The previous section established that conditions for the NVC model could obtain in Poland. However, I have yet to establish that this model is actually better at explaining Poland’s backsliding trajectory than Svolik’s theory, which—recall—does not rely on uncertainty about the incumbent’s true identity. In this section, I adjudicate between these two models further by developing two contrasting hypotheses about the determinants of voting for PiS. Both models predict that voting for PiS ought to increase with polarization. However, in addition to elite polarization, the NVC requires polarization in the electorate and a negative effect of beliefs that the policies implemented by PiS are those of a closet autocrat. Specifically, voting for PiS is more likely when respondents believe that judicial reforms are none other than the continuation of a TJ program that should have reached the judiciary. Relatedly, while both models underscore polarization as a leading cause for reelecting the autocrat, the latter model is compatible with voters exhibiting a strong aversion to authoritarian rule.
I will first present evidence in support of both models. These first data come from historical surveys and illustrate increasing levels of polarization between political elites in Poland in the twenty first century. This can be used to support either the Svolik or the NVC model. Following this, I provide evidence from a contemporary survey that allows for adjudication between the two theories by showing that respondents’ willingness to reelect closet authoritarians impersonating ideological conservatives decreases with their beliefs that the incumbent is indeed a closet autocrat.
Elite polarization in historical surveys
The measuring of elite polarization is possible thanks to reliable public opinion polls that have been conducted according to a uniform sampling technique between 2001 and 2011. The surveys come from the Center of Public Opinion Research, CBOS, a public opinion polling company with more than 35 years of experience in conducting surveys.Footnote 9 It fields regular surveys on stratified samples representative of Polish citizens between the ages of 18 and 75. It uses face-to-face interviews and has consistently included in its questionnaire the same battery of questions pertaining to political preferences and trust of political institutions. The question I use from the historical surveys is “Persons in public life—in their behavior, in what they say, and what they try to achieve—can arouse more or less trust. We will show you now a list of persons active in public life and ask you to what extent this person is trustworthy. − 5 means that you feel great distrust towards this person, 0 means you are indifferent and 5 means that you trust them completely. And of course, you can make use of any numbers between − 5 and 5 to express your trust towards this person. Please let us know if you do not know this person.”
In addition, based on current political events, CBOS supplements the regularly-asked questions with additional ones (I will make use of these surveys in the next section of this article). The period 2001–2011 coincides with the time when PiS and PO—the two largest parties in Poland—alternated in government and were led by the same two leaders. Both parties are descendants of Electoral Action Solidarity (AWS), the umbrella party organization uniting former anti-communist dissidents with roots in the independent Solidarity trade union.Footnote 10 Beginning in 2003, the main cleavage dividing Polish voters was no longer allegiance to the former communist autocrats or their opposition, but a more classical conservative-liberal cleavage (Carroll and Nalepa 2019). At that time, the two parties started separating ideologically.
To measure elite polarization, I use CBOS’s feeling thermometer question, described above, gauging respondents’ sympathy towards Kaczynski and Tusk, the leaders of PiS and PO, respectively. The formula describing the polarization measure is
$$\text{Elite Polarization}_{t} = \mathop \sum \limits_{it}^{n} \frac{{\left| {TrustPO_{it} - {\text{TrustPiS}}_{it} } \right|}}{n} .$$
(1)
TrustPO represents trust towards the leader of PO, Donald Tusk (and in 2017, Grzegorz Schetyna); TrustPiS is the trust towards the leader of PiS; n is the total number of respondents.Footnote 11 The absolute value is necessary because we are interested simply in the absolute distance between the two leaders as perceived by individual respondents. If the absolute value is not taken before distances are averaged, differences may cancel each other out. For instance, if one respondent expresses highest sympathy for Kaczynski and lowest sympathy for Tusk, while another respondent expresses highest sympathy for Tusk and lowest for Kaczynski, the mean of their differentials would be zero without the use of the absolute value. Yet, such preferences are consistent with the highest elite polarization possible as measured by these questions.
Measuring polarization as the mean differential between sympathy for each of the primary party leaders is a slight departure from the way polarization is measured in the American Politics context, where individual respondents are asked for their views on specific issues and the responses of Democrats and Republicans are compared (see Fiorina and Abrams 2008). In this context, the main question is the extent to which polarization is explained by partisan sorting. I depart from this because I want to pin down how distant political elites are from one another in the eyes of the public. This measure is closer to the concept of voters’ perceptions from both Svolik and NVC.
Figure 2, below, shows the Elite Polarization (marked as “diff_mean”) trend starting in 2001 with the 92nd survey conducted by CBOS and ends in 2011 with the 165th survey. The figure shows an increase in polarization around the 140th survey (in 2005). This is the year PiS won a plurality in the parliamentary elections for the first time (and entered into a cabinet coalition with two other parties: Samoobrona and the League for Polish Families). The increase continues until 2011, the year when PO took over the reins of government, a situation that persisted until 2015. Elite polarization in the contemporary survey discussed in the next section is 4.707 with a standard deviation of 3.21, indicating that current elite polarization surpasses its highest value from the 2001–2011 period.
This result suggests that one of the conditions for the NVC equilibrium is satisfied in the Polish case. However, this evidence is also consistent with other models of authoritarian backsliding, such as that of Svolik (2020). To find more direct support for NVS, we need to look beyond increases in elite polarization over time and establish conditions for obtaining the equilibrium in which voters reelect incumbents because they are uncertain whether they are dealing with an ideologue or with a closet autocrat.
Polarization in the electorate: contemporary surveys
For this further evidence, I turn to a contemporary CBOS survey from August 2017, the month directly following a presidential veto over the bills proposing the politicization of the National Council of the Judiciary and the Supreme Court. Recall that the Supreme Court is the highest court of appeal and for all practical purposes overlaps with the Supreme Court of the US except for in verifying the constitutionality of legislation. The National Council of the Judiciary, on the other hand, is the sole advisory body nominating candidates for judges and initiating disciplinary action against judges.
The survey I use here was conducted within days of intense and nationwide protests that prompted the President’s decision to veto two bills that would have brought the judiciary under complete political control.
Although observational data from surveys does not allow me to establish that uncertainty about the kind of incumbent voters are facing causes voters to support the incumbent, I show evidence that the association between uncertainty and incumbent voting increases with anti-authoritarian attitudes, exactly as the NVC model would lead us to expect.
In addition to predicting this effect of beliefs, the NVC model leads to expectations that respondents perceiving elite polarization to be higher and respondents who are further away from the opposition candidate are likely to vote for the incumbent, even when their anti-authoritarian predisposition is high. I quantify Subjectively Perceived Elite Polarization (SPEP) with a measure related to the one described in Eq. (1), but I do not average the measure across respondents. Hence:
$$\text{SPEP}_{i} = |TrustPO_{i} - {\text{TrustPiS}}_{i}| .$$
(2)
As before, I take the absolute value of the differential between trust in PiS leader Kaczynski and in the current PO leader (who is Schetyna and no longer Tusk). The index i indicates that this polarization is measured at the level of each respondent. The index over time is dropped because I now only deal with one survey.
Finally, I need to measure polarization in the electorate—that is, the distance between the citizen and the opposition candidate. This is best captured with simply the citizen’s sympathy toward Schetyna, the PO leader:
$$\text{Polarization in the Electorate}_{i} = TrustPO_{i}$$
(3)
Next, to measure citizens’ beliefs as to whether they are facing a closet autocrat or an ideological conservative, I use respondents’ expressed attitudes on the July 2017 protests; these protests in defense of the National Council of Judiciary and Supreme Court independence led the President to veto the two controversial bills. Since the protesters argued that both reforms violated the constitution, a respondent’s support for the protest can serve as an indicator of a belief that the incumbent is a closet authoritarian. This variable is labeled Support_Protestsi
Finally, in order to measure how much citizens care for the authoritarian dimension, I use the degree to which respondents agree with the following four questions gauging respondents’ sensitivity to authoritarianism. The four questions that measure the respondents’ sensitivity to the authoritarian dimension asked the respondent about the extent to which he or she agreed with the following four separate statements.
-
1.
“Democracy is superior to any other form of rule” (variable labeled Authoritarian Ii)
-
2.
“For people like me, it does not matter whether the regime is authoritarian or democratic” (variable labeled Indifferenti)
-
3.
“Sometimes Non-democratic rule is better than democratic rule” (variable labeled Authoritarian IIi)
-
4.
“Government by a strong leader is decidedly better than democratic rule” (variable labeled- Strong Leaderi)
Respondents could “agree strongly,” “agree somewhat,” “rather disagree,” or “strongly disagree” with the above statements. The answers to the first of the four questions were recoded to match the other three, so that that all four answers increase with pro-democratic values. Higher values of these variables represent stronger disagreement and therefore more pro-democratic values.
It is important to point out that these questions are not simply picking up on sympathy for the opposition. A correlation table in the Appendix shows that the association of these measures with Polarization in the Electorate is low.
In the results reported below, I use an interaction term between Support_Protests and Polarization in the Electorate to reflect the fact that voting for the incumbent in equilibrium of the NVC model requires both sufficiently high polarization of the electorate and sufficiently low beliefs that the incumbent is a closet autocrat. The dependent variable is a dummy, PiSvoteri, indicating that a respondent i answered “PiS” to the question “Were the elections to the Sejm to take place this Sunday, which party would you vote for?” I used the question “Would you vote in the parliamentary elections were they to take place this Sunday?” to filter out non-voters. Thus, the regressions are run only on voters and the sample is slightly smaller than the representative sample of Poles between 18 and 75.
Table 1 presents a regression of PiSvoteri on this set of independent variables obtained from the survey along with demographics such as age, gender, and education.
Table 1 Predictors of PiSvoting corroborating the NVC model (logit models) Because of the dichotomous nature of my dependent variable, I chose a nonlinear probability model—specifically, a logit in the following format:
$$\text{Pr}(PiSvoter_{i} = 1) = \frac{1}{{1 + e^{{ - x_{i} \beta }} }}$$
$$\begin{aligned} \text{{where} } \mathbf{x_{i} \beta} = \beta_{1} TrustPO_{i} + \beta_{2} SPEP_{i} + \beta_{3} {\text{PiS}}\_{\text{Protest}}_{i} + \beta_{4} SPEP_{i} *{\text{PiS}}\_{\text{Protest}}_{i} \\ \quad + \beta_{5} Authoritarian I_{i} + \beta_{6} {\text{Indifferent}}_{i} + \beta_{7} Authoritarian II_{i} + \beta_{8} {\text{Strong Leader}}_{i} \\ \quad + \beta_{J} Demographics \\ \end{aligned}$$
“Demographics” includes gender, employment status, urbanization, education, and religiosity.Footnote 12
I expect the effect of Subjective Perception of Elite Polarization and Polarization in the Electorate to be positive and \(SPEP_{i} *{\text{PiS}}\_{\text{Protest}}_{i}\) to be negative (and more negative for higher values of Polarization in the Electorate).
Results of four models run using different operationalizations of anti-authoritarian attitudes are presented in Table 1 below.
The effects of Polarization in the Electorate and SPEP are significant and in the expected direction, increasing the probability of voting for PiS. The individual effect of beliefs that the incumbent is a closet autocrat is negative. What is of interest, however, is the combined effect of beliefs and Polarization in the Electorate on the probability of voting for PiS. We see that while one of the constituent terms is negative, the other is positive. The optimal way to visualize these effects is by graphing the average effect of beliefs for different levels of polarization.
Figure 3 graphs the marginal effect of citizen beliefs that they are dealing with a closet autocrat (measured by PiSProtest) for different levels of Polarization in the Electorate (measured by TrustPO). It clearly shows that the interaction is negative. An increase in the belief that the voter is facing a closet autocrat decreases the probability of reelection for all levels of polarization in the electorate, but particularly for low levels of polarization. Concretely, when polarization is low, the average effect of the belief that the incumbent is a closet authoritarian is a decrease in the probability of reelection by about 52%, but when polarization is high, the effect is only about 20%.
The expectations for the effect of sensitivity to authoritarian rule (Authoritarian I, Authoritarian II, Indifferent and Strong Leader) is harder to interpret. Therefore, in order to illustrate the effect of this sensitivity, in the context of the survey at hand, I proceed as follows. First, I split the sample into respondents who answered the sensitivity questions positively (yes and rather yes → High) and those who answered the sensitivity questions negatively (no and rather no → Low). The results of these split sample tests are presented in Table 2 below.
Table 2 Logit regressions of PiS voting on split samples according to high or low democratic attachments In line with expectations, the effects of anti-authoritarian sensitivity (Antiauthoritarian2, Strong Leader, and Indifferent) are significant and in the expected direction in two out of three of the models representing citizens who are deeply anti-authoritarian, but entirely insignificant and/or in the wrong direction in all models representing citizens who weight the authoritarian dimension lightly. Furthermore, the substantive effects of these variables are higher than in the pooled sample from Table 1. In addition, that the remaining effects of the parameters remain stable in the models where respondents assign heavy authoritarian weight adds confidence to my findings.