In terms of second-order nonlinearity, nonlinear distortion in an EAM-based IMDD OFDM system is deterministically modeled as subcarrier-to-subcarrier intermixing interference (SSII), such that DSP-based nonlinear compensation can be provided by decision-aided SSII cancellation at the receiver. The effectiveness of SSII cancellation heavily relies on the accuracy of SSII calculation, which is determined mainly by the modulation model adopted. This study made comprehensive comparisons of three small-signal models (constant-chirp, dynamic-chirp, and nonlinear-E/O models). We then combined theory and experiment results to examine the characteristics of SSII estimation and cancellation using various models, such as SSII spectra, regression coefficients, and correlation coefficients between the SSII and the total received errors, as well as the relationship between improvements in performance and the correlation coefficient. The nonlinear-E/O model is the most comprehensive, and as a result presents the flattest regression coefficient and highest correlation coefficient, which results in the most pronounced improvement in performance.