(1) uses half the number of subjects

(2) Maintains same rate of Type 1 error

(3) Can look at interactions

(1) Each score is influenced by both independent variables

(2) These influences can be separated out by margins

(3) variance is not affected by treatments

What does an interaction measure?

Advantages of Within ANOVA

The standard deviation will be smaller than a between subj design

Size of Denominator of F effects on power

Within ANOVA: SSw is replaced with what?

Yes, because it reduces intrinsic variablity

True or False

Interactions reflect nonlinear effects of independent variables

True

True or False

Main effects reflect linear summationof the effects of independent variables

What does the row factor reflect in regular two factor ANOVA?

any differences in the effects of specific treatment conditions on specific subjects

If subject variability increases in Within ANOVA what happens to the SSbet and SSw?

One factor (one way) ANOVA

Factorial ANOVA

Repeated Measures (within subjects) ANOVA

(1) Experiments with nominal variables

(2) To analyze proportional results

Whether the frequencies observed in our experiment are significant or due to chance

The distribution of the Chi Square under the null depends on what?

What are the properties of the Chi Square Distribution (under the null)?

(1) The mean of the disribution=number of degrees of freeedom

(2) the variance= 2 times df

(3) As the df gets bigger, the Chi Square approaches a normal.

Chi Square: As the df increases, the critical value increases

T and F: As the df decreases the critical value decreases

(1) population is normal

(2) groups are sampled randomly

(3) the treatment only changes the means, not the variances

Increase in type 2 error, sometimes type 1

What are the possible reasons for normality being violated?

(1) the dependent variable is not distributed normally

(2)floor/ceiling effects

(3) distribution is skewed

(4) sample comes from more than one population

(5) dependent variable is influenced by direct random effects

(6) noise is not normally distributed

(1) Data transformation

(2) Nonparametric tests

(1) rank order

(2) permutation/randomization tests

(3) Mann Whitney U

(4) Kruskal Wallis H

(5) Wilkoxen signed rank test

What data transformation should you use for positively skewed data?

How do we estimate the appropiate data transformation to use on our data?

(1) operate on the dependent variable

(2) plot histogram

(3)guess which data transformation you should use

(4)apply transformation

(5) plot distribution

(6) check that signal and noise is normal

Constraints on transformations

(1) same transformation must be applied to all groups in the experiment

(2) apply it before hypothesis testing

(3) transformation can't affect variance

(4) transformation cannot change the ranks

Advantages of randomization

(1) it doesn't assume your population is normal

(2) will give the correct significance value no matter the population distribution

(3) if population is normal, randomization and a conventional test will give same level of significance

(1) if n is low can't do many permutations

(2) we can only make conclusions about the sample NOT the population

Why do we use bootstrapping?

To get a better estimate of the population parameter.

(1) its normally distributed

(2) its unbiased

(3) the standard deviation is a good estimate after resampling

(1)bootstrapping samples with replacement

What is the difference between sampling with replacement in bootstrapping and shuffling scores?

If you sample with replacement the scores will repeat, shuffling your scores uses the exact same scores in different order

There is more than one independent variable

(1) x variables are independent

(2) the coefficients are used the same as in linear regression

(3) the noise is normally distributed

(4) variance is the same for all the data points

What test do we use to see if the coefficients in a regression equation is statistically significant?

Partial correlation is the relationship between two variables with the effects of the others removed.

Your 'r' if you normalized your y data

Yes. If there are only two groups

What is the general linear model?

It refers to the way ANOVA can be turned into regression.

(1) treatment only shifts the means, not the variance

(2) samples are randomly selected

(3) the variance is the same for all the predictors (coefficients)