Week 4 ANCOVA from the Text Books Flashcards
To elucidate on our knowledge of ANCOVA from our learned text book writing scholars
What is ANCOVA in a nutshell?
ANCOVA is a more powerful version of ANOVA
• ANCOVA brings ANOVA & Correlation together
• ANCOVA is an extension to any type of ANOVA including MANOVA
What does ANCOVA do?
• ANCOVA statistically controls for a Covariate(s)
i.e. another variable that is related to the DV
o therefore ANCOVA removes the effect of background error variance
o actually changing scores on the DV
o The DV scores are adjusted to what we would predict they would have been if everyone had scored the same on the covariate
What is the F Ratio again?
• The F ratio is the ratio of between-groups variance divided by error variance
o Reducing the error variance (the denominator) produces a larger F ratio
- This larger F ratio has a greater likelihood of achieving significance
What is ANCOVA based on?
•ANCOVA is based on the correlation between the covariate(s) & the DV using regression to predict DV scores from the covariate(s)
o The part of the DV score that can be predicted by the covariate(s) is, literally removed from the observed DV score for each participant to produce an adjusted DV score
What’s an example of the partitioning of the variance?
In the case of a positive correlation: participants with high scores on a covariate would have their DV scores reduced somewhat whilst those with low scores on the covariate would have their DV scores increased somewhat – to remove the portion of the DV that can be accounted for by the covariate
I often hear that ANCOVA needs to be used with discretion. What are the ANOVA assumptions my results must adhere to in order to consider ANCOVA?
•assumes the independence,
normality and homogeneity of the variances
•require homoscedasticity,
I often hear that ANCOVA needs to be used with discretion. What are some of the assumptions my results must adhere to in order to consider ANCOVA?
- covariates = interval or ratio (dichotomous can work)
- The correlation between covariates & the DV should be significant
- Multiple covariates need to be chosen sparingly based on a sound theoretical understanding of their relevance. -CV’s should not be too highly correlated with one another
- The CV(s) should be normally distributed (in each group)
- There needs to be a linear relationship between the CV(s) & the DV (in each group)
- The CV(s) should be reliable
- There must be homogeneity of regression.
Why should the the correlation between covariates & the DV should be significant?
To ensure reliability of ANCOVA results
Why must there be homogeneity of regression?
There must be homogeneity of regression, that is, relationship between the covariate(s) & the DV must be the same across groups.
• This is because adjustments are made to the DV scores according to the overall correlation between the covariate(s) & the DV.
• Such adjustments would be nonsensical if the correlations were actually different across groups
When running a true experiment, the CV must not be related to the IV. Why is this?
Otherwise there is a confound – even if covariates are statistically removed using ANCOVA there is no way of knowing what else may be associated with them, hence causal inference is compromised
When running a true experiment, why do participants need to be measured on the covariate(s) before being exposed to the experimental manipulations?
To avoid the possibility of the manipulation also affecting the covariate
- Were this to happen, removing the effect of the covariate might also remove part of the effect of the IV
What are the rules when running Quasi experiments & Natural Group Designs?
- The covariate(s) must be measured prior to any experimental manipulation in Quasi experiments that use ‘intact’ groups
- For both quasi experiments & non-experimental natural group designs, the covariate(s) may well be related to the IV.
It is common for the covariate(s) may well be related to the IV when running Quasi experiments & Natural Group Designs. When does this happen and what can be done about it?
This can occur in quasi experiments when groups differ on the pre-test & also the post-test) & the pre-test is used as the covariate to statistically remove the effects of the pre-existing differences, thereby ‘equating’ the groups.
• Correlations between the covariate(s) & DV are common in natural group designs.
o In these cases ‘interpretation is fraught with difficulty’
o Researchers really need to know what they are doing to use ANCOVA appropriately & need to consult specialised texts beforehand
What is the word of caution Hills offers with regard to Populations & Individuals?
*The adjustment of DV scores are only ever predicted adjustments
Thus there is always a degree of error in such predictions
•Whilst ANCOVA is a powerful technique to help establish whether population groups differ overall (i.e., in their means), it tells us very little about what to expect with particular individuals
o This is true for all ANOVA designs, & to all statistical tests; though even more so for ANCOVA (I guess because we are removing the effect of background error variance
i.e., actually changing scores on the DV)
According to Andy Field, What are continuous or Covariates?
- Continuous variables predict the outcome / dependent variable
- Continuous or covariates are not part of the main experimental manipulation but have an influence on the DV