R 中的多重插补（solve.default(xtx + diag(pen)) 中的错误：系统在计算上是奇异的：倒数条件数 =）

Question

我想分析有关 Covid-19 的数据。我已经完成了部分数据清理，最终得到了 this 数据集（160260 行和 34 列）。我已将变量 Continental,Location,Tests_Units 转换为因子。我想检查缺失值，所以我计算了缺失值的百分比，结果是：

> (colMeans(is.na(dataset1)))*100
          continent                location                    date             total_cases 
          0.0000000               0.0000000               0.0000000               1.9699239 
          new_cases            total_deaths              new_deaths       reproduction_rate 
          2.0366904               8.0094846               8.1130663              14.0078622 
       icu_patients           hosp_patients   weekly_icu_admissions  weekly_hosp_admissions 
         84.7747410              83.7021091              96.2386123              92.5851741 
        total_tests               new_tests           positive_rate          tests_per_case 
         54.4465244              56.6966180              43.9292400              44.7154624 
        tests_units people_fully_vaccinated        new_vaccinations        stringency_index 
         38.0974666              73.6390865              76.2298765              15.7138400 
         population      population_density              median_age           aged_70_older 
          0.0000000               4.3073755              10.5291401              11.0077374 
     gdp_per_capita         extreme_poverty   cardiovasc_death_rate     diabetes_prevalence 
         11.9381006              42.0897292              11.0077374               6.7003619 
     female_smokers            male_smokers  handwashing_facilities         life_expectancy 
         32.9963809              33.9535754              55.9690503               0.4785973 
        human_development_index        excess_mortality
         13.3738924                    96.1225509 

我不想分析缺失值的数据集，因此我进行了很多搜索，以找到填充这些 NA 的方法。我发现我可以使用 mouse 函数来填充这些 NA。我的目标是:

1. 以不将变量日期用作预测变量的方式使用 mouse 函数。
2. 不要估算变量中的值：大陆、位置、日期、人口，因为它们没有 NA。
3. 估算变量中的值：total_cases、new_cases、total_deaths、new_deaths、reproductive_rate、icu_患者、hosp_患者、weekly_icu_admissions、weekly_hosp_admissions、total_tests、new_tests、positive_rate、tests_per_case、people_complete_vaccinated、new_vaccinations、stringency_index、population_密度、median_age、ged _70_老年人，人均国内生产总值，极端贫困， Cardiovasc_death_rate、diabetes_prevalence、female_smokers、male_smokers、handwashing_facilities、life_expectancy、 human_development_index、excess_mortality 使用方法 pmm（预测均值匹配），因为这些变量是数字。
4. 使用方法polyreg（多分逻辑回归）插补变量tests_units中的值，因为该变量是一个具有4个水平的因子。

我按照this链接中的每一步进行操作，并运行此代码：

library(mice)

init = mice(dataset1,maxit = 0)
meth = init$method
predM = init$predictorMatrix

predM[, c("date")] = 0 #goal number 1

meth[c("continent","location","date","population")] = "" #goal number 2

meth[c("total_cases","new_cases","total_deaths","new_deaths","reproduction_rate",
   "icu_patients","hosp_patients","weekly_icu_admissions",
   "weekly_hosp_admissions","total_tests","new_tests","positive_rate",
   "tests_per_case","people_fully_vaccinated",
   "new_vaccinations","stringency_index","population_density","median_age",
   "aged_70_older","gdp_per_capita","extreme_poverty",
   "cardiovasc_death_rate","diabetes_prevalence","female_smokers",
   "male_smokers","handwashing_facilities","life_expectancy",
   "human_development_index","excess_mortality")]="pmm" #goal number 3

meth[c("tests_units")] = "polyreg" #goal number 4

set.seed(103)

imputed = mice(dataset1, method=meth, predictorMatrix=predM, m=5)

我得到的结果是

> library(mice)
> init = mice(dataset1,maxit = 0)
Warning message:
Number of logged events: 1 
> meth = init$method
> predM = init$predictorMatrix
> predM[, c("date")] = 0
> meth[c("continent","location","date","population")] = ""
> meth[c("total_cases","new_cases","total_deaths","new_deaths","reproduction_rate",
+        "icu_patients","hosp_patients","weekly_icu_admissions",
+        "weekly_hosp_admissions","total_tests","new_tests","positive_rate",
+        "tests_per_case","people_fully_vaccinated",
+        "new_vaccinations","stringency_index","population_density","median_age",
+        "aged_70_older","gdp_per_capita","extreme_poverty",
+        "cardiovasc_death_rate","diabetes_prevalence","female_smokers",
+        "male_smokers","handwashing_facilities","life_expectancy",
+        "human_development_index","excess_mortality")]="pmm"
> meth[c("tests_units")] = "polyreg"
> 
> set.seed(103)
> imputed = mice(dataset1, method=meth, predictorMatrix=predM, m=5)

 iter imp variable
  1   1  total_casesError in solve.default(xtx + diag(pen)) : 
  system is computationally singular: reciprocal condition number = 2.80783e-24

这不太令人愉快。我应该更改什么或应该运行哪些代码？

提前致谢！

Answer 1

您检查过您记录的事件吗？

view(init$loggedEvents)

也许是因为您使用的插补方法（“polyreg”）。您是否尝试过使用更强大的方法，例如

pmm

？

Answer 2

我认为这是因为与矩阵求逆相关的问题（可能与一个或多个变量中有太多类别有关）。