Periodic Reporting for period 4 - BPI (Bayesian Peer Influence: Group Beliefs, Polarisation and Segregation)
Okres sprawozdawczy: 2021-02-01 do 2022-01-31
The objective of this research is to provide a new framework to model and analyse dynamics of group beliefs, in order to study phenomena such as group polarisation, segregation and discrimination. To do so we were planning to introduce a simple heuristic (BPI) that captures how individuals learn from others’ beliefs, and to explore its properties, and analyse the implications of belief dynamics for economic and political interactions.specific aims were to: (i) Analyse rational learning from others’ beliefs and characterise the BPI. (ii) Use the BPI to account for cognitive biases in information processing. (iii) Use the BPI to analyse the diffusion of beliefs in social networks. (iv) Apply the BPI to understand the relation between belief
polarization, segregation in education, and labour market discrimination. (v) Apply the BPI to understand the relation between belief polarization and political outcomes. In line with this goal, the achievements of this project so far are as follows.
The problems being addressed such as political polarisation, discrimination, segregation, cognitive biases in information processing, are extremely important for the well functioning of democratic and equal societies, let alone efficient ones that will implement optimal outcomes.
In addition, during the project, I added the following aims: (iv) Understanding how individuals can strategically manipulate voters with correlation neglect; (v) Understand the effect of misspecified models more generally (beyond correlation neglect) affects political competitions, polarisation, and the rise of populism.
The overall objectives were met with successful publications or research papers as specified below.
polarization, segregation in education, and labour market discrimination. (v) Apply the BPI to understand the relation between belief polarization and political outcomes. In line with this goal, the achievements of this project so far are as follows.
The problems being addressed such as political polarisation, discrimination, segregation, cognitive biases in information processing, are extremely important for the well functioning of democratic and equal societies, let alone efficient ones that will implement optimal outcomes.
In addition, during the project, I added the following aims: (iv) Understanding how individuals can strategically manipulate voters with correlation neglect; (v) Understand the effect of misspecified models more generally (beyond correlation neglect) affects political competitions, polarisation, and the rise of populism.
The overall objectives were met with successful publications or research papers as specified below.
In this project my aim was to provide a framework to understand how individuals process information in a world with complex information networks, in order to better understand topical political process such as polarisation, segregation, and discrimination. I suggest a new formula to calculate how individuals form opinions given the information sources that they are exposed to (such as newspapers, social media, or friends). This formula, which I call the Bayesian Peer Influence heuristic, BPI in short, allows me to focus attention on the implications of different, well-established, behavioural biases in processing information. First, individuals that are exposed to many sources of information often neglect to consider that some sources of information are correlated as, for example, arise from identical sources. Another common mistake in processing information is selection bias, when individuals do not consider the fact that they are exposed to opinions, that are potentially more like their own. In the work I have carried out in this project my objective was to show that my formula and its extensions can capture these two biases, to show how these two biases affect the distribution of beliefs or opinions in society, and in turn how this affects economic and political outcomes.
Specifically, in Levy and Razin (2018) I show how, when individuals interact in social networks, and when they suffer from correlation neglect, this creates polarisation in opinions. In Levy, Moreno de Barreda and Razin (2021, 2022) we focus on political opinions and show how political campaigns can use correlation neglect to manipulate voters’ opinions creating polarisation in the population. We show that in some cases, this manipulation can be extreme. In Levy and Razin (2019) we analyse how echo chambers together with selection bias imply homogenous beliefs within chambers but polarisation across them.
In Levy and Razin (2017) we study discrimination between graduates of private and state school education in the UK. We use the above framework we show how selection bias in applying the BPI implies that this type of discrimination can be sustained in the long term when individuals in society choose to segregate into chambers where they hold similar opinions and influence each other to create more polarisation.
In Levy and Razin (2022) we establish theoretical foundations for the BPI. We consider a situation in which individuals can potentially account for some correlation, even large degrees, but are exposed to a very large amount of information. We show that in this case individuals may end up behaving as if they neglect all levels of correlation. In Laohakunakorn, Levy and Razin (2019) we use this extended framework of the BPI to analyse what happens when individuals are averse to uncertainty about the correlation between their information sources. We analyse outcomes of auctions when bidders are concerned about the correlation between their information and that of the other bidders and show that this can lead in some cases to lower revenues to the seller.
During the lifetime of the project I have also adopted and changed my formula to discuss related biases in information processing. In Levy and Razin (2021) we adapt the framework to analyse a mirror image of correlation neglect, that arises when people explain what they observe by potentially considering too much correlation in their data sources, and so are looking for patterns where these potentially do not exist. A key insight there is that extremism and polarisation may still arise. Our study of correlation neglect also led us to the paper Levy, Razin and Young (2022) in which we explore the effect of correlation neglect on political outcomes in society. We show that when some groups in society neglect to take into account the correlation between policy outcomes and some policy tools, we end up with political cycles in which society is perpetually bouncing between periods of political rule of politicians that espouse the correct model of the world and periods of political rule of (potentially populist) politicians holding simplified views of the world.
During the lifetime of the project, I have concluded and published 9 papers in major economic academic journals. I disseminated these papers and results in many conferences and seminar presentations.
Specifically, in Levy and Razin (2018) I show how, when individuals interact in social networks, and when they suffer from correlation neglect, this creates polarisation in opinions. In Levy, Moreno de Barreda and Razin (2021, 2022) we focus on political opinions and show how political campaigns can use correlation neglect to manipulate voters’ opinions creating polarisation in the population. We show that in some cases, this manipulation can be extreme. In Levy and Razin (2019) we analyse how echo chambers together with selection bias imply homogenous beliefs within chambers but polarisation across them.
In Levy and Razin (2017) we study discrimination between graduates of private and state school education in the UK. We use the above framework we show how selection bias in applying the BPI implies that this type of discrimination can be sustained in the long term when individuals in society choose to segregate into chambers where they hold similar opinions and influence each other to create more polarisation.
In Levy and Razin (2022) we establish theoretical foundations for the BPI. We consider a situation in which individuals can potentially account for some correlation, even large degrees, but are exposed to a very large amount of information. We show that in this case individuals may end up behaving as if they neglect all levels of correlation. In Laohakunakorn, Levy and Razin (2019) we use this extended framework of the BPI to analyse what happens when individuals are averse to uncertainty about the correlation between their information sources. We analyse outcomes of auctions when bidders are concerned about the correlation between their information and that of the other bidders and show that this can lead in some cases to lower revenues to the seller.
During the lifetime of the project I have also adopted and changed my formula to discuss related biases in information processing. In Levy and Razin (2021) we adapt the framework to analyse a mirror image of correlation neglect, that arises when people explain what they observe by potentially considering too much correlation in their data sources, and so are looking for patterns where these potentially do not exist. A key insight there is that extremism and polarisation may still arise. Our study of correlation neglect also led us to the paper Levy, Razin and Young (2022) in which we explore the effect of correlation neglect on political outcomes in society. We show that when some groups in society neglect to take into account the correlation between policy outcomes and some policy tools, we end up with political cycles in which society is perpetually bouncing between periods of political rule of politicians that espouse the correct model of the world and periods of political rule of (potentially populist) politicians holding simplified views of the world.
During the lifetime of the project, I have concluded and published 9 papers in major economic academic journals. I disseminated these papers and results in many conferences and seminar presentations.
Our novel methodology is to use insights from information theory to model how individuals perceive correlation. We formulate the BPI heuristic and explore how to generalize this approach using tools from information theory.
This led us to novel and unconventional models of decision making:
1. A new approach using maximum likelihood heuristic, to combining forecasts.
2. A new model to analyse degrees of correlation neglect using the BPI, and application to financial markets in which correlation is sometimes misperceived.
3. A novel approach to modelling populism as a simple misspecified model of the world. Application to political competition and unravelling the possibility of cycles of populism.
4. A new approach for the analysis of political polarisation as well as populism arising from cognitive biases and correlation neglect using the BPI, which is important mainly considering social media.
5. A new approach for the analysis of strategic manipulation of voters who exhibit correlation neglect, using the BPI.
6. A new approach for the analysis of discrimination and segregation stemming from polarised beliefs that arise from endogenous networks and selection bias, using the BPI.
This led us to novel and unconventional models of decision making:
1. A new approach using maximum likelihood heuristic, to combining forecasts.
2. A new model to analyse degrees of correlation neglect using the BPI, and application to financial markets in which correlation is sometimes misperceived.
3. A novel approach to modelling populism as a simple misspecified model of the world. Application to political competition and unravelling the possibility of cycles of populism.
4. A new approach for the analysis of political polarisation as well as populism arising from cognitive biases and correlation neglect using the BPI, which is important mainly considering social media.
5. A new approach for the analysis of strategic manipulation of voters who exhibit correlation neglect, using the BPI.
6. A new approach for the analysis of discrimination and segregation stemming from polarised beliefs that arise from endogenous networks and selection bias, using the BPI.