Chapter 3 Visualization

Stata has robust graphing capabilities that can both generate numerous types of plots, as well as modify them as needed. We’ll only cover the basics here, for a reference we would recommend A Visual Guide to Stata Graphics by Michael Mitchell, which lays out step-by-step syntax for the countless graphs that can be generated in Stata.

Let’s reload the auto dataset to make sure we’re starting on the same page.

. sysuse auto, clear
(1978 Automobile Data)



3.1 The graph command

Most (though not all, see some other graphs below) graphs in Stata are created by the graph command. Generally the syntax is

graph <type> <variable(s)>, <options>


The “type” is the subcommand.

For example, to create a bar chart of price by rep78, we could run

. graph bar price, over(rep78)



For further information, we could instead construct a boxplot.

. graph box price, over(rep78)



There are a few other infrequently used graphs, see help graph for details.

There is a plot subcommand, twoway, which takes additional sub-subcommands, and supports a wide range of types.

graph twoway <type> <variable(s)>, <options>


twoway creates most of the scatterplot-esque plots. The “types” in twoway are subcommands different from the subcommands in non-twoway graph, it takes options such as scatter to create a scatterplot:

. graph twoway scatter mpg weight



Note: For graph twoway commands, the graph is optional. E.g., these commands are equivalent:

graph twoway scatter mpg weight
twoway scatter mpg weight


This is not true of commands like graph box.

The options in the graphing commands are quite extensive and enable tweaking of many different settings. Rather than a full catalog of the options, here’s an example:

. twoway scatter mpg weight, msymbol(s) ///
>                           mcolor(blue) ///
>                           mfcolor(yellow) ///
>                           msize(3) ///
>                           title("Mileage vs Weight") ///
>                           xtitle("Weight (in lbs)") ///
>                           ytitle("Mileage") ///
>                           ylabel(15 "15" 25 "25" 35 "35")

.


Graphs made using twoway have an additional benefit - it is easy to stack them. For example, twoway lfit creates a best-fit line between the points:

. twoway lfit mpg weight



This isn’t really that useful. It would be much better to overlap those two - generate the scatter plot, then add the best fit line. We can easily do that by passing multiple plots to twoway:

. twoway (scatter mpg weight) (lfit mpg weight)



Note that the order of the plots matters - if you can tell, the best-fit line was drawn on top of the scatter plot points. If you reversed the order in the command (twoway (lfit mpg weight) (scatter mpg weight)), the line would be drawn first and the points on top of it.

Finally, note that options can be passed to each individual plot:

. twoway (scatter mpg weight, msymbol(t)) ///
>        (lfit mpg weight, lcolor(green))



Putting these options “globally”, as twoway (…) (…), msymbol(to) would NOT work, as msymbol is an option specifically for twoway scatter (and a few others), not for the more general twoway. There are options that apply to the twoway command, see help twoway_options for details.

There is an alternate way to specify the overlaid plots. These two commands are equivalent:

twoway (scatter mpg weight) (lfit mpg weight)
twoway scatter mpg weight || lfit mpg weight


We prefer the former as it makes it’s cleaner to distinguish when you have multiple overlaid plots with their own options, but some authors may chose the latter.

3.2 Other graphs

There are a very large number of graphs which do not exist under the graph command. Most are very niche, but the most important general example is histogram, which has its own command.

. histogram mpg
(bin=8, start=12, width=3.625)



You can see a full list of the non-graph plots by looking at

help graph other


3.3 Plotting by group

All graph commands accept a by(<grouping var>) option which will repeat the graphing command for each level of the grouping variable, and display all graphs on the same output. For example,

. twoway (scatter mpg weight) (lfit mpg weight), by(foreign)



It often looks better to see the two plots overlaid on each other for a more direct comparison. To do this, rather than using by(…), we’ll instead add each overlay conditionally:

. twoway (scatter mpg weight if foreign ==  0) ///
>          (scatter mpg weight if foreign == 1) ///
>          (lfit mpg weight if foreign ==  0) ///
>          (lfit mpg weight if foreign == 1)



Notice that Stata automatically made each plot a separate color, but not in a logical fashion. Here’s a cleaned up version:

. twoway (scatter mpg weight if foreign ==  0, mcolor(orange)) ///
>          (scatter mpg weight if foreign == 1, mcolor(green)) ///
>          (lfit mpg weight if foreign ==  0, lcolor(orange) lwidth(1.4)) ///
>          (lfit mpg weight if foreign == 1, lcolor(green) lwidth(1.4)), ///
>       legend(label(1 "Domestic") label(2 "Foreign") order(1 2)) ///
>       title("Mileage vs Weight") xtitle("Weight (lbs)") ///
>       ytitle("Mileage")



(Since its not entirely clear from the code, the order(1 2) argument inside legend serves two purposes - first, it “orders” the entries in the legend box, but secondly and more importantly, it does not contain 3 or 4. If you look at the previous plot, it had four entries in the legend for the two scatters plus two lfits. By excluding 3 and 4 from order [3 and 4 corresponding to the two lfits], their legend entries are ignored.)

3.4 Getting help on Graphs

There are a ton of options in all these graphs. Rather than list them all, we instead direct you to some various help pages.

help graph


Each individual type of graph has its own help page:

help graph box
help graph twoway
help twoway scatter
help histogram


There are various generalized options which are the same over the variety of plots. These can be found in the documentation of each individual graph, or you can access them directly:

Topics Help command
Help with titles, subtitles, notes, captions. help title_options
Axis labels, tick marks, scaling, etc. help axis_options
Manipulating the legend help legend_options
Modifying points (e.g. scatter) help marker_options
Adding labels to markers help marker_label_options
Options for any lines (e.g. lfit) help cline_options

3.5 Displaying multiple graphs simultaneously

You may have noticed that opening a new plot closes the old one. What if you wanted to compare the plots? The behind-the-scenes reason that the old plots are closed is that Stata names each plot and each plot can only be open once. The default name is “Graph”, so with each new plot, the “Graph” plot is overridden. If you closed a plot and wanted to re-open it, you can run the following at any point until you run another graph just like with estimation commands.

graph display Graph


When we create a new plot with the default name, we lose the last one.

If we give a plot a non-default name, it will be saved (so that it can be re-displayed later) and more importantly, will open a new window without closing the last. Running two plots with custom names opens two separate windows. (These are not run in the notes because obviously this won’t demonstrate well, but try them on your own.)

hist price, name(g1)
hist mpg, name(g2)


Names can be re-used (and plots re-generated) easily:

hist price, title("Histogram of Price") name(g1, replace)


We can also list (using dir), re-display (using display), or drop graphs (using drop):

graph dir
graph display g1
graph drop g1
graph drop _all


Finally, if you’d rather have all the graphs in one window with tabs instead of separate windows, use

set autotabgraphs on


(You can pass the permanently option to not have to do this every time you open Stata.) You still need to name graphs separately.

3.6 Exercise 2

Reload the NHANES data if you haven’t:

webuse nhanes2, clear


Using twoway scatter and twoway lfit, create a scatter plot of diastolic and systolic blood pressure, by gender. Be sure to color the lines and points consistenly and to clean up the legend.