wide2long {quest}

Reshape Multiple Sets of Variables From Wide to Long

Description

wide2long reshapes data from wide to long. This if often necessary to do with multilevel data where multiple sets of variables in the wide format seek to be reshaped to multiple rows in the long format. If only one set of variables needs to be reshaped, then you can use stack2 or melt.data.frame - but that does not work for *multiple* sets of variables. See details for more information.

Usage

wide2long(
  data,
  vrb.nm.list,
  grp.nm = NULL,
  sep = ".",
  rtn.obs.nm = "obs",
  order.by.grp = TRUE,
  keep.attr = FALSE
)

Arguments

data

data.frame of multilevel data in the wide format.

vrb.nm.list

A unique argument for the quest package such that it can take on different types of inputs. The conventional use is to provide a list of character vectors specifying each set of colnames to be reshaped. In longitudinal panel data, each list element would contain a score with multiple timepoints. The advanced use is to provide a single character vector specifying the colnames to be reshaped (not organized by sets). See details.

grp.nm

character vector specifying the colnames in data corresponding to the groups. Because data is in the wide format, data[grp.nm] must have unique rows (aka groups); if this is not the case, an error is returned. grp.nm can be NULL, in which case the rownames of data will be used. In longitudinal panel data this variable would be the participant ID variable.

sep

character vector of length 1 specifying the string in the column names provided by vrb.nm.list that separates out the name prefix from the number suffix. If sep = "", then that implies there is no string separating the name prefix and the number suffix (e.g., "outcome1").

rtn.obs.nm

character vector of length 1 specifying the new colname in the return object indicating which observation within each group the row refers to. In longitudinal panel data, this would be the returned time variable.

order.by.grp

logical vector of length 1 specifying whether to sort the return object first by grp.nm and then obs.nm (TRUE) or by obs.nm and then grp.nm (FALSE).

keep.attr

logical vector of length 1 specifying whether to keep the "reshapeLong" attribute (from reshape) in the return object.

Details

wide2long uses reshape(direction = "long") to reshape the data. It attempts to streamline the task of reshaping wide to long as the reshape arguments can be confusing because the same arguments are used for wide vs. long reshaping. See reshape if you are curious.

IF vrb.nm.list IS A LIST OF CHARACTER VECTORS: The conventional use of vrb.nm.list is to provide a list of character vectors, which specify each set of variables to be reshaped. For example, if data contains data from a longitudinal panel study with the same scores at different waves, then there might be a column for each score at each wave. vrb.nm.list would then contain an element for each score with each element containing a character vector of the colnames for that score at each wave (see examples). The names of the list elements would then be the colnames in the return object for those scores.

IF vrb.nm.list IS A CHARACTER VECTOR: The advanced use of vrb.nm.list is to provide a single character vector, which specify the variables to be reshaped (not organized by sets). In this case (i.e., if vrb.nm.list is not a list), then wide2long (really reshape) will attempt to guess which colnames go together as a set. It is assumed the following column naming scheme has been used: 1) have the same name prefix for columns within a set, 2) have the same number suffixes for each set of columns, 3) use, *and only use*, sep in the colnames to separate the name prefix and the number suffix. For example, the name prefixes might be "predictor" and "outcome" while the number suffixes might be "0", "1", and "2", and the separator might be ".", resulting in column names such as "outcome.1". The name prefix could include separators other than sep (e.g., "outcome_item.1"), but it cannot include sep (e.g., "outcome.item.1"). So "outcome_item1.1" could be acceptable, but "outcome.item1.1" would not.

Value

data.frame with nrow equal to nrow(data) * length(vrb.nm.list[[1]]) if vrb.nm.list is a list (i.e., conventional use) or nrow(data) * number of unique number suffixes in vrb.nm.list if vrb.nm.list is not a list (i.e., advanced use). The columns will be in the following order: 1) grp.nm of the groups, 2) rtn.obs.nm of the observation labels, 3) the reshaped columns, 4) the additional columns that were not reshaped and instead repeated. How the returned data.frame is sorted depends on order.by.grp.

See Also

long2wide reshape stack2

Examples

# SINGLE GROUPING VARIABLE
dat_wide <- data.frame(
   x_1.1 = runif(5L),
   x_2.1 = runif(5L),
   x_3.1 = runif(5L),
   x_4.1 = runif(5L),
   x_1.2 = runif(5L),
   x_2.2 = runif(5L),
   x_3.2 = runif(5L),
   x_4.2 = runif(5L),
   x_1.3 = runif(5L),
   x_2.3 = runif(5L),
   x_3.3 = runif(5L),
   x_4.3 = runif(5L),
   y_1.1 = runif(5L),
   y_2.1 = runif(5L),
   y_1.2 = runif(5L),
   y_2.2 = runif(5L),
   y_1.3 = runif(5L),
   y_2.3 = runif(5L))
row.names(dat_wide) <- letters[1:5]
print(dat_wide)

# vrb.nm.list = list of character vectors (conventional use)
vrb_pat <- c("x_1","x_2","x_3","x_4","y_1","y_2")
vrb_nm_list <- lapply(X = setNames(vrb_pat, nm = vrb_pat), FUN = function(pat) {
   str2str::pick(x = names(dat_wide), val = pat, pat = TRUE)})
# without `grp.nm`
z1 <- wide2long(dat_wide, vrb.nm = vrb_nm_list)
# with `grp.nm`
dat_wide$"ID" <- letters[1:5]
z2 <- wide2long(dat_wide, vrb.nm = vrb_nm_list, grp.nm = "ID")
dat_wide$"ID" <- NULL

# vrb.nm.list = character vector + guessing (advanced use)
vrb_nm <- str2str::pick(x = names(dat_wide), val = "ID", not = TRUE)
# without `grp.nm`
z3 <- wide2long(dat_wide, vrb.nm.list = vrb_nm)
# with `grp.nm`
dat_wide$"ID" <- letters[1:5]
z4 <- wide2long(dat_wide, vrb.nm = vrb_nm, grp.nm = "ID")
dat_wide$"ID" <- NULL

# comparisons
head(z1); head(z3); head(z2); head(z4)
all.equal(z1, z3)
all.equal(z2, z4)
# keeping the reshapeLong attributes
z7 <- wide2long(dat_wide, vrb.nm = vrb_nm_list, keep.attr = TRUE)
attributes(z7)

# MULTIPLE GROUPING VARIABLES
bfi2 <- psych::bfi
bfi2$"person" <- unlist(lapply(X = 1:400, FUN = rep.int, times = 7))
bfi2$"day" <- rep.int(1:7, times = 400L)
head(bfi2, n = 15)

# vrb.nm.list = list of character vectors (conventional use)
vrb_pat <- c("A","C","E","N","O")
vrb_nm_list <- lapply(X = setNames(vrb_pat, nm = vrb_pat), FUN = function(pat) {
   str2str::pick(x = names(bfi2), val = pat, pat = TRUE)})
z5 <- wide2long(bfi2, vrb.nm.list = vrb_nm_list, grp = c("person","day"),
   rtn.obs.nm = "item")

# vrb.nm.list = character vector + guessing (advanced use)
vrb_nm <- str2str::pick(x = names(bfi2),
   val = c("person","day","gender","education","age"), not = TRUE)
z6 <- wide2long(bfi2, vrb.nm.list = vrb_nm, grp = c("person","day"),
   sep = "", rtn.obs.nm = "item") # need sep = "" because no character separating
   # scale name and item number
all.equal(z5, z6)