Anomalies in experimental data for the EPR-Bohm experiment: Are both classical and quantum mechanics wrong?

We analyze anomalies in data to test the violation of Bell's inequality for the EPR-Bohm experiment. We found that the experimental correlations for photon polarization have an intriguing property. In the experimental data there are visible non-negligible deviations of probabilities $P_{++}^{\rm{exp}}(\alpha, \beta), P_{+-}^{\rm{exp}}(\alpha, \beta), P_{-+}^{\rm{exp}}(\alpha, \beta), P_{--}^{\rm{exp}}(\alpha, \beta)$ from the predictions of quantum mechanics, namely, $P_{++}(\alpha, \beta)=P_{--}(\alpha, \beta)= {1/2}\cos^2(\alpha-\beta)$ and $P_{+-}=P_{-+}(\alpha, \beta)={1/2}\sin^2(\alpha-\beta).$ However, in some mysterious way those deviations compensate each other and finally the correlation $E^{\rm{exp}}(\alpha, \beta)= P_{++}^{\rm{exp}}(\alpha, \beta)- P_{+-}^{\rm{exp}}(\alpha, \beta)- P_{-+}^{\rm{exp}}(\alpha, \beta)+ P_{--}^{\rm{exp}}(\alpha, \beta)$ is in the complete agreement with the QM-prediction, namely, $E(\alpha, \beta)= P_{++}(\alpha, \beta)- P_{+-}(\alpha, \beta)- P_{-+}(\alpha, \beta)+ P_{--}(\alpha, \beta)= \cos 2(\alpha-\beta).$ Therefore such anomalies play no role in the Bell's inequality framework. Nevertheless, other linear combinations of experimental probabilities do not have such a compensation property. There can be found non-negligible deviations from predictions of quantum mechanics. Thus neither classical nor quantum model can pass the whole family of statistical tests given by all possible linear combinations of the EPR-Bohm probabilities. Does it mean that both models are wrong?