Significant correlation, not significant coefficient

Click For Summary

Discussion Overview

The discussion revolves around the discrepancies observed between correlation coefficients and regression analysis results in a panel data regression test. Participants explore issues related to multicollinearity, the significance of coefficients, and the use of statistical commands in Stata.

Discussion Character

  • Technical explanation
  • Debate/contested
  • Mathematical reasoning

Main Points Raised

  • One participant notes a significant correlation coefficient of 0.177 for "Anleggsmidler/eiendeler" with the dependent variable, but finds it statistically insignificant in regression analysis.
  • Another participant suggests that substantial multicollinearity in the data could lead to unreliable estimates of individual coefficients in the regression.
  • A participant shares VIF-test results indicating that the VIF values for the independent variables are below 10, which they interpret as generally acceptable, although they acknowledge variability in standards.
  • One participant proposes that the lack of statistical significance for "Anleggsmidler/eiendeler" may be due to the effects of other variables, recommending its removal from the model for a re-run of the regression.
  • Another participant suggests running a reduced model by deleting "totalrentabilitat" as a potential solution.

Areas of Agreement / Disagreement

Participants express differing views on the implications of the VIF results and the appropriate course of action regarding the regression model. There is no consensus on the best approach to address the issues raised.

Contextual Notes

The discussion highlights potential limitations related to multicollinearity and the interpretation of statistical significance, but does not resolve these issues or provide definitive solutions.

monsmatglad
Messages
75
Reaction score
0
A couple of questions today. First. I am running a panel data regression test. First I check the correlations between the independent variables and the dependent variable. these are the results.
upload_2017-4-30_17-32-50.png


The D/(D+Em) is the dependent variable, and the independent are the 4 variables most adjacent. Disregard the two outer variables (the red area). The independent variable "Anleggsmidler/eiendeler" has a correlation coefficient (pearson's) to the dependent variable of ,177 and this is a very significant result as you can see. However, when I do the regression analysis (as shown below), the relation between "Anleggsmidler/eiendeler" and the dependent variable is not significant at all. How come the results are so different in terms of significance?

upload_2017-4-30_17-38-18.png

Second question is is there any command to include standardised coefficients to fixed effects regression results in stata? the beta command does not work when I use fixed effects regression (i am pretty green when it comes to Stata). Any advice on these two questions? Thanks in advance!
 
Physics news on Phys.org
You have pretty substantial multicolinearity in this data. The estimates of individual coefficients will be unreliable in the regression.
 
hey. I used a VIF-test and it produced the following numbers: I understand that it differs what is regarded as acceptable, but as far as I know, below 10 is usually no disaster (based on google searches, not on my acumen).

Variable VIF 1/VIF
lndriftsin~r 5.72 0.174921
anleggsmid~r 5.02 0.199399
totalrenta~t 1.34 0.743779
markedbok 1.18 0.845347
Mean VIF 3.31
 
By the time the (estimated) effects of the other variables are taken into account, there is not enough left for Anleggsmidler/eiendeler to be statistically significant. It should be removed from the model and the linear regression should be re-run without it.
 
monsmatglad said:
I used a VIF-test
Instead, just try running a reduced model deleting totalrentabilitat
 

Similar threads

Undergrad Coefficient sign flip in linear regression with correlated predictors
  • · Replies 13 ·
Replies
13
Views
5K
Undergrad Linear regression, feature scaling, and regression coefficients
  • · Replies 8 ·
Replies
8
Views
3K
Undergrad Multicollinearity and Interactions
  • · Replies 5 ·
Replies
5
Views
2K
Undergrad Expected coefficient change from simple to multiple linear regression
  • · Replies 4 ·
Replies
4
Views
2K
Undergrad Linear regression and random variables
  • · Replies 30 ·
2
Replies
30
Views
5K
Graduate "Population-averaged"regression on panel data using Stata
  • · Replies 1 ·
Replies
1
Views
2K
Undergrad Variable Normalization for different variable ranges....
  • · Replies 3 ·
Replies
3
Views
2K
Undergrad Beta coefficient vs. Significance coefficient
  • · Replies 4 ·
Replies
4
Views
4K
Graduate Indirect effect and spuriousity
  • · Replies 1 ·
Replies
1
Views
1K
Undergrad Using correlation coefficients as x in a regression?
  • · Replies 2 ·
Replies
2
Views
2K