regression

awkhan

New Member
1) Why do we do statistical tests to check if the independent variables used in the regression are statistically significant or not, when it can be checked using correlation?
2) My understanding is when we have only one continuos variable - we check if the data given is normally distributed or not and then compute statistics for the historical data - which we better coin this term as descriptive statistics.
Not sure where it is used in real -life, normally my understanding is that continuous variables are better understood based on the context of one dim. Let me know your thoughts on this?

hlsmith

Less is more. Stay pure. Stay poor.
1) Why do we do statistical tests to check if the independent variables used in the regression are statistically significant or not, when it can be checked using correlation?
Regression allows you to calculate the MSE, intercept, slope, and R-2 estimates.
Moreover, in regression you can examine the affect of a independent variable on the dependent variable while while controlling for other independent variables.

awkhan

New Member
Regression allows you to calculate the MSE, intercept, slope, and R-2 estimates.
Moreover, in regression you can examine the affect of a independent variable on the dependent variable while while controlling for other independent variables
Understand , it does all these that you mentioned. Along with that we do test statistic(Hypothesis Testing) to check if 2 variables are statistically significant or not. Why do we do this, if we can do it by using correlation, which can check if 2 variables are related or not

hlsmith

Less is more. Stay pure. Stay poor.
Because you can get estimates that you can use in practice. If I tell you two variables have a correlation coefficient of 0.54, what are you going to do with that? If I tell you while controlling for X1 and X2, variable Y increases 5 units for every 1 unit increase of X3 - that is more helpful - right?

What more do you want?

noetsi

Fortran must die
Univariates relationships (shown through correlations) are different than the multivariate relationship tested in regression. X can have a very strong univariate relationship on Y. When you throw another X in the first relationship may largely vanish. Regression test the marginal, controlling for other predictors, influence of X on Y which univariate correlations can not.