R中使用anesrake报错‘无变量偏差超5%’的技术求助

错误提示：“使用所选方法时，无变量偏差超过5%，要么无需加权，要么应选择更小的预耙限制”

尝试在数据集上应用anesrake方法实现加权时收到上述错误，已确认变量无空水平、变量名匹配且已转为因子，但问题仍未解决。代码中通过Age与CustomerType生成了AgePack交互变量，并将所有类别的目标占比设为5.5555556%，相关R代码如下：

#LoadDataSet
NPSSurvey_df <- read.csv('C:/Users/andavies/Desktop/Maru_NPS_CSAT_RAWDATA_13_12_2022_F1/Test.csv')
NPSSurvey_df <- as.data.frame(NPSSurvey_df)


new_data <- NPSSurvey_df %>%
  mutate(AgePack = case_when(Age == '18-24' & CustomerType =='In-Life' ~ '18-24 & In-Life',
                            Age == '25-34' & CustomerType =='In-Life' ~ '25-34 & In-Life',
                            Age == '35-44' & CustomerType =='In-Life' ~ '35-44 & In-Life',
                            Age == '45-54' & CustomerType =='In-Life' ~ '45-54 & In-Life',
                            Age == '55-64' & CustomerType =='In-Life' ~ '55-64 & In-Life',
                            Age == '65 years or more' & CustomerType =='In-Life' ~ '65+ & In-Life',
                           
                            Age == '18-24' & CustomerType =='Lapsed' ~ '18-24 & Lapsed',
                            Age == '25-34' & CustomerType =='Lapsed' ~ '25-34 & Lapsed',
                            Age == '35-44' & CustomerType =='Lapsed' ~ '35-44 & Lapsed',
                            Age == '45-54' & CustomerType =='Lapsed' ~ '45-54 & Lapsed',
                            Age == '55-64' & CustomerType =='Lapsed' ~ '55-64 & Lapsed',
                            Age == '65 years or more' & CustomerType =='Lapsed' ~ '65+ & Lapsed',
                            
                            Age == '18-24' & CustomerType =='New' ~ '18-24 & New',
                            Age == '25-34' & CustomerType =='New' ~ '25-34 & New',
                            Age == '35-44' & CustomerType =='New' ~ '35-44 & New',
                            Age == '45-54' & CustomerType =='New' ~ '45-54 & New',
                            Age == '55-64' & CustomerType =='New' ~ '55-64 & New',
                            Age == '65 years or more' & CustomerType =='New' ~ '65+ & New'))


new_data$AgePack <- as.factor(new_data$AgePack)
levels(new_data$AgePack) <- c('18-24 & In-Life','25-34 & In-Life', '35-44 & In-Life', '45-54 & In-Life', '55-64 & In-Life', '65+ & In-Life',
                              '18-24 & Lapsed','25-34 & Lapsed', '35-44 & Lapsed', '45-54 & Lapsed', '55-64 & Lapsed', '65+ & Lapsed',
                              '18-24 & New','25-34 & New', '35-44 & New', '45-54 & New', '55-64 & New', '65+ & New')
                          
AgePack <- c(5.555555555555556,5.555555555555556,5.555555555555556,5.555555555555556,5.555555555555556,5.555555555555556,
             5.555555555555556,5.555555555555556,5.555555555555556,5.555555555555556,5.555555555555556,5.555555555555556,
             5.555555555555556,5.555555555555556,5.555555555555556,5.555555555555556,5.555555555555556,5.555555555555556)

names(AgePack) <- c('18-24 & In-Life','25-34 & In-Life', '35-44 & In-Life', '45-54 & In-Life', '55-64 & In-Life', '65+ & In-Life',
                    '18-24 & Lapsed','25-34 & Lapsed', '35-44 & Lapsed', '45-54 & Lapsed', '55-64 & Lapsed', '65+ & Lapsed',
                    '18-24 & New','25-34 & New', '35-44 & New', '45-54 & New', '55-64 & New', '65+ & New')

target <- list(AgePack)

names(target) <- c("AgePack")

outsave <- anesrake(target, new_data, caseid = new_data$Response_ID,
                    verbose= TRUE, cap = 5, choosemethod = "total",
                    type = "pctlim", pctlim = .05 , nlim = 5,
                    iterate = TRUE , force1 = TRUE)

summary(outsave)

new_data$weightvec <- unlist(outsave[1])

排查步骤建议： #

• 核实样本分布与目标值的偏差：运行prop.table(table(new_data$AgePack)) * 100计算每个AgePack类别的实际占比，对比目标值5.555%。如果所有类别的偏差都小于5%，说明样本已经符合目标分布，无需加权，这是报错的直接原因。
• 调整预耙限制参数：若需要强制对齐目标（即使偏差较小），可将pctlim参数调小，比如设置为pctlim = 0.01（1%的偏差阈值），这样即使偏差小于5%但大于1%，anesrake会执行加权操作。
• 确认交互变量的完整性：用table(new_data$AgePack)检查每个AgePack类别是否都有样本（无零频数），同时确保类别名称与目标变量AgePack的命名完全一致（包括空格、符号等）。
• 验证目标值总和：确认sum(AgePack)是否等于100，anesrake要求目标占比总和必须为100%（这里18*5.555≈100，理论上没问题，但建议实际计算验证）。
• 尝试不同的加权触发逻辑：修改type参数为"total"或"nlim"，结合调整nlim的值，或者暂时关闭iterate = FALSE执行单次耙平，观察是否能生成权重。

内容的提问来源于stack exchange，提问作者Andrew Davies