How to re-assign values from different rows to same row

Question

I have this dataframe; specifically, the column phase distinguishes alphabetical groups with values from A through E and the column duration contains the durations of these phases:

df
   speaker                                                 action duration phase
39   ID2-A                                    <no: yeah: it 's:>=    0.872  <NA>
40   ID1-G          ((m: r hand holds up three fingers ifo face))    1.769     A      # group 1
41   ID1-P     ((m: r hand holds up three fingers ifo face)) prep    0.679     B      # group 1
42   ID1-P   ((m: r hand holds up three fingers ifo face)) stroke    0.399     C      # group 1
56   ID1-A                             °>that 's crazy<°  anyway=    1.514  <NA>
57   ID1-G                                 ((m: r hand airslaps))    0.938     A      # group 2
58   ID1-P                            ((m: r hand airslaps)) prep    0.299     B      # group 2
59   ID1-P                          ((m: r hand airslaps)) stroke    0.261     C      # group 2
60   ID1-P                           ((m: r hand airslaps)) relax    0.374     E      # group 2
61    <NA>                                                (0.057)    0.057  <NA>
62   ID2-A                                  =yeah >I don 't know<    0.582  <NA>

I want to create new columns prep, stroke, hold, and relax, filled, where available, with the duration values for phase B, C, D(not shown in the example), and E all put side by side on the same row as the respective A values. The expected output is this:

df
   speaker                                                 action duration phase  prep stroke hold relax
39   ID2-A                                    <no: yeah: it 's:>=    0.872  <NA>    NA     NA   NA    NA
40   ID1-G          ((m: r hand holds up three fingers ifo face))    1.769     A 0.679  0.399   NA    NA
41   ID1-P     ((m: r hand holds up three fingers ifo face)) prep    0.679     B    NA     NA   NA    NA
42   ID1-P   ((m: r hand holds up three fingers ifo face)) stroke    0.399     C    NA     NA   NA    NA
56   ID1-A                             °>that 's crazy<°  anyway=    1.514  <NA>    NA     NA   NA    NA
57   ID1-G                                 ((m: r hand airslaps))    0.938     A 0.299  0.261   NA 0.374
58   ID1-P                            ((m: r hand airslaps)) prep    0.299     B    NA     NA   NA    NA
59   ID1-P                          ((m: r hand airslaps)) stroke    0.261     C    NA     NA   NA    NA
60   ID1-P                           ((m: r hand airslaps)) relax    0.374     E    NA     NA   NA    NA
61    <NA>                                                (0.057)    0.057  <NA>    NA     NA   NA    NA
62   ID2-A                                  =yeah >I don 't know<    0.582  <NA>    NA     NA   NA    NA

To get there I've created the new columns using ifelse:

df$prep <- ifelse(df$phase=="B", df$duration, NA)
df$stroke <- ifelse(df$phase=="C", df$duration, NA)
df$hold <- ifelse(df$phase=="D", df$duration, NA)
df$relax <- ifelse(df$phase=="E", df$duration, NA)

This works fine. The problematic part is the transfer of the duration values to the A rows. I've tried using lead, for example:

library(dplyr)
df$prep <- lead(df$prep, 1)

The issue here is that the number of positions to lead can vary if not all 5 phases A through E are present per group. For example, if the D phase is missing as in the example (cf. rows 59-60), then the number of positions to lead the duration value associated with phase E is not 4 but 3.

Any suggestion for how to solve this?

Reproducible data:

df <- dput(t[c(39:42,56:62), c(2:3,5:6)])
structure(list(speaker = c("ID2-A", "ID1-G", "ID1-P", "ID1-P", 
"ID1-A", "ID1-G", "ID1-P", "ID1-P", "ID1-P", NA, "ID2-A"), action = c("   <no: yeah: it 's:>=", 
"   ((m: r hand holds up three fingers ifo face))", "  ((m: r hand holds up three fingers ifo face)) prep", 
"  ((m: r hand holds up three fingers ifo face)) stroke", "   °>that 's crazy<°  anyway=", 
"   ((m: r hand airslaps))", "  ((m: r hand airslaps)) prep", 
"  ((m: r hand airslaps)) stroke", "  ((m: r hand airslaps)) relax", 
"(0.057)", "   =yeah >I don 't know<"), duration = c(0.872, 1.769, 
0.679, 0.399, 1.514, 0.938, 0.299, 0.261, 0.374, 0.057, 0.582
), phase = c(NA, "A", "B", "C", NA, "A", "B", "C", "E", NA, NA
)), row.names = c(39L, 40L, 41L, 42L, 56L, 57L, 58L, 59L, 60L, 
61L, 62L), class = "data.frame")

EDIT:

The phasegroups are not always neatly separated by NA, as in this example:

df
   speaker                                        action duration phase
29    <NA>                                      canceled    3.672  <NA>
30   ID1-G          ((m: r hand imitates throwing away))    1.478     A
31   ID1-P     ((m: r hand imitates throwing away)) prep    0.254     B
32   ID1-P   ((m: r hand imitates throwing away)) stroke    0.775     C
33   ID1-P     ((m: r hand imitates throwing away)) hold    0.450     D
34   ID1-G                            ((m: r hand nods))    1.584     A
35   ID1-P                       ((m: r hand nods)) prep    0.466     B
36   ID1-P                     ((m: r hand nods)) stroke    0.324     C
37   ID1-P                      ((m: r hand nods)) relax    0.785     E
38    <NA>                                       (0.071)    0.071  <NA>

Answer 1

Using reshaping data long-to-wide pivot_wider, then merge left_join:

library(tidyverse)

# add a group column
x <- df %>% 
  mutate(grp = cumsum(is.na(phase)),
         grp = ifelse(is.na(phase), NA, grp))

# reshape
y <- x %>% 
  select(grp, phase, duration) %>% 
  filter(!is.na(grp)) %>% 
  pivot_wider(id_cols = grp, names_from = phase, values_from = duration) %>% 
  mutate(phase = "A")

# merge
left_join(x, y, by = c("phase", "grp"))

#    speaker                                                 action duration phase grp     A     B     C     E
# 1    ID2-A                                    <no: yeah: it 's:>=    0.872  <NA>  NA    NA    NA    NA    NA
# 2    ID1-G          ((m: r hand holds up three fingers ifo face))    1.769     A   1 1.769 0.679 0.399    NA
# 3    ID1-P     ((m: r hand holds up three fingers ifo face)) prep    0.679     B   1    NA    NA    NA    NA
# 4    ID1-P   ((m: r hand holds up three fingers ifo face)) stroke    0.399     C   1    NA    NA    NA    NA
# 5    ID1-A                             °>that 's crazy<°  anyway=    1.514  <NA>  NA    NA    NA    NA    NA
# 6    ID1-G                                 ((m: r hand airslaps))    0.938     A   2 0.938 0.299 0.261 0.374
# 7    ID1-P                            ((m: r hand airslaps)) prep    0.299     B   2    NA    NA    NA    NA
# 8    ID1-P                          ((m: r hand airslaps)) stroke    0.261     C   2    NA    NA    NA    NA
# 9    ID1-P                           ((m: r hand airslaps)) relax    0.374     E   2    NA    NA    NA    NA
# 10    <NA>                                                (0.057)    0.057  <NA>  NA    NA    NA    NA    NA
# 11   ID2-A                                  =yeah >I don 't know<    0.582  <NA>  NA    NA    NA    NA    NA