Term
| What is the main purpose of ANOVA? |
|
Definition
-contrast the means between different groups
aka: compare the relative size of the boxes: variance due to treatment vs. variance due to error |
|
|
Term
|
Definition
- Factors: your IV's
-Levels: subsets of the IV's (subcategories) |
|
|
Term
| What is the difference between a 1 x 3 ANOVA and a 3 x 2 ANOVA? (and give an example of each) |
|
Definition
- A 1 by 3 has one factor with three levels (Ex: Suggestion and Alcohol: control, placebo, alcohol)
- A 3 by 2 has two factors, one was 3 levels and the other has 2 levels (Touch and Tip study: type of touch and gender of customer) |
|
|
Term
| What are problems associated with doing multiple t tests?(2) |
|
Definition
- t-test is a pair-wise comparison, so alpha rapidly inflates beyond what is acceptable
- also, limited experimental design |
|
|
Term
| How does the F-test set alpha? |
|
Definition
| for the whole experiment, so keeps the probability of a Type I error flat no matter how many comparisons you make |
|
|
Term
| Conceptually (think deviation scores), what is meant by: SS betw? |
|
Definition
| deviation of tx mean from the grand mean |
|
|
Term
| Conceptually (think deviation scores), what is meant by: SS within |
|
Definition
| how much each individual deviates from their own mean |
|
|
Term
| Conceptually (think deviation scores), what is meant by: SS total |
|
Definition
| How much individuals deviate from the grand mean |
|
|
Term
| What is the difference between SS and MS |
|
Definition
| MS (means squares) is adjusted for sample size (SS/df) |
|
|
Term
| Which is additive (SS or MS?) |
|
Definition
SS: columns add up to the total
- also, SS is what is used in partioning of variance |
|
|
Term
| Logic behind df term for: between subjects (tx) |
|
Definition
df: k-1
- because you're looking at how much the means of each tx deviate from the grand mean, so you are using k # of values and how far away they are from the grand mean |
|
|
Term
| Logic behind df term for: within subjects |
|
Definition
- df: N-k
- looking at how much individuals deviated from their own mean (so use everyone and use k #of means) |
|
|
Term
| Logic behind df term for: total |
|
Definition
df: N-1
- looking at how much individuals deviate from the grand mean (use everybody, but need just one mean--group mean) |
|
|
Term
| If the null hypothesis is true (no difference between means), what would you expect regarding the relationship between MS bet and MS within? |
|
Definition
| they would be about equal, no tx effect |
|
|
Term
| How does ANOVA use variance estimates to test for differences between means? |
|
Definition
| the F stat is a ratio of the variance related to the tx (spread among the means) to the variance within (due to error) |
|
|
Term
| Key characteristics of the F-distrib: composed of |
|
Definition
| F values, F-ratios due to chance (ratios of variance) |
|
|
Term
| Key characteristics of the F-distrib: shape (and what does the amount of skew depend on?) |
|
Definition
skewed: positive, b/c variance can't be less than zero
- amount of skew depends on df |
|
|
Term
| What is the function of post-hoc tests? |
|
Definition
- pairwise comparisons
- safeguard alpha inflation |
|
|
Term
| And when are post-hoc tests conducted? (2) |
|
Definition
1. when find significant F
2. if factor has more than 2 levels |
|
|
Term
| How does a Tukey HSD test attempt to prevent an inflated chance of a Type I error? |
|
Definition
calculates the minimal difference required for significance
- the q-values are adjusted based on how man comparisons you are doing
- more comparisons -> bigger q -> tougher to find significance |
|
|
Term
| How does the Scheffe test attempt to safeguard alpha? |
|
Definition
- like doing a mini-ANOVA on each pair of means
- safeguards alpha by multiple adjustments: df is more conservative, error term (use global error term), F crit is more cautious/conservative |
|
|
Term
| Assumptions made for independent groups ANOVA (3)...how robust? |
|
Definition
1. variance are about the same (s.d.'s are essentially the same)--extremely robust, so much so that no one bothers to test it
2. normal distributions: also very robust
3. data are independent |
|
|
Term
| Compare and contrast eta squared and r squared |
|
Definition
BOTH are proportion of variance accounted for out of the total variance
- with eta squared you can do it for multiple tx's: pie has more slices |
|
|
Term
| What is the source of the "Subjects effect" in repeated measures ANOVA's? |
|
Definition
consistent individual differences ex: depression scores, how good/bad of a driver you are
***CONSISTENCY IN THE INDIVIDUALS |
|
|
Term
| How does the partitioning of variance differ for a repeated measures ANOVA than with an independent? |
|
Definition
| - more boxes, "Variance due to subjects", this variance is taken out of the error term |
|
|
Term
| When might the "subject's effect" be large? |
|
Definition
| when there is lots of consistency in the individuals |
|
|
Term
| Compare "subject's effect" to the minus 2r term |
|
Definition
| get to subtract it our of the error term |
|
|
Term
| Advantages of using repeated measures |
|
Definition
get to measure those individual consistencies and then dump them
- decrease error, increase power |
|
|
Term
| Some concerns for repeated measures designs (3) |
|
Definition
1. some things are not reversible (ex: destroying the rat's hypothalamus)
2. carry-over effects: practice, fatigue, memory
3. greater awareness/demand characteristics |
|
|
Term
| What does it mean to assess the proportion of variance explained by each ANOVA component? |
|
Definition
| tells of much of the variance in the DV can be accounted for by the tx, subjects, error, etc. |
|
|
Term
| Semi-partial eta squared (what we calculate) vs. Partial eta squared |
|
Definition
| in the partial eta squared: no longer have total variance in the denominator |
|
|
Term
| What is meant by a two-way ANOVA? |
|
Definition
|
|
Term
| main effect (def and which means do we use?) |
|
Definition
impact of one IV (by itself) on the DV, regardless of the influence of any other variable
- uses marginal means |
|
|
Term
| What is meant by an interaction? |
|
Definition
| the joint impact of 2 or more IV's, where the impact of one IV depends on the level of other ex: task difficulty had an impact on males, but not on females |
|
|
Term
| What values are used in testing for the presence of an interaction? |
|
Definition
|
|
Term
| Advantages of a two-way ANOVA (5) |
|
Definition
1. more realistic: better reflects multi-determined behavior (explores the complexity of behavior)
2. decrease error: converts error variance to tx var. (potentially increase power)
3. Permits checking on generalizability
4. More efficient use of participants
5. Allows exploration of interaction effects (often of theoretical interest) |
|
|
Term
| one disadvantage of using two0way ANOVA's |
|
Definition
| difficult to interpret multi-way interactions |
|
|
Term
| What are simple main effects? When are they conducted? |
|
Definition
| conducted when a sig. interaction is found and more than 2 levels |
|
|
Term
| Options for IV'S (3 Categories) |
|
Definition
1. Manipulative (customer need) vs Non-manipulative (store busyness)
2. Qualitative (type of touch) vs Quantitative (prejudice of the subject: high/low)
3. Independent groups (prejudice of subjects) vs. Repeated Measures (ethnicity of speaker: each subject heard all three levels of this factor) |
|
|
Term
| What is similar between correlation and regression? |
|
Definition
| use same data set, same r |
|
|
Term
| Difference between correlation and regression |
|
Definition
- correlation: descriptive statistic, test r for significance, just a descriptive measure
- regression: takes it a step further, uses data make predictions |
|
|
Term
| What is meant by errors in prediction? (graphically and mathematically) |
|
Definition
- graphically: on scatter plot there are the actual data points above or below the regression line, and error is the amount that you differed from the regression line
- mathematically: Y minus Y hat |
|
|
Term
| Criteria for determining the prediction/regression line (4) |
|
Definition
1. sum of errors balances out to zero
2. sum of squared errors is minimized around it (least squares criterion)
3. passes through mean of X and mean of Y (pivot point for the ideal line)
4. Linear: optimal pattern/line through two variables (bivariate data) |
|
|
Term
| Compare criteria for line of best fit with the mean (3) |
|
Definition
1. sum of deviation scores balanced out to zero around the mean
2. sum of squared deviations is minimized
3. mean is the ideal center pt for univariate data |
|
|
Term
| 3 ways to evaluate accuracy of our prediction |
|
Definition
1. correlation (Pearson r)
2. standard error of the regression line
3. ratio of variance: r squared which is the variance of Y-hat over the variance of Y (actual scores) |
|
|
Term
| How can we identify the proportion of variance accounted for by a given predictor (in a regression problem)? |
|
Definition
| r squared: is the variance of the y hats over the variance of the actual Y values |
|
|
Term
| How is r squared like eta squared |
|
Definition
| both are the percent of variance that is identifiable over the total variance |
|
|
