In this dissertation, I analyze the drivers of wind power development in the United States as well as the relationship between renewable power plant location and transmission congestion and emissions levels. I first examine the role of government renewable energy incentives and access to the electricity grid on investment in wind power plants across counties from 1998-2007. The results indicate that the federal production tax credit, state-level sales tax credit and production incentives play an important role in promoting wind power. In addition, higher wind power penetration levels can be achieved by bringing more parts of the electricity transmission grid under independent system operator regulation. I conclude that state and federal government policies play a significant role in wind power development both by providing financial support and by improving physical and procedural access to the electricity grid.

Second, I examine the effect of renewable power plant location on electricity transmission congestion levels and system-wide emissions levels in a theoretical model and a simulation study. A new renewable plant takes the effect of congestion on its own output into account, but ignores the effect of its marginal contribution to congestion on output from existing plants, which results in curtailment of renewable power. Though pricing congestion removes the externality and reduces curtailment, I find that in the absence of a price on emissions, pricing congestion may in some cases actually increase system-wide emissions.

The final part of my dissertation deals with an econometric issue that emerged from the empirical analysis of the drivers of wind power. I study the effect of the degree of censoring on random-effects Tobit estimates in finite sample with a particular focus on severe censoring, when the percentage of uncensored observations reaches 1 to 5 percent. The results show that the Tobit model performs well even at 5 percent uncensored observations with the bias in the Tobit estimates remaining at or below 5 percent. Under severe censoring (1 percent uncensored observations), large biases appear in the estimated standard errors and marginal effects. These are generally reduced as the sample size increases in both N and T.