Is a linear estimator really the best choice for this data?
That was my first thought as well. Apparently mediocre minds think alike.
The data would obviously fit an exponential decay much better. Try f(x) = A e^(-Bx) where A and B are the paramters, rather than f(x) = Ax+B.
Other than the poor choice of regression function, the point is well made with the graphs. But to put a finer point on it, you might consider separating subsistence agriculture from production agriculture. I have no doubt that those societies in which agriculture is performed primarily by small farmers who feed only themselves would have a low per-capita GDP, but what about comparing economies of scale as PPP versus %non-farm employees, where "farm" is defined by corporate farms. You'd still probably be right, but it'd be a more compelling argument to make because no one will accuse you of comparing apples to oranges, so to speak.
