Figure 5. General Additive Model Plot, Crop Farming income versus Relative Change in Rainfall, CBMS 2007-2010 Although not utilized in this text, consider a simple General Additive Model (GAM) (Tibshirani, 1986) to fit a non-parametric regression of crop farming income on relative change in rainfall. It can be seen in Figure 5 that the relationship is far from linear4. In fact, chi-square test of linearity between GAM and a usual linear model of crop farming on rainfall yields a significant result (deviance=-7.998). Hence, assuming a linear model might not fit the data so well. However, fitting a GAM may pose interpretability issues on the result not to mention that we have a structural model. Hence, piecewise regression was adopted in estimating a non-linear relationship between crop farming income and change in climate measures. Knots at (-20, 10, 0, 10, 20) have been adopted for rainfall relative measures while increase/decrease for temperature measures. For instance, Figure6 and 7 show demonstrative plot of piecewise regressions (Baum, 2006) of crop farming income on relative change 4 CBMS data is usually used due to larger number of observations. CBMS-FAO | 27