The theory of molecular-field-induced splitting of the 2p(3/2) peak in x-ray photoelectron spectra of second-row atoms is reexamined, using PH3 and PF3 as examples. Existing theory, which includes crystal field and pseudoatomic valence effects, is found to describe correctly the electrostatic field in the 2p region, but, nevertheless, to severely underestimate the magnitude of the resulting molecular-field splitting. Numerical simulations show that core-valence correlation in the core ionized, final states is responsible for amplifying the molecular-field splitting to the extent that it is observable with the present resolution. A detailed analysis based on second-order Moller-Plesset theory is presented of these correlation effects. A qualitative model is developed which predicts that the contribution from core-valence correlation scales with the asymmetry of the valence shell at the atom of interest. This represents an extension of the pseudoatomic valence theory, to take into account not only electrostatic and exchange effects, but also core-valence correlation.