i have 30 observations of yearly data and i try to do quantile but results is all of my variables is not significant. im wondering if we can use quantile regression on yearly data? or is there any minimun observation?
Asked
Active
Viewed 138 times
0
-
Which quantiles do you want to estimate from the 30 data points? Also, see these threads to learn more about quantile regression: What are the assumptions for applying a quantile regression model? and When is quantile regression worse than OLS?. – dipetkov Dec 29 '22 at 14:34
-
0.5,0.75.0.25 sir – Ummi Nurannisa Jan 05 '23 at 02:57
-
Some comments: (a) Checking whether any predictors are significant isn't a good way to evaluate model fit. (b) I couldn't find explicit guidance about minimum sample size (other than warnings quantile regression is less efficient than OLS). However, Koenker's Quantile Regression has a section to evaluate methods for computing confidence intervals for the median, and the sample size used is (at least) $n = 100$. (See Section 3.10.) So I would be hesitant to use only $n = 30$ to model all three quartiles. – dipetkov Jan 15 '23 at 17:05
-
(c) An alternative approach to consider might be ordinal regression. See the resources by Prof. F. Harrell here. – dipetkov Jan 15 '23 at 17:05