Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
26 Cards in this Set
- Front
- Back
3 Difference scores issues (with mixed designs S/AxB)
|
1. masks main effects; only shows extent of change in A across B (interaction)
2. interpretation: one-way between groups ANOVA is computed but results must be interpreted as an interaction 3. if levels of B > 2, this fails |
|
what is Spearman Rho correlation? give example
|
correlation between ranks on two dimensions, ie hs class rank and college class rank
|
|
Pearson product moment:
ex: |
correlation between two continuous variables
self esteem and depression |
|
what does r estimate? between what and what?
|
r estimates rho. btw 1 and -1
|
|
what are the determinants of power for predicting y hat?
|
small N, small r
|
|
name the 3 cautions of hierarchical regression:
|
1. multicollinearity
2. suppression 3. shrinkage |
|
define mulitcollinearity, explain how it is fixed
|
high correlation among predictors, which explain the same variance and lower DF.
FIX: examine correlation matrix and eliminate a predictor, or combine predictors |
|
define suppression, explain fix
|
suppression is when the prediction of x1 is enhanced only because x2 suppresses variance irrelevant to y.
FIX: examine correlation matrix, if b's and r's have different signs, you have suppression! |
|
define shrinkage, explain fix
|
R squared is inflated as K increases (worse with small N).
FIX 20-30 cases per predictor |
|
name the 3 assumptions of r
|
rectilinearity, homoschedasticity, x measured without error
|
|
define rectilinearity
|
relation is a STRAIGHT LINE
|
|
define homoschedasticity
|
variances of ys at each x are equal
|
|
what are the three caveats of r?
|
1. do not try to predict y beyond original range 2. do a scatter plot to look for leverage (outliers or extremes in mean) 3. truncated distributions reduce the size of r
|
|
what is the common language size effect?
|
if person A's x exceeds B's x, what is the probability that A's y will exceed B's y? CLE= y67, 67% chance
|
|
what does the multiple correlation and regression equation do?
|
comprizes more than one predictor, and minimizes squared differences of predicted y's from observed y's (least squares regression)
|
|
what is R?
|
correlation between obtained and predicted ys
|
|
R = ?
|
r times y times yhat
ALWAYS positive R is the best linear combination of the predictors of y |
|
explain a standard multiple regression
|
all variables added at once, each predictor is evaluated as though no overlap exists.
|
|
what is crucial about standard multiple regression?
|
each predictor is evaluated in terms of what it adds to the prediction of the criterion, that differs from predictability given by others
|
|
what test allows for assessment of unique contribution in standard multiple regression?
|
significance test of each regression coefficient (b)
|
|
what do these divisions of S/AxB design mean?: A, S/A, B, AxB, S/AxB
|
A = average variation between scores
S/A = error B = repeated measures factor AxB = interaction S/AxB = unexplained variability |
|
define these variables of S/AxB design: A, B, AB, a, an, b, bn
|
A = column totals
B = row totals AB = cell totals a = # levels of IV an = # of subjects b = # of times subject gives score bn = # of x's added to get A |
|
DF for S/AxB design:
total btw A S/A within B AxB S/AxB |
total: N -1
btw: an-1 A: a-1 S/A: btw-A within: total-btw B: b-1 AxB: (a-1)(b-1) S/AxB: (S/A) times (B) |
|
what N is needed given projected effect for power?
|
4/rsquared = N-1
this gives you 50% power double for 80-90% power |
|
least squares regression:
|
Sigma (y - yhat)squared
|
|
in uncorrelated predictors, what does Rsquared tell us? R?
|
Rsquared: percent of variance in y explained by x
R: how strongly x1 x2 and y are correlated |