在一个简单模型中使用奇异拟合,而在使用lmer R

时间:2020-01-15 20:11:59

标签: r lme4 multilevel-analysis lmertest

我正在使用lmerTest运行多级模型,其中员工嵌套在团队和部门中。我采用的是模型比较方法,因此我仅使用随机效果来构建模型。当我使用两个随机效应(团队成员和部门成员)来预测剧烈运动时,以下是结果:

library(lme4)
summary(m0_ev_io <- lmer(exer_vig ~ 1 + (1 | team_num) + (1 | dept_client), data = clean_data_0))


Linear mixed model fit by REML. t-tests use Satterthwaite's method ['lmerModLmerTest']
Formula: exer_vig ~ 1 + (1 | team_num) + (1 | dept_client)
   Data: clean_data_0

REML criterion at convergence: 527.5

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-1.6783 -0.6071 -0.2324  0.4233  2.1587 

Random effects:
 Groups      Name        Variance Std.Dev.
 team_num    (Intercept) 0.16687  0.4085  
 dept_client (Intercept) 0.03047  0.1746  
 Residual                1.14821  1.0715  
Number of obs: 169, groups:  team_num, 58; dept_client, 33

Fixed effects:
            Estimate Std. Error      df t value Pr(>|t|)    
(Intercept)   2.6743     0.1081 14.6284   24.74  2.4e-13 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

此模型以及所有后续模型运行良好,没有错误。但是,当我在精简运动中使用相同的数据运行相同的模型时,我得到一个奇异警告,突然部门成员之间没有差异:


summary(m0_el_io <- lmer(exer_lite ~ 1 + (1 | team_num) + (1 | dept_client), data = clean_data_0))


boundary (singular) fit: see ?isSingular
Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: exer_lite ~ 1 + (1 | team_num) + (1 | dept_client)
   Data: clean_data_0

REML criterion at convergence: 542

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-1.6403 -0.5925 -0.3208  0.4440  2.0776 

Random effects:
 Groups      Name        Variance Std.Dev.
 team_num    (Intercept) 0.1471   0.3835  
 dept_client (Intercept) 0.0000   0.0000  
 Residual                1.3027   1.1414  
Number of obs: 169, groups:  team_num, 58; dept_client, 33

Fixed effects:
            Estimate Std. Error      df t value Pr(>|t|)    
(Intercept)   2.7160     0.1037 42.5453    26.2   <2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
convergence code: 0
boundary (singular) fit: see ?isSingular 

除了因变量外,数据是相同的,所以我很困惑为什么会这样。我有信心这不是由于过拟合(例如在该线程(How to cope with a singular fit in a linear mixed model (lme4)?中),因为即使剧烈运动模型包含更多变量,它也不会发出奇异的警告。

您对发生这种情况的原因有什么想法,如何在不删除部门成员的情况下解决此问题?我已经尝试过其他站点的建议,包括将REML = FALSE和更改优化器[control = lmerControl(optimizer ='optimx',optCtrl = list(method ='L-BFGS-B')]],但没有任何效果。

谢谢!

编辑:这是数据示例。注意:team_num和dept_client是因素。

library(tidyverse)
clean_data_0 <- tibble(
  exer_lite = c(5, 4, 4, 5, 2, 4, 3, 1, 2, 2, 5,3, 4, 5, 2, 2, 2, 5, 5, 2, 3, 3, 1, 2, 5),
  exer_vig = c(4, 2, 4, 1, 2, 2, 3, 1, 2, 2, 5, 3, 3, 5, 2, 2, 3, 5, 5, 2, 3, 2, 1, 3, 5),
  dept_client = factor(c(17, 17, 45, 45, 80, 100, 90, 14, 2, 80, 100, 90, 121, 121, 121, 2, 90, 90, 90, 2, 100, 14, 14, 76, 76)),
  team_num = factor(c(509, 509, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 3, 3, 3, 4, 4, 5, 5, 6, 6, 13, 13, 14, 14)),
  id = c(1:25))

0 个答案:

没有答案