If you are comparing both algorithms you need first to have a reference, and then you need the performance index for the comparison. In your example you are not really able to compare because your "real data" are only 4 points and you are using the same points for the fitting and you do not have any information between the points. I suggest, you generate for example 100 points and use only 50 for the fitting, then you can evaluate the deviation of the generated curves (spline and pchip) in the points that you know but that you are not using for the fitting (i.e. the other 50). And for the "evaluation" you can uses, mean value, square mean value and the correlation factor (as performance indices).