You are on page 1of 35

The Concept of Analysis of Co-variance

BY . PRASAD JADHAV

What exactly a Analysis of Covariance?

Covariance is a measure of how

much two variables change together and how strong the relationship is

between them..

ANCOVA is really ANOVA with covariates" or, more simply, a combination of ANOVA and regression used when you have some categorical factors and some quantitative predictors.

Analysis of covariance may be looked upon as a special

provision or procedure for exercising necessary statistical


control over the variable or variables that have been left uncontrolled at the start of the experiment or study on account of practical limitations and difficulties associated with the conducting such experiments or studies.

By using this technique, we try to partial out the side effects if any on our study due to lack of exercise proper experimental control over the intervening variables, after having conducted the actual study of covariance.

Assumptions Underlying of Covariance

The method of analysis of covariance requires some basic assumptions for its application. 1. The dependent variable which is under measurement should be normally distributed in the population. 2. The treatment groups should be selected at random from the same population. 3. Within-groups variances must be approximately equal 4. The regression of the final socres (Y) on initial scores (X) Should be basically the same in all groups 5. There should exist a linear relationship between X & Y.

We will illustrate ANCOVA with a Example.


Three groups of five students each (Randomly selected from a Class VIII of a school) were initially rated for their leadership qualities, and their scores were recorded. Then they were subjected to different treatments ( Three different approaches or training techniques for leadership), and after 15 days of such training, they were again evaluated for their leadership qualities, and these final scores were also recorded. The data so collected are given in the next slide

Group A Initial values X1 4 3 5 2 1 Final scores Y1 6 5 6 4 4

Group B Initial values X2 8 6 7 4 5 Final scores Y2 8 5 9 7 6

Group C Initial values X3 7 9 10 12 12 Final scores Y3 9 8 11 12 15

Use this data to compare the relative effectiveness of the treatments given to the groups

Solution.
There were many observable differences amongst the initial scores of the three groups, but no attempt was made to make these groups as equivalent groups at the start of the study, i.e .before subjecting these to different treatments. In absence of such an experimental control, the researcher was forced to exercise statistical control by applying the technique of analysis of covariance. The procedure may be understood through computation . Let us begin with arranging the given data .

Group A X1 4 3 Y1 6 5 X1Y1 24 15 X21 16 9 Y21 36 25 X2 8 6 Y2 8 5

Group B X2Y2 64 30 X22 64 36 Y22 64 25 X3 7 9 Y3 9 8

Group C X3Y 3 63 72 X23 49 81 Y 23 81 64

5
2 1 S U M M S

6
4 4

30
8 4

25
4 1

36
16 16

7
4 5

9
7 6

63
28 30

49
16 25

81
49 36

10
12 12

11
12 15

110
144 180

100
144 144

121
144 225

15

25

81

55

129

30

35

215

190

255

50

55

569

518

635

10

11

For all the three groups,


X = X1 + X2 + X3 = 15 + 30 + 50 = 95 Y = Y1 + Y2 + Y3 = 25 + 35 + 55 = 115

X2 = X21 + X22 + X23 = 55 + 190 + 518 = 763


Y2 = Y21 + Y22 + Y23 = 129 + 255+ 635 = 1019 XY = X1Y1 + X2Y2 + X3Y3 = 81+ 215+ 569 = 865
After Computing various sums and means thus, the following steps can be adopted

STEP 1.

Computation of correction terms (Cs) .


Different corrections are applied to different sums of Sqaures as in the case of analysis of variance. These can be computed by using the formulae shown in next slide.

For all the three groups,


i. Cx = ( X)2
N
=

95 x 95
15

601.67

ii. Cy= ( Y)2


=

115 x 115
=

881.67

15

iii. Cxy = X Y
=

95 x 115
=

728.33

15

STEP 2.

Computation of total sum of squares (total SS)

Computation of total sum of squares (total SS)


SSX = X2 CX = 769 601.67 = 161.33
SSY = Y2 CY = 1019 881.67 = 137.33 SSXY = XY CXY = 865 728.33 = 136.67

STEP 3.

Computation of Sum of squares (SS) among the means of the groups.

Computation of Sum of squares (SS) among the means of the groups.


i) SS Amongst Means for X =

X21
N1

X22
N2

X23
N3

-Cx

15 2 + 30 2 + 50 2
5

- 601.67

123.33

Computation of Sum of squares (SS) among the means of the groups.


i) SS Amongst Means for y =

y 21
N1

y 22
N2

y 23
N3

-Cy

25 2 + 35 2 + 55 2
5

- 881.67

93.33

Computation of Sum of squares (SS) among the means of the groups.


i) SS Amongst x y 1 1 Means for xy =
N1
=
+

x 2y 2 +
N2

x 3y 3
N3

- Cxy

15x25 +30x35 + 50 x 55
5
=

- 728.33

106.67

STEP 4

Computation of Sum of squares (SS) with in group

Computation of Sum of squares (SS) with in group


(i) With in groups SS For X = SSx SS among-means for x = 161.33 123.33 = 38 (ii) With in groups SS For Y = SSy SS among-means for y = 137.33 93.33 = 44 (iii) With in groups SS For XY= SSxySS among-means for xy = 136.67 106.67 = 30

STEP 5

Computation of the number of degree of freedom

Computation of the number of degree of freedom


i. Amongst means (df) = K-1 = no of groups = 3 1 = 2

ii. Within groups (df) = N-K = 15 3 = 12

STEP 6

Analysis of variance of X & Y scores taken saperately

Source Df of variation

SSx ( Sum of squares for X)

Ssy ( Sum of squares for Y)

MSx ( Mean square variance for X or X 123.33/ 2= 61.66 38/12 = 3.17

MSy ( Mean square variance for y or y 93.33/2 = 46.66 44/12 = 3.67

Amongst 2 -means Withingroups total 12 14

123.33

93.33

38 161.33

44 137.33

Mean square variance of among-groups (for x) Fx = = Mean square variance of within-groups 61.66 3.17 Mean square variance of among-groups (for y) Fy = Mean square variance of within-groups 46.66 = 12.71 = 19.45

3.67
Where Fx = F ratio for X Scores Fy= F Ratio for Y Scores

From Table R of the Appendix for df (2, 12), We can have the critical value of F At 0.05 level = 3.88 and at 0.01 level = 6.93

Computation of adjusted sum of squares (SS for y i.e. SS for x) The initial differences in the groups X scores may cause variablility in their final scores measured after giving treatment. It needs to be checked and controlled. For this purpose, necessary adjustments are made in various sum of squares (SS) for Y by using the following general formula.

STEP 7

Computation of adjusted sum of squares (SS for y i.e. SS for x)

Rule : If Fx is significant the H0 is to be rejected showing that initially groups were different. Hence covariance is needed. If not significant we can have only analysis of variance. The computed value of of F for X scores is significant at both the levels, and similar is the case with the computed F for Y Scores. Hence H0 for X Scores as well as Y scores are rejected, leading to the conclusion that (i) There is significant difference in intial (X) Scores and (ii) there are significant difference in final (Y) Scores
= SS YX= Ssy (SS xy)2 SSx

(Here SSyx stands for the sum of squares of Y adjusted for X difference.)

The Specific adjusted SS for Y May be computed as follows : (a)Adjusted Sum of squares for total , i.e.,
SSYx (Total) = SSy Total (SS xy)2 total SSx (136.67)2 161.33

= 137.33-

= 137.33 115.77 = 21.56

(b)Adjusted Sum of squares for within-group means, i.e.,


SSYx (within-means) = SSy = 44 within (SS xy)2 within SSx (30)2 38

= 44 23.68= 20.32

(c)Adjusted Sum of squares for among-group means, i.e.,


SSYx (among-means) = Ssyx (Total ) - Ssyx (within) = 21.56 20.32 = 1.24

STEP 8

Computation of Analysis of Covariance :

Computation of Analysis of Covariance :


It is carried out as shown in the table
Source of variation Msyx or Vyx

df

SSx

SSy

SSxy

SSyx

Sdyx

AmongGroup means Within group means Total

123.33

93.33

106.67

0.44

0.22 1.76 1.76

11* 13

38 161.33

44 137.33

30 136.67

20.32 20.76

=1.32

(*1 df is lost because of regression of Y on X)

(
Fyx = Vyx (among) Vyx (within)

= 0. 22 1.76

=0.125

From Table R of the appendix for df (2,11) Critical F at 0.05 level = 3.98 And Critical F at 0.01 level = 7.20 The computed value of F(Fyx) is not significant at both the levels. Hence H0 is to be accepted with the conclusion that groups do not different significantly after giving treatments . Hence out of the three treatments cannot be termed better than another.

You might also like