Correlation vs regression analysis?

#1
Hi!. All, I am a bit confused about the difference between correlation and regression analysis. My friends has been telling me that correlation analysis is designed for understanding if relationships exist between variables, which can be either positive or negative. However, doesn't it look like the same as regression analysis? because ones can not only know if there is a relationship among variables with regression analysis, but they can also know if the relationship is directional or not. And I have noticed that many study tend to perform correlation analysis before performing regression analysis. Why bother to perform correlation analysis first, why not just perform regression?

If possible, could you give me examples of research question of correlation and regression analysis respectively?
 
Last edited:

hlsmith

Less is more. Stay pure. Stay poor.
#2
Correlation provides information about a linear relationship between two variables. Its coefficient is between -1 and 1, and yes conveys information about the directionality. Of note, there are multiple types of correlation to address how the variables are formatted (e.g., Peason, Spearman, point, tetra).

Linear regression provides information about the change in the dependent variable when the value of the independent variable is changed. It also conveys information about directionality. You can perform correlation on more than two variables, but as you mention, regression is more informative. Also, regression allows you to look at relationships while controlling for other covariates as well, called multiple linear regression. You can perform a bunch of bivariate (two variable) correlations before conducting multiple linear regression, but there is no real purpose, since given how variables are generated the correlations can mask some relationships.

Correlations are good for quick exploratory analyses and regression survey to potentially understand dependencies between variables.
 
#3
Correlation provides information about a linear relationship between two variables. Its coefficient is between -1 and 1, and yes conveys information about the directionality. Of note, there are multiple types of correlation to address how the variables are formatted (e.g., Peason, Spearman, point, tetra).

Linear regression provides information about the change in the dependent variable when the value of the independent variable is changed. It also conveys information about directionality. You can perform correlation on more than two variables, but as you mention, regression is more informative. Also, regression allows you to look at relationships while controlling for other covariates as well, called multiple linear regression. You can perform a bunch of bivariate (two variable) correlations before conducting multiple linear regression, but there is no real purpose, since given how variables are generated the correlations can mask some relationships.

Correlations are good for quick exploratory analyses and regression survey to potentially understand dependencies between variables.

Thanks for your prompt reply, but I have heard that people need to conduct correlation analysis first in structural equation modeling analysis(as you can see in the attachment below), is that true? and how about regression analysis , do I also have to perform correlation before performing regression analysis?

1640797578065.png
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
I haven't ran SEM before. But I know they report direct, indirect and adjusted values. @spunky what is your take on this?

Conducting bivariate correlations before regression is misleading and not needed. See topics such at colinearity, simpson's pardox, mediation, etc. All candidate terms should be examined at the same time and based on context knowledge. Basing feature selection on bivariate correlation is an old and very dated process.