I have approximated up to t=2 with step sizes h=(1/2)^n for n=1,...,10 using a Adam-Bashford (order 4) that uses the Runge-Kutta (order 4) method to determine the starting values. I've run my MATLAB code for a few examples and they match. However, when I look at my absolute errors for y(2), I find that the next error isn't 1/16 of the previous error, which I think it should be, since I'm increasing step size of 1/2 and the order should be C(h^4).
Am I wrong that the error should be C(h^4)?