Hello Everyone,
I am making a visualization for treatment effects that looks something like this:
The code I used to make this is as follows (please note that I am using the Auto dataset from within Stata for this):
Probably not the most efficient code i must say lol but does the job!
Coming to my question. There are text boxes generated (highlighted in red within the code block) at the bottom of the visualization that mark 'baseline' 'intervention' and 'endline' that neatly encapsulate the text with adequate margins. An issue i faced with visualization is that if the minimum value starts getting close to 0, the text box size starts to vertically grow too large.
Is there a way of fixing the text box size in a manner that it maintains its look as seen above?
The text box below is an example of how it looks like if lowest point gets close to 0:
I hope the example has adequate information. Please let me know if clarity is required.
Also, if there is any way to improve on the code, then that help will be very much appreciated!
I am making a visualization for treatment effects that looks something like this:
The code I used to make this is as follows (please note that I am using the Auto dataset from within Stata for this):
Code:
* Generating Sample Data
* Requires a baseline variable, and a treatment variable (one group)
sysuse auto, clear
generate treat = 0 if _n <= _N/5
replace treat = 1 if missing(treat)
tab treat, gen(treatment)
replace treatment1 = . if treatment1 != 1
* Generating sample baseline price
generate price1 = price - 3000, after(price)
* Note:
* NO need to edit from this point onwards unless you want to modify the look
* of the visual and possibly change the way data is displayed or fix errors
********************************************************************************
* Regression with first treatment arm
quietly regress price treatment2 trunk, robust
matrix A1 = r(table)'
* Control mean figures
matrix define C1a = J(1, 9, .)
ci means price if treatment2 == 0
matrix C1a[1,1] = r(mean)
matrix C1a[1,5] = r(lb)
matrix C1a[1,6] = r(ub)
matrix colnames C1a = b e t pvalue ll ul df crit eform
matrix rownames C1a = control
matrix define C1b = J(1, 9, .)
ci means price1 if treatment2 == 0
matrix C1b[1,1] = r(mean)
matrix C1b[1,5] = r(lb)
matrix C1b[1,6] = r(ub)
matrix colnames C1b = b e t pvalue ll ul df crit eform
matrix rownames C1b = controlbase
* Append Regression results with control mean
matrix G1 = A1\C1a\C1b
* Generating column vector of 1 to denote treatment arm 1
local rows1 : rowsof G1
matrix T1 = J(`rows1', 1, 1)
matrix colnames T1 = treatarm
* Combining columnwise the regression results and treatment arm
matrix combined = G1,T1
matrix list combined
********************************************************************************
* Saving matrix as a dataset
* get original row names of matrix (and row count)
local rownames : rowfullnames combined
local c : word count `rownames'
* get original column names of matrix and substitute out _cons
local names : colfullnames combined
local newnames : subinstr local names "_cons" "cons", word
* rename columns of matrix
matrix colnames combined = `newnames'
* convert to dataset
clear
svmat combined, names(col)
* add matrix row names to dataset
gen rownames = ""
forvalues i = 1/`c' {
replace rownames = "`:word `i' of `rownames''" in `i'
}
order rownames
********************************************************************************
* Plotting
generate treatment = treatarm
replace treatment = 1 if treatment == 1 & rownames == "controlbase"
replace treatment = 5 if treatment == 1 & rownames == "control"
replace treatment = 5 if treatment == 1 & rownames == "treatment2"
* Generate full coefficient
sort treatarm, stable
by treatarm: generate srno = _n
reshape wide rownames b se t pvalue ll ul df crit eform treatment, i(srno) j(treatarm)
generate control1 = b1 if rownames1 == "control", after(b1)
ereplace control1 = max(control1)
replace b1 = b1 + control1 if rownames1 == "treatment2"
replace ll1 = ll1 + control1 if rownames1 == "treatment2"
replace ul1 = ul1 + control1 if rownames1 == "treatment2"
* Intercept point calculation
* Since our intervention point is X=3 on the horizontal axis, we need Y value
* We will use Y = MX + C
* First we need slope: M = (y2 - y1)/(x2 - x1)
local X = 3
local bl : display `=_N'
local ctrl : display (`=_N' - 1)
local M = (b1[`ctrl'] - b1[`bl']) / (treatment1[`ctrl'] - treatment1[`bl'])
local C = b1[`ctrl'] - (`M' * treatment1[`ctrl'])
local Y = (`M' * `X') + `C'
local newobs : display `=_N' + 1
set obs `newobs'
replace treatment1 = `X' if _n == _N
replace b1 = `Y' if _n == _N
replace rownames1 = "intercept" if _n == _N
* Plot
if (b1[`ctrl'] > b1[`bl']) & (b1[`ctrl'] > b1[1]) {
local min : display b1[`bl'] - abs(b1[`bl']*0.25)
local max : display b1[`ctrl'] + abs(b1[`ctrl']*0.25)
local textpos: display `min' + abs(b1[`bl']*(1/8))
}
else if (b1[`ctrl'] > b1[`bl']) & (b1[`ctrl'] < b1[1]) {
local min : display b1[`bl'] - abs(b1[`bl']*0.25)
local max : display b1[`ctrl'] + abs(b1[`ctrl']*0.25)
local textpos: display `min' + abs(b1[`bl']*(1/8))
}
else if (b1[`ctrl'] < b1[`bl']) & (b1[`ctrl'] > b1[1]) {
local min : display b1[1] - abs(b1[1]*0.25)
local max : display b1[`bl'] + abs(b1[`bl']*0.25)
local textpos: display `min' + abs(b1[1]*(1/8))
}
else {
local min : display b1[`ctrl'] - abs(b1[`ctrl']*0.25)
local max : display b1[`bl'] + abs(b1[`bl']*0.25)
local textpos: display `min' + abs(b1[`ctrl']*(1/8))
}
local trt : display b1[1]
local ctr : display b1[`ctrl']
local teff : display round(abs(`ctr' - `trt'), .01)
if `trt' > `ctr' {
local val : display `trt' - abs(`teff'/2)
}
else {
local val : display `trt' + abs(`teff'/2)
}
* Text Box values
local topb : display `textpos' + abs(`textpos'*0.05)
local botb : display `textpos' - abs(`textpos'*0.05)
#delimit ;
twoway (connected b1 treatment1 if inlist(rownames1, "treatment2", "intercept"), ms(i))
(connected b1 treatment1 if inlist(rownames1, "control", "controlbase"), ms(i))
(scatteri `trt' 5.1 `ctr' 5.1, recast(line) lcolor(gs4))
(scatteri `ctr' 5.05 `ctr' 5.1, recast(line) lcolor(gs4))
(scatteri `trt' 5.05 `trt' 5.1, recast(line) lcolor(gs4))
(scatteri `min' 1 `max' 1, recast(line) lcolor(dkgreen%40))
(scatteri `min' 3 `max' 3, recast(line) lcolor(gs3%50) lpattern(dash))
(scatteri `min' 5 `max' 5, recast(line) lcolor(red%40))
(scatteri `val' 5.35 "`teff'", ms(i) mlabpos(0))
(scatteri `botb' 0.7 `botb' 1.3 `topb' 1.3 `topb' 0.7, recast(area) color(white) lwidth(0.2) lcolor(dkgreen%40))
(scatteri `textpos' 1 "Baseline", ms(i) mlabpos(0) mlabsize(2.25))
(scatteri `botb' 2.7 `botb' 3.3 `topb' 3.3 `topb' 2.7, recast(area) color(white) lwidth(0.2) lcolor(gs3%50))
(scatteri `textpos' 3 "Intervention", ms(i) mlabpos(0) mlabsize(2.25))
(scatteri `botb' 4.7 `botb' 5.3 `topb' 5.3 `topb' 4.7, recast(area) color(white) lwidth(0.2) lcolor(red%40))
(scatteri `textpos' 5 "Endline", ms(i) mlabpos(0) mlabsize(2.25))
,
title("{bf} Treatment Effect Visualization", pos(11) margin(b+3 t=-1) size(*.6))
subtitle("Single group automation of graph using sysuse auto", pos(11) margin(b+6 t=-3 l=1.75) size(*.6))
xlabel(none)
xtitle("")
ylabel(none)
ytitle("")
yscale(range(`min' `max') off)
legend(order(2 1) label(1 "Treatment") label(2 "Control") size(2.25))
scheme(white_tableau)
;
#delimit cr
Coming to my question. There are text boxes generated (highlighted in red within the code block) at the bottom of the visualization that mark 'baseline' 'intervention' and 'endline' that neatly encapsulate the text with adequate margins. An issue i faced with visualization is that if the minimum value starts getting close to 0, the text box size starts to vertically grow too large.
Is there a way of fixing the text box size in a manner that it maintains its look as seen above?
The text box below is an example of how it looks like if lowest point gets close to 0:
I hope the example has adequate information. Please let me know if clarity is required.
Also, if there is any way to improve on the code, then that help will be very much appreciated!
Comment