在 R 中有条件地删除观察结果（在使用 MatchIt 包之后）
[英]Removing observations conditionally (after use of MatchIt package) in R
问题描述
I have used the package MatchIt to conduct an exact matching for treatment ( treat = 1 ) and control groups ( treat = 0 ) -- the matching was made through age .
我使用 package MatchIt对治疗组 ( treat = 1 ) 和对照组 ( treat = 0 ) 进行了精确匹配——匹配是通过age进行的。 The variable subclass reveals the matched units.变量subclass显示匹配的单元。
I would like to have one control unit selected randomly for each treated unit if it is matched to more than one control.如果它与多个控制匹配，我想为每个处理单元随机选择一个控制单元。 It is important that it be random.重要的是它是随机的。
If I have more than one treatment unit matched to only 1 control (case of subclass 4), I would like to discard such control unit as to keep the same number of controls and units for each subclass.如果我有多个治疗单元只匹配 1 个控制（ subclass 4 的情况），我想丢弃这样的控制单元，以便为每个子类保留相同数量的控制和单元。 In the end, I expect to have an equal number of observations for which treat = 1 and treat = 0.最后，我希望 treat = 1 和 treat = 0 的观察次数相等。
My real dataset is huge and consists of more than a million subclasses.我的真实数据集很大，包含超过一百万个子类。
structure(list(id = c("NSW1", "NSW57", "PSID6", "PSID84", "PSID147",
"PSID349", "PSID361", "PSID400", "NSW2", "NSW6", "NSW9", "NSW60",
"NSW77", "NSW80", "NSW127", "NSW161", "NSW169", "NSW177", "NSW179",
"PSID15", "PSID31", "PSID41", "PSID62", "PSID92", "PSID93", "PSID150",
"PSID167", "PSID178", "PSID254", "PSID292", "PSID300", "PSID308",
"PSID309", "PSID314", "PSID330", "NSW3", "NSW55", "NSW109", "PSID1",
"PSID69", "PSID91", "PSID165", "PSID166", "PSID302", "PSID378",
"ASID9033", "ASID9034", "ASID9036"), treat = c(1L, 1L, 0L, 0L,
0L, 0L, 0L, 0L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 0L,
0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 1L,
1L, 1L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 1L, 1L, 0L), age = c(37L,
37L, 37L, 37L, 37L, 37L, 37L, 37L, 22L, 22L, 22L, 22L, 22L, 22L,
22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L,
22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 30L, 30L, 30L, 30L, 30L,
30L, 30L, 30L, 30L, 30L, 29L, 29L, 29L), race = c("black", "black",
"black", "hispan", "white", "white", "white", "black", "hispan",
"black", "black", "white", "black", "black", "black", "black",
"black", "hispan", "white", "black", "hispan", "black", "white",
"white", "white", "hispan", "white", "white", "white", "white",
"black", "black", "white", "white", "black", "black", "black",
"black", "white", "black", "white", "white", "white", "white",
"white", "black", "white", "black"), married = c(1L, 0L, 1L,
0L, 1L, 1L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 1L, 1L, 0L,
1L, 0L, 1L, 1L, 1L, 0L, 0L, 1L, 0L, 1L, 1L, 1L, 1L, 1L, 0L, 0L,
0L, 1L, 0L, 1L, 0L, 0L, 1L, 1L, 1L, 1L, 1L, 0L, 0L), subclass = c(1L,
1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L,
2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L,
2L, 2L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 4L, 4L, 4L)), class = "data.frame", row.names = c(NA,
-48L))
3 个解决方案
解决方案1
2 已采纳 2022-04-19 12:40:56
Here's a (maybe a bit convoluted) way using group_split and map_dfr .这是使用group_split和map_dfr的（可能有点复杂）方式。
library(tidyverse)
df %>%
group_split(subclass) %>%
map_dfr(~ if(sum(.x$treat) > (nrow(.x) / 2)) bind_rows(.x[.x$treat == 0, ], sample_n(.x[.x$treat == 1, ], nrow(.x[.x$treat == 0, ])))
else if(sum(.x$treat) < (nrow(.x) / 2)) bind_rows(.x[.x$treat == 1, ], sample_n(.x[.x$treat == 0, ], nrow(.x[.x$treat == 1, ])))
else .x)
# A tibble: 34 x 6
id treat age race married subclass
<chr> <int> <int> <chr> <int> <int>
1 NSW1 1 37 black 1 1
2 NSW57 1 37 black 0 1
3 PSID400 0 37 black 0 1
4 PSID84 0 37 hispan 0 1
5 NSW2 1 22 hispan 0 2
6 NSW6 1 22 black 0 2
7 NSW9 1 22 black 0 2
8 NSW60 1 22 white 0 2
9 NSW77 1 22 black 0 2
10 NSW80 1 22 black 0 2
# ... with 24 more rows
解决方案2
2 2022-04-20 14:54:55
Another (base R) approach:另一种（基础 R）方法：
md <- do.call("rbind", unname(lapply(split(md, ~subclass),
function(x) {
x[c(which(x$treat == 1)[1],
which(x$treat == 0)[1]),]
})))
Grabs the first treated and first control unit from each subclass then rbind s them all together.从每个子类中获取第一个处理单元和第一个控制单元，然后将它们全部rbind在一起。 If your data are randomly ordered this is equivalent to randomly selecting one treated and one control unit.如果您的数据是随机排序的，这相当于随机选择一个处理单元和一个控制单元。
解决方案3
0 2022-04-19 13:05:02
Here's one simple approach这是一种简单的方法
library(tidyverse)
set.seed(999)
mydata %>%
mutate(r = runif(n = nrow(mydata))) %>%
arrange(r) %>%
group_by(treat, subclass) %>%
mutate(max_r = max(r)) %>%
filter(r == max_r) %>% select(-c(r, max_r)) -> mydata.filtered
I first create a random number r , then I arrange the data based on r .我首先创建一个随机数r ，然后我根据r排列数据。 Thereafter I calculate max(r) for each subclass x treat cell and drop everything where max(r) != r .此后，我为每个子类 x 处理单元计算max(r)并删除max(r) != r的所有内容。
This results in 1 treated and 1 non-treated obs for each subclass.这导致每个子类有 1 个已处理和 1 个未处理的 obs。
> table(mydata.filtered$treat, mydata.filtered$subclass)
1 2 3 4
0 1 1 1 1
1 1 1 1 1
data数据
mydata<- structure(list(id = c("NSW1", "NSW57", "PSID6", "PSID84", "PSID147",
"PSID349", "PSID361", "PSID400", "NSW2", "NSW6", "NSW9", "NSW60",
"NSW77", "NSW80", "NSW127", "NSW161", "NSW169", "NSW177", "NSW179",
"PSID15", "PSID31", "PSID41", "PSID62", "PSID92", "PSID93", "PSID150",
"PSID167", "PSID178", "PSID254", "PSID292", "PSID300", "PSID308",
"PSID309", "PSID314", "PSID330", "NSW3", "NSW55", "NSW109", "PSID1",
"PSID69", "PSID91", "PSID165", "PSID166", "PSID302", "PSID378",
"ASID9033", "ASID9034", "ASID9036"), treat = c(1L, 1L, 0L, 0L,
0L, 0L, 0L, 0L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 0L,
0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 1L,
1L, 1L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 1L, 1L, 0L), age = c(37L,
37L, 37L, 37L, 37L, 37L, 37L, 37L, 22L, 22L, 22L, 22L, 22L, 22L,
22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L,
22L, 22L, 22L, 22L, 22L, 22L, 22L, 22L, 30L, 30L, 30L, 30L, 30L,
30L, 30L, 30L, 30L, 30L, 29L, 29L, 29L), race = c("black", "black",
"black", "hispan", "white", "white", "white", "black", "hispan",
"black", "black", "white", "black", "black", "black", "black",
"black", "hispan", "white", "black", "hispan", "black", "white",
"white", "white", "hispan", "white", "white", "white", "white",
"black", "black", "white", "white", "black", "black", "black", "black", "white", "black", "white", "white", "white", "white",
"white", "black", "white", "black"), married = c(1L, 0L, 1L,
0L, 1L, 1L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 1L, 1L, 0L,
1L, 0L, 1L, 1L, 1L, 0L, 0L, 1L, 0L, 1L, 1L, 1L, 1L, 1L, 0L, 0L,
0L, 1L, 0L, 1L, 0L, 0L, 1L, 1L, 1L, 1L, 1L, 0L, 0L), subclass = c(1L,
1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L,
2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L,
2L, 2L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 4L, 4L, 4L)), class = "data.frame", row.names = c(NA,
-48L))
?MatchIt seems to also supply the ratio argument, which can be used to force 1-to-1 matching within the matching function call. ?MatchIt似乎还提供了ratio参数，可用于在匹配 function 调用中强制进行一对一匹配。
声明:本站的技术帖子网页，遵循CC BY-SA 4.0协议，如果您需要转载，请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.