Analysis of Covariance

By:
Haftom Temesgen (PhD)
Mekell University
 Analysis of Covariance
A procedure for comparing treatment means that
incorporates information on a quantitative explanatory
variable, X, sometimes called a covariate.
The procedure, ANCOVA, is a combination of ANOVA
with regression. Combines linear regression and ANOVA
Can be used to compare g treatments, after controlling for
quantitative factor believed to be related to response (e.g. pre-
treatment score)
Can be used to compare regression equations among g groups
(e.g. common slopes and/or intercepts)
Model: (X quantitative, Z1,...,Zg-1 dummy variables)
E(Y ) X 1Z1 g 1Z g 1
2 Haftom Temesgen (PhD) 1/18/2017
 Tests for Additive Model
Relation for group i (i=1,..., g-1):
E(Y)=a+bX+bi
Relation for group g: E(Y)=a+bX
H0: b1=...=bg-1=0 (Controlling for covariate, no
differences among treatments)

3 Haftom Temesgen (PhD) 1/18/2017

 Interaction Model
Regression slopes between Y and X are allowed to vary
among groups
E(Y ) X 1Z1 g 1Z g 1 1 XZ1 g 1 XZ g 1

Group i (i=1,...,g-1): E(Y)=+X+i+iX=(+ i)+( +i)X

Group g: E(Y)=+X

No interaction means common slopes: 1=...=g-1=0

4 Haftom Temesgen (PhD) 1/18/2017

 Inference in ANCOVA
Model: E(Y ) X 1Z1 g 1Z g 1 1 XZ1 g 1 XZ g 1
Construct 3 sets of independent variables:
{X} , {Z1,Z2,...,Zg-1}, {XZ1,...,XZg-1}
Fit Complete model, containing all 3 sets.
Obtain SSEC (or, equivalently RC2) and dfC
Fit Reduced, model containing {X}, {Z1,Z2,...,Zg-1}
Obtain SSER (or, equivalently RR2) and dfR
H0:1=...=g-1=0 (No interaction). Test Statistic: F

5 Haftom Temesgen (PhD) 1/18/2017

 Inference in ANCOVA (cont)
Test for Group Differences, controlling for covariate

E(Y ) X 1Z1 g 1Z g 1
Fit Complete, model containing {X}, {Z1,Z2,...,Zg-1}
Obtain SSEC (or, equivalently RC2) and dfC
Fit Reduced, model containing {X}
Obtain SSER (or, equivalently RR2) and dfR
H0: 1=...=g-1=0 (No group differences) Test
Statistic: F
6 Haftom Temesgen (PhD) 1/18/2017
 Inference in ANCOVA (cont)
Test for Effect of Covariate controlling for qualitative variable

E(Y ) X 1Z1 g 1Z g 1
H0:b=0 (No covariate effect) Test Statistic: t

b
tobs ^
b

7 Haftom Temesgen (PhD) 1/18/2017

 Adjusted Means
Goal: Compare the g group means, after controlling for the
covariate
Unadjusted Means: Y 1 ,..., Y g

' '
Adjusted Means: Y ,..., Y
1 g Obtained by evaluating
regression equation at X X

Comparing adjusted means (based on regression equation):

' ' ' '
bi Y Y i g bi b j Y Y j
i

8 Haftom Temesgen (PhD) 1/18/2017

Multiple Comparisons of Adjusted Means

Comparisons of each group with group g:

^
bi t C / 2, N g 1 bi i 1,..., g 1

Comparisons among the other g-1 groups:

b b t
^ 2 ^ 2
i j C / 2, N g 1 b b 2COV (bi , b j )
i j

Variances and covariances are obtained from computer

software packages (SPSS, SAS, Stata)
9 Haftom Temesgen (PhD) 1/18/2017