In the first chapter, I develop a theoretical model to investigate why and how information senders are biased. In this paper, a rational Bayesian consumer decides whether or not to purchase a new product. His utility from purchasing depends on the quality of the product and his idiosyncratic preference for the product. Before making his decision, the consumer can receive a signal of the product's quality by actively choosing an information sender. An information sender would like to attract more consumers by providing more accurate signals, but it is costly. In the paper, an information structure consists of a probability of recommending the product when the quality is high and that of not recommending it when the quality is low. An information sender's bias is defined as the difference between the accuracy of the signal in the high quality state and that in the low quality state. On the other hand, the sender's overall accuracy depends on the sum of the accuracy of the signals. A
consumer does not have direct utility from biased information but it is shown that his expected utility from an information sender is increasing in the sender's bias when the consumer subscribe to a like-minded information sender holding the sender's overall accuracy constant. The indirect demand for information bias gives an incentive to the sender to be biased. As a result, no matter how many information senders are in the market, they have an incentive to be biased. Moreover, as more information senders are potentially able to enter the market, overall accuracy is weakly increasing and bias is weakly decreasing due to competition effects.
In the second chapter, I explore a rational social learning model in which a consumer can observe other consumers' ratings for a product and past purchase decisions. In this paper, I demonstrate how ratings work as an additional information source in a social learning model, and investigate whether or not additional ratings information improves learning. It is common in the social learning literature to model the history of the purchase as the observable, but in this paper, product ratings are also incorporated as an additional source of information; consumers can observe the history of both purchase decisions and ratings. Ratings provide additional information about how previous buyers felt about the product they purchased. I find that with ratings, low quality products are not chosen, but high quality products may also be ignored which causes incorrect herds in the long-run. Interestingly, the probability of high quality products being ignored is not monotonic in the accuracy of ratings. The more accurate information might lower the probability of making correct decisions in the long-run. When consumers believe that ratings are more accurate than private signals, they rely more on ratings and ignore other information sources. In that case, a small number of negative ratings can prevent later buyers from purchasing high quality products. Moreover, the probability of incorrect herds with high quality products might be higher in an environment with ratings, even when the ratings are more accurate than private signals. These results imply that the additional information source does not always guarantee better learning.
Finally, the third chapter studies the case in which observing others actions is noisy in diffusion of innovations. I consider a three stage game where only finite number of early adopters receive private signals about the state of nature and make their decision about whether to adopt the new technology in the first stage. In the second stage, all remaining decision makers observe the early adopters' decisions, but the observation is noisy. I show how the noisy observation affects the dynamics in diffusion of innovations. The dynamics are related to inefficiency of the model because it is desirable that the majority adopts the new technology earlier when its quality is good. The inefficiency tends to decrease as agents can observe others' actions with higher probability, however, it might locally increase in some areas.