Timeline for Kruskal-Wallis or Fligner test to check homogeneity of variances?
Current License: CC BY-SA 3.0
14 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 13, 2014 at 16:15 | comment | added | Baumann |
white.test from package tseries is not about homoscedasticity, but nonlinearity. This function "generically computes the White neural network test for neglected nonlinearity".
|
|
| Sep 13, 2011 at 8:15 | comment | added | Dail | I get: Estimate Std. Error t value Pr(>|t|) (Intercept) -0.02994069 0.09444855 -0.317 0.7513 prices[, 2] 0.06069896 0.00064885 93.549 <2e-16 ***, Do i have to look at <2e-16 ? and what about white.test example I wrote above, correct? | |
| Sep 13, 2011 at 8:07 | comment | added | caracal |
@Dail Re Fox & Weisberg, for the example above: library(car); library(lmtest); coeftest(fit, vcov=hccm). This is basically the normal t-test for regression parameters, just with the different standard errors.
|
|
| Sep 13, 2011 at 7:08 | comment | added | Dail | @¢aracal, then how to see the pages you told me in "Fox & Weisberg (2011), An R Companion to Applied Regression." ? I saw hccm)= but I didn't understand how to calculate the p.value. | |
| Sep 13, 2011 at 6:43 | comment | added | Dail | caracal, perfect! Thank you so much.. only one doubt about white.test. Using BP test i pass the model (linear reg), but white.test doesn't accept that model, i see x and y do i have to do something like: white.test(prices[,1],prices[,2]) ? | |
| Sep 13, 2011 at 6:39 | vote | accept | Dail | ||
| Sep 12, 2011 at 19:10 | comment | added | caracal | @Dail I've updated my answer for functions that implement the White-Test and the heteroscedasticity-consistent standard errors. | |
| Sep 12, 2011 at 19:09 | history | edited | caracal | CC BY-SA 3.0 |
added 520 characters in body
|
| Sep 12, 2011 at 18:02 | comment | added | Dail | caracal, it seems that BP test always return a very loow p-value if I do: bptest(mod) where mod is my linear regression model. DO i have to "format" my data before doing BP test? maybe use this formula: diff(resid(mod))~1 ? | |
| Sep 12, 2011 at 16:10 | comment | added | Dail | it seems that only BP test there is on R...do you know a package that has HW test? | |
| Sep 12, 2011 at 16:05 | comment | added | caracal | @Dail Yes, you can consider the Breusch-Pagan-Godfrey-Test or the White-Test. You can also have a look at heteroscedasticity-consistent standard errors (Huber-White). There may be more methods specifically suited for time series (GARCH, ...), but I'm not familiar with them. | |
| Sep 12, 2011 at 16:02 | history | edited | caracal | CC BY-SA 3.0 |
added 175 characters in body
|
| Sep 12, 2011 at 15:46 | comment | added | Dail | wow! wonderful explanation, so are you advincing me to use an homoscedastic test like the Breush-Pagan test? I remember that i used it before BUT it doesn't works well with ouliers(I mean, if my series are outliers It could tell the series is Homoscedastic but is not), what is the best method to check homoscedastic in timeseries? | |
| Sep 12, 2011 at 15:36 | history | answered | caracal | CC BY-SA 3.0 |