SAS中day30.5日期差计算转R代码求助：结果不一致

原SAS代码 #

data data_new;  
set data_old;  
months=intck('day30.5',date1,date2);  
run;

用户尝试的R代码（结果不匹配） #

test_data <- data.frame(
  months2_sas = c(5, -2, -8, 10, 4, 7, -8, -7, -2, 2, 7, -7, -8, -11, -9, -8, 0, 11, 0, -2, -1, 7, -4, 7, -6, 10, 6, 9),
  date1 = as.Date(c("2021-06-16","2021-06-16","2020-09-23","2020-09-23","2020-09-23","2022-05-05","2022-03-10","2022-03-10","2022-03-10","2022-03-10","2022-03-10","2022-03-16","2020-02-04","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-04-28","2017-04-28","2020-10-31","2020-10-31","2023-06-22","2023-06-22")),
  date2 = as.Date(c("2021-12-03","2021-05-05","2020-02-10","2021-08-11","2021-02-09","2022-11-15","2021-08-02","2021-09-07","2022-02-02","2022-06-07","2022-10-31","2021-08-30","2019-06-21","2016-11-08","2017-01-06","2017-02-10","2017-10-02","2018-09-03","2017-10-10","2017-08-08","2017-09-04","2018-04-30","2016-12-09","2017-11-09","2020-05-18","2021-08-13","2023-12-05","2024-03-05"))
)

sas_epoch <- as.Date("1960-01-01")

test_data %>%
  mutate(
    months_r = floor(as.numeric(date2 - sas_epoch) / 30.5) - 
      floor(as.numeric(date1 - sas_epoch) / 30.5)
  ) %>%
  select(months2_sas,months_r)

正确的等价R实现 #

根据SAS官方文档，INTCK('DAY30.5', start, end) 的核心逻辑是：将日期转换为从1960-01-01开始的天数，分别对两个日期的天数按30.5天间隔向下取整，再计算差值。为避免浮点数精度误差，可将运算转换为整数乘法（30.5 = 61/2，除以30.5等价于乘以2再除以61）：

library(dplyr)

test_data <- data.frame(
  months2_sas = c(5, -2, -8, 10, 4, 7, -8, -7, -2, 2, 7, -7, -8, -11, -9, -8, 0, 11, 0, -2, -1, 7, -4, 7, -6, 10, 6, 9),
  date1 = as.Date(c("2021-06-16","2021-06-16","2020-09-23","2020-09-23","2020-09-23","2022-05-05","2022-03-10","2022-03-10","2022-03-10","2022-03-10","2022-03-10","2022-03-16","2020-02-04","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-10-28","2017-04-28","2017-04-28","2020-10-31","2020-10-31","2023-06-22","2023-06-22")),
  date2 = as.Date(c("2021-12-03","2021-05-05","2020-02-10","2021-08-11","2021-02-09","2022-11-15","2021-08-02","2021-09-07","2022-02-02","2022-06-07","2022-10-31","2021-08-30","2019-06-21","2016-11-08","2017-01-06","2017-02-10","2017-10-02","2018-09-03","2017-10-10","2017-08-08","2017-09-04","2018-04-30","2016-12-09","2017-11-09","2020-05-18","2021-08-13","2023-12-05","2024-03-05"))
)

sas_epoch <- as.Date("1960-01-01")

test_data %>%
  mutate(
    days1 = as.numeric(date1 - sas_epoch),
    days2 = as.numeric(date2 - sas_epoch),
    # 用整数运算避免浮点数精度偏差
    months_r = floor((days2 * 2) / 61) - floor((days1 * 2) / 61)
  ) %>%
  select(months2_sas, months_r)

结果差异说明 #

如果仍存在结果不匹配，可能的原因包括：

1. 浮点数精度误差：直接使用/30.5可能因浮点数精度导致取整结果偏差，整数乘法方案可解决此问题。
2. SAS参数差异：确认SAS代码是否使用了INTCK的可选参数（如'C'或'E'），这类参数会改变间隔计算逻辑。
3. 测试数据一致性：验证测试数据中的months2_sas是否确实由intck('day30.5', date1, date2)生成。

内容的提问来源于stack exchange，提问作者Ando