1/53
Looks like no tags are added yet.
Ways to estimate population variance
Sample variances by taking variance of each sample then averaging it
Sample mean variances by multiplying vairance of means with n (o2 = nox2)
F distribution
Probability curve of variance ratios assuming samples are from the same population
What does the F score mean?
F = explained/unexplained = MSgroup/MSerror
If F = 1
owithin and obetween are similar and they’re both samples from the same population
What happens when groups are from different populations?
obetween will be very difference, creating a large ratio
ANOVAs
Partitions variation into group and error variances to evalulate H0
Other terms for within group variances
Error, residuals
Other terms for between group variances
Groups, conditions
SStotal
Measures total variation/total squared spread of raw data
SStotal Steps
Calculate grand mean
Subtract grand mean from each score
Square
Sum
Differences of scores
Measures variation as it represents how far each number is far from grand mean
Grand Mean
Represents best guess at summarizing data
What is SStotal kinda equivalent to
Stnadard error of a sum, it will continusouly increase as N increase because it’s not averaged
Variance
Average square distance from mean
SSerror
Measures variation with groupos, the unexplained variation
SSerror Steps
Calculates group mean
Subtract the group mean from each score
Sum the square deviations
SS group
Number of variation that’s from difference between the mean/variation that researcher can explain from manipulation
SSgroup Steps
Replace each score within a group with its group mean (gets rid of all within var)
Subtract grand mean from each score
Sum the squares deviation
Mean Squares
SS/df = variance(mean) x n = mean(varerror) => variance
MSerror
Variances of deviations from group means regardless of position
MSgroup
Variances amongst group means with respect to position
Contrast
Specific way to compare group means by giving each group a number to indicate how much each group counts in a substraction problem to compare just the groups you care about
Linear Contrast
Breaks up SSgroup into smaller, specific pieces to look for speicific patterns of difference between groups to evaluate a specific hypothesis to focus statistical power on a particular group
Orthogonal Contrasts
Contrasts are mutually exclusive so that all weights and prod of coeff are 0
Why do the error values remain contrast in linear contrasts?
All data used to calculate regardless of which group because assumes groups are from the same population
k
Number of groups
N
Number of people
n
Number of people per group
t2
F
Why are linear contrasts used?
Focuses statistical power on a particular group because MS tends to be larger from less df so F test is more sensitive
Pairwise Tests
Compares each level of group variables with special t test
Tukey HSD
Evaluates all pairwise (variation of groups) differences between groups and gives CI, difference, p value
As # of tests increases,
Chances of getting a type I error compounds (1 - 0.95#)
How does Tukey HSD accounts for compounding type I errors?
Adjusts individual alpha level and t value so that overall/familywise alpha rate is 0.05
Planned Comparisons
Tests decided before research looks at data
Post-Hoc Comparisons
Tests decided after researcher looks at data and increases type I error because multiple informal tests have already been done
Scheffé Method
Increases Fcritical to control for type I error rate in post-hoc tests
Factorial ANOVAs
Tests the effect of 2+ conditions and how they interact by partitioning IV on DV (SSA,SSB), both IVs on DV (SSA + B) and error (SSerror)
Factorial
Going across the variables
Balanced Factorial Design
Equal n per cell/group
Marginal Mean
Mean of a level/category (condition mean)
SSA and SSB
Variation by factor A and B, equivalent to SSgroup
SSA or B Calculation
Calculate marginal mean
Replace each score with marginal mean for each group for that factor
Sub grand mean and sum squares
Degres of freedome for SSA or B
k - 1
SSinteraction
Variability of factor A and B combined that isn’t explained by either factors alones
SSinteraction Calculation
Calculate each specific cell mean
Replace every entry with its cell mean
Subtract grand mean
Square and sum
Subtract from SStotal or SScells
SSinteraction Degrees of freedom
dfA * dfB
SStotal =
SSA + SSB + SSinteraction + SSerror
SScells =
SSA + SSB + SSinteraction
dferror
N - k
When there are 0 effects
Each MS value is an estimate of o2error and should be similar to MSerror
If an interaction is significant
Effect of A and B goes both ways
Interaction
Effects changes outcomes and makes outcome not consistent across conditions
Crossover Effect
Interaction produces results that swap between conditions