UC San Diego

Chapter 1 of this dissertation seeks to understand the sources and nature of measurement errors in temperature variables, and how they can influence climate impact estimations. Measurement error in temperature variables is usually assumed to be “classical'' i.i.d. normal and “small''. This type of measurement error leads to attenuation bias, that is often small enough to be ignored. We show, however, that measurement errors in temperature variables are often large, distinctly non-normal, and vary systematically across space and time. The divergence between our empirical results and conventional wisdom stems from the fact that the construction of the temperature variables involves a series of steps, each of which introduces a distinct source of measurement error. This work is the first to formally characterize sources of measurement error in temperature variables and how they interact. Simulation results are provided that illustrate the influence of these sources of measurement errors on climate impact estimates. We further propose a correction method that can be used to obtain consistent estimates of the parameters of interest under the conditions identified.

In Chapter 2, I zoom in to examine one particular source of measurement error: measurement error induced by using coarse data to approximate daily mean temperature. Much of the historical records from weather stations around the world reports on minimum and maximum temperature and this practice is still followed by most stations. The minimum and maximum are averaged to approximate daily mean temperature, the exposure variable most commonly used in climate impact studies. Empirically, I find substantive differences between daily mean temperatures constructed using hourly temperature data and daily minimum and maximum temperature. This single source of measurement error can easily result in a 5 to 10\% bias in estimates of the impact of daily mean temperature on an output measure of economic interest, with the direction of the bias dependent on the location and season of the year.

Chapter 3 investigates climate impact models from another perspective: the representation of climate variables. The “bin” regression model has been put forward as a flexible semi-parametric method for representing a climate variable and it has emerged as the workhorse approach for empirical work (e.g., Deschênes and Greenstone, 2011). Our work is the first to formally explore econometric properties of the bin regression approach. We show that, although the bin regression approach often produces reasonable results, that the approach produces consistent parameter estimates only under very stringent and highly unlikely assumptions about the true data generating procedure. Problems with the bin regression approach are likely to be most severe in the tail bin categories, where most policy interest with respect to climate change impacts lies. We propose alternatives to bin model for the climate change impacts that produce consistent estimates and generally have better efficiency properties.