Assumptions Flashcards
Homogeneity of Variance
Populations being compared have similar error variance. Tested using Levene’s Test, Spread vs Level Plot.
Normality
Results fall in a normal distribution around the mean. Tested using a histogram or skewness/kurtosis statistics.
Independence of Observations
Each observation (score) is independent of the others. Controlled by using a between-subjects design and/or only having subjects participate once.
Sphericity
Variance of difference between treatment means is constant. Usually violated with 3+ IV levels. Tested using Mauchly’s Test and controlled using a Greenhouse-Geisser Correction.
Sample Size
How small or large a given sample is. In regression, sample size must be over 100 participants.
(Multi) Collinearity
A high correlation between predictors (above .8). Tested using Tolerance and VIF statistics. To control, remove one of the high-correlating variables (statistically or manually).
Linearity
Linear relationship between the IV/s and DV/s.
Homoscedacity
Error variance should be the same at each level of the predictor variable (variability in one variable should be the same at all values of the other variables). Tested using Levene’s test on parts of the data or plots. Plot should have shapeless cloud of data points.
Multivariate Normal
Means of DVs and all linear combinations must be normal (no outliers). Tested using plots and equal sample sizes. To control remove outliers from data.
Homogeneity of variance-covariance matrices
Homogeneity of variance and correlation between DVs must be the same in each group. Tested using Box’s M, however is sample size is equal this should be disregarded.
Independence of Errors
Error terms should not be correlated and should be normally distributed. Tested using a histogram of residuals and a probability plot. Points should be close to the line.
Linearity of the Logit
Assumption of logistic regression. There must be a linear relationship between the continuous IV and the log of the DV (the logit).
