Backscattering is a sensitive probe of the accuracy of electron scattering
algorithms implemented in Monte Carlo codes. The capability of the Geant4
toolkit to describe realistically the fraction of electrons backscattered from
a target volume is extensively and quantitatively evaluated in comparison with
experimental data retrieved from the literature. The validation test covers the
energy range between approximately 100 eV and 20 MeV, and concerns a wide set
of target elements. Multiple and single electron scattering models implemented
in Geant4, as well as preassembled selections of physics models distributed
within Geant4, are analyzed with statistical methods. The evaluations concern
Geant4 versions from 9.1 to 10.1. Significant evolutions are observed over the
range of Geant4 versions, not always in the direction of better compatibility
with experiment. Goodness-of-fit tests complemented by categorical analysis
tests identify a configuration based on Geant4 Urban multiple scattering model
in Geant4 version 9.1 and a configuration based on single Coulomb scattering in
Geant4 10.0 as the physics options best reproducing experimental data above a
few tens of keV. At lower energies only single scattering demonstrates some
capability to reproduce data down to a few keV. Recommended preassembled
physics configurations appear incapable of describing electron backscattering
compatible with experiment. With the support of statistical methods, a
correlation is established between the validation of Geant4-based simulation of
backscattering and of energy deposition