Upload
dorothy-mccarthy
View
221
Download
0
Tags:
Embed Size (px)
Citation preview
Descriptive Statistics IIIREVIEW
• Variability• Range, variance, standard deviation
• Coefficient of variation (S/M): 2 data sets• Value of standard scores?
2SS
S
MXZ
zT 1050
1
2
2
n
MXS
Correlation(Pearson Product Moment or r)
•Are two variables related?•Car speed & likelihood of getting a ticket•Skinfolds & percent body fat
•What happens to one variable when the other one changes?
•Linear relationship between two variables
Scatterplot of correlation between pull-ups and chin-ups
(direct relationship/+)
0
2
4
6
8
10
12
14
16
0 2 4 6 8 10 12 14
Pull-ups (#completed)
Ch
in-u
ps
(#co
mp
lete
d)
Scatterplot of correlation betweenbody weight and pull-ups
(indirect relationship/-)
0
2
4
6
8
10
12
14
120 130 140 150 160 170 180
Weight (lb)
Pu
ll-u
ps
(#co
mp
lete
d)
Correlation issues
• Causation• -1.00 < r < +1.00• Coefficient of Determination (r2) (shared variance)• Linear or Curvilinear (≠ no relationship)• Range Restriction• Prediction (relationship allows prediction)• Error of Prediction (for r ≠ 1.0)• Standard Error of Estimate (prediction error)
Limitations of r
Figure 4.5Curvilinear relationship
Example of variable?
Figure 4.6Range restriction
Uses of Correlation
• Quantify RELIABILITY of a test/measure
• Quantify VALIDITY of a test/measure
• Understand nature/magnitude of bivariate relationship
• Provide evidence to suggest possible causality
Misuses of Correlation
• Implying cause/effect relationship
• Over-emphasize strength of relationship due to “significant” r
Correlation/PredictionREVIEW
• Bivariate nature• Strength (-1 to 1)• Linear relationships (curvilinear?)• (In)Direct relationships• Coefficient of determination: what is it and
what does it tell you?• Uses/Misuses of correlation?
Variables
Independent• Presumed cause• Antecedent• Manipulated by researcher• Predicted from• Predictor• X
Dependent• Presumed effect• Consequence• Measured by researcher• Predicted• Criterion• Y
We have data from a previous study on weight loss. Predict the expected weight
loss (Y; dependent) as a function of #days dieting (X; independent)for a new
program we are starting
To get regression equation, calculate b & c
b=r(sy/sx) b=.90(1.5/15) b=.09On average, we expect a daily wt loss of .09# while dieting
c=Ybar–bXbar c=8.0-.09(65) c=2.15
Y’ = bX + cY’ = .09x + 2.15
Predicted wt loss = .09(days dieting) + 2.15
Y=weight loss Ybar=8.0# sy=1.5#X=days dieting Xbar=65 days sx=15 daysrxy=.90
Standard Error of Estimate(SEE)
21* rSySe As r ↑, error ↓As r ↓, error ↑
Is ↑r good? Why/Not?
Is ↑ error good? Why/Not?