In this paper, the Bayesian robustness of the posterior predictive p-value is studied. First of all, it is proved that Lavine's linearization technique can be extended for analyzing this problem. Then, the result is applied to the E-contamination class of prior distributions.