Chapter 1: A Study of the Statistical Evidence for Global Warming: Can the Normal Linear Regression Model Detect Global Warming?

2 " Greenhouse gases are accumulating in Earth's atmosphere as a result of human activities, causing surface air temperatures and subsurface ocean temperatures to rise. From the Kyoto Protocol to domestic policies to restrict carbon emissions to proposed taxes on fuel and coal derivatives, national and international governments have embraced the call to combat global warming. 1 There remains little doubt in the minds of many in the scientific community that global warming is real and verifiable. There are many ways to measure climate change; one factor that has received considerable attention by researchers and the public is surface air temperature. 2 The data available from monitoring stations and satellites around the world indicate the existence of a measurable upward trend in surface temperatures. 3 Because of such evidence and the increasingly evident link between the rate of temperature increases and human economic activity, governments have proposed numerous environmental and economic policies, often times at great social and financial cost, in order to slow down the rate of warming in the atmosphere. 4 The justification for these expenditures is that " temperatures are, in fact, rising, " as noted in the introductory quote. In the face of such costs, an important question to ask is how reliable are the empirical analyses upon which the evidence of global warming is founded. These analyses rely on data collected from several sources. The empirical analysis of global temperature data consists of the use of climate and statistical models (Angell, 2002; the empirical analysis of temperature data using statistical models. Statistical analysis of trends in the temperature data involves the search for a statistically significant positive trend in the data—interpreted as evidence of global warming The traditional method has been to fit a linear trend model to the data in order to increased recognition that the linear trend assumption is not appropriate for temperature series and that a non-linear function may better characterize trends in temperature Polsky et. al. (2000) estimated temperature trends in the northeastern United States using non-linear trend functions. Benestad (2003) modeled non-linear temperature trends in the Nordic region of Scandinavia using a cubic polynomial and found significant periods of warming and cooling. Seidel and Lanzante (2004) testes three alternative models that incorporate linear slopes and instantaneous step changes using surface temperature anomaly data and found that two of the three alternative models provide a better fit to the data than …