project_kids_items

[ben-domingue]

this one needs to be fixed. in particular, i want to uncouple the wave and person id but i also want the wave to be ordinal numeric values. so i think we need to go back to the code.
table: https://redivis.com/datasets/as2e-cv7jb41fd/tables/r4fk-ckdx7s2mr?tablesList-search=project_ki&tablesList-orderDirection=DESC&tablesList-orderBy=relevance&tablesList-tableToken=tablesList#cells

see also: https://domingue-lab.slack.com/archives/C05AKJGU3UJ/p1733849254844299?thread_ts=1733766196.979549&cid=C05AKJGU3UJ

@saviranadela & @KingArthur0205 any chance one of you could take a look at this?

[saviranadela]

hi @ben-domingue i have a question

i saw wave has these values

they distinguished the waves into "wave" and "grade", how would you want it to be transformed to ordinal numeric?

codebook:
Codebook_itemleveldata_PKKIDS.xlsx

[ben-domingue]

@saviranadela let me dive in

[ben-domingue]

I actually think that this just needs to be redone from the ground-up. The description reads:

Data on behavioral and achievement assessment items from elementary school children

These data were processed from before we were being as strict about 'different scales -> different tables'. But I think we can redo this somewhat efficiently using the original code:
https://github.com/ben-domingue/irw/blob/main/data/project_kids_items.R

[saviranadela]

got it! will re-do from the ground-up

[saviranadela]

@ben-domingue one more question: i see that one item level could have multiple types of waves:


i'm thinking to stick with non-numeric waves, but happy to hear your suggestion!

[ben-domingue]

let me see if i follow: i think the issue here is that they are discussing things in a confusing way. they talk about grades and waves and it isn't really clear what is going on. wave 3 presumably occurs after wave 1. but what does that have to do with grade 1/3 followup? it's kind of difficult to know. does this seem right?

my sense is that wave should be ordinal given that it captures a temporal element of data collection and time is always ordered. what if we split the data here into the chunk we can order by wave (so the first two told rows) and that we can order by 'follow' (so the latter two?). or, failing that, maybe i should email the authors asking for clarity?

[saviranadela]

@ben-domingue do you mean splitting them like this?
project_kids_items_split.zip
makes more sense to me as we split items also based on different "types" of waves, but please let me know if this isn't what you meant

[saviranadela]

code:

library(tidyverse)
library(readr)

df_raw <- read_csv('PK_ItemLevelData.csv')

names(df_raw) <- tolower(names(df_raw))

# find variables with no response or single responses to drop
# put them in a list to drop
drop_vars <- c()

for (i in 1:ncol(df_raw)) {
  unique_vals <- unique(df_raw[[i]])
  unique_len <- length(unique_vals)
  
  if (unique_len == 1 & is.na(unique(unique_vals[1]))) {
    drop_vars <- append(drop_vars, names(df_raw)[i])
  }
  
  if (unique_len == 2 & (is.na(unique_vals[1]) | is.na(unique_vals[2]))) {
    drop_vars <- append(drop_vars, names(df_raw)[i])
  }
}


df_raw <- df_raw |>
  # drop unneeded variables
  select(-all_of(drop_vars),
         -pk_id,
         -starts_with('ctrs'),
         -starts_with('swan'),
         -starts_with('ssrs'),
         -starts_with('tq')) |>
  # create participant ID
  mutate(id = row_number()) 

# transform tosrec assessment variables
tosrec <- df_raw |>
  select(id,
         starts_with('tosrec_g2c')) |>
  pivot_longer(cols = -id,
               names_to = c('pt1', 'wave', 'pt2', 'pt3'),
               names_sep = '_',
               values_to = 'resp',
               values_drop_na = T) |>
  mutate(wave = 'g2_end',
         item = paste0(pt1, '_', pt2, '_', pt3),
         wave_temp = '3') |>
  select(id, item, wave, wave_temp, resp) 

# transform variables with three underscores       
three <- df_raw |>
  select(id,
         starts_with('ctopp'),
         starts_with('told'),
         starts_with('wj_ak'),
         starts_with('wj_ap'),
         starts_with('wj_lw'),
         starts_with('wj_pc'),
         starts_with('wj_pv'),
         starts_with('wj_qc'),
         starts_with('wj_sa'),
         starts_with('wj_spell'),
         starts_with('wj_wa'),
         starts_with('wj_wf')) |>
  pivot_longer(cols = -id,
               names_to = c('pt1', 'pt2', 'pt3', 'wave'),
               names_sep = '_',
               values_to = 'resp',
               values_drop_na = T) |>
  mutate(item = paste0(pt1, '_', pt2, '_', pt3)) |>
  mutate(wave_temp = case_when(wave == 'g1' ~ '1',
                          wave == 'g2' ~ '2',
                          wave == 'g3' ~ '3',
                          wave == 'w1' ~ '1',
                          wave == 'w2' ~ '2',
                          wave == 'w3' ~ '3')) |>
  select(id, item, wave, wave_temp, resp) 

# transform kbit assessment variables
kbit <- df_raw |>
  select(id,
         starts_with('kbit')) |>
  pivot_longer(cols = -id,
               names_to = 'item',
               values_to = 'resp',
               values_drop_na = T) |>
  mutate(wave = NA, wave_temp = NA) |>
  select(id, item, wave, wave_temp, resp)

# transform variables with two underscores
two <- df_raw |>
  select(id,
         starts_with('swan'),
         starts_with('topel')) |>
  pivot_longer(cols = -id,
               names_to = c('pt1', 'pt2', 'wave'),
               names_sep = '_',
               values_to = 'resp',
               values_drop_na = T) |>
  mutate(item = paste0(pt1, '_', pt2)) |>
  mutate(wave_temp = case_when(wave == 'g1' ~ '1',
                               wave == 'g2' ~ '2',
                               wave == 'g3' ~ '3',
                               wave == 'w1' ~ '1',
                               wave == 'w2' ~ '2',
                               wave == 'w3' ~ '3')) |>
  select(id, item, wave, wave_temp, resp)

tosrec2 <- df_raw |>
  select(id,
         starts_with('tosrec_g1c'),
         starts_with('tosrec_g2a')) |>
  pivot_longer(cols = -id,
               names_to = c('pt1', 'wave', 'pt2', 'pt3', 'pt4'),
               names_sep = '_',
               values_to = 'resp',
               values_drop_na = T) |>
  # mutate(wave = case_when(wave == 'g1c' ~ 'g1_end',
  #                         wave == 'g2a' ~ 'g2_beginning'),
  #        item = paste0(pt1, '_', pt2, '_', pt3, '_', pt4)) |>
  mutate(wave_temp = case_when(wave == 'g1c' ~ '1',
                          wave == 'g2a' ~ '2'),
         item = paste0(pt1, '_', pt2, '_', pt3, '_', pt4)) |>
  select(id, item, wave, wave_temp, resp)

# transform variables with four underscores
four <- df_raw |>
  select(id,
         starts_with('wj_mf')) |>
  pivot_longer(cols = -id,
               names_to = c('pt1', 'pt2', 'pt3', 'pt4', 'wave'),
               names_sep = '_',
               values_to = 'resp',
               values_drop_na = T) |>
  mutate(item = paste0(pt1, '_', pt2, '_', pt3, '_', pt4)) |>
  mutate(wave_temp = case_when(wave == 'g1' ~ '1',
                               wave == 'g2' ~ '2',
                               wave == 'g3' ~ '3',
                               wave == 'w1' ~ '1',
                               wave == 'w2' ~ '2',
                               wave == 'w3' ~ '3')) |>
  select(id, item, wave, wave_temp, resp) 

df <- rbind(four, kbit, three, tosrec, tosrec2, two)

df$check <- str_sub(df$item, 1, 5)

df_ctopp <- df %>%
  filter(grepl("ctopp",df$check)) %>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_mf <- df %>%
  filter(grepl("wj_mf",df$check)) %>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_kbit <- df %>%
  filter(grepl("kbit",df$check)) %>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")


df_wj_lw_grade <- df %>%
  filter(grepl("wj_lw",df$check), grepl("g",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_lw_wave <- df %>%
  filter(grepl("wj_lw",df$check), grepl("w",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_pc_grade <- df %>%
  filter(grepl("wj_pc",df$check), grepl("g",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_pc_wave <- df %>%
  filter(grepl("wj_pc",df$check), grepl("w",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_pv_grade <- df %>%
  filter(grepl("wj_pv",df$check), grepl("g",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_pv_wave <- df %>%
  filter(grepl("wj_pv",df$check), grepl("w",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_ak_grade <- df %>%
  filter(grepl("wj_ak",df$check), grepl("g",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_ak_wave <- df %>%
  filter(grepl("wj_ak",df$check), grepl("w",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_sa <- df %>%
  filter(grepl("wj_sa",df$check)) %>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_wa_grade <- df %>%
  filter(grepl("wj_wa",df$check), grepl("g",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_wa_wave <- df %>%
  filter(grepl("wj_wa",df$check), grepl("w",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_wf <- df %>%
  filter(grepl("wj_wf",df$check)) %>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_ap <- df %>%
  filter(grepl("wj_ap",df$check)) %>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_qc <- df %>%
  filter(grepl("wj_qc",df$check)) %>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_told_grade <- df %>%
  filter(grepl("told",df$check), grepl("g",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_told_wave <- df %>%
  filter(grepl("told",df$check), grepl("w",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_spell_grade <- df %>%
  filter(grepl("wj_sp",df$check), grepl("g",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_wj_spell_wave <- df %>%
  filter(grepl("wj_sp",df$check), grepl("w",wave))%>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_tosrec <- df %>%
  filter(grepl("tosre",df$check)) %>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

df_topel <- df %>%
  filter(grepl("topel",df$check)) %>%
  select(id, item, wave_temp, resp) %>%
  rename("wave" = "wave_temp")

write.csv(df_ctopp, "project_kids_ctopp.csv", row.names=FALSE)
write.csv(df_wj_mf, "project_kids_wj_mf.csv", row.names=FALSE)
write.csv(df_kbit, "project_kids_kbit.csv", row.names=FALSE)
write.csv(df_wj_lw_grade, "project_kids_wj_lwid_grade.csv", row.names=FALSE)
write.csv(df_wj_lw_wave, "project_kids_wj_lwid_wave.csv", row.names=FALSE)
write.csv(df_wj_pc_grade, "project_kids_wj_pc_grade.csv", row.names=FALSE)
write.csv(df_wj_pc_wave, "project_kids_wj_pc_wave.csv", row.names=FALSE)
write.csv(df_wj_pv_grade, "project_kids_wj_pv_grade.csv", row.names=FALSE)
write.csv(df_wj_pv_wave, "project_kids_wj_pv_wave.csv", row.names=FALSE)
write.csv(df_wj_ak_grade, "project_kids_wj_ak_grade.csv", row.names=FALSE)
write.csv(df_wj_ak_wave, "project_kids_wj_ak_wave.csv", row.names=FALSE)
write.csv(df_wj_sa, "project_kids_wj_sa.csv", row.names=FALSE)
write.csv(df_wj_wa_grade, "project_kids_wj_wa_grade.csv", row.names=FALSE)
write.csv(df_wj_wa_wave, "project_kids_wj_wa_wave.csv", row.names=FALSE)
write.csv(df_wj_wf, "project_kids_wj_wf.csv", row.names=FALSE)
write.csv(df_wj_ap, "project_kids_wj_ap.csv", row.names=FALSE)
write.csv(df_wj_qc, "project_kids_wj_qc.csv", row.names=FALSE)
write.csv(df_told_grade, "project_kids_told_grade.csv", row.names=FALSE)
write.csv(df_told_wave, "project_kids_told_wave.csv", row.names=FALSE)
write.csv(df_wj_spell_grade, "project_kids_wj_spell_grade.csv", row.names=FALSE)
write.csv(df_wj_spell_wave, "project_kids_wj_spell_wave.csv", row.names=FALSE)
write.csv(df_tosrec, "project_kids_tosrec.csv", row.names=FALSE)
write.csv(df_topel, "project_kids_topel.csv", row.names=FALSE)

[ben-domingue]

woo boy. this data is tricky! let's focus on just one pair:

==> project_kids_wj_ak_grade.csv <==
"id","item","wave","resp"
2551,"wj_aka_10s","1",1
2551,"wj_aka_11s","1",1
2551,"wj_aka_12s","1",1
2551,"wj_aka_13s","1",1
2551,"wj_aka_14s","1",1
2551,"wj_aka_15s","1",1
2551,"wj_aka_16s","1",0
2551,"wj_aka_17s","1",1
2551,"wj_aka_18s","1",1

==> project_kids_wj_ak_wave.csv <==
"id","item","wave","resp"
2,"wj_aka_1s","1",1
2,"wj_aka_10s","1",1
2,"wj_aka_11s","1",1
2,"wj_aka_12s","1",1
2,"wj_aka_13s","1",1
2,"wj_aka_14s","1",0
2,"wj_aka_15s","1",0
2,"wj_aka_16s","1",0
2,"wj_aka_10s","3",1

from the index, these should be:

[going to followup below]

[ben-domingue]

what i think is happening:

• the _grade file is the one 'grade' row (row 4)
• if you look at the original documentation you can see that the data come from two studies: "These data represent a combined dataset of Drs. Stephanie Al Otaiba and Carol Connor's RCT projects. Therefore, these data are not unique, in that the original investigators may have shared the individual datasets in other places that we are not aware of."
• my guess is that 'grade' is one study and 'wave' is another.
  if that is the case, i think we are ok. but can someone double check my thinking. a very thorny puzzle....

excellent work @saviranadela !

[saviranadela]

my guess is that 'grade' is one study and 'wave' is another.

yes, it might be! agree with you. but still, we can’t be 100% sure what was going on just yet. either way, i think splitting those waves is already a good move. however, i am happy to email the project PIC if you think it’s necessary!

[ben-domingue]

i'm willing to live on the edge. i feel pretty confident that is the right thing! unless you remain dubious @saviranadela i would be ok processing the ones we have here :)

[saviranadela]

sounds good!

PR: #746

[ben-domingue]

thanks @saviranadela ! adding these now