The realistic prediction of spray dynamics and dispersion is one of the main tasks of the combustion simulations in diesel engines. Usually, droplet dispersion modeling accuracy can be improved by the development of turbulence models and stochastic dispersion models. Although droplet dispersion due to turbulence is an important phenomenon in high-velocity fuel sprays, studies on the effect of these models are relatively few. Among these studies, the influence of a common used stochastic dispersion model on spray dynamics is not well understood and the application of hybrid turbulence models to fuel spray simulation in the Eulerian-Lagrangian framework with two-way coupling is still at starting stage. In the present study, a better understanding of the effect of the stochastic model on spray dynamics can be obtained by using a statistical analysis of stochastic dispersion time and velocity of dispersed droplets. In order to assess the ability of the hybrid model to improve the radial spread and the inhomogenous distribution of small droplets produced by a high-pressure injector, a partially averaged Navier-Stokes (PANS) model is implemented in an open-source code OpenFOAM and is compared with an unsteady Reynolds averaged Navier-Stokes (RANS) model and a large eddy simulation (LES) model. The results show the importance of resolving large-scale turbulent structures on droplet dispersion modeling in high-velocity fuel spray simulation. The common stochastic dispersion model used in RANS developed for homogenous, isotropic turbulent flows leads to the non-physical dynamics and distribution of droplets and hence is not applicable to the droplet dispersion which is dominated by the non-stochastic motion of large-scale structures in free shear layers. Using PANS instead of RANS should improve the droplet dispersion modeling accuracy of high-velocity fuel sprays in the development of diesel engines with an advanced injection system.