如何替换第一个事件中缺失的观察值的缺失值？
How to replace missing values for observations that are missing on the first event?
我在面板数据中输入社会 class 变量。我想用第一波出现的第一个社交 class 替换所有社交 class 变量。但是，有些人缺少第一波的数据。我还想归因于这些人的社交 class，但只是在第一次出现社交 class 变量之后。如果我使用以下命令：
df %>%
group_by(id) %>%
dplyr::mutate(class_imputed = first(class))
# Groups: id [4]
id wave year class class_imputed
<dbl> <dbl> <dbl> <fct> <fct>
1 1 1 2007 3 3
2 1 2 2008 2 3
3 1 3 2009 2 3
4 1 4 2010 2 3
5 2 1 2005 . .
6 2 2 2006 2 .
7 2 3 2007 3 .
8 2 4 2008 . .
9 3 1 2007 . .
10 3 2 2008 . .
11 3 3 2009 1 .
12 3 4 2010 1 .
13 4 1 2009 2 2
14 4 2 2010 . 2
15 4 3 2011 3 2
16 4 4 2012 . 2
我们可以看到，ID 2 和 3 的人没有被插补。但我只想估算第一次观察到社会 class 变量的时期。这意味着 ID 号为 2 的人将在 2006 年、2007 年和 2008 年获得社交 class = 2，但在 2005 年不会。有人可以帮忙吗？
这是数据：
structure(list(id = c(1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4,
4, 4, 4), wave = c(1, 2, 3, 4, 1, 2, 3, 4, 1, 2, 3, 4, 1, 2,
3, 4), year = c(2007, 2008, 2009, 2010, 2005, 2006, 2007, 2008,
2007, 2008, 2009, 2010, 2009, 2010, 2011, 2012), class = structure(c(4L,
3L, 3L, 3L, 1L, 3L, 4L, 1L, 1L, 1L, 2L, 2L, 3L, 1L, 4L, 1L), .Label = c(".",
"1", "2", "3"), class = "factor"), class_imputed = structure(c(3L,
3L, 3L, 3L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L), .Label = c(".",
"2", "3"), class = "factor")), row.names = c(NA, -16L), class = "data.frame")
library(tidyverse)
data <- structure(list(id = c(1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4,
4, 4, 4), wave = c(1, 2, 3, 4, 1, 2, 3, 4, 1, 2, 3, 4, 1, 2,
3, 4), year = c(2007, 2008, 2009, 2010, 2005, 2006, 2007, 2008,
2007, 2008, 2009, 2010, 2009, 2010, 2011, 2012), class = structure(c(4L,
3L, 3L, 3L, 1L, 3L, 4L, 1L, 1L, 1L, 2L, 2L, 3L, 1L, 4L, 1L), .Label = c(".",
"1", "2", "3"), class = "factor"), class_imputed = structure(c(3L,
3L, 3L, 3L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L), .Label = c(".",
"2", "3"), class = "factor")), row.names = c(NA, -16L), class = "data.frame")
data %>%
na_if(".") %>%
mutate(class_imputed = class) %>%
fill(class_imputed)
#> id wave year class class_imputed
#> 1 1 1 2007 3 3
#> 2 1 2 2008 2 2
#> 3 1 3 2009 2 2
#> 4 1 4 2010 2 2
#> 5 2 1 2005 <NA> 2
#> 6 2 2 2006 2 2
#> 7 2 3 2007 3 3
#> 8 2 4 2008 <NA> 3
#> 9 3 1 2007 <NA> 3
#> 10 3 2 2008 <NA> 3
#> 11 3 3 2009 1 1
#> 12 3 4 2010 1 1
#> 13 4 1 2009 2 2
#> 14 4 2 2010 <NA> 2
#> 15 4 3 2011 3 3
#> 16 4 4 2012 <NA> 3
由 reprex package (v2.0.1)
于 2021 年 10 月 1 日创建
您可以试试这个选项-
library(dplyr)
df %>%
group_by(id) %>%
dplyr::mutate(class_imputed = replace(class,
row_number() > match(TRUE, class != '.'),
class[class != '.'][1]))
# id wave year class class_imputed
# <dbl> <dbl> <dbl> <fct> <fct>
# 1 1 1 2007 3 3
# 2 1 2 2008 2 3
# 3 1 3 2009 2 3
# 4 1 4 2010 2 3
# 5 2 1 2005 . .
# 6 2 2 2006 2 2
# 7 2 3 2007 3 2
# 8 2 4 2008 . 2
# 9 3 1 2007 . .
#10 3 2 2008 . .
#11 3 3 2009 1 1
#12 3 4 2010 1 1
#13 4 1 2009 2 2
#14 4 2 2010 . 2
#15 4 3 2011 3 2
#16 4 4 2012 . 2
match(TRUE, class != '.') 将 return 值不同于 '.' 的第一个位置，而 class[class != '.'][1] 将 return 被估算的值，即第一个社会class变量。
使用 data.table
更容易
library(data.table)
i1 <- setDT(df)[, .I[cumsum(class != "." & any(class == ".")) > 0],id]$V1
df[i1, class_imputed := first(class), id]
-输出
> df
id wave year class class_imputed
1: 1 1 2007 3 3
2: 1 2 2008 2 3
3: 1 3 2009 2 3
4: 1 4 2010 2 3
5: 2 1 2005 . .
6: 2 2 2006 2 2
7: 2 3 2007 3 2
8: 2 4 2008 . 2
9: 3 1 2007 . .
10: 3 2 2008 . .
11: 3 3 2009 1 1
12: 3 4 2010 1 1
13: 4 1 2009 2 2
14: 4 2 2010 . 2
15: 4 3 2011 3 2
16: 4 4 2012 . 2
我在面板数据中输入社会 class 变量。我想用第一波出现的第一个社交 class 替换所有社交 class 变量。但是，有些人缺少第一波的数据。我还想归因于这些人的社交 class，但只是在第一次出现社交 class 变量之后。如果我使用以下命令：
df %>%
group_by(id) %>%
dplyr::mutate(class_imputed = first(class))
# Groups: id [4]
id wave year class class_imputed
<dbl> <dbl> <dbl> <fct> <fct>
1 1 1 2007 3 3
2 1 2 2008 2 3
3 1 3 2009 2 3
4 1 4 2010 2 3
5 2 1 2005 . .
6 2 2 2006 2 .
7 2 3 2007 3 .
8 2 4 2008 . .
9 3 1 2007 . .
10 3 2 2008 . .
11 3 3 2009 1 .
12 3 4 2010 1 .
13 4 1 2009 2 2
14 4 2 2010 . 2
15 4 3 2011 3 2
16 4 4 2012 . 2
我们可以看到，ID 2 和 3 的人没有被插补。但我只想估算第一次观察到社会 class 变量的时期。这意味着 ID 号为 2 的人将在 2006 年、2007 年和 2008 年获得社交 class = 2，但在 2005 年不会。有人可以帮忙吗？
这是数据：
structure(list(id = c(1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4,
4, 4, 4), wave = c(1, 2, 3, 4, 1, 2, 3, 4, 1, 2, 3, 4, 1, 2,
3, 4), year = c(2007, 2008, 2009, 2010, 2005, 2006, 2007, 2008,
2007, 2008, 2009, 2010, 2009, 2010, 2011, 2012), class = structure(c(4L,
3L, 3L, 3L, 1L, 3L, 4L, 1L, 1L, 1L, 2L, 2L, 3L, 1L, 4L, 1L), .Label = c(".",
"1", "2", "3"), class = "factor"), class_imputed = structure(c(3L,
3L, 3L, 3L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L), .Label = c(".",
"2", "3"), class = "factor")), row.names = c(NA, -16L), class = "data.frame")
library(tidyverse)
data <- structure(list(id = c(1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4,
4, 4, 4), wave = c(1, 2, 3, 4, 1, 2, 3, 4, 1, 2, 3, 4, 1, 2,
3, 4), year = c(2007, 2008, 2009, 2010, 2005, 2006, 2007, 2008,
2007, 2008, 2009, 2010, 2009, 2010, 2011, 2012), class = structure(c(4L,
3L, 3L, 3L, 1L, 3L, 4L, 1L, 1L, 1L, 2L, 2L, 3L, 1L, 4L, 1L), .Label = c(".",
"1", "2", "3"), class = "factor"), class_imputed = structure(c(3L,
3L, 3L, 3L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L), .Label = c(".",
"2", "3"), class = "factor")), row.names = c(NA, -16L), class = "data.frame")
data %>%
na_if(".") %>%
mutate(class_imputed = class) %>%
fill(class_imputed)
#> id wave year class class_imputed
#> 1 1 1 2007 3 3
#> 2 1 2 2008 2 2
#> 3 1 3 2009 2 2
#> 4 1 4 2010 2 2
#> 5 2 1 2005 <NA> 2
#> 6 2 2 2006 2 2
#> 7 2 3 2007 3 3
#> 8 2 4 2008 <NA> 3
#> 9 3 1 2007 <NA> 3
#> 10 3 2 2008 <NA> 3
#> 11 3 3 2009 1 1
#> 12 3 4 2010 1 1
#> 13 4 1 2009 2 2
#> 14 4 2 2010 <NA> 2
#> 15 4 3 2011 3 3
#> 16 4 4 2012 <NA> 3
由 reprex package (v2.0.1)
于 2021 年 10 月 1 日创建您可以试试这个选项-
library(dplyr)
df %>%
group_by(id) %>%
dplyr::mutate(class_imputed = replace(class,
row_number() > match(TRUE, class != '.'),
class[class != '.'][1]))
# id wave year class class_imputed
# <dbl> <dbl> <dbl> <fct> <fct>
# 1 1 1 2007 3 3
# 2 1 2 2008 2 3
# 3 1 3 2009 2 3
# 4 1 4 2010 2 3
# 5 2 1 2005 . .
# 6 2 2 2006 2 2
# 7 2 3 2007 3 2
# 8 2 4 2008 . 2
# 9 3 1 2007 . .
#10 3 2 2008 . .
#11 3 3 2009 1 1
#12 3 4 2010 1 1
#13 4 1 2009 2 2
#14 4 2 2010 . 2
#15 4 3 2011 3 2
#16 4 4 2012 . 2
match(TRUE, class != '.') 将 return 值不同于 '.' 的第一个位置，而 class[class != '.'][1] 将 return 被估算的值，即第一个社会class变量。
使用 data.table
library(data.table)
i1 <- setDT(df)[, .I[cumsum(class != "." & any(class == ".")) > 0],id]$V1
df[i1, class_imputed := first(class), id]
-输出
> df
id wave year class class_imputed
1: 1 1 2007 3 3
2: 1 2 2008 2 3
3: 1 3 2009 2 3
4: 1 4 2010 2 3
5: 2 1 2005 . .
6: 2 2 2006 2 2
7: 2 3 2007 3 2
8: 2 4 2008 . 2
9: 3 1 2007 . .
10: 3 2 2008 . .
11: 3 3 2009 1 1
12: 3 4 2010 1 1
13: 4 1 2009 2 2
14: 4 2 2010 . 2
15: 4 3 2011 3 2
16: 4 4 2012 . 2