Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
April/May 2010
Outline
Course will cover:
• Path analysis and SEM
• Gain familiarity with some of common statistical procedures,
coupled with their application through the use of statistical
software AMOS:
• Mediation analysis
• Confirmatory factor analysis
• Simple SEM
• Multigroup comparison (“Factorial invariance”)
Today:
What are path analyses and SEMs?
– Recap correlation and regression
– Basic concepts of path analysis and SEM
– Unobservable latent traits
– Relationship between correlation,
regression, factor analysis, path analysis
and SEM
Introduction to AMOS
– Simple analyses with AMOS
Correations
Correlations
5.0
Pearson’s r=0.66
Water consumption (dl)
4.0
3.0
2.0
1.0
0.0
17 18 19 20 21 22 23
Temperature
9
Examples of correlations
correlation=0
4 correlation = 0.35
2
0
0
-2
-2
-2 0 2
-3 -2 -1 0 1 2 3
correlation=.73 correlation=.94
10
10
5
0 0
-5
-10 -10
-2 0 2 -2 0 2 10
βtemp
Temperature Water consumption
βthirst
Thirst Water consumption
5.0
Water consumption (dl)
4.0
3.0
b=0.8 dl
2.0
1C
1.0
R Sq Linear = 0.434
0.0
17 18 19 20 21 22 23
Temperature 12
Unstandardized Standardized
Coefficients Coefficients
Model B Std. Error Beta t Sig.
1 (Constant) -14.320 2.886 -4.961 .000
temp Temperature .836 .138 .659 6.071 .000
a. Dependent Variable: water Water consumption (dl)
Answer to Question 1:
Temperature = 20 degree
Water consumption= ?
Water consumption = -14.32 + 0.836 *20
= -14.32+16.72
= 2.4 dl
Question 2:
water consumption= -14.32 + 0.836 *Temperature
Temperature = 0 degree
Water consumption= ?
14
Temperature
Water consumption
Thirst
15
y = β0 + β1 x1 + β2 x2 + ... + βk xk + ε where
Water consumption =
-5.068 + 0.2681*temperature + 0.779*thirst + ε(N,σ2)
17
Temperature
β1
Water consumption
β2
Thirst
19
Multiple regression
Error
Tem perature 1
β1
β2
Thirst
21
Stimulus-Organism-Response Model
Theory suggests that the effect of a stimulus on a response is
mediated by the organism:
Higher temperature increases thirst, which then leads to water
consumption.
(Woodworth 1928) Stimulus
Organism
Response
22
Temperature
Error
1
Thirst
Error
1
Water consumption
Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social
psychological research: Conceptual, strategic and statistical considerations. Journal of 23
Personality and Social Psychology, 51, 1173-1182.
Path analysis
24
25
Path analysis
Independent variable X
Error
1
Error
Indirect 1
effect from
X on Y via
M Dependent variable Y
26
27
28
1 1 1 1 1
e1 e2 e3 e4 e5
1 1 1 1 1
31
Temperature 1
Item 1 e1
1 1
e7 Item 2 e2
1
Latent trait "Thirst" Item 3 e3
1
Item 4 e4
e6 1
Item 5 e5
Water consumption
32
1 1 1
SEM
Item 1 Item 2 Item 3
Pain
error
1
1 Item 1 Error 1
1
Depression Item 2 Error 2
1
Item 3 Error 3
Function
1 1 1
34
Error 3 Error 2 Error 1
35
1 1 1 Measurement model
Pain
1
1 Item 1 Error 1
1
Depression Item 2 Error 2
1
Item 3 Error 3
Function
1 1 1
Structural model
36
Error 3 Error 2 Error 1
37
1 1 1 1
1
Item 5
e5 contri- Item 1 Item 2 Item 3 Item 4
bution
1 Item 4 1
e4
Coworker
1
Item3 1
e3 other Self Esteem e10
Depression
people
1
Item 2
e2 super- 1
visor
1 Item 1
e1
"work"
Age
38
e
Structural
Multiple Path
Bivariate Regression
Equation
Analysis
Correlation Modeling
Confirmatory
Factor Factor
Analysis Analysis
Exploratory
Factor
Analysis
39
1 1
Change in Temperature
quicksilver column How thirsty are you? Latent trait "Thirst"
40
41
Unobservable traits
• In psychology and health sciences we are often
concerned with questions which are more subjective
than questions in other fields of science.
• These includes measurements of:
abilities, knowledge, emotions, feelings, attitudes or
personality traits.
• All traits have got in common that they are unobservable
traits = latent traits.
42
43
45
46
“I am proud of my relationship
with my supervisor” 1 2 3 4 5
48
Item 1 1
e1
"work"
1
Item 2 1
super- e2
visor
Item3 1
Self Esteem other e3
people
Item 4 1
e4
Coworker
Item 5 1
contri- e5
bution
49
50
1 1 1 1
1
Item 5
e5 contri- Item 1 Item 2 Item 3 Item 4
bution
1 Item 4 1
e4
Coworker
1
Item3 1
e3 other Self Esteem e10
Depression
people
1
Item 2
e2 super- 1
visor
1 Item 1
e1
"work"
Age
51
e
1
1 pain e1
e3 function
1
1 Well-being Well-being 1
e2 depress depress e2
Time 1 Time 2
1
1
1 function e3
e1 pain
53
error
1
1 1 paragrap err_p
1
Age verbal sentence err_s
1
wordmean err_w
54
55
1 1 1 1
X1 X2 X3 X4
ICEPT SLOPE
56
0, g1 0, g2 0, g3 0, g4
e1 e2 e3 e4
1 1 1 1
mg1 mg2 mg3 mg4
y1 y2 y3 y4
4
2 6
1 1
1 1 0
ICEPT Slope
57
58
AMOS
60
61
Observed variable
1
Latent variable with
disturbance or error
1 Observed variable
with measurement
error (“endogenous Unidirectional path
variable”) (“regression”)
Correlation between
1 1 variables
Latent variable with
items (= observed
1 variables) Reciprocal relation
between variables
62
16. Save
Exercise 1
64
e1
1
e2
Tem perature 1
Thirst
65
Model 2
Temp
1 1
e1 Thirst activity e2
water consumption
e3
66
e1 e2 e3 e4 e5
1 1 1 1 1
a b c d e
F1
67
Model 4
e1 e2 e3 e4 e5
1 1 1 1 1
a b c d e
F1
1
F2 e12
p q r s t u
1 1 1 1 1 1
68
e11 e10 e9 e8 e7 e6
69
70
•
• See results on graph Temperature W ater consumption (dl)
71
Example
Data set:
• Water_consumption.sav
72
.66
73
Regression
e1
1
.92
.43
74
.92 e1
1
Temperature
.27
Thirst
Standardised estimates
e1
Temperature .61
.21
Thirst
75
Exercise
Use SPSS and open the data file “mediation.sav”
1. Do the following analysis with SPSS:
• Correlation matrix of all three variables
• Simple linear regressions between:
– Function Depression
– Pain Depression
• Multiple regression:
– Pain +Function Depression
2. Do the same analysis with AMOS
3. Optional: Delete a few observations of the independent
variables the data set and rerun the multiple regression
in SPSS and AMOS. Are the results still the same?
Compare the sample sizes of both analyses. 76
77
78
Pain Depression
79
Regression results
0, .40
Unstandardised regression coefficient
e1 Error
Variance Variance
1
4.93, 4.08 1.00
.11
Pain Depression
Explained
e1 variance r2
Standardised regression coefficient
.11
.34
Pain Depression
80
1
Pain .06 2.61
Body Function
Explained
Standardised regression coefficient e1 variance r2
-.46 Depression
Correlation -.34
between
independent Body Function
81
variables
pain
Error 2
pain
1
Error 1
1
depress
function
function
Error 2
1
1
Error 1 depress
82