You are on page 1of 34

Analysis of Covariance

ANCOVA
Uses of ANCOVA

 Two-group pre-test/post-test design


 The scores on the pre-test are treated as a covariate to ‘control’ for pre-existing
differences between the groups.
 This makes ANCOVA very useful in situations when you have quite small sample
sizes and only small or medium effect sizes
Choosing appropriate covariates

 ANCOVA can be used to control for one or more covariate at the same time
 The variables that you choose as your covariates should be
 Continuous variables
 Measured reliability
 Correlate significantly with dependent variables

 The covariate must be measured before the treatment or experimental


manipulation is performed.
Assumptions of ANCOVA

 Each group sample is drawn from a normally distributed population


 Within each sample, the observations are sampled randomly and independently of each other
 the variances of two populations are equal.
 Levene-test (p>.05)
 Correlations among the covariate
 If you are using more than one covariate, make sure the correlation to one another is not too strong correlated.
 If only one covariate, do not need to do this
 Linearity
 Scatter plot – the relationship between the DV and your covariate is linear
 Homogeneity of regression slope
 Relationship between the covariate and DV for each group is the same
 Similar slopes on the regression line for each group (p>.05)
One-Way ANCOVA

 Involves
 One Independent variables (two or more levels/conditions/groups)
 One dependent variables
 One or more continuous covariates

 Null hypothesis
 There is no significant difference in (Dependent variables) for independent variables
(level 1) and (level 2), while controlling for their pre-test scores.

 Alternative Hypothesis
 There is a significant difference in (Dependent variables) for independent variables
(level 1) and (level 2), while controlling for their pre-test scores.
Hypothesis
Case 1

 A one-way between-groups analysis of covariance was conducted to compare the


effectiveness of two different interventions designed to reduce participants’ fear
of statistics.
 The independent variable was the type of intervention (maths skills, confidence
building), and the dependent variable consisted of scores on the Fear of Statistics
Test administered after the intervention was completed.
 Participants’ scores on the pre-intervention administration of the Fear of
Statistics Test were used as the covariate in this analysis.
Procedure for one-way ANCOVA

 Click Analyse, then click on General Linear Model, then on Univariate


 In the DV box, put your DV
 In the Fixed Factor box, put your IV
 In the Covariate box, put your covariate
 Click on the Model button. Click on Full Factorial in the Specicy Model section.
Click on Continue
 Click on the Options button
 In the top section labelled Estimated Marginal Means, click on your IV
 Click on the arrow to move it into the box labelled Display Means for. This will
provide you with the mean score on your dependent variable for each group,
adjusted for the influence of the covariate.
 In the bottom section of the Options dialogue box, choose Descriptive statistics,
Estimates of effect size and Homogeneity tests.
 Click on Continue and then OK.
Linearity

The relationship is clearly linear, this assumption is


fulfilled
Homogeneity of regression slope

 Homogeneity of regression slope


 Relationship between the covariate and DV for each group is the same
 Similar slopes on the regression line for each group (p>.05)

P>.05, p= 816, this assumption is fulfilled


SPSS OUTPUT

Descriptive Statistics
SPSS OUTPUT

 Leven’s Test of Equality of Error Variance

P>.05, the variance are equal.


SPSS OUTPUT

 Main ANCOVA results - Test of Between-Subjects Effects

Source : group : F(1,27) = .763, p=.39, partial eta squared = .027, the result is not significant.
There is not a significant difference in the Fear of Statistics Test scores for participants in the maths skills
group and the confidence-building group after controlling for scores on the Fear of Statistics Test
administered prior to the intervention.
Partial eta squared

 Partial Eta Squared value


 The value in this case is only .027 (a small effect size according to Cohen’s 1988
guidelines).
 This value also indicates how much of the variance in the dependent variable is
explained by the independent variable.
 In this example, we are able to explain only 2.7 per cent of the variance.
 This indicates whether there is a significant relationship between the covariate and the dependent variable while
controlling for the independent variable (group).
 In the line corresponding to Fost1 (our covariate- Fear of Statistics-pre-test), you will see that the Sig. value is .000
(which actually means less than .0005).
 This is less than .05, so our covariate is significant. Partial eta square (.75). It explained 75 per cent of the variance in
the dependent variable.
SPSS OUTPUT

 Estimated Marginal Means

• with the adjusted means on the dependent variable for each of our groups.
• ‘Adjusted’ refers to the fact that the effect of the covariate has been statistically removed.
Presenting the results

 After adjusting for pre-intervention scores, there was no significant difference


between the two intervention groups on post-intervention scores on the Fear of
Statistics Test, F (1, 27) = .76, p = .39, partial eta squared = .03. There was a
strong relationship between the pre-intervention and post-intervention scores on
the Fear of Statistics Test, as indicated by a partial eta squared value of .75.
Example…

 A teacher want to investigate whether three different teaching methods such as


cooperative learning, problem-based learning and inquiry-based learning affect
students’ science achievement. S/he knows that students’ achievements in pre-
test would affect their science achievements in their post-test.
 Steps
 State the null and alternative hypotheses
 Set the significance level for rejecting the null hypothesis
 Check the assumptions
 Analyse the data using SPSS
 Make a decision about the null hypothesis
 Draw a conclusion
Step 1

 Null Hypothesis
 There is no significant difference in science achievement for cooperative learning,
problem-based learning and inquiry-based learning cooperative learning, while
controlling for their pre-test scores.

 Alternative Hypothesis
 There is a significant difference in science achievement for cooperative learning,
problem-based learning and inquiry-based learning, while controlling for their pre-
test scores.
Step 2

 Set α = .01,
 If p ≤ .01, reject the null hypothesis.
 If p > .01, fail to reject the null hypothesis.
Step 3

State the assumptions


Step 4 : SPSS DATA

 Data
 Run the SPSS
 Check the assumptions.
 What is the result?
Step 5

 Make a decision about the null hypothesis

p< .05, reject null hypothesis


Step 6

 Draw a conclusion
 After adjusting for pre- test scores, there was a significant difference between the three
intervention groups on post-test scores on the students’ science achievement, F (2, 19) =
6.96, p = .005, partial eta squared = .42. There was a strong relationship between the
pre-test and post-test scores on the students’ science achievement, as indicated by a
partial eta squared value of .77.
Another example…

 SPSS Output.
 Mathematics anxiety and types of tests (MCQ and Structured test)
 In this example, you want to explore the impact of MCQ (Group 1) and
Structured test (Group 2) on participants’ scores on mathematics anxiety, while
controlling for the scores on this test administered before the program.
 Steps
 State the null and alternative hypotheses
 Set the significance level for rejecting the null hypothesis
 Check the assumptions
 Make a decision about the null hypothesis
 Draw a conclusion

You might also like