build.panel {psidR}

build.panel: Build PSID panel data set

Description

Builds a panel data set with id variables pid (unique person identifier) and year from individual PSID family files and supplemental wealth files.

Usage

build.panel(
  datadir = NULL,
  fam.vars,
  ind.vars = NULL,
  heads.only = FALSE,
  current.heads.only = FALSE,
  sample = NULL,
  design = "balanced",
  loglevel = INFO
)

Arguments

datadir

either NULL, in which case saves to tmpdir or path to directory containing family files ("FAMyyyy.RData") and individual file ("IND2009ER.RData").

fam.vars

data.frame of variable to retrieve from family files. Can contain see example for required format.

ind.vars

data.frame of variables to get from individual file. In almost all cases this will be the type of survey weights you want to use. don't include id variables ER30001 and ER30002.

heads.only

logical TRUE if user wants household heads only. Household heads in sample year.

current.heads.only

logical TRUE if user wants current household heads only. Distinguishes mover outs heads.

sample

string indicating which sample to select: "SRC" (survey research center), "SEO" (survey for economic opportunity), "immigrant" (immigrant sample), "latino" (Latino family sample). Defaults to NULL, so no subsetting takes place.

design

either character balanced or all or integer. balanced means only individuals who appear in each wave are considered. All means all are taken. An integer value stands for minimum consecutive years of participation, i.e. design=3 means present in at least 3 consecutive waves.

loglevel

one of INFO, WARN and DEBUG. INFO by default.

Details

There are several supported approches. Approach one downloads stata data, uses stata to build each wave, then puts it together with 'psidR'. The second (recommended) approach downloads all data directly from the psid servers (no Stata needed). For this approach you need to supply the precise names of psid variables - those variable names vary by year. E.g. total family income will have different names in different waves. The function getNamesPSID greatly helps collecting names for all waves.

Value

resulting data.table. the variable pid is the unique person identifier, constructed from ID1968 and pernum

Merging

The variables interview number in each family file map to the interview number variable of a given year in the individual file. Run example(build.panel) for a demonstration.

Supplements

Notice that support for wealth supplements is disabled! Recent releases of the main family file have wealth data included. Earlier waves must be merged manually, again by variable interview number as above.

Examples

## Not run: 
# ################################################
# Real-world example: not run because takes long.
# Build panel with income, wage, age and education
# optionally: add wealth supplements!
# ################################################

# The package is installed with a list of variables
# Alternatively, search for names with \code{\link{getNamesPSID}}
# This is the body of function build.psid()
# (so why not call build.psid() and see what happens!)
r = system.file(package="psidR")
if (small){
  f = fread(file.path(r,"psid-lists","famvars-small.txt"))
  i = fread(file.path(r,"psid-lists","indvars-small.txt"))
} else {
  f = fread(file.path(r,"psid-lists","famvars.txt"))
  i = fread(file.path(r,"psid-lists","indvars.txt"))
}
setkey(i,"name")
setkey(f,"name")
i = dcast(i[,list(year,name,variable)],year~name)
f = dcast(f[,list(year,name,variable)],year~name)
  d = build.panel(datadir="~/datasets/psid/",fam.vars=f,
                 ind.vars=i, 
                 heads.only =TRUE,sample="SRC",
                 design="all")
  save(d,file="~/psid.RData")

## End(Not run)

# ######################################
# reproducible example on artifical data. 
# run this with example(build.panel).
# ######################################

## make reproducible family data sets for 2 years
## variables are: family income (Money) and age

## Data acquisition step:
## run build.panel with sascii=TRUE

# testPSID creates artifical PSID data
td <- testPSID(N=12,N.attr=0)
fam1985 <- data.table::copy(td$famvars1985)
fam1986 <- data.table::copy(td$famvars1986)
IND2019ER <- data.table::copy(td$IND2019ER)

# create a temporary datadir
my.dir <- tempdir()
#save those in the datadir
# notice different R formats admissible
save(fam1985,file=paste0(my.dir,"/FAM1985ER.rda"))
save(fam1986,file=paste0(my.dir,"/FAM1986ER.RData"))
save(IND2019ER,file=paste0(my.dir,"/IND2019ER.RData"))

## end Data acquisition step.

# now define which famvars
famvars <- data.frame(year=c(1985,1986),
                      money=c("Money85","Money86"),
                      age=c("age85","age86"))

# create ind.vars
indvars <- data.frame(year=c(1985,1986),ind.weight=c("ER30497","ER30534"))

# call the builder
# data will contain column "relation.head" holding the relationship code.

d <- build.panel(datadir=my.dir,fam.vars=famvars,
                 ind.vars=indvars,
                 heads.only=FALSE)	

# see what happens if we drop non-heads
# only the ones who are heads in BOTH years 
# are present (since design='balanced' by default)
d <- build.panel(datadir=my.dir,fam.vars=famvars,
                 ind.vars=indvars,
                 heads.only=TRUE)	
print(d[order(pid)],nrow=Inf)

# change sample design to "all": 
# we'll keep individuals if they are head in one year,
# and drop in the other
d <- build.panel(datadir=my.dir,fam.vars=famvars,
                 ind.vars=indvars,heads.only=TRUE,
                 design="all")	
print(d[order(pid)],nrow=Inf)

file.remove(paste0(my.dir,"/FAM1985ER.rda"),
            paste0(my.dir,"/FAM1986ER.RData"),
            paste0(my.dir,"/IND2019ER.RData"))

# END psidR example

# #####################################################################
# Please go to https://github.com/floswald/psidR for more example usage
# #####################################################################