我使用程序包afex运行具有二项式链接函数的广义线性混合模型（GLMM），以使用以下代码分析任务的准确性：
model1.ds = mixed(corr ~ groupID*cond*degrees*axis + gender +
centerage + gender:groupID + centerage:groupID
+ (cond*degrees*axis|subjectID) + (1|target.id),
data = ds.data, method = 'LRT',control =
glmerControl(optCtrl=list(maxfun=2e8)),
family = binomial(link = 'logit'), all_fit=TRUE, cl = cl)
数据结构如下：
$ subjectID : Factor w/ 40 levels "101","102","103",..:
$ groupID : Factor w/ 2 levels "DS","HN":
$ corr : num 1 1 1 1 1 1 1 1 1 1 ...
$ target.id : Factor w/ 15 levels "1","10","11",..:
$ degrees : Factor w/ 4 levels "0","60","120",..:
$ axis : Factor w/ 2 levels "horizontal","vertical":
$ cond : Factor w/ 2 levels "same","mirrored":
$ gender : Factor w/ 2 levels "male","female":
$ centerage : atomic 8.61 8.61 8.61 8.61 8.61 ...
运行此模型时，收到以下警告：
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'groupID':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'cond':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'degrees':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'axis':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'gender':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'centerage':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'groupID:cond':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'groupID:degrees':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'cond:degrees':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'groupID:axis':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'cond:axis':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'degrees:axis':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'groupID:gender':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'groupID:centerage':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'groupID:cond:degrees':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'groupID:cond:axis':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'groupID:degrees:axis':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'cond:degrees:axis':
* boundary (singular) fit: see ?isSingularlme4 reported (at least) the following warnings for 'groupID:cond:degrees:axis':
* boundary (singular) fit: see ?isSingular
如果我理解正确，那么当随机效应方差估计为零或相关性估计为+/- 1时，通常会发生此类警告。但是，事实并非如此：
Model: gender:groupID + centerage:groupID + (cond * degrees * axis |
Model: subjectID) + (1 | target.id)
Data: ds.data
Df full model: 173
Effect df Chisq p.value
1 groupID 1 0.53 .47
2 cond 1 14.26 *** .0002
3 degrees 3 63.69 *** <.0001
4 axis 1 0.09 .77
5 gender 1 0.00 .95
6 centerage 1 5.34 * .02
7 groupID:cond 1 0.06 .81
8 groupID:degrees 3 7.92 * .05
9 cond:degrees 3 39.19 *** <.0001
10 groupID:axis 1 0.18 .67
11 cond:axis 1 0.27 .60
12 degrees:axis 3 17.56 *** .0005
13 groupID:gender 1 1.07 .30
14 groupID:centerage 1 1.36 .24
15 groupID:cond:degrees 3 7.26 + .06
16 groupID:cond:axis 1 0.58 .45
17 groupID:degrees:axis 3 0.70 .87
18 cond:degrees:axis 3 4.34 .23
19 groupID:cond:degrees:axis 3 2.27 .52
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘+’ 0.1 ‘ ’ 1
Generalized linear mixed model fit by maximum likelihood (Laplace Approximation) ['glmerMod']
Family: binomial ( logit )
Formula: corr ~ groupID * cond * degrees * axis + gender + centerage +
gender:groupID + centerage:groupID + (cond * degrees * axis | subjectID) + (1 | target.id)
Data: dots$data
Control: ctrl
AIC BIC logLik deviance df.resid
5551.3 6791.7 -2602.7 5205.3 9427
Scaled residuals:
Min 1Q Median 3Q Max
-10.0742 0.0984 0.2056 0.3345 2.9059
Random effects:
Groups Name Variance Std.Dev. Corr
subjectID (Intercept) 0.47951 0.6925
cond1 0.18497 0.4301 0.09
degrees1 0.37840 0.6151 0.04 -0.41
degrees2 0.07897 0.2810 0.14 0.74 0.09
degrees3 0.08058 0.2839 -0.03 -0.15 -0.71 -0.61
axis1 0.09348 0.3057 -0.13 -0.05 0.06 -0.12 0.33
cond1:degrees1 0.43913 0.6627 -0.03 -0.19 0.24 -0.16 0.11 0.63
cond1:degrees2 0.03610 0.1900 0.51 0.05 -0.37 0.01 0.32 0.31 0.01
cond1:degrees3 0.21020 0.4585 -0.23 -0.14 -0.08 -0.10 -0.02 -0.65 -0.83 -0.34
cond1:axis1 0.06011 0.2452 0.24 -0.28 -0.16 -0.56 0.62 0.48 0.25 0.47 -0.15
degrees1:axis1 0.39570 0.6290 -0.34 -0.59 0.40 -0.61 0.07 0.45 0.65 -0.11 -0.38 0.42
degrees2:axis1 0.15866 0.3983 0.46 0.06 -0.30 -0.05 0.19 -0.54 -0.82 0.20 0.69 0.13 -0.65
degrees3:axis1 0.09526 0.3086 -0.11 0.75 -0.16 0.93 -0.44 -0.18 -0.27 -0.03 0.05 -0.65 -0.68
cond1:degrees1:axis1 0.47493 0.6892 -0.27 -0.67 0.35 -0.74 0.23 0.45 0.65 -0.14 -0.33 0.53 0.97
cond1:degrees2:axis1 0.18847 0.4341 0.13 0.50 -0.49 0.49 -0.03 -0.23 -0.67 0.37 0.25 -0.42 -0.80
cond1:degrees3:axis1 0.13904 0.3729 0.52 0.72 0.13 0.76 -0.50 -0.05 0.07 0.04 -0.43 -0.37 -0.44
target.id (Intercept) 0.20089 0.4482
-0.03
-0.54 -0.80
0.47 0.63 -0.86
-0.05 0.56 -0.52 0.26
我尝试简化模型，但是即使是最简单的模型，每个主题“（1 | subjectID）”仅具有随机截距，也会引起关于奇点的警告。有关如何进行的任何建议？