Introduction
From buying stocks to getting vaccinated or even crossing the street, people rely greatly on others to help make their decisions. In all these examples, people decide sequentially which will indefinitely lead to information cascades. I am going to dive a little deeper into this concept and look at some of the intricacies of information cascades. First I will look at how personal and social choices are made and how their accuracies compare. Secondly, I am going to discuss the effects of confidence and speed of choosing their choice. Lastly, I will talk about how initial choices affect the future choice and their significance. The reason I picked this topic was that I was curious and wanted to learn more about the mechanisms of information cascades since this idea greatly affects my and everyone else’s daily life.
During lectures, we learned about how information can diffuse and cascade through the herding model. The herding model is when each person can make decisions from their private info and although they can’t observe what others know, they can see what others choose. In these experiments, they tested using this concept where participants are asked a question and first provide their personal choice and then get to see social information about other individual’s choices where they can then change and submit a second answer.
From 141 participants they made these graphs looking at the accuracy of personal and social choices. Graph A clearly shows that the average social choice will be higher than the average personal choice (the average is the black dot). Overall, there was an average accuracy of 79% on the social choice which was 5% better than overall personal choice. From graph B you can see a graph comparing confidence in personal choice and personal accuracy.
Now with the second Graph A, we can see that people who reported having the lowest confidence in their choice had the largest improvements after looking at the social choice (15%) while those most confident rarely benefited. This is likely related to the fact that the reason users are more confident in their answers is that they have proper justification as to why they arrived at their decision, thus making their answer more likely correct. It’s also noted that when personal choice and social choice aligned, they would rarely change which makes sense as it reconfirms their idea that they made the correct choice. Second Graph B shows that confidence in personal choice leads to faster social choices. Graph D showed that the personal choice of those who chose later was much less accurate than those who chose first but did greatly improve when choosing their social choice. The main conclusion you can draw is that the overall social choice will be more likely to be correct. Also, the group’s answers will self-organize according to the quality of the information with confident and more accurate participants choosing early which then provides high-quality information for the less accurate and less confident people, who then choose later on.
From this data, you can see the group choice is better but not always accurate. Graph E looks at confident individuals and the effect of how having more accurate answers affects their later final response. It shows people improved their choice when more confident people were more accurate. If confident individuals who are more likely to choose early, choose wrong, this leads to less improvement in social choice. With this idea and looking at graph C, which shows that when the larger majority favors the opposing choice, the participants are more likely to switch, it causes a snowball effect. More people choose the wrong answer and because more people chose the wrong answer before them, they are more inclined to choose the wrong answer themselves. Relating this to our class, this is like the example in lecture where a fake news post can receive thousands of retweets and likes even though it’s not real. This emphasizes that those who make the decision confidently and early have a greater effect than the choice of those later on as they can dictate the original choice to snowball from.
These three graphs show the improvement as more people make choices (choice threshold where low means fewer people choose and high means a lot have made their choice). From graph A you can see that unless confident individuals were less accurate than unconfident ones, they still improved on their accuracy. Groups B and C with lower personal accuracy only improved when the confident ones were always better than the unconfident ones. You can see from all 3 graphs that the improvement of unconfident agents is amplified based on how accurate the confident individuals are. The more accurate they are, the better, and the less accurate, the worse they do. This is likely due to unconfident participants relying on the social choice more since they are hesitant about their answers. This happens even though they may be more right which further and clearly shows the snowball effect.
Conclusion
Information diffusion and cascades are largely influenced by the herding model and looking at a person’s confidence and how quickly they make their choice can tell you their accuracy as well as the likely accuracy of subsequent unconfident individuals. Hopefully, you learned more in-depth about the decision dynamics of information cascades. This knowledge may also be able to help us question or improve our decision-making when we are unsure about our choices.
Thanks for reading!
Reference/Sources
https://www.science.org/doi/10.1126/sciadv.abb0266