Lecture 9 - ANCOVA Flashcards
When do we use an ANCOVA?
When we can’t experimentally control for error - ANCOVA can do this statistically
How does ANCOVA statistically control for error?
It adjusts the analysis in two ways:
1 - reducing the estimates of experimental error
2 - adjusting treatment effects with response to the covariate
We collect data on the covariate, usually before the experimental treatment
How does ANCOVA actually work?
It uses linear regression to estimate the effect of treatment, given the covariate information.
We look at how scores deviate from the regression line (instead of looking at how far the scores are from the mean, which is what we do in an ANOVA)
What are the assumptions of an ANCOVA
Same as ANOVA
- normality
- independence
- equal variances
- random sampling
Two ANCOVA specific assumptions
- assumption of linear regression
- assumption of homogeneity of regression coefficients
What does the assumption of linear regression mean?
Assuming that all the deviations from the regression equation have normal distributions, means of 0 and homoscedacity
If this isn’t true then the ANCOVA is basically useless
How do we test for homogeneity of regression coefficients
We look at the interaction between the IV and the covariate
What are the two main limitations of ANCOVA
- only a very small amount of covariates
- covariates must be independent of treatment (this is why we collect covariate data before treatment begins)
