Header
Research Article
A Monte Carlo Study of Multilevel Latent Class Regression
Luohua Jiang1* and Shuai Chen2
1Department of Epidemiology and Biostatistics, School of Rural Public Health, Texas A&M Health Science Center, College Station, Texas, USA
2Department of Statistics, Texas A&M University, Collect Station, Texas, USA

ABSTRACT
Multilevel latent class analysis (MLCA) has been built into a few statistical software to analyze nested data that do not satisfy the conditional independence assumption of simple Latent class analysis (LCA). Multilevel latent class regression (MLCR) is also available in those software packages to analyze the relationships between latent class membership and covariates. The impact of using simple latent class regression (SLCR) instead of MLCR for nested data has not been investigated empirically. In this study, we conduct Monte Carlo simulations to examine the influence of intra-class correlation (ICC) on the estimation bias and coverage of regression coefficients using MLCR vs. SLCR. We also evaluate the consequences of assuming perfect correlation when the random intercepts are actually not perfectly correlated. The results indicate that, with the increase of ICC, the biases in regression coefficients increased while the coverage probabilities decreased. Further, we find that the bias caused by the misspecification of perfect correlation assumption in MLCR estimation was slight, especially when the ICC was low. Thus, MLCR with perfect correlation might be a computationally efficient method without substantial loss of accuracy, and hence could be a reasonable substitute for MLCR procedure with ordinary correlation when computation burden is a concern
KEYWORDS
Latent class analysis; Multilevel models; Correlated random intercepts

Introduction
Latent class analysis (LCA) is a widely used statistical method in many fields. It assumes the subjects belong to some latent subgroups, referred as latent classes. Although the classes are not directly observed, they can be inferred from a set of observed categorical variables using LCA. Further, the relationships between latent class membership and covariates may be assessed using latent class regression (LCR). In simple LCA, it is assumed that subjects are independent conditional on the latent class membership. However, if the subjects are clustered in groups, the conditional independence assumption may not be met. Thus, multilevel techniques are in need to incorporate the intra-cluster dependence for those types of data.
In recent years, multilevel latent class analysis (MLCA) has been developed by a few groups [1-3] to apply LCA with nested data. They also extended MLCA to include Level 1 and Level 2 covariates in the model [1,3]. In multilevel LCR (MLCR), a two-level multinomial logistic regression is adopted, by introducing random intercepts across Level 2 (cluster) units. When the number of latent classes C is more than 2, the C-1 random intercepts are allowed to be correlated with one another. However, the computational burden of this model grows exponentially with C. Thus, Vermunt [1] suggested to model all the random intercepts using a common factor, which assumes the random intercepts are perfectly correlated.
To the best knowledge of the authors, the impact of using simple latent class regression (SLCR) instead of MLCR for nested data has not been investigated empirically. In this article, we present a Monte Carlo simulation study to examine the influence of intra-class correlation (ICC) on the estimation bias and coverage of regression coefficients using MLCR vs. SLCR. We also evaluate the consequences of assuming perfect correlation when the random intercepts are actually not perfectly correlated. More specifically, the performance of SLCR, MLCR with perfect correlation (MLCR-P), and MLCR with ordinary correlation (MLCR-O) are compared in a MLCR model with 3 latent classes.
Materials and Methods
SLCR
In a SLCR model, let Yi = (Yi1,,YiM) bean observed response vector for the ith individual, where variable Yim takes possible values 1, 2, …, rm, and ci = 1,2, …, C denote the latent class membership of the ith individual. Let xi = (xi1, xi2,, xip) ’ be a vector of explanatory variables for the ith individual. A LCR model can be expressed as:
Pr( Y 1 = y 1 ,, Y n = y n )= i=1 n c=1 C γ c ( x i ) m=1 M k=1 r m ρ mk|c I( y im =k) , MathType@MTEF@5@5@+=feaaguart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacPqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=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@7C55@
where ρmk|c= Pr (Yim = k | ci = c) is the conditional probability of response k to the mth item given class membership c; γc (xi) is the class membership probabilities for the cth class given xi, which is related to γc through a multinomial logistic regression:
γ c ( x i )=Pr( c i =c| x i )= exp( x i ' β c ) 1+ j=1 C1 exp( x i ' β j ) , MathType@MTEF@5@5@+=feaaguart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacPqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=He9q8qqQ8frFve9Fve9Ff0dmeaabaqaciGacaGaaeqabaWaaqaafaaakeaaqaaaaaaaaaWdbiabeo7aN9aadaWgaaWcbaWdbiaadogaa8aabeaak8qadaqadaWdaeaaieWapeGaa8hEa8aadaWgaaWcbaWdbiaadMgaa8aabeaaaOWdbiaawIcacaGLPaaacqGH9aqpciGGqbGaaiOCamaabmaapaqaa8qacaWGJbWdamaaBaaaleaapeGaamyAaaWdaeqaaOWdbiabg2da9iaadogacaqG8bGaa8hEa8aadaWgaaWcbaWdbiaadMgaa8aabeaaaOWdbiaawIcacaGLPaaacqGH9aqpdaWcaaWdaeaapeGaaeyzaiaabIhacaqGWbGaaiikaiaa=HhapaWaa0baaSqaa8qacaWGPbaapaqaa8qacaGGNaaaaOGaeqOSdi2damaaBaaaleaapeGaam4yaaWdaeqaaOWdbiaacMcaa8aabaWdbiaaigdacqGHRaWkdaqfWaqabSWdaeaapeGaamOAaiabg2da9iaaigdaa8aabaWdbiaadoeacqGHsislcaaIXaaan8aabaWdbiabggHiLdaakiaabwgacaqG4bGaaeiCaiaacIcacaWF4bWdamaaDaaaleaapeGaamyAaaWdaeaapeGaai4jaaaakiabek7aI9aadaWgaaWcbaWdbiaadQgaa8aabeaak8qacaGGPaaaaiaacYcaaaa@6DAB@
where βc is a vector of logistic regression coefficients, with the reference class as C.
MLCR
If we denote cij as the class membership of the ith individual coming from the jth cluster, γjc (xij,wj) as the probability that cij =c with level 1 covariate xij andlevel 2 covariate wj , assuming random intercept ujc ~ N (0, 1), a MLCR model can be written as:
Level 1 (individual):
γ jc ( x ij , w j )=Pr( c ij =c| x ij , w j )= exp( β 0jc + β 1c x ij ) 1+ k=1 C1 exp( β 0jk + β 1k x ij ) , MathType@MTEF@5@5@+=feaaguart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacPqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=He9q8qqQ8frFve9Fve9Ff0dmeaabaqaciGacaGaaeqabaWaaqaafaaakeaaiiGaqaaaaaaaaaWdbiab=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@8498@
Level 2 (cluster): β 0jc = α 0c + α 1c w j + σ c u jc , MathType@MTEF@5@5@+=feaaguart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacPqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=He9q8qqQ8frFve9Fve9Ff0dmeaabaqaciGacaGaaeqabaWaaqaafaaakeaaiiaaqaaaaaaaaaWdbiab=j7aI9aadaWgaaWcbaWdbiaaicdacaWGQbGaam4yaaWdaeqaaOWdbiabg2da9GGaciab+f7aH9aadaWgaaWcbaWdbiaaicdacaWGJbaapaqabaGcpeGaey4kaSIaeqySde2damaaBaaaleaapeGaaGymaiaadogaa8aabeaak8qacaWG3bWdamaaBaaaleaapeGaamOAaaWdaeqaaOWdbiabgUcaRiab+n8aZ9aadaWgaaWcbaWdbiaadogaa8aabeaak8qacaWG1bWdamaaBaaaleaapeGaamOAaiaadogaa8aabeaakiaacYcaaaa@5332@
The intra-class correlation (ICC) for class c in MLCR is defined as the proportion of the variance of the random effects out of the total variance, i. e. , i.e., .  
r c = σ c 2 σ c 2 + π 2 /3   (1). MathType@MTEF@5@5@+=feaaguart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacPqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=He9q8qqQ8frFve9Fve9Ff0dmeaabaqaciGacaGaaeqabaWaaqaafaaakeaaqaaaaaaaaaWdbiaadkhapaWaaSbaaSqaa8qacaWGJbaapaqabaGcpeGaeyypa0ZaaSaaa8aabaWdbiabeo8aZ9aadaqhaaWcbaWdbiaadogaa8aabaWdbiaaikdaaaaak8aabaWdbiabeo8aZ9aadaqhaaWcbaWdbiaadogaa8aabaWdbiaaikdaaaGccqGHRaWkcqaHapaCpaWaaWbaaSqabeaapeGaaGOmaaaakiaac+cacaaIZaaaaiaacckacaqGGaGaaiikaiaaigdacaGGPaGaaiOlaaaa@4FFC@
When C is more than 2, MLCR-O allows the C-1 random intercepts ujc to be correlated with one another. On the other hand, the MLCR-P uses a common factor to model all the random intercepts (i. e. , ujc= ujfor all c), which assumes the random intercepts are perfectly correlated.
Monte carlo simulation
In the Monte Carlo simulation study, we generated data from a 3-class MLCR model using the R 2.15. 2 package. The observed vector, Yi = (Yi1,,Yi3) , have 3 categorical variables, each of which has 5 categories. The model includes 2 covariate variables. One covariate is a Level 1 continuous variable with standard normal distribution, and the other is a Level 2 binary covariate. We assign Class 3 as the reference class, i. e. , α03 = α13 = β13= σ3= 0. We set the regression parameters as α0102 = -1, α11 = 0.5,α12 = -0.5, β11= 0.5,β12 = -0.5. The random intercept uj1is correlated with uj2through a bivariate normal distribution:
( u j1 u j2 )~BVN( ( 0 0 ),( 1 0.5 0.5 1 ) ). MathType@MTEF@5@5@+=feaaguart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacPqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=He9q8qqQ8frFve9Fve9Ff0dmeaabaqaciGacaGaaeqabaWaaqaafaaakeaaqaaaaaaaaaWdbmaabmaapaqaauaabeqaceaaaeaapeGaamyDa8aadaWgaaWcbaWdbiaadQgacaaIXaaapaqabaaakeaapeGaamyDa8aadaWgaaWcbaWdbiaadQgacaaIYaaapaqabaaaaaGcpeGaayjkaiaawMcaaiaac6hacaWGcbGaamOvaiaad6eadaqadaWdaeaapeWaaeWaa8aabaqbaeqabiqaaaqaa8qacaaIWaaapaqaa8qacaaIWaaaaaGaayjkaiaawMcaaiaacYcadaqadaWdaeaafaqabeGacaaabaWdbiaaigdaa8aabaWdbiaaicdacaGGUaGaaGynaaWdaeaapeGaaGimaiaac6cacaaI1aaapaqaa8qacaaIXaaaaaGaayjkaiaawMcaaaGaayjkaiaawMcaaiaac6caaaa@547F@
In addition we set different values of σ1= σ2= σ to obtain ICCs at various levels. A value of σ = 0.416 generates data with ICC of 0.05, a value of σ = 0.6 generates data with ICC of 0.1, and a value of σ = 1 gives us data with ICC of 0.25. Finally, the conditional item response probabilities for m = 1,2,3 were chosen as:
( ρ m1|1 , ρ m2|1 , ρ m3|1 ,  ρ m4|1 ,  ρ m5|1 )= ( 0.05, 0.8, 0.05, 0.05, 0.05 ), MathType@MTEF@5@5@+=feaaguart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacPqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=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@7423@
( ρ m1|2 , ρ m2|2 ,ρ , m3|2 ,  ρ m4|2 ,  ρ m5|2 )= ( 0.05, 0.05, 0.1, 0.4, 0.4 ) MathType@MTEF@5@5@+=feaaguart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacPqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=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@72AB@
( ρ m1|3 , ρ m2|3 , ρ m3|3 ,  ρ m4|3 ,  ρ m5|3 )= ( 0.3, 0.3, 0.3, 0.05, 0.05 ). MathType@MTEF@5@5@+=feaaguart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacPqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=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@72B2@
We applied the SLCR, MLCR-P and MLCR-O approaches to estimate the regression coefficients of the two covariates using Mplus 7 software package [4]. Since MLCR-O adopts the true model, we expect MLCR-O would perform the best. We generated 500 replications of 3000 subjects with different ICCs. The 3000 subjects were grouped into 30 or 300 equally sized clusters, half of which were assigned with 0 for the Level 2 covariate, and the other half were assigned with 1.
Results and Discussion
Table 1 presents the biases and 95% confidence interval coverage rates for the estimates of regression coefficients using different models. It clearly shows that SLCR had the largest bias and worst confidence interval coverage among the three methods, especially for the Level 2 covariate. With the increase of ICC, the biases in regression coefficients increased while the coverage probabilities decreased. SLCR also had worse performance when only a limited number of clusters were available in the data comparing to the scenario when a large number of clusters were collected. In general, these results are consistent with the simulation results of multilevel logistic regression models [5], suggesting the importance of using multilevel analysis techniques when you have clustered/correlated data that do not satisfy the conditional independence assumption, especially when the regression coefficients for level 2 covariates are of interest.
Table 1 Summary of the relative bias ( θ ^ θ θ ) MathType@MTEF@5@5@+=feaaguart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacNqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=He9q8qqQ8frFve9Fve9Ff0dmeaabaqaciGacaGaaeqabaWaaqaafaaakeaadaqadaqaamaalaaabaGafqiUdeNbaKaacqGHsislcqaH4oqCaeaacqaH4oqCaaaacaGLOaGaayzkaaaaaa@4284@ and 95% confidence interval coverage for the regression coefficients.

Table 1

# of groups

Group Size

ICC

Class

Coefficient

Relative Bias

95% CI

Coverage Rate

 

 

 

 

 

SLCR

MLCR-P

MLCR-O

SLCR

MLCR-P

MLCR-O

30

100

0.05

Class 1

Level 1 Covariate

-.02

.00

.00

.93

.94

.95

 

 

 

 

Level 2 Covariate

-.03

.02

-.03

.69

.94

.94

 

 

 

 

Intercept

.03

-.00

.01

.84

.93

.94

 

 

 

Class 2

Level 1 Covariate

.01

.00

-.01

.96

.95

.94

 

 

 

 

Level 2 Covariate

-.01

-.01

-.01

.87

.94

.93

 

 

 

 

Intercept

.02

.03

.01

.93

.94

.93

30

100

0.1

Class 1

Level 1 Covariate

-.05

-.00

.00

.92

.93

.93

 

 

 

 

Level 2 Covariate

-.06

-01

.01

.59

.94

.95

 

 

 

 

Intercept

.06

.00

-.00

.70

.94

.92

 

 

 

Class 2

Level 1 Covariate

.01

.02

.01

.94

.93

.95

 

 

 

 

Level 2 Covariate

-.02

-.02

-.01

.76

.94

.94

 

 

 

 

Intercept

.05

.04

.00

.85

.91

.92

30

100

0.25

Class 1

Level 1 Covariate

-.14

-.01

.00

.68

.92

.93

 

 

 

 

Level 2 Covariate

-.13

.05

-.00

.44

.93

.94

 

 

 

 

Intercept

.13

-.00

.02

.51

.93

.91

 

 

 

Class 2

Level 1 Covariate

.05

.09

.00

.93

.90

.95

 

 

 

 

Level 2 Covariate

.09

.12

.03

.66

.92

.94

 

 

 

 

Intercept

.11

.13

-.01

.75

.88

.93

300

10

0.05

Class 1

Level 1 Covariate

-.03

.00

-.00

.94

.95

.95

 

 

 

 

Level 2 Covariate

-.05

-.01

-.01

.92

.96

.94

 

 

 

 

Intercept

.04

-.01

.01

.94

.96

.96

 

 

 

Class 2

Level 1 Covariate

.00

.00

-.01

.96

.95

.95

 

 

 

 

Level 2 Covariate

.01

.01

-.01

.93

.94

.96

 

 

 

 

Intercept

.03

.01

-.00

.95

.96

.97

300

10

0.1

Class 1

Level 1 Covariate

-.06

.01

.01

.92

.95

.94

 

 

 

 

Level 2 Covariate

-.07

-.00

.01

.87

.95

.95

 

 

 

 

Intercept

.06

-.01

-.00

.90

.95

.97

 

 

 

Class 2

Level 1 Covariate

.02

.04

-.02

.93

.93

.94

 

 

 

 

Level 2 Covariate

.02

.02

.01

.95

.95

.95

 

 

 

 

Intercept

.04

.04

.01

.94

.95

.93

300

10

0.25

Class 1

Level 1 Covariate

-.16

-.01

.01

.63

.95

.95

 

 

 

 

Level 2 Covariate

-.16

-.01

-.01

.75

.96

.96

 

 

 

 

Intercept

.16

.02

-.00

.63

.96

.94

 

 

 

Class 2

Level 1 Covariate

.04

.10

.01

.95

.92

.95

 

 

 

 

Level 2 Covariate

.04

.08

-.00

.93

.95

.96

 

 

 

 

Intercept

.15

.18

-.01

.87

.87

.94

Table 1 Summary of the relative bias ( θ ^ θ θ ) MathType@MTEF@5@5@+=feaaguart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbdfwBIjxAHbqedmvETj2BSbqefm0B1jxALjhiov2DaerbuLwBLnhiov2DGi1BTfMBaebbnrfifHhDYfgasaacNqpw0le9v8qqaqFD0xXdHaVhbbf9v8qqaqFr0xc9pk0xbba9q8WqFfea0=yr0RYxir=Jbba9q8aq0=yq=He9q8qqQ8frFve9Fve9Ff0dmeaabaqaciGacaGaaeqabaWaaqaafaaakeaadaqadaqaamaalaaabaGafqiUdeNbaKaacqGHsislcqaH4oqCaeaacqaH4oqCaaaacaGLOaGaayzkaaaaaa@4284@ and 95% confidence interval coverage for the regression coefficients.

×
When the MLCR-P procedure was compared to the MLCR-O procedure, it appears that the loss of efficiency was not substantial especially when ICC was low. Even for the cases with an ICC of 0.25, MLCR-P was only slightly worse than MLCR-O in terms of biases and coverage rates. Meanwhile, MLCR-P procedure took much less computation time than MLCR-O (2.5 minutes vs. 30 minutes for each simulated dataset on a PC with CPU of Intel i5 2.40GHz).
Conclusions
Consistent with previous studies of multilevel logistic regression models [5] and multilevel data analysis techniques broadly [6], the results of this empirical study demonstrate the importance of using multilevel regressions, more specifically, MLCR when performing LCR for clustered/correlated data structure.
When the number of latent classes (C) is more than 2, the computational complexity of the estimation procedure for MLCR-O increases rapidly with the increase of C. To alleviate the computational intensity and reduce computation time, the perfect correlation assumption for the random intercepts may be adopted to use MLCR-P in those situations. However, attention should be paid to the possible bias brought by such misspecification. Based on our Monte Carlo simulation results, we conclude that the bias caused by the misspecification of perfect correlation among random intercepts in MLCR-P model estimation is slight, especially when the ICC is low. Therefore, MLCR-P might serve as a computationally efficient method without substantial loss of accuracy in parameter estimates, hence could be a reasonable substitute for MLCR-O procedure when computation burden is a concern.
Acknowledgements
Manuscript preparation was supported in part by American Diabetes Association (ADA #7-12-CT-36, L. Jiang).

Cite this article: Jiang L, Chen S (2013) A Monte Carlo Study of Multilevel Latent Class Regression. Ann Biom Biostat 1: 1004.
Right Table
Current Issue Vol.1.1
Footer
Content:   Home  |  Aims & Scope  |  Early Online  |  Current Issue  | 
Journal Info:   Editorial Board  |  Article Processing Charges  |  FAQs
Contact Us
2952 Market Street, Suite 140
San Diego, California 92102, USA
Tel: 1-619-373-8030
Fax: 1-619-793-4845
Toll free number: 1-800-762-9856
Copyright © 2013 JSciMed Central. All rights reserved.