I am under the impression that quantile regression uses all observations at each quantile analyzed (only with different weights on the residuals), so that there are not issues we would find with subsamples/sample size. Is this true?
I don't feel like an expert on this, but it seems that the reliability of your results deteriorate as the number of regressors increases. Another thought might be to use some simulations to get an idea about the size of the standard errors corresponding to your application.
Regression, in general, uses all the observations. Quantile regression enables us to examine the pattern of relationships & measures variability across samples which is expected to be statistically significant.It all depends further on nature of distribution of the population.How quantile regression helps us in non-normal population is a question to be answered.
Ensure that sample of 300,which statistically large, is normally distributed.
Are there non-parametric tests ? I am not sure.
Dr.N.S.Viswanath
Dear Amanda,
a few points about qreg:
1) Quantile regression, like any other regression model, uses all the informations. The "quantile" is referred to the point of the distribution of your dependent variable at which you're estimating (i.e. studying the effect). For instance, the widely used linear regression does it at the mean.
[Note: from this standpoint, using quantile regression might give an advantage, as you can study the effect at any point of the distribution. Some statistical packages - Stata does that for sure - allow you to estimate regressors, i.e. effect, per each quantile and then plot them to see if the effect changes in magnitude or even in direction when you move across the distribution of your dependent variable].
2) Quantile regression does NOT require the dependent variable to be normally distributed.
3) Be aware that, when using quantile regression, some theoretical statisticians recommend to use bootstrapping for the estimation of the standard errors of regressors.
HTH!
Cheers,
Marco
I would add some more points on regression.We need a strong theoretical construct before using acquired data. The frame work that one develops must have data points to suit a type of regression model.
Test the assumptions of the model for the data set be fore we venture to analyze. These preparations will enable us to interpret data set for drawing valid data based conclusions.
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running