1: \begin{abstract}

2: We develop a univariate, differentially private mean estimator, called the private \textit{modified winsorized mean}, designed to be used as the aggregator in subsample-and-aggregate. 

3: We demonstrate, via real data analysis, that common differentially private multivariate mean estimators may not perform well as the aggregator, even with a dataset with 8000 observations, motivating our developments.

4: We show that the modified winsorized mean is minimax optimal for several, large classes of distributions, even under adversarial contamination. 

5: We also demonstrate that, empirically, the modified winsorized mean performs well compared to other private mean estimates. 

6: We consider the modified winsorized mean as the aggregator in subsample-and-aggregate, deriving a finite sample deviations bound for a subsample-and-aggregate estimate generated with the new aggregator. 

7: This result yields two important insights: (i) the optimal choice of subsamples depends on the bias of the estimator computed on the subsamples, and (ii) the rate of convergence of the subsample-and-aggregate estimator depends on the robustness of the estimator computed on the subsamples. 

8: % Lastly, as a by-product of our analysis, we introduce a zero-concentrated differentially private version of the unbounded private quantile algorithm, as well as some concentration inequalities for these quantile estimates, 

9: \end{abstract}

10: