Text this: Empirical versus estimated accuracy of imputation: optimising filtering thresholds for sequence imputation