I am doing a multiple regression on several variables (i.e. 13 vars). The variables have few continuous variables and many categorical variables. The dataset is for only a year involving 7891 school applicants for entrance exams. I am using the average scores as my dependent var. The R-squared of the multiple linear model is very low 0.022. I have attempted using other methods such as nonlinear OLS but the R-squared is still very low. Also, I attempted using quantile regression but I think it's not applicable for the current data at hand. Can I get any recommendations in terms of pre and post estimations methodologies for the current data? Thank you!
Hi,
It is not clear what your goal is:
- Do you need a good predictive model?
- Do you need to model the data using a pre-specified known statistical model?
Anyway, Rsquared values like that basically mean that the model is not very useful, i.e. the variables you have are not very good at predicting the outcome.
There is probably no fix to this, unless you include other predictors.
Quantile regression probably IS applicable, but it is not clear why you would want to use it.
Could you elaborate?
- To investigate about the impact of factors affecting students’ examinations results or performance by using regression. The specific objective is aimed to empirically examine the econometric models that would look at some relationships of applicants’ characteristics on their performance scores (i.e. average scores). Please tell me how I can employ the Quantile regression since I have two subjects (i.e math and english scores) plus huge categorical variables relating to students specifically. If this is applicable, please send me some materials on test scores as well as sofware scripts (i.e. R/Stata/Eviews). Thank you very much!
It might not sound appropriate to use inferential statistical test on nominal or ordinal variables. I guess that is the cause of low regression and wide stochastic disturbance term. You can employ factor analysis or relative importance index.
Huh?
Joe Greaves you can totally use regression if your dependent variable is continuous and it is a score, as it seems. Would you like to share a screenshot of your dataframe, indicating the outcome and the predictors? What do you mean you have two subjects? Two outcomes?
****
Adetayo Olaniyi Adeniran if the outcome is categorical you can use a GLM, I do not see the problem here.
@ Stefano...I partly agree with you. Categorical variable is not suitable for Regression analysis. There is no need to patch it.
- Badges
- Science topic