1: \begin{abstract}

2: {\renewcommand{\baselinestretch}{1}

3: This paper provides an alternative to penalized estimators for

4: estimation and variable selection in high dimensional

5: linear regression models with measurement error or missing

6: covariates. We propose estimation via bias corrected

7: least squares after model selection. We show that by separating

8: model selection and estimation, it is

9: possible to achieve an improved rate of convergence of

10: the $\ell_2$ estimation error compared to the rate $\sqrt{s\log p/n}$

11: achieved by simultaneous estimation and variable selection

12: methods such as $\ell_1$ penalized corrected least squares.

13: If the correct model is selected with high probability then the $\ell_2$

14: rate of convergence for the proposed method is indeed the oracle rate of $\sqrt{s/n}.$ Here $s,$

15: $p$ are the number of non zero parameters and the model dimension,

16: respectively, and $n$ is the sample size.

17: Under very general model selection criteria, the proposed

18: method is  computationally simpler and statistically at

19: least as efficient as the $\ell_1$ penalized corrected least squares method,

20: performs model selection without the availability of the  bias correction matrix, and

21: is able to provide estimates with only a small sub-block of the bias correction covariance matrix of

22: order $s\times s$ in comparison to the $p\times p$  correction

23: matrix required for computation of the $\ell_1$ penalized version.

24: Furthermore we show that the model selection requirements are met by a correlation

25: screening type method and the $\ell_1$ penalized corrected least squares method.

26: Also, the proposed methodology when applied to the estimation

27: of precision matrices with missing observations, is seen to

28: perform at least

29: as well as existing $\ell_1$ penalty based methods. All results are supported empirically

30: by a simulation study. }

31: \end{abstract}