R（dplyr）：通过ID计算列中“1”之前的“0”观察值

Question

I have a dataset with two variables: ID , repeatvisit , and timeperiod . ID represents the individual that visits the clinic, while referredvisit represents whether that observation has been recommended a referral. In other words, referredvisit == 0 means that the individual is not referred to go to another clinic, while referredvisit == 1 represents a patient that is recommended a referral. timeperiod shows the sequence in which the individuals come in.

My dataset looks like this:

timeperiod <- 1:18
ID <- c("TOM", "TOM", "SALLY", "SALLY", "RICHIE", "TOM", "TOM", "SALLY", "RICHIE", "RICHIE", "RICHIE", "SALLY", "TOM", "TOM", "TOM", "RICHIE", "RICHIE", "RICHIE")
referredvisit <- c(0, 1, 1, 1, 0, 1, 0, 1, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0)
df <- cbind(timeperiod, ID, referredvisit)
df <- as.data.frame(df)

What I aim to do is to for every referredvisit == 0 , I would like to count how many rows of "1"s preceded it until it hits the beginning of the column (for the first 0) or until it hits another 0 (for the rest of the 0s) by ID. I want to create a column that store this count. My result for the dataset should look like this:

df$result <- c(0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 3, 0, 0, 1, 0, 2, 0)

I am actually trying to follow this link , but it can't seem to work as the link assumes that the ID is nicely sorted out. I was thinking that perhaps dplyr might help, but can't seem to figure something out too. Deeply appreciate if anyone could help me in this!

Thank you in advance!

EDIT: For better visualisation, the result will look like this. But this is only after I manually sort it by ID. Cause my original data set will contain thousands of rows, and it is difficult for me to sort the ID manually.

Answer 1

The differences of the positions of the zeros minus 1 gives the number of preceding ones and count_ones performs that calculation for a single ID where its argument is assumed to be a logical vector which is TRUE in zero positions. ave is then used to apply it to every ID . No packages are used.

count_ones <- function(is0) replace(is0, is0, diff(which(c(TRUE, is0))) - 1)    
transform(df, result = ave(referredvisit == 0, ID, FUN = count_ones))

giving:

   timeperiod     ID referredvisit result
1           1    TOM             0      0
2           2    TOM             1      0
3           3  SALLY             1      0
4           4  SALLY             1      0
5           5 RICHIE             0      0
6           6    TOM             1      0
7           7    TOM             0      2
8           8  SALLY             1      0
9           9 RICHIE             0      0
10         10 RICHIE             0      0
11         11 RICHIE             1      0
12         12  SALLY             0      3
13         13    TOM             0      0
14         14    TOM             1      0
15         15    TOM             0      1
16         16 RICHIE             1      0
17         17 RICHIE             0      2
18         18 RICHIE             0      0

Answer 2

Here is a tidyverse approach that reproduces your expected result (in column result2 )

df %>%
    mutate(referredvisit = as.numeric(as.character(referredvisit))) %>%
    arrange(ID) %>%
    group_by(ID) %>%
    mutate(
        flag = c(0, diff(referredvisit) < 0),
        grp = cumsum(flag)) %>%
    group_by(ID, grp) %>%
    mutate(cms = cumsum(referredvisit)) %>%
    ungroup() %>%
    mutate(result2 = ifelse(flag == 1, lag(cms), 0)) %>%
    select(-cms, -grp, -flag)
## A tibble: 18 x 5
#   timeperiod ID     referredvisit result result2
#   <fct>      <fct>          <dbl>  <dbl>   <dbl>
# 1 5          RICHIE            0.     0.      0.
# 2 9          RICHIE            0.     0.      0.
# 3 10         RICHIE            0.     0.      0.
# 4 11         RICHIE            1.     0.      0.
# 5 16         RICHIE            1.     0.      0.
# 6 17         RICHIE            0.     2.      2.
# 7 18         RICHIE            0.     0.      0.
# 8 3          SALLY             1.     0.      0.
# 9 4          SALLY             1.     0.      0.
#10 8          SALLY             1.     0.      0.
#11 12         SALLY             0.     3.      3.
#12 1          TOM               0.     0.      0.
#13 2          TOM               1.     0.      0.
#14 6          TOM               1.     0.      0.
#15 7          TOM               0.     2.      2.
#16 13         TOM               0.     0.      0.
#17 14         TOM               1.     0.      0.
#18 15         TOM               0.     1.      1.

---

Update

To keep the original ordering you could do

df %>%
    rowid_to_column("row") %>%
    mutate(referredvisit = as.numeric(as.character(referredvisit))) %>%
    arrange(ID) %>%
    group_by(ID) %>%
    mutate(
        flag = c(0, diff(referredvisit) < 0),
        grp = cumsum(flag)) %>%
    group_by(ID, grp) %>%
    mutate(cms = cumsum(referredvisit)) %>%
    ungroup() %>%
    mutate(result2 = ifelse(flag == 1, lag(cms), 0)) %>%
    arrange(row) %>%
    select(-cms, -grp, -flag, -row)
## A tibble: 18 x 5
#   timeperiod ID     referredvisit result result2
#   <fct>      <fct>          <dbl>  <dbl>   <dbl>
# 1 1          TOM               0.     0.      0.
# 2 2          TOM               1.     0.      0.
# 3 3          SALLY             1.     0.      0.
# 4 4          SALLY             1.     0.      0.
# 5 5          RICHIE            0.     0.      0.
# 6 6          TOM               1.     0.      0.
# 7 7          TOM               0.     2.      2.
# 8 8          SALLY             1.     0.      0.
# 9 9          RICHIE            0.     0.      0.
#10 10         RICHIE            0.     0.      0.
#11 11         RICHIE            1.     0.      0.
#12 12         SALLY             0.     3.      3.
#13 13         TOM               0.     0.      0.
#14 14         TOM               1.     0.      0.
#15 15         TOM               0.     1.      1.
#16 16         RICHIE            1.     0.      0.
#17 17         RICHIE            0.     2.      2.
#18 18         RICHIE            0.     0.      0.