Initially, these actions reflected broad support among Tunisians unhappy with the performance of the country’s elected leaders, especially the Islamist Ennahda party. Yet since President Saied assumed primary responsibility for governing by shutting down the elected parliament in 2021, the country has plunged into ever-increasing economic difficulties. These troubles have affected both young and old, rich and poor, and they could be cutting into President Saied’s popularity.
Recently, numerous political and media figures have been put on trial for making statements critical of the state or the military. When Tunisians see these trials, they may conclude that they should avoid sharing criticism of the government with others.
As a result, learning the true level of opposition to Saied’s regime has become more difficult because people are afraid to answer questions from pollsters.
To account for this bias, Amr Yakout and Robert Kubinec fielded an online survey over two weeks with an encryption technique for anonymizing responses, and their data shows that–for the first time–at least 50 percent of Tunisians oppose Saied’s shutdown of the parliament. Furthermore, they estimate that approximately 20 percent of the population is afraid to report their opposition when asked directly–a remarkably large number as it suggests one in five Tunisians are unwilling to give their true opinion about President Saied’s performance.
The Challenges of Polling Under an Authoritarian Regime
According to the researchers, polling people about their political opinions is a difficult task even in established democracies. Reaching the entire population and designing questions that capture diverse opinions is tricky, especially in a world where people communicate through smartphones. When people develop a fear that there also might be others who want to spy on their opinions–whether government agencies or others who live around them–accurately capturing opinions becomes even more difficult.
Up until Saied’s coup, polling companies in Tunisia had grown considerably, taking on more ambitious projects such as regular longitudinal* polls of approval of the government. This type of information is essential in a democracy so that politicians can understand how people are responding to their policies and proposals.
At the same time, by Saied’s shutdown of parliament in the summer of 2021, Tunisian polling companies had approximately ten years of experience since 2011. Little research had been done to evaluate polling quality and know whether existing polling strategies were capturing population opinion accurately.
For these reasons, knowing exactly how much Saied’s popularity has fallen since the coup is no easy task. There have been rumors that Saied’s popularity has fallen in the past year, but precious little evidence of it.
A household survey in early 2022 found that almost 90 percent of Tunisians approved of Saied’s decision to close the parliament. One of the few polling agencies still in operation shows that his popularity declined to about 60 percent as of June of this year, but the numbers show a surprisingly enduring majority in Saied’s favor.
Crucially, all of these polls use direct questions asking respondents what they think about Saied, and cannot offer anonymity because they are completed over the telephone or face to face. As the ongoing speculation about Putin’s approval rating since the outbreak of the Russian-Ukraine war has made clear, interpreting polls under authoritarian regimes is a difficult endeavor.
According to the researchers, knowing how many people oppose or support Saied’s government is a crucial piece of information because it affects Tunisians’ political decisions.
If Saied has become more unpopular, then he is also more vulnerable to criticism and mobilization by his opponents. If Saied can continue to convince Tunisians that he has unassailable popularity, then support for Saied can become a self-fulfilling prophecy: people support him because they believe others do, and so on.
Hacking Repression with a Survey Response Encryption Tool
To obtain accurate opinions about President Kais Saied, it is essential to give people a way of responding without anyone knowing what their opinion is. The solution to this dilemma is a form of encryption which we will inject into the survey design. In a similar way that encrypted messaging apps like WhatsApp make use of random “keys” to secure connections, our survey question made use of the respondent’s private information to allow them to share their opinion about Saied.
A first online survey experiment wa conducted from July 26 to August 18 that enrolled 830 Tunisians from all 24 of the country’s governorates. For half of the sample, we asked them a direct question about Kais Saied’s overthrow of the parliament:
Do you oppose President Kais Saied’s moves to change Tunisia’s constitution and close the parliament?
The possible answers are either yes or no.
This question is expected to be sensitive–some may not want to answer that they oppose Kais Saied because they are concerned that someone could find out their true opinion. Exactly what the risk is–whether they think the survey is being run by the government or some other issue–isn’t as important as the fact that fear exists.
For the other half of the sample, participants’ responses were subjected to a form of encryption, which is known as a randomized response question. This question makes use of naturally-occurring randomness–the birth month of the respondent’s mother–to encrypt the respondents’ answers to the same question about Saied’s actions. Essentially, without knowing the respondent’s mother’s date of birth, it is impossible to decrypt their answers. This question takes the following form:
As long as the respondent reads and follows the instructions, they can report their opposition (or support) for Kais Saied’s power grab without their true response being identifiable. Essentially, we can’t separate their answer from whether their mother’s birthday is in a given month, and as a result, the individual answers are encoded. For any respondent, we have no idea whether they oppose Saied or not.
Even though individual responses are encrypted when using the randomized response technique, it is still possible to estimate the total number of people opposed to the President. To do this, researchers used existing data, which indicated that approximately 25% of people are born in the first quarter of the year. By subtracting away this random noise* and employing statistical models, they were able to obtain an approximate estimate of the number of people opposed to Saied.
Importantly, if some respondents do not trust the technique, the results will be biased in a conservative direction, which ensures that Saied’s opposition is not overestimated.
Saied Only Has At Most 50% of the Country Behind Him
In this section, the authors will go over the results of the poll. They also have released our raw data and code to allow others to verify our conclusions.
For the first direct question about supporting the president’s coup, only 30 percent of respondents report that they oppose the coup. This number is quite low.
It is important to note that no questions were asked about approval of the president’s current job. For that reason, this question may over-state support because some respondents may have backed his coup but are unhappy with his work at present. Again, these figures can be considered conservative in terms of the number of Tunisians who oppose Saied.
Looking at the raw results from the randomized response question, it is impossible to determine whether or not someone is opposed to Saied. It could be that they are opposed to Saied or their mother was born in the first three months of the year (or both of these statements are true or false).
The statistical model described in the technical details allows to determine just how different the responses to the direct question are from the encrypted question.
The survey was conducted in two phases: 50% of participants were exposed only to the direct question, while the other half saw only the encrypted version. The difference is equal to the proportion of people who might be afraid to answer a direct question. This estimate is shown in the graphic below.
The blue plot above shows the range of plausible estimates for the percentage of the Tunisian population that does oppose Saied’s coup but would prefer not to say so directly.
This value is significantly higher than what was initially anticipated by the researchers. In a previous document, they had estimated that 5 to 10% of Tunisian citizens could be influenced and declare their approval of Saied's coup. However, it appears that the most likely proportion is 20%, with a high likelihood that this proportion exceeds 10% and may even reach 30%.
It is important to interpret this number carefully. These people may not be strongly anti-Saied; they could be people who are uncomfortable with Saied but not enough to make their preferences known. For these reasons, it’s better to think of them as a silent minority who feels trapped by the climate of fear and repression. However, it is also true that these people are unhappy with Saied and even his shutdown of parliament in 2021.
Another graphic created from the encrypted question results helps estimate the number of people who oppose Saied’s coup:
The blue area in the plot shows all the plausible estimates for how many people oppose Saied’s coup. Our best estimate is that it is only almost exactly 50 percent, but because the encrypted response question has additional randomness, there are a range of plausible answers. Total opposition could be as low as 40 percent or as high as 60 percent. When we use a technique like randomized response, we get less precise answers in exchange for anonymity and security. However, even with this wide range, we can still say that the number who report opposition is much higher than the direct question without protection.
The blue area in the plot shows all the plausible estimates for how many people oppose Saied’s coup. The results with the randomized response technique, although less precise, ensure participant anonymity and security. The best estimate suggests it is very close to 50%. However, due to the additional randomness of the encrypted response question, there is a range of plausible answers. Total opposition could be as low as 40 percent or as high as 60 percent. Even with this wide range, the number who report opposition is much higher than the direct question without protection indicates.
When comparing the direct question to the encrypted response, it is clear that the reported opposition to Saied (32 percent) is almost twice as low compared to when respondents are given protection (50 percent). Even when potential sampling variation is taken into account, it is very unlikely that the number of those who are afraid to report their opposition is less than 10 percent of the population.
The researchers addressed issues related to using an online sample by adjusting it with Tunisian census data, as detailed in their blog post, and found that this only results in minor differences. The online sample includes participants from all governorates of Tunisia, making it a reasonable representation of the population.
Opposition to Saied varies significantly among Tunisians. One of the most notable distinctions relates to age, with significant differences between young and older individuals:
Distinctions in opposition to Saied by age
This graphic shows the estimated proportion who oppose Saied’s coup by age categories from 15 - 19 up to more than 80 years old. While there is uncertainty–each line represents the range of plausible estimates–it does seem like the young are more opposed to Saied than the old. This would seem to follow from the growing number of reasons young people have to be unhappy with Saied’s leadership given the disastrous collapse of the economy.
Moreover, the study reveals another difference within respondents when distinctions are made based on gender. Male respondents tend to express greater opposition to Saied than their female counterparts with 10% difference between both.
Distinctions in opposition to Saied by gender
In addition to age and gender differences, the research also shows some regional variations. The data portrays relatively small differences between different governorates in Tunisia, with median values ranging from 48.38% (Mahdia) to 52.29% (Ariana) in opposition to Saied's coup. These minor regional disparities suggest a consistent level of opposition across various parts of the country.
Distinctions in opposition to Saied by governorate
The survey evidence shows that Tunisians are increasingly hostile with Saied’s handling of the government, but a significant number are afraid to speak their minds. With economic conditions steadily worsening, the political situation is looking more similar to the end of the regime of the former dictator, Ben Ali, when widespread unhappiness co-existed with a veneer of acquiescence to the security state. As the Arab Spring revolutions and the collapse of the Soviet Union revealed, authoritarian repression can result in sudden political changes that are difficult to foresee. Forecasting political outcomes is always tricky, but it is plausible that Tunisia is in for another round of political upheaval.