direction: the invisible player
DESCRIPTION
Direction: The Invisible Player. Robert E. Johnson Dept. of Mathematical Sciences VCU Based on - Johnson and Herr, “Direction: The Invisible Player”, ASA Proceedings of the Statistical Education Section, 48-52, 1993. Workload Related to Anesthesiology Service. Dependent Variable - PowerPoint PPT PresentationTRANSCRIPT
Direction:The Invisible Player
Robert E. JohnsonDept. of Mathematical Sciences
VCU
Based on - Johnson and Herr, “Direction: The Invisible Player”, ASA Proceedings of the Statistical Education Section, 48-52, 1993.
Workload Related to Anesthesiology Service
• Dependent Variable– WORKLOAD: man-hours
• Independent Variables– CASES: number of surgical cases– ELIGIBLE: rate of service eligibility per 1000
patients
Source: Procedures and Analyses for Staffing Standards Development: Data/Regression Analysis Handbook. San Diego, CA: Navy Manpower and Material Analysis Center, 1979. [From: Myers, Classical and Modern Regression with Applications, PWS-Kent Publishing Co., Boston, MA, 1990 (pages 381-383)].
CorrelationsWorkload/Anesthesiology Service
Cases Eligible
Workload 0.980 0.971(<0.001) (<0.001)
Regression AnalysisWorkload/Anesthesiology Service
Analysis of Variance Section
Sum of Mean ProbSource DF Squares Square F-Ratio LevelModel 2 1.457E+07 7284372.7 124.752 <0.001Error 9 525517.4 58390.82Total(Adj.) 11 1.509E+07
Root Mean Square Error 241.642 R-Squared 0.965
Regression AnalysisWorkload/Anesthesiology Service
Regression Equation Section
Indep. Regression Standard T-Value ProbVariable Coefficient Error (Ho: B=0) LevelIntercept 137.2353 115.0564 1.193 0.264Cases 2.960778 1.214657 2.438 0.038Eligible 3.085881 2.919934 1.057 0.318
CorrelationsWorkload/Anesthesiology Service
Cases Eligible
Workload 0.980 0.971(<0.001) (<0.001)
Cases 0.975(<0.001)
cZcX
ZX c
XZ c
X=Cases; Z=Eligible; Y=Workload
cY
Growth Rate in Experimental Rats
• Dependent Variable– GROWTH: growth rate
• Independent Variables– DOSE: dosage of a dietary supplement– DOSESQ: dosage squared
Hypothetical Data Source: SAS Institute Inc., SAS/STAT User’s Guide, Version 6, Fourth Edition, Volume 2, SAS Institute Inc., Cary, NC, 1989 (page 1438).
Correlations Growth Rate in Experimental Rats
Dose DoseSq
Growth .186 .359(0.608) (0.309)
Regression AnalysisGrowth Rate in Experimental Rats
Analysis of Variance Section
Sum of Mean ProbSource DF Squares Square F-Ratio LevelModel 2 655.706 332.853 51.555 <0.001Error 7 45.1938 6.45626Total(Adj.) 9 710.9 78.9889
Root Mean Square Error 2.541 R-Squared 0.936
Regression Analysis Growth Rate in Experimental Rats
Regression Equation Section
Indep. Regression Standard T-Value ProbVariable Coefficient Error (Ho: B=0) LevelIntercept 35.65744 5.617927 6.3471 4E-04Dose 5.262896 0.558022 9.4313 3E-05DoseSq 0.12767 0.012811 9.966 2E-05
Correlations Growth Rate in Experimental Rats
Dose DoseSq
Growth .186 .359(0.608) (0.309)
Dose 0.983(<0.001)
cZ
cX
cY
ZX c
XZ c
X=Dose; Z=DoseSq; Y=Growth
Unique Contribution of an Independent Variableto the Total Variation (CSS)
“When independent variables are correlated, there is no unique sum of squares which can be ascribed to an independent variable as reflecting its effect in reducing the total variation in Y. The reduction in total variation ascribed to an independent variable must be viewed in the context of the other independent variables in the model…”
- Neter & Wasserman
Neter and Wasserman, Applied Linear Statistical Models,Richard D. Irwin, Inc., 1974 (page 253).
cZcX
XZ c
Contribution of an Independent Variableto the Total Variation (CSS)
cYType 3:
Partial SSThe SS ascribed to the direction in the XZc-plane orthogonal to X.
This is the reduction in
error SS when Z joins X in the
model.
Type 1 SSThe SS
ascribed to the direction of Z.
This is the regression SS when Z is the only variable in
the model.
cZcX
XZ c
Contribution of an Independent Variableto the Total Variation (CSS)
cY
W
cX
XZ c
Contribution of an Independent Variableto the Total Variation (CSS)
cY W XZc-plane
W
cX
XZ c
Contribution of an Independent Variableto the Total Variation (CSS)
cYGiven Xc and
the XZc-plane, there is no
unique direction in the
plane which can be
ascribed to this SS.
W XZc-plane
cX
Variance Inflation Factor (VIF)
cX̂ cY
cZcX
Variance Inflation Factor (VIF)
cX̂
cY
ab
)cos1(1
2
abVIF
VIF: Variance Inflation Factor
2
2
2
1
2
( )1
1( )
1
i
x i
n
x ii
ii
VarSS R
SS X X
VIFR
β
β