Timeline for Simulate linear regression with heteroscedasticity
Current License: CC BY-SA 3.0
21 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 21, 2017 at 15:18 | history | edited | kjetil b halvorsen♦ |
edited tags
|
|
| Feb 27, 2017 at 14:33 | history | edited | kjetil b halvorsen♦ |
edited tags
|
|
| Jan 27, 2017 at 19:32 | vote | accept | user44796 | ||
| Jan 27, 2017 at 19:17 | history | tweeted | twitter.com/StackStats/status/825060185836310528 | ||
| Jan 27, 2017 at 18:53 | answer | added | gung - Reinstate Monica | timeline score: 11 | |
| Jan 27, 2017 at 18:34 | vote | accept | user44796 | ||
| Jan 27, 2017 at 19:32 | |||||
| Jan 27, 2017 at 17:30 | comment | added | Nick Cox | On the spelling, see (or on the other hand don't see if you don't care one bit) stats.stackexchange.com/questions/153526/… | |
| Jan 27, 2017 at 17:28 | history | edited | Nick Cox | CC BY-SA 3.0 |
added 6 characters in body; edited title
|
| Jan 27, 2017 at 16:46 | answer | added | kjetil b halvorsen♦ | timeline score: 5 | |
| Jan 27, 2017 at 16:23 | comment | added | user44796 | Yes, my question is how do I estimate sigma2, which is equivalent to n^1.3. If I have a formula to estimate sigma2, then I can simulate the data. But in the toy example, I only know it because I wrote it. | |
| Jan 27, 2017 at 15:44 | comment | added | whuber♦ | OK, but my last comment still seems apt: you need to postulate some kind of model for the behavior of the residuals. In the model you use to generate the data, the residual SD is proportional to the regressor. Is that exactly the kind of model you want to fit? Or do you need to contemplate other forms of relationship between the error variance and the regressor (or response) values? | |
| Jan 27, 2017 at 15:38 | comment | added | user44796 | Hopefully I have clarified my question with edits. In the above question, n and y represent the empirical data. I want to fit a model to the data and then use the model to generate simulated data that matches the mean and residuals of the original data. | |
| Jan 27, 2017 at 15:36 | history | edited | user44796 | CC BY-SA 3.0 |
added 247 characters in body
|
| Jan 27, 2017 at 15:23 | history | reopened | whuber♦ | ||
| Jan 27, 2017 at 15:23 | history | closed | whuber♦ | Needs details or clarity | |
| Jan 27, 2017 at 15:22 | comment | added | whuber♦ | I do not understand your question, because your code accomplishes exactly what you seem to be asking for in its title: it simulates a linear regression with heteroscedastic errors. Are you asking for methods to estimate some kind of model for the heteroscedasticity? If so, then you need to specify a model! | |
| Jan 27, 2017 at 15:19 | history | edited | user44796 | CC BY-SA 3.0 |
changed lm(n ~ y) to lm(y ~n)
|
| Jan 27, 2017 at 15:04 | comment | added | kjetil b halvorsen♦ | There is aN ERROR IN YOUR CODE, YOU MUST USE ` lm( y ~ n)` | |
| Jan 27, 2017 at 14:55 | comment | added | user44796 | You're right. I am trying to use a linear model to capture the heterogeneity. I think that I should be using a generalized least squares model. If there are any other recommendations, I will try them. | |
| Jan 27, 2017 at 14:44 | comment | added | generic_user | So the linear model won't capture conditional heteroskedasticity unless it explicitly tries to do so, using one of a few approaches. Standard econometric techniques adjust the standard errors on parameters to account for heteroskedasticity, but they don't explicitly model it. | |
| Jan 27, 2017 at 14:40 | history | asked | user44796 | CC BY-SA 3.0 |