R语言分组后计算当前weight与最后观测weight的差值问题

按country分组、wave排序后，计算每个wave的当前weight与该分组内上一个非缺失weight的差值。例如country "a"的wave5，需计算当前weight（69）与wave5之前最后一个有观测的weight（65）的差；wave8则计算当前weight（82）与wave8之前最后一个有观测的weight（69）的差。

set.seed(42000)
data <- data_frame(
  country = sample(letters[1:20], size = 100, replace = TRUE),
  weight = round(runif(100, min = 48, max = 90)))

data <- data %>%
  group_by(country)  %>%
  arrange(weight) %>%
  mutate(wave = seq_along(weight))

n_rows <- nrow(data)
perc_missing <- 10
data[sample(1:n_rows, sample(1:n_rows, round(perc_missing/100 * n_rows, 0))), c("weight")] <- NA 

最初使用lag()函数的方法无法满足需求，因为它只会取前一行的值，无法跳过缺失值找到最近的非缺失值：

data <- 
  data %>%
  group_by(country) %>%
  arrange(wave) %>%
  mutate(lag_weight = weight - lag(weight, default = first(weight, na.rm = TRUE))) 

使用zoo包中的na.locf()函数（全称Last Observation Carried Forward，即最近观测值向前填充）可以解决问题：

library(zoo)

data <- 
  data %>%
  group_by(country) %>%
  arrange(wave) %>%
  # 用最近的非缺失值填充，得到每个位置之前最后一个有效weight
  mutate(weight.io = na.locf(weight)) %>%
  # 当前weight减去上一个填充后的有效weight，得到目标差值
  mutate(lag_weight = weight - lag(weight.io))

内容的提问来源于stack exchange，提问作者R and C.F