1. Bibliographic Information

1.1. Title

Does anxiety increase policy learning?

1.2. Authors

• Moulay Lablih (Swiss Graduate School of Public Administration (IDHEAP), University of Lausanne, Lausanne, Switzerland)
• Pirmin Bundi (Swiss Graduate School of Public Administration (IDHEAP), University of Lausanne, Lausanne, Switzerland)
• Lea Portmann (Interface Politikstudien, Switzerland)

1.3. Journal/Conference

Policy Studies Journal. This journal is a peer-reviewed academic publication covering a wide range of topics related to public policy. It is generally well-regarded in the field of public policy and political science, serving as a platform for scholarly research, theoretical developments, and empirical analysis.

1.4. Publication Year

2024

1.5. Abstract

The paper investigates the effect of anxiety on policy learning among local public officials, building on the idea that a lack of policy learning contributes to policy failures. While prior research often focuses on socioeconomic and partisan factors, this study highlights the role of emotions, specifically anxiety, in information processing. It introduces Marcus' Affective Intelligence Model (AIM) to the policy learning literature. Using a novel experimental design with Swiss local elected officials, where anxiety-inducing images are randomly displayed alongside policy information, the study finds that anxiety has a positive causal effect on learning. While the relationship is not significantly moderated by the strength of priors or the perceived complexity of policies, these variables are found to be substantially correlated with policy learning. The findings have significant implications for understanding how information influences policymaking.

1.6. Original Source Link

/files/papers/694f95da9c764da3f20e36db/paper.pdf (Published at 2024-02-23T00:00:00.000Z)

2. Executive Summary

2.1. Background & Motivation

The core problem the paper addresses is the lack of policy learning among decision-makers, which is often cited as a reason for policy failures. Traditional research in public policy has primarily focused on socioeconomic and partisan variables to explain how public officials process policy information. However, this approach has largely overlooked the role of emotions, such as anxiety, in influencing this critical process.

This problem is important because effective policymaking relies on decision-makers' ability to update their beliefs and adapt policies based on new information and evidence, a process known as policy learning. If key emotional factors are ignored, our understanding of policymaking remains incomplete, potentially leading to less effective governance. The existing research gap concerns the individual-level psychological mechanisms that influence how public officials engage with and learn from policy information.

The paper's innovative idea and entry point is to explicitly investigate the causal effect of anxiety on policy learning among local office holders. It does this by integrating Marcus' Affective Intelligence Model (AIM)—a prominent theory from political psychology about how emotions affect information processing—into the policy learning literature. This integration posits that anxiety, as a response to perceived threat or novelty, should enhance systematic information processing and, consequently, learning.

2.2. Main Contributions / Findings

The paper makes several primary contributions:

• Causal Effect of Anxiety: It provides empirical evidence through a novel survey experiment that anxiety has a positive causal effect on policy learning at the individual level among local policymakers. This is a significant finding, as it directly demonstrates how an emotion can influence a crucial cognitive process in policymaking.

• Application of AIM to Policy Learning: The study successfully introduces and applies Marcus' Affective Intelligence Model to the field of policy learning, bridging political psychology with public policy research. This opens new avenues for understanding the microfoundations of policymaking.

• Role of Prior Beliefs: It confirms that the strength of prior beliefs is negatively correlated with policy learning. Policymakers with stronger priors are less likely to update their beliefs, which is consistent with Bayesian updating principles where strong priors require more overwhelming evidence to change.

• Role of Perceived Policy Complexity: The research reveals that perceived policy complexity is also negatively correlated with policy learning. Counter-intuitively, policies perceived as more complex lead to less learning, suggesting a backlash effect or reliance on heuristics in complex situations, rather than increased systematic processing.

• Moderation Effects: Importantly, the study found that neither the strength of priors nor perceived policy complexity significantly moderated the causal effect of anxiety on policy learning, despite initial hypotheses. However, these variables are still substantially correlated with learning outcomes.

• Observational Insights: Beyond the experimental findings, observational analyses show that a right-wing ideology is negatively correlated with policy learning, while having a university degree is positively correlated.

  These findings solve the problem of understanding the micro-level psychological mechanisms that drive or hinder policy learning. By identifying anxiety as a motivator for learning and confirming the resistance posed by strong priors and perceived complexity, the paper helps explain why decision-makers sometimes fail to adapt to new information, ultimately contributing to a better understanding of policy failures and policymaking processes.

3. Prerequisite Knowledge & Related Work

3.1. Foundational Concepts

To fully understand this paper, several core concepts are essential:

• Policy Learning: In the context of this paper, policy learning is defined as "the acquisition of new relevant information that allows beliefs to be updated" (Braun & Gilardi, 2006). It fundamentally involves a belief change process, where individuals modify their existing views or knowledge in response to new information. This process is often conceptualized through Bayesian decision theory.

  • Bayesian Decision Theory: This is a statistical framework for making decisions under uncertainty. In Bayesian learning, an individual starts with prior beliefs (initial probability distribution over possible states of the world) and then updates these beliefs to posterior beliefs (updated probability distribution) after observing new evidence or information. The degree to which one's beliefs change depends on both the strength of the prior beliefs and the persuasiveness of the new information.
  • Prior Beliefs: These are the initial beliefs or knowledge an individual holds about a particular policy issue before encountering new information.
  • Posterior Beliefs: These are the updated beliefs an individual holds after processing new information. Policy learning occurs when there is a measurable difference between prior beliefs and posterior beliefs in the direction suggested by the new information.
  • Belief Updating: The process of moving from prior beliefs to posterior beliefs based on new evidence.

• Emotions in Politics: This refers to the study of how various emotional states (e.g., anxiety, anger, enthusiasm, fear) influence political attitudes, behaviors, and decision-making. Historically, political science has focused more on rational choice, but increasingly, emotions are recognized as integral to political cognition and action.

• Marcus' Affective Intelligence Model (AIM): This is a prominent two-channel theory in political psychology that explains how emotions affect individuals' information processing.

  • Two Channels:
    1. Surveillance System (Novelty/Threat): This channel monitors the environment for novelty or threat. When individuals encounter unfamiliar or potentially dangerous situations, this system is activated, triggering emotions like anxiety.
    2. Dispositional System (Familiarity/Routine): This channel assesses the appropriateness of familiar actions, routines, or practices based on their past success or failure. It triggers emotions like enthusiasm (for success) or depression (for failure).
  • Anxiety's Role in AIM: According to AIM, anxiety (triggered by the surveillance system when a threat or novelty is perceived) plays a crucial role in information processing. It prompts individuals to disengage from habitual behaviors and engage in more systematic, in-depth processing of new information. This increased attention and deeper analysis can lead to political learning and a greater openness to updating beliefs. Conversely, enthusiasm is associated with reliance on prior beliefs and habitual actions.

3.2. Previous Works

The paper contextualizes its research within several streams of literature:

• Policy Learning Literature: The concept of policy learning has been defined in various ways (Bennett & Howlett, 1992; Hall, 1993; Heclo, 2010), but most definitions revolve around belief change and updating knowledge. Nowlin (2020) suggests that policy learning is a function of the strength of prior beliefs. Dunlop & Radaelli (2017) emphasize coherent updating of prior beliefs when faced with new information. The literature has also identified different types of learning (Dunlop & Radaelli, 2017, 2018) and examined policy diffusion (Gilardi, 2010; Graham et al., 2013) and policy transfer (Dolowitz, 2018).
• Role of Emotions in Politics: While largely overlooked in policy learning, recent studies highlight the importance of emotions in political campaigns (Boussalis et al., 2021; Nai et al., 2017) and political decisions (Colombo, 2021; Suiter et al., 2020). Specific emotions like anger and anxiety have been shown to be relevant in the policy context (Maor & Capelos, 2023; Pierce, 2021). Durnová (2015) investigates emotions in policy deliberation.
• Policy Complexity Literature: Scholars have noted that policy domains differ in complexity (Hurka et al., 2022; Limberg et al., 2021). Complex policies pose greater evaluation challenges and require more resources and expertise (Adam et al., 2018). This distinction between technically complex (e.g., energy, health) and less technical issues (e.g., security, unemployment) is relevant (Eshbaugh-Soha, 2006; Gormley Jr, 1983).
• Cognitive Biases: The paper implicitly refers to cognitive conservatism, where prior beliefs are overweighted, making updating difficult. Bullock (2009) demonstrated that weak priors are more easily overturned than strong priors. Tversky and Kahneman (1974) explain how heuristics are used in complex situations due to uncertainty.
• Political Theories of Change: The paper connects its findings to existing theories like the Advocacy Coalition Framework (ACF) (Weible & Sabatier, 2006), which emphasizes belief systems and learning in policy change; Policy Feedback Theory (Mettler & SoRelle, 2018), which suggests existing policies shape receptiveness to new information; and the Multiple Streams Framework (Blum, 2018; Fowler, 2022), which considers how perceptions of complexity influence the convergence of problem, policy, and political streams.

3.3. Technological Evolution

The field of policy studies has evolved from focusing primarily on structural factors (e.g., socioeconomic conditions, institutional arrangements) and rational-actor models to increasingly incorporate psychological insights and individual-level variables. Early work on policy learning often examined institutional learning or organizational learning. More recently, there's been a shift towards understanding the microfoundations of policy processes, including how individual policymakers process information and make decisions. This has led to the integration of concepts from cognitive psychology and political psychology.

This paper's work fits into this timeline by explicitly moving beyond traditional socioeconomic and partisan explanations to investigate the causal role of emotions in policy learning at the individual level, using experimental methods to isolate effects. This represents a more nuanced and psychologically informed approach to understanding policy dynamics.

3.4. Differentiation Analysis

Compared to the main methods in related work, this paper's approach offers several core differences and innovations:

• Focus on Emotions: Unlike much of the policy learning literature that emphasizes socioeconomic, partisan, or institutional factors, this paper uniquely focuses on the causal role of a specific emotion (anxiety).
• Experimental Design for Causality: Many studies infer relationships through observational data. This paper employs a controlled survey experiment with randomized treatment (anxiety-inducing images), allowing for the identification of a causal effect of anxiety on learning, which is a stronger claim than mere correlation.
• Application of AIM to Policy Learning: The explicit integration and empirical testing of Marcus' Affective Intelligence Model within the policy learning framework is a significant theoretical contribution, bridging two previously distinct academic domains.
• Individual-Level Analysis on Policymakers: The study directly surveys and experiments with local elected officials, providing insights into the psychological processes of actual policymakers, rather than relying on student samples or aggregated data.
• Measuring Belief Updating: The operationalization of policy learning as the difference between prior and posterior beliefs (Bayesian updating) provides a concrete and measurable dependent variable for individual learning.
• Moderation Testing: The paper specifically tests for moderating effects of strength of priors and perceived complexity, addressing theoretical expectations from both Bayesian learning and policy studies literature on how these factors might influence anxiety's impact.

4. Methodology

4.1. Principles

The core idea of the method used in this paper is to investigate the causal effect of anxiety on policy learning among local public officials by experimentally manipulating anxiety levels and observing subsequent belief changes. The theoretical basis for this approach is Marcus' Affective Intelligence Model (AIM), which posits that anxiety, as a response to threat or novelty, enhances systematic information processing and learning by prompting individuals to disengage from habitual behaviors and engage in deeper analysis.

The intuition behind this is that when policymakers feel anxious about a situation or a policy issue, they become more alert and motivated to understand the situation thoroughly. This heightened state of alert should lead them to pay more attention to new information, scrutinize it more carefully, and consequently, be more willing to update their existing beliefs in light of this new evidence. The study operationalizes policy learning as Bayesian updating, where a change from prior beliefs to posterior beliefs (in the direction of scientific evidence) indicates learning. The experimental design allows for isolating the effect of anxiety by comparing treatment groups (exposed to anxiety-inducing stimuli) with control groups (not exposed to such stimuli), while all groups receive the same policy information.

4.2. Core Methodology In-depth (Layer by Layer)

The study employed a survey experiment with a between-subjects design to test its hypotheses.

4.2.1. Participants and Recruitment

• Target Population: Local office holders in Switzerland.
• Recruitment: 3600 elected office holders from municipalities with more than 5000 inhabitants were invited via email to participate in an online survey.
• Response Rate: 39% of invited officials responded, resulting in a sample of 1404 participants.
• Inclusion Criterion: Participants had to be an incumbent member of a Swiss municipal government.
• Ethical Approval: The study received ethical approval from the Research Ethics Commission of the authors' university on April 28, 2022. It was also pre-registered.

4.2.2. Experimental Design

The experiment utilized a between-subjects design with block randomization to ensure balance.

• Blocking Variable: The spoken language (German, French, Italian) in Switzerland was used as a blocking variable, acknowledging its role as a crucial indicator of political culture and its potential influence on attitudes and information processing.
• Random Assignment: Within each language block, participants were randomly assigned to one of two treatment groups or two control groups.
  • Policy Assignment: Both treated and control units were further randomly assigned to one of two policy issues: environmental policy or security policy.

4.2.3. Treatment Manipulation (Anxiety Induction)

The anxiety treatment was administered through the display of anxiety-inducing images.

• Anxiety Variable: The anxiety variable was binary: 1 if the subject received the treatment (image displayed), 0 if not (control group).
• Image Source: Images were selected from the Socio-Moral Image Database (SMID) (Crone et al., 2018), a database specifically calibrated to elicit specific emotional responses.
• Image Selection Criteria: Images were chosen to fit the definition of anxiety, characterized by negative valence and high arousal.
  • "Security" Image: Had a valence score of 1.71 (out of 5, indicating negative tone) and an arousal score of 4.29 (out of 5, indicating high activation).
  • "Environment" Image: Had a valence score of 1.588 (negative tone) and an arousal score of 3.088 (reasonable activation).
• Treatment Validation: Treatment validation checks were conducted within the survey to assess the valence and arousal levels elicited by the images, confirming their effectiveness in inducing anxiety.

4.2.4. Policy Information Exposure

All respondents, regardless of their treatment or control group assignment, were exposed to scientific evidence related to the policy issue (environment or security) to which they were assigned.

• Information Source: The scientific information was derived from reputable scientific journals (e.g., Nature Communications, Policy Practice and Research).
• Information Characteristics: The evidence was chosen to provide a clear direction on what scientific knowledge suggests, facilitating belief updating. Efforts were made to standardize the amount of information presented across policy issues by rephrasing or summarizing texts.
• Fictitious Policy Proposals: Each piece of scientific information was paired with a fictitious policy proposal consistent with the scientific findings (see Appendix A in the original paper for examples).
  • Environmental Policy Information: Focused on the need to accelerate the phase-out of carbon-intensive infrastructure to meet the 1.5°C target.
    • Policy proposal: "We should be moving away from carbon-intensive infrastructure much faster."
  • Security Policy Information: Highlighted that public surveillance camera systems reduce crime.
    • Policy proposal: "The government should increase the number of cameras in public places."

4.2.5. Outcome Variable: Policy Learning

The dependent variable, policy learning, was constructed by measuring the variation between prior and posterior beliefs.

• Prior Beliefs ($t_{-1}$): Before the anxiety treatment ($t_0$) and exposure to scientific information, respondents were asked to rate their agreement/disagreement (on a scale of 0-10) with a statement about the assigned policy area.
• Posterior Beliefs ($t_{1}$): After the anxiety treatment and exposure to scientific information, respondents were asked the same question again.
• Calculation: Policy Learning is calculated as the direct difference: $ \text{Policy Learning} = \text{Posterior beliefs} - \text{Prior beliefs} $ A positive value indicates learning (updating beliefs towards the scientific evidence), while a negative value or zero indicates no learning or a backlash effect (updating beliefs contrary to the evidence).

4.2.6. Other Variables (Moderators and Controls)

• Strength of Prior Beliefs: Measured by asking respondents how knowledgeable they considered themselves to be (on a scale from 0 to 10) about the assigned policy. The assumption is that higher self-rated knowledge correlates with stronger priors. This was measured early in the survey to limit overreporting.
• Perceived Complexity: Measured by asking policymakers to rate the degree of complexity (on a scale from 0 to 10) of the environmental or security policies. This captures their subjective perception of complexity, not an objective measure.
• Ideology: Measured by asking officials to position themselves on a left-right ideological scale.
• Sociodemographic Variables: Age, gender, and education level (e.g., university degree) were collected as control variables for observational analyses.

4.2.7. Methodology for Hypothesis Testing

The study primarily uses linear regression to test the hypotheses.

• Hypothesis 1 (H1): Anxiety increases the extent of policy learning.

  • Estimator: Linear regression.
  • Estimand: Average Treatment Effect (ATE).
  • Specification: Regression model includes Anxiety as the main independent variable.

• Hypothesis 2 (H2): The effect of anxiety on policy learning is stronger when office holders exhibit weak priors.

  • Estimator: Linear regression.
  • Estimand: Conditional Average Treatment Effect (CATE).
  • Specification: Regression model includes an interaction term between Anxiety and Strength of Prior Beliefs.

• Hypothesis 3 (H3): The effect of anxiety on policy learning is stronger when office holders perceive a policy as complex.

  • Estimator: Linear regression.
  • Estimand: Conditional Average Treatment Effect (CATE).
  • Specification: Regression model includes an interaction term between Anxiety and Complexity.

• SUTVA (Stable Unit Treatment Value Assumption): The paper notes that SUTVA (which assumes the treatment of one unit does not affect another, and there's only one version of treatment) is not tested in the case of H2 and H3 because these hypotheses propose conditional effects, implying the treatment effect depends on other factors. This makes the ATE less informative than the CATE.

  The following are the results from Table 1 of the original paper:

  Hypothesis	Estimator	Estimand	Specification
  H	Linear regression	ATE	Anxiety
  H2	Linear regression	CATE	Anxiety × Strength of Prior Beliefs
  H3	Linear regression	CATE	Anxiety × Complexity

5. Experimental Setup

5.1. Datasets

The study collected novel experimental data through an online survey administered to local elected officials in Switzerland.

• Source: Direct survey of 1404 government members from Swiss municipalities with over 5000 inhabitants.

• Characteristics: The dataset comprises individual-level responses to survey questions, including:

  • Prior beliefs (agreement/disagreement with policy statements, 0-10 scale).

  • Posterior beliefs (same as prior, after treatment, 0-10 scale).

  • Self-rated knowledge about the policy (0-10 scale, proxy for strength of priors).

  • Perceived complexity of the policy (0-10 scale).

  • Ideology (left-right self-placement).

  • Sociodemographic information (age, gender, education level).

  • Treatment assignment (anxiety or control), and policy domain (environment or security).

    While no raw data samples are provided in the paper, an example of a policy statement for which beliefs were measured is:

• For the environmental policy: "We should be moving away from carbon-intensive infrastructure much faster."

• For the security policy: "The government should increase the number of cameras in public places."

  These datasets were chosen because they allow for direct observation and experimental manipulation of anxiety and policy learning among actual policymakers, making them highly relevant for validating the paper's hypotheses at the individual level. The Swiss context was selected due to the availability of elected officials' contact information and the presence of militia politicians (part-time officials), which the authors argue represents a "least likely case" for policy learning, thus making positive findings particularly robust.

The following are the results from Table A1 of the original paper:

The following are the results from Table A2 of the original paper:

5.2. Evaluation Metrics

The primary evaluation metric used in this paper is Policy Learning, which quantifies the change in an individual's beliefs after being exposed to new information and, in the treatment group, anxiety-inducing stimuli.

1. Conceptual Definition: Policy Learning measures the extent to which an individual updates their prior beliefs in the direction suggested by new scientific information. It aims to capture the cognitive process of belief updating in response to external stimuli, serving as a proxy for how effectively an individual integrates new evidence into their understanding of a policy issue. A positive value indicates that beliefs have shifted towards the scientific evidence, while a negative value suggests a shift away (a backlash effect).

2. Mathematical Formula: $ \text{Policy Learning} = \text{Posterior beliefs} - \text{Prior beliefs} $

3. Symbol Explanation:

   • $\text{Policy Learning}$: The numerical value representing the change in belief.
   • $\text{Posterior beliefs}$: The numerical rating (on a 0-10 scale) of a respondent's agreement with a policy statement after receiving the experimental treatment (anxiety stimulus and scientific information).
   • $\text{Prior beliefs}$: The numerical rating (on a 0-10 scale) of a respondent's agreement with the same policy statement before receiving the experimental treatment.

5.3. Baselines

The paper's method is compared against control groups within its experimental design.

• Control Group (No Anxiety Treatment): Participants in the control groups were exposed to the same policy information and scientific evidence as the treatment groups, but they were not exposed to the anxiety-inducing images. This serves as the primary baseline to isolate the causal effect of anxiety. By comparing the policy learning in the anxiety treatment group versus the control group, the researchers can determine if anxiety itself, beyond mere exposure to information, influences learning.

6. Results & Analysis

6.1. Core Results Analysis

The experimental findings provide clear insights into the hypotheses.

Hypothesis 1 (H1): Anxiety increases the extent of policy learning. The study found strong evidence supporting H1. Anxiety has a significant and positive causal effect on policy learning. This means that policymakers exposed to anxiety-inducing stimuli were more likely to update their opinions towards the scientific evidence. This result aligns with Marcus' Affective Intelligence Model (AIM), suggesting that when information is perceived as threatening (due to anxiety induction), systematic information processing is enhanced, leading to belief updating. Figure 1 visually confirms this, showing a positive learning effect in the treatment group and a negative effect (beliefs updated contrary to evidence) in the control group.

Hypothesis 2 (H2): The effect of anxiety on policy learning is stronger when office holders exhibit weak priors. The results partially support the underlying premise but ultimately lead to the rejection of the moderation hypothesis.

• Main Effect of Prior Strength: The strength of priors (self-rated knowledge) was negatively correlated with policy learning. This means that the more knowledgeable a policymaker claimed to be, the less likely they were to update their beliefs. This is a significant finding consistent with Bayesian learning—strong priors are harder to overturn.
• Moderation Effect: However, the interaction effect between anxiety and strength of priors was not statistically significant. This implies that anxiety did not have a stronger effect on learning for those with weak priors compared to those with strong priors, contrary to expectations. Figure 2 shows that while higher levels of learning are generally found with weaker prior knowledge, anxiety exposure did not significantly differentiate learning across different levels of prior strength. The paper notes a slight tendency for those with strong priors to learn more when anxious, which is an unexpected direction, though not significant.

Hypothesis 3 (H3): The effect of anxiety on policy learning is stronger when office holders perceive a policy as complex. This hypothesis was not confirmed.

• Main Effect of Perceived Complexity: The perception of complexity was negatively correlated with policy learning. Policymakers who perceived a policy as more complex were less likely to learn. This contradicts the initial expectation that complexity would increase threat perception and thus promote systematic processing. Instead, it suggests a backlash effect, where perceived complexity might discourage learning or lead to reliance on heuristics (Tversky and Kahneman, 1974).
• Moderation Effect: The interaction effect between anxiety and perceived complexity was not statistically significant. This means anxiety did not have a stronger effect on learning for complex policies.

Observational Findings: Beyond the experimental hypotheses, an observational analysis provided additional insights:

• Complexity and strength of priors were both negatively and significantly correlated with learning, reinforcing the experimental findings regarding their main effects.

• Ideology played a small but significant role: right-wing self-placement was negatively correlated with policy learning, suggesting right-leaning individuals might be less open to updating beliefs.

• University education was positively correlated with policy learning, indicating that higher education might foster greater openness to new information.

• Age and gender showed no significant effect on belief updating, suggesting that the psychological mechanism of learning transcends some sociodemographic differences.

  Overall, the core result is the positive causal effect of anxiety on policy learning. While the proposed moderation effects were not statistically significant, the main effects of prior strength and perceived complexity on learning (both negative) are important findings.

6.2. Data Presentation (Tables)

The following are the results from Table 2 of the original paper:

$^*$ $\hat{p} < 0.1$; $^{**}$ $\hat{p} < 0.05$; $^{***}$ $\hat{p} < 0.01$

The following figure (Figure 1 from the original paper) shows the estimation of the effect of anxiety treatment on policy learning:

该图像是一个图表，展示了焦虑治疗对政策学习的影响。X轴表示焦虑治疗（0表示未治疗，1表示治疗），Y轴表示学习效果。数据点显示出治疗组的学习效果显著高于对照组，表明焦虑对政策学习有正向影响。

The following figure (Figure 2 from the original paper) shows the estimation of the moderating effect of strength of the priors:

该图像是图表，展示了在不同的先前信念强度下，焦虑对学习的影响。红色和蓝色曲线分别代表不同的焦虑处理组，随着先前信念强度的变化，学习表现出现不同趋势。

The following are the results from Table 3 of the original paper:

$^*$ $\hat{p} < 0.1$; $^{**}$ $\hat{p} < 0.05$; $^{***}$ $\hat{p} < 0.01$

The following figure (Figure A1 from the original paper) shows the distribution of three variables: learning ability, strength of prior beliefs, and perceived complexity under different treatment groups (0 and 1).

该图像是一个示意图，展示了不同处理组（0和1）下，学习能力、先前信仰的强度和感知复杂性三个变量的分布情况。图中红色曲线代表处理组0，青色曲线代表处理组1。

The following figure (Figure A2 from the original paper) shows the perceived complexity of environment and security policies.

该图像是图表，展示了环境和安全两个领域的复杂性感知。从图中可以看出，不同复杂性水平下，环境（粉红色）和安全（青色）的感知密度分布情况，反映了政策学习中对情绪的影响。

6.3. Ablation Studies / Parameter Analysis

The paper conducted robustness checks using an alternative modeling approach to verify the consistency of its findings.

• Alternative Model: Instead of linear regressions, multinomial logistic regressions were employed.

• Alternative Dependent Variable: This alternative dependent variable captured three different mechanisms:

  1. Backlash effects: Where incoming information leads to updating that is inconsistent with the information (e.g., beliefs shifting in the opposite direction).
  2. No updating process: Where beliefs remain unchanged.
  3. Learning: Where beliefs update consistently with the new information.

• Results: This alternative version of the analysis, detailed in Appendix C (Table A3), brought confirmation to the findings from the main linear regression models. This suggests that the core relationships observed (e.g., positive effect of anxiety, negative effect of priors/complexity) are robust across different statistical specifications and definitions of learning outcomes.

  The following are the results from Table A3 of the original paper:

  	Dependent variable
  		Backlash effect Learning H	Backlash effect Learning H2		Backlash effect Learning H3
  Anxiety	−0.157 (0.158)	0.209* (0.122)	0.085 (0.586)	−0.165 (0.431)	0.009 (0.373)	−0.019 (0.256)
  Strength of the priors			−0.011 (0.057)	−0.278*** (0.047)
  Anxiety × Strength of the priors			−0.034 (0.083)	0.060 (0.065)
  Complexity					0.069* (0.040)	−0.132*** (0.032)
  Anxiety × Complexity					−0.034 (0.059)	0.041 (0.045)
  Constant	−0.828*** (0.107) −0.184** (0.088) −0.763* (0.409) 1.582***(0.312)				−1.204*** (0.261) 0.494*** (0.184)
  Akaike Inf. Crit.	2699.129	2699.129	2627.660	2627.660	2630.099	2630.099

$^*$ $\hat{p} < 0.1$; $^{**}$ $\hat{p} < 0.05$; $^{***}$ $\hat{p} < 0.01.$

7. Conclusion & Reflections

7.1. Conclusion Summary

This research successfully demonstrated that anxiety has a positive and significant causal effect on policy learning at the individual level among local elected officials. This finding introduces emotions as a critical factor in how public officials process information and update their beliefs, aligning with Marcus' Affective Intelligence Model. The study also found that the strength of prior beliefs is negatively correlated with belief updating, meaning confident policymakers are less likely to learn. Similarly, perceived policy complexity was negatively correlated with policy learning, suggesting a backlash effect where complex issues discourage learning. Crucially, while these factors correlate with learning, they did not significantly moderate the causal effect of anxiety. The study contributes to the policy learning literature by highlighting the microfoundations of learning and bridging insights from political psychology.

7.2. Limitations & Future Work

The authors acknowledge several limitations:

• Resource Constraints: The sample size of approximately 1400 observations limited the statistical power, particularly for detecting interaction effects. A larger sample would be beneficial for more robust moderation analyses.
• Generalizability (Swiss Context): While the psychological mechanisms should be broadly applicable, the study's focus on Swiss militia politicians (part-time officials with regular jobs) means the findings may not be perfectly generalizable to professional politicians in other institutional contexts. However, the authors argue that militia politicians represent a least likely case for policy learning, making the observed positive effect of anxiety particularly robust.
• Need for Replication: Future research should replicate this experiment in other contexts using a comparative approach to enhance generalizability.
• Other Individual-Level Variables: The paper suggests exploring other individual-level variables that might explain information use in policymaking, such as personality traits.
• Information Quantity/Entropy: Future research could investigate the causal effect of information quantity or entropy on knowledge updating and belief modification.

7.3. Personal Insights & Critique

This paper offers a compelling and experimentally rigorous investigation into the often-overlooked role of emotions in policymaking. The finding that anxiety can positively drive policy learning is highly insightful, challenging the traditional view of emotions as mere distractions or irrational forces. It suggests that a certain level of emotional arousal can be a functional component of effective governance, prompting deeper engagement with complex issues.

Potential Issues & Unverified Assumptions:

• Ecological Validity of Anxiety Induction: While the SMID images are validated, the intensity and duration of anxiety induced in a lab-like survey setting might differ from real-world political decision-making contexts. Policymakers face chronic stress and various forms of anxiety, and the effect of a momentary, induced anxiety might not perfectly translate.
• Nature of "Learning": The definition of policy learning as belief updating towards a given scientific consensus is well-defined. However, real-world policy learning is often more complex, involving negotiation, value-based considerations, and learning about problem definitions or normative aspects, not just factual adjustments. The experiment captures one important facet, but not the full spectrum.
• Lack of Moderation Effects: The non-significant moderation effects for priors and complexity are somewhat counter-intuitive given the theoretical expectations. While the main effects are strong, the lack of interaction suggests that anxiety's influence might be more universal than initially thought across these dimensions, or that the experimental design wasn't sensitive enough to capture these nuances. The "backlash effect" of complexity is particularly interesting and warrants further qualitative exploration to understand the underlying psychological resistance.
• Manipulation Risk: The discussion on the potential for manipulation through emotional frames is crucial. If anxiety can increase learning, it also opens the door for interest groups or other actors to strategically use emotional appeals to influence policymakers in ways that may not serve the public good. This highlights the importance of information quality and ethical standards in policy communication.

Transferability & Applications:

• Policy Design & Communication: Understanding that anxiety can spur learning suggests that policymakers might be more receptive to evidence-based arguments when issues are framed in a way that evokes a measured sense of concern or urgency, especially in areas like climate change or public health. However, this must be balanced against the risk of panic or disengagement if anxiety is too high.

• Addressing Cognitive Biases: The negative correlation between strong priors and learning confirms the pervasive nature of cognitive conservatism. Strategies to promote learning might involve explicitly challenging strong priors or creating environments that encourage policymakers to question their existing assumptions without triggering defensive mechanisms.

• Education for Policymakers: The positive effect of university education on learning suggests that continuous education or training for policymakers could be beneficial, potentially fostering critical thinking and openness to new information.

• Interdisciplinary Research: The paper successfully integrates political psychology with public policy. This approach could be extended to other emotions (e.g., anger, hope) or other cognitive biases (e.g., confirmation bias, availability heuristic) to build a more comprehensive model of policymaker cognition.

  This study lays excellent groundwork for further research into the emotional and cognitive microfoundations of policy learning, which is increasingly vital in a world grappling with complex and uncertain challenges.