One-sided p-values of each model’s Root Mean Square Error (RMSE) score, using the Student’s t-test, against the distribution of per-fold cross-validation means achieved by the best model in Table 4.

The best model is the Text Only SVR (RBF Kernel) without sentiment, indicated as Ref below. The null hypothesis is that the mean of this best reference model is greater than (or in other words, worse than) the means shown in Column 4 of Table 4. Models specified with (*) do not include any zip code-level features. The p-values for the predictive models, with and without sentiment as a feature, are reported as without sentiment / with sentiment. Since cross-validation is not applicable for the constant-value baseline models, they are excluded in this table.