Consider the Hartley-Ross unbiased ratio estimator $\hat{Y}_{HR} = \frac{n-1}{n} \sum_{i=1}^n \frac{y_i}{r_i} \cdot \bar{r} + \frac{1}{n} \sum_{i=1}^n y_i$, where $r_i = y_i / x_i$. Under what conditions is the variance of this estimator smaller than the standard ratio estimator $\hat{Y}_R$?