Are the unhappy unemployed to blame for unrest? Scrutinising participation in the Arab Spring uprisings

1 Introduction

In December 2010 in a Tunisian provincial town, the fruit vendor Mohamed Bouazizi set himself on fire to protest against the actions of corrupted police. This event sparked unrests which also erupted in the capital Tunis, and soon unseated president Ben Ali. Revolts inspired by the peaceful revolution in Tunisia rolled through most of the Arab world in 2011. These uprisings that came to be known as the Arab Spring toppled several rulers who had governed for decades, but they also sparked brutal civil wars. Although mostly what is considered as the Arab Spring occurred in 2011, its impact is clearly evident still today. The war in Syria has continued for more than 8 years and caused reportedly over 500,000 fatalities. There are ongoing civil wars also in Yemen and Libya, and there is much turbulence and insecurity elsewhere in the region.

Unemployment is often nominated as a driver behind these uprisings, and the Arab Spring was not the first time that unemployment has been blamed for riots in the region (Ottaway & Hamzawy, 2011, pp. 4–6; Sadiki, 2000). Also outside the Arab world, unemployment is commonly seen as a factor which considerably increases the likelihood of various forms of political instability from protesting and rioting to terrorism and civil wars. As it is assumed that unemployment causes political instability, this is then followed by assuming that creating employment is a way to foster peace (Berman, Callen, Felter, & Shapiro, 2011, p. 497). However, there is hardly any empirical evidence to confirm either the link between unemployment and political instability or the link between job creation and increased stability (Cramer, 2011; Holmes, McCord, Hagen-Zanker, Bergh, & Zanker, 2013). Nor has, in the case of the Arab Spring, the role of unemployment been studied thoroughly. This study aims at filling this gap. The article will primarily focus on the specific case of the Arab Spring, but it will also analyse the relationship between unemployment and protest in the Arab countries more generally.

To understand its potential role behind the rebellions in the Arab world, a few notions about unemployment in the region are worth making. Figure 1 shows that unemployment in the Arab world is substantial compared to the world average. Notably, however, the Arab countries as a whole did not see the increase in unemployment after the 2008 financial crisis. Actually, in 2010, on the eve of the Arab Spring, the average of the unemployment rates of the Arab countries was at the lowest level since 1991.

Figure 1:

Unemployment rates in the Arab world, the European Union, and globally 1991–2018.

In the Arab world, three groups especially suffered from high unemployment rates: the youth, the educated and women. A higher unemployment rate among the youth is not uncommon. As Figure 1 shows, the young were unemployed more often also elsewhere. In many Arab countries, unemployment rates among the highly educated were higher compared to those with lower education levels. This, in turn, was a unique situation in the world (Chaaban, 2010, pp. 18–20; Goldstone, 2011, p. 12; Mirkin, 2013, p. 24; World Bank, 2019).

Figure 2 shows the unemployment rates in the Arab countries prior to the eruption of the Arab Spring. The unemployment rates presented in Figure 1 and Figure 2 are taken from the data collected by the International Labour Organization. This data is mainly based on labour force surveys. There are certainly some inaccuracies in it but they do not affect the overall picture of Figure 2; the Arab countries were very heterogeneous in terms of unemployment when the Arab Spring started. Where many Arab countries faced significant problems in providing employment opportunities for their citizens, there has been a lot of labour migration to oil exporting Gulf countries (Mirkin, 2013, p. 17).

Figure 2:

Unemployment rates in the Arab League member countries in 2010.

The Arab Spring has been seen as an outcome of various factors; high unemployment, a large share of the youth population, autocratic and bad governance, human right violations, increased food prices, an increase in educational attainment, social media, a state dominated economic system, and oil. Unemployment is maybe the most often mentioned of these factors (see for example Allansson, Baumann, Taub, Themnér, & Wallensteen, 2012; Campante & Chor, 2012; Gelvin, 2012; Goldstone, 2011; Hoffman & Jamal, 2012; Malik & Awadallah, 2013; Mirkin, 2013; Ottaway & Hamzawy, 2011).

Some have seen the relationship between unemployment and the Arab Spring clearly as causal: the International Labour Organization (2011) titled its analysis straightforwardly “Youth unemployment in the Arab world is a major cause for rebellion.” Some have been more cautious in their wording. Malik and Awadallah (2013, p. 296) posed that “Arab revolutions were fueled by poverty, unemployment and lack of economic opportunity.” The linkage has also been promoted in mass media, for example in an article by The Guardian (2011) “The frustrated generation at the heart of the protests tell how their progress is being stifled by unemployment and corruption.”

The relationship between unemployment and the Arab Spring is seldom analysed more closely; still, some have done slightly more than just mentioned it. The results of Campante and Chor (2012) indicate that people are more willing to protest in countries with high unemployment and a rapid increase in years of schooling, and most Arab countries appeared in this group of countries. Byun and Hollander (2015) find that unemployment did not have a significant impact on the level of unrest a country faced during the Arab Spring.

Some studies have, based on different survey data sets, analysed drivers of individual level protest participation. Doherty and Schraeder (2015) study Tunisia and find that there were no statistically significant differences between the unemployed and the employed as to how often they took part in protests before the ousting of Ben Ali and in protests after elections. Shafiq and Vignoles (2015) note that “perhaps surprisingly” their descriptive data do not show that the unemployed would have been more likely to protest.

Utilising the first wave of the Arab Barometer survey, Hoffman and Jamal (2012, pp. 181–184) report that – “[c]ontrary to what might be expected” – being unemployed lowers the likelihood for an individual to protest. Beissinger, Jamal, & Mazur (2012, p. 13) as well as Hoffman and Jamal (2014, pp. 599, 603–604) have studied protest participation in Tunisia and Egypt based on the second wave of the Arab Barometer. Both studies find that unemployment was not a statistically significant predictor of participation in the Arab Spring insurgencies in either country. Taken together, this existing research provides a mixed picture about the relationship between unemployment and the Arab Spring.

This article contributes to the existing research in many ways. The above-mentioned studies have made some observations about unemployment and the Arab Spring. However, in these studies, unemployment has been one factor among others in regression analyses, or it has otherwise received minor attention. The articles that have analysed individual level participation, have mostly studied just Tunisia or Tunisia and Egypt (Beissinger et al., 2012; Doherty & Schraeder, 2015; Hoffman & Jamal, 2014; Shafiq & Vignoles, 2015). In addition, earlier studies have often modelled participation in the Arab Spring based on some other protests, they have not discussed further their scattered findings about unemployment or the inspection of the participation of the unemployed has been descriptive.

Even though unemployment is a difficult concept in the developing economies, previous studies have not paid much attention to how to operationalise it. For example, Hoffman and Jamal (2012) have counted all those who have not reported being employed as unemployed. Thus, they have counted also the retired, housewives, and students as “unemployed,” which is apparently an imperfect solution. Here I will put more emphasis on tackling the problem of measuring unemployment in the Arab world. I also pay special attention to the young unemployed and the educated unemployed. This has not been done previously. To my knowledge, there are no prior studies that would have analysed the relationship of the unemployed and the Arab Spring as thoroughly as I will do. Further, as will be pointed out, there is in general surprisingly little research on the relationship between unemployment and political instability. This scarcity highlights the importance of this paper as one new empirical case study about the struggles of the unemployed.

2 Literature survey and hypotheses

The Arab Spring manifested itself in various forms of political instability, thus it is relevant to survey the connection of unemployment both to protesting and to organised violence. There are two main theoretical approaches suggesting that unemployment increases the likelihood of political instability. The first of these two is economic and it is often traced back to Becker (1968) who studied crimes. Later Grossman (1991) and Collier and Hoeffler (1998), among others, have promoted the economic approach and applied it to conflicts. The assumption is that individuals take part in collective acts of violence because of material inducements. When options for non-violent employment are scarce, soldiering might offer people alluring opportunities to earn money. Consequently, high unemployment rates can fuel conflicts by facilitating the recruitment of rebels. Further, Lichbach (1995, p. 342) argues that the unemployed will participate also in protests more probably because they face no opportunity-costs from losing time at paid work.

The second prominent approach is relative deprivation theory which has been more in favour of social psychologists. According to Gurr (1970), an influential contributor to the theory, relative deprivation is experienced by those whose actual circumstances and commodities do not equal what they expected to gain and what they think they are rightfully entitled to. The individual’s perception of deprivation is important, people whose expectations are not met can feel themselves deprived even if an “objective” outsider would consider otherwise. The relatively deprived use political violence to struggle against their situation. Scholars of social movements have linked relative deprivation also with protesting; those relatively deprived protest to redress the deprivation they face (Klandermans, 2011, pp. 40–42).

However, various arguments challenge this assumed positive linkage between unemployment and political instability. First, stigmatisation has been considered one of the most important obstacles for the collective action of the unemployed (Perry & Reiss, 2011, p. 32). If public opinion and elected officials consider unemployment to be a personal fault of an individual, the insurgence of the unemployed lacks legitimacy (Chabanet & Faniel, 2011, pp. 392–393).

Second, to think that the employed are wealthy and happy and the unemployed are deprived and unhappy might be a simplistic and a Western mindset. Cramer (2011, p. 17) reminds that globally there are millions of people whose working conditions are “so dreadful that it is probably no more difficult to recruit [rebels and soldiers] among them” than among the unemployed. There are also some issues related to the concept of being unemployed. The definition of unemployment requires that an individual is looking for work. In the developing economies, people are maybe not looking for work because they consider seeking work by conventional means useless if the informal economy and self-employment offer the major employment opportunities (World Bank, 2019). In the Arab world more than 10% of the employed work in the informal sector (Chaaban, 2010, p. 18). Another issue is that in the developing countries state welfare benefits are mostly scant or non-existent so people cannot afford to be unemployed in the sense understood in the West (Cramer, 2011, p. 15).

Third, the unemployed are a socially heterogeneous group, and many of them want to get rid of being unemployed. Both these factors hinder the formation of a collective identity which is important in collective action such as protesting (Chabanet & Faniel, 2011, p. 394; Klandermans, 2011, p. 46).

Fourth, social movements are never spontaneous events (Chabanet & Faniel, 2011, p. 397). Although the Arab Spring revolts have been labelled spontaneous from time to time, Gelvin (2012, pp. 54–55) reminds how the April 6 Movement among other organisations was coordinating the Egyptian uprisings. The movement, for example, organised twenty-one marches to Tahrir Square because a single march could easily have been broken by the security forces.

Fifth, the unemployed lack resources such as money, political skills, leadership, organisational resources, and social networks which are necessary for mobilising a protest (Klandermans, 2011, pp. 45–46; Schlozman & Verba, 1979, pp. 12–20).

Further, protests do not provide employment opportunities to the protesters (Klandermans, 2011, p. 55). The rational unemployed can thus be expected to devote their time and skills to finding work rather than to protesting (Schlozman & Verba, 1979, p. 351). This so-called free rider problem is not less notable in the case of authoritarian Arab states where rioting is indeed risky.

The contention of the economic approach that money is an essential motive for an individual to join groups that use violence is also challenged. For example, people have given manifold motivations for joining the Colombian guerrilla movement FARC: the allure of military life, getting a gun, personal problems, the boredom and miseries of daily life, sexual harassment at home, the wish to get training and an education, and political ideologies (Gutierréz-Sanín, 2008, pp. 14, 17, 20–25).

Last, if insurgence is considered something people do by choice, people possibly start fighting when their basic needs are filled. This rather suggests that an improved economic situation would lead to higher levels of violence at the same time that it leads to lower levels of unemployment (Berman et al., 2011, pp. 497–498). Similarly, Klandermans (2011, p. 46) suggests that those who protest are maybe not the poor and the powerless but those who are slightly better off.

3 Data, methods, and measurements

Used data sets are the third wave of the Arab Barometer (referred to as ABIII) and the sixth wave of the World Values Survey (WVS6). In total ABIII has over 14,000 respondents from twelve different Arab countries. From WVS6, eight Arab countries and their roughly 10,000 respondents are included. Later, when I refer to “all countries” of WVS6, this always stands for all these eight analysed Arab countries of WVS6. Following the most common definition, I consider the member states of the Arab League to be Arab countries. The data sets represent adults 18 years old and older in each country. Table 1 presents the countries included in the analyses and those Arab countries of WVS6 which are excluded because some questions were not asked. The table also shows when the interviews have been conducted in these countries.

The article mainly studies all Arab countries in which there is data as one sample. To supplement the picture of the mobilisation of the unemployed, hypothesis 1 is also studied in some individual countries and country groups. There could be country-specific differences in the participation of the unemployed because, for example, labour unions play a role in protest mobilisation and they have different relationships to the unemployed in different countries (Chabanet & Faniel, 2011, pp. 398–404). It is also suggested that in different Arab Spring countries different socio-economic groups mobilised the rebellions (Anderson, 2011, p. 2). Because the article has a special focus on the Arab Spring, the individually studied countries are Egypt, Libya, Tunisia, and Yemen. According to Allansson et al. (2012), p. 46 these countries were hardest hit by the uprisings together with Bahrain and Syria, from the two last-mentioned countries there is no data.

Linear regression – and more precisely linear ordinary least squares regression – is used with hypotheses 4 and 5. In other cases, binary logistic regression is applied. Some assumptions related to the use of regression are worth discussing here. The independence of observations is an assumption in all regression analysis. It should not cause concern in this study as the observations here are single respondents interviewed separately. A source of debate is the use of Likert scale variables in a linear regression model. In all my linear regression models the explained variable is a Likert scaled question or a combination of two Likert scaled questions. Following Vehkalahti (2008, p. 37) and Newsom (2013, p. 2) I consider that the use of Likert scale variables is justified here.

Lumley, Diehr, Emerson, and Chen (2002) argue that it “is widely but incorrectly believed” that linear regression would be valid only with normally distributed explained variables. Some previous studies have found that the normality condition loses its significance with sample sizes of some 100 observations, and Lumley et al. conclude that 500 is enough even with an extremely non-normal data. Here all explained variables of linear regression models are not normally distributed. However, the sample sizes are above 9,000, so this should not be a problem.

To tackle the challenge of measuring unemployment in developing economies I will use three ways to operationalise unemployment. First, I will simply utilise questions in which the respondents themselves report their employment status (q1005 and q1006 in ABIII, V229 in WVS6). The options given differ slightly between ABIII and WVS6, but both surveys offer options of being “unemployed” and working “full-time,” i.e. 30 hours a week or more. Those respondents who answer that they are unemployed will obviously form the group under scrutiny, the unemployed. In my analyses, those who work full-time constitute the group of employed. Respondents can also have chosen to be part-time employed, self-employed, retired, housewife, and student. All these respondents who are neither full-time workers nor unemployed will be grouped as having other employment status. Part-time workers are classified as having other employment status because it is unknown whether they work one hour or 29 hours a week. Categorising them as employed or unemployed would, either way, disturb the analysis. This division of respondents into three groups is coded by two separate binary variables. In the “Employed” variable the employed are coded as 1 and all the rest of the respondents as 0. In the “Others” variable respondents with other employment status are coded as 1 and unemployed and employed respondents as 0.

Because the unemployed are coded as 0 in both variables, they form the reference category. Consequently, the “Employed” variable will enable us to compare the employed with the unemployed (Long, Long, & Freese, 2006, pp. 417–418). We are not interested in comparing the unemployed and the “Others” group because the “Others” group is so heterogeneous. It would have been possible to exclude respondents with any other employment status than employed or unemployed but including “Others” increases the sample size. Bigger sample size in turn means that the control variables are better adjusted.

As argued earlier, the nature of labour markets in the Arab world challenges the definition of being unemployed. Robbins, project director of the Arab Barometer, assumes that some respondents who are not currently working might still have answered that they work when they mean that they work in general. This can particularly be the case with those working in the informal sector (Robbins, 2014). Thus, there can be some bias in this measure of unemployment as the responses to this question may not fully reflect the formal definition of unemployment. To handle these issues, I also use income as a proxy for employment. Since most people arguably get most of their money from work it is reasonable to assume that people who are mostly unemployed have a low income.

I use two ways to measure income which I call absolute income and relative income. These are the two other operationalisations of unemployment. In WVS6, respondents are asked (v239) to specify themselves in which of the ten shown income group they belong. In ABIII I have calculated the income groups myself based on the monthly income of the respondents (q1014). I have divided all respondents in each country into ten groups so that the groups are as equal in size as possible. The respondents of both ABIII and WVS6 are coded so that those in group 1 have the lowest income and those in group 10 the highest income within a country. For various reasons, these questions about absolute income do not necessarily tell whether the respondents’ income covers their needs; for example, there are differences in price levels within countries. In ABIII the relative income is based on question (q1016) where respondents are asked about the sufficiency of their household income. Those facing significant difficulties in meeting their needs are coded as 1 and at the opposite end respondents able to save as 4. In WVS6 there is no question to measure relative income. In analyses of relative and absolute income only those respondents are included who have to the employment status question answered that they are unemployed, part-time employed, full-time employed, or self-employed.

In ABIII, respondents are asked (q800a) whether they in 2011 and 2012 had participated in demonstrations and rallies to which the Arab Spring led. Following from this wording in ABIII the article concentrates in the Arab Spring related protests in years 2011 and 2012. WVS6 provides information on whether respondents have during the last year attended peaceful demonstrations (v87 and v92). This is coded as a binary variable where 1 is for respondents who did protest during the last year and 0 for those who did not.

The best question to measure satisfaction in life in ABIII is (q102a) “Generally speaking, how would you compare your living conditions with the rest of your fellow citizens?” The five options given are coded from 1, “much worse,” to 5, “much better.” WVS6, instead, provides question (v23) asking “All things considered, how satisfied are you with your life as a whole these days?.” The scale for answers ranges from 1, “completely dissatisfied,” to 10, “completely satisfied.” The relative deprivation theory assumes that people who feel deprived after comparing themselves to some other people are dissatisfied and start rebelling. Thus, questions both from ABIII and WVS6 measure whether the respondent is relatively deprived in the sense of the theory.

There are two four-step questions both in ABIII (q404 and q405) and in WVS6 (v7 and v84) where respondents rate their interest in politics. For both data sets, I have calculated the average of the two items. The scale goes from 1.0 to 4.0: the more interest in politics the respondent has, the higher the value.

The control variables in each model are age, education level, gender, country of residence, and whether the respondent lives in an urban or rural area. For binary logistic regression, education (question q1003 in ABIII, v248 in WVS6) is coded on a four-step scale where 1 implies the lowest educational level and 4 the highest. The “Age” variable equals simply the age of the respondent in years. Gender is coded as 1 for males and 0 for females. In most countries of ABIII, the respondents are coded as urban, 1, or rural, 0, based on question (q13) which specifically asks this. For respondents from Kuwait this information is not available but 100.0% of Kuwaitis live in urban areas (World Bank, 2019), so I have coded all Kuwaiti respondents as urban. In Palestine “refugee camp” is available as a third option, Palestinians living in refugee camps are excluded from analyses. In WVS6 respondents living in towns with a population greater than 10,000 are considered urban and respondents living in towns smaller than that, rural (v253). Country is used as a control variable when more than one country is included in the sample. There is a separate binary variable for each country except for the country serving as reference category. Even though these binary country variables are not reported in the tables, the country variables are included in all analyses that study more than one country. In hypotheses 4 and 5 the results are controlled for marital status (q1010, v57), those married are coded as 1 and all others as 0.

Descriptive statistcs, frequency tables and correlation matrices of the studied variables can be found as appendices.

When respondents are asked about protesting and interests in politics, they might try to avoid answering or answer dishonestly. This is a possible source of error. ABIII includes question (q513) asking how satisfied respondents are with their government. If respondents would have been afraid of consequences, one could expect good grades for governments in this question. However, on a scale from 0 to 10, the most common answer has been zero: the respondent is absolutely unsatisfied with the government. About half, 52.1%, of the respondents have evaluated their satisfaction to be lower than five, which is the median, and 29.1% have considered themselves more satisfied than five. In addition, in this question the share of missing answers, 3.6%, is relatively low and close to the share of missing answers for example in question (q701a) concerning trade with other countries, there 4.0% have failed to answer. The question about foreign trade could have had less missing answers if many respondents had refused to answer in fear of state surveillance. These figures do not support suspicions of unwillingness to answer truthfully in fear of consequences. Last, there is no apparent reason why this potential source of error would affect the unemployed and the employed differently.

4 Results

Figure 3 shows that, maybe not surprisingly, students were most active in the Arab Spring uprisings, the retired and the housewives the most passive. Therefore, it appears that employment status is linked to the likelihood of protest participation. However, the difference between the unemployed and the employed seems marginal.

Figure 3:

Protest participation by employment status.

In the two leftmost analyses of Table 2, which utilise the employment status question, the results of the hypothesis can be read from the “Employed” variable. According to both data sets the employed were slightly less active protesters than the unemployed, however, these differences are not statistically significant. This model can be considered as the base model which will be modified when studying hypotheses 2 and 3. The three rightmost analyses show the alternative measurements for unemployment, absolute and relative income. Here the coefficients suggest that the higher the income, the higher the likelihood of an individual to take part in protests. In the case of absolute income and WVS6 this relationship comes close to statistical significance (p = 0.071). Although the results in hypothesis 1 differ to some extent depending on which of the three measurements of unemployment is used, the central finding is obvious. None of these analyses provides appreciable support for the hypothesis that the unemployed would be more active protesters than the employed. Because there are no great differences between the three operationalisations, the direct question of employment status will be the only measure of unemployment used from now on.

Table 3 shows the results of the leftmost model of Table 2 in the subsamples of some countries and country groups. When Egypt, Libya, Tunisia, and Yemen are studied together, the employed are statistically significantly more active in protesting than the unemployed. However, in the subsample of “other countries,” which includes all countries except the four countries studied separately, the employed protest statistically significantly less probably than the unemployed. If all these countries of the ABIII data were studied one by one, it would turn out that the coefficients of five countries support the hypothesis, two of these countries suggesting a statistically significant relationship, and seven countries contradict our assumption, one of these seven, Libya, being statistically significant. The coefficients show that in some countries the difference between the unemployed and the employed is notable, whereas in some countries, as in Tunisia, there are practically no differences between these groups. Taken together, no unambiguous relationship appears between protest participation and being unemployed.

To study hypothesis 2, I have run the base model of hypothesis 1 but included only respondents 18–35 years old. Youth is defined following Bjorvatn and Høigilt (2016, p. 54), who argue that 35 years is an appropriate upper limit for youth in the context of the Arab world. When studying hypothesis 3, I have run the model only for respondents with higher than secondary education.

Table 4 shows that there is faintly more support among the Arab youth for hypothesis 2 than there is among the all age population for hypothesis 1. Both data sets give a negative coefficient for the “Employed” variable, as expected based on our hypotheses, and both results come closer to being statistically significant (p = 0.077, p = 0.094). In turn, there is no considerable evidence that the educated unemployed would be more active insurgents than the educated employed.

The results in Table 5 substantiate hypothesis 4, the unemployed are indeed statistically significantly less satisfied. Compared to the unemployed, the employed are 0.2 or 0.8 units more satisfied on a scale of 1–5 of the ABIII or 1–10 of the WVS6, respectively. The coefficients suggest that the unemployed are less interested in politics than the employed, as assumed in hypothesis 5, but the difference between the groups is small. Neither is the difference statistically significant, although with WVS6 it comes close to statistical significance (p = 0.059).

Table 6 shows that, unlike what is assumed in hypothesis 6, satisfaction with life does not have an effect on the likelihood of protest participation. By contrast, hypothesis 7 gets robust backing, those interested in politics are clearly more likely to protest. Further analyses in the two leftmost columns suggest that the slim interest in politics of the unemployed is a factor lowering their participation in protests. In other words, when interest in politics is added into the base model, being unemployed becomes a slightly stronger predictor of participation in a protest. The changes are minor, but the same shift can be observed also when adding interest in politics to the subsamples of Table 3; the coefficients of the Employed-variable move in the negative direction, and in case the coefficients are negative their p-values decrease and in case the coefficients are positive they turn negative or their p-values increase.

Control variables have shown that higher education increases the likelihood for protesting very clearly, higher age reduces the likelihood for protesting almost as clearly, and males are obviously more active protesters than females. Results also indicate that those living in urban areas are generally more likely to protest. Although not reported, the country has been a control variable in all models where more than one country is analysed, and in every one of these models, there have been considerable differences between countries.

4.2 Comparisons with earlier research

The results of all hypotheses are summarised in Figure 4.

Figure 4:

Findings of the hypotheses.

The arrows in the figure indicate assumed relationships, and the colours of the arrows tell whether the results suggested a positive relationship (blue) or no relationship at all (blank). None of the relationships presented in the figure was predominantly negative. The figure partly generalises the results but in short, my analyses show that the likelihood of the unemployed of taking to the streets is not higher (arrow 1) despite the fact that they are more dissatisfied (arrow 2). Further, this passivity of the unemployed can be partly explained with the finding that, in general, dissatisfaction with life does not increase the likelihood of protesting (arrow 3). Being interested in politics then again increases the likelihood of protesting (arrow 4). But the unemployed are rather less interested in politics (arrow 5), and my results hint that this lessens their participation in protests.

On the other hand, one should not make too bold claims about causality on the basis of this article. For example, it might be that individual-level personality traits, such as being enterprising or sociable person, make people both politically active and employed. It seems more plausible to think that interest in politics drives people to participate in protests than the other way around, i.e. that participating in protests would make people interested in politics. On the other hand, while being unemployed can arguably make people dissatisfied with life, being dissatisfied might also make it harder to get a job. But even we cannot be sure about causalities between the different factors, this does not change the core results that the unemployed are unhappy but relatively passive protesters compared to the employed. After all, it perhaps seems understandable that the unemployed were less eager to take part in the Arab Spring protests. Maybe people who have problems in feeding themselves are not so interested in whether the country is governed democratically or not.

Earlier literature has emphasised the important role labour unions often play in the mobilisation of the unemployed and the differences between countries in how labour unions take the jobless into consideration (Chabanet & Faniel, 2011, pp. 398–404; Klandermans, 2011, p. 45). This might offer one explanation for the suggestive result that the unemployed were more passive in some Arab countries than in others. Langhor (2014) has found that countries differed considerably as to which role the labour unions took in the Arab Spring uprisings.

Until now we have traced the relationship between unemployment and political instability by observing whether the unemployed have been active rebels compared to the employed. But people who are not unemployed themselves can take political action concerning unemployment (Schlozman & Verba, 1979, p. 20). This can happen because of solidarity or because of a large number of unemployed makes those who are working concerned about their own future. The employed and unemployed are connected through social networks. They can also live in the same households.

In Tunisia the most chanted slogan of the Arab Spring protesters is said to have been “Employment, freedom, and dignity” (Ben Mhenni, 2014). Soon after the uprising had toppled Tunisia’s former ruler Ben Ali, a survey found that 60% regarded unemployment as the biggest or second biggest problem facing the country (International Republican Institute, 2011). Thus, it seems that people brought out their frustration with unemployment, even though in Tunisia the unemployment rate had rather steadily decreased during the decades before the Arab Spring (World Bank, 2019). Arampatzi, Burger, Ianchovichina, Röhricht, and Veenhoven (2018) have demonstrated that as a whole there were substantial improvements in economic and social indicators throughout the Middle East and North Africa region during the decades preceding the Arab Spring. But simultaneously, in the years before the Arab Spring people had become increasingly less satisfied with their lives. Arampatzi et al. identify this as the “unhappy development” paradox.

We know now that the unemployed were not more likely to revolt and that in most Arab countries the unemployment rates decreased or remained steady in the decade preceding the Arab Spring (World Bank, 2019). It also seems that people can be dissatisfied with the employment situation even if the unemployment rate has decreased. People who are not unemployed themselves might riot because of unemployment. But on the other hand, those who are less satisfied are not more likely to protest. The results of this study should not be interpreted as a proof that unemployment was irrelevant for the Arab Spring. Still, the results give us a strong reason to question whether unemployment was as important a factor in this chain of events as many have suggested.

5 Conclusion

It is important to study the significance of unemployment for political instability to get a better understanding of the mechanisms behind rebellions, but also to provide guidance for policies aiming at preventing political instability.

This study found that compared to the employed, the unemployed generally were no more likely to participate in the Arab Spring uprisings or some other protests of the region. This outcome is robust and all three different ways used to operationalise unemployment support it. Unlike what was hypothesised, in several cases it turned out that the unemployed actually are less likely to protest than the employed. The results suggest that, due to their low interest in politics, the dissatisfaction of the unemployed probably does not erupt as rioting. Theoretical considerations also provide other possible explanations for why the unemployed do not take action. This might be due to the fact that they, for example, lack social networks, are marginalised or believe that rebelling is unlikely to bring them jobs. Taken together, although my results do not provide a complete picture, all the results of the study form a logical whole, where one part supports the other.

The results show clearly and consistently that being younger and being more educated are, as expected, factors which increase the likelihood of an individual to protest. This finding further underlines the slight and controversial effect of being unemployed.

Earlier research studying other countries indicates that unemployment is not a strong causal factor predicting political instability at an individual or country or region level. The key result of this study is in line with this finding; neither does unemployment seem to have been a major driving force behind the Arab Spring. Against the background of earlier research this result might seem expected, but taken into account the numerous assumptions about the linkage of the Arab Spring and unemployment, these findings are rather surprising.

There are – as far as I know – no prior studies which would have focused on the role of the unemployed in the revolts of the Arab Spring. So first, this study has shed more light on the net of drivers behind the Arab Spring. Second, this study is novel in the thoroughness with which it examines the role of unemployment in the protests of the Arab countries more generally. But in addition, neither has the significance of unemployment for protest movements and rebellions globally been studied too much. This study is one new case study to increase our knowledge in this field.

It is not advisable to make, based on these results, country-specific assumptions about the passivity or activity of the unemployed elsewhere. One indicative result of this study is that there are considerable differences between countries in how unemployment influences the protest engagement of individuals. Thus one direction for future research would be to explore whether the differences really are country specific. And if country-specific differences do exist, could they at least partly explain contradictory results of earlier research in relation to unemployment and political instability? And further, what causes these differences between countries in the first place?

On the other hand, something from all this could be generalised. Namely, based on both my results and earlier studies, the significance of unemployment as a driver of political instability generally appears to be weaker and more complex than is often assumed. As to policy recommendations, this study joins those previous analyses which suggest that programmes aiming at fostering peace solely through creating jobs are likely to miss their target.