PS2010 Lecture 10 PDF
Document Details
Uploaded by FragrantCynicalRealism
Royal Holloway, University of London
Tags
Summary
This lecture discusses non-parametric statistical analysis methods such as Kruskal-Wallis, Wilcoxon, and Mann-Whitney tests. It contrasts these with parametric methods, highlighting the assumptions and limitations of each approach.
Full Transcript
Non-parametric approaches: Non-parametric analyses: - Do not rely on too many rules or assumptions. - Fairly limited use for complex designs. We'd prefer to run parametric analyses. Independence of observations -- measurements are not influencing one another. Assumptions of parametric anal...
Non-parametric approaches: Non-parametric analyses: - Do not rely on too many rules or assumptions. - Fairly limited use for complex designs. We'd prefer to run parametric analyses. Independence of observations -- measurements are not influencing one another. Assumptions of parametric analyses -- independence of observations -- measurements are not influencing one another. Interval or ratio level data -- data being analysed presents as a score along continuum. Not categorical or ordinal -- ranked. For factorial ANOVAs, the model is generally robust to minor deviations from normality. For one-way ANOVAs and t-tests, there are alternative tests you could use instead. t-test -- the dependent variable should be normally distributed -- independent t-test, the raw dependent variable data. Repeated t-test -- the difference scores. How many hours do you revise -- psychology, geography, biology- iv DV -- ave number hours spent revising -- during exam period. 1 -- independent variable with three groups. - Run ANOVA and then evaluate model by checking the assumptions. - If they don't meet the assumptions, you may need to use a non-parametric For non-parametric tests -- use median -- extreme values can skew mean when data is skewed. Homogeneity of variance -- Levene's test Kruskal-Wallis, Wilcoxon, Mann- Whitney.
