Thesis Structure
The structure will build up the concepts to achieve the main goal: suggest a experiment to test whether a social bond contract decreases the level of complaints in a region. Thus, the first chapter will present the overlaying context that motivates this thesis: the ultimate goal of using city and citizens generated data to guide public policies. This requires a debate about the current public administration status and the new data-guided proposals of public contracts, such as the Social Impact Bonds. Then, we will present our view of how those contracts should be developed in order to avoid known deficiencies of replicability, effectiveness proof and live monitoring.
The second chapter will describe de problem and describe the Waze dataset we use. I will restrict the problem to road infrastructure on cities arguing that there is a reliable and comprehensive source of this type of data. Then, I will characterise Waze data, showing its format and how it was processed to the analysis. Also, I will explain how the data is collected, in order to understand possible biases. This will provide a basis for a discussion about possible analytical impacts of those biases and what are the available mitigations for them.
All the following chapters are methodological steps to achieve the main goal. However, we pretend to develop this project with an open source paradigm. This means that the design of the method has to simplify future contributions and criticisms. We see two main requirements: modularity and error estimation. The first requirement allow a contribution on a single step of the method without affecting the process chain. Thus, improving the final goal. Error estimation is a common tool to compare different methodologies and simplify community discussion. Therefore, each methodology step is thought to be independent and, each of them, has to have an error estimation.
The third chapter will develop a methodology to divide the city in regions and to aggregate complains geographically. The first topic is required due to the need of running an hypothesis test to ensure that the social bond contract [SIB] is effective. Thus, this division has to attend minimum requirements of variable controls. The former topic, aggregate complains geographically, is needed because Waze data provides the coordinates where the alert was reported.
The fourth chapter will suggest a way to measure the level of complains on a certain region. This measurement has to consider temporal user variations at the region and the city. For instance, an increase of complaints on a certain region can be the outcome of a increase of app usage at the city. Also, it has to consider the seasonality of complaints that can lead to a mistaken estimation of citizen satisfaction. Thus, the desired outcome is a robust and reliable index of citizen satisfaction at a certain region that can be used on a contract.
Finally, the fifth chapter will propose an experiment to test whether the social bond contracts are effective. It will use all the methods developed previously. It has to control all possible variables that can affect the results. Furthermore, it should provide results on a weekly basis to allow contractors to verify the company effectiveness.