You are on page 1of 42

Structural equation

modeling
Advanced topics
Rex B Kline
Concordia

QICSS Set A

A1

Power

A2

A3

A4

Agenda
o M: Power, means,
LGM
o T:

LGM, measurement
invariance

o W: Moderation & mediation


A5

A6

Power
o Sometimes required
o Grant application
o Model tests in SEM
A7

Power
o Input (1):
H1 (population ES)
H0, , N, statistic
A8

Power
o Output (1):
p (reject H0 | H1 true)
1
A9

Power
o Input (2):
Target power (e.g., .80)
H1 (population ES)
H0, , statistic
A10

Power
o Output (2):
Target N

A11

Power in SEM
o Levels:
Individual parameter
Entire model
A12

Power in SEM
o Methods:
MacCallum et al. (1996)
Monte Carlo simulation
A13

Power in SEM
o

Bandalos, D. L., & Gagn, P. (2012). Simulation


methods in structural equation modeling. In
R. H. Hoyle (Ed.), Handbook of structural
equation modeling (pp. 92108). New York:
Guilford.

Bandalos, D. L., & Leite, W. (2013). Use of Monte


Carlo studies in structural equation modeling.
In G. R. Hancock & R. O. Mueller (Eds.),
Structural equation modeling: A second course
(2nd ed.) (pp. 625666). Charlotte, NC: IAP.

A14

Power in SEM
o MacCallum et al.:
RMSEA
0, 1

Type of H0
A15

Power in SEM
o RMSEA:
2

= max (0, M dfM )

dfM(N 1)
A16

Power in SEM
o RMSEA thresholds:
.05

.05 < < .08


.10
A17

Power in SEM
o , 90% CI:
[L , U]
E.g., = .02 [0, .15]
A18

Type of H0

Accept
support

Exact fit

Close fit

Reject
support

Not close fit

Poor fit

A19

Power in SEM
o Accept-support:
Logically weak
Low power favours model
A20

Power in SEM
o Reject-support:
Conventional logic
Low power against model
A21

Power in SEM
o Low power:
Exact fit, close fit
Low p (reject false model)
A22

Power in SEM
o Low power:
Not close fit
Low p (detect close model)
A23

Power in SEM
o Low power:
Poor fit
Low p (detect unpoor model)
A24

Test

Null

Exact fit

H0 : 0 = 0

Close fit

H0: 0 .05

Not close fit

H0: 0 .05

Poor fit

H0: 0 .10

A25

Power in SEM
o Given , [L , U]
o Exact fit:
L 0, reject H0: 0 = 0
2
M

Same p as (dfM )
A26

Power in SEM
o Given , [L , U]
o Close fit:
L > .05, reject H0: 0 .05

Some tools print p


A27

Power in SEM
o Given , [L , U]
o Not close fit:
U > .10, retain H0: 0 .05

A28

Power in SEM
o Given , [L , U]
o Poor fit:
U > .10, retain H0: 0 .10

A29

N = 373, dfM = 5
DFi
1

Fitness
DIl

Exercise

Illness
Hardiness
Stress
1

DSt

A30

Goodness of Fit Statistics


Degrees of Freedom for (C1)-(C2)
Maximum Likelihood Ratio Chi-Square (C1)

5
11.107 (P = 0.0493)

Browne's (1984) ADF Chi-Square (C2_NT)


Estimated Non-centrality Parameter (NCP)
90 Percent Confidence Interval for NCP

11.103 (P = 0.0494)
6.107
(0.0167 ; 19.837)

Minimum Fit Function Value


Population Discrepancy Function Value (F0)
90 Percent Confidence Interval for F0
Root Mean Square Error of Approximation (RMSEA)
90 Percent Confidence Interval for RMSEA
P-Value for Test of Close Fit (RMSEA < 0.05)

0.0298
0.0164
(0.000 ; 0.0532)
0.0572
(0.00299 ; 0.103)
0.336

A31

Power in SEM
o Results:
= .057 [.003, .103]
2
M

(5) = 11.107, p = .049


A32

Power in SEM
o Exact fit (H0: 0 = 0):
1. L = .003 > 0; reject
2. p = .049 < .05; reject
A33

Power in SEM
o Close fit (H0: 0 .05):
1. L = .003 < .05; retain
2. pclose = .336 > .05; retain
A34

Power in SEM
o Close yet failing:
Fail exact fit

Inspect
residuals

Pass close fit


A35

Power in SEM
o Not close fit (H0: 0 > .05):
U = .103 > .05; retain

A36

Power in SEM
o Poor fit (H0: 0 .10):
U = .103 > .10; retain

A37

Summary
Exact fit
Close fit
Not close fit
Not poor fit
= .057 [.003, .103]
A38

semTools for R
http://cran.r-project.org/web/packages/semTools/index.html

A39

date()
library(semTools)
# power for test of close fit hypothesis for N = 373
findRMSEApower(.05, .08, 5, 373, .05, 1)
# sample size for target power = .80 for close fit hypothesis
findRMSEAsamplesize(.05, .08, 5, .80, .05, 1)
# power for test of not close fit hypothesis for N = 373
findRMSEApower(.05, .01, 5, 373, .05, 1)
# sample size for target power = .80 for not close fit hypothesis
findRMSEAsamplesize(.05, .01, 5, .80, .05, 1)

A40

Statistic
N
dfM
Power
Close fita
Not close fitb
Minimum N for power .80
Close fit
Not close fit

373
5
.317
.229
1,464
1,216

aH0:

0 .05, 1 = .08, = .05

bH0:

0 > .05, 1 = .01, = .05

A41

Other tools
o STATISTICA Power Analysis:
http://www.statsoft.com

o Generate SPSS, R syntax:


http://timo.gnambs.at/en/scripts/powerforsem

o SAS/STAT syntax:
http://www.datavis.ca/sasmac/csmpower.html

A42

You might also like