The Strategic Use of Fear Appeals in Political Communication

Abstract

Fear appeals constitute a frequent theme of populist rhetoric. One potential motive for this is that they decrease people’s reliance on partisan habits and increase openness to new information. Political actors can use this effect to attract more ideologically distant groups of voters, but not without drawbacks.

This paper analyses the strategic use of fear appeals in the framework of the Bounded-Confidence model. It is shown that attracting undecided voters between two opinion clusters is decisive for the success of a party’s fear appeal strategy. Hence, fear appeals can increase a party’s reach for new supporters, yet only if the party manages to clearly differentiate itself form ideological competitors.

Keywords:

• fear appeals
• affective intelligence theory
• strategic communication
• bounded confidence
• agent-based modelling

Supplementary Material

Supplemental data for this article can be accessed on the publisher’s website at https://doi.org/10.1080/10584609.2019.1631918.

Notes

1. The terms ‘fear’ and ‘anxiety’ are often used synonymously in the literature on political psychology (and also in colloquial language and politics). In the psychological literature more generally, some authors (e.g. Davis, Walker, Miles, and Grillon (2009)) distinguish the two with regards to whether some event is appraised as certain or uncertain. For reasons of complying with the former (and for this paper arguably more relevant) literature, I will use the terms interchangeably, but will refer to an understanding of fear/anxiety as an uncertainty-inducing emotional response, as explained later in more detail.

2. Some authors, e.g. Ladd and Lenz (2008) provide a more critical perspective on AIT’s findings, arguing that affect transfer and endogenous affect formation might also explain some of the results. These are, in turn disputed in Marcus, MacKuen, and Neuman (2011). Further, note that a rich literature on the impact of other discrete emotions (e.g. enthusiasm, anger) exists as well, which cannot be appropriately dealt with here due to reasons of focus (Huddy, Feldman, & Cassese, 2007; Parker & Isbell, 2010; Rico et al., 2017; Salmela & von Scheve, 2017; Steenbergen & Ellis, 2006; Weeks, 2015), but could be fruitfully considered in a consecutive analysis.

3. There are, of course, also limits to extent to which fear or other emotional appeals can be effective, particularly for framed threats that concern societal and political issues (Albertson & Gadarian, 2015). The ability to induce fear is limited for instance by people’s feeling of self-efficacy in a given situation (Witte, 1992; Witte & Allen, 2000), a message’s perceived manipulativeness (Dillard & Shen, 2005), or partisanship (Albertson & Gadarian, 2015).

4. Formally, the model is characterised by $n\in N$ agents with $x_i(t)\in [0,1]$. Further, $x(t)$ denotes the profile of all opinions at time $t$. Agent $i$’s influencing set $I_i$ is given as $I_i({ϵ}_i,x(t))=\{j||x_i(t)-x_j(t)|\le {ϵ}_i\}$. Logically, $I_i$ always includes agent $i$ herself. This implies the following rule for opinion-updating:

$x_i(t+1)=\frac{{\sum }_{j\in I_i}{x}_j(t)}{\mathrm{\#}{I}_i},$

where $\mathrm{\#}{I}_i$ is the number of agents in the influencing set $I_i$.

5. Note that the BC model can explain polarisation only in the narrow sense that two distinct opinion clusters form; it does not explain polarisation in the sense of two clusters moving further apart over time.

6. Analogous considerations apply to similar antagonistic dynamics, e.g. Vasilopoulos et al. (2018a); Redlawsk, Civettini, and Emmerson (2018); Vasilopoulou and Wagner (2017), or interaction effects that would complicate the role of partisanship (MacKuen et al., 2007).

7. See for instance Virag (2008) for a similar model.

8. In a series of robustness analyses, the parameter $\phi$ generalised the party’s strength beyond values of 10. This variation produced qualitatively stable results and expectable limiting cases for extreme parameter values. This robustness analyisis is documented in the online-appendix.

9. Formally, a party $k$’s updating rule for her position $\mathrm{\Phi }$ is: $\mathrm{\Phi }(t+1)=\mathrm{\Phi }(t)=\mathrm{\Phi }$. If $|x_i(t)-\mathrm{\Phi }|\le {ϵ}_i$, the influencing-set of agent $i$ includes the party $k$, and does not include the party if $|x_i(t)-\mathrm{\Phi }|\ge {ϵ}_i$. Thus, the updating-rule for an agent’s opinion is formally expressed as

$x_i(t+1)=\frac{{\sum }_{j\in I_i}{x}_j(t)+\mathrm{\Phi }\times 10\times Ind}{\mathrm{\#}{I}_i+10\times Ind},\mathrm{w}\mathrm{i}\mathrm{t}\mathrm{h}Ind=\left\{\begin{array}{c}1\mathrm{f}\mathrm{o}\mathrm{r}|x_i(t)-\mathrm{\Phi }|\le {ϵ}_i,\\ 0\mathrm{o}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{w}\mathrm{i}\mathrm{s}\mathrm{e}\end{array}.\right.$

10. Appeals that aim at reducing agents’ confidence bounds could analogously be interpreted as appeals to anger or enthusiasm. For reasons of space, they are not explicitly considered in this paper.

11. Technically, confidence bounds of regular agents are updated according to

${ϵ}_i(t+1)=\frac{{\sum }_{j\in I_i}{ϵ}_j(t)+\mathrm{\Pi }\times 10\times \mathrm{I}\mathrm{n}\mathrm{d}}{\mathrm{\#}{I}_i+10\times \mathrm{I}\mathrm{n}\mathrm{d}}$

A party’s emotional appeal is assumed to be constant during the course of a model run. Further, note that a somewhat similar dynamic process has also been studied in (Deffuant, Amblard, Weisbuch, & Faure, 2002).

12. In Figure 2, the party did not yet exert any influence on $ϵ$, hence desired and actual sphere of influence were trivially congruent.

13. Figure S1 in the appendix provides data for larger parameter ranges.

14. For a more detailed justification of the equidistant distribution, see Hegselmann and Krause (2015). Note that Lorenz (2006) identifies certain drawbacks of using this distribution, particularly that its symmetry artificially increases convergence times. Despite those issues (and also because they are not directly relevant to this analysis), the equidistant distribution is chosen for this analysis because it most easily allows us to study the individual model runs that produced a certain outcome. With averaging done over many runs with different random values, one would have to stumble across these tipping phenomena coincidentally. A robustness analysis with several random initial distributions that were fixed over the whole parameter space showed that the qualitative results described below also occur. Some examples can be found in the online-appendix, where also other modifications of $x(0)$ are considered.

15. Robustness checks have been carried out for various values of $ϵ(0)$, showing that the substential findings persist over the relevant parameter range. See online-appendix.

16. Practically, a model run stops as soon as agent movements are all smaller than $0.001$ per round. Every voter no further than $0.1$ from the party’s position is considered a follower of the party.

17. All results presented in this section are qualitatively robust against variation in the parameters $ϵ(0)$ and $\phi$ and only become overlain by other effects when the parameters take exteme values. These robustness checks are documented in the separate online-appendix.

18. The analysis deliberately excludes more moderate positions in the analysis due to the initially prescribed focus on extreme parties. As Figure S1 in the appendix shows, the pattern repeats itself in that parameter region. So do the underlying causal mechanisms.

Additional information

Funding

This work was supported by the German Research Foundation and the Humboldt-Foundation.

Notes on contributors

Simon Scheller

Simon Scheller studied philosophy, economics and political science in Bayreuth, Leeds and Bamberg. He holds a PhD in political science from Bamberg University and is currently a postdoctoral researcher at the Munich Center for Mathematical Philosophy.