Who fakes support for the military? Experimental evidence from Tunisia

ABSTRACT

Surveys around the world report exceptionally high levels of support for the military. This is particularly relevant for countries in transition from authoritarian rule to democracy, where militaries can play a vital role for democratic consolidation or autocratic backsliding. Given the sensitive nature of the issue, we suspect that figures indicating strong support for the military are at least partly driven by sensitivity bias. We explore this possibility through list experiments in two nationally representative surveys in Tunisia. We find that misreporting of support for the military in Tunisia is substantial, with respondents overreporting positive attitudes by 40–50 percentage points. Moreover, misreporting is not random, but instead varies systematically by incumbency, with supporters of governing parties misreporting support for the military to a significantly higher degree than opposition supporters or non-voters. Our results suggest that public opinion researchers should be wary of using direct questions to measure support for the military.

KEYWORDS:

• Military
• trust
• list experiment
• misreporting
• Tunisia

Introduction

In the wake of the Arab Spring, scholars have paid increasing attention to the political role of military organizations in successful and failed democratic transitions.¹ Despite their political centrality amid transitions from authoritarian rule, we have little robust knowledge about just how popular military organizations are, or about which social groups support the armed forces politically.

Conventional public opinion surveys are of limited help. They frequently find that militaries are among the most trusted institutions in the developing world. We contend, however, that these high levels of support for the military may be distorted by sensitivity bias.² Some respondents, in other words, say they support the military in public, while secretly harbouring more critical views of the armed forces. To the best of our knowledge, scholars have yet to explore this issue in the context of emerging democracies and hybrid regimes in the Global South and hence to expand experiment-driven scholarship on popular perceptions of the military beyond the US and European context. We address this lacuna by asking two interrelated questions: first, do people in emerging democracies misreport positive perceptions of the military? And second, if we find such misreporting, which social and political groups are most likely to lie in public?

Gaining robust knowledge on popular support of the military is important for the research programme on democratic transitions. First, pacted transitions often generate “reserved domains” for former members of the ancien regime, including military elites, which threatens to compromise the quality of democracy.³ It remains vital to gauge popular support for such political forces beyond the control of democratic institutions. Relatedly, militaries have become increasingly important agents not only in popular mass uprisings against authoritarian rulers, but also in the political dynamics of democratic backsliding.⁴

Tunisia is an excellent case to study popular perceptions of the military for both empirical and theoretical reasons. Empirically, Tunisia represents the political dynamics of democratic transition and consolidation very well, in that it has embarked on an uncertain post-Arab Spring democratization process that may just have met its fate with the monopolization of power by President Kais Saied in July 2021. Second, Tunisia also remains an intriguing case for studying the misreporting of popular support for the military since the role of its military in politics has remained marginal, at least in contrast to many of its Arab neighbouring countries.⁵ Tunisia thus introduces a “hard case” in that one should not expect popular perceptions of the military to be characterized by preference falsification.⁶ We do, however, find significant levels of misreporting in revealed attitudes towards the military across two independent nationally representative surveys in Tunisia. We leverage this evidence to explore the drivers of misreporting.

Why do people misreport support for the military? We begin with addressing the commonly assumed drivers of sensitivity bias: general social desirability concerns in response to social norms and conventions and preference falsification as a consequence of a fear of punishment. We expect the first source of sensitivity bias to be prevalent across different regime types, while the second source of sensitivity bias is largely associated with authoritarian state-society relations. The latter appears less relevant for democratizing Tunisia, while the former might be too broad a psychological disposition to account for the actual drivers of misreporting. We therefore theorize sources of sensitivity bias beyond general social desirability concerns or outright fear of punishment. In particular, we contend that political expediency is an important driver of misreporting. More specifically, we generate and test two alternative hypotheses on the political sources of misreporting. Partisan bias would suggest that misreporting varies by respondents’ party-political alignment, particularly whether their party is under- or over-represented in the ranks of the military. Incumbency bias, in turn, would suggest that faking support for the troops is most substantial where a respondent’s party is part of the government. Comparing respondents’ answers to an experimental and direct measure of support for the military, we find considerable variation in misreporting in our data and demonstrate that this variation best fits the incumbency bias hypothesis.

The remainder of this article proceeds as follows. The next section examines the findings from conventional public opinion research on the military and the potential for sensitivity bias. The third section discusses the empirical context of the Tunisian case and develops testable hypotheses on the political sources of misreporting. The fourth section introduces our experiment, tests these hypotheses, and provides evidence for our findings. For a robustness test, we draw on further empirical insights using a second survey, and then conclude with broader implications for research on civil–military relations and public opinion surveys.

Popular perceptions of the military

Existing public opinion data emphasize overwhelmingly positive perceptions of the military. Available surveys often report very high levels of trust in the military – and do so almost uniformly across countries and across time. The three waves of the Arab Barometer survey, for instance, report trust in the armed forces at levels between 76% and 82% for all countries included, and even higher levels when focusing only on Tunisia.⁷ Beyond the Middle East and North Africa, the Pew Global Attitudes surveys and the World Values Survey reveal similar findings.⁸ Figure 1 shows levels of trust in the armed forces in different countries from the 7th wave of the World Value Survey. As can be seen from this figure, in 63 out of the 75 countries included, respondents reporting trust in the armed forces outnumber those who do not. Relying on available data, the inevitable conclusion is that the military is among the most trusted institutions in the world.

Figure 1. Trust in the armed forces (WVS, wave 7).

To what extent are such figures valid measures of popular attitudes towards the military? Public opinion researchers have long argued that survey responses to particular items considered sensitive by respondents might be biased. Respondents have been found to consistently overreport behaviour they see as socially desirable and to underreport behaviour judged socially undesirable.⁹ Social desirability bias has led respondents to overreport turnout in elections¹⁰ or support for veterans,¹¹ for example, or to underreport socially less desirable attitudes or behaviour, such as anti-immigrant sentiment,¹² racist views,¹³ and vote buying.¹⁴

Scholars expect sensitivity bias in attitudes towards the military for two broad reasons: social pressure and the threat of political punishment. Soldiers are putting their lives on the line for the nation, producing a “societal injunction” to support the troops.¹⁵ “[N]o one […] ever wants to be accused of not supporting the troops,” writes Diane Mazur in the U.S. context. “The most effective conversation stopper ever invented in contemporary American dialogue is the charge that someone doesn’t respect the military.”¹⁶ Similar social desirability pressure could be expected to affect respondents’ answers in other countries as well.

Beyond public pressure to correspond to existing social conventions, exposure to physical harm and fear of punishment can prompt respondents to publicly falsify their preferences as well.¹⁷ For instance, Matanock and García-Sanchez find that fear of punishment in Colombia may move respondents to overreport support for the military’s counterinsurgency campaign.¹⁸ In authoritarian regimes, respondents may fear government monitoring and punishment, leading them to overreport support for incumbents and underreport support for militant groups.¹⁹

Who lies in public?

There is value in reaching beyond broad social desirability bias and the threat of punishment as sources of misreporting, however. For one, assuming a general social desirability bias relies on broad psychological dispositions among the population writ large and obscures variation in misreporting across social groups, some of which may be more prone to misreporting trust in the armed forces. For another, people appear to lie in public in democracies as much as in authoritarian regimes or amid violent conflicts, which would make the threat of punishment a weak predictor of misreporting.

Our curiosity to look beyond the threat of punishment and general psychological dispositions is inspired by existing work studying individual-level variation in the way people look at the military. Simon and Lovrich, for instance, unpack different sources of social support for military spending in the United States;²⁰ Sarigil shows variation between social and political identities in Turkey;²¹ and Hines et al. have studied public attitudes towards the military in the United Kingdom.²² While interest in popular support for the military has inspired a thriving research agenda, our main curiosity is the degree to which people misreport support, rather than the level of support for the military as such.

Why might Tunisians who are unsupportive of the military misreport support in public? We differentiate between two alternative sources of misreporting, namely partisan motivations and incumbency bias. Both sources have been identified in studies of public opinion in established democracies, notably in the United States. Yet, unpacking social variation in the degree to which people lie in public has not been attempted in other regime types, including in the volatile political dynamics of democratic transitions.

To begin with, respondents might misreport support for the military for partisan reasons. Addressing attitudes on affirmative action in the U.S. context, for example, Gilens, Sniderman, and Kuklinski show that misreporting of support for affirmative action was significantly more pronounced for liberal than conservative respondents.²³ They suggest that differences in the ideological orientation of the Democratic and Republican parties produced different degrees of pressure to publicly toe the party line. In a similar manner, the partisan divisions in post-revolutionary Tunisia might also structure misreporting.

A second source of bias is what we term incumbency bias. Scholars of the U.S. military have found that Democrats and Republicans become more supportive of the military in surveys when their party controls the presidency.²⁴ We argue that such increases in support may not be genuine, but rather reflective of sensitivity bias: when a respondent’s party is controlling the government, they may express more positive opinions about all state institutions, as they feel pressure to praise the state and therefore their party. In Egypt, for instance, government employees have publicly overreported their support for the country’s strongman Abdel Fattah al-Sisi.²⁵ With an association between political incumbency and the military, supporters of the incumbent party/parties may express more support for the military regardless of their genuine political and ideological attitudes.

Tunisia and the military

Tunisia is an excellent case for our inquiry aimed at reaching beyond conventional explanations of preference falsification and broad social desirability bias. To begin with, Tunisia at the time of our surveys (2017 and 2019) represents a “hard case” for finding preference falsification driven by the threat of punishment due to its comparatively liberal political environment and general absence of the military in public life. Detecting sensitivity bias in popular attitudes in Tunisia would therefore provide leverage for exploring the political sources of misreporting support for the troops.

The military in Tunisian politics

Of all the countries in the Middle East and North Africa, Tunisia is the one where we would expect to find low levels of misreporting when it comes to attitudes towards the military. At the time of our surveys, it was the lone Arab democracy, with Tunisians free to criticize the government and state institutions, and routinely doing so in public opinion polls. In the Arab Barometer surveys, for instance, Tunisians openly reported very low support for the government and police forces.²⁶ While one activist (Yassine Ayari) had been prosecuted for defaming the military, there have also been trials for defaming the police and governmental officials,²⁷ and those had not censored criticism.

Moreover, compared to other military organizations in the region, the Tunisian military has played a relatively marginal role in the country’s post-colonial history.²⁸ While military officers stood at the helm of post-independence state- and regime building projects in most Arab republics, Tunisia avoided military intervention and developed a civilian authoritarian regime based on a dominant party and strong personalist elements. Under both presidents Habib Bourguiba (1956-1987) and Zine El-Abidine Ben Ali (1987-2011), the Tunisian military was largely absent from the public eye. Other than a brief skirmish with the French in 1961, the Tunisian military never went to war, and both Bourguiba and Ben Ali relied instead on the police and internal security forces for repression and counterterrorism operations. The military was, since the early 1990s, heavily involved in international peacekeeping missions, which has contributed to its professionalization over the past three decades.²⁹ The military itself was one of the smallest in the Arab world, numbering less than 40,000 in a population of 11 million. As a retired Tunisian officer lamented to one of us, “we had no relations with society at all. In public life the army was absent.”³⁰ Another claimed that “the army was imprisoned in the barracks. We were kept isolated from society.”³¹

This isolation was temporarily broken in January 2011, when it was widely reported that General Rachid Ammar, army chief of staff, had refused Ben Ali’s order to fire on protesters during the Arab Spring. In fact, Ben Ali had not asked the military to repress protesters.³² But this rumour, combined with Ammar’s later statement that the “army would protect the revolution” and the role the military played in ensuring security in its wake, likely shaped many Tunisians’ first few (positive) images of the military. The military has since seen a slightly more public role and became increasingly involved in counterterrorism operations, election security, and the protection of economic production sites.³³ Its image of siding with protesters was also confirmed in 2017, when an army unit allowed protesters to storm an oil site in Tataouine despite the president’s orders to defend it.³⁴ Security concerns and economic crisis have also fuelled increasing expectations for the military to play a greater role in politics and the economy.³⁵

Despite these changes, however, contemporary Tunisian politics have retained their civilian character. While militaries have assumed a dominant position in other post-Arab Spring countries – namely in Syria’s civil war and in Egypt’s resurrection of authoritarianism – Tunisia’s transition to democracy since 2011 was characterized by the absence of the military from politics and a more prominent role for civilian forces, such as the country’s trade union federation.³⁶ Ministerial positions, including the defence minister, continue to be staffed by civilians,³⁷ and there is still no military involvement in the economy. In short, given its level of freedom and marginal role for the military, Tunisia is thus a hard case for finding preference falsification.

Potential sources of sensitivity bias in Tunisia

In this section, we develop two broad hypotheses on why Tunisians might still misreport support for the military. To begin with, we hypothesize that misreporting might be driven by partisan bias, originating from respondents’ support for ideological camps in political competition. In the United States, for example, the increasing overrepresentation of Republican partisans within the ranks of the military has been seen as one driver of increased support for the armed forces among Republicans at large.³⁸ A similar dynamic could be at play in Tunisia. The most prominent ideological divide in post-revolutionary Tunisia is that between Islamist and secular forces. As the ranks of the military are largely secular, Islamists may support the armed forces less than secularists. Islamists have been informally banned from the officer corps since the 1980s, a ban that was even more strictly applied after an Islamist coup plot in 1987 and a fake Islamist coup plot in 1991.³⁹

Given the military’s composition, Islamists may not only be less supportive of the military, but may also be wary of openly expressing those beliefs. The threat of a military coup, while unlikely, has been a remote possibility amid Tunisia’s early democratic transition.⁴⁰ Rached Ghannouchi, president of the Islamist party Ennahda, told Salafists in 2012 not to push too far as “Secularists still control the economy, the media, and the administration … the army and police also are not guaranteed.”⁴¹ These fears came to the fore in 2013, when a military coup in Egypt ousted the Muslim Brotherhood. A fringe group within the secular camp formed a Tunisian version of the Egyptian Tamarod (rebel) group and explicitly called for military intervention.⁴² There were reports about cooperation between Nidaa Tunis – Ennahda’s main secular competitor – and the Tunisian Tamarod movement, and Nidaa-leader Béji Caïd Essebsi referred to the Egyptian coup as the “second revolution.”⁴³ Military officers indeed felt pressure to intervene by secular politicians’ calls for a repetition of the Egyptian scenario,⁴⁴ though they ultimately chose to stay in the barracks. Given that the military is largely secular and has been egged on to intervene by secularists against Islamists, we can expect considerable partisan bias according to respondents’ placement on the secular-Islamist cleavage.

Hypothesis 1: In their perceptions of the military, supporters of the Islamist Ennahda party should exhibit a greater degree of misreporting than supporters of secular parties.

Second, we consider the presence of incumbency bias, that is, bias among supporters of political forces representing the government. Research on consolidated democracies has established that respondents report more positive views of the economy,⁴⁵ foreign policy decisions,⁴⁶ and the military,⁴⁷ if their preferred party is in power. This strand of research has also demonstrated that this incumbency bias can be mitigated if respondents are given additional incentives to report accurately, suggesting that the incumbency effect does not change respondents’ true perceptions, but rather the way in which they report them. As Prior, Sood, and Khanna conclude, even if “respondents hold accurate beliefs, their motivation to give an answer consistent with their partisan dispositions may outweigh their motivation to give an accurate response.”⁴⁸

Government supporters might thus overreport support for the military because they want to give an answer consistent with their political orientation, irrespective of their actual attitude towards the military. This suggests a different alignment of political groups than the hypothesis of a partisan bias. In Tunisia at the time of our first survey, the government was led by a coalition of four parties: the Islamist party Ennahda and three secular parties, Nidaa Tounes, Afek Tounes, and the Free Patriotic Union (UPL). Hence our alternative hypothesis:

Hypothesis 2: In their perceptions of the military, supporters of incumbent parties should exhibit a greater degree of sensitivity bias than supporters of opposition parties.

Empirical test: a list experiment

To explore the extent and potential sources of sensitivity bias, we conducted a nationally representative telephone survey in Tunisia in July 2017. Working with a local team from One to One for Research and Polling,⁴⁹ we surveyed 1038 adult Tunisians in Tunisian dialect. In order to assess the extent to which respondents misreport attitudes on the military, we combined a direct question with an experimental approach relying on the item count technique, more commonly known as a list experiment.⁵⁰ In a list experiment, respondents are presented with a list of items and are asked to report the number of items they endorse. In the treatment group, the list includes the sensitive item; in the control group it does not. Respondents in the treatment group can thus endorse the sensitive item without revealing this fact to the interviewer. Given random assignment and the fact that the only difference between the two lists is that one includes the sensitive item and the other does not, the proportion of respondents endorsing the sensitive item can be estimated by comparing the treatment and control groups.

In our context, we are not only interested in the true proportion of respondents holding a sensitive attitude, however, but in the extent to which respondents misreport such attitudes. In general terms, we use the term “misreporting” to refer to the difference between respondents’ true beliefs and their self-reported attitudes.⁵¹ Measuring variation in levels of misreporting across individuals relies on a comparison between respondents’ answer to the experimental question and their answer to a direct question capturing the same quantity. We use the maximum-likelihood estimator developed by Blair and Imai as well as Eady’s extensions of this procedure to estimate predicted probabilities of misreporting for each respondent.⁵²

Methodologically, the validity of the list experiment, and the comparison with a direct question, depends on three main assumptions.⁵³ The first is that the inclusion of the sensitive item in the list experiment does not alter the way respondents deal with the control items (“no design effect”). This assumption we can test for directly using the ict.test function in R; the analysis reveals no design effect. A second assumption, particularly for comparing the list experiment to the direct question, is monotonicity: that any difference between them is driven by respondents who were hesitant to openly reveal that they oppose the military, not those who were for some reason hesitant to reveal they support the military. In our survey, this assumption seems to be met, as only four out of 1038 respondents clearly report support in the experimental, but not in the direct question. We exclude these respondents from the analysis.⁵⁴

Finally, the no-liars assumption requires that respondents do not lie in the list experiment, as well. This assumption is frequently violated, for example, if respondents generally endorse all control items (ceiling effect) or, conversely, if they reject all control items (floor effect). In both of these cases, respondents might no longer believe that their endorsement of the sensitive item will remain hidden. We accordingly took care to develop the control items to avoid ceilings and floors. Other sources of violation might be if respondents do not understand the additional layer of protection afforded by the experimental setup.⁵⁵ In general terms, however, we would expect responses to the direct and experimental questions to be similar if the no-liars assumption was violated since respondents would then either misreport on both items, or misreport on neither. The fact that we find large differences between the direct and experimental questions across two independent experiments thus suggests that a systematic violation of the no-liars assumption is unlikely.

List experiments have come under some criticism. To begin with, list experiments rely on a rather complex procedure and there is evidence that these difficulties can lead to the breakdown of list experiments, particularly among less literate and numerate respondents.⁵⁶ In the Tunisian context, such concerns are mitigated by high general levels of literacy which, according to World Bank data, stood at 79% in 2014.⁵⁷ In our survey, the percentage of respondents without any formal education is low with 5.4% (56 respondents). Moreover, as we will discuss below, we find consistent effects in two list experiments conducted independently from each other, one embedded in a phone survey, the other implemented face to face.

In the analysis below, we use the following questions. Directly, we asked respondents:

What kind of influence is the military having on the way things are going in Tunisia?

1. Very bad

2. Somewhat bad

3. Somewhat good

4. Very good

5. No influence

6. Don’t know

7. Refuse (do not read)

Figure 2 presents respondents’ answers to the direct question. About 77% claimed that the military is having a “somewhat good” or “very good” influence on the way things are going in Tunisia. This is consistent with the high levels of support for the military found in the Arab Barometer, Pew, and World Values Surveys mentioned above. Notably, only 11% were openly willing to criticize the military.

Figure 2. Direct question on perceptions of the military in Tunisia (N = 1038).

We compare this 77% support in the direct question with an indirect measure of support obtained through the list experiment. In our experiment, respondents were asked: “I will name a number of institutions. Please tell me how many of them have a good or very good influence on the way things are going in Tunisia.” In the control group, respondents saw four items: political parties, parliament, the police, and the labour union federation (UGTT). The average number of items supported here was 1.31 (see Table 1). In the treatment group, respondents saw those four items plus a fifth: the armed forces. In the treatment group, respondents supported an average of 1.68 items, a difference of 0.37 compared to the control. That is, in the list experiment, we can estimate that about 37% supported the additional fifth item, the military.

Table 1. Direct and experimental measures (survey 1).

	List Experiment		Direct
Control	1.31	[540]
Treatment	1.68	[493]
Estimated Support	36.65***	(4.62)	76.88^+++	(1.31)
N	1033		1038

***Two-tailed significance of comparison between treatment and control group.

⁺⁺⁺ Two-tailed significance of comparison between experimental and direct measure.

N in square brackets; Standard errors in round brackets.

In short, when asked directly, about 77% of the sample said the military is having a good or very good influence. But when asked indirectly through the list experiment, only 37% did so. This difference – 40 percentage points – is large and statistically significant (p < 0.001). It suggests that around 40% of the sample are lying about their support for the military when asked directly.⁵⁸

Political sources of misreporting

In this section, we provide evidence for our assumption that sensitivity bias is in fact not random but rather varies across political cleavages along the two dimensions outlined above: partisanship and incumbency. If misreporting was driven only by general social desirability bias, we would expect the size of the difference between the experimental and direct measure of trust in the military to vary with the psychological disposition of the respondent (such as the ability to withstand social pressure), but not across social groups. While different respondents might well have individual reasons to succumb to or withstand social pressure, such individual differences should average out and the overall size of the bias should be comparable across social groups. Moreover, assuming that the size of the bias is roughly the same for all respondents, at least on average, the direct measure should be just as effective a predictor of other variables as the list experiment.

Figure 3 visually summarizes the levels of our experimental and direct measures of support for the military, as well as the extent of misreporting across different groups of respondents. For each variable, the figure displays the estimated response to the experimental (green) and direct questions (red), as well as the level of misreporting (blue); these quantities are displayed for respondents who do (triangle) or do not (circle) belong to a specific social group or hold a specific opinion. At the top of Figure 3 we thus see that levels of support as captured in the experimental measure do not differ significantly between those respondents who do support Nidaa Tounes and those who do not (33.56% vs. 35.05%; green, left panel), but that supporters misreport their attitudes towards the military slightly more than non-supporters (47.97 vs. 39.8%; blue, right panel). Estimates for the experimental question and the level of misreporting are obtained with the list package in R,⁵⁹ while estimates for the direct measure are based on a logit model. The triangles and circles represent the predicted percentage of support for the military (or misreporting), while the bars give 95% confidence intervals.

Figure 3. Sensitivity bias across social groups.

The main conclusion emerging from Figure 3 is that sensitivity bias with respect to support for the Tunisian military is not random. To begin with, the level of misreporting varies between 20.39% for respondents who do not believe that the military is having a positive influence on political stability, to values in the 60% range for voters of Ennahda (62.2%) and Moncef Marzouki (60.92%) in the 2014 legislative or presidential elections, respectively, as well as for respondents who think the military is making a positive contribution to democracy (61.3%). Overall, levels of misreporting thus vary up to 40 percentage points. Moreover, the difference between the experimental and direct question is not significant for all respondents. Respondents who do not think the military has a positive impact on political stability or is doing a good job with fighting terrorism, for example, do not significantly overreport support for the military–even though they still report higher levels of support when asked directly. The same observation holds for respondents who served in the military themselves, as well as for unemployed respondents, students, and housewives.

Partisan bias

We start by examining whether sensitivity bias varies by political preferences, particularly respondents’ voting behaviour in the 2014 parliamentary elections.⁶⁰ We hypothesized that respondents who voted for the Islamist Ennahda party would be less supportive of the military, but unwilling to express that sentiment openly.

Figure 4 provides suggestive evidence of this partisan bias. Ennahda voters indeed exhibit the largest degree of misreporting when it comes to their attitudes towards the military. While 84% of Ennahda voters say they support the military when asked directly, the list experiment suggests that only 17% actually do so. Estimating the degree of misreporting directly, we find that 63% of Ennahda voters misreport their attitudes towards the military–the highest level of misreporting overall (see Figure 3). This effect is significant at p = 0.0412, using the misreport package in R.⁶¹ At the same time, Nidaa voters misreport only slightly above the overall average (48%, p = 0.0304), while non-voters are less likely to misreport than the average (34%, p = 0.0011); misreporting among voters for other parties is not significantly different from the average (31%, p = 0.8243). In short, the data provide some initial evidence of partisan bias. (Figure 4).

Figure 4. Partisan bias.

Incumbency bias

Next, we explore whether sensitivity bias varies by whether a respondent’s preferred party is part of the government. This test builds on the previous analysis combining voters of all four coalition partners: Ennahda (N = 141), Nidaa Tounes (N = 308), Afek Tounes (N = 7), and the Free Patriotic Union (N = 15). We compare these ruling parties to supporters of other parties.

Here we find our strongest evidence of systematic bias. In line with our hypothesis that government supporters should want to publicly praise all state institutions, we find that respondents who voted for the ruling parties indeed were more likely than opposition voters to openly state support for the military in the direct question (82 v. 75%, p = 0.0003). However, in reality, government supporters were actually slightly less supportive of the military according to the list experiment (30 vs. 43%). As a result, respondents who voted for the ruling parties show considerably greater misreporting (52 percentage points) compared to respondents who voted for the opposition (30 percentage points, see Figure 5). This difference is significant at p = 0.0007.

Figure 5. Incumbency bias.

Multivariate regression

The two sources of misreporting just identified – partisan and incumbency bias – potentially overlap. Ennahda, for instance, was part of the government at the time of our survey. Hence, a simple comparison across groups cannot isolate the effect. We therefore test both biases while controlling for the other, to ascertain which is truly driving the variation. In addition, we control for a host of demographic variables in case those, and not the political ones identified above, are in fact the cause. Table 2 contains two regression models. First, a base model with the two biases voting for Ennahda, and voting for any ruling party. Model 2 then adds a control for the military’s counterterrorism activities in a respondent’s governorate as a proxy for both fear of punishment as well as social pressure to support the troops.⁶² It also adds a variety of demographic controls, including age, gender, education, unemployment, student, housewife, married, income, coast, urban, member of union, and prior military experience.

Table 2. Misreporting support for the military (survey 1).

	Model 1	Model 2
Government	0.4714**	0.4805**
	(0.1728)	(0.1831)
Nahda	0.1805	0.2236
	(0.2698)	(0.2741)
Terrorism		−0.17
		(0.1669)
Protest 2011		−0.2825
		(0.1851)
Age		0.0306
		(0.0382)
Female		0.0631
		(0.1944)
Education		0.0152
		(0.0595)
Unemployed		−0.2897
		(0.2497)
Student		−0.3733
		(0.3168)
Housewife		0.0665
		(0.2583)
Married		−0.4607*
		(0.2155)
Income		−0.0306
		(0.054)
Income refused		0.3777
		(0.4158)
Coast		0.2214
		(0.1707)
Military service		−0.2052
		(0.2303)
Treatment	−0.0715	−0.0886
	(0.1506)	(0.1524)
Intercept	1.0008***	1.2233**
	(0.1219)	(0.4028)
N	1029	1029
Log-likelihood	−1974.81	−1945.38

Note: ⁺p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.

Regressions conducted with misreport package in R (Eady 2017).

Model 1 suggests that incumbency bias is in fact driving the variation, remaining significant at p = 0.006. The effect of voting for Ennahda – partisan bias – remains positive but drops to insignificance. Model 2 shows that incumbency bias remains significant (p = 0.009) in the presence of demographic controls. The only demographic variable that reaches significance is marriage (p = 0.03). In short, our survey suggests that bias towards the military is large and varies systematically according to incumbency.

Robustness check: survey 2

Results from our first survey indicate substantially larger degrees of misreporting than the average list experiment.⁶³ We returned to Tunisia in September 2019 to further examine this phenomenon for methodological and substantive reasons. For one, given the large scale of the results from survey 1, a second survey at a different time was to rule out that this was simply a one-off event. Moreover, our second survey helped us account for possible floor effects in the design of the first survey. Finally, conducting a second survey helped us control for possible differences across survey techniques, with phone surveys found to be more vulnerable to respondents’ sensitivity bias than face-to-face interviews.⁶⁴

We therefore administered a face-to-face survey where we found even larger sensitivity bias: roughly 50 percentage points. The second survey asked respondents directly: “how much do you support the military?” to which 95% responded “a great deal” or “a lot.” We compare that 95% in the direct question once again to a list experiment. The list experiment asked, “Please tell me how many of the following institutions you support.” To avoid the potential floor effects from survey 1, we modified the control items to be: the UGTT labour union, the UTICA employers’ union, the anti-corruption authority, and the judiciary. This increased the level of support in the control group to 1.91 items (Table 3 below). In the treatment group, which featured a fifth item (the military), respondents on average supported 2.34 items, a difference of about 0.44. The list experiment thus suggests that the true level of support for the military is around 44%, compared to the 95% we found in the direct question, a statistically significant (p < 0.001) 51-point difference. Survey 2 thus confirms a massive level of misreporting of support for the military.

Table 3. Direct and experimental measures (survey 2).

	List Experiment		Direct
Control	1.91	[518]
Treatment	2.35	[490]
Estimated Support	43.88***	(7.89)	94.84^+++	(0.70)
N	1008		1008

***Two-tailed significance of comparison between treatment and control group.

⁺⁺⁺ Two-tailed significance of comparison between experimental and direct measure.

N in square brackets; Standard errors in round brackets.

Incumbency bias

We again find significant evidence that government supporters are more likely to overstate their support for the military. At the time of the second survey (2019), Tunisia featured a divided government, with the president hailing from Nidaa Tounes, but the prime minister receiving parliamentary support from Ennahda, Tahya Tounes, Machrou Tounes, and Afek Tounes. Survey 2 therefore classified as “Government” (N = 601) anyone who supported either side of the divided government, with the rest as “Opposition” (N = 412). The survey asked respondents to rate the performance of both the president and the government. It also asked about their intended voting behaviour in the upcoming 2019 elections. Government supporters are those who either ranked the performance of the president or government as “very good” or “somewhat good” or who intended to vote for candidates or parties from the government.⁶⁵ Figure 6 plots each group’s support for the military.

Figure 6. Incumbency bias (survey 2).

As in survey 1, when asked directly, government supporters are slightly, though significantly, more likely to support the military than opposition supporters (96 v. 93%, p = 0.0106). However, in reality, they are no different in their actual support for the military, as measured in the list experiment (44 v. 47%). Accordingly, government supporters show greater sensitivity bias (51 percentage points) compared to the opposition (46 percentage points). Though not substantively as large as in survey 1, this difference remains statistically significant (p = 0.0432). Table 4 demonstrates that incumbency bias is robust to the inclusion of a number of political and demographic covariates.⁶⁶

Table 4. Misreporting support for the military (survey 2).

	Model 3	Model 4
Government	0.6942*	0.7137*
	(0.3434)	(0.3537)
Nahda		−0.6881
		(0.4852)
Terrorism		−0.2309
		(0.3676)
Age		0.1086
		(0.0706)
Female		−0.2415
		(0.4468)
Education		0.1648
		(0.1632)
Unemployed		−0.8239^+
		(0.456)
Student		1.1947
		(1.0939)
Housewife		−0.3892
		(0.5371)
Income		0.0617
		(0.0764)
Income refused		−5.945
		(7.2084)
Coast		−1.0774
		(1.0433)
Military service		0.325
		(0.664)
Treatment	0.048	0.0679
	(0.3422)	(0.3496)
Intercept	2.7338***	2.8267*
	(0.2813)	(1.2936)
N	959	959
Log-likelihood	−1621.7	−1584.27

Note: ⁺p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.

Regressions conducted with misreport package in R (Eady 2017).

Conclusion

This article provides substantial empirical evidence for misreporting of popular support for the military in Tunisia. Sensitivity bias is real and varies systematically rather than randomly. In particular, we find evidence of an incumbency bias: supporters of the ruling parties are significantly more likely than other social and political constituencies to publicly state that they support the military when they in fact do not.

These findings have important empirical and theoretical implications. Selecting Tunisia helps us reach beyond preference falsification as a result of authoritarian pressure, but also supports the claim that our findings will travel beyond our individual case. First, our findings help open an empirical black box evident in the study of Middle East politics and the Global South more broadly. Existing survey measures of support for the military are problematic. Not only are respondents likely to overstate their support for the military when asked directly, but in addition, bias varies by concepts political scientists are interested in, such as voting for a ruling party. Direct questions about support for the military should therefore be approached with caution, including when simply exploring correlations. Second, from a broader perspective, our results suggest that existing findings in American politics that partisans become more supportive of the military when their party is in power, may actually reflect sensitivity bias, rather than a genuine increase in support of the troops.

Scholars and policymakers would do well to take sensitivity bias into account when examining public support for the military. First, these findings are important for research on democracy. The presumed high levels of support for the military, especially relative to lower support for the government and for democracy in general, have led some observers to predict a global crisis of democracy,⁶⁷ the erosion of civilian control over the military,⁶⁸ and future civil–military tensions and coups.⁶⁹ Moreover, public support can help to legitimize a military coup and allow coup leaders to avoid international sanctions.⁷⁰ In the specific case of Tunisia, the political crisis following President Kais Saied’s decision to suspend parliament on 25 July 2021 made the question of popular support for the military particularly pressing.⁷¹ Understanding true popular perceptions of the military thus has important implications for arguments about the crisis of democracy and political stability more generally. Second, our findings suggest that sensitivity bias may be far more common than anticipated in large-scale survey projects. The mechanism we explore helps explain why sensitivity bias towards state institutions can be substantial even in democracies where citizens enjoy the freedom to state their views in public. Finally, policymakers more broadly will find interest in understanding robust measures of popular perceptions of state institutions.

Acknowledgements

An earlier version of this paper was presented at the 2019 convention of the American Political Science Association and received the 2020 best paper award of the MENA Politics Section. We would like to thank Ellen Lust and the other participants in the APSA panel for their feedback. We also benefited from input provided by colleagues at the departmental research seminar at Leiden University as well as at the European University Institute. All remaining errors are ours.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Kevin Koehler

Kevin Koehler is an Assistant Professor of Comparative Politics (with tenure) at the Institute of Political Science at Leiden University. His research focuses the political role of militaries as well as on regime change in the Middle East and North Africa. For more information, please visit https://kevinkoehler.org/.

Sharan Grewal

Sharan Grewal is an Assistant Professor of Government at the College of William & Mary and a Nonresident Fellow at the Brookings Institution. His research focuses on civil-military relations, religion, and democratization in the Middle East and North Africa. He received a PhD in Politics from Princeton University in 2018.

Holger Albrecht

Holger Albrecht is a professor of Political Science at the University of Alabama. His research interests include civil-military relations, authoritarian regimes, and Middle East politics.

Notes

1 Brownlee, Masoud, and Reynolds, Arab Spring; Albrecht, Croissant, and Lawson, Armies and Insurgencies; Koehler, “Political Militaries.”

2 Blair, Coppock, and Moor, “When to Worry.”

3 Croissant et al., “Fallacy of Coup-ism.”

4 Koehler, “Political Militaries”; Albrecht, Croissant, and Lawson, Armies and Insurgencies.

5 Brooks, “Abandoned at the Palace”; Jebnoun, “In the Shadow”; Koehler, “Officers and Regimes.”

6 Rapport, “Hard Thinking.”

7 The data are available at: https://www.arabbarometer.org/survey-data/. See also the Appendix (Figure A1).

8 Johnson, “Trust in the Military.”

9 Krumpal, “Determinants of Social Desirability.”

10 Karp and Brockington, “Social Desirability.”

11 Kleykamp, Hipes, and MacLean, “Who Supports US Veterans?”

12 Janus, “Influence of Social Desirability.”

13 Kuklinski et al., “Racial Prejudice.”

14 Gonzalez-Ocantos et al., “Vote Buying.”

15 Kleykamp, Hipes, and MacLean, “Who Supports US Veterans,” 93; Golby and Feaver, Thank You for Your Lip Service?

16 Mazur, A More Perfect Military, 3.

17 Kuran, Private Truths; Lyall, Blair, and Imai, “Explaining Support.”

18 Matanock and García-Sanchez, “Counterinsurgent Success.”

19 Jiang and Yang, “Lying or Believing”; Truex and Tavana, “Implicit Attitudes”; Blair et al., “Poverty and Support.”

20 Simon and Lovrich, “Citizen Support.”

21 Sarigil, “Deconstructing the Turkish Military’s Popularity.”

22 Hines et al., “Are the Armed Forces Understood.”

23 Gilens, Sniderman, and Kuklinski, “Affirmative Action.”

24 Burbach, “Gaining Trust.”

25 Truex and Tavana, “Implicit Attitudes.”

26 Lotito, “Public Trust.”

27 Jebli, “Freedom of Expression.”

28 Brooks, “Abandoned at the Palace”; Jebnoun, “In the Shadow of Power”; Koehler, “Officers and Regimes.”

29 Albrecht, “Diversionary Peace.”

30 Interview with retired Colonel Major Mahmoud Mezoughi, Tunis, October 9, 2015.

31 Interview with retired Colonel Major Hedi Kolsi, Sfax, September 21, 2015.

32 Holmes and Koehler, “Myths of Military Defection.”

33 Grewal, A Quiet Revolution.

34 Grewal, “Military Defection.”

35 Albrecht, Bufano, and Koehler, “Role Model.”

36 Bishara, “Legacy Trade Unions.”

37 Buehler and Ayari, “Autocrat’s Advisors.”

38 Liebert and Golby, “Midlife Crisis?”

39 Grewal, “Tunisia’s Foiled Coup.”

40 Kinney, “Sharing Saddles.”

41 Amara, “Tunisia Islamist Leader.”

42 Achcar, Morbid Symptoms, 159.

43 Marks, “Tunnisia’s Ennahda,” 8.

44 Grewal, A Quiet Revolution, 8.

45 Prior, Sood, and Khanna, “You Cannot Be Serious.”

46 Bartels, “Beyond the Running.”

47 Burbach, “Gaining Trust.”

48 Prior, Sood, and Khanna, “You Cannot Be Serious,” 492.

49 The survey was conducted in August-September 2017 and used a national quota sample of adult Tunisians. Tunisian enumerators used Computer Assisted Telephone Interview technology and randomly selected respondents from One-to-One’s sampling frame of landlines, as well as phone numbers from all mobile phone operators in Tunisia. Interviews were solicited proportionally to population size according to the following quotas: governorate, urban/rural, gender, and age.

50 Corstange, “Sensitive Questions”; Glynn, “What Can We Learn.”

51 Eady, “Statistical Analysis.”

52 Blair and Imai, “Statistical Analysis”; Eady, “Statistical Analysis.” These estimators are implemented in the list and misreport packages in R, respectively.

53 Blair and Imai, “Statistical Analysis.”

54 This procedure of course only detects the strongest form of non-monotonicity, namely if respondents in the control group endorse all experimental items (and therefore also the sensitive item), but do not report support for the military in the direct question. Nevertheless, the fact that only 4 respondents fit this profile increases our confidence in our experiment.

55 Li, “Relaxing the No Liars Assumption,” p. 543. We formally test for floor and ceiling effects below.

56 Kramon and Weghorst, “(Mis)Measuring Sensitive Attitudes.”

57 https://data.worldbank.org/indicator/SE.ADT.LITR.ZS?end=2014&locations=TN&start=1984&view=chart.

58 Robustness checks rule out any design effects in the list experiment (p = 1). The experiment was placed ten questions before the direct question, with the treatment condition having no priming effect (see Aronow et al., “Combining List Experiment”) on the direct question (p = 0.46). There are possible floor effects (see Ahlquist, “List Experiment Design”) in this survey, for which we correct in survey 2.

59 Blair and Imai, “Statistical Analysis.”

60 Results are nearly identical for the 2014 presidential elections, but we choose to present the parliamentary ones as they give us a direct measure of support for the Islamist party Ennahda. In the presidential elections, Ennahda did not endorse either secular candidate.

61 Eady, “Statistical Analysis.”

62 These include the governorates of Tunis, Sousse, Medenine, Kef, and Kasserine. The first two suffered terrorist attacks in 2015, the third in 2016, and the latter two have seen a small-scale insurgency.

63 Blair, Coppock, and Moor, “When to Worry.”

64 Holbrook, Green, and Krosnick, “Telephone versus Face-to-Face.”

65 These candidates and parties included: Abdelfattah Mourou (Ennahda), Youssef Chahed (Tahya Tounes), Mohsen Marzouk (Machrou Tounes), Abdelkrim Zbidi (Nidaa Tounes), Selma Elloumi Rekik (Amal Tounes), Slim Riahi (UPL), Mehdi Jomaa (Badil Ettounsi), Neji Jalloul, and Afek Tounes.

66 The second survey did not ask about support for the 2011 revolution, or marital status.

67 Foa and Mounk, “Danger of Deconsolidation.”

68 Schake and Mattis, Warriors and Citizens; Karlin and Friend, “Military Worship.”

69 Lotito, “Public Trust.”

70 Grewal and Kureshi, “How to Sell a Coup.”

71 Yee, “Tunisia’s Democracy”; Albrecht et al., “Popular Support.”