The document discusses various measures that can be used to detect multicollinearity in regression analysis. It describes rule of thumb measures like high R-squared and insignificant t-ratios. It also discusses examining correlations between independent variables, performing auxiliary regressions, and using tolerance, VIF, eigenvalues and condition indices. Auxiliary regressions are run for each independent variable by regressing it on the other independents. High R-squared values from auxiliary regressions compared to the overall regression indicate multicollinearity based on Klein's rule of thumb. Tolerance close to 0 and VIF greater than 10 also suggest multicollinearity between variables.
The document discusses various measures that can be used to detect multicollinearity in regression analysis. It describes rule of thumb measures like high R-squared and insignificant t-ratios. It also discusses examining correlations between independent variables, performing auxiliary regressions, and using tolerance, VIF, eigenvalues and condition indices. Auxiliary regressions are run for each independent variable by regressing it on the other independents. High R-squared values from auxiliary regressions compared to the overall regression indicate multicollinearity based on Klein's rule of thumb. Tolerance close to 0 and VIF greater than 10 also suggest multicollinearity between variables.
The document discusses various measures that can be used to detect multicollinearity in regression analysis. It describes rule of thumb measures like high R-squared and insignificant t-ratios. It also discusses examining correlations between independent variables, performing auxiliary regressions, and using tolerance, VIF, eigenvalues and condition indices. Auxiliary regressions are run for each independent variable by regressing it on the other independents. High R-squared values from auxiliary regressions compared to the overall regression indicate multicollinearity based on Klein's rule of thumb. Tolerance close to 0 and VIF greater than 10 also suggest multicollinearity between variables.
The document discusses various measures that can be used to detect multicollinearity in regression analysis. It describes rule of thumb measures like high R-squared and insignificant t-ratios. It also discusses examining correlations between independent variables, performing auxiliary regressions, and using tolerance, VIF, eigenvalues and condition indices. Auxiliary regressions are run for each independent variable by regressing it on the other independents. High R-squared values from auxiliary regressions compared to the overall regression indicate multicollinearity based on Klein's rule of thumb. Tolerance close to 0 and VIF greater than 10 also suggest multicollinearity between variables.
Download as PPTX, PDF, TXT or read online from Scribd
Download as pptx, pdf, or txt
You are on page 1/ 26
Multicollinearity
Measures for Detection
There are following measures for detection of multicollinearity; 1.
Rule of thumb: High R2 and insignificant t-ratios
2.
Correlation between X-variables
3.
Auxiliary regressions
4.
Kliens rule of thumb: multicollinearity is troublesome only if R 2 from
auxiliary-regression > R2 from regular-regression
5.
Tolerance and VIF
6.
Eigenvalues and CI
Rule of thumb: High R 2 and insignificant t-ratios
When value of R2 is high and the F-statistic is significant, but value of tstatistic for large number of individual coefficients i are statistically insignificant. The existence of this problem reflects multicollinearity among explanatory variables. For Example: Model
R Squr
Adj R Squr
Sig. F
.987
.876
.869
40.274
.000
Sig
2.37
.01
1.03
.38
Unstd Coeffint
Std Coeffint
Model
Std Er
Beta
Const
.804
.330
PJ
.021
.550
.0237
Correlation b/w Explanatory Variables
Correlations DJ
PearsonCorrelation
DJ
PJ
IJ
INJ
AEE
Sig.(1-tailed)
DJ
PJ
IJ
INJ
AEE
PJ
INJ
AEE
1.00
0.68
0.51
0.57
0.21
0.68
1.00
0.56
0.66
0.13
0.51
0.56
1.00
0.54
0.11
0.57
0.66
0.54
1.00
0.12
0.21
0.13
0.11
0.12
1.00
0.00
0.00
0.00
0.00
0.00
0.00
0.014
0.00
0.035
.000
IJ
.000
0.00 .
.000
0.00
.000
0.014
0.00 . 0.035
0.024 0.024 .
Auxiliary Regression By putting an independent variables and test the significance of the coefficients on added variables. e.g. test of repressor in IV context is called auxiliary regression Regression function for Job Satisfaction JS
F(DJ, PJ, IJ, INJ, AEE)
We have run five times auxiliary regressions on five independent
variables, as follows: DJ = F(PJ, IJ, INJ, AEE) PJ = IJ = INJ =
F(DJ, IJ, INJ, AEE)
AEE=
F(DJ, PJ, IJ, INJ)
F(DJ, PJ, INJ, AEE)
F(DJ, PJ, IJ, AEE)
Auxiliary regression The Regression results of five independent variables and R- Square( R2 ) of each regression are as following. Variables
R- Square Auxiliary regression
Regular Regression
DJ
0.516
0.255
PJ
0.596
0.255
IJ
0.383
0.255
INJ
0.494
0.255
AEE
0.040
0.255
Auxiliary regression The F-statistics is calculated by using value R2 in auxiliary regression of each variables with following formula
Fi = {R2/(k-2)}/{(1-R2)/(n-k+1)} n= number of observations k= number of independent variables K-2= k-(1+1) as another independent variables get out from independent list of the original JS regression and consequently reduce degree of freedom for independent variables measurement. n-k+1= explanation as above, and consequently it will reduce degree of freedom for error term measurement.
Auxiliary regression F-statistics for DJ
FDJ =
{R2/(k-2)}
{0.516/(4-2)}
/{(1-R2)/(n-k+1)}
/{(1-0.516)/(264-4+1)
= {0.516/2}/{(0.484)/(261) = {0.258}/(0.001854) = 139.1281 F-calculated = 139.1281 > F-tabulated = 4.61, with DF = 2 & 261 at p < 0.01, suggesting explanatory variable DJ is strongly correlated with other explanatory variables.
Auxiliary regression:(Cont..) F-statisticsforPJ
FPJ =
{R2/(k-2)}
/{(1-R2)/(n-k+1)}
= {0.596/(2)} {(0.404)/(261)}
= {0.298} {(0.001548)} = 192.5198 F-calculated = 192.5198 > F-tabulated = 4.61, with DF = 2 & 261 at p < 0.01, suggesting explanatory variable PJ is strongly correlated with other explanatory variables.
Auxiliary regression:(Cont..) F-statisticsforIJ
FIJ =
{R2/(k-2)}
/{(1-R2)/(n-k+1)}
= {0.383/(2)} {(0.617)/(261)}
= {0.1915} {(0.002364) = 81.00729 F-calculated = 81.00729 > F-tabulated = 4.61, with DF = 2 & 261 at p < 0.01, suggesting explanatory variable IJ is strongly correlated with other explanatory variables.
Auxiliary regression:(Cont..) F-statisticsforINJ
FINJ =
{R2/(k-2)}
/{(1-R2)/(n-k+1)}
= {0.494/(2)} {(0.506)/(261)}
= {0.247} {(0.001939) = 127.4051 F-calculated = 127.4051 > F-tabulated = 4.61, with DF = 2 & 261 at p < 0.01, suggesting explanatory variable INJ is strongly correlated with other explanatory variables.
Auxiliary regression:(Cont..) F-statisticsforAEE
FAEE
{R2/(k-2)}
/{(1-R2)/(n-k+1)}
= {0.040/(2)}/{(0.960)/(261)} = {0.020}/{(0.003678) = 5.4375 F-calculated = 5.4375 > F-tabulated = 4.61, with DF = 2 & 261 at p < 0.01, suggesting explanatory variable INJ is moderately correlated with other explanatory variables.
Kliens Rule of Thumb
According to Kliens rule of thumb, if Auxiliary
R2s are greater than the R2 obtained from regular regression than multicollinearity will be a troublesome.
Kliens Rule of Thumb
Variables
R- Square Auxiliary regression
Regular Regression
DJ
0.516
0.255
PJ
0.596
0.255
IJ
0.383
0.255
INJ
0.494
0.255
AEE
0.040
0.255
Except AEE R2 of auxiliary regression all R2 of auxiliary regression are greater
than regular regression R2 , which show severe multicollinearity.
Model Sum of Squares DF Mean Square F Sig. 1 Regression .471 4 .118 1.576 .196 Residual 3.590 48 .075 Total 4.061 52 A. Predictors: (Constant), LC, EXT, DEBT, TANG B. Dependent Variable: DPR
Model Sum of Squares DF Mean Square F Sig. 1 Regression .471 4 .118 1.576 .196 Residual 3.590 48 .075 Total 4.061 52 A. Predictors: (Constant), LC, EXT, DEBT, TANG B. Dependent Variable: DPR
