We present a stochastic analysis of turbulence data, which provides access to the joint probability of finding velocity increments at several scales. The underlying stochastic process in form of a Fokker-Planck equation can be reconstructed from given data. Intermittency effects are included. The stochastic process is Markovian for scale separations larger than the
Einstein-Markov coherence length
, which is closely related to the Taylor microscale
. We extend our analysis to turbulence generated by a fractal square grid. We find that in contrast to other types of turbulence, like free-jet turbulence, the n-scale statistics of the velocity increments and the leading coefficients of the Fokker-Planck equation do not depend strongly on the Reynolds number.