| dc.description.abstract | Much of science depends on random sampling, and random sampling always involves variance — samples of the same population can have different results. In this dissertation, I explore how variance between samples manifests as variance between scientists’ opinions as well as unpredictability of the community as a whole, depending on how the community handles the samples. I show choices between network structures, strategies for sharing information, and strategies for trusting information impose trade-offs concerning the goals that a community might have. Finally, I consider how the variance of studies might be distorted by two forms of inherent reliability of scientists, imprecision and bias, and evaluate solutions to the resulting challenges.
To conduct this exploration, I design computer models to simulate scientific research. Scientists update their opinions using data that they produce individually and share amongst each other. I model core aspects of science and bracket away further complications, which obfuscate underlying dynamics. By isolating particular features of scientific communities in this way, I am able to identify novel dynamics inherent to science on a fundamental level.
In particular, I show how certain ways of diversifying opinions of a community can be useful by enhancing the predictability of that community. I follow this by uncovering how a community can achieve predictability without diversifying opinions. This sets up an investigation into such methods as potential solutions to severe challenges of scientific communities that are caused by data manipulation and insidious forms of information sharing.
This project provides a foundation for how to approach distributing statistically generated evidence (i.e. random samples) on networks. One might think that interesting questions concerning information sharing only emerge for complicated environments and that sharing scientific information alone is largely trivial. I show that this intuition is false. I show there are fundamental dynamics concerning how statistically generated data is shared, and exploring these is vital for understanding the more complex situations. | |
