# TB8/9 RM Overview

HideShow resource information
• Created by: mint75
• Created on: 16-01-16 15:59
• TB8 RM Overview
• Lecture 1; One-way ANOVA
• T-Tests
• T-Tests allow you to compare 2 means/ *two models*
• For t-tests, the data is categorical as opposed to continuous on x-axis
• The regression version
• Formula
• A  = B0 + B1 x G + eta
• A=The DV for each condition
• G = The 'code' for each condition
• Two conditions are given a code (0 or 1), called 'dummy variables'
• Control is usually = 0
• F-Statistic used to assess regression models appropriate-ness
• However, T-Tests can ONLY COMPARE 2 MEANS
• ANOVA
• Can compare several means
• 'Omnibus test'
• Tests for ANY difference between groups and IF the group means are different
• Does NOT tell you WHICH means differ
• Output
• Homogeneity of variance tests
• Homo-skedasticity
• Errors
• Type 1 and 2 errors
• Familywise errors
• Post-Hoc Tests and Planned Contrasts
• Control for the familywise error rate
• Bonferri method (simplest post hoc)
• Alpha level / Number of tests
• REGWQ, Tukey HSD (standard ones), Gabriels, Hochberg GT2, Games-Howell
• Different criterion!
• Post hoc tests look for the mean differences and significances
• Lecture 2; Planned Contrasts and 2-way Independent ANOVAs
• ANOVA by hand
• The F Ratio
• Theory of ANOVA
• Total variance in the data = Variance explained by the model and unexplained variance
• The steps for ANOVA by hand
• 2 Way Independent ANOVA
• 2 Independent Variables
• Diferent pps in ALL conditions
• Several independent variables = a factorial design
• Can look at how variables interact
• By hand!
• Steps involved
• Can be worked out as long as you have the;
• Group means
• Group STDDEVs
• Number of items
• Planned Contrasts
• Explains how is the variance partitioned?
• Because ANOVA is an omnibus test and doesn't tell you which variable is explaining the variance between groups
• Rules when CHOOSING  contrasts
• Independent chunks
• Only 2 chunks
• K - 1 aka you should always have one less contrast than the number of groups
• Control groups; the first contrast should always be a comparison between CONTROL and EXPERI-MENTAL
• Rules when CODING contrasts
• Groups coded with + weights compared to groups coded with - weights
• The sum of weights for a comparison should be 0
• If a group is NOT involved in a comparison, code it as 0
• For a given contrast, weights in one chunk should = the opposite chunk
• If a group is 'singled out' in a comparison, it should NOT be used in further comparisons
• e.g Helmert Contrast
• Effect Sizes
• Eta2, same as R2
• Partial Eta2; The proportion of variance that a variable explains that is NOT EXPLAINED by other variables
• Lecture 3; Repeated Measures ANOVA
• Violation of assumption of independence = core problem
• (Assumption of) Sphericity
• "Variances in the differences between conditions are equal"
• Deviations in sphericity = not enough DFs
• Correct to make them smaller using 'correction factor'
• Correction factors (multiply DF by)
• Greenhouse-Geisser estimate (CON-SERVATIVE)
• Huynh Feldt Estimate (LIBERAL)
• Lower bound estimate
• Mauchlys test
• Lecture 4; Mixed ANOVA
• Reduction v.s refinement
• Use more subjects than pure repeated measures
• When 1 or more IV uses the same pps, and 1 or more uses different pps
• Interactions are important
• Also generate an F-ratio alongside the F for the main effect
• Errors
• Type 1 and 2 errors
• Familywise errors