Analysis of Covariance

Analysis of Covariance
Analysis of Covariance (ANCOVA) is a statistical technique that combines elements of both
analysis of variance (ANOVA) and regression. It is used to compare group means while statistically
controlling for the effects of one or more continuous variables that are not of primary interest,
known as covariates.

The major objective of a covariance analysis is to obtain a reduced estimate of experimental error
by taking into account the regression.

The concomitant variate, not itself of experimental interest, will be called covariate and designated
as X. The other variate, which is of experimental interest, will be termed the criterion and
designated as Y.

Thus, in the design we obtain two observations (X and Y) from each subject, one before applying
the treatment (X) and the other after the treatment (Y).
Key components of ANCOVA:

1. Dependent Variable (DV): This is the outcome variable that you are trying to
analyze and compare across different groups.

2. Independent Variable (IV): This is the categorical variable that defines the groups
you are comparing. It could represent different treatment conditions, groups, or levels.

3. Covariate(s): These are continuous variables that are not the primary focus of the
study but are included in the analysis to control for their potential influence on the
dependent variable. ANCOVA aims to remove the variance associated with these
covariates from the dependent variable, allowing for a more accurate assessment of
group differences.
4. Assumptions:

● Linearity: The relationship between covariates and the dependent variable is

linear.
● Homogeneity of Regression Slopes: The effect of the independent variable on
the dependent variable is consistent across different levels of the covariate(s).
● Homogeneity of Variances: The variance of the dependent variable is equal
across all levels of the independent variable.
 Numerical Example
An experimenter was interested in studying the effect of two types of reinforcement in a
maze learning task. The experimental units consisted of rats and the experimenter decided
to have a control group with no reinforcement.

Thus, three groups of rats were studied – control group, experimental 1 and experimental 2.
The experimenter did not have the choice to assign the subjects at random to three
conditions. However, the groups were randomly assigned to the three conditions.

Before applying the experimental treatment, performance in terms of number of errors in

running the maze was obtained from each experimental unit. The scores on this trial
constituted the covariate measure X. After obtaining the X measures, trials on the same
maze were run with reinforcement (treatment) in two groups and no reinforcement in the
control group.

In each group 5 rats were taken.

 Control Experimental 1 Experimental 2
S.no
X Y X Y X Y

1 5 4 3 2 4 3

2 4 4 2 1 4 2

3 4 5 2 2 3 2

4 5 3 2 1 5 3

5 6 5 4 3 2 1

Total 24 21 13 9 18 11

Mean 4.8 4.2 2.6 1.8 3.6 2.2

Total X = 55; total Y = 41

Partitioning of Total Variation and Df for ANOVA

Total
kn - 1
3 x 5 - 2 = 13

Treatment Between Treatment Within

k-1 k(n - 1)
3-1=2 3(5 -1)
3 x 4 = 12

Here, k = 3 (number of treatments)

n = 5 (number of subjects in each group)
Partitioning of Total Variation and Df for ANCOVA

Total
kn - 2
3 x 5 - 2 = 13

Treatment Between Treatment Within

k-1 k(n - 1)-1
3-1=2 3(5 -1) -1
3x4-1
12 - 1 = 11

Here, k = 3 (number of treatments)

n = 5 (number of subjects in each group)
 Computation
ANOVA of X Scores

1. Total SS = (5² + 4² + 4² +..................+ 2²) – (55)² ÷ 15

= 225 - 201.67 = 23.33

1. Between Group SS = (24²/5 + 13²/5 +18²/5) - (55)² ÷ 15

= 213.8 - 201.67 = 12.13

1. Within Group = Total – Between

= 23.33 – 12.13 = 11.20

Summary of ANOVA of X Scores

Source of SS df MS (SS/df) F (MS

variation between/MS
within)

Between 12.13 2 12.13/2 = 6.06 6.06/0.93 =

6.51*

Within 11.20 12 11.20/12 = 0.93

Total 23.33 14

Tabled value at 0.05 level for df 2 and 12 is 3.80 and our observed value 6.51 is greater
than the critical value (3.80). Though it is observed that the three groups differ
significantly on the covariate measure X.
ANOVA of Y Scores
1. Total SS = (4² + 4² + 5² +..................+ 1²) – (41)² ÷ 15

= 137.0 - 112.07 = 24.93

1. Between Group SS = (21²/5 + 9²/5 +11²/5) - (41)² ÷ 15

= 128.6 - 112.07 = 16.53

1. Within Group = Total – Between

= 24.93 – 16.53 = 8.40

Summary of ANOVA of Y Scores

Source of SS df MS (SS/df) F (MS

variation between/MS
within)

Between 16.53 2 16.53/2 = 8.265 8.265/0.7 =

11.8**

Within 8.40 12 8.40/12 = 0.7

Total 24.93 14

Tabled value at 0.01 level for df 2 and 12 is 6.93 and our observed value 11.8 is greater
than the critical value (6.93). Though it is observed that the three groups differ
significantly on the criterion measure Y. This reflects the differences in the three groups
on the variable of Y (criterion).
Analysis of Covariance
Total Sum of Products = [(5x4) + (4x4) + (4x5) +...........(2x1)] - (55x41) ÷ 15

= 170.0 – 150.33 = 19.67

Sum of Products Between Groups = (24x21/5 + 13x9/5 + 18x11/5) - (55x41)÷15

= 163.80 –150.33 = 13.47

Sum of Products Within Groups = Total – Between

= 19.67 – 13.47 = 6.20

Summary of Sums of Squares of X and Y and Sum of Products

Total Between Within

Sum of Squares Y 24.93 16.53 8.40

Sum of Squares X 23.33 12.13 11.20

Sum of Products XY 19.67 13.47 6.20

Now, further computations are made for the purpose of correcting the Y scores for differences in X scores (means
of final Y scores after ruling out the initial difference in X scores).

Formula for obtaining the adjusted total and within sum of squares for Y scores.

Adjusted total sum of squares of Y scores = SSyt – (SPt)²/SSxt

Adjusted Within Sum of Squares of Y scores = SSyw – (SPw)²/SSxw

Here, SSyt is total sum of squares of Y scores

SPt is total sum of products

SSxt is total sum of squares of X

SSyw is within sum of squares of Y scores

SPw is within sum of products

SSxw is within sum of squares of X

Total adjusted SS for Y = SSyt – (SPt)²/SSxt

= 24.93 – (19.67)²/ 23.33

= 24.93 –16.58 = 8.35

Within Group adjusted SS for Y = SSyw – (SPw)²/SSxw

= 8.40 – (6.20)²/11.20

= 8.40 – 3.43 = 4.97

Between Group Adjusted SS for Y = Total – Within

= 8.35 – 4.97 = 3.38

 Summary of the Covariance Analysis
Source of SS df MS (SS/df) F (MS
variation between/MS
within)

Between 3.38 2 3.38/2 = 1.69 1.69/0.45 =

Groups 3.75

Within Groups 4.97 11 4.97/11 = 0.45

(error)

Total 8.35 13

Tabled value at 0.05 level for df 2 and 11 is 3.98 and our observed value 3.75 is less than the critical value
(3.98). Though our F value is considered non significant.
Thus, analysis of covariance of the adjusted Y scores does not show a significant difference between the
groups.
It has been observed that, the three groups differ significantly in respect of X as well as Y scores. However,
when the Y scores are adjusted, the groups no longer show significant difference after giving treatments
(reinforcements). Hence, out of the three treatments, one treatment cannot be termed better than another.
It appears that the former differences of Y scores was simply reflection of differences in X scores not due to the
treatments (reinforcement) that was given to the rats.