zambia (Hern Reference Hern, Banda, Kaaba, Hinfelaar and Ndulo2020; Seekings Reference Seekings, Banda, Kaaba, Hinfelaar and Ndulo2020, Reference Seekings2022) and other countries where the structure of the economy is spatially uneven (e.g. Lewis-Beck & Stegmaier Reference Lewis-Beck and Stegmaier2008; Wimpy & Whitten Reference Wimpy and Whitten2017; Rhee Reference Rhee2021). The findings underscore how commonly used economic metrics, such as inflation, GDP growth and unemployment, may vary in relevance across urban and rural locations.
African party switching and economic voting
A certain level of voter mobility is crucial for any functioning democracy. Without at least a subsection of the electorate willing to re-evaluate vote choice between elections, democracy becomes a census without the prospects for meaningful accountability (Riker Reference Riker1982). Mobile voters have become a main preoccupation in the political behaviour literature, and important work has tried to develop further understandings of the micro-level characteristics of so-called ‘party switchers’ (van der Meer et al. Reference Van der Meer, van Elas, Lubbe and van der Brug2015; Dassonneville et al. Reference Dassonneville, Blais and Dejaeghere2015). These are voters who vote for one party in one election only to change their vote choice to a different party in the subsequent election.Footnote 2
While African elections have been known for extreme levels of party volatility between electoral cycles (Weghorst & Bernhard Reference Weghorst and Bernhard2014; Kuenzi et al. Reference Kuenzi, Tuman, Rissmann and Lambright2019), such volatility has generally been understood as reflecting elite-level realignments rather than ‘true’ voter mobility. Particularly, in accounts of African voting focused on stable cleavages, such as ethnicity, electoral volatility has been regarded almost exclusively from the prism of coalition politics (e.g. Arriola Reference Arriola2013).
Despite the overall emphasis on stable political cleavages, a few noteworthy contributions to the African voting literature have focused on party switching. In pioneering work by Lindberg & Morrison (Reference Lindberg and Morrison2005) from Ghana, the authors show that a sizeable 18% of Ghanaian voters could be considered party switchers. They also find that party switchers tend to be motivated by candidate evaluation and perceptions of performance. In later contributions, Weghorst & Lindberg (Reference Weghorst and Lindberg2013) show that some party switchers may also be motivated by clientelism. Horowitz (Reference Horowitz2019), studying Kenya, shows that voters without a co-ethnic in the race are more likely to change their voting intentions during the campaign period. Utilising constituency-level election results from 28 elections in 7 countries (Botswana, Ghana, Kenya, Malawi, Tanzania, Zambia and Zimbabwe), Wahman & Boone (Reference Wahman and Boone2018) consider the constituency-level propensity for ‘government-opposition swing’, showing that urban constituencies are more likely to ‘swing away’ from the government party in favour of the opposition.
The literature on party switching is highly related to theories on economic voting (Powell &Whitten Reference Powell and Whitten1993; Duch & Stevenson Reference Duch and Stevenson2008). Economic voting literature has held that switching can be understood as a reflection of either aggregate economic evaluations or the evaluation of voters' private financial situation (Kramer Reference Kramer1983). While the relationship between macro-economic performance and electoral outcomes is complex and mitigated by the extent to which economic blame can easily be attributed to the ruling party (Powell & Whitten Reference Powell and Whitten1993; Hobolt et al. Reference Hobolt, Tilley and Banducci2013), economic voting hypotheses have been generally supported in the literature on industrialised consolidated democracies. In general, governments presiding over low growth economies, contracting labour markets, or high inflation rates tend to be punished at the ballot box (Nannestad & Paldam Reference Nannestad and Paldam1994; Lewis-Beck & Stegmaier Reference Lewis-Beck and Stegmaier2008).
The relationship between economic outputs and voting does, however, seem more ambiguous in developing economies dominated by subsistence agriculture. While Rhee (Reference Rhee2021) finds a strong relationship between voters' evaluation of governments' economic performance and vote choice in African democracies, Wimpy & Whitten (Reference Wimpy and Whitten2017) find that the degree to which growth and inflation determine incumbent party success is related to levels of media freedom. Others have questioned to what extent performance evaluations may be overshadowed by clientelistic and ethnic considerations (Conroy-Krutz Reference Conroy-Krutz2013; Carlson Reference Carlson2015).
While all this research has been important for our understanding of economic voting in Africa, there has not been much questioning of the extent to which classic measures of performance- such as employment, inflation and growth- are relevant indicators in African economies where some subnational economies are dominated by subsistence farming. And if so, whether such measures are relevant to a different extent across vastly different political economies in urban and rural locations. In rural African economies, formal wage-earning employment rates remain low. In such economies, citizens are less likely to depend on markets for purchasing the major goods they consume daily (Ruel et al. Reference Ruel, Garrett, Morris, Maxwell, Oshaug, Engle, Menon, Slack and Haddad1998; Baiphethi & Jacobs Reference Baiphethi and Jacobs2009; Wodon & Zaman Reference Wodon and Zaman2010). They are particularly unlikely to consume many imported goods.
In cities, however, citizens are net consumers of agricultural goods and are more integrated into markets for employment and consumption (Christiaensen & Demery Reference Christiaensen and Demery2007). They are more sensitive to inflation both on imported goods and domestically produced food goods. In densely populated areas, reliant on natural resource extraction or service provision, citizens are highly dependent on the state of the labour market for their subsistence.
The political economy of urban opposition in Zambia
Urban Africa has historically been the epicentre of opposition activity. This was true in the colonial period, where civil society, trade unions and a marginalised African clerical class became pivotal for the anti-colonial struggle (Coleman Reference Coleman1954; Henderson Reference Henderson1973). It was also true in the post-independence period, where urban populations proved to be more mobilisable in the face of deteriorating labour conditions, poor housing, inflation and poor economic outputs (Baylies & Szeftel Reference Baylies, Szeftel, Cherry, Baylies and Szeftel1992; Bates & Collier Reference Bates and Collier1995; Cheeseman & Larmer Reference Cheeseman and Larmer2015; Sishuwa Reference Sishuwa2021).
As theorised by Bates (Reference Bates1981), African economic policy during authoritarian rule was biased towards urban interest to mitigate the risk of urban mobilisation.Footnote 3 In Zambia, such urban bias manifested itself as prioritising consumer over agricultural interest, high supply of foreign currency, price control and market regulation (Bates & Collier Reference Bates and Collier1995). While such policies may have reduced political pressure among urbanites, they also led to increased foreign debt and bloated government budgets. Structural adjustment programmes (SAPs), the supposed remedy to these economic ills, became a hard pill to swallow for urban populations. The initial consequences of government austerity, reduced public sectors, and inflation led to urban economic protests around the continent (Bratton & van de Walle Reference Bratton and van de Walle1992). In Zambia, violent riots erupted in 1986 in Lusaka and the Copperbelt due to the lifting of government subsidies on the staple food, mealie-meal, leading to a 100% price hike in consumer prices (Sano Reference Sano1988). Economic protest, often led by urban trade unions and student organisations, ignited the hasty transition to multipartyism across the continent (Bratton & van de Walle Reference Bratton and van de Walle1992).
In Zambia's first multiparty election, the opposition MMD built a strong and broad coalition against the ruling UNIP. The MMD was an alliance of trade union leaders and ‘local capitalists’ with significant business interests who opposed Kenneth Kaunda's one-party rule and statist economic policies (Baylies & Szeftel Reference Baylies, Szeftel, Cherry, Baylies and Szeftel1992). The initial organisation and mobilisation of the MMD yet again originated in the Copperbelt, building on the organisational strength of trade union structures (LeBas Reference LeBas2011). The urban food riots of the late 1980s and the general economic decline accompanying UNIP policies also fuelled demands for economic and political reforms, especially in urban areas (Rakner Reference Rakner2003).
Once in power, the MMD implemented SAPs supported by the International Monetary Fund and the World Bank. These reforms aimed to attain substantial debt relief and implement an economic recovery programme. However, Zambia's economic conditions worsened rather than improved throughout the 1990s. Urban unemployment skyrocketed, and cities were struck by a cost-of-living crisis as well as a general decline in both economic growth and per capita income (Resnick Reference Resnick2014).
By the late 1990s, the MMD's dominance began to unravel with the formation of viable opposition parties. One new opposition party created in 2001 was the Patriotic Front (PF), led by Michael Sata. While the PF initially attained only limited success, the party took off in the 2006 election after Sata adopted ‘populist’ strategies to mobilise urban informal economy workers in Lusaka and mine workers on the Copperbelt (Larmer & Fraser Reference Larmer and Fraser2007; Resnick Reference Resnick2014; Sishuwa Reference Sishuwa2021). By 2011, PF remarkably defeated MMD in an election where the ruling party's urban support base had been completely depleted.
The PF's urban base: from strength to collapse
The PF came to power promising to address the economic concerns of urban Zambians who were disillusioned with the failures of economic growth to translate into improved living conditions for the majority (Cheeseman et al. Reference Cheeseman, Ford and Simutanyi2014). Most PF urban supporters were informal economy workers who make up 81% of the urban labour force (Hinfelaar et al. Reference Hinfelaar, Resnick and Sishuwa2020). This includes market traders, bus and taxi drivers, street vendors, and hawkers who expected policy adjustments to serve their interests. The PF's core promises included lowering taxes, creating jobs, and improving the quality of basic service delivery through the expansion of universal education and healthcare (Larmer & Fraser Reference Larmer and Fraser2007).
Once in government, the PF adopted a ‘leftist’ policy platform.Footnote 4 After one year in office, the party had implemented measures that benefitted urban citizens, including wage increases for public sector workers, increases in the mineral royalty tax, a 108% increase in the minimum wage, and adjustments to strengthen social security for semi-skilled workers such as shop assistants and domestic workers (Helle & Rakner Reference Helle and Rakner2012; Kim Reference Kim2017; Hinfelaar & Sichone Reference Hinfelaar and Sichone2019). University students at public institutions benefited from a 100% increment in the monthly living allowances paid for by the government. The government also lifted restrictions on street vending and hawking, which were imposed by the previous administration (Hinfelaar et al. Reference Hinfelaar, Resnick and Sishuwa2020).
The PF further implemented agricultural reforms that had implications for urban and rural Zambians alike. Under the MMD, the government implemented a large-scale fertiliser subsidy programme that increasingly benefitted rural districts where the party was dominant (Kim Reference Kim2017). With Sata as president, the PF government cut fertiliser subsidies for small-scale farmers from 75% to 50% and limited other agriculture spending (Ibid: 30). The savings made from the fertiliser subsidies were reportedly channelled to expand a social cash transfer programme, which the PF regarded to be a more efficient means of addressing rural poverty than subsidies for small-scale farmers (Siachiwena Reference Siachiwena2021b).
Perhaps the most significant reform under the PF was the massive expansion of infrastructure, focusing on roads, railways, power, education and health facilities (Siachiwena Reference Siachiwena2021b). Zambia had seen little infrastructure development since the early 1990s, when the MMD began implementing structural adjustment reforms. Sata initiated a Link 8,000 project aimed at constructing 8,000 km of tarred roads to address the country's dilapidated road infrastructure within five years.
Much due to these popular and short-term beneficial policies, PF's urban base remained robust in two presidential elections organised in 2015 and 2016. Michael Sata died in office in 2014, but his successor, Edgar Lungu, managed to win a by-election with significant urban support in 2015. He later also managed to fend off opposition from UPND's Hakainde Hichilema in the regular 2016 election (albeit with a somewhat smaller majority). However, as the economy went from bad to worse, the party's popularity in urban areas dwindled further after 2016.
In the long run, PF's focus on infrastructure as their major policy achievement turned out to be out of sync with the priorities of urban voters. Infrastructure investments were meant to create semi-skilled jobs for the unemployed, contribute to economic growth, and, in turn, generate revenues to repay debts. Instead, it contributed to spiralling foreign debt and was generally used as a form of patronage for PF elites without many benefits to local urban economies (Hinfelaar et al. Reference Hinfelaar, Resnick and Sishuwa2020).
Zambia experienced a significant economic decline in 2015, which resulted in the collapse of real GDP growth from an average of 7.7% between 2005 and 2014 to 2.9% in 2015 (Cheelo & Haatongo-Masenke Reference Cheelo and Haatongo-Masenke2018). The government attributed the challenges to external shocks such as adverse weather conditions and a drop in global commodity prices which affected copper prices (Republic of Zambia 2020). However, a study by a Zambian government think-tank attributed the economic crisis to flawed budget management, increased debt levels, and lack of fiscal discipline (Banda-Muleya et al. Reference Banda-Muleya, Nalishebo, Kalikeka, Mungomba, Masilokwa and Cheelo2019).
Critics of the government also observed increased political interference in the management of the economy, especially after Lungu's election in 2015 (Hinfelaar & Sichone Reference Hinfelaar and Sichone2019). The PF was able to stave off the effects of the crisis at the time of the 2016 elections, but the crisis worsened over the next five years. The effects were palatable among several core urban constituencies as the debt and economic crises persisted. The government struggled to pay local contractors, experienced delays in paying civil servant salaries and was unable to sustain the monthly meal allowances paid to university students (Hinfelaar et al. Reference Hinfelaar, Resnick and Sishuwa2020). From September 2015, Zambia also experienced exchange rate and inflation shocks, which had implications for macroeconomic stability and the rate of inflation (Cheelo & Haatongo-Masenke Reference Cheelo and Haatongo-Masenke2018).
Some of the measures implemented by the PF government to respond to these crises included the removal of fuel and electricity subsidies (Cheelo & Haatongo-Masenke Reference Cheelo and Haatongo-Masenke2018). The removal of subsidies had implications for the cost of basic foodstuffs, which are mostly produced in rural areas but commonly consumed in urban zones. Electricity tariffs were increased which led to an increase in production costs of sugar and cooking oil which are also largely consumed in urban areas (Ibid). National statistics show that in 2015, 89% of rural households earned their livelihood from the agriculture sector (mostly subsistence agriculture) compared to only 18% in urban areas (Republic of Zambia 2016). Moreover, 91% of rural households owned their housing units, compared to 41% in urban areas (Ibid: XV). Crucially, only 4% of rural households were connected to the national electricity grid compared to 67% in urban areas (Ibid: 130). These dynamics meant that rural households had lower costs of energy, food and housing. Rural areas were also cushioned from the effects of the cost of living because of policies such as agricultural input subsidies that expanded massively under Lungu. Such policies do not typically benefit urban areas where economic activity revolves around informal sector trading.
Zambia is an interesting case for studying the economic basis of urban party-switching because of the PF's origin as an urban-based party. Moreover, while earlier research on African turnovers has often focused on coalition formation, elite realignments, and how such factors relate to ethnic voting (Arriola Reference Arriola2013), such explanations are not persuasive for the 2021 Zambian turnover. Party volatility has certainly been an important part of the Zambian party system historically, with parties emerging and disappearing and individual elites crossing the floor to join new parties (Rakner & Svåsand Reference Rakner and Svåsand2004; Sichone & Bwalya Reference Sichone and Bwalya2016; Mudenda Reference Mudenda2020). But by 2021, Zambia had maintained the same two main parties (PF and UPND) for three consecutive presidential elections (2015, 2016, and 2021). Moreover, the two presidential candidates have not changed since 2015. While some elites had changed parties between elections, both parties benefited from the reconfigurations. Voter volatility in 2021 seems to be much more an expression of performance voting than a reflection of a reconfigured party system.
Constituency-level change in PF-support
The quantitative sections of the paper aim to achieve two things. First, to establish that the seemingly urban/rural divide in party switching can genuinely be attributed to differences in urbanisation. Second, to investigate to what extent such urban party switching may be explained by economic performance voting. As a first gauge of how the PF coalition broke down in the 2021 election, we study constituency-level presidential results for PF in 2016 and 2021. We here use official election data from all of Zambia's 156 constituencies as presented by the Electoral Commission of Zambia. In the Appendix Table A1 we present the vote share for each province in 2016 and 2021.
For our analysis, we code all constituencies with a population density higher than 100 inhabitants/km2 as urban.Footnote 5 This threshold is the one proposed by Fraser (Reference Fraser2017). Data have been collated and aggregated to the constituency level by the Zambia Statistics Agency, based on the 2010 census. The threshold of 100 inhabitants/km2 leaves us with 32 urban and 124 rural constituencies.Footnote 6 Figure 1 below shows the relationship between the change in PF vote share between 2021 and 2016 and the constituency vote share in 2021. We see a large general tendency towards diminishing PF vote shares. The average constituency saw the PF vote share drop by a considerable 11.3% points. There is a strong negative correlation, meaning that those constituencies with very low PF vote-share in 2016 saw only smaller drops in the vote share for the ruling party. This relationship can be understood as a ‘floor effect.’ When the baseline PF vote share was minuscule, there is little room for decline. However, the figure reveals some interesting relative difference between urban and rural constituencies. Many of the urban constituencies belong to the group of constituencies where PF had their most significant losses. Moreover, looking at the regression line summarising the predicted value of PF vote share loss based on previous PF vote share, 79% of all urban constituencies saw a larger PF loss than predicted in the entire sample.
Figure 1. PF vote share in 2016 and change in PF vote share between 2021 and 2016.
Note: Line shows expected change in the change in PF Vote share given PF vote share in 2016.
One possible objection is that the urban-rural differences may be due to ethnic calculus. Urban Zambia is much more ethnically diverse than rural Zambia, and the PF electorate in the cities, especially in Lusaka, is less dominated by the PF core ethnic Bemba-speaking group than rural strongholds in areas such as Northern, Luapula and Muchinga. To investigate whether the urban/rural divide can be explained by ethnicity, we represent the results from two regression analyses in Table 1. In these models, the dependent variable is change in PF vote share. Standard errors are clustered by province. Model 1 introduces a dummy variable for urban and controls for PF vote-share in 2016. In Model 2, we introduce several ethno-linguistic controls. We control for the share of Bemba, Nyanja, Lozi and Tonga speakers in the constituency. We also control for the constituency ethno-lingual fractionalisation.Footnote 7
Table 1. OLS regression of constituency-level change in PF vote share.
*** p < 0.001, ** p < 0.01, * p < 0.05.
Note: Entries are OLS regression coefficients with standard errors in parentheses. Standard errors clustered by province.
Model 1 shows that in control for previous vote share, urban constituencies saw a significantly larger drop in PF vote share than rural constituencies. Model 2 adds additional confidence to the argument that the urban effect is not mainly due to differences in ethnic composition. Even in control for ethnicity, urban constituencies experienced a significantly larger drop in PF support (significant at the 99.9% level). The predicted drop in PF vote share is almost 6% points larger in urban than in rural constituencies, holding all other variables constant.
New panel data on Zambian voting
This paper introduces a new panel data set of Zambian voters from 2019–2021. Panel data allows us to make better inferences on individual-level vote choice and control for other individual-level covariates that may explain urban/rural variations in voting behaviour. The baseline dataset for this study was conducted in 2019 by the University of Gothenburg's Governance and Local Development Institute (GLD) and included a sample of 9,864 respondents drawn predominantly from two regions: the urban areas in and around Lusaka and the predominately rural area along Zambia's eastern border with Malawi, Mozambique and Tanzania (Eastern and Muchinga Provinces) (Lust et al. Reference Lust, Kao, Landry, Harris, Dulani, Metheney, Nickel, Carlitz, Gakii Gatua, Jöst, Mechkova, Mujenja, Tengatenga, Grimes, Ahsan Jansson, Alfonso, Nyasente, Ben Brahim, Jordan, Bauhr, Boräng, Ferree, Hartmann and Lueders2020). Figure A1 shows the sample map for the original baseline survey. The second dataset is the Zambia Election Panel Survey (ZEPS) (Lust et al. Reference Lust, Beardsworth, Krönke, Seekings and Wahman2021), a three-round election-panel survey conducted in 2021 using the same sample frame as the 2019 survey. While the original survey was conducted face-to-face, the election panel used telephone surveys due to COVID-19. Ubuntu Research conducted interviews in Lusaka in Nyanja and English.
The first two rounds were conducted prior to the August 12, 2021 election, with the third wave following the election. The sample sizes for the election panel surveys are 1,691 in Round 1, 1,536 in Round 2, and 1,299 in Round 3. Recruitment for the 2021 election panel was done by calling phone numbers provided by respondents in the 2019 baseline survey.Footnote 8 When the original 2019 respondents were not available, individuals contacted at the same phone number were interviewed in place of the initial 2019 respondents. We are able to link 949 respondents from the baseline survey to one or more of the election panel rounds (roughly 10% of the initial 2019 sample). While the surveys are not designed to be nationally representative, they provide a particularly useful sample for studying the decline of PF. The sample is divided between rural PF strongholds (Eastern and Muchinga) and Lusaka, one of PF's two major urban strongholds.Footnote 9
Categorising respondents as urban or rural is not entirely straightforward. The ZEPS data are not geo-coded, but the survey asked respondents to name their constituency. As a first rule, we coded all respondents who lived in a constituency with more than 100 inhabitants/km2 as urban (using the above-mentioned threshold). However, one complicating factor is the four constituencies on the outskirts of Lusaka: Chilanga, Chongwe, Kafue and Katuba.Footnote 10 These are all geographically large constituencies containing urban, semi-urban and rural areas, but the majority of our respondents are sampled from the parts of the four constituencies closer to Lusaka (i.e. the more urban parts). Ideally, we would have used geo-coordinates to identify whether these respondents lived in the urban or rural parts of the constituencies. In lieu of this information, we used self-reported information about whether the respondents categorised their area as urban or rural. We categorised respondents in these constituencies as urban unless they identified their own location as rural.Footnote 11 If respondents did not know the name of their constituency, we coded respondents as urban if they named Lusaka as their province and did not describe their area as rural.Footnote 12 Another possible difficulty is that the most urban constituency in Eastern Province, Chipata Central, falls short of the 100 inhabitants/km2 threshold. This means that in the main analysis, respondents from Chipata are not considered urban. However, in the appendix, we include robustness tests where we record respondents from Chipata Central who did not classify their area as rural as urban respondents.
Microlevel analysis using survey data
The aggregate constituency-level results make clear that PF lost more support in urban than rural constituencies around the country. However, micro-level analysis of survey responses allows for a more careful examination of urban-rural disparities in preferences and party switching. There are, however, three limitations. First, the timeframe for analysis is more limited since the baseline survey was conducted in 2019. If some of the movement away from the PF occurred prior to the 2019 survey, our ability to capture between-election trends will be limited. Second, as with all pre-election surveys, particularly in less than democratic settings, a considerable share of respondents are unwilling or unable to reveal their vote intention (Tannenberg Reference Tannenberg2022). More than 30% of respondents chose not to provide an answer to the vote choice question, reporting instead that they, would not vote, did not know or simply refused to provide an answer (see Table A5 in the appendix). Finally, we observe an implausibly large shift in reported support for the UPND after the 2021 election, which we attribute to the well-documented tendency to over-report support for electoral winners after the fact in African elections (Adida et al. Reference Adida, Ferree, Posner and Robinson2016).Footnote 13 For this reason, we privilege the pre-election surveys and exclude the post-election survey from our analysis when measuring vote-choice.
Despite these limitations, results from the survey data are consistent with the constituency-level analysis using election returns. In all models, we control for important confounders including poverty, age, female and ethnic identity (including the most common ethnic groups in the survey sample).Footnote 14 Table 2 shows the results from a logistic regression of party switching. Model 1 maintains the full sample of respondents who stated a vote choice in 2019. The dependent variable is whether the respondent planned to vote for PF in 2019 but changed to another party in the 2021 election panel (Round 2). Model 2 is the more crucial model. In this model, we limit the sample to respondents who were planned PF voters in 2019 and model the probability that these respondents declare the intention to vote for another party in the 2021 Election Panel (Round 2).
Table 2. Logistic regression micro-level party switching from PF.
*** p < 0.001, ** p < 0.01, * p < 0.05.
Note: Entries are logistic regression coefficients with standard errors in parentheses.
The results from Model 1 show that urban respondents were more likely to belong to the group of voters switching away from PF, but the relationship is not statistically significant. However, this model does not consider that some voters will have been opposition voters already in 2019 and were thus unable to switch away from PF. More importantly, in Model 2 we find that among stated PF voters in 2019, urban respondents were significantly more likely to state a different vote choice in 2021. The relationship is significant at the 95%-level. To add some substantive interpretation to these results, the models suggest that the predicted probability of changing from PF to another party was 14% points higher for urban than rural voters. This is a large and substantial difference in predicted probability. As a robustness test, in Table A6 (Model 2) of the appendix, we re-run the analysis, re-coding respondents in Chipata Central as urban.Footnote 15 In this analysis, the urban coefficient falls just short of the 95% confidence threshold (p = 0.053). However, the coefficient remains large and substantial, with the predicted probability of changing from PF to another party being 12% points higher for urban than rural voters.
Economic evaluations as an explanation for urban party switching
As noted earlier, Zambia's 2021 election took place in the context of an acute economic crisis. While all Zambians likely felt the effects of the economic downturn, urban voters may have been especially hard hit. To examine performance evaluations, we draw on a series of questions that probed ratings of President Lungu's performance since 2016 in three areas: managing the economy, reducing corruption, and improving roads and other infrastructure. In each of these areas, we ask whether the respondent believes that the president performed well or badly.
To examine urban/rural differences, we estimate a series of logistic regressions. Results in Table 3 show that evaluations of Lungu's performance on the economy and corruption were substantially lower among urban respondents.Footnote 16 Urban respondents were 9% points less likely to approve of President Lungu's economic performance and 11% points less likely to rate his record on corruption favourably.Footnote 17 The finding on corruption is interesting and suggests that urban voters may have reacted not only to PF's poor economic performance but also to its poor record on democracy and governance. Indeed, earlier work has suggested that both of these factors were important for reducing voters' trust in the governing party (Siachiwena Reference Siachiwena2021a). On roads and infrastructure, however, evaluations were higher among urban respondents. Urban voters were 12% points more likely to rate Lungu favourably on infrastructure than rural respondents.Footnote 18 These findings suggest that PF's strategy to point to infrastructure improvements was not successful with the urban electorate. While the urban voters do give the ruling party credit for infrastructure development, positive evaluations of infrastructure did not translate into votes. Overall economic performance is cited as the overwhelming priority among both urban and rural respondents, consistent with aggregate-level findings on previous Zambian elections (Hern Reference Hern, Banda, Kaaba, Hinfelaar and Ndulo2020; Seekings Reference Seekings, Banda, Kaaba, Hinfelaar and Ndulo2020) and Eastern Province voters in 2021 (Seekings Reference Seekings2022). Our study goes further and shows that urbanites are somewhat more likely to emphasise broad economic priorities with 59.3% of the respondents in the first round of the election panel survey citing economic management, unemployment, poverty, or wages as their top concern, relative to 44.6% of rural respondents. By contrast, only 5% identified infrastructure and roads as the most pressing problem, and 2% corruption.
Table 3. Logistic regression of President Lungu economic, corruption and infrastructure performance.
*** p < 0.001, ** p < 0.01, * p < 0.05.
Note: Entries are logistic regression coefficients with standard errors in parentheses.
In Table 4, we evaluate the connection between performance assessments and electoral preferences by running a series of bivariate logistic models of PF support using data from the final pre-election survey conducted before the 2021 election. While there are obvious limitations related to endogeneity and possible omitted variable bias, these models serve as a plausibility probe for theories that relate vote choice to economic performance evaluations. Results in Table 4 (Models 1–3) show that each of the performance measures is associated with voting intentions in the expected way: a better evaluation of Lungu is associated with a higher propensity to state an intention to vote for the PF. This is true for evaluations in relation to the economy, corruption and infrastructure. The strongest association is between economic performance and intended vote choice for PF.
Table 4. Logistic regression of intention to vote for PF.
*** p < 0.001, ** p < 0.01, * p < 0.05.
Note: Entries are logistic regression coefficients with standard errors in parentheses.
Lastly, one possibility is that the drop in PF vote choice in urban areas is due not only to a generally weaker evaluation of PF performance in cities, but also a higher propensity to punish the government party at the polls when performance evaluation is poor. That is, urban voters may be more capable and willing to use their vote to ‘throw the bums out’ when they are performing poorly. To investigate this possibility, we also ran interactive logistic models where we introduced multiplicative interaction effects between the three forms of performance evaluation (economic, corruption and infrastructure) and urban residency. The post-estimations of these models are presented in Table 5. Full regression models are presented in Table A5 of the appendix.
Table 5. Urban and rural predicted probability to vote for PF given performance evaluation.
Note: Postestimations with covariates held at means. Full regression results available in Table A8 of the appendix. Entries are predicted probabilities. 95% confidence interval in parentheses.
These models show some interesting findings. Most importantly, we find that urban voters with a poor evaluation of the economy are less likely to state an intention to vote for the PF than rural voters with a weak evaluation of the PF's economic performance. However, the difference between the urban and rural voters' likelihood of voting when they rate the economy as poor falls just short of the 95% significance threshold. Among those who state that they are unhappy with PF's economic performance, the predicted probability of voting for PF is 15% for urban voters and 22% for rural voters. Similarly, urban voters are more likely to punish the government for poor corruption performance. Lastly, we find the opposite to be true for infrastructure. Among those who say that PF has performed well on infrastructure, the predicted probability of voting for the PF is 30% among urban respondents compared to 49% for rural respondents (the difference is statistically significant at the 99.9%-level). In all, this analysis indicates that urban voters may have been more inclined than rural respondents to punish the PF for poor performance on the economy and corruption, yet they were also less likely to reward the PF for performance on infrastructure development.
Alternative mechanisms
We have argued that economic performance voting is the main mechanism explaining the larger drop in PF support in urban areas compared to rural areas. This is not to say that other mechanisms could not also have been in play. Therefore, we conclude our analysis by evaluating three other possible mechanisms: campaign exposure, voter autonomy and repression. It has been argued that cash-strapped opposition parties have concentrated their campaign efforts on urban areas where they can reach many (and often volatile) voters with fewer resources (Beardsworth Reference Beardsworth, Banda, Kaaba, Hinfelaar and Ndulo2020), while they have shied away from rural strongholds of the government party (Horowitz Reference Horowitz2016). It has also been argued that rural voters are less autonomous in their vote choice due to social, economic and communal pressure and coercion (Conroy-Krutz Reference Conroy-Krutz2018; Wahman & Boone Reference Wahman and Boone2018; Lust Reference Lust2022) and that urban areas have been particularly exposed to repression (Beardsworth et al. Reference Beardsworth, Fraser, Resnick and Siame2021).
To evaluate these alternative mechanisms, we again rely on the election panel. First, Table 6 investigates the mobilisation argument using data from the election panel (Round 3). To examine patterns of electoral mobilisation, we draw on questions about rally attendance and whether respondents had been contacted by each of the two leading parties in 2021. These analyses yield mixed results. First, we find that urban respondents are less likely to attend both PF and UPND rallies (Model 1 and 2). This finding suggests that while the parties may organise more rallies in urban areas, the draw of such events may be lower in the higher-paced contexts of urban Zambia. We find that urban respondents are more likely to have been contacted by the PF, but the difference is not statistically significant. We do, however, find that urban respondents were significantly more likely (significant at the 99.9% level) to have been contacted by the UPND than rural respondents. However, we remain doubtful that the higher UPND-contact frequency in urban areas is the main driver of the PF urban decline. In Table A9 of the appendix (Model 4), we model the likelihood of voting for the PF, given that you have been contacted by the UPND. We do not find that those contacted by the UPND are significantly less likely to vote for the PF.
Table 6. Correlates of partisan campaign interaction.
*** p < 0.001, ** p < 0.01, * p < 0.05.
Note: Entries are logistic regression coefficients with standard errors in parentheses.
Second, we focus on perceived social sanctions for deviating from the candidate favoured by: (1) members of respondents' ethnic groups, (2) family members and friends, and (3) others in one's village or neighbourhood. In each case, we asked whether respondents expected that deviating from group preferences would result in having others think poorly of them (disapprove) or suffering material or physical damage (harm). Results in Table 7 show that urban respondents are less fearful of social sanctions, less often expecting punishments from their ethnic communities or families for deviating from group voting norms, though we observe no differential regarding locality sanctioning. In other words, there is some support for the alternate explanation related to voter autonomy. However, in Table A7 of the appendix, we model the relationship between social sanctions and voting for the PF. None of these correlations are significant.
Table 7. Correlates of social sanctions.
*** p < 0.001, ** p < 0.01, * p < 0.05.
Note: Entries are logistic regression coefficients with standard errors in parentheses.
Lastly, we evaluate the possible influence of repression. Since 2015, the PF increasingly relied on repression to maintain its grip on power. In particular, the ruling party cultivated a culture of cadreism, where young party members were given permission to control markets and other commercial zones. In exchange, these cadres provided repressive resources to the PF and limited the opposition's ability to campaign (Wahman Reference Wahman2023). While cadres were active around the country, their presence was particularly felt in urban areas (Hinfelaar et al. Reference Hinfelaar, Resnick and Sishuwa2020).
It is difficult to evaluate the effects of repression on the urban vote, especially as an alternative explanation to economic voting. On the one hand, earlier research has shown that many Zambians were concerned about the repression and found it one further reason to vote PF out of office (Beardsworth et al. Reference Beardsworth, Fraser, Resnick and Siame2021). This conclusion aligns with new research on election violence, arguing that election violence may backfire on perpetrators (Rosenzweig Reference Rosenzweig2023). On the other hand, high repression in urban areas may be seen as endogenous to poor performance evaluation among the urban electorate. That is, the PF was forced to enhance repression as the momentum among urbanites turned against the regime due to economic decline. Moreover, violence may also have indirectly affected urban voters, which could have benefitted the PF. Violence does not only affect voters directly but also shapes electoral environments and reduces the opposition's campaign capacity (Wahman Reference Wahman2023).
Nevertheless, our survey did include a question related to cadres. Respondents in ZEPS were asked whether the ‘actions of PF cadres have changed your view of the PF.’ Table 8 shows the results from a regression analysis where the dependent variable is whether respondents respond that the actions of cadres have negatively affected their view of the PF. We find that urban respondents are significantly more likely to have had their views of PF negatively impacted due to cadres. While this finding does not challenge our findings regarding the impact of the economy, it does indicate that other dynamics potentially contributed to PF's urban electoral collapse.
Table 8. Correlates of PF views negatively affected by cadres.
*** p < 0.001, ** p < 0.01, * p < 0.05.
Note: Entries are logistic regression coefficients with standard errors in parentheses.
Conclusion
The 2021 Zambian election was an extraordinary event. On a continent where elections generally feature high levels of incumbent advantage (Bleck & van de Walle Reference Bleck and van de Walle2018), incumbent defeat of the Zambian magnitude remains rare. While some earlier incumbent defeats on the African continent may be attributed to changes in elite coalitions and defections from the ruling party (Arriola Reference Arriola2013), the Zambian election did not follow this pattern. Instead, the PF defeat was a consequence of large-scale and genuine party-switching between stable political alternatives.
This paper digs deeper into patterns of party-switching, zeroing in particularly on urban/rural differences. This paper shows that while PF lost support across the board, the party was particularly likely to lose supporters in urban areas. These patterns are established using both constituency-level and micro-level data. We particularly identify economic grievances as the main driver of urban opposition support.
While the PF came to power on promises of economic transformation for Zambia's hard-tested urban populations, they ultimately failed to live up to these high expectations. The severe economic crisis in the lead-up to the 2021 election had left urbanites struggling from inflation, high unemployment and disastrous economic projections. Such economic hardship undoubtedly affected Zambians in both urban and rural Zambia, but most specifically urbanites relying on formal employment and access to cheap agricultural produce and imported goods. Using panel data, we show that urbanites evaluated PF economic performance significantly worse than rural voters and that those rating PF economic performance negatively were less likely to support the governing party. We also find that urban voters were more likely to vote for the opposition when they considered government economic performance poor.
This paper focuses on one single election, and it is impossible to generalise across the continent to suggest that urban party-switching is always a reflection of economic performance evaluation. Nevertheless, the paper shows clearly that African elections are not entirely structured by stable cleavages but also feature significant degrees of volatility. This is good news for thinking about the prospects of real democratic accountability. The probability for party-switching may vary across space, but this may also be because perceptions of performance will vary among different constituencies.
In extension, the paper also provides insight into why many African newly elected parties shifted their electoral base in favour of a more rural-focused coalition. A coalition dependent on urban voters will be hard to maintain with the inevitability of urban economic discontent, while a more rural coalition can be more easily contained with the use of clientelism. From the perspective of understanding Zambian politics, the results also suggest that UPND, the new ruling party, may find it challenging to maintain its current coalition and hold on to votes from a volatile and disgruntled urban electorate.
Finally, an important caveat is needed. The paper has focused particularly on the political economy basis of urban party switching. This is not to say that urban voters were not also concerned with other forms of performance, including basic service delivery (Hern Reference Hern, Banda, Kaaba, Hinfelaar and Ndulo2020). The 2021 election was held in a context of democratic backsliding and enhanced repression (Hinfelaar et al. Reference Hinfelaar, Rakner, Sishuwa and Van de Walle2022a, Reference Hinfelaar, Rakner, Van de Walle, Arriola, Rakner and Van de Walle2022b). This paper hints that- apart from being worried about the economy- urban voters were also particularly concerned about corruption and the actions of lawless cadres. The urban electorate certainly used the power of the ballot box to protect the country's fragile democracy (Siachiwena Reference Siachiwena2021a; Resnick Reference Resnick2022). Further research into this topic is needed and will add substantially to our understanding of subnational variations in performance voting in new African democracies.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0022278X24000077.
Acknowledgements
The collection of the ZEPS data was enabled by a Swedish Research Council Recruitment Grant (E0003801) and funding from the University of Cape Town. Erica Metheney and Kirk Ammerman provided data collection support. The survey was implemented masterfully by Ubuntu Research and Rural Development Company. The study was determined exempt by the Michigan State University (MSU) IRB Office (STUDY00006161). We are grateful to Ellen Lust for her support. Jeremy Horowitz played an important role in the early stages of conceptualising this paper. Anne Pitcher and Jeremy Seekings provided useful feedback on earlier drafts.
Open access
