What 4 things are required for establishing causality?

Theoretical credibility, co-variation, elimination of confounds and temporal precedence.

Describe necessary causation.

If you see Effect(B), you must be at level B.

Describe sufficient causation.

If you see Effect(B), you may be at level B.

Describe Simpson’s paradox.

Where a correlation seems to go one direction, but adding a factor means there are many correlations, possibly in a different direction.

What is observed in 2-way ANOVA?

Grand mean + F1 effect + F2 effect + Interaction effect + error.

What does adding a factor that matters do?

Decrease SS, df and error term.

What does adding a factor that doesn’t matter do?

No decrease in SS, decrease in df, increase in error term and power, possible loss of significance.

What is used to control for multiple t tests?

The overall effect size given by R squared.

What is eta-squared?

The proportion of all variability explained the effects/interactions.

What is partial eta-squared?

The proportion f the variance that could be explained, that the effect/interaction explains.

What does including participants as a random factor allow us to do?

Account for variability in a dataset due to individual differences.

Why do we calculate the average across participant means rather than the average across trials?

We usually want to talk about ”people on average” rather than the average trial.

What makes up a nesting effect?

The mean for one level of nested factor 2 - (grand mean + one level of factor 1 effect).

What is observed in a nested ANOVA?

Grand mean + F1 effect + F2(F1) effect + error.

How can bivariate data be analysed?

Pearson’s correlation and simple linear regression/1-way ANOVA.

How can multivariate data be analysed?

Partial and multiple correlation and multiple linear regression/n-way ANOVA/ANCOVA.

What does partial correlation ask about?

The partial variance of X3.

What does semi-partial correlation ask about?

The total variance of X3.

Give 2 key aspects of partial correlation.

It is a “pure” measure uncontaminated by other variables and its uniqueness of variance makes it theoretically simpler.

Give 2 key aspects of semi-partial correlation.

It has a more intuitive baselines and it allows easy comparison of coefficients because X3 is constant.

How can ANCOVA be used to control for a variable?

You can add an extra variable, and use a regression line to estimate when X is the same.

What is observed in an ANCOVA?

Grand mean + factor effect + participant effect + individual differences + noise.

How can a covariant remove unwanted variability?

If some DV variability is attributed to the covariate it is removed from the error sum of squares.

When can adding a covariate increase degrees of freedom?

When the covariate is not related to the DV.

When can adding a covariate reduce power?

When it is co-linear with the first.

What assumptions does ANCOVA share with ANOVA?

Normality of error distributions, homogeneity of error variance and independence of “error term.”

What assumptions does ANCOVA share with regression?

Interval/ratio scale data and linear relationship.

Give 3 reasons to use ANCOVA compared to ANOVA.

Accounts for more variability, “controls for” other variables and quantifies which subject characteristics are important.

Give 3 things a good covariate should be.

Related to the dependent variable, unrelated to the other independent variables and unrelated to the other covariates.