dgraph

A Stata package for t-test graphs - Update: tables with significance levels in Stata console and LaTeX

v. 2.3 - Minor bugs fixed

This package generates a stacked coefficient plot for two-tailed t-tests. It is very useful for showcasing, through a very simple line of code, a graphical representation of covariates imbalance between groups. Several dependent variables can be tested in a single run. Results can be stored in a LaTeX table.

On 30th March 2023, dgraph was uploaded to the Statistical Software Components (SSC) archive. Details about the functionalities of this package can be accessed by typing on the Stata prompt (after the installation):

help dgraph

Setup

To install the package, copy and paste in your Stata prompt this line (install from SSC):

ssc install dgraph, replace

or this line (install from GitHub):

net install dgraph, from("https://raw.githubusercontent.com/DiegoCiccia/dgraph/main") replace

Syntax

syntax varlist(min=1) [if] [in], by(string)  
  [ 
    label long reverse echo suppress ci(num) 
    title(string) subtitle(string) 
    lc(string) lp(string) lw(num) 
    mc(string) msize(string) 
    sl(num) sw(num)
    seplc(string) seplp(string) seplw(num) 
    labsize(num) labangle(num) 
    scheme(string) ysize(num) xsize(num) 
    saving(string) replace tabsaving(string)
  ]

Baseline Options

varlist: (required) outcome variables. Only numeric variables can be included in varlist.
by(): (required) dummy variable defining two estimation groups. It can be numeric or string, but it must take only two distinct values.
label: use variable labels in place of variable names.
long: stack confidence intervals horizontally (starting from the left side);. By default, confidence intervals are stacked vertically (starting from the bottom).
reverse: shown point estimates as the average y(1) - y(0). As in the Stata ttest command, by default y(0) - y(1) is shown.
echo: print in the Stata console a table with the numeric values of the average difference between groups and upper/lower bounds for the estimate for each of the variables in varlist.
suppress: (often used in combination with echo and tabsaving) the graph is not produced, while values can be still displayed in console or printed in file.
ci(): level of confidence. By default, ci(95) is specified.

Graph Options

title(), subtitle(): specify title and subtitle of the graph. By default, no title nor subtitle.
lc(), lw(), lp(): change the color, width and patters of the lines in confidence intervals.
mc(), ms(): change color and size of scatter points for point estimates and upper/lower bounds.
sl(), sw(): change length and width of segments at the edges of confidence intervals.
seplc(), seplp(), seplw(): change the color, width and pattern of the separation line, by default at 0.
labsize(), labangle(): change size and angle orientation of variable labels on the graph.
scheme(): change the graph scheme.
ysize(), xsize(): change the axes sizes.
saving(): save the graph as a .gph file.
replace: replaces previous savings.
tabsaving(): saves the echo output as a TeX tabular (automatic replace).

Example

clear
set seed 0
set obs 20000
forv i = 1/30 {
    gen var_`i' = rnormal()
    label var var_`i' "Dep Var `i'"
}
gen D = runiform() > 0.5
tostring D, replace

The previous block of code generates a dataset of 20,000 observations with 30 random variables (from a normal distribution). The group variable (D) is drawn from a random uniform distribution and indicates values above 0.5. To showcase that the command works for any by() variable as long as only two values are in its support, the variable D is reformatted as a string.

dgraph var_*, by(D) long labangle(45) label scheme(white_tableau) title("Graph") reverse mc(black) msize(1) lw(0.2) ci(90) labsize(vsmall) saving(gr_sample) replace echo tabsaving(table)

Output:

Figure

Console

TeX file

This is a beta version. Please report bugs in the Issues section. For any other information, contact me by e-mail: cicciadiego99@gmail.com