R Mplus vs Lavan中报告的可用参数数

我在R的

lavan

中运行了一个SEM模型，现在正试图在Mplus中复制它。对于因子加载、系数和拟合度量，我得到了几乎相同的结果，因此复制看起来是成功的。但是我很困惑，因为

lavan

和Mplus报告了不同数量的自由估计参数

以下是

lavan

summary的摘录，作为说明：

lavaan 0.6-5 ended normally after 174 iterations

  Estimator                                         ML
  Optimization method                           NLMINB
  Number of free parameters                        254
  Number of equality constraints                    30
  Row rank of the constraints matrix                30

  Number of observations                          1622
  Number of missing patterns                       536

Model Test User Model:

  Test statistic                              2237.435
  Degrees of freedom                              1206
  P-value (Chi-square)                           0.000

Model Test Baseline Model:

  Test statistic                             19296.614
  Degrees of freedom                              1326
  P-value                                        0.000

User Model versus Baseline Model:

  Comparative Fit Index (CFI)                    0.943
  Tucker-Lewis Index (TLI)                       0.937

Mplus.version         "7"         
AnalysisType          "GENERAL"   
DataType              "INDIVIDUAL"
Estimator             "ML"        
Observations          "1622"      
NGroups               "1"         
NDependentVars        "51"        
NIndependentVars      "1"         
NContinuousLatentVars "12"        
Parameters            "224"       
ChiSqM_Value          "2237.424"  
ChiSqM_DF             "1206"      
ChiSqM_PValue         "0"         
ChiSqBaseline_Value   "19296.6"   
ChiSqBaseline_DF      "1326"      
ChiSqBaseline_PValue  "0"         
LL                    "-78566.54" 
UnrestrictedLL        "-77447.83" 
CFI                   "0.943"     
TLI                   "0.937"     
AIC                   "157581.1"  
BIC                   "158788.8"  
aBIC                  "158077.1"  
RMSEA_Estimate        "0.023"     
RMSEA_90CI_LB         "0.021"     
RMSEA_90CI_UB         "0.024"     
RMSEA_pLT05           "1"         
SRMR                  "0.048"     
AICC                  "157653.2" 

以下是Mplus摘要的摘录：

lavaan 0.6-5 ended normally after 174 iterations

  Estimator                                         ML
  Optimization method                           NLMINB
  Number of free parameters                        254
  Number of equality constraints                    30
  Row rank of the constraints matrix                30

  Number of observations                          1622
  Number of missing patterns                       536

Model Test User Model:

  Test statistic                              2237.435
  Degrees of freedom                              1206
  P-value (Chi-square)                           0.000

Model Test Baseline Model:

  Test statistic                             19296.614
  Degrees of freedom                              1326
  P-value                                        0.000

User Model versus Baseline Model:

  Comparative Fit Index (CFI)                    0.943
  Tucker-Lewis Index (TLI)                       0.937

Mplus.version         "7"         
AnalysisType          "GENERAL"   
DataType              "INDIVIDUAL"
Estimator             "ML"        
Observations          "1622"      
NGroups               "1"         
NDependentVars        "51"        
NIndependentVars      "1"         
NContinuousLatentVars "12"        
Parameters            "224"       
ChiSqM_Value          "2237.424"  
ChiSqM_DF             "1206"      
ChiSqM_PValue         "0"         
ChiSqBaseline_Value   "19296.6"   
ChiSqBaseline_DF      "1326"      
ChiSqBaseline_PValue  "0"         
LL                    "-78566.54" 
UnrestrictedLL        "-77447.83" 
CFI                   "0.943"     
TLI                   "0.937"     
AIC                   "157581.1"  
BIC                   "158788.8"  
aBIC                  "158077.1"  
RMSEA_Estimate        "0.023"     
RMSEA_90CI_LB         "0.021"     
RMSEA_90CI_UB         "0.024"     
RMSEA_pLT05           "1"         
SRMR                  "0.048"     
AICC                  "157653.2" 

如您所见，

lavan

报告了254个自由估计的参数，而Mplus报告了224个

我注意到，如果从254中减去相等约束数（30），则返回224，因此我想知道这是否解释了差异。我的猜测正确吗

---

提前感谢您的建议

在尝试使用mplus验证LaVan时（或反之亦然），我认为使用LaVan函数：mimic='mplus'
当使用模拟功能时，您应该发现自由参数的数量将匹配。此外，输出将更紧密地匹配Mplus输出的顺序，使您的生活更轻松。如果你还没有，Lavan作者的论文非常有用，可以在这里找到：jstatsoft.org/article/view/v048i02/v48i02.pdf

下面是一些示例代码

set.seed(9999)  
mat=matrix(nrow=500,ncol=3)  
mat[,1]<-rnorm(mean=5,sd=1,500)  
mat[,2]<-rnorm(mat[,1],n=500)  
mat[,3]<-rnorm(mat[,2],n=500)  
colnames(mat)<-c(paste0('Q',1:3))  
cfasyntax='F1=~Q1+Q2+Q3'  
model=cfa(cfasyntax,data=mat,mimic='mplus')  
summary(model)

set.seed（9999）
mat=矩阵（nrow=500，ncol=3）
根据我的经验，MPLU和Lavaan之间自由估计参数的差异确实是由于不平等的约束。然而，如果没有一个最小的可重复的例子，就不可能回答这类问题。为此，我们需要MPLU和Lavan模型规范。