Relationship Stats Flashcards
Pearson Correlation
-r= +1(positive) to -1 (negative)
-how close to the line of best fit
-interval/ratio data, methological
-is there a relationship and how strong is it?
-nothing to do with slope or agreement or cause/effect
Pearson Correlation Interpretation
.1 small effect
.3 medium
.5 large effect
-must be significant in SPSS
r^2
-coefficient of determination
-how much the variability in one variable can be redicted by the other
r^pb
-point-biseral correlation
-correclation between 2 level of a categorical variable
-run the same as pearson
-must be significant
Spearman Coefficiet
-non-parametric equivalent to pearson’s
-ordinal data
-rank order
-1 to 1
-D^2= larger the difference the smaller the association
Internal Consistency
-how closely related the items in an outcome measure are as a group
Crohbach’s a
-calculated by averaging all possible pairwise correlations between items
-0-1
-.7 acceptable
-.8 good
-.9 excellent
Standard Error of Mean (SEM)
-value that describes the diff between the sample mean and true pop mean
-always smaller than SD
-smaller=less sampling error
-sample SD/Square root (n)
Standard Error of Measurement
-related to test reliability (test-retest)
-how much error should be expected
Intraclass Correlation Coefficient
-ICC
-reliability coefficient
-measure agreeemnt or association between 3 raters
-0-1
ICC Model
Model 1: one way random, each participant is assessed differently, rare
Model 2: two-way random, each participant assessed by each rater, raters are radomized, common
Model 3: two way mixed, each participant assessed by each rater, raters are only of interest
ICC Form
1: single measurement
2: average of 2 measurements
3: average of 3 measurements
G-Theory
-generalizability
-sophisticated way to measure reliability
-relative contribution of error
-of the amount of error, what is contributing the most
-higher G= higher reliability
-generalizability coefficient in articles
Regression
-preficts an outcome
-interval/ratio/nominal
-more data than pearson
Linear, Multiple Linear, Logistic, Multinominal logistic, ordinal logistic
Linear
-one predictor and one outcome
ex: does GRE predict GPA
SPSS:
-ANOVA, will show significance for entire model but not where
-Model Summary: Adjusted R square, Significant F change, Durbin-Watson
-Coefficients: Unstandardized B
Unstandardized B
-coefficients for regression that show where it was significant
-has constant and predictor value
Linear Regression Equation
Outcome= Constant + (Predictor Score x Unstandardized B of Predictor)
Multiple Linear
-multiple predictors and 1 outcome
ex: GPA and GRE predict PT GPA
-R square change for how much above and beyond
-sig F change for significance
Logistic
-1 or more predictors and one categorical, 2 levels
ex: Does TUG predict fallers and non fallers
Catergory: Falling
Level: fall, non falls
-uses logarithmic curve
Multinominal Logistic
3 or more predictors and one categorical, 2+ levels
Ordinal Logistic
-1 or more predictors and 1 ordinal, 2+ levels
Regression Assumptions
-Must be linear (scatterplot)
-Normality (Shapiro-Wilk Test if violates)
-Homoscedasticity
-Free of outliers (cook’s distance)
-Data Independence (durbin-Watson test)
Multiple Regression:
-Multicollinearity
Shapiro-Wilk Test
-normality in regression
-dont want significance
Cook’s Distance
-checks for outliers
->1 is bad
Durbin-Watson Test
-check for independence of observation in
-0-4 and 2 is perfect
Adjusted R Square
-generalize results to population
Multicollinearity
-multiple regression assumption
-too much correlation of predictors
-do not want this
-<0.9 correlation, VIF should be <10, Tollerance should be >0.1