Only keep variables that are significantly different from 0

[Lipty]

A super silly question to ask.

I would like to do a robustness test, in which I will only keep variables that are significantly different from 0. What is the most efficient method? Thank you very much. : )

[sustagens]

Check out this link: PROC REG: Simple Linear Regression 

"The table also contains the statistics and the corresponding -values for testing whether each parameter is significantly different from zero."

[PaigeMiller]

---

@Lipty wrote:

I would like to do a robustness test, in which I will only keep variables that are significantly different from 0. What is the most efficient method? Thank you very much. : )

---

Which robustness test? Be specific.

[Lipty]

There are about 10,000 observations and some are very close to 0. For example, -0.005, 0.0007,-0.006, etc. I conducted some tests with the original 10,000 observations. Now I could like to only keep observations that are significantly different from 0 in the sub-set to do the same tests. Then I need to drop other observations.

Thank you. : )

[PaigeMiller]

---

@Lipty wrote:

There are about 10,000 observations and some are very close to 0. For example, -0.005, 0.0007,-0.006, etc. I conducted some tests with the original 10,000 observations. Now I could like to only keep observations that are significantly different from 0 in the sub-set to do the same tests. Then I need to drop other observations.

---

Again, I ask which specific test are you doing to determine if something is significantly different than 0?

Also, originally you spoke of keeping variables, now you speak of keeping observations. Which is it?

[Lipty]

Thank you, Paige. I meant observations. It should be observations in the in subject. 

Once the observations are 5% significantly different from 0, I will keep them in the sub-sample.

[PaigeMiller]

Okay, you haven't answered my question. So, let me try another way.

How do you want to determine if the observation is statistically different than zero? What statistic? What method? What technique? You mentioned "robust", but that isn't enough for me to know how you want to do this. Please explain.

Also, would you please be so kind as to change the subject line on your original message to indicate that the problem is about keeping observations?