statar

R-CMD-check CRAN_Status_Badge codecov

This package contains R functions corresponding to useful Stata commands.

The package includes: - panel data functions (monthly/quarterly dates, lead/lag, fillin) - data.frame functions (tabulate, merge) - vector functions (xtile, pctile, winsorize) - graph functions (binscatter)

Data Frame Functions

sum_up = summarize

sum_up prints detailed summary statistics (corresponds to Stata summarize)

N <- 100
df <- tibble(
  id = 1:N,
  v1 = sample(5, N, TRUE),
  v2 = sample(1e6, N, TRUE)
)
sum_up(df)
df %>% sum_up(starts_with("v"), d = TRUE)
df %>% group_by(v1) %>%  sum_up()

tab = tabulate

tab prints distinct rows with their count. Compared to the dplyr function count, this command adds frequency, percent, and cumulative percent.

N <- 1e2 ; K = 10
df <- tibble(
  id = sample(c(NA,1:5), N/K, TRUE),
  v1 = sample(1:5, N/K, TRUE)       
)
tab(df, id)
tab(df, id, na.rm = TRUE)
tab(df, id, v1)

join = merge

join is a wrapper for dplyr merge functionalities, with two added functions

Vector Functions

# pctile computes quantile and weighted quantile of type 2 (similarly to Stata _pctile)
v <- c(NA, 1:10)                   
pctile(v, probs = c(0.3, 0.7), na.rm = TRUE) 

# xtile creates integer variable for quantile categories (corresponds to Stata xtile)
v <- c(NA, 1:10)                   
xtile(v, n_quantiles = 3) # 3 groups based on terciles
xtile(v, probs = c(0.3, 0.7)) # 3 groups based on two quantiles
xtile(v, cutpoints = c(2, 3)) # 3 groups based on two cutpoints

# winsorize (default based on 5 x interquartile range)
v <- c(1:4, 99)
winsorize(v)
winsorize(v, replace = NA)
winsorize(v, probs = c(0.01, 0.99))
winsorize(v, cutpoints = c(1, 50))

Panel Data Functions

Elapsed dates

The classes “monthly” and “quarterly” print as dates and are compatible with usual time extraction (ie month, year, etc). Yet, they are stored as integers representing the number of elapsed periods since 1970/01/0 (resp in week, months, quarters). This is particularly handy for simple algebra:

 # elapsed dates
 library(lubridate)
 date <- mdy(c("04/03/1992", "01/04/1992", "03/15/1992"))  
 datem <- as.monthly(date)
 # displays as a period
 datem
 #> [1] "1992m04" "1992m01" "1992m03"
 # behaves as an integer for numerical operations:
 datem + 1
 #> [1] "1992m05" "1992m02" "1992m04"
 # behaves as a date for period extractions:
 year(datem)
 #> [1] 1992 1992 1992

lag / lead

tlag/tlead a vector with respect to a number of periods, not with respect to the number of rows

year <- c(1989, 1991, 1992)
value <- c(4.1, 4.5, 3.3)
tlag(value, 1, time = year)
library(lubridate)
date <- mdy(c("01/04/1992", "03/15/1992", "04/03/1992"))
datem <- as.monthly(date)
value <- c(4.1, 4.5, 3.3)
tlag(value, time = datem) 

In constrast to comparable functions in zoo and xts, these functions can be applied to any vector and be used within a dplyr chain:

df <- tibble(
    id    = c(1, 1, 1, 2, 2),
    year  = c(1989, 1991, 1992, 1991, 1992),
    value = c(4.1, 4.5, 3.3, 3.2, 5.2)
)
df %>% group_by(id) %>% mutate(value_l = tlag(value, time = year))

is.panel

is.panel checks whether a dataset is a panel i.e. the time variable is never missing and the combinations (id, time) are unique.

df <- tibble(
    id1    = c(1, 1, 1, 2, 2),
    id2   = 1:5,
    year  = c(1991, 1993, NA, 1992, 1992),
    value = c(4.1, 4.5, 3.3, 3.2, 5.2)
)
df %>% group_by(id1) %>% is.panel(year)
df1 <- df %>% filter(!is.na(year))
df1 %>% is.panel(year)
df1 %>% group_by(id1) %>% is.panel(year)
df1 %>% group_by(id1, id2) %>% is.panel(year)

fill_gap

fill_gap transforms a unbalanced panel into a balanced panel. It corresponds to the stata command tsfill. Missing observations are added as rows with missing values.

df <- tibble(
    id    = c(1, 1, 1, 2),
    datem  = as.monthly(mdy(c("04/03/1992", "01/04/1992", "03/15/1992", "05/11/1992"))),
    value = c(4.1, 4.5, 3.3, 3.2)
)
df %>% group_by(id) %>% fill_gap(datem)
df %>% group_by(id) %>% fill_gap(datem, full = TRUE)
df %>% group_by(id) %>% fill_gap(datem, roll = "nearest")

Graph Functions

stat_binmean

stat_binmean() (a stat for ggplot2) returns the mean of y and x within 20 bins of x. It’s a barebone version of the Stata command binscatter

ggplot(iris, aes(x = Sepal.Width , y = Sepal.Length)) + stat_binmean()
# change number of bins
ggplot(iris, aes(x = Sepal.Width , y = Sepal.Length, color = Species)) + stat_binmean(n = 10) 
# add regression line
ggplot(iris, aes(x = Sepal.Width , y = Sepal.Length, color = Species)) + stat_binmean() + stat_smooth(method = "lm", se = FALSE)

Installation

You can install