Latent Class Análysis
Latent Class Análysis
Tarani Chandola
methods@manchester
Many names‐ similar methods
• (Finite) Mixture Modeling
• Latent Class Analysis
• Latent Profile Analysis
Latent class analysis (LCA)
• LCA is a similar to factor
analysis, but for
categorical responses.
• Like factor analysis, LCA A C
addresses the complex
pattern of association
that appears among B D
observations….
3
Factor Analysis
We observe a correlation
between two variables. Why?
X Y
X Y
4. A third, unmeasured
cause ?
Z
2. Y causes X ?
X Y
X Y
4
Unmeasured Causes:
Factor Models
Variables may be related due to the action of unobserved
influences.
Sometimes these are confounding variables, but many
constructs of interest are not directly observed (or even
observable)
Unobserved Construct Observed Measures
Social Capital Bowling club membership
Local newspaper reading
Ethnic prejudice Housing segregation
Ethnic intermarriage
5
Factor Models
Correlations may not be due to causal relations among the
observed variables at all, but due to these unmeasured, latent
influences - factors
Y1 Y2 Y3 Y4
Y1 1.0
Y2 0.6 1.0
6
Factor Models
The observed correlations may be due to each observed
measure sharing an unobserved component (F)
F
Y1 Y2 Y3 Y4
Y1 1.0
Y2 0.6 1.0
Y1 Y2 Y3 Y4
Y3 0.7 0.6 1.0
7
Factor Models
Example: Four questionnaire items that have highly correlated
answers
Y1 “I Often feel blue”
Y2 “I dislike myself”
Y3 “I have a low opinion
of myself”
Y4 “My life lacks
direction” Y1 Y2 Y3 Y4
8
Factor Models
The items may be correlated due to the influence of the
respondent’s mood state, which we can’t observed directly
Y1 “I Often feel blue” F F = Depressed?
Y2 “I dislike myself”
Y3 “I have a low opinion
of myself”
Y4 “My life lacks Y1 Y2 Y3 Y4
direction”
E1 E2 E3 E4
9
Factor Models
10
Model Fit
• Standard measure of ‘observed’ vs. ‘expected’
fit?
– Pearson χ2 (Chi‐Square) test
– Sum of the squared differences between observed
(O) and expected (E) (co)variances divided by the
expected
χ2 = Σ[(O‐E)2/E]
– The larger the χ2 the greater the model misfit
– Can test if χ2 = 0 using the model df
11
In LCA, the underlying unobserved variables are not
continuous (dimensions) but classes/categories/discrete
Class Class
1 2
A B C D
SUGI 31 ‐ Contributed paper
12
201‐31
What if you do not know how to classify people into (depressed
vs not depressed) groups? What if there is no gold standard to
assess a pattern of “yes/no”signs and symptoms?
Rindskopf, R., & Rindskopf, W. (1986). The value of latent class analysis in
medical diagnosis. Statistics in Medicine, 5, 21‐27.
13
LCA of Depression (Dep) indicators
LCA predicts latent class membership such that the observed
variables are independent.
14
P(Dep) P(Not Dep)
LCA estimates
Latent class prevalences
Conditional probabilities: probabilities of specific
response, given class membership
15
LCA works on unconditional contingency table (no
information on latent class membership)
Feel Low Dislike Life lacks nijkl
blue opinion myself direction
0 0 0 0 15
0 0 0 1 14
0 0 1 0 11
0 0 1 1 8
0 1 0 0 23
. . . . .
1 1 1 1 9
LCA’s goal is to produce
a complete (conditional) table
that assigns counts for each latent class:
Feel Dislike Low Life Latent nijklt
blue myself opinion lacks Class
directi X=t
on
0 0 0 0 1 9
0 0 0 1 2 6
0 0 1 0 1 3
0 0 1 1 2 11
. . . . . .
1 1 1 1 2 9
Estimating LC parameters
• Maximum likelihood approach
• Because LC membership is unobserved, the likelihood
function, and the likelihood surface, are complex.
18
EM algorithm calculates L
when some data (X) are unobserved
“M” step
produces ML estimates from
complete table
“E” step
uses parameter estimates
to update expected values
for cell counts nijklt
in complete contingency table
19
EM algorithm requires initial estimates
“M” step
1st “E” step:
Provide initial
(random) estimates to
“fill in” missing “E” step
information on LC
membership
20
Mixture modeling
21
Latent Profile Latent Class
Analysis Model Analysis Model
y1 y2 y3
Continuous indicators Dichotomous (0/1)
y: y1, y2, … , yr indicators
Categorical latent variable u: u1, u2, … , ur
c: c = k ; k = 1, 2, … , K. Categorical latent variable
c: c = k ; k = 1, 2, … , K.
22
Model Results
Two-Tailed
Estimate S.E. Est./S.E. P-Value
Latent Class 1
Means
BMI 25.166 0.139 181.262 0.000
Variances
BMI 15.305 1.279 11.970 0.000
Mean BMI of 25.2 (and variance of 15.3) in the whole population
23
Histogram of BMI (1 class solution)
24
Class 1
Class 2
25
Mixture Model of BMI with 3 classes
Two-Tailed
Estimate S.E. Est./S.E. P-Value
Latent Class 1
Means
BMI 32.100 16.503 1.945 0.052
Variances
BMI 8.319 4.653 1.788 0.074
Latent Class 2
Means
BMI 40.685 18.724 2.173 0.030
Variances
BMI 8.319 4.653 1.788 0.074
Latent Class 3
Means
BMI 24.414 1.342 18.190 0.000
Variances
BMI 8.319 4.653 1.788 0.074
! Those in Class 2 have much higher mean BMI than those in classes 1 and 3
26
Latent Profile/Class analysis with 2 and 3 latent
classes
27
Deciding on number of latent classes
‐ Start with the simplest (a one class) solution, and add more
classes stepwise.
‐ Examine the model evaluation statistics: Chi‐square difference
tests are not appropriate for likelihood ratio test comparisons of
models with higher numbers of classes. Models that maximize
the log likelihood are generally better fitting, although this
comes at the expense of fitting more parameters to the model.
Look for low values on the Akaike’s Information Criterion (AIC),
Bayesian Information Criterion (BIC) and sample size adjusted
BIC statistics. In addition, Tech 11: modification to the likelihood
ratios test that adjusts the conventional likelihood ratio test for K
vs K‐1 classes for violation of regularity conditions (p>0.05
indicates K‐1 classes are sufficient).
‐ Examine entropy measure (higher values indicate better fit).
‐ Usefulness of the latent classes in practice. This can be
determined by examining the trajectory shapes for similarity, the
number of individuals in each class, and whether the classes are
associated with observed characteristics in an expected manner.
28
Deciding on number of classes‐ BMI example
LRT p-value for
No. of classes Loglikelihood # par. AIC BIC Entropy k-1
1 -2209.712 2 4423.424 4432.778 NA NA
2 -2144.898 4 4297.797 4316.505 0.952 0.0000
3 -2137.826 6 4287.652 4315.714 0.901 0.8237
4 -2133.359 8 4282.718 4320.135 0.745 0.0326
Loglikelihood AIC
-2080 4450
1 2 3 4
-2100
4400
-2120
4350
-2140
-2160 4300
-2180
4250
-2200
4200
-2220 1 2 3 4
29
Mixture modeling with categorical dependent
variables
Latent classes (‘normal’
Normal weight and ‘obese’) predict
weight? health problems: logistic
Without regression c
health
Obese?
problems?
With health
problems?
30
Mixture model with covariates and categorical
dependent variables
X predicts
membership into
normal weight and
obese latent classes x c
31
Are you a joiner or a splitter?
Factor Analysis
vs. Latent Profile/Class Analysis
32
Resources
Introduction to LCA:
http://www.john‐uebersax.com/stat/faq.htm
http://www.ccsr.ac.uk/methods/festival/programme/wiwp/francis.pdf
McCutcheon AC. Latent class analysis. Beverly Hills: Sage Publications, 1987
Software:
http://www.john‐uebersax.com/stat/soft.htm
Short courses:
Latent Trait and Latent Class Analysis for Multiple Groups Using Mplus
http://www.ccsr.ac.uk/courses/congnitiveInterviewing/LatT.html
Introduction to Structural Equation Modelling using Mplus
http://www.ccsr.ac.uk/courses/semintro/