# Weijie Su - When Does the First Spurious Variable Get Selected by Sequential Regression Procedures?

286
views
0
13 months ago by

0
13 months ago by
Weijie's new paper: https://arxiv.org/abs/1708.03046

An earlier related paper: https://arxiv.org/abs/1511.01957
0
13 months ago by
In the theorems on Slides 18 and 19, is there any assumption on the ratio of strongest to smallest signal? Or are you using $\beta_{j}=C\sqrt{2 \log p}$ as on preceding slides? (with C a constant)
0
13 months ago by
In your simulation setting, values of  coefficients are the same, what if the value of coefficients varies, same result holds?
For the second part, it will be different. T is likely to be larger. In Figure 2(d) we have examined this issue a bit. In the most extreme case, suppose b1=b2=...=b40=1000 and b41=b42=...=b80=0.001, T will behave as if k=40. In this kind of case, we might want to define an "effective" sparsity.
written 13 months ago by Weijie Su