Part 8: Regressions

There are all sorts of models out there for performing regressions. This page focuses on 3 that are used for binary outcomes:

  1. Logistic regression
  2. Modified Poisson regression
  3. Cox proportional hazards models.

Getting set up

Before you get going, you want to explicitly define your outcome of interest (aka dependent variable), primary exposure (aka independent variable) and covariates that you are adjusting for in your model (aka other independent variables). You’ll also want to know your study design (eg case-control? cross-sectional? prospective? time to event?).

Are your independent variables continuous or factor/categorical?

In these regression models, you can specify that independent variables (primary exposure and covariates that are in your model) are continuous or factor/categorical. Continuous variables can be specified with a leading “c.”, and examples might include age, height, or length. (“c.age”.) In contrast, factor/categorical variables might be New England states (e.g., 1=Connecticut, 2=Maine, 3=Mass, 4=New Hampshire, 5=RI, and 6=Vermont). So, you’d want to specifiy a variable for “newengland” as i.newengland in your regression. Stata defaults to treating the smallest number (here, 1=Connecticut) as the reference group. You can change that by using the “ib#.” prefix instead, where the # is the reference group, or here ib2.newengland to have Maine as the reference group.

Read about factor variables in –help fvvarlist–.

What about binary/dichotomous variables (things that are ==0 or 1)? Well, it doesn’t change your analysis if you specify a binary/dichotomous as either continuous (“c.”) or factor/categorical (“i.”). The math is the same on the back end.

Checking for interactions

In general, when checking for an interaction, you will need to specify if the two variables of interest are categorical or factor/categorical and drop two pound signs in between the two. See details on –help fvvarlist–. Here’s an example of how this would look, looking for an interaction between sex on age groups.

regress bp i.sex##c.agegrp

      Source |       SS           df       MS      Number of obs   =       240
-------------+----------------------------------   F(3, 236)       =     23.86
       Model |   9519.9625         3  3173.32083   Prob > F        =    0.0000
    Residual |  31392.8333       236   133.02048   R-squared       =    0.2327
-------------+----------------------------------   Adj R-squared   =    0.2229
       Total |  40912.7958       239  171.183246   Root MSE        =    11.533

------------------------------------------------------------------------------
          bp | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         sex |
     Female  |     -4.275   3.939423    -1.09   0.279    -12.03593    3.485927
      agegrp |     7.0625   1.289479     5.48   0.000      4.52214     9.60286
             |
sex#c.agegrp |
     Female  |      -1.35   1.823599    -0.74   0.460    -4.942611    2.242611
             |
       _cons |   143.2667   2.785593    51.43   0.000     137.7789    148.7545
------------------------------------------------------------------------------

You’ll see the sex#c.agegrp P-value is 0.460, so that wouldn’t qualify as a statistically significant interaction.

Regressions using survey data?

If you are using survey analyses (eg need to account for pweighting), generally you have to use the svy: prefixes for your analyses. This includes svy: logistic, poisson, etc. Type –help svy_estimation– to see what the options are.

Logistic regression

Logistic regressions provide odds ratios of binary outcomes. Odds ratios don’t approximate the risk ratio if the outcome is common (eg >10% occurrence) so I tend to avoid them as I study hypertension, which occurs commonly in a population.

There are oodles of details on logistic regression in the excellent UCLA website. In brief, you want want to use a regression command, you can use “logit” and get the raw betas or “logistic” and get the odds ratios. Most people will want to use “logistic” to get the odds ratios.

Here’s an example of logistic regression in Stata. In this dataset, “race” is Black/White/other, so you’ll need to specify this as a factor/categorical variable with the “i.” prefix, however “smoke” is binary so you can either specify it as a continuous or as a factor/categorical variable. If you don’t specify anything, then it is treated as continuous, which is fine.


webuse lbw, clear
codebook race // see race is 1=White, 2=Black, 3=other
codebook smoke // smoke is 0/1
logistic low i.race smoke 

// output: 
Logistic regression                                     Number of obs =    189
                                                        LR chi2(3)    =  14.70
                                                        Prob > chi2   = 0.0021
Log likelihood = -109.98736                             Pseudo R2     = 0.0626

------------------------------------------------------------------------------
         low | Odds ratio   Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
        race |
      Black  |   2.956742   1.448759     2.21   0.027     1.131716    7.724838
      Other  |   3.030001   1.212927     2.77   0.006     1.382616     6.64024
             |
       smoke |   3.052631   1.127112     3.02   0.003     1.480432    6.294487
       _cons |   .1587319   .0560108    -5.22   0.000     .0794888    .3169732
------------------------------------------------------------------------------
Note: _cons estimates baseline odds.

So, here you see that the odds ratio (OR) for the outcome of “low” is 2.96 (95% CI 1.13, 7.72) for Black relative to White participants and the OR for the outcome of “low” is 3.03 (95% CI 1.38, 6.64) for Other race relative to White participants. Since it’s the reference group, you’ll notice that White race isn’t shown. But what if we want Black race as the reference? You’d use “ib2.race” instead of “i.race”. Example:

logistic low ib2.race smoke 
// output:

Logistic regression                                     Number of obs =    189
                                                        LR chi2(3)    =  14.70
                                                        Prob > chi2   = 0.0021
Log likelihood = -109.98736                             Pseudo R2     = 0.0626

------------------------------------------------------------------------------
         low | Odds ratio   Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
        race |
      White  |   .3382101   .1657178    -2.21   0.027     .1294525    .8836136
      Other  |   1.024777   .5049663     0.05   0.960     .3901157    2.691938
             |
       smoke |   3.052631   1.127112     3.02   0.003     1.480432    6.294487
       _cons |   .4693292   .2059043    -1.72   0.085     .1986269    1.108963
------------------------------------------------------------------------------
Note: _cons estimates baseline odds.


Modified Poisson regression

Modified Poisson regression (sometimes called Poisson Regression with Robust Variance Estimation or Poisson Regression with Sandwich Variance Estimation) is pretty straightforward. Note: this is different from ‘conventional’ Poisson regression, which is used for counts and not dichotomous outcomes. You use the “poisson” command with subcommands “, vce(robust) irr” to use the modified Poisson regression type. Note: with svy data, robust VCE is the default so you just need to use the subcommand “, irr”.

As with the logistic regression section above, race is a 3-level nominal variable so we’ll use the “i.” or “ib#.” prefix to specify that it’s not to be treated as a continuous variable.

webuse lbw, clear
codebook race // see race is 1=White, 2=Black, 3=other
codebook smoke // smoke is 0/1
// set Black race as the reference group
poisson low ib2.race smoke, vce(robust) irr
// output:

Iteration 0:  Log pseudolikelihood = -122.83059  
Iteration 1:  Log pseudolikelihood = -122.83058  

Poisson regression                                      Number of obs =    189
                                                        Wald chi2(3)  =  19.47
                                                        Prob > chi2   = 0.0002
Log pseudolikelihood = -122.83058                       Pseudo R2     = 0.0380

------------------------------------------------------------------------------
             |               Robust
         low |        IRR   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
        race |
      White  |    .507831    .141131    -2.44   0.015     .2945523    .8755401
      Other  |   1.038362   .2907804     0.13   0.893     .5997635      1.7977
             |
       smoke |   2.020686   .4300124     3.31   0.001     1.331554    3.066471
       _cons |   .3038099   .0800721    -4.52   0.000     .1812422    .5092656
------------------------------------------------------------------------------
Note: _cons estimates baseline incidence rate.

Here you see that the relative risk (RR) of “low” for White relative to Black participants is 0.51 (95% CI 0.29, 0.88), and the RR for other race relative to Black participants is 1.04 (95% CI 0.60, 1.80). As above, the “Black” group isn’t shown since it’s the reference group.

Cox proportional hazards model

Cox PH models are used for time to event data. This is a 2-part. First is to –stset– the data, thereby telling Stata that it’s time to event data. In the –stset– command, you specify the outcome/failure variable. Second is to use the –stcox– command to specify the primary exposure/covariates of interest (aka independent variables). There are lots of different steps to setting up the stset command, so make sure to check the –help stset– page. Ditto –help stcox–.

Here, you specify the days of follow-up as “days_hfhosp” and failure as “hfhosp”. Since follow-up is in days, you set the scale to 365.25 to have it instead considered as years.

stset days_hfhosp, f(hfhosp) scale(365.25)

Then you’ll want to use the stcox command to estimate the risk of hospitalization by beta blocker use by age, sex, and race with race relative to the group that is equal to 1.

stcox betablocker age sex ib1.race

You’ll want to make a Kaplan Meier figure at the same time, read about how to do that on this other post.

Printing hazard ratio on Kaplan Meier curve in Stata

I recently made a figure that estimates a hazard ratio and renders it right on top of a Kaplan Meier curve in Stata. Here’s some example code to make this.

Good luck!


// Load example dataset. I got this from the --help stset-- file
webuse diet, clear

// First, stset the data. 
stset dox /// dox is the event or censor date
, ///
failure(fail) /// "fail" is the failure vs censor variable
scale(365.25)


// Next, estimate a cox ph model by "hienergy"
stcox hienergy
// now grab the bits from output of this
local hrb=r(table)[1,1]
local hrlo=r(table)[5,1]
local hrhi=r(table)[6,1]
local pval = r(table)[4,1]
// now format the p-value so it's pretty
if `pval'>=0.056 {
	local pvalue "P=`: display %3.2f `pval''"
}
if `pval'>=0.044 & `pval'<0.056 {
	local pvalue "P=`: display %5.4f `pval''"
}
if `pval' <0.044 {
	local pvalue "P=`: display %4.3f `pval''"
}
if `pval' <0.001 {
	local pvalue "P<0.001"
}
if `pval' <0.0001 {
	local pvalue "P<0.0001"
}

di "original P is " `pval' ", formatted is " "`pvalue'"
di "HR " %4.2f `hrb' " (95% CI " %4.2f `hrlo' "-" %4.2f `hrhi' "; `pvalue')"

// Now make a km plot. this example uses CIs
sts graph ///
, ///
survival /// 
by(hienergy) ///
plot1opts(lpattern(dash) lcolor(red)) /// options for line 1
plot2opts(lpattern(solid) lcolor(blue)) /// options for line 2
ci /// add CIs
ci1opts(color(red%20)) /// options for CI 1
ci2opts(color(blue%20)) /// options for CI 2
/// Following this is the legend, placed in the 6 O'clock position. 
/// Only graphics 5 and 6 are needed, but all 6 are shown so you 
/// see that other bits that can show up in the legend. Delete 
/// everything except for 5 and 6 to hide the rest of the legend components
legend(order(1 "[one]" 2 "[two]" 3 "[three]" 4 "[four]" 5 "First group" 6 "Second group") position(6)) ///
/// Risk table to print at the bottom:
risktable(0(5)20 , size(small) order(1 "First group" 2 "Second group")) ///
title("Title") ///
t1title("Subtitle") ///
xtitle("Year of Follow-Up") ///
ytitle("Event-Free Survival") ///
/// Here's how you render the HR. Change the first 2 numbers to move it:
text(0 0 "`: display "HR " %4.2f `hrb' " (95% CI " %4.2f `hrlo' "-" %4.2f `hrhi' "; `pvalue')"'", placement(e) size(medsmall)) ///
yla(0(0.2)1) 

Stata graph box boxplots with different colors for –over– groups

I recently had to make some boxplots with Stata’s –graph box– command. I find this to be challenging to use since it varies from syntax from the –twoway– command set that I use all the time. I was using the –over– subcommand x2 and wanted to change the colors of each box & dots by group from the first –over– subcommand. I found some helpful details on the Statalist Forum here and here. Here’s what I did to accomplish this, using the help from the Statalist forum.

Some tweaks here: I wanted to show rotate some labels 45 degrees with –angle– and I also aggressively labeled variables and their values so I didn’t need to manually relabel the figure (which is done with the –relabel– subcommand if needed). It takes an extra 30 seconds to label variables and values, and will save you lots of headbanging fiddling with the –relabel– command, so just label your variables and values from the start.

This example uses fake data. Code follows the picture. Good luck!

clear all
set obs 1000 // blank dataset with 1000 observations
set seed 8675309 // jenny seed
gen group = round(runiform()) // make group that is 0 or 1
gen time = round(3*runiform()) // make 4 times, 0 through 3
replace time = time+1 // now time is 1 through 4
gen tommy2tone = 100*rbeta(3,10) // fake skewed data

// now apply labels to variables.
// technically you only need to label the 3rd one
// of these since categorical variable value labels
// are shown instead of the variable label itself,
// but might as well do all 3 in case you need them
// labeled somewhere else.
label variable group "Group"
label variable time "Time"
label variable tommy2tone "Jenny Lyrics remembered, %"

// now make value labels.
* group
label define grouplab 0 "Tommy" 1 "Tutone"
label values group grouplab //DON'T FORGET TO APPLY LABELS

* time
label define timelab 1 "Time 1" 2 "Time 2" 3 "Time 3" 4 "Time 4"
label values time timelab //DON'T FORGET TO APPLY LABELS

// code for boxplot
graph box tommy2tone ///
, ///
over(group, label(angle(45))) ///
over(time) ///
scale(1.3) /// embiggen labels & figure components
box(1, color(red)) marker(1, mcolor(red)) ///
box(2, color(blue)) marker(2, mcolor(blue)) ///
asyvars showyvars leg(off)

Making a table in Stata for regression results (and other output) using frames

Frames were introduced in Stata 16 and are handy for (a) storing/manipulating multiple datasets simultaneously and (b) building datasets on the fly. I’ve had good luck making a table using frames. This strategy includes (1) making a new frame with as many columns as you need, specifying they are long strings (strL), and printing the top row, (2) running regressions or whatnot, (3) saving components of the regression as local macros that print as text, (4) making “display” macros for each column, (5) saving those “display” local macros to the new frame, repeating steps 2-5 as many times as needed, and (6) exporting the new frame as an Excel file.

Stata has recently introduced “tables” and “collect” features that allow you to do similar things. I find those features overly confusing, so I’ve developed this approach. I hope you find it useful! Here’s what we are trying to make:

Note: these details depend on use of concepts discussed on this other post (“return list” “ereturn list” “matrix list r(table)”, local macros, etc.). Make sure to familiarize yourself with that post.

Problems you might run into here:

  • This uses local macros, which disappear when a do file stops running. You need to run the entire do file from top to bottom every time, not line by line.
  • Stata’s -frames post- functionality is finicky. It expects exactly as many posted variables as there are variables in the new frame, and that each posted format matches the predefined format on the frame’s variable.
  • Three forward slashes (“///”) extends things across lines. Two forward slashes (“//”) allows commenting but does not continue to the next line. Check your double and triple forward slashes.
  • Stata can’t write to an open excel file. Close the excel file before re-running.

Code

* Reset frames and reload data
frames reset
sysuse auto, clear
*
* STEP 1 - Make a new frame with long strings
*
* Make new frame called "table" and 
* define all columns/variables as being strings.
* Name all variables as col#, starting with col1.
* You can extend this list as long as you'd like.
frame create table /// <--TRIPLE SLASH
strL col1 /// name 
strL col2  /// n
strL col3 /// beta
strL col4 /// 95% CI
strL col5 // <-- DOUBLE SLASH, P-value 
* If you want to add more col#s, make sure you change 
* the last double slashes to triple slashes, all new 
* col#s should have triple slashes except the last,
* which should have double (or no) slashes
*
* Prove to yourself that you have made a new frame
* called "table" in addition to the "default" one
* with the auto dataset. 
frames dir
* You could switch to the new table frame with 
* the "cwf table" command, if interested. To switch
* back, type "cwf default". 
*
* STEP 1B - print out your first row
* 
frame post table /// <--TRIPLE SLASH
("Variable name") /// col1
("N") /// col2
("Beta") /// col3
("95% CI") /// col4
("P-value") // <--DOUBLE SLASH col5 
*
* You can repeat this step as many times as you want 
* to add as many rows of custom text as needed. 
* Note that if you want a blank column, you need
* to still include the quotations within the parentheses.
* eg, if you didn't want the first column to have 
* "variable name", you'd put this instead:
* strL ("") /// col1
*
* If you wanted to flip over to the table frame and 
* see what's there, you'd type:
*cwf table 
*bro
*
* ...and to flip back to the default frame, type:
*cwf default
*
* STEP 2 - run your regression and look where the 
* coefficients of interest are stored 
* and
* STEP 3 - saving components of regression as local
* macros
*
regress price weight
*
ereturn list
* The N is here under e(N), save that as a local macro
local n1_1 = e(N)
*
matrix list r(table)
* The betas is at [1,1], the 95% thresholds
* are at [5,1] and [6,1], the se is
* at [2,1], and the p-value is at [4,1].
* We weren't planning on using se in this table,
* but we'll grab it anyway in case we change our minds
local beta1_1 = r(table)[1,1]
local ll1_1 = r(table)[5,1]
local ul1_1 = r(table)[6,1]
local se1_1 = r(table)[2,1]
local p1_1 = r(table)[4,1]
*
* STEP 4 - Making "display" macros
* 
* We are going to use local macros to grab things
* by column with a "display" commmand. Note that 
* column 1 is name, which we are going to call "all" here. 
* You could easily grab the variable name here with 
* the "local lab:" command detailed here:
* https://www.stata.com/statalist/archive/2002-11/msg00087.html
* or: 
*local label_name: variable label foreign 
* ...then call `label_name' instead of "all"
* 
* Now this is very important, we are not going to just
* capture these variables as local macros, we are going
* to capture a DISPLAY command followed by how we
* want the text to be rendered in the table. 
* Immediately after defining the local macro, we are going
* to call the macro so we can see what it will render 
* as in the table. This is what it will look like:
*local col1 di "All"
*`col1'
*
* IF YOU HAVE A BLANK CELL IN THIS ROW OF YOUR TABLE, 
* USE TWO EMPTY QUOTATION MARKS AFTER THE "di" COMMAND, eg: 
* local col1 di ""
* 
* now we will grab all pieces of the first row, 
* column-by-column, at the same time:
local col1 di "All" // name
`col1'
local col2 di `n1_1' // n
`col2'
local col3 di %4.2f `beta1_1' // betas
`col3'
local col4 di %4.2f `ll1_1' " to " %4.2f `ul1_1' // 95% ci
`col4'
local col5 di %4.3f `p1_1' // p-value
`col5' 
*
* note for the P-value, you can get much fancier in 
* formatting, see my script on how to do that here: 
* https://blog.uvm.edu/tbplante/2022/10/26/formatting-p-values-for-stata-output/
*
* STEP 5 - Posting the display macros to the table frame
* 
* Now post all columns row-by-row to the table frame
frame post table ///
("`: `col1''") ///
("`: `col2''") ///
("`: `col3''") ///
("`: `col4''") ///
("`: `col5''") // <-- DOUBLE SLASH
*
* Bonus! Insert a text row
*
frame post table /// <--TRIPLE SLASH
("By domestic vs. foreign status") /// col1
("") /// col2
("") /// col3
("") /// col4
("") // <--DOUBLE SLASH col5 
* Now repeat by foreign status
* domestic
regress price weight if foreign ==0 
ereturn list
local n1_1 = e(N)
*
matrix list r(table)
local beta1_1 = r(table)[1,1]
local ll1_1 = r(table)[5,1]
local ul1_1 = r(table)[6,1]
local se1_1 = r(table)[2,1]
local p1_1 = r(table)[4,1]
*
* note: you could automate col1 with the following command,
* which would grabe the label from the foreign==0 value of
* foreign:
*local label_name: label foreign 0
* ... then call `label_name' in the "local col1 di". 
local col1 di "  Domestic" // name, with 2 space indent
`col1'
local col2 di `n1_1' // n
`col2'
local col3 di %4.2f `beta1_1' // betas
`col3'
local col4 di %4.2f `ll1_1' " to " %4.2f `ul1_1' // 95% ci
`col4'
local col5 di %4.3f `p1_1' // p-value
`col5' 
*
frame post table ///
("`: `col1''") ///
("`: `col2''") ///
("`: `col3''") ///
("`: `col4''") ///
("`: `col5''") // <-- DOUBLE SLASH
*
* foreign
regress price weight if foreign ==1
ereturn list
local n1_1 = e(N)
*
matrix list r(table)
local beta1_1 = r(table)[1,1]
local ll1_1 = r(table)[5,1]
local ul1_1 = r(table)[6,1]
local se1_1 = r(table)[2,1]
local p1_1 = r(table)[4,1]
*
local col1 di "  Foreign" // name, with 2 space indent
`col1'
local col2 di `n1_1' // n
`col2'
local col3 di %4.2f `beta1_1' // betas
`col3'
local col4 di %4.2f `ll1_1' " to " %4.2f `ul1_1' // 95% ci
`col4'
local col5 di %4.3f `p1_1' // p-value
`col5' 
*
frame post table ///
("`: `col1''") ///
("`: `col2''") ///
("`: `col3''") ///
("`: `col4''") ///
("`: `col5''") // <-- DOUBLE SLASH
*
* STEP 6 - export the table frame as an excel file
* 
* This is the present working directory, where this excel
* file will be saved if you don't specify a directory:
pwd
*
* Switch to the table frame and take a look at it:
cwf table
bro 
* Now export it to excel
export excel using "table.xlsx", replace
*
* That's it!

Generating overlapping/overlaying decile frequency histograms in Stata

I recently had a dataset with two groups (0 or 1), and a continuous variable. I wanted to show how the overall deciles of that continuous variable varied by group. Step 1 was to generate an overall decile variable with an –xtile– command. Step 2 was to make a frequency histogram. BUT! I wanted these histograms to overlap and not be side-by-side. Stata’s handy –histogram– is a quick and easy way to make histograms by groups using the –by– command, but it makes them side-by-side like this, and not overlapping. (Note: see how to use –twoway histogram– to make overlapping histograms at the end of this post.)

I instead used a collapse command to generate a count of # in each decile by group (using the transparent color command as color percent sign number), like this:

Here’s the code to make both:

clear all

// make fake data
set obs 1000
set seed 8675309
gen id=_n // ID of 1 though 100
gen var0or1 = round(runiform())
gen continuousvalue = 100*runiform()

// make overall deciles of continuousvalue
xtile decilesbygroup = continuousvalue, nq(10)

// now make a frequency histogram of those deciles
set scheme s1color // I like this scheme
hist decilesbygroup, by(var0or1) frequency bin(10)

// make a variable equal to 1 that we will sum in collapse
gen countbygroup = 1
// now sum that variable by the 0 or 1 indicator and deciles
collapse (sum) countbygroup, by(var0or1 decilesbygroup)
// now render the count from above as a bar graph:
set scheme s1color // I like this scheme
twoway ///
(bar countbygroup decilesbygroup if var0or1==0, vertical color(red%40)) ///
(bar countbygroup decilesbygroup if var0or1==1, vertical color(blue%40)) ///
, ///
legend(order(1 "var0or1==0" 2 "var0or1==1")) ///
title("Title!") ///
xtitle("Decile of continuousvalue") ///
xla(1(1)10) ///
yla(0(10)70, angle(0)) ///
ytitle("N in Decile")

You could also offset the deciles by the var0or1 and shrink the bar width a bit to get a frequency histogram where the bars are next to each other, like this:

clear all

// make fake data
set obs 1000
set seed 8675309
gen id=_n // ID of 1 though 100
gen var0or1 = round(runiform())
gen continuousvalue = 100*runiform()

// make overall deciles of continuousvalue
xtile decilesbygroup = continuousvalue, nq(10)

// now make a frequency histogram of those deciles
set scheme s1color // I like this scheme
hist decilesbygroup, by(var0or1) frequency bin(10)

// offset the decilesbygroup by var0or1 a bit:
replace decilesbygroup = decilesbygroup - 0.2 if var0or1==0
replace decilesbygroup = decilesbygroup + 0.2 if var0or1==1

// make a variable equal to 1 that we will sum in collapse
gen countbygroup = 1
// now sum that variable by the 0 or 1 indicator and deciles
collapse (sum) countbygroup, by(var0or1 decilesbygroup)
// now render the count from above as a bar graph:
set scheme s1color // I like this scheme
twoway ///
(bar countbygroup decilesbygroup if var0or1==0, vertical color(red%40) barwidth(0.4)) ///
(bar countbygroup decilesbygroup if var0or1==1, vertical color(blue%40) barwidth(0.4)) ///
, ///
legend(order(1 "var0or1==0" 2 "var0or1==1")) ///
title("Title!") ///
xtitle("Decile of continuousvalue") ///
xla(1(1)10) ///
yla(0(10)70, angle(0)) ///
ytitle("N in Decile")

A few quick notes here: The way that I am specifying the “bins” for the histograms here is different than how Stata specifies bins for histograms, since I’m forcing it to render by decile. If you were to generate a histogram of the “continuousvalue” instead of the above example using “decilebygroup”, you’ll notice that the resulting histograms looks a bit different from each other:

clear all

// make fake data
set obs 1000
set seed 8675309
gen id=_n // ID of 1 though 100
gen var0or1 = round(runiform())
gen continuousvalue = 100*runiform()

// make overall deciles of continuousvalue
xtile decilesbygroup = continuousvalue, nq(10)

// now make a frequency histogram of those deciles
set scheme s1color // I like this scheme
hist decilesbygroup, title("hist decilesbygroup") by(var0or1) frequency bin(10) name(a)
hist continuousvalue, title("hist continuousvalue") by(var0or1)  frequency bin(10) name(b)

Also, this code will only render frequency histograms, not density histograms, which are the default in Stata. You can also use the –twoway hist– command to overlay two bar graphs, but these might not perfectly align with the deciles. But, using the –twoway hist– allows you to use density histograms instead. See the example that follows. I suspect that most people will get what they need with the –twoway hist– command in Stata.

clear all

// make fake data
set obs 1000
set seed 8675309
gen id=_n // ID of 1 though 100
gen var0or1 = round(runiform())
gen continuousvalue = 100*runiform()

set scheme s1color // I like this scheme
twoway ///
(hist continuousvalue if var0or1==0, bin(10) color(red%40) density) ///
(hist continuousvalue if var0or1==1, bin(10) color(blue%40) density) ///
, ///
legend(order(1 "var0or1==0" 2 "var0or1==1")) ///
title("Title!") ///
xtitle("Grouping in 10 Bins") 

Making Box Plots in Stata from scratch

Stata has some handy built-in features to make boxplots. If you are trying to make a typical boxplot in Stata, go read up on the –graph box– command as described on this post.

I was asked to abstract a boxplot from an old paper and re-render it in Stata.

I used the excellent WebPlotDigitizer to abstract the points in these figure. It took me a few rounds of data abstraction to get all of the data. I generated the following variables

  • (1) “row” – rows 1-8 — the “A-D” labels on the x axis are offset at 1.5, 3.5, 5.5, and 7.5, respectively,
  • (2) “group” – an indicator for group 1 or 2,
  • (3-5) “boxlow” “boxmid” and “boxhigh” – the lower, mid, and upper bounds of the box,
  • (6-7) “bar1” and “bar2” – the low and upper end of the bars, and
  • (8) “dot” – extreme values.

I used “input” at the top of a do file to load these data. The colors were taken from ColorBrewer2. The “box” are just really wide pcspike lines with overlaying horizontal bars as floating text boxes to indicate the bottom, median, and top of the box. One problem is the placement of the horizontal lines on and above/below the box itself, since the “―” ascii character isn’t perfectly in the middle of the vertical space that is used to render it. To get around this, I included an offset value that could be modified to shift these lines up and down so they lined up with the boxes. The “whiskers” are an rcap. The extreme values are just scatterplot dots.

Code to make this follows, data here are fake.

// abstracted with this: https://apps.automeris.io/wpd/
clear all

input row group boxlow boxmid boxhigh bar1 bar2 dot
1 1 110 130 140 100 150 .
2 2 115 125 135 90 150 .
3 1 135 145 155 120 175 . 
4 2 80 110 115 70 125 .
5 1 160 175 180 140 200 .
6 2 120 130 140 110 160 .
7 1 145 160 170 135 190 .
8 2 120 135 155 110 160 .
1 . . . . . . 95
1 . . . . . . 100
1 . . . . . . 155
1 . . . . . . 160
2 . . . . . . 80
2 . . . . . . 85
2 . . . . . . 155
2 . . . . . . 160
3 . . . . . . 110
3 . . . . . . 115
3 . . . . . . 180
3 . . . . . . 185
3 . . . . . . 190
4 . . . . . . 60
4 . . . . . . 65
4 . . . . . . 130
4 . . . . . . 135
5 . . . . . . 130
5 . . . . . . 135
5 . . . . . . 210
5 . . . . . . 140
6 . . . . . . 100
6 . . . . . . 105
6 . . . . . . 170
6 . . . . . . 175
7 . . . . . . 125
7 . . . . . . 130
7 . . . . . . 200
7 . . . . . . 210
8 . . . . . . 100
8 . . . . . . 105
8 . . . . . . 170
8 . . . . . . 175
end



// offset to move the bars on the box up or down on Y-axis,
// tweak as needed:
local offset = 2

set scheme s1mono

twoway ///
(rcap bar1 bar2 row, vert lcolor(black)) /// code for 95% CI
(pcspike boxlow row boxhigh row if group==1, vert lwidth(vvvthick) lcolor("141 211 199")) ///
(pcspike boxlow row boxhigh row if group==2, vert lwidth(vvvthick) lcolor("255 255 179")) ///
(scatter dot row , mcolor(black) msymbol(o) msize(tiny)) ///
, ///
text(`=boxmid[1]+`offset'' 1 "―", placement(c) size(medlarge) color(black)) ///
text(`=boxlow[1]+`offset'' 1 "{bf:―}", placement(c) size(medlarge) color(black)) ///
text(`=boxhigh[1]+`offset'' 1 "{bf:―}", placement(c) size(medlarge) color(black)) ///
///
text(`=boxmid[2]+`offset'' 2 "―", placement(c) size(medlarge) color(black)) ///
text(`=boxlow[2]+`offset'' 2 "{bf:―}", placement(c) size(medlarge) color(black)) ///
text(`=boxhigh[2]+`offset'' 2 "{bf:―}", placement(c) size(medlarge) color(black)) ///
///
text(`=boxmid[3]+`offset'' 3 "―", placement(c) size(medlarge) color(black)) ///
text(`=boxlow[3]+`offset'' 3 "{bf:―}", placement(c) size(medlarge) color(black)) ///
text(`=boxhigh[3]+`offset'' 3 "{bf:―}", placement(c) size(medlarge) color(black)) ///
///
text(`=boxmid[4]+`offset'' 4 "―", placement(c) size(medlarge) color(black)) ///
text(`=boxlow[4]+`offset'' 4 "{bf:―}", placement(c) size(medlarge) color(black)) ///
text(`=boxhigh[4]+`offset'' 4 "{bf:―}", placement(c) size(medlarge) color(black)) ///
///
text(`=boxmid[5]+`offset'' 5 "―", placement(c) size(medlarge) color(black)) ///
text(`=boxlow[5]+`offset'' 5 "{bf:―}", placement(c) size(medlarge) color(black)) ///
text(`=boxhigh[5]+`offset'' 5 "{bf:―}", placement(c) size(medlarge) color(black)) ///
///
text(`=boxmid[6]+`offset'' 6 "―", placement(c) size(medlarge) color(black)) ///
text(`=boxlow[6]+`offset'' 6 "{bf:―}", placement(c) size(medlarge) color(black)) ///
text(`=boxhigh[6]+`offset'' 6 "{bf:―}", placement(c) size(medlarge) color(black)) ///
///
text(`=boxmid[7]+`offset'' 7 "―", placement(c) size(medlarge) color(black)) ///
text(`=boxlow[7]+`offset'' 7 "{bf:―}", placement(c) size(medlarge) color(black)) ///
text(`=boxhigh[7]+`offset'' 7 "{bf:―}", placement(c) size(medlarge) color(black)) ///
///
text(`=boxmid[8]+`offset'' 8 "―", placement(c) size(medlarge) color(black)) ///
text(`=boxlow[8]+`offset'' 8 "{bf:―}", placement(c) size(medlarge) color(black)) ///
text(`=boxhigh[8]+`offset'' 8 "{bf:―}", placement(c) size(medlarge) color(black)) ///
///
legend(order(2 "Group 1" 3 "Group 2") row(2) size(small)) ///
scale(1.3) ///
///
xtitle("X Title") /// 
ytitle("Y Title") ///
///
title("Title") ///
xsize(5) ///
ysize(6) ///
yla(0(50)200, angle(0)) ///
xscale(r(.5 8.5)) ///
xla(1.5 "A" 3.5 "B" 5.5 "C" 7.5 "D") ///
xline(2.5 4.5 6.5)

Making a subgroup analysis figure in Stata

I was the analyst for the myPACE trial (published here in JAMA Cardiology), and needed to put together a subgroup analysis figure. I didn’t find any helpful stock code, so I wrote my own. The code uses the Frames feature that was introduced in Stata 16. It will (a) make a new frame with the required variables but no data, (b) generate a new dichotomous variable from continuous variables, (c) generate labels, (d) grab the Ns, point estimates, and 95% CI from a logistic regression, (e) grab the P-value for interaction for the primary exposure dependent variable*, (f) write the point estimate/95% CI/P-value for interaction to the new frame, then (g) switch to the new frame and make this figure. This script uses local macros so needs to be run all at once in a do file, not line by line.

This uses a stock Stata dataset called “catheter”. The outcome of interest/dependent variable is “infect”, the primary exposure/independent variable of interest is “time”, and the subgroups are age, sex, and patient number. This uses logistic regression, but you can easily swap this model out for any other model.

*You can get more complex code to format the P-values here.

Here’s the code!

frame reset // drop all existing frames and data
webuse catheter, clear // load analytical dataset
version 16 // need Stata version 16 or newer
*
* Make an empty frame with the variables we'll add later row by row
* The variables "rowname" and "pvalue" will be strings so when
* you add to these variables with the --frame post-- command,
* you need to use quotes. 
frame create subgroup str30 rowname n beta low95 high95 str30 pval
*
*** Age
* Need to generate a dichotomous age variable 
* You don't need to do this if the variable is already dichotomous, 
* ordinal, or nominal
generate agesplit = (age>=50) if !missing(age) // below 50 is 0, 50 and above is 1
* 
* Now generate a label for the overall grouping
local label_group "{bf:Age}" 
* 
* Group 0, below age 50
* Generate label for this subgroup
local label_subgroup_0 "Under 50y" 
* Now run the model for this subgroup
logistic infect time if agesplit==0 
* Now save the N, beta, and 95% CI as local macros.
* There's lots you can save after a regression, type --return list--, 
* --ereturn list--, and --matrix list r(table)-- to see what's there
local n_0 = e(N) 
local beta_0 = r(table)[1,1] 
local low95_0 = r(table)[5,1]
local high95_0 = r(table)[6,1]
* print above local macros to prove you collected them correctly: 
di "For `label_subgroup_0', n=" `n_0' ", beta (95% CI)=" %4.2f `beta_0' " (" %4.2f `low95_0' " to " %4.2f `high95_0' ")" 
*
* Group 1, at least 50 years old
local label_subgroup_1 "At least 50y" 
logistic infect time if agesplit==1 
local n_1 = e(N)
local beta_1 = r(table)[1,1] 
local low95_1 = r(table)[5,1]
local high95_1 = r(table)[6,1]
di "For `label_group', subgroup `label_subgroup_1', n=" `n_1' ", beta (95% CI)=" %4.2f `beta_1' " (" %4.2f `low95_1' " to " %4.2f `high95_1' ")" 
* 
* Now run the model with an interaction term between the primary exposure
* and the subgroup. 
logistic infect c.time##i.agesplit
* Grab the p-value as a local macro saved as pval
local pval = r(table)[4,5] 
* Print that local macro to see that you've grabbed it correctly
di "P-int = " %4.3f `pval'
* Format that P-value and save that as a new local macro called pvalue
local pvalue "P=`: display %3.2f `pval''"
di "`pvalue'"
* 
* Now write your local macros to the the "subgroup" frame
* Each "frame post" command will add 1 additional row to the frame.
* We will graph these line by line.
* First line with overall group name a P-value for interaction:
frame post subgroup ("`label_group'") (.)  (.) (.) (.) ("`pvalue'") 
* Now each subgroup by itself:
frame post subgroup ("`label_subgroup_0'") (`n_0')  (`beta_0') (`low95_0') (`high95_0') ("") 
frame post subgroup ("`label_subgroup_1'") (`n_1')  (`beta_1') (`low95_1') (`high95_1') ("") 
* Optional blank line:
frame post subgroup ("") (.) (.) (.) (.) ("") 
*
*** Female sex
* This is already dichotomous, so don't need to create a new variable
* like we did for age.
local label_group "{bf:Sex}"  
*group 0, males
local label_subgroup_0 "Males" 
logistic infect time if female==0 
local n_0 = e(N) 
local beta_0 = r(table)[1,1] 
local low95_0 = r(table)[5,1]
local high95_0 = r(table)[6,1]
*group 1, females
local label_subgroup_1 "Females" 
logistic infect time if female==1 
local n_1 = e(N) // N
local beta_1 = r(table)[1,1] 
local low95_1 = r(table)[5,1]
local high95_1 = r(table)[6,1]
*interaction P-value
logistic infect c.time##i.female
local pval = r(table)[4,5]
local pvalue "P=`: display %3.2f `pval''"
*write to subgroup frame
frame post subgroup ("`label_group'") (.)  (.) (.) (.) ("`pvalue'") 
frame post subgroup ("`label_subgroup_0'") (`n_0')  (`beta_0') (`low95_0') (`high95_0') ("") 
frame post subgroup ("`label_subgroup_1'") (`n_1')  (`beta_1') (`low95_1') (`high95_1') ("") 
frame post subgroup ("") (.) (.) (.) (.) ("") 
*
*** patient
* need to generate a patient dichotomous variable
generate patientsplit = (patient>=20) if !missing(patient) // below 20 is 0, 20 and above is 1
local label_group "{bf:Patient}" 
*group 0, below 20
local label_subgroup_0 "Under 20th patient" 
logistic infect time if patientsplit==0 
local n_0 = e(N) 
local beta_0 = r(table)[1,1] 
local low95_0 = r(table)[5,1]
local high95_0 = r(table)[6,1]
*group 1, 20 and above
local label_subgroup_1 "At least the 20th patient" 
logistic infect time if patientsplit==1 
local n_1 = e(N) 
local beta_1 = r(table)[1,1] 
local low95_1 = r(table)[5,1]
local high95_1 = r(table)[6,1]
*interaction P-value
logistic infect c.time##i.agesplit
local pval = r(table)[4,5] 
local pvalue "P=`: display %3.2f `pval''"
*write to subgroup frame
frame post subgroup ("`label_group'") (.)  (.) (.) (.) ("`pvalue'") 
frame post subgroup ("`label_subgroup_0'") (`n_0')  (`beta_0') (`low95_0') (`high95_0') ("") 
frame post subgroup ("`label_subgroup_1'") (`n_1')  (`beta_1') (`low95_1') (`high95_1') ("") 
frame post subgroup ("") (.) (.) (.) (.) ("") 
*
*** Now make the figure. You'll have to modify this so the number of rows 
*   in your subgroup frame matches the labels and whatnot
set scheme s1mono // I like this scheme
* Change frame to the subgroup frame
cwf subgroup
* Generate a row number by the current order of the data in this frame
gen row=_n
* Here's the code to make the figure
twoway ///
(scatter row beta, msymbol(d) mcolor(black) msize(medium)) ///
(rcap low95 high95 row, horizontal lcolor(black) lwidth(medlarge)) ///
, ///
legend(off) ///
xline(1, lcolor(red) lpattern(dash) lwidth(medium)) ///
title("Title") ///
yti("Y Title") ///
xti("X Title") ///
yscale(reverse) ///
yla( ///
1 "`=rowname[1]'" ///
2 "`=rowname[2]', n=`=n[2]'" ///
3 "`=rowname[3]', n=`=n[3]'" ///
4 " " /// blank since it's a blank row
5 "`=rowname[5]'" ///
6 "`=rowname[6]', n=`=n[6]'" ///
7 "`=rowname[7]', n=`=n[7]'" ///
8 " " /// blank since it's a blank row
9 "`=rowname[9]'" ///
10 "`=rowname[10]', n=`=n[10]'" ///
11 "`=rowname[11]', n=`=n[11]'" ///
12 " " /// blank since it's a blank row
, angle(0) labsize(small) noticks) ///
xla(0.8(.2)2.2) ///
text(1 1.1 "`=pval[1]'", placement(e) size(small)) /// these are the p-value labels
text(5 1.1 "`=pval[5]'", placement(e) size(small)) ///
text(9 1.1 "`=pval[9]'", placement(e) size(small)) 
*
* Now export your figure as a PNG file
graph export "myfigure.png", replace width(1000)

Formatting P-values for Stata output

If you like automating your Stata output, you have probably struggled with how to format P-values so they display in a format that is common in journals, namely:

  • If P>0.05 and not close to 0.05, P has an equals sign and you round at the hundredth place
    • E.g., P=0.2777 becomes P=0.28
  • If P is close to 0.05 , P has an equals sign and you round at the ten thousandth place
    • E.g., P=0.05033259 becomes P=0.0503
    • E.g., P=0.0492823 become P=0.0493
  • If well below 0.05 but above 0.001, P has an equals sign and you round at the hundredth place
    • E.g., P=0.0028832 becomes P=0.003
  • if below 0.001, display as “P<0.001"
  • if below 0.0001, display as “P<0.0001"

Here’s a loop that will do that for you. This contains local macros so you need to run it all at once from a do file, not line by line. You’ll need to have grabbed the P-value as a from Stata as a local macro named pval. It will generate a new local macro called pvalue that you need to treat as a string.

local pval 0.05535235 // change me to play around with this
if `pval'>=0.056 {
	local pvalue "P=`: display %3.2f `pval''"
}
if `pval'>=0.044 & `pval'<0.056 {
	local pvalue "P=`: display %5.4f `pval''"
}
if `pval' <0.044 {
	local pvalue "P=`: display %4.3f `pval''"
}
if `pval' <0.001 {
	local pvalue "P<0.001"
}
if `pval' <0.0001 {
	local pvalue "P<0.0001"
}
di "original P is " `pval' ", formatted is " "`pvalue'"

Here’s how you might use it to format the text of a regression coefficient that you put on a figure.

sysuse auto, clear
regress weight mpg
// let's grab the P-value for the mpg
matrix list r(table) //notice that it's at [4,1]
local pval = r(table)[4,1] 

if `pval'>=0.056 {
	local pvalue "P=`: display %3.2f `pval''"
}
if `pval'>=0.044 & `pval'<0.056 {
	local pvalue "P=`: display %5.4f `pval''"
}
if `pval' <0.044 {
	local pvalue "P=`: display %4.3f `pval''"
}
if `pval' <0.001 {
	local pvalue "P<0.001"
}
if `pval' <0.0001 {
	local pvalue "P<0.0001"
}
di "original P is " `pval' ", formatted is " "`pvalue'"

// now make a scatterplot and print out that formatted p-value
twoway ///
(scatter weight mpg) ///
(lfit weight mpg) ///
, ///
text(3500 35 "`pvalue'", size(large))

Adding overlaying text “boxes”/markup to Stata figures/graphs

What’s up with adding text to figures in Stata?

It’s handy to add text to your plots to orient your readers to specific observations in your figures. You might opt to highlight a datapoint, add percentages to bars, or say what part of a figure range is good vs. bad.

Adding text to Stata is relatively straightforward, once you get over the initial learning hump. Check out the added text help files by typing —help added_text_options— and —help textbox_options— for details.

Added text syntax

First, make your figure, then add your comma, then add your syntax. Like everything(?) in Stata, the location is specified as Y then X. Then add your text in quotes (or multiple quotes in a row if you want to insert a line break). You can bold or italicize by using the it: and bf: formatting and wrapping the text you want to apply it to in brackets. Then you drop a comma, specify placement as centered (c) or cardinal directions (e.g., ne, s, w) of that text box relative to the Y,X coordinates, justification of the text as left-aligned (left), centered (center), or right-aligned (right), and size of the text (vsmall, small, medsmall, medium, medlarge, large, vlarge, huge as shown in —help textsizestyle–).

Text box without outline, playing around with bolding and italicizing:

sysuse auto, clear
twoway ///
(scatter mpg price) ///
, ///
text(35 12000 "{bf:First} line" "Second {it:line}", placement(c) justification(left) size(large))

Text box with outline

The main difference is that you add the word “box” to your text box options.

sysuse auto, clear
twoway ///
(scatter mpg price) ///
, ///
text(35 12000 "{bf:First} line" "Second {it:line}", placement(c) justification(left) size(large) box)

You can further customize your box by specifying options shown under —help textbox_options–. Here, I’m making a box with no fill, and a thick/dashed/red outline with fcolor (fill color, which can be “none” for blank/see-through), lcolor (outline color), lpattern (outline pattern), and lwidth (thickness of outline).

sysuse auto, clear
twoway ///
(scatter mpg price) ///
, ///
text(35 12000 "{bf:First} line" "Second {it:line}", placement(c) justification(left) size(large) box fcolor(none) lcolor(red) lpattern(dash) lwidth(thick)) 

That looks pretty cool, but I’d like there to be more space between the text and the border. That’s specified with “margin”. There’s lots of complex options for this command you can read about in —help marginstyle–, but specifying “small” margin will add just a little buffer.

sysuse auto, clear
twoway ///
(scatter mpg price) ///
, ///
text(35 12000 "{bf:First} line" "Second {it:line}", placement(c) justification(left) size(large) box fcolor(none) lcolor(red) lpattern(dash) lwidth(thick) margin(small))

Special characters, and offsetting the Y,X coordinates

Sometimes you need a superscript 2 or 3, an arrow, or some other special character. Here are some that you can copy/paste into your text:

≥
≤
±
↑
↓
←
→
↔
🗸
χ² (chi-squared)
°
α
β
δ
Δ
γ
² (squared)
³ (cubed)

Here’s how you might use the above arrow. Here, I’ve changed the placement to “e” for “east” so the text is to the right and centered of the specified Y,X coordinates of this point (which I looked up, it’s 41,5397). I’ve dropped the box and it’s only one line. I need to manually specify the Y,X coordinates, so it’s a bit clunky still.

sysuse auto, clear
twoway ///
(scatter mpg price) ///
, ///
text(41 5397 "←look at this point!", placement(e) justification(left) size(large))

You’ll notice that the left pointing arrow is inside the point of interest. The left pointing arrow in Ariel text isn’t exactly vertically centered (that’s a quirk of Ariel). You can “nudge” the text box to the right and up by adding an offset to the Y and X coordinates. To do math on an x,y point and add an offset, you place the point in an opening tick (to the left of 1 on your keyboard) and an apostrophe, and putting an equal sign at the beginning. After a bit of trial and error, we want to add an offset of +0.3 to the Y coordinate and +200 to the X coordinate.

sysuse auto, clear
twoway ///
(scatter mpg price) ///
, ///
text(`=41+0.3' `=5397+200' "←look at this point!", placement(e) justification(left) size(large))

If we want to automate the extreme numbers on either axis, we can use the sum command to grab the maximum value aka r(max) of the Y axis and the corresponding x value for that extreme. We can make a local macro to grab that value. Since this uses local macros, you need to run all lines at once in a do file, not line by line.

sysuse auto, clear
// grab yaxis (mpg) extreme
sum mpg, d
// now look at the r-value scalars we can call.
// note that r(max) is 41 and is the maximum. 
return list
// now save r(max) as a local macro
local ymax = r(max)
// now let's find the x value for that y value.
// remember that the x axis variable is "price" 
sum price if mpg==`ymax', d
// now look at r-value scalars you can call.
// since there's only 1 value, the mean, min, max,
// and sum are all the same. 
return list
// let's now grab the r(mean) of as a local macro
local ymax_x = r(mean)
// now you can prove that you grabbed the correct point
di "Y max =" `ymax' ", and that point is at X=" `ymax_x'
// now place those local macros in the Y and X, keeping
// the other offsets
twoway ///
(scatter mpg price) ///
, ///
text(`=`ymax'+0.3' `=`ymax_x'+200' "←look at this point!", placement(e) justification(center) size(large))

You can do the same for the x-axis extreme in the same script. for the X extreme, we’ll change the placement so it’s northwest and also tweak the Y,X coordinate offsets.

sysuse auto, clear
sum mpg, d
local ymax = r(max)
sum price if mpg==`ymax', d
local ymax_x = r(mean)
di "Y max =" `ymax' ", and that point is at X=" `ymax_x'
//
sum price, d
local xmax=r(max)
sum mpg if price==`xmax', d
local xmax_y=r(mean)
di "X max =" `xmax' ", and that point is at Y=" `xmax_y'

twoway ///
(scatter mpg price) ///
, ///
text(`=`ymax'+0.3' `=`ymax_x'+200' "←look at this point!", placement(e) justification(center) size(large)) ///
text(`=`xmax_y'+1' `=`xmax'+260' "look at this point!↓", placement(nw) justification(center) size(large))

What about adding text labels bar graphs?

There’s a simple –graph bar– command, which isn’t terribly customizable. Here, we’ll be using the –twoway bar– bar graphs and add some fancy error bars with –twoway rcap–. The simplest way to do this is to make a new “dataset” from summary statistics in a separate frame. Frames are new in Stata 16, so if you are using an older version, this won’t work.

After making a frame with a bunch of summary statistics by group (most of which we won’t use here, but you might opt to later so we’ll keep it all there), you get this figure of mean and +/- 1.98*SD error bars.

// this has local macros so need to run from top to bottom in a do file!
// clear all frames
version 16
frames reset
sysuse auto, clear
// make a new frame called "bardata" that will store the values for bars. 
frame create bardata str20 rowname n low95 high95 mean sd median iqrlow iqrhigh bar_order
//
// now grab summary statistics for foreign==0 
// grab 2.5th and 95th %ile and save as local macros
// in case you might want them ever
_pctile mpg if foreign==0, percentiles(2.5 97.5)
local low95 = r(r1)
local high95= r(r2)
// now sum with detail one row 
sum mpg if foreign==0, d
// post these to the new frame/bardata
// note the final row, which will be the x axis value that will correspond with bar placement
frame post bardata  ("Foreign==0") (r(N)) (r(mean)) (`low95') (`high95') (r(sd)) (r(p50)) (r(p25)) (r(p75)) (1)
// 
// now repeat for foreign==1
_pctile mpg if foreign==1, percentiles(2.5 97.5)
local low95 = r(r1)
local high95= r(r2)
sum mpg if foreign==1, d
// note the final row is now 3, leaving a gap of between the prior row 
frame post bardata  ("Foreign==1") (r(N)) (r(mean)) (`low95') (`high95') (r(sd)) (r(p50)) (r(p25)) (r(p75)) (3)

// now change to the frame with the summary statistics
cwf bardata 
// let's make a new variable that's the mean +/- 1.98*sd
gen sd198low = mean-(sd*1.98)
gen sd198high = mean+(sd*1.98)

// now graph those data 
twoway ///
(bar mean bar_order) ///
(rcap sd198low sd198high bar_order, vert lcolor(black)) ///
, ///
yla(0(10)50) ///
xla(1 "Domestic" 3 "Foreign") ///
ytitle("MPG") 

Let’s make things a bit more fancy with labels, and place a text box with the mean and SD immediately above/north of the bars, which themselves are means of MPG by group. One quirk of Stata is that when formatting printed text from stored values, you need to place the label in an opening and closing tick, put a colon, type display, then apply the formatting (e.g., %3.1f for 1 digit after the decimal) then the value you are trying to show. You’ll notice that formatting in the last two lines of the following code.

// clear all frames
version 16
frames reset
sysuse auto, clear
// make a new frame called "bardata" that will store the values for bars. 
frame create bardata str20 rowname n mean low95 high95  sd median iqrlow iqrhigh bar_order
//
// now grab summary statisticis for foreign==0 
// grab 2.5th and 95th %ile and save as local macros
_pctile mpg if foreign==0, percentiles(2.5 97.5)
local low95 = r(r1)
local high95= r(r2)
// now sum with detail one row 
sum mpg if foreign==0, d
// post these to the new frame/bardata
// note the final row, which will be the x axis value that will correspond with bar placement
frame post bardata  ("Foreign==0") (r(N)) (r(mean)) (`low95') (`high95') (r(sd)) (r(p50)) (r(p25)) (r(p75)) (1)
// 
// now repeat for foreign==1
_pctile mpg if foreign==1, percentiles(2.5 97.5)
local low95 = r(r1)
local high95= r(r2)
sum mpg if foreign==1, d
// note the final row is now 3, leaving a gap of between the prior row 
frame post bardata  ("Foreign==1") (r(N)) (r(mean)) (`low95') (`high95') (r(sd)) (r(p50)) (r(p25)) (r(p75)) (3)

// now change to the frame with the summary statistics
cwf bardata 
// let's make a new variable that's the mean +/- 1.98*sd
gen sd198low = mean-(sd*1.98)
gen sd198high = mean+(sd*1.98)

// let's grab the mean and SD for each group. This frame has only two 
// rows of data, so we'll use the row notatition to generate a local from
// the first and second rows, with row number specified in brackets
local group0_mean = mean[1]
local group0_sd =sd[1]
local group0_n = n[1]
// print to check your work
di "for group 0, mean =" `group0_mean' ", SD=" `group0_sd' ", n=" `group0_n'
// do the other group 
local group1_mean = mean[2]
local group1_sd =sd[2]
local group1_n = n[2]
// print to check your work
di "for group 1, mean =" `group1_mean' ", SD=" `group1_sd' ", n=" `group1_n'

// now graph those data 
// we'll add labels for the mean and SD and also
// label the N in the xlabels
twoway ///
(bar mean bar_order) ///
(rcap sd198low sd198high bar_order, vert lcolor(black)) ///
, ///
yla(0(10)50) ///
xla(1 "Domestic, n=`group0_n'" 3 "Foreign, n=`group1_n'") ///
ytitle("MPG") ///
text(`group0_mean' 1 "Mean (SD)= `: display %3.1f `group0_mean'' (`: display %3.1f `group0_sd'')", placement(n)) ///
text(`group1_mean' 3 "Mean (SD)= `: display %3.1f `group1_mean'' (`: display %3.1f `group1_sd'')", placement(n)) 

Generate random data, make scatterplot with fitted line, and merge multiple figures in Stata

I recently made these two figures with a Stata do file that (A) generates some fake data, (B) plots the data overall with a fitted line, (C) plots each participant individually with fitted line, and (D) merges these four individual plots into one overall plot. One tricky piece of this was to get the –graph combine– command to get the four figures to be square, I had to fiddle with the –ysize– to get them to be square. I had originally tried using the –aspect– option, but it builds in lots of white space to between the figures. this way seemed to work.

Here’s the Stata code to make the above figures.

************************
*****clear data/graphs**
****set graph schemes***
************************
clear
graph drop _all
set scheme s1mono

************************
*****generate data******
************************
set seed 8675309
// generate 40 random values
set obs 40
//...called x and y that range from - to +9
gen y = runiform(-9,9)
gen x = runiform(-9,9)
// now make an indicator for each grouping of 10
gen n =.
replace n= 1 if _n>=0 & _n<=10
replace n= 2 if _n>10 & _n<=20
replace n= 3 if _n>20 & _n<=30
replace n= 4 if _n>30 & _n<=40
// now for each successive 10 rows, add 100, 120, 140, and 160
replace y = y+100 if n==1
replace x = x+100 if n==1

replace y = y+120 if n==2
replace x= x+120 if n==2

replace y= y+140 if n==3
replace x= x+140 if n==3

replace y= y+160 if n==4
replace x= x+160 if n==4

************************
*****single figure******
************************
twoway ///
/// line of unity (45 degree line): 
(function y = x, ra(90 170) clpat(solid) clwidth(medium) clcolor(black)) ///
/// line of fit: 
(lfit y x, lcolor(red) lpattern(dash) lwidth(thick)) ///
/// dots for each group, using colors from colorbrewer2
(scatter y x if n==1, mcolor("230 97 1") msymbol(O) mlcolor(black)) ///
(scatter y x if n==2, mcolor("253 184 99") msymbol(O) mlcolor(black)) ///
(scatter y x if n==3, mcolor("178 171 210") msymbol(O) mlcolor(black)) ///
(scatter y x if n==4, mcolor("94 60 153") msymbol(O) mlcolor(black)) ///
, ///
/// force size:
ysize(4) xsize(4) ///
/// axis titles and rows:
xtitle("Reference BP, mm Hg") ///
ytitle("Cuffless BP, mm Hg") ///
xla(90(20)170) ///
yla(90(20)170) ///
legend(off) ///
/// labels:
text(102 109 "←Participant A", placement(e)) ///
text(119 127 "←Participant B", placement(e)) ///
text(142 135 "Participant C→", placement(w)) ///
text(160 154 "Participant D→", placement(w)) 
/// export figure:
graph export figure1av2.png, replace width(1000)


************************
***four-piece figures***
*****to merge***********
************************
twoway ///
(function y = x if n==1, ra(90 110) clpat(solid) clwidth(medium) clcolor(black)) ///
(lfit y x if n==1, lcolor(red) lpattern(dash) lwidth(thick)) ///
(scatter y x if n==1, mcolor("230 97 1") msymbol(O) mlcolor(black)) ///
, ///
title("Participant A") ///
xtitle(" ") ///
ytitle(" ") ///
yla(90(5)110) ///
xla(90(5)110) ///
legend(off) ///
name(figure1b1)

twoway ///
(function y = x if n==2, ra(110 130) clpat(solid) clwidth(medium) clcolor(black)) ///
(lfit y x if n==2, lcolor(red) lpattern(dash) lwidth(thick)) ///
(scatter y x if n==2, mcolor("253 184 99") msymbol(O) mlcolor(black)) ///
, ///
title("Participant B") ///
xtitle(" ") ///
ytitle(" ") ///
yla(110(5)130) ///
xla(110(5)130) ///
legend(off) ///
name(figure1b2)

twoway ///
(function y = x if n==3, ra(130 150) clpat(solid) clwidth(medium) clcolor(black)) ///
(lfit y x if n==3, lcolor(red) lpattern(dash) lwidth(thick)) ///
(scatter y x if n==3, mcolor("178 171 210") msymbol(O) mlcolor(black)) ///
, ///
title("Participant C") ///
xtitle(" ") ///
ytitle(" ") ///
yla(130(5)150) ///
xla(130(5)150) ///
legend(off) ///
name(figure1b3)

twoway ///
(function y = x if n==4, ra(150 170) clpat(solid) clwidth(medium) clcolor(black)) ///
(lfit y x if n==4, lcolor(red) lpattern(dash) lwidth(thick)) ///
(scatter y x if n==4, mcolor("94 60 153") msymbol(O) mlcolor(black)) ///
, ///
title("Participant D") ///
xtitle(" ") ///
ytitle(" ") ///
yla(150(5)170) ///
xla(150(5)170) ///
legend(off) ///
name(figure1b4)

***********************
***combine figures*****
***********************
graph combine figure1b1 figure1b2 figure1b3 figure1b4 /*figure1b5*/, ///
b1title("Reference BP, mm Hg") ///
l1title("Cuffless BP, mm Hg") ///
/// this forces the four scatterplots to be the same height/width
/// details here:
/// https://www.stata.com/statalist/archive/2013-07/msg00842.html
cols(2) iscale(.7273) ysize(5.9) graphregion(margin(zero))

/// export to figure
graph export figure1bv2.png, replace width(1000)