MSD calculation window

Hi everyone,

I have a simulation of protein in water for 100 ns and all the necessary trjconv is done.
I wanted to calculate the diffusion coefficient of the water around certain residues using MSD curve slopes.

I notices that, when I change the MSD calculation window (200ps, 500 ps, 1000 ps) the slopes getting steeper and diffusion coefficients increased.

200ps 500ps 1000ps
1.47E-05 1.64E-05 1.77E-05

My question is, what is the protocol to use the ideal/best MSD calculation window?

FYI: I used same region with linear regression for each calculation