3.2.1 Perceptions and Experiences of Hardship

In the Afrobarometer survey, we measure how people perceive the IMF SAP of their country to have affected their personal economic circumstances. Comparativists refer to such assessments as pocketbook evaluations, which most respondents will find easy to answer as they can draw on their own experience (Reference Duch and StevensonDuch and Stevenson 2006).

To capture pocketbook evaluations, we draw on the following Afrobarometer survey question: “What effect do you think [the SAP] has had on the way you live your life? Has it made it worse, had no effect, or made it better?” Answer categories include: “A lot worse,” “worse,” “neither better nor worse,” “better,” “a lot better,” and separately, non-responses or missing data.

Figure 3 shows the distribution of pocketbook evaluations of IMF SAPs, available for individuals who have heard about the IMF SAP of their country. What is striking is that there is considerable heterogeneity in responses. While negative assessments of IMF SAPs dominate (59.9%), a sizable share of respondents say their economic circumstances slightly improved (19.3%); some respondents did not experience any difference (15.0%), and a small remainder improved significantly (5.9%). Considering that IMF SAPs are often portrayed as uniformly challenging in the policy discourse (Reference Reinsberg, Stubbs and KentikelenisReinsberg, Stubbs, and Kentikelenis 2022), these patterns are remarkable and merit further investigation.

Figure 3 Distribution of pocketbook evaluations of IMF SAPs

Figure A1 in the appendix plots the distribution of responses to this question separately for all countries in which it was asked. The variation in SAP evaluations is not driven by specific countries. In most countries, IMF SAP evaluations hover around the middle category. A clear exception is South Africa, where the modal experience with the IMF SAP seems to have been very negative. Zimbabwe is another case in which the modal experience has been negative. In Mali, relatively more people were satisfied with the IMF SAP. We confirm that SAP evaluations differ more within countries than across countries by decomposing variation in the outcome. This indicates a general pattern whereby IMF SAPs create both winners and losers, with the losers being in the relative majority.

Because the surveys for Asia and Latin America do not directly ask respondents about their perceptions of IMF SAPs, we need to draw on alternative evaluation questions. In the Asian Barometer, we have the unique opportunity to understand whom people blame for the possible deterioration of their economic circumstances. We draw on two related questions. The first asks respondents to evaluate the extent to which their government treats everyone equally or not: “Now I am going to read to you a list of statements that describe how people often feel about the state of affairs in [country]. Please tell me whether you strongly agree, somewhat agree, somewhat disagree, or strongly disagree with each of these statements: Everyone is treated equally by the government” (Q108). The second probes respondent evaluations of the IMF among those individuals who have heard about the organization: “Please let us know about your impression of the following organizations: the International Monetary Fund” (Q163). Answer categories are on a scale from 1 to 10, where 1 is “very bad” and 10 is “very good.”

Figure 4 shows the distribution of responses for perceptions of government biasedness in Asian Barometer countries with exposure to IMF programs. People are divided in their views on whether their government treats people equally or not: 40 percent disagree with the statement that their government treats everyone equally (10 percent disagree even strongly), and almost 60 percent think their government is unbiased, while only a small percentage cannot judge either way. In the appendix, we find similar patterns concerning perceptions of government biasedness in all Asian program countries (Figure A2).

Figure 4 “Everyone is treated equally by the government”

Figure 5 shows how people perceive the IMF. In contrast to their own governments, most people have a positive impression of the IMF: 60 percent give it a high rating of at least 7 out of 10 and fewer than 5 percent give a low rating of at most 3 out of 10. This result is somewhat surprising, given the bad reputation of the IMF, especially in Asia after the Asian Financial Crisis (Reference ItoIto 2012). In the appendix, we plot the distributions by country. We find similar patterns across program countries concerning perceptions of government bias, with a slightly higher perception that the government is biased in Mongolia (Figure A2) and impressions of the IMF, with slightly more IMF-critical views in Indonesia (Figure A3).

Figure 5 Impression of the IMF across Asian Barometer countries

Like the Asian Barometer, the Latinobarometer does not ask respondents about their perceptions of the IMF SAP. Instead, the survey includes questions on the perceived quality of public services and their satisfaction with the functioning of the market economy. These are important issues because they are directly affected by IMF conditionality: an abundant literature documents how the IMF has agreed with governments on reforms that seek to reduce public spending, privatize public service provision and liberalize markets and prices (Reference BabbBabb 2013; Reference Kentikelenis, Stubbs and KingKentikelenis, Stubbs, and King 2016; Reference Reinsberg, Kentikelenis and StubbsReinsberg et al. 2019; Reference Rickard and CarawayRickard and Caraway 2019). To the extent that governments implement at least some of these reforms, we would expect their consequences to be felt by citizens, with changes in their perceptions of public service quality as a result.

We draw on two survey questions from the Latinobarometer asking respondents for their views of how the quality of public services has changed over the past year. The survey questions read as follows: “Would you say that in [country], in the last 12 months, the quality of the [public schools/public hospitals] has gone down, gone up or stayed the same?” (P90STB/P90STA). Answers on a five-point Likert scale include “has gone up a lot,” “has gone up a little,” “has stayed the same,” “has gone down a little,” “has gone down a lot,” and separately, non-responses. For our purposes, these questions are advantageous because they allow us to compare how public service quality evolved in countries under IMF programs compared to countries not under such programs. This provides for stronger inferences compared to survey items that just capture the current level of satisfaction with public services and thus are missing an individual-level baseline. Moreover, this question wording is closest to the question in the Afrobarometer survey asking respondents to assess the IMF SAP.

The results show most people have experienced no change in how they perceive the quality of public services in their country. The remaining responses divide roughly equally between perceived improvements and perceived deteriorations in public schooling (Figure 6) and public hospitals (Figure 7). In the appendix, we further analyze current levels of satisfaction with public service provision, finding that respondents are exactly split as to whether they are satisfied or not with the quality of public schools in their country – both in the aggregate (Figure A4) and across individual countries (Figure A5). Similarly, people are divided on their perceived quality of public health services – both in the aggregate (Figure A6) and in individual countries (Figure A7).

Figure 6 Evolution of perceived quality of public education in Latin American program countries

Figure 7 Evolution of perceived quality of public hospitals in Latin American program countries

Across all regional surveys, we have found considerable variation in how individuals assess the effects of IMF SAPs concerning their own fortunes and the quality of public-service provision, and whether or not they conceive governments as biased implementers of IMF-mandated reforms.

In the World Values Survey, we can capture experiences of hardship through two measures. The first is an index of deprivation that draws on three items asking whether the respondent (and their family) have respectively gone without enough food to eat, gone without medicine or treatment, and gone without cash income in the past twelve months (H008). The frequency of each marker of deprivation can be: never (0), rarely (1), sometimes (2), or often (3). We create a deprivation index by adding up the categorical response values over all three types of deprivation. The resulting index can range from zero (never any deprivation in any dimension) to nine (often deprived for all three essential items). The second outcome is the income group of a respondent. Within each country, respondents can be in one of three income groups relative to the living standard in that country: low income, medium income, and high income (X047CS).

Figure 8 shows the prevalence of different forms of deprivation across all respondents of the World Values Survey. While over 40 percent of respondents never experienced any material hardship, almost 60 percent did experience some hardships during the year before the survey. Where they occurred, however, hardship experiences were relatively mild (for example, an index value of “one” means hardship was rare and only in one dimension). There are somewhat fewer respondents who sometimes or often experience hardship across more than one dimension. The most extreme form of deprivation – being deprived “often” of all three types of essential goods – is extremely rare (index value of “nine”).

Figure 8 Prevalence of deprivation among respondents in the World Values Survey

Figure 9 shows a range of income inequality across the respondents of all survey countries. Most people have a medium income. About one-third of the respondents have low incomes. About one-sixth of the respondents have high incomes.

Figure 9 Income distribution of respondents in the World Values Survey

3.2.2 Protest

Our second outcome of interest is protest. Comparativists have debated how to measure protest in surveys, specifically whether to only measure protest behavior or also include protest inclination. On the one hand, only considering actual protest provides a conservative measurement strategy because it requires a behavioral response. On the other hand, capturing protest inclination could be advantageous because it likely is a first step toward actual protest. At the same time, protest inclination measures individual preferences not imbued by considerations about collective mobilization and potential repression of protest by the government that may discourage people from actually protesting (Reference Ritter and ConradRitter and Conrad 2016). We consider both measures substantively interesting. We proceed with actual protest because it is available in all surveys. For some surveys, we also use the inclusive measure that combines actual protest and protest inclination. Our results are not sensitive to this choice.

In the Afrobarometer, the corresponding survey question on protest is: “Here are a number of different actions people might take if government were to do something they thought was wrong or harmful. For each of these, please tell me whether you have engaged in this activity or not: Attend demonstration or protest march?” (pardem). This question has five answer categories, including “Yes, often,” “Yes, a few times,” “Yes, once,” “No, but would do if I had the chance,” and “No, would never do.” Most people never protest, and only a few posit to have engaged in protest. A sizable share of about 10 percent would protest if given the chance (Figure 10). In the appendix, we plot the distribution of responses separately for all countries in the survey (Figure A8). We obtain qualitatively similar patterns, suggesting that the protest results are unlikely to be driven by individual countries.

Figure 10 Protest in sub-Saharan African program countries

In the Asian Barometer, the protest question is somewhat simpler: “Did you attend a demonstration or protest march in the past twelve months?” (Q88). The four possible answer categories include “Yes, more than once,” “Yes, once,” “No, never,” and non-responses. The drawback is that we cannot capture protest inclinations in the Asian Barometer. Instead, the survey includes a question on violent protest that can be used for robustness tests. Figure 11 shows the distribution of responses, indicating that protest is very uncommon in Asia: over 90 percent of the respondents have never protested. At the same time, protest (or lack thereof) is not unique to any Asian country in particular – the appendix shows similar patterns across all Asian program countries (Figure A9).

Figure 11 Protest in Asian program countries

In the Latinobarometer, respondents are asked to indicate participation in a whole battery of contentious behaviors. The introductory text reads: “I am going to read out a political activity. I would like you to tell me, if you have ever done it, if you would ever do it, or if you would never do it?”. This is followed by a list of activities including signing a petition; taking part in an authorized demonstration; participating in riots; occupying land, buildings, or factories; blocking traffic; taking part in unauthorized demonstrations; and contacting officials, parliamentarians, or civil society actors. We do not use all these activities but focus on the items that are closest to protest. Hence, we include responses on authorized demonstrations, unauthorized demonstrations, and riots. Figure 12 shows the distribution of responses based on lawful demonstrations only – the type of activity with the most affirmative responses. In line with the other surveys, most people do not engage in protest, while some would consider doing so, and only a few have actually done so. In the appendix, we verify that these patterns look similar across Latin American program countries (Figure A10).

Figure 12 Protest in Latin American program countries

We have learned that a majority of individuals across all barometers have not engaged in protest. Comparing across regions, protest is more common in Latin America than in Asia, and slightly more common than in Africa. While only a minority has engaged in protest, significant proportions of respondents have a latent potential to do so.

In the World Values Survey, we measure whether respondents have engaged in a lawful protest in the past twelve months (E221B). Figure 13 plots the distribution of responses across all respondents, showing that about 24 percent have protested while 76 percent have not. Protest thus appears to be more prevalent in the global sample than in the regional barometers. It remains to be seen whether grievances induced by IMF-mandated reforms translate into greater inclinations for protest and protest behavior.

Figure 13 Protest among World Values Survey respondents

3.2.3 Partisan Allegiances

To test our argument, we require individual-level data on the partisan allegiance of respondents. The survey datasets differ concerning the extent to which they offer a straightforward measure of partisan allegiance, although we are always able to (re)construct these partisan allegiances using additional information (see Tables A4 and A5). The Afrobarometer includes a readily usable variable indicating whether a respondent supported the government, the opposition, or no party at all in the past election. Figure 14 shows that besides 48 percent who are non-partisans, a plurality of 33 percent supported the governing party in the previous election, while 19 percent supported opposition parties.

Figure 14 Partisan allegiances in sub-Saharan African countries

The Asian Barometer also has a pre-constructed variable capturing whether individuals supported the governing party in the last election (Q39a). Figure 15 shows the distribution of individuals across different partisan camps, indicating that half of the respondents in program countries are opposition supporters, 38 percent are government supporters, and 12 percent are non-partisans.

Figure 15 Partisan allegiances in Asian program countries

The Latinobarometer is more challenging because it does not include a readily usable measure of partisan alignment but only a forward-looking party choice variable for upcoming elections. Therefore, we construct a measure of partisan allegiance by matching the left-right ideological self-placement of a respondent to the partisan orientation of the incumbent. We describe this matching procedure in more detail in the appendix (Table A4). We also replicate this approach for the World Values Survey because it does not include a readily usable measure of partisan alignment either (Table A5). In the Latinobarometer, we can partly enhance the validity of our arguably crude proxy of partisan alignment by requiring individuals to have voted in the past election (p49stu). Figure 16 presents the distribution of partisan allegiances in Latin American program countries, showing that about 28 percent of respondents have not voted in the past election, 38 percent share the partisan orientation of the government, and about 34 percent do not share the partisan orientation of the government.

Figure 16 Partisan allegiances in Latin American program countries

3.2.4 Individual-Level Control Variables

In multivariate analyses, we draw on several individual-level control variables, directly available from the surveys. The reason for including these controls is that they could confound the relationship between partisan allegiances and relevant outcomes such as IMF SAP evaluations. Our demographic controls include (logged) age, dummies for whether the respondent is male, lives in an urban area, has a college education, and is unemployed (Reference Almond and VerbaAlmond and Verba 1963; Reference Kaufman and ZuckermannKaufman and Zuckermann 1998; Reference PilatiPilati 2011; Reference Robertson and TeitelbaumRobertson and Teitelbaum 2011). We also measure political interest, capturing respondents who said they often listen to the news on the radio and who consider themselves as “always interested” in politics. As an objective indicator of political knowledge, we measure whether the respondent recalled the correct name of the finance minister. As to attitudes, beliefs, and values, we use a binary indicator to capture whether the respondent is satisfied with democracy as it is currently practiced in the country. We also measure support for free markets, the public sector, and the privatization of state-owned enterprises. Controlling for these economic values is important to the extent that such values condition evaluations of IMF-sponsored measures. Finally, our most demanding controls gauge respondent evaluations of the state of the economy (Reference Kaufman and ZuckermannKaufman and Zuckermann 1998). Specifically, we capture whether respondents think the economy is doing worse than twelve months ago and whether they expect it to be deteriorating in the next twelve months. These variables are included to ensure that our results are not driven by individuals who are generally more pessimistic about the economy. For similar reasons, we include a binary variable indicating whether respondents are dissatisfied with the performance of the incumbent. This variable has two important functions. First, it helps us isolate evaluations of the IMF SAP as opposed to evaluations of more general government policies. Second, it proxies the extent to which respondents “like” a particular government – a potential confounder for our analysis because such partisan predispositions could shade their evaluations of IMF SAPs and their vote choice in the last election.

To ensure our results are comparable across surveys, we seek to construct sets of control variables that are as similar as possible across the barometers. Perfect congruence is not possible because some questions may not be available or are phrased differently. Nonetheless, we hold that we employ sufficiently similar sets of controls across all surveys to collectively mitigate for the same alternative explanations to our argument. The appendix shows definitions and descriptive statistics for all variables in the Afrobarometer (Table A2), the Asian Barometer (Table A3), the Latinobarometer (Table A4), and the World Values Survey (Table A5).

4.1.1 Adjustment Policies of the Rawlings Administration

For two decades, Ghana was ruled by Jerry Rawlings, first as military chief (1981–93) and then as the elected leader of the National Democratic Congress (NDC) (1993–2001). Rawlings took power on December 31, 1981, in the fifth coup that had taken place in thirty-five years, and began a near consecutive series of agreements with the IMF in 1983 (Reference AkonorAkonor 2013). Facing re-election for the first time in 1996, Rawlings asserted himself through distributive politics. He targeted rural areas with capital spending on roads, rather than cutting capital spending as agreed upon with the IMF, with a total expenditure 20 percent higher than the government had originally budgeted (Reference AkonorAkonor 2013, 99–101). Reference NinsinNinsin (1998, 226–27) astutely observes how Rawlings used the state to reward would-be supporters with the distribution of limited national resources:

Both ethnicity and anticipated material benefits prompted voters to support the winning party (Reference Abdulai and HickeyAbdulai and Hickey 2016; Reference YoudeYoude 2005). The president used his executive authority to appoint teachers across the country for political advantage. Citizens rewarded these appointments in presidential elections with larger shares of votes for the incumbent (Reference HardingHarding 2015). Drawing on international aid, Rawlings successfully used electrification to both reward supporters and bring new voters into the fold as NDC voting increased in constituencies that received electrification (Reference BriggsBriggs 2012, 603).

4.1.2 Ghanaian Adjustment Politics under the Kufuor Administration

On January 7, 2001, Rawlings’s nominee, vice-president, John Atta Mills, lost the NDC’s bid to retain power losing to the opposition leader John Agyekum Kufuor from the New Patriotic Party (NPP). The NPP was strongly tied to the indigenous entrepreneurial elite, a group that suffered under Rawlings (GhanaWeb 2001c; Reference HandleyHandley 2007). The new Kufuor administration continued its engagement with the IMF. Facing dire economic circumstances, Kufuor instructed his administration to adopt the HIPC initiative to unlock debt relief and free up fiscal space. The initiative reduced Ghana’s debt overhang to facilitate investment into the domestic economy. The new administration enacted many adjustment reforms to promote the business sectors of the economy, which favored Kufuor’s entrepreneurial base, while at the same time ignoring the demands of civil society organizations (Reference Crawford and AbdulaiCrawford and Abdulai 2009, 102–3). This is illustrated by a union leader who complained about the spending cuts proposed by the Kufuor administration:

There is also evidence that the Kufuor administration directed infrastructure projects to regions of the country that supported his election. As part of the new lending by the IMF and the World Bank, Kufuor’s administration received a total of $462 million to support Ghana’s developmental projects (GhanaWeb 2001b). Part of this package contained “$220 million … for road projects, notably Accra-Kumasi and Accra-Cape Coast dual carriage roads” (GhanaWeb 2001b). Kumasi is the capital of the Ashanti region, and the region which provided Kufuor with the highest percentage of support, totaling 80.5 percent of the vote in the presidential run-off increasing from 65.8 percent of the vote in 1996 (PeaceFMOnline 1996a). While Cape Coast was also one of the largest supporters for Kufuor, it also represented an important swing region moving from 43.0 percent support for Kufuor in the 1996 election to 60.2 percent of the vote in the presidential run-off in 2000 (PeaceFMOnline 1996b). Kufuor appears to reward both his support base in the Ashanti region as well as thanking the newly supportive region of the Cape Coast with substantial infrastructure investment. The behavior of the Kufuor administration ties in with research on Ghana where voters can attribute the provision of a particular good (roads) to a particular politician, they will reward those politicians with their votes (Reference HardingHarding 2015).

Our illustrative case of Ghana provides us with some evidence that governments may utilize the flexibility given to them in complying with IMF mandates to advantage their partisans and place the burdens of adjustment disproportionately on supporters of the opposition. We found that such patterns of distributional politics are particularly salient where governments have direct control over policy implementation, notably for public-sector employment, public infrastructure, and the divestiture process. To probe the plausibility of these findings beyond the case of Ghana, we now turn to our large-N analysis using survey data.

4.2.1 Research Design

We use a sample of 21,531 respondents in twelve sub-Saharan African countries from the first wave of Afrobarometer (1999–2001). Our effective sample is smaller due to the omission of countries in which relevant questions were not asked, or individuals for whom the question is not relevant because they are unaware of the IMF SAP. Our main dependent variable draws on the Afrobarometer survey question asking people about the pocketbook effects of the IMF SAP of their country, measured on a five-point Likert scale. For our analysis, we create the dichotomous variable SAP made my life worse by combining the two response categories for deteriorating life circumstances.

Our main independent variable is based on whether the individual is a supporter of the governing party, a supporter of an opposition party, or not a supporter of any party at all. It is available from the survey, constructed by comparing the incumbent party and the party identification variable in the survey dataset. As our argument builds on differences in partisan allegiances, we drop the neutral category in our main analysis and construct the binary indicator opposition supporter. This allows us to directly compare pocketbook evaluations between opposition supporters and government supporters. In robustness tests, we use a trichotomous variable comparing government supporters (supporter of the winning party), opposition supporters (supporter of the losing party), and the neutral category (those who do not support any party).

Independent variables at the country level include a dummy for whether the country has been under an IMF program in the year before the survey, as well as the total number of IMF conditions in that year. We draw both pieces of information from the IMF Monitor Database. Using these variables allows us to examine whether distributive politics intensifies in the presence of IMF SAPs, especially when such programs leave governments with greater potential for distributive politics by including more adjustment burdens. We therefore expect partisan-driven differences in IMF SAP evaluations to be bigger where the country has an IMF program, and the government faces a greater number of binding policy conditions.

For control variables, we focus on individual-level controls because country-fixed effects already account for the confounding impact of country-level variables on our outcomes. While our first set of controls only includes these fixed effects, the second set includes individual-level demographic characteristics and respondent values, as described earlier. The third set of controls, which were also described in the previous section adds variables capturing respondent evaluations of the government more generally.

A final consideration pertains to the choice of respondent sample. Here we need to consider that people who have not heard about the IMF SAP do not record any perceptions about it. The sample therefore only includes respondents who have heard about the SAP in their country. We may be concerned that people who heard about the SAP are different from people who have not heard about it and that this difference is related to partisanship and potentially other unobserved respondent characteristics. As this selection process could induce bias into our estimates, we also estimate a two-stage model in which we jointly estimate the determinants of respondent awareness about the IMF SAP and program evaluations for those being aware. The two-stage model is necessary only if unobserved factors are driving both IMF SAP awareness and IMF SAP evaluations. Reassuringly, the main results are not affected by either modeling alternative.

4.2.2 Illustrative Evidence

We now present illustrative evidence for our argument from the Afrobarometer sample. Figure 17 provides suggestive support for the argument that negative evaluations of IMF SAPs are more prevalent among opposition supporters. While 45.1 percent (95% CI: 43.1–47.7%) of government supporters consider that the IMF SAP made their life worse, the corresponding figure is 63.5 percent (95% CI: 60.8–66.2%) for opposition supporters. A t-test finds the difference of 18.4 percentage points to be statistically significant (p<0.001).

Figure 17 Partisan allegiances and IMF SAP pocketbook evaluations

Figure 18 probes the notion that the scope for distributive politics – and therefore the partisan-based differences in IMF SAP evaluations – should be larger where countries are under an IMF program. Indeed, we find a similar prevalence of negative pocketbook evaluations of IMF SAPs between government supporters and opposition supporters in countries that were not under an IMF program. A t-test confirms that the 2.4 percentage-point difference in outcomes is not statistically significant. In contrast, where countries are under an IMF program, partisan-based differences are large: For about 42.3 percent (95% CI: 40.1–44.9%) of government supporters, IMF SAPs have made their lives worse, compared to about 64.5 percent (95% CI: 61.6–67.5%) for opposition supporters. The difference of 22.2 percentage points is statistically significant (p<0.001).

Figure 18 Partisan allegiances, IMF program exposure, and IMF SAP pocketbook evaluations

Finally, we exploit differences in IMF program design across countries. Figure 19 shows that there is no discernible partisan difference in pocketbook evaluations for IMF SAPs with different numbers of conditions. We will scrutinize this finding again after controlling for observable confounders.

Figure 19 Partisan allegiances, IMF program design, and IMF SAP pocketbook evaluations

4.2.3 Regression Results

Table 1 shows the relationship between partisan allegiances and whether respondents think that the IMF SAP made their life worse. Our key quantity of interest is the coefficient for opposition supporters, which indicates the percentage-point difference in the prevalence of a negative pocketbook evaluation of the IMF SAP compared to government supporters. Different columns show models with three different sets of control variables.

Our results provide evidence for our first hypothesis that partisan supporters of the opposition view the consequences of IMF program lending more negatively than partisans who support the government. We find that opposition supporters have a significantly higher likelihood of a negative pocketbook evaluation of the IMF SAP, compared to government supporters (p<0.05). For example, in the last model, opposition supporters have a likelihood of 64.3 percent (95% CI: 50.3–78.3%) of a negative pocketbook evaluation compared to 52.2 percent (95% CI: 32.5–71.9%) for government supporters.

Coefficient estimates of control variables are consistent with theoretical expectations. Among the respondents who have heard about the IMF SAP, older people tend to be unhappier with SAPs. A similar finding emerges for unemployed people. Urban residents are less likely to have a negative IMF SAP evaluation. We also find evidence that political interest and radio consumption are negatively related to pocketbook evaluations, while politically knowledgeable people are generally more critical of IMF SAPs. There is some evidence that people who prefer capitalism and support privatization view the pocketbook effects of IMF SAPs more negatively. In the models with extended control variables, we find that individuals with generally more pessimistic evaluations of the economy and notably the performance of the incumbent are more likely to express negative evaluations of IMF SAPs. This may look surprising at first but makes sense when considering the role of unfulfilled expectations for these respondents. None of the remaining controls reaches statistical significance. The fact that our core result is consistent across the three specifications is reassuring, especially when considering that evaluations of the state of the economy and the level of dissatisfaction with incumbent performance absorb substantive variation in the outcome of interest.

Table 2 presents the results from split-sample models in which we allow for the partisan-induced difference in IMF SAP evaluations to vary by whether or not countries are under an IMF SAP. We find that in countries with IMF programs, 45.0 percent (95% CI: 15.3–74.7%) of government supporters and 64.8 percent (95% CI: 40.1–89.6%) of opposition supporters have negative pocketbook evaluations of IMF SAPs. As the confidence intervals of effects for each partisan group do not include the point prediction of the other group, the difference is statistically significant. In countries without IMF program exposure, this difference becomes insignificant – here 60.9 percent of government supporters and 64.7 of the opposition supporters have negative pocketbook evaluations of IMF SAPs. This suggests that IMF program participation intensifies distributive politics and thus increases the observable partisan difference in IMF SAP pocketbook evaluations.

4.2.4 Threats to Inference, Robustness Tests, and Further Analyses

One objection to our results so far is that governments may not be able to identify different partisans, which would limit the extent of distributional politics. We argue that the scope of distributive politics may indeed be limited to sectors in which the government has direct control over the allocation of resources, such as employment contracts in the public sector. Hence, individuals who are in public-sector employment and support the opposition view the consequences of IMF program lending more negatively than those who are not. The results corroborate our expectations. While different partisans differ in their IMF SAP evaluations, effect magnitudes are consistently larger among public-sector workers compared to workers in other sectors (p<0.01) (Table A6).

Another objection to our results is that different IMF SAP evaluations are driven purely by perceptions. Different partisans may be inclined to interpret their experience with the IMF SAP differently, regardless of actual differences in their treatment by the government. To allay such concerns, we exploit an index of objective hardship, constructed from a battery of Afrobarometer questions. These questions capture the frequency with which respondents have gone without food, health care, potable water, and income. For example, the food deprivation question reads: “In the last twelve months, how often have you or your family: gone without enough food to eat?” Answer categories for all questions include “never,” “rarely,” “sometimes,” “often,” and non-responses. We create an index deprivation by adding responses from all four items (where “never” is 0, “rarely” is 1, “sometimes” is 2, and “often” is 3). The empirical range is from zero to twelve. Because the survey question is not specifically about deprivation induced by the IMF SAP, we need to leverage additional macro-level variation in IMF program participation and IMF program design to identify the effect of distributive politics in the context of the IMF SAP.

Our results provide strong support for the notion that partisan-based differences in IMF SAP pocketbook evaluations are not simply driven by perceptions but objective hardships. Compared to countries without IMF program experience, being an opposition supporter is significantly related to deprivation in countries under an IMF program (Table A7). Comparing across programs with different designs, we obtain less significant patterns, although the direction of the effect is indicative: In a high-conditionality scenario, being an opposition supporter tends to be related to more deprivation than being a government supporter, and compared to a low-conditionality scenario (Table A8).

Table 3 shows the results of split-sample models in which we allow for the partisan-induced difference in IMF SAP evaluations to vary by the number of IMF conditions. Here we only consider countries with IMF programs and split the sample at the median number of conditions. Consistent with our argument that more conditions imply a greater potential for governments to allocate burdens along partisan lines, we find a significant relationship between partisan allegiances and IMF SAP evaluations only in the high-conditionality case, but not in the low-conditionality case. In the former case, we find at least a 10.6 percentage-point difference in IMF SAP evaluations (p<0.05) – with 30.5 percent for government supporters and 41.1 percent for opposition supporters. In the latter case, we find a 10.4 percentage-point difference that is not statistically significant.

We have so far only included respondents who have heard about the IMF program. This could induce selection bias into our results to the extent that unobserved respondent characteristics that correlate with partisanship drive both their interest in the IMF SAP and their related pocketbook evaluation. To mitigate this bias, we perform a two-stage selection model in which we first predict whether respondents heard about the IMF program, before examining the determinants of a negative pocketbook evaluation among aware individuals. We predict awareness about the IMF SAP with two binary variables that are unlikely to directly affect IMF SAP evaluations. The first is political interest – whether respondents consider themselves interested in politics. The second is objective knowledge about politics, specifically whether a respondent correctly recalled the name of the finance minister. Both variables are available from the survey. We find significant partisan differences in IMF SAP evaluations, even considering that incumbent supporters tend to be more aware of IMF SAPs (Table A9). When comparing across IMF SAPs, we also find that as a government faces more conditions, the partisan difference concerning negative IMF SAP evaluations widens (Table A10).

Some have argued that politics in an African context is organized along both ethnic and partisan lines (Reference Bratton, Bhavnani and ChenBratton, Bhavnani, and Chen 2012). From a methodological point of view, a related critique is that political allegiances may change quickly and thus become endogenous to distributional outcomes. Probing whether our results also hold for ethnicity, therefore, holds promise to strengthen our inferences. We would expect that individuals belonging to an ethnically discriminated group will have less favorable evaluations of the IMF SAP than members of a powerful group. We rely on group status information from the Ethnic Power Relations dataset (Reference VogtVogt et al. 2015). The remaining challenge is to identify to which ethnic group a respondent belongs, which is not directly available from the Afrobarometer. From the Afrobarometer, we identify the main language that respondents speak. In some cases, it is possible to infer the ethnic group from the language spoken. Our focus is to distinguish groups that control the government from those that are powerless – if not entirely discriminated against. We disregard groups that have junior-partner status in the government. Our relevant comparisons are thus between powerful groups and powerless groups. Where such ascriptions are not possible, for example, because ethnicity is not politically relevant, we exclude respondents of such countries. Our remaining sample has 6,136 individuals from five countries, of which 5,760 individuals are members of the powerful group and 376 individuals are from powerless groups. The sample further reduces when requiring that individuals must have heard about the SAP in their country.

Replicating our main results with markers of ethnicity, we find a significantly higher likelihood of negative pocketbook evaluations of IMF SAPs among members of powerless ethnic groups, compared to members of powerful groups (Table A11). Specifically, being a member of an ethnically powerless group increases the likelihood of a negative pocketbook evaluation by about 14.7 percent (95% CI: 8.9–20.5%), relative to being a member of an ethnically powerful group. In further analysis, we find a wider ethnicity gap in pocketbook evaluations between IMF program countries and non-program countries (Table A12). The effect of being a member of an ethnic minority on a negative pocketbook evaluation is amplified by an additional ten percentage points (95% CI: 7.9–11.1%) if a country has an active IMF program.Footnote ⁶

Finally, we probe the robustness of our findings to alternative econometric modeling choices. We re-run our models using survey-adjusted probit regressions. Our results are qualitatively similar, showing a statistically significant difference in IMF SAP evaluations across partisan groups (Table A13). Partisan differences are significantly larger when the country is under an IMF program (Table A14) and when the government faces more IMF conditions (Table A15). In addition, our results hold for random-intercept multi-level models. While we are careful not to over-interpret the findings given the small number of country clusters (Reference MuellerStegmueller 2013), we corroborate the statistically significant difference in IMF SAP evaluations across partisan groups (Table A16), which grows even larger when conditioning on IMF program participation (Table A17) and the number of IMF conditions (Table A18).

4.3.1 Illustrative Evidence

Figure 20 shows the perceived biasedness of the government for different partisan groups and across Asian program countries with different conditionality. Several observations stand out. First, we find that more people think their government is biased in countries with high-conditionality programs compared to those with low-conditionality programs. Second, in the high-conditionality scenario, opposition supporters are significantly more likely than government supporters to assess the government as biased. A t-test confirms the difference of 14.1 percentage points as statistically significant (p<0.001). This is not the case in the low-conditionality scenario, where the difference is just 1.2 percentage points and not significant (p=0.63). Both pieces of evidence are in line with our argument that more conditions provide governments with additional burdens to allocate, notably upon opposition supporters.

Figure 20 Partisan allegiance, IMF conditionality, and government bias

Figure 21 replicates this analysis concerning the prevalence of a bad impression of the IMF as the outcome variable. Interestingly, we obtain qualitatively different response patterns. In the high-conditionality scenario, government supporters are more likely to dislike the IMF than opposition supporters, while the reverse holds in the low-conditionality scenario. A plausible interpretation of these findings is that people are capable of correctly attributing responsibility for their fortunes. Opposition supporters do not seem to blame the IMF when governments implement high-discretion programs with many conditions, but may attribute blame to the government.

Figure 21 Partisan allegiance, IMF conditionality, and bad impression of the IMF

4.3.2 Regression Results

Table 4 examines the relationship between partisan allegiance, IMF program exposure, and (the perception of) government bias under various sets of control variables. We find a statistically significant positive interaction between partisan allegiance and IMF program participation that is robust across model specifications. In substantive terms, 34.4 percent (95% CI: 31.7–37.2%) of opposition supporters and 30.7 percent (95% CI: 29.2–32.3%) of government supporters conceive the government as biased when the government does not participate in an IMF program – a partisan gap of 3.7 percent (95% CI: 0.5–6.9%). When the government is under an IMF program, this partisan gap increases to 7.4 percent (95% CI: 0.2–14.6%). This is consistent with an intensification of partisan-based distributive politics under IMF SAPs.

Table 5 examines the relationship between partisan allegiance, IMF conditionality, and (the perception of) government bias under various sets of control variables. We find a statistically significant interaction effect between partisan allegiance and IMF conditionality that is robust across model specifications. Because it is difficult to directly read off marginal effects, we predict the outcome for different constellations of the predictors. At the mean of IMF conditionality, the probability that opposition supporters conceive the government as biased is 53.2 percent (95% CI: 49.4–56.9%) and respectively 47.6 percent (95% CI: 43.3–51.9%) for government supporters. This partisan gap widens under a high-conditionality scenario compared to a low-conditionality scenario. These patterns corroborate our argument that partisan allegiance drives distributive politics when governments have lots of opportunities to allocate adjustment burdens of IMF programs. We cannot think of an alternative explanation for these patterns.

In the appendix, we seek to establish the robustness of our findings, considering potential selection at the individual level. Specifically, we include a selection model to predict whether respondents have heard about the IMF. Our excluded instruments – political interest and political knowledge – are strongly significant in the selection model. We find our estimates of interest to be qualitatively unaffected, although less statistically significant in the first two columns (Table A19). Substantively, at the empirical maximum of IMF conditionality, about 74.2 percent (95% CI: 68.6–79.9%) of opposition supporters believe the government is biased, but only 61.8 percent (95% CI: 57.2–66.5%) of government supporters do so.

In addition, we provide another piece of evidence supporting our interpretation. Specifically, we can show that different patterns emerge when people are asked about their impression of the IMF. This outcome is only available for people who have heard about the IMF. We find that partisan differences in the prevalence of bad impressions about the IMF are no less pronounced in countries with IMF exposure compared to countries without IMF exposure (Table A20). A higher number of IMF conditions in an IMF program significantly reduces the partisan gap in negative evaluations of the IMF, compared to a program with fewer conditions (Table A21). These results suggest that people can identify the government as the main actor responsible for the biased implementation of IMF programs.

4.4.1 Illustrative Results

Across our outcomes of interest, we find no significant partisan differences comparing cases under IMF programs and cases without IMF programs. Neither do we find a noticeable change in partisan differences when we compare low-conditionality programs to high-conditionality programs (Figure 22). This could be due to several reasons. It may be that our proxy of partisan allegiance is too rough. It may also be that IMF programs do not drive partisan differences in these outcomes, in part because these outcomes are too remote from IMF policy demands. We now probe these patterns further using multivariate analysis.

Figure 22 Partisan allegiance, IMF conditionality, and perceived deterioration of public schools

4.4.2 Regression Results

Table 6 shows the relationship between partisan allegiances, IMF program exposure, and our two evaluative outcomes of interest. Two findings stand out. First, the extent of perceived deterioration of public services is generally higher where countries have been exposed to IMF programs. This can be seen by inspecting the coefficient estimates for IMF exposure and those for the interaction term. Second, opposition supporters tend to be no more dissatisfied with public services than government supporters. This result holds both in the absence of IMF program exposure, as well as in the presence of program exposure. Before we draw definitive conclusions about this issue, we turn to our subsample analysis of countries with IMF program exposure.

Table 7 shows the relationship between partisan allegiances, average IMF conditionality, and the two evaluative outcomes in Latin American program countries. We find that opposition supporters tend to be marginally more likely to have worse evaluations of public services in these IMF program countries. This partisan difference appears to be unaffected by the stringency of IMF conditionality. In sum, this set of results provides the weakest support for our argument among all our analyses. We will revisit how to interpret these results in the conclusion, given our other findings.

4.5.1 Illustrative Evidence

Figure 23 shows the extent of deprivation after a financial crisis as experienced by respondents with different partisan allegiances. The left panel includes countries that attempted crisis resolution without IMF programs, while the right panel includes countries with IMF programs. The results are striking. Even under highly turbulent times of financial crisis, there is no apparent partisan difference in deprivation in countries without IMF involvement. In contrast, there is a significant partisan difference in deprivation in crisis-affected countries that involved the IMF. While government supporters also appear to suffer from increased hardship under IMF programs, the burdens for opposition supporters are inordinately higher. The partisan difference in deprivation is 0.528 (95% CI: 0.336–0.722) – a tangible difference given the ten-point range of the index points that is also statistically significant in a t-test (p<0.001). These patterns suggest that governments under IMF-mandated programs lump additional adjustment burdens onto opposition supporters. The patterns also suggest that the intensification of partisan politics is not the inevitable consequence of financial turmoil but requires IMF-mandated policies that facilitate such partisan politics.

Figure 23 Partisan allegiance and deprivation under financial crises.

While we focus on deprivation as our main outcome of interest, the patterns look similar for income as an alternative outcome (Figure A11). As confirmed by t-tests, opposition supporters have lower income than government supporters if their country faced a financial crisis and the country underwent an IMF program but not if the crisis-affected country avoided the IMF program.

4.5.2 Regression Analysis

Table 9 shows the relationship between partisan allegiances and deprivation in the context of financial crises, distinguishing cases of IMF programs and cases without IMF programs. In line with the illustrative evidence, we find a significantly positive interaction effect between the incidence of financial crisis and being an opposition supporter, but only in the subsample of countries with IMF program participation. Where countries facing financial turmoil attempted to resolve their crises without IMF programs, we do not find a significant partisan gap concerning experienced deprivation.

Model predictions underscore the substantive significance of these differences. Focusing on countries that experienced financial turmoil in the three years before the survey, the partisan gap in deprivation is estimated to be at least 0.266 (95% CI: 0.104–0.428) on a ten-point scale. In crisis countries without IMF involvement, this partisan gap is indistinguishable from zero. Our findings are based on up to forty-two countries and nine survey years, which eases concerns that they could be driven by outlying cases. Moreover, the results are robust to different sets of individual-level control variables. The estimates of the control variables are consistent with expectations that deprivation is more prevalent among women (especially during IMF programs), larger households, un(der)employed people, and less educated people.

In the appendix, we demonstrate the robustness of these findings to alternative measures and modeling choices. First, we show similar results when using the income group as the outcome variable (Table A22). We find that crisis resolution with an IMF program is associated with significantly lower income among opposition supporters relative to opposition supporters. However, the partisan gap in incomes does not exist for crisis countries that did not turn to the IMF. The results are based on a high number of countries across several survey waves. Second, we probe whether results hold for a narrower timing of macro-events. Specifically, we capture whether a country underwent a financial crisis two years before the survey, which is the year before respondents are asked to assess whether they suffered any hardships. We use the same operationalization to capture whether a country had an IMF program. We continue to find a strongly bifurcated pattern whereby the partisan gap in experienced deprivation is large for crisis countries that participated in an IMF program but indistinguishable from zero when the country did not undergo an IMF program (Table A23). For the results on income, the findings are less clear-cut, although they tend to reproduce a larger partisan gap in the crisis scenario with IMF participation (Table A24).

5.1.1 Moi Administration (1997–2002)

Daniel arap Moi, leader of the Kenya African National Union (KANU), took office after the death of Jomo Kenyatta in 1978. Moi had a relatively narrow support base, relying on the Kalenjin ethnic group and minority groups from the Rift Valley. Moi’s Kalenjin group benefited from plum jobs in the civil service, army, and SOEs (Reference WrongWrong 2010, 51).

Moi agreed to two Extended Credit Facility loans with the IMF – respectively for 1996–1999 and 2000–2003, $216 million each. A key IMF demand was to retrench a total of 32,348 public-sector workers by 2002, with 25,783 civil servants slated for dismissal by October 2000 (Reference KamauKamau 2000). Other areas of the package included parastatal reform, price liberalization, and divestiture of public enterprises (IMF 1996).

Moi had ultimate responsibility for the retrenchment of civil servants – a task that he delegated to his permanent secretaries (Reference WarigiWarigi 2000). These appointees discriminated in the hiring of civil servants (The Nation 2014); there is no reason to believe they would behave any differently in their dismissal of employees. Under the threat of a national strike and facing the upcoming re-election in 1997, his government acceded to the demands of about 260,000 teachers for a 150–200 percent pay rise in October 1997 (The Nation 2013). To further this sense of unfairness, Moi excluded the security services and teachers from the IMF-mandated retrenchment in 2000.

The contradictory set of criteria for targeting staff to retrench included age, gender, health, the record of past service, and academic and professional qualifications (Reference OtienoOtieno 2009, 66). In practice, these criteria particularly favored Kikuyu and Kalenjin ethnic groups, who tended to be more educated and have better health outcomes compared to minority ethnic groups, and men who were better educated than women (Reference OtienoOtieno 2009, 66). At the same time, the government sought to increase the salaries of civil servants who remained in government employment, effectively concentrating the benefits of employment even during the midst of adjustment lending (Daily Reference NationNation 2000).

Kenyans challenged the politically driven implementation of retrenchment. Societal groups like the Retrenchees Welfare Association claimed that the process was “biased, unfair, political and riddled with irregularities.” They objected to “the rush to beat the deadline and the use of poorly selected committees dominated by members who gave room to prejudice, witch-hunting and ethnicity … and claimed that certain political affiliations and communities were being favored in the exercise” (Reference NationSunday Nation 2000). These critiques chime with reporting elsewhere about the promotion of under-qualified individuals who are co-ethnics of senior managers (Reference WrongWrong 2010). Even junior ministers in the governing party, like assistant minister Charfano Guyo Mokku, said that “the civil-service retrenchment program was discriminatory, flawed and unfair to certain ethnic communities” (Reference NationSunday Nation 2000). National Labour Party chairman Kennedy Kiliku said the retrenchment process was illegal, arguing that “opposition sympathizers were being victimized through retrenchment … and that the so-called downsizing was actually ‘down-sizing’ the opposition sympathizers” (Reference KithiKithi 2000).

Kenyans mobilized to protest civil-service cuts and fiscal austerity (Reference Ellis-JonesEllis-Jones 2002, 18). In May 2001, state-employed air traffic controllers in Mombasa airport went on strike demanding better terms of employment and salary increases. In September 2001, Mombasa council workers dumped piles of rubbish on the street all over the city and inside city hall to protest the non-payment of three months salaries (Reference Ellis-JonesEllis-Jones 2002, 18). The strike, which lasted for over a week, ended up in running battles with the police. Mombasa was an opposition stronghold. Moi’s KANU party received only about a third of the votes across each parliamentary constituency, holding only one of the four Mombasa districts in the 1997 election (Reference Rakodi, Gatabaki-Kamau and DevasRakodi, Gatabaki-Kamau, and Devas 2000, 160). Moi’s administration also withheld the salaries of local government workers in Kakamega, which led to a series of strikes in 2001. Kakamega is in Kenya’s Lurambi constituency and was won by opposition politician Newton Kulundu in 1997.

5.1.2 Kibaki Administration (2003–2007)

In late 2002, Moi’s nominee Uhuru Kenyatta lost KANU’s bid to retain power to opposition leader Mwai Kibaki’s National Rainbow Coalition (NARC). While NARC primarily reflected Kibaki’s own Kikuyu ethnic group it also included a dozen other ethno-regional parties and represented a broader electoral coalition than the previous Moi administration (Reference BarkanBarkan 2004). Kibaki’s approach to distributional politics in the context of IMF lending manifested itself in broad-based sectoral and infrastructural initiatives rewarding areas and regions of support with government spending facilitated by the IMF. Kibaki dismissed the heads of parastatal, permanent secretaries, and other high-ranking members of the civil service and those on the judicial bench who had been installed by ex-president Moi.

Kibaki concluded a three-year Extended Credit Facility of over $252 million with the IMF, in exchange for IMF demands to combat corruption in the judiciary, passage of a strengthened anti-corruption bill, reform of parastatals, and civil-service retrenchment (Reference OpiyoOpiyo 2003a). Notably, Kibaki’s administration was tasked with dismissing another 24,000 workers during the period of its loan (Reference Githahu and OmangaGithahu and Omanga 2003).

In his early statements as president, Kibaki demonstrated his commitment to representing all forty-two ethnic communities in his new government (Reference OchiengOchieng 2002). He resisted calls by the IMF to cut civil-service spending during his first year in office (Reference OpiyoOpiyo 2003b). In his first ten months, his administration already spent more on wages than the allotted amount for the year in the IMF Medium Term Expenditure Framework (Reference OpiyoOpiyo 2003a). The Kibaki administration also sought to enlarge portions of the civil service as part of its approach to distributional politics (The Nation 2009). It continued with the commitment of the previous Moi administration promising increases in teacher salaries, although the IMF warned that this would undermine progress toward the envisaged cuts in the public-sector wage bill (Reference KamauKamau 2003).

While Kibaki’s administration was more reluctant to engage in targeted cuts given his broad coalition of support, protests did take place. For example, on March 3, 2005, a group of civil society organizations used the occurrence of a mini-ministerial World Trade Organization conference taking place at a luxury Kenyan South Coast resort to demand greater African participation at the conference in general and protest the role of international financial institutions in Kenya in particular. About 500 protesters took part in a peaceful protest, 40 of whom were arrested by the police (CIS 2005).

Kibaki’s administration was very sensitive to demands from international institutions which had clear distributional impacts on his support base that could encourage protest, despite potentially benefiting the country overall. For example, the World Bank pressured the Kibaki administration to liberalize the maize market in Kenya to solve its long-standing issues of maize shortages. While unrestricted imports of maize may have helped to solve issues of food security within the country, the government was reluctant to enact these changes because maize farmers in the North Rift would be hardest hit (Reference OpiyoOpiyo 2003a). Kibaki more than doubled his vote share in the Rift Valley from 20.9 percent to 43 percent in the December 2002 election (Reference ThroupThroup 2003). Maintaining the benefits of the status quo to his supporters was one issue that would be at stake in liberalizing the market.

In summary, we find evidence for distributional politics in the context of structural adjustment. Despite facing similar IMF pressures, Moi’s narrower coalition base meant that he could target more opposition supporters for cuts in comparison to Kibaki’s administration, whose broad coalition made him more reluctant to also make cuts. Facing similar IMF demands, both incumbents used their discretion and implemented IMF-sponsored reforms to protect their partisan supporters, leading to grievances and protests mobilized by opposition supporters.

5.2.1 Illustrative Evidence

We proceed with simple comparisons and bivariate t-tests. Figure 24 shows the average prevalence of protest in sub-Saharan program countries depending on whether or not respondents consider that the IMF SAP has made their life worse. We find a remarkable difference in protest behavior across both groups. While 18.6 percent (95% CI: 17.2–20.0%) of those who think that the IMF SAP made their life worse protest, it is only 15.4 percent (95% CI: 14.0–16.8%) of those who think otherwise. The percentage-point difference of 3.2 percent is statistically significant according to a t-test (p<0.01).

Figure 24 Pocketbook evaluations of IMF SAPs and protest

Of course, it may again be the case that, for reasons that we cannot capture, respondents may both hold negative views about the IMF SAP and be more likely to protest. This would make the relationship of interest spurious. Our main remedy to this challenge is to exploit variations in the number of conditions in IMF SAPs. Specifically, programs with more conditions imply a higher burden of adjustment that governments can lump onto opposition supporters. The increased opportunity for partisan politics in this case should affect partisan differences in outcomes. Figure 25 confirms our expectation. We find a larger difference in protest frequencies depending on how people evaluate the pocketbook effect of the IMF SAP in the high-conditionality scenario, compared to the same difference in the low-conditionality scenario. In the former, the difference is 12.1 percent (p<0.0001), but only 2.9 percent in the latter (p=0.51).Footnote ¹²

Figure 25 Program design, pocketbook evaluations of IMF SAPs, and protest

An alternative approach is to relate protest to partisanship and exploit variation in IMF exposure and program design. Figure 26 shows the results. We find that among countries not under an IMF program, the frequency of protest is roughly the same across different partisan groups – with only a difference of 1.7 percent (p=0.60). In contrast, in countries with an IMF SAP, opposition supporters have a significantly higher likelihood of protest than government supporters. A t-test confirms that the percentage-point difference of 10.7 percent is significant (p<0.001). These patterns provide strong suggestive evidence for the protest-inducing effect of distributive politics facilitated by government exposure to IMF programs.

Figure 26 Partisan allegiances, IMF program exposure, and protest

Figure 27 shows the relationship between partisan allegiances, IMF conditionality, and protest. Exploiting program design assumes that the scope for distributive politics increases as governments face more extensive burdens of adjustment. Our illustrative evidence using countries under IMF programs is consistent with this expectation. We find that the partisan-induced gap in average protest is higher in the high-conditionality scenario, compared to a low-conditionality scenario. In the former case, the percentage-point difference in protest across partisan groups is 12.1 percent (p<0.001), while it is only 2.9 percent (p=0.51) and statistically insignificant in the latter case. Furthermore, protest frequencies are relatively higher in the high-conditionality cases.

Figure 27 Partisan allegiances, IMF conditionality, and protest

5.2.2 Regression Analysis

We now scrutinize the above relationships using multivariate analysis. Table 10 shows that people who feel adversely affected by IMF SAPs are more likely to protest: their likelihood of protest is 14.9 percent (95% CI: 7.5–22.3%), compared to 11.4 percent (95% CI: 3.4–19.4%) for those who do not think that the IMF SAP made their life worse. These results hold across different model specifications (p<0.01).

In the next step, we identify differential treatment by the government based on partisan allegiances during the implementation of IMF SAPs as a source of grievance-induced protest. Since the question on protest is not formulated concerning IMF SAPs, we exploit variation in exposure to IMF programs. Table 11 shows the results from split-sample regressions by program exposure. We find that partisan allegiance tends to be positively related to protest when the country had IMF exposure, although the coefficient is statistically significant only in one model. In contrast, we do not find any effect of being an opposition supporter on protest when the country had not been under an IMF SAP. In other words, the differences across partisan groups concerning the propensity for protest are larger for program countries than for countries not under an IMF program. Despite lacking statistical significance, the findings are consistent with our argument that IMF program participation can increase the prevalence of protest due to intensified distributive politics.

A final step is to examine partisan-induced differences in protest exploiting variation in IMF conditionality. Our relevant sample is all countries with an IMF SAP. Table 12 shows the results from split-sample regressions at the median of IMF conditionality for program countries. We find that partisan allegiance is significantly related to protest when the government has faced an above-median number of IMF conditions, but not for a below-median number of IMF conditions, which is exactly what the distributive politics argument would lead us to expect. Substantively, the difference across partisan groups is 4.9 percentage points (p<0.01) in the former scenario. In the latter scenario, it is larger but not precisely estimated.

5.2.3 Threats to Inference, Robustness Tests, and Further Analyses

We now probe the robustness of our findings and address potential threats to inference. One issue – that perceptive biases rather than objective hardships affect IMF SAP evaluations – is less problematic in the study of protest. This is because even perceptions can have real consequences concerning protest. However, it would be interesting to know whether partisan-related protest in the context of IMF SAPs is primarily due to unfair treatment or the perception thereof. Therefore, we examine how deprivation – under different scenarios of IMF conditionality – affects protest. In the appendix, we confirm that as conditionality increases, more deprivation increases the likelihood of protest (Table A25).Footnote ¹³

As our analyses so far only included respondents who are aware of the IMF SAP, findings could be liable to selection bias. This would be the case if unobserved respondent characteristics were to drive both their awareness of the IMF SAP and their decision to protest. We therefore model IMF SAP awareness of respondents through an additional selection equation using all respondents. To predict IMF SAP awareness, we measure whether respondents are politically interested and politically knowledgeable. Both variables strongly correlate with IMF SAP awareness but should not directly affect protest. Using this estimation approach, the second-stage results concerning protest are qualitatively similar (Table A26).

In addition, we probe the applicability of our theoretical mechanism to ethnic politics. To that end, we compare protest patterns by respondents from ethnically powerless groups to respondents from powerful groups in countries with different exposure to IMF SAPs. The results indicate a protest-inducing effect of being a member of an ethnically powerless group when the country was under an IMF program, but not otherwise (Table A27). We probe whether and how ethnicity and partisanship co-produce protest under IMF SAP exposure. We find that individuals from ethnically powerless groups are more likely to protest when they also support the political opposition but less likely when they support the government (Table A28).

Furthermore, we utilize an alternative definition of protest that also includes protest inclinations. This is substantively interesting given that individuals may face obstacles to realizing protest preferences in many countries, for example, due to government repression. Our results are virtually unaffected: When individuals believe the IMF SAP made them worse off, they are more likely to consider protest. The percentage-point difference between partisans is at least 4.6 percent (p<0.05) (Table A29).

Finally, we re-run our analyses with alternative econometric models. One alternative is to pool all observations and estimate probit models (Table A30). A more advanced alternative, which honors the nested nature of the data, is a multi-level random-intercept model, which assumes baseline country averages to be distributed normally around the mean (Table A31). In both cases, we find a strongly significant positive relationship between negative pocketbook evaluations and increased protest.

5.3.1 Illustrative Evidence

We first conduct a subsample analysis, comparing the prevalence of protest among respondents who perceive their government as biased and those respondents who do not. As would be expected, the prevalence of protest is higher – twice as high – for the former respondents relative to the latter respondents (Figure A12), and highly statistically significant (p<0.001). Furthermore, we find that protest is more likely among individuals who have a bad impression of the IMF compared to those with a good impression. The difference is statistically significant but less pronounced than for government biasedness (Figure A13)

Figure 28 helps us to link protest to the underlying markers of distributive politics – the partisan allegiance of respondents – while also considering the context of IMF SAPs. It shows that when a country is not under an IMF program, there is a significant difference in the probability of protest between opposition supporters and government supporters (of about 2.8 percentage points). However, when a country undergoes an IMF SAP, the absolute difference in protest probabilities across different partisans increases (to about 3.6 percentage points). These patterns are consistent with our argument about the intensification of partisan conflict under IMF SAPs.

Figure 28 Partisan allegiance, IMF program exposure, and protest in Asia

Figure 29 probes these patterns further by examining protest across different partisans and for different burdens of adjustment in IMF SAPs. We find that the partisan-based difference in protest is substantively sizable and strongly statistically significant in the high-conditionality scenario. About 12.0 percent (95% CI: 8.3–15.8%) of opposition supporters and 4.6 percent (95% CI: 2.9–6.4%) of government supporters protest under a high-conditionality scenario. In contrast, the partisan difference is smaller in the low-conditionality scenario, with 9.9 percent of opposition supporters (95% CI: 7.4–12.4%) and 6.7 percent of government supporters (95% CI: 2.9–10.6%) participating in protest respectively.

Figure 29 Partisan allegiance, IMF conditionality, and protest in Asian program countries

5.3.2 Regression Results

Table 13 shows the relationship between government biasedness, IMF program exposure, and protest in Asia. Two findings stand out. First, perceptions of government biasedness are related to significantly higher probabilities for protest. Second, where a government is under IMF program exposure, the effect of perceiving the government as biased on protest is higher compared to when a government is not under IMF exposure. In substantive terms, the average baseline probability of protest among individuals who perceive the government is biased is 3.8 percent (95% CI: 3.0–4.6%), which increases to 6.5 percent (95% CI: 4.0–8.9%) under an IMF program. These findings seem to suggest that IMF program exposure – which we know increases government biasedness on its own – amplifies protest proclivities among those who believe their government is biased.

Table 14 analyzes the relationship between government biasedness, IMF conditionality, and protest in Asian program countries. We find that as the number of IMF conditions increases, the protest-inducing effect of government biasedness increases. At the mean of IMF conditionality, the likelihood of protest is 5.5 percent (95% CI: 4.1–6.9%) for those who consider the government biased and 4.1 percent (95% CI: 3.2–4.9%) for those who do not. In the appendix, we plot the marginal effect of being an opposition supporter across the range of IMF conditions. Under a low-conditionality scenario, partisanship has no relationship with protest. Under a high-conditionality scenario, being an opposition supporter is related to an increased protest probability of two percentage points (p<0.05) (Figure A14).

Table 15 shows the relationship between partisan allegiances, IMF program participation, and protest. Opposition supporters are significantly more likely to protest when their country is under an IMF program. In contrast, the link between being an opposition supporter and protesting is not robust when a country is not under an IMF program. Substantively, this difference in opposition-driven protest amounts to about 5.4 percentage points (p<0.05). This finding supports the notion that IMF SAPs can amplify pre-existing partisan differences in protest inclination.

Table 16 relates protest to partisanship and IMF conditionality. Our conclusion is similar: as conditionality increases, the protest-inducing effect of being an opposition supporter increases. The effect is most precisely estimated without controls but remains marginally significant even when adding a host of controls including attitudes, values, and beliefs. In the appendix, we plot the marginal effect of being an opposition supporter on protest for different levels of IMF conditionality (Figure A15). Albeit less precisely estimated, the partisan-based difference in protest is almost seven percentage points (p<0.1) across both program design scenarios.

5.3.3 Robustness Tests and Further Analyses

In the appendix, we probe whether our results hold when considering selection bias at the individual level. To that end, we include a selection equation modeling whether individuals have heard about the IMF SAP. We again use political interest and political knowledge to predict respondent awareness of the IMF SAP. We find that the partisan difference concerning protest is higher in countries that undergo an IMF program (Table A32). Among IMF SAP countries, we find that opposition supporters are more likely to protest especially when programs entail more adjustment burdens (Table A33). Taken together, these results are in line with our argument about the implications of partisan-based politics in the implementation of IMF SAPs.

5.4.1 Illustrative Evidence

Using simple bivariate analysis, we examine whether protest is more likely when people are more dissatisfied with the evolution of public services that are plausibly affected by IMF SAPs. Figure 30 does not yield a clear-cut picture. While protest is significantly higher among those who believe that the quality of public schools had worsened, protest is unaffected by different perceptions of how the quality of public hospitals had evolved. The figures are based only on countries that had IMF program exposure in the three years before the survey.

Figure 30 Evaluations of public services and protest in Latin American program countries

We also examine how partisan differences in protest evolve if a country undergoes an IMF program, especially if that program has many conditions. Figure 31 shows that among countries that had IMF exposure, the partisan difference in protest is significantly higher for high-conditionality programs, compared to low-conditionality programs (p<0.01).

Figure 31 Partisan allegiances, IMF conditionality, and protest

5.4.2 Regession Results

We now turn to multivariate regression analysis for the above relationships. Table 17 examines the relationship between partisan allegiances, IMF program participation, and protest. We find that people who were aligned with the opposition are no more likely to protest than government supporters. This result holds regardless of whether a country was under an IMF program or not.

Table 18 shows the relationship between partisan allegiances, IMF conditionality, and protest in Latin American program countries. We find a significantly positive interaction effect between being an opposition supporter and the number of binding conditions in the program in two models. We consider the barebones model as less credible as it does not adjust for demographic differences. In sum, the results provide some evidence to suggest that governments lump the adjustment costs of more demanding IMF programs on opposition supporters, whose likelihood of protest increases as a result.

5.5.1 Illustrative Evidence

Figure 32 shows the partisan difference in protest for countries facing financial turmoil. Where countries attempted crisis resolution without the IMF, the partisan gap is 5.8 (95% CI: 2.3–9.4) percentage points, a relatively small difference. In contrast, with IMF involvement, the partisan gap is much higher, namely 20.5 (95% CI: 15.5–25.5) percentage points. We also observe lower absolute levels of protest for both partisans in the cases with IMF programs compared to crisis cases without IMF programs. However, it must be borne in mind that these results are correlational and should not be interpreted as evidence of a protest-dampening causal effect of IMF programs.Footnote ¹⁴

Figure 32 Partisan allegiances, financial crisis, and protest

5.5.2 Regression Analysis

Table 19 shows the relationship between partisan allegiances and protest in the context of financial crises, distinguishing cases of IMF programs and cases without IMF programs. We find a significantly positive interaction effect between the incidence of a financial crisis and being an opposition supporter on the likelihood of protest in the subsample of countries with IMF program participation. The partisan difference in protest is at least 9.6 (95% CI: 1.0–18.2) percentage points. This is a sizable effect, considering that only about 24.2 percent of the population actively engaged in protest in the entire sample. In contrast, where countries facing financial turmoil attempted to resolve their crises without IMF programs, we do not find a significant partisan gap concerning protest.

The results are stable across different sets of controls, and the estimates of the control variables are consistent with theoretical expectations. For example, we find that protest is more prevalent among males, the unmarried, fully employed respondents (who have the most to lose from cuts), the educated, and the politically interested. We also find that respondents who prioritize economic growth and who have confidence in their government are less likely to protest. Our estimates are based on a moderate number of countries (N≤44) and survey waves (N≤9), and the overall model fit is sizable.

In the robustness checks, we probe an alternative operationalization of the key predictors using twice-lagged macro-predictors for financial crises and IMF programs. We find some evidence for a higher partisan gap with respect to protest when a crisis country underwent an IMF program. The partisan difference is only marginally significant in one model (Table A34). These patterns suggest that the timing of financial turmoil and IMF intervention is crucial when considering protest outcomes, as there seems to be a time lag between implementation of reforms, mobilization of protesters, and actual protest.