Statistical Analysis

with Latent Variables

using Mplus Version 3

Bengt Muthen, UCLA

bmuthen@ucla.edu

Overview
Overview of Mplus modeling capabilities
n

Emphasis on growth models

Future seminars in this series:

n
n
n

Latent class analysis (Karen Nylund)

Discrete-time survival analysis (Katherine Masyn)
Modeling with a preponderance of zeros (Frauke Kreuter)

The Mplus web site

Examples of Mplus analyses
n
n

Real data
Monte Carlo simulations

Recorded

Program Background
Inefficient dissemination of statistical methods:
n

Many good methods contributions from biostatistics,

psychometrics, etc are underutilized in practice

Fragmented presentation of methods:

n
n

Technical descriptions in many different journals

Many different pieces of limited software

Mplus: Integration of methods in one framework

n
n

Easy to use: Simple language, graphics

Powerful: General modeling capabilities

The General Modeling Framework of Mplus

Achieves its flexibility from using a combination of
categorical and continuous latent variables
Special cases:
n
n
n
n
n

Factor analysis and structural equation modeling

Growth modeling
Mixture (latent class) modeling
Multilevel modeling
Combinations

General Latent Variable Modeling Framework

General Latent Variable Modeling Framework

General Latent Variable Modeling Framework

General Latent Variable

Modeling Framework
Muthn, B. (2002). Beyond SEM: General latent
variable modeling. Behaviormetrika, 29, 81-117
Asparouhov & Muthen (2004). Maximum-likelihood
estimation in general latent variable modeling
Muthen & Muthen (1998-2004). Mplus Version 3
Mplus team: Linda Muthen, Bengt Muthen, Tihomir
Asparouhov, Thuy Nguyen, Michelle Conn
(see www.statmodel.com)
7

Continuous Latent Variables

Factor analysis, structural equation modeling
- Constructs measured with multiple indicators

Growth modeling
- Growth factors, random effects: random intercepts and
random slopes representing individual differences of
development over time (unobserved heterogeneity)

Survival analysis
- Frailties

Missing data modeling

8

Path Analysis with a Categorical Outcome

and Missing Data on a Mediator
Logistic Regression

female
mothed
homeres
expect
lunch
expel
arrest
droptht7
hisp
black
math7
math10

hsdrop

Path Analysis

female
mothed
homeres
expect
lunch
expel
arrest
droptht7
hisp
black
math7

math10

hsdrop

Continuous Latent Variables:

Two Examples
Muthen (1992). Latent variable modeling in
epidemiology. Alcohol Health & Research World,
16, 286-292
- Blood pressure predicting coronary heart disease

Nurses Health Study (Rosner, Willet & Spiegelman,

1989). Nutritional study of 89,538 women.
- Dietary fat intake questionnaire for everyone
- Dietary diary for 173 women for 4 1-week periods at 3month intervals
10

Measurement Error in a Covariate

1.0

Proportion With Coronary Heart Disease

Without measurement error

(latent variable)
With measurement error
(observed variable)

0.8

0.6

0.4

0.2

0.0
0

20

40

60

80

100

120

Blood Pressure (millimeters of mercury)

11

Measurement Error in a Covariate

y1

y2

y3

12

Structural Equation Model

y7

y8

y9

y10

y11

y12

y1
f1
y2
y3
f3

f4

y4
y5
f2
y6
13

Structural Equation Model with

Interaction between Latent Variables
y7

y8

y9

y10

y11

y12

y1
f1
y2
y3
f3

f4

y4
y5
f2
y6
14

Growth Modeling
Exploring the data using Mplus graphics
Math achievement in grades 7 10 of the
Longitudinal Study of American Youth (LSAY)

15

Growth Modeling with Time-Varying Covariates

math7

math8

math9

math10

s
stvc
female
mothed
homeres

mthcrs7 mthcrs8 mthcrs9 mthcrs10

16

A Generalized Growth Model

math7

math8

math9

math10

s
stvc

female
mothed
homeres

mthcrs7 mthcrs8 mthcrs9 mthcrs10

17

A Generalized Growth Model

math7

math8

math9

math10

s
stvc

female
mothed
homeres

mthcrs7 mthcrs8 mthcrs9 mthcrs10

18

A Generalized Growth Model

math7

math8

math9

math10

dropout
stvc

female
mothed
homeres

mthcrs7 mthcrs8 mthcrs9 mthcrs10

19

Growth Modeling with a

Latent Variable Interaction
math7

math8

math9

math10

mthcrs7
20

Growth in Factors Measured

by Multiple Categorical Indicators
u11-u14

u21-u24

u31-u34

u41-u44

f1

f2

f3

f4

x1

x2

x3

x4

x0

21

Modeling with a Preponderance of Zeros

Outcomes: non-normal continuous count categorical

Censored-normal modeling
Two-part (semicontinuous modeling): Duan et al. (1983),
Olsen & Shafer (2001)
Onset (survival) followed by growth: Albert & Shih (2003)
Mixture models, e.g. zero-inflated (mixture) Poisson
(Roeder et al., 1999), censored-inflated, mover-stayer
latent transition models, growth mixture models
22

Two-Part (Semicontinuous) Growth Modeling

u1

u2

iu

su

iy

sy

y1

y2

u3

u4

y3

y4

23

Onset (Survival) Followed by Growth

u1

u2

u3

u4
Event History

iy

sy
Growth

y1

y2

y3

y4

24

General Latent Variable Modeling Framework

25

Categorical Latent Variables

Mixture regression
Latent class analysis
Latent transition analysis
Missing data modeling

26

Randomized Preventive Interventions and

Complier-Average Causal Effect Estimation
(CACE)

Angrist, Imbens & Rubin (1996)

Yau & Little (1998, 2001)
Jo (2002)
Dunn et al. (2003)

Compliance status observed for those invited for

treatment
Compliance status unobserved for controls
27

CACE Mixture Modeling

y

tx

28

Latent Class Analysis

Item Profiles
u1

u2

u3

u4

Class 1

c
Class 2
Class 3

Item u4

Item u3

Item u2

Item u1

29

Hidden Markov Modeling

u1

u2

u3

u4

c1

c2

c3

c4

30

Latent Transition Analysis

Transition Probabilities

Time Point 2

c2

Mover Class
1

0.6

0.4

0.3

0.7

(c=1)

Time Point 1
u11

u12

u13

u14

u21

u22

u23

u24

c1

Stayer Class

c1

c2
1

0.90

0.10

0.05

0.95

(c=2)

c2

c1

31

Loglinear Modeling of Frequency Tables

u1

c1

c2

u2

c3

u3

32

General Latent Variable Modeling Framework

33

Combinations of
Continuous and Categorical Latent Variables

Mixture CFA, SEM

Generalized latent class analysis
Second-order latent class analysis (twin modeling)
Growth mixture modeling
Longitudinal Complier-Average Causal Effect
(CACE) modeling in randomized preventive
interventions
Non-ignorable missing data modeling

34

Factor Mixture - Non-Parametric Factor Modeling

y1

y2

y3

y4

y5

35

Latent Class Analysis

With A Random Effect
u1

u2
c

f
u3

u4

36

Twin Modeling

37

Growth Mixture Modeling

Muthn, B. & Shedden, K. (1999). Finite mixture
modeling with mixture outcomes using the EM
algorithm. Biometrics, 55, 463-469.
Muthn, B., Brown, C.H., Masyn, K., Jo, B., Khoo,
S.T., Yang, C.C., Wang, C.P., Kellam, S., Carlin, J.,
& Liao, J. (2002). General growth mixture modeling
for randomized preventive interventions.
Biostatistics, 3, 459-475.
38

Growth Mixture Modeling

Outcome

y1

y2

y3

y4

Escalating
Early Onset

Normative
Age

39

8w
ee
ks

4w
ee
ks

2w
ee
ks

30

30

25

25
25

25

20

20
20

20

15

15
15

15

10

10
10

10

5
5

0
0

Placebo Group

1w
ee
k

W
48 ashho in
urs

Ba
se
lin
e

8w
ee
ks

4w
ee
ks

2w
ee
ks

1w
ee
k

W
48 ashho in
urs

Ba
se
lin
e

30

Hamilton Depression Rating Scale

30

A Clinical Trial
of Depression Medication:
Conventional Growth Modeling
Medication Group

40

8w
ee
ks

4w
ee
ks

2w
ee
ks

1w
ee
k

30

30
30

30

25

25
25

25

20

20
20

20

15

15
15

15

10

10
10

10

5
5

0
0

Placebo Non-Responders, 55%

W
48 ashho in
urs

Ba
se
lin
e

8w
ee
ks

4w
ee
ks

2w
ee
ks

1w
ee
k

W
48ashho in
ur
s

Ba
se
lin
e

Hamilton Depression Rating Scale

A Clinical Trial
of Depression Medication:
2-Class Growth Mixture Modeling
Placebo Responders, 45%

41

Growth Mixture Modeling in Mplus

Mplus input for LSAY math achievement

42

Growth Mixture Modeling:

LSAY Math Achievement Trajectory Classes
and the Prediction of High School Dropout

Dropout:

9
Grades 7-10

10

69%

100
80
40

40
7

Good Development: 52%

60

80

100

Moderate Development: 28%

60

80
60
40

Math Achievement

100

Poor Development: 20%

9
Grades 7-10

10

8%

9
Grades 7-10

10

1%

43

Longitudinal CACE,
Non-Ignorable Missing Data
Yau & Little (2001). Inference for the complier-average
causal effect from longitudinal data subject to noncompliance
and missing data, with application to a job training assessment
for the unemployed. Journal of the American Statistical
Association, 96, 1232-1244.
Frangakis & Rubin (1999). Addressing complications of
intention-to-treat analysis in the combined presence of all-ornone treatment-noncompliance and subsequent missing
outcomes. Biometrika, 86, 365-379.
Muthn, Jo, & Brown (2003). Comment on the Barnard,
Frangakis, Hill & Rubin article, Principal stratification
approach to broken randomized experiments: A case study of
school choice vouchers in New York City. Journal of the
American Statistical Association, 98, 311-314.
44

Growth Mixture Modeling with Non-Ignorable

Missingness as a Function of Latent Variables
y1

y2

u1

u2

y3

y4

u3

u4

Outcome
Escalating
Early Onset
c

Normative
Age

45

Growth Mixture Modeling with Non-Ignorable

Missingness as a Function of Latent Variables
y1

y2

u1

u2

y3

y4

u3

u4

Outcome
Escalating
Early Onset
c

Normative
Age

46

Growth Mixture Modeling with Non-Ignorable

Missingness as a Function of Latent Variables
y1

y2

u1

u2

y3

y4

u3

u4

Outcome
Escalating
Early Onset
c

Normative
Age

47

General Latent Variable Modeling Framework

48

Multilevel Modeling with

Continuous and Categorical Latent Variables

Multilevel regression
Multilevel CFA, SEM
Multilevel growth modeling
Multilevel discrete-time survival analysis

Multilevel regression mixture analysis (CACE)

Multilevel latent class analysis
Multilevel growth mixture modeling
49

Within

Between

per_adva

m92

private

s1

catholic

s2

female
s1
m92
s2
stud_ses

mean_ses

50

Two-Level Factor Analysis with Covariates

Within

x1

fw1

Between

y1

y1

y2

y2

y3

y3
w

x2

fw2

fb

y4

y4

y5

y5

y6

y6

51

Multilevel Modeling with a

Random Slope for Latent Variables
Student (Within)
y1

y2

y3

School (Between)
s

y4

ib

s
iw

sw

sb

52

Two-Level CACE Mixture Modeling

Individual level
(Within)

Cluster level
(Between)

Class-varying
x

tx

53

Two-Level Latent Class Analysis

Within
u1

u2

u3

u4

Between
u5

u6

c#1

c#2

54

Multilevel Growth Mixture Modeling

Math7

Math8

Math9

Math10

Female
Hispanic
Black
Mother s Ed.
High School
Dropout

Home Res.
Expectations

Drop Thoughts
Arrested

ib

cb

sb

hb

Expelled
School-Level Covariates

55

Monte Carlo Simulations in Mplus

Data generation, analysis, and results summaries
across replications
Studies of tests of model fit, parameter estimation,
standard errors, coverage, and power as a function of
model variations, parameter values, and sample size
Model Population, Model Missing, Model for
analysis
Full modeling framework available: continuous and
categorical latent variables, multilevel data, different
types of outcomes

56

References
See the Mplus web site www.statmodel.com

57

General Latent Variable Modeling Framework

58