如何构建我的数据以进行周期性事件分析？答案

【问题标题】：How can I structure my data for a recurrent event analysis?如何构建我的数据以进行周期性事件分析？
【发布时间】：2021-12-20 01:39:06
【问题描述】：

我对 R 很陌生，所以也许这看起来很简单，但我想不通。我的数据看起来像 Df，它需要看起来像 Df2：

Df <- data.frame(country = c("A", "A", "A", "A", "A", "B","B", "B", "B"),
                 year = c("1950", "1951", "1952", "1953", "1954", "1950", "1951", "1952", "1953"), 
                 start_year = c("NA", "1951", "1951", "NA", "1954", "1950", "NA", "1951", "1951"), 
                 end_year= c("NA", "NA", "1952", "NA", "1954", "1950", "NA", "NA", "NA"),
                 status = c(0, 1, 1, 0, 1, 1, 0, 1, 1),
                 treatment = c(10, "NA", 20, 5, "NA", "NA", 30, 100, 10))

Df2 <- data.frame(country = c("A", "A", "A", "A", "B","B", "B"),
                  time1 = c("1950", "1951", "1953", "1954", "1950", "1951", "1952"),
                  time2 = c("1951", "1953", "1954", "1955", "1951", "1952", "1954"),
                  status = c(0, 1, 0, 1, 1, 0, 1),
                  treatment = c(10, 20, 0, "NA", "NA", 30, 110))

我们的目标是在一个结构中进行 PWP 循环事件分析。 Df2 中的处理应该是间隔时间 1 到时间 2 的处理值之和。

有什么想法可以到达那里吗？谢谢！

【问题讨论】：

为什么 B 带有 time1 = 1952 和 treatment = 110 但 A has time1 = 1951` 和 treatment = 20。这里选择year或start_year的逻辑是什么？
@Martin Gal Treatment 在我的数据中是对妇女组织的资助金额。而year 是观察年份，也就是资金发放的年份。 start_year 显示冲突何时开始。现在有意义吗？

标签： r survival-analysis

【解决方案1】：

你可以使用

library(dplyr)

Df %>% 
  mutate(across(where(is.character), ~na_if(.x, "NA")),
         time1 = as.numeric(coalesce(start_year, year)),
         treatment = as.numeric(treatment)) %>% 
  group_by(country, time1, status) %>% 
  summarise(treatment = sum(treatment, na.rm = TRUE), .groups = "drop") %>% 
  group_by(country) %>% 
  mutate(time2 = lead(time1, default = last(time1) + 1)) %>% 
  select(country, time1, time2, status, treatment) %>% 
  ungroup()

得到

# A tibble: 7 x 5
  country time1 time2 status treatment
  <chr>   <dbl> <dbl>  <dbl>     <dbl>
1 A        1950  1951      0        10
2 A        1951  1953      1        20
3 A        1953  1954      0         5
4 A        1954  1955      1         0
5 B        1950  1951      1         0
6 B        1951  1951      0        30
7 B        1951  1952      1       110

这并不完全是您想要的输出（请参阅我的评论），而是解决您的问题的开始。

【讨论】：

【解决方案2】：

Df2 <- Df %>% mutate(episode = data.table::rleid(status))

library(tidyverse)
Df2 <- Df2 %>%
  arrange(country, year) %>%
  group_by(country, episode) %>%
  mutate(time1 = min(year))

Df2 <- Df2 %>%
  arrange(country, year) %>%
  group_by(country, episode) %>%
  mutate(time2 = (max(as.numeric(year) + 1)))

我创建了一个剧集标识符并设法为每个episode 识别time1 和time2。现在，我仍然需要合并由episode 分组的行，以便每集有一行显示treatment 的总和。任何想法如何做到这一点？

【讨论】：