Introduction to Stata 8

Svend Juul

Department of Epidemiology and Social Medicine, University of Aarhus, September 2004

Contents
1. Installing, customizing and updating Stata 3
2. Windows in Stata 5
3. Suggested mode of operation 7
4. Getting help 9
5. Stata file types and names 10
6. Variables and observations 11
6.1. Variable names 11
6.2. Numeric variables 11
6.3. Missing values 12
7. Command syntax 13
8. Getting data into Stata 16
9. Documentation commands 18
10. Modifying data 20
10.1. Calculations 20
10.2. Selections 21
10.3. Renaming and reordering variables 23
10.4. Sorting data 24
10.5. Numbering observations 24
10.6. Combining files 25
10.7. Reshaping data 26
11. Description and analysis 27
11.1. Categorical data 27
11.2. Continuous data 30
12. Regression models 32
12.1. Linear regression 32
12.2. Logistic regression 33
13. Survival and related analyses 34
14. Graphs 38
15. Miscellaneous 54
15.1. Memory considerations 54
15.2. String variables 55
15.3. Dates. Danish CPR numbers 57
15.4. Random samples, simulations 59
15.5. Immediate commands 60
15.6. Sample size and power estimation 61
15.7. Ado-files 62
15.8. Exchange of data with other programs 63
15.9. For old SPSS users 63
16. Do-file examples 65

Appendix 1: Purchasing Stata and manuals 67

Appendix 2: Entering data with EpiData 68
Appendix 3: NoteTab Light: a text editor 70

Alphabetic index 71
2
Preface
Stata is a software package designed for data management and statistical analysis; the primary
target group is academia.
This booklet is intended mainly for the beginner, but knowledge of fundamental Windows
functions is necessary. Only basic commands and a few more advanced analyses are shown.
You will find a few examples of output, and you can find more in the manuals. The booklet
does not replace the manuals.
You communicate with Stata by commands, either by typing them or by using the menus to
create them. Every action is elicited by a command, giving you a straightforward
documentation of what you did. This mode of operation is a contrast to working with
spreadsheets where you can do many things almost by intuition, but where the documentation
of what you did – including the sequence of actions – may be extremely difficult to
reconstruct.
Managing and analysing data is more than issuing commands, one at a time. In my booklet
Take good care of your data1 I give advice on documentation and safe data management, with
SPSS and Stata examples. The main message is: Keep the audit trail. Primarily for your own
sake, secondarily to enable external audit or monitoring. These considerations are also
reflected in this booklet.
To users with Stata 7 experience: You can go on working as usual, and your version 7 do-files
should work as usual. However, most users will love the new menu system and the much-
improved graphics.
To users with SPSS experience: By design Stata is lean compared to SPSS. Concerning
statistical capabilities Stata can do a lot more. Stata has a vivid exchange of ideas and
experiences with the academic users while SPSS increasingly targets the business world. Just
compare the home pages www.spss.com and www.stata.com. Or try to ask a question or give
a suggestion to each of the companies. In section 15.9 I show some SPSS commands and their
Stata counterparts.
Vince Wiggins of Stata Corporation has given helpful advice on the graphics section.

I welcome any comments and suggestions. My e-mail address is: sj@soci.au.dk.

Aarhus, September 2004

Svend Juul

1) Juul S. Take good care of your data. Aarhus, 2003. (download from
www.biostat.au.dk/teaching/software).

1
Notation in this booklet
Stata commands are shown like this:
tabulate agegr sex , chi2
tabulate and chi2 are Stata words, shown with italics, agegr and sex is variable
information, shown with ordinary typeface.
In the output examples you will often see the commands preceded by a period:
. tabulate agegr sex , chi2
This is how commands look in output, but you should not type the period yourself when
entering a command.
Optional parts of commands are shown in light typeface and enclosed in light typeface square
brackets. Square brackets may also be part of a Stata command; in that case they are shown in
the usual bold typeface. Comments are shown with light typeface:
save c:\dokumenter\proj1\alfa1.dta [ , replace]
summarize bmi [fweight=pop] // Weights in square brackets

In the examples I use c:\tmp as my folder for temporary files.

In the Stata manuals it is assumed that you use c:\data for all of your Stata files. I strongly
discourage that suggestion. I am convinced that files should be stored in folders reflecting the
subject, not the program used; otherwise you could easily end up confused. You will therefore
see that I always give a full path in the examples when opening (use) or saving files (save).

Throughout the text I use Stata's style to refer to the manuals:

[GSW] Getting Started with Stata for Windows
[U] User's Guide
[R] Base Reference manual (4 volumes)
[G] Graphics manual
[ST] Survival analysis and epidemiological tables
See more on manuals in appendix 1.

2
1. Installing, customizing and updating Stata
1.1. Installing Stata [GSW] 1
No big deal, just insert the CD and follow the instructions. By default Stata will be installed in
the c:\Stata folder. 'Official' ado-files will be put in c:\Stata\ado, 'unofficial' in
c:\ado. To get information about the locations, enter the Stata command sysdir. The
folders created typically are:
c:\Stata the main program
c:\Stata\ado\base 'official' ado-files as shipped with Stata
c:\Stata\ado\updates 'official' ado-file updates
c:\ado\plus downloaded 'unofficial' ado-files
c:\ado\personal for your own creations (ado-files, profile.do etc.)

A must: Update now – and do it regularly

Right after installing Stata, in he menu bar click:
Help ► Official updates
and select http://www.stata.com to get the most recent modifications and corrections of bugs.
Do this regularly to keep Stata updated.

Recommended: Also install NoteTab Light

As described in section 2, Stata has some shortcomings in handling output, and you will
benefit a lot from a good text editor. I recommend NoteTab Light; see appendix 3.

1.2. Customizing Stata

Create desktop shortcut icon
In the Start Menu you find the Stata shortcut icon. Right-click it and drag it to the desktop to
make a copy. Next right-click the desktop shortcut icon and select Properties (Egenskaber).
In the Path field you see e.g. c:\stata\wstata.exe /m1, meaning that 1 MB of memory is
reserved for Stata. You might increase this, see section 15.1 on memory considerations.
As the start folder you probably see c:\data. I strongly suggest to change the start folder to
c:\dokumenter or whatever your personal main folder is. The reasons for this suggestion:
• You should put your own text-, graph-, data- and do-files in folders organised and
named by subject, not by program, otherwise you will end up confused.
• All of your 'own' folders should be sub-folders under one personal main folder, e.g.
c:\dokumenter. This has at least two advantages:
- You avoid mixing your 'own' files with program files
- You can set up a consistent backup strategy.

3
The profile.do file [GSW] A7
If you put a profile.do file in the c:\ado\personal folder2 the commands will be
executed automatically each time you open Stata. Write your profile.do using any text-
editor, e.g. Stata's Do-file editor or NoteTab (see appendix 3) – but not Word or WordPerfect.
// c:\ado\personal\profile.do
set logtype text // simple text output log
log using c:\tmp\stata.log , replace // open output log
cmdlog using c:\tmp\cmdlog.txt , append // open command log
// See a more elaborate profile.do in section 16

set logtype text writes plain ASCII text (not SMCL) in the output log, to enable
displaying it in e.g. NoteTab.
log opens Stata's log file (stata.log) to receive the full output; the replace option
overwrites old output. The folder c:\tmp must exist beforehand.
cmdlog opens Stata's command log file (cmdlog.txt); the append option keeps the
command log from previous sessions and lets you examine and re-use past commands.

Fonts. Window sizes and locations [GSW] 18, 20

Start by:
Prefs ► General Preferences ► Windowing
and let the Review and Variables windows not be floating. Next, maximize the main Stata
window (click the upper right [□] button). Arrange windows as shown in section 2, and:
Prefs ► Save Windowing Preferences

If you somehow lost the setting, you can easily recreate it by:
Prefs ► Load windowing preferences

By default the Results window displays a lot of colours; to me they generate more smoke than
light. I chose to highlight error messages only and to underline links:
Prefs ► General preferences ► Result colors ► Color Scheme: Custom 1
Result, Standard, Input: White
Errors: Strong yellow, bold
Hilite: White, bold
Link: Light blue, underlined
Background: Dark blue or black

In each window (see section 2) you may click the upper left window button; one option is to
select font for that type of window. Select a fixed width font, e.g. Courier New 9 pt.

Windows 98 and ME users only:

These Windows versions have restricted capacity for dialogs. To avoid crashes:
set smalldlg on , permanently

2) [GSW] appendix A7 recommends otherwise. I stick to my recommendation; this place is safer.

4
2. Windows in Stata [GSW] 2

I Suggest arranging the main windows as shown below. Once you made your choices:
Prefs ► Save Windowing Preferences

█ Intercooled Stata 8.2

File Edit Prefs Data Graphics Statistics User Window Help

█ Review
sysuse auto.dta █ Stata Results
summarize
generate gp100m = 100/mp . generate gp100m = 100/mpg
label variable gp100m "Gallo
. label variable gp100m "Gallons per 100 miles"
regress gp100m weight
. regress gp100m weight
Source | SS df MS Number of obs = 74
-------------+------------------------------ F( 1, 72) = 194.71
Model | 87.2964969 1 87.2964969 Prob > F = 0.0000
█ Variables Residual | 32.2797639 72 .448330054 R-squared = 0.7300
-------------+------------------------------ Adj R-squared = 0.7263
make Make and M Total | 119.576261 73 1.63803097 Root MSE = .66957
price Price
-----------------------------------------------------------------------
mpg Mileage (mp
gp100m | Coef. Std. Err. t P>|t| [95% Conf. Interval]
rep78 Repair recor
-------------+---------------------------------------------------------
headroom Headroom (i
weight | .001407 .0001008 13.95 0.000 .001206 .0016081
trunk Trunk space
_cons | .7707669 .3142571 2.45 0.017 .1443069 1.397227
weight Weight (lbs.)
-----------------------------------------------------------------------
length Length (in.)
-more-
turn Turn Circle (
displacement Displaceme
gear_ratio Gear Ratio █ Stata Command
foreign Car type
gp100m Gallons per regress gp100m weight

Command line window

In this one line window you may enter single commands and execute them by [Enter].

Review window
This window displays the most recent commands. Click a command in the Review window to
paste it to the Command line window where you may edit and execute it. You may also scroll
through past commands using the PgUp and PgDn keys.
Save past commands to a do-file by clicking the upper left Review window button and: Save
Review Contents. Another option is to use the command log file (cmdlog.txt; section 1.2).

Variables window
You see a list of the variables in memory. Paste a variable name to the Command line window
by clicking it. You may also paste a variable name to an active dialog field.

Results window
This is the primary display of the output. Its size is limited, and you can't edit it. You may
print the contents, but in general my suggestion to let NoteTab print the log-file is handier.

5
Viewer window [GSW] 3
The main use of this window is viewing help files (see help and search, section 4). You
may select part of the viewer window for printing, using the mouse – but unfortunately not
the keyboard – to highlight it.

█ Stata Viewer [Advice on Help]

Back Refresh Search Help! Contents What's New News
Command: help viewer

Using the Viewer manual: [GS] 3 Using the Viewer

In the Viewer, you can

see help for contents or help for any Stata command

search help files, documentation, and FAQs (advice on using search)

find and install SJ, STB, and user-written programs from the net
review, manage, and uninstall user-written programs

check for and optionally install official updates

view your logs or any file

launch your browser

see the latest news from www.stata.com

Stata also suggests that you use the Viewer window for viewing and printing output (the log
file), but it does not work well, and I find it much handier to use a general text editor (see
section 3) for examining, editing and printing output.

Data window [GSW] 9

The Data window looks like a spreadsheet, and you may use it for entering small amounts of
data, but I don't recommend that, see section 8. A data entry program proper should be
preferred, e.g. EpiData, see appendix 2.
In [GSW] 9 it is demonstrated how to use the edit command to make corrections directly
in the data window. For reasons of safety and documentation I strongly discourage that.
Modifications to data should be made with do-files; see Take good care of your data.
The browse command enables you to see but not modify data in the Data window. To see a
specific cell issue the command:
browse age in 23 // the variable age in the 23rd observation

Do-file editor [GSW] 15

This is a standard text editor used for writing do-files (see section 3.1). The do-file editor has
the special feature that you may request Stata to execute all or a selection of the commands by
clicking the Do-button. I prefer NoteTab Light to the do-file editor; see section 3.

6
3. Suggested mode of operation
The recommendations in this section only partly follow what is recommended in the [GSW]
manual; I try to explain why, when my recommendations differ.

3.1. Issuing commands

The command line window
In the Command line window you may enter commands, one at a time, and execute them.
This works well in simple tasks, but for complex tasks it is a lot safer and more efficient to
write the commands in a do-file before execution.

The dialogs (menu system) [GSW] 2

The menu system enables you to generate quite complex commands without looking them up
in the manuals. You may need to do some editing if you are not quite satisfied with the result.
For graphs and many analyses the menus are a blessing. But for documentation commands
(section 9) and calculation commands (section 10) it is a lot easier to write the commands
directly in the command window or in a do-file than to generate them by the menus.

Do-files [U] 19
See examples of do-files in section 16. A do-file is a series of commands to be executed in
sequence. For any major tasks this is preferable to entering single commands because:
• You make sure that the commands are executed in the sequence intended.
• If you discover an error you can easily correct it and re-run the do-file.
• The do-file serves as documentation of what you did.
Use the do-file editor or any text editor like NoteTab (see appendix 3) to enter the commands.
I prefer NoteTab to the do-file editor, because I have direct access to several do-files and the
output in one place, although I cannot execute commands directly from NoteTab.
You may, after having issued a number of more or less successful commands, click the upper
left Review window button to save the contents of the Review window as a do-file. The
command log file (cmdlog.txt) may be used for the same purpose.

The do command
Find and execute a do-file by clicking: File ► Do...
or by entering path and filename in the command window:
do c:\dokumenter\proj1\alpha.do

You may also from the do-file editor execute the entire or a highlighted part of the current file
by clicking the do-button (number two from the right). The disadvantage of this method is
that the name of your do-file is not reflected in the output. I recommend issuing a do
command with full path and name of the do-file, for reasons of documentation.

7
3.2. Handling output
Stata's output facilities are less than optimal. In this section I show how you can use the third-
party program NoteTab to handle output for editing and printing.

The Results window

The output immediately appears in the Results window. You may print all of the Results
window (be careful!) or a selected part of it. However, manoeuvring is more restricted than in
a text editor, and you can use only the mouse, not the keyboard, to highlight a section. You
cannot edit its contents. If you, like me, dislike the output interruptions by -more- you can:
set more off [, permanently]

The size of the Results window buffer is restricted, and you only have access to the last few
pages of output. To increase the buffer size (default 32,000 bytes) permanently:
set scrollbufsize 200000

The Viewer window

This window is excellent to examine help files, see section 4. [GSW] section 3 and 17 also
describe how to use it to examine and print output, but it is much too inflexible. The SMCL-
formatted output is a mixed blessing, with parts of a table in bold, parts in plain text. And
actually only the [GSW] manual uses SMCL-formatted output, the others use plain text.

The log files and NoteTab

You may generate two log files, a full log and a command log. [GSW] section 17 tells how to
open and close log files. That didn't work well for me; I often forgot to open a log file. Now I
let the start-up file profile.do create the log files automatically, see section 1.2.

The full log (c:\tmp\stata.log) is a copy of what you saw in the Results window. I use it
to inspect, edit and print output in NoteTab. I selected plain ASCII text for the full log; it is
overwritten next time you start Stata or when you issue the newlog command.

The nt command gives you rapid access to your output in NoteTab. See appendix 3 on how
to create both commands.
The command log (c:\tmp\cmdlog.txt) includes all commands issued. It is cumulative,
i.e. new commands are added to the file, which is not overwritten next time Stata is opened.
You may use it instead of the Review window to see and recover previous commands.

Copying a table to a word processor document.

You might want to use a Stata table for publication. Copying a table directly to a word
processor document does not work well, but you may use Excel as an intermediary:
1. Highlight the table in the Results window. Right-click it and select Copy Table
2. Open Excel and paste the table to it ([Ctrl]+V). Edit the table if needed.
3. Copy-and-paste the table from Excel to your document.
To do the trick correctly, Windows must be set to display decimal periods (unless you in Stata
chose set dp comma).
8
4. Getting help [GSW] 4, 19, 20; [U] 2, 8

4.1. The manuals

See recommendations in Appendix 1.

4.2. Online help [GSW] 4; [U] 8, 32

Online help is best displayed in the Viewer window (see section 2 and 3). Issue help and
search from the Viewer's command line, whelp and findit from the Stata command
line. Keep your Stata updated – then your online help is updated too.

help and whelp

If you know the command name (e.g. tabulate) see the help file (tabulate.hlp) by:
help tabulate from the Viewer command line: or
whelp tabulate from Stata’s command line window
The help file is displayed in the Viewer window, and from here you may print it. You may
also use the links included. Try it.

search and findit. Using the menus

You need not know the command name. Get information about nonparametric tests by:
search nonparametric from the Viewer command line
To search Stata and the net for information on goodness-of-fit tests with poisson regression:
findit poisson goodness from Stata’s command line window
You may also use the menus to locate a command:
Statistics ► Summaries, tables & tests ► Nonparametric tests
and find quite a few procedures.

FAQs (Frequently asked questions)

Try www.stata.com/support/faqs. This site includes a lot of advice on various topics.

Statalist and the Danish Statanewcomerlist

Statalist is a forum for questions from users; see www.stata.com/support/statalist.
A Danish list especially for newcomers has been established by University of Southern
Denmark and Aarhus University; see www.biostat.sdu.dk/statalist.html.

Error messages
Stata's short error messages include a code, e.g. r(131). The code is a link, and by clicking
it you get more clues. Some error messages, however, are not very informative.

9
5. Stata file types and names [U] 14.6

.dta files: Stata data

The extension for a Stata data set is .dta. Stata data sets can only be created and interpreted
by Stata itself.

.do files: command files

A do-file with the extension .do is a number of commands to be executed in sequence. Do-
files are in standard ASCII format and can be edited and displayed by any text editor.
You may issue single commands in the command line window, but if you are doing anything
substantial you should do it with a do-file. You find examples in section 16 and some useful
examples in Take good care of your data. In both places I emphasize the importance of a
system for naming do-files.

.ado files: programs

An ado-file with the extension .ado is a program. Ado-files are in standard ASCII format.
For more information see section 15.7.

.hlp files: Stata help

Stata's online documentation is kept in .hlp files, written in SMCL format (somewhat like
HTML). SMCL-formatted files can be displayed in the Viewer window.

.gph files: graphs [GSW] 16; [G] (Graphics manual)

Stata graphs can be saved as .gph files; see section 14.8.

.dct files: dictionary files [U] 24; [R] infile

Fixed format ASCII data may be read with infile using a dictionary file. I don't
demonstrate this option in section 8.

10
6. Variables
A Stata data set is rectangular; here is one with five observations and four variables:
Variables
obsno age height weight
1 27 178 74
Observations

2 54 166 67
3 63 173 85
4 36 182 81
5 57 165 90

6.1. Variable names

Variable names can be 1-32 characters, but Stata often abbreviates long variable names in
output, so I recommend to use only 8 characters. The letters a-z (but not æøå), the numbers
0-9 and _ (underscore) are valid characters. Names must start with a letter (or an
underscore, but this is discouraged because many Stata-generated variables start with an
underscore). These are valid variable names:
a q17 q_17 pregnant sex

Stata is case-sensitive
Variable names may include lowercase and uppercase letters, but Stata is case-sensitive: sex
and Sex are two different variable names. Throughout this booklet I use lowercase variable
names; anything else would be confusing.

6.2. Numeric variables [U] 15.2

Most often you don't need worry about numeric types, but if you encounter memory
problems, you should know this (see section 15.1 on Memory considerations):

Type Bytes Precision (digits) Range (approx.)

Integer byte 1 2 ±100
int 2 4 ±32,000
long 4 9 ±2×109
Floating point float 4 7 ±1036
double 8 16 ±10308

compress can reduce the size of your dataset considerably by finding the most economical
way of storage.

11
Numeric formats [U] 15.5.1
The default is General format, presenting values as readable and precisely as possible. In most
cases you need not bother with numeric formats, but you may specify:
format dollars kroner %6.2f

Format Formula Example 2 1,000 10,000,000

General %w.dg %9.0g 1.414214 1000 1.00e+07

Fixed %w.df %9.0f 1 1000 10000000

%9.2f 1.41 1000.00 1.00e+07

Exponential %w.de %10.3e 1.414e+00 1.000e+03 1.000e+07

w: The total width, including period and decimals. d: Number of decimals

You may use comma formats with Stata, but there are confusing limitations, and I don't
recommend it. To display commas rather than periods (probably most relevant for graphs):
set dp comma

6.3. Missing values [U] 15.2.1

Missing values are omitted from calculations. There are two types of missing values:
The system missing value is shown as a . (period). It is created in input when a numeric field
is empty; by invalid calculations, e.g. division by 0, or calculations involving a missing value.
User-defined missing values are .a, .b, .c, ... .z . It is a good idea to use a general
principle consistently, e.g.:
.a Question not asked (complications to an operation not performed)
.b Question asked, no response
.c Response: Don't know

Unfortunately no data entry program accepts .a in a numeric field. In EpiData you might
choose the codes -1 to -3 (provided, of course, that they could not be valid codes) and let
Stata recode them:
recode _all (-1=.a)(-2=.b)(-3=.c)

Missing values are high-end numbers; the ordering is:

All valid numbers < . < .a < .b < ... < .z
You need not bother about the actual numerical values behind the missing values, but you
need to know the logics to avoid mistakes. In calculations missing values behave as expected,
but not so in conditions. The following command initially surprised me by listing all whose
age was > 65, and those with missing age:
list age if age > 65

To exclude the missing:

list age if age > 65 & age < .

To list the missing only, including the user-defined missing values:

list id age if age >= . or
list id age if missing(age)

12
7. Command syntax [U] 14.1

Stata is case-sensitive, and all Stata commands are lowercase. Variable names may include
lowercase and uppercase letters, but sex and Sex are two different variable names.
Throughout this booklet I use lowercase variable names; anything else would be confusing.
Uppercase variable names and are sometimes used within programs (ado-files, see section
15.7) to avoid confusion with the variable names in the data set.
The general syntax of Stata commands can be written like this:
[prefix:] command [varlist][if expression][in range][weight][, options]

Qualifiers and options

Qualifiers are general to many commands. See below on if, in and weights.
Options are specific to a command. A comma precedes the option list. Missing or misplacing
a comma is a frequent cause of error messages.

Command examples
Here are examples with the command summarize (mean, minimum, maximum etc.):
Prefix Command Varlist Qualifiers Options Comments
summarize No varlist: All variables
summarize _all _all: all variables
summarize sex age Two variables
summarize sex-weight Variables in sequence
summarize pro* All variables starting with
pro
summarize age if sex==1 Males only
summarize bmi in 1/10 First 10 observations
summarize bmi [fweight=n] Weighted observations
sort sex Separate table for each sex.
by sex: summarize bmi Data must be sorted first.
summarize bmi , detail Option: detail

Variable lists [U] 14.1.1

A variable list (varlist) calls one or more variables to be processed. Examples:
(nothing) sometimes the same as _all
_all all variables in the data set
sex age pregnant three variables
pregnant sex-weight pregnant and the consecutive variables sex to weight
pro* all variables starting with pro

13
In commands that have a dependent variable, it is the first in the varlist:
oneway bmi sex bmi is the dependent variable
regression bmi sex age bmi is the dependent variable
scatter weight height scatterplot, weight is y-axis
tabulate expos case The first variable defines the rows

Conditional commands. The if qualifier [U] 14.1.3.

The operators used in conditions are defined in section 10.1. Here are a few examples:
summarize age if sex==1 statistics for males only
list id age if age < 35 list only if age < 35
replace npreg=. if sex==1 set npreg to missing for males

Numeric ranges. [U] 14.1.4

Numeric ranges are marked by a slash:
recode age (1/9.999=1)(10/19.999=2) , generate(agegrp)
list sex age weight in 1/10 // observations 1 to 10

Number lists. The in qualifier [U] 14.1.8

A number list (numlist) is a list of numbers; there are some shorthand possibilities:
1(3)11 means 1 4 7 10
1(1)4 4.5(0.5)6 means 1 2 3 4 4.5 5 5.5 6
4 3 2 7(-1)1 means 4 3 2 7 6 5 4 3 2 1 (Danish CPR number test)
1/5 means 1 2 3 4 5
4/2 7/1 means 4 3 2 7 6 5 4 3 2 1 (Danish CPR number test)
Example:
list sex age weight in 1/10 // observations 1 to 10
twoway line mort year , xlabel(1900(20)2000) // x-axis labels

Weighting observations [U] 14.1.6, [U] 23.13

A typical use is to 'multiply' observations when the input is tabular:
Cases Controls
Exposed 21 30
Unexposed 23 100
Total 44 130

. input expos case pop // see section 8

1 1 21
1 0 30
0 1 23
0 0 100
. end
. tabulate expos case [fweight=pop] , chi2 // see section 11.1
. cc expos case [fweight=pop] // see section 11.1

14
by and bysort prefix [U] 14.5
Makes a command display results for subgroups of the data. Data must be pre-sorted:
sort sex
by sex: summarize age height weight
or, in one line:
bysort sex: summarize age height weight

Text strings, quotes

Stata requires double quotes around text strings, but you may omit quotes unless the string has
embedded blanks or commas:
label define sex 1 male 2 female 9 "sex unknown"

You need not use quotes around filenames:

save c:\dokumenter\proj1\alfa1.dta
unless they include blank space:
save "c:\dokumenter\project 1\alfa1.dta"

Comments [U] 19.1.2

The following is interpreted as comments, to include short explanations in a do-file:
• Lines that begin with *
• text surrounded by /* and */
• Text following // (the easy way; used in this booklet)

The purpose of comments is to make do-files more readable to yourself – Stata does not care
whatever you write.
// C:\DOKUMENTER\PROFILE.DO executes when opening Stata
summarize bmi , detail // Body mass index

Long command lines [U] 19.1.3

By default a command ends when the line ends (carriage return), and no special delimiter
terminates commands. However, command lines in do- and ado-files should be no longer than
80 characters. The problem is solved by /// telling Stata that the following line is a
continuation.
infix str10 cprstr 1-10 bday 1-2 bmon 3-4 byear 5-6 ///
control 7-10 using c:\dokumenter\p1\datefile.txt

Another option is to define ; (semicolon) as the future command delimiter:

#delimit ; // Semicolon delimits future commands
infix str10 cprstr 1-10 bday 1-2 bmon 3-4 byear 5-6
control 7-10 using c:\dokumenter\p1\datefile.txt ;
tab1 opagr ;
#delimit cr // Back to normal: Carriage return delimiter

15
8. Getting data into Stata [U] 24; [GSW] 7

On exchange of data with other programs, see section 15.8.

Open Stata data [R] save

Read an existing Stata data set from disk into memory by:
use c:\dokumenter\p1\a.dta [ , clear]
If there are data in memory, use will be rejected unless you specify the clear option.
You may also issue a clear command before the use command:
clear

If you want only observations that meet a condition:

use c:\dokumenter\p1\a.dta if sex==1

If you want the first 100 observations only:

use c:\dokumenter\p1\a.dta in 1/100

If you want to work with only a subset of variables:

use age sex q1-q17 using c:\dokumenter\p1\a.dta

Save Stata data [R] save

Save the data in memory to a disk file by:
save c:\dokumenter\p1\a.dta [ , replace]

If you already have a disk file with this name, your request will be rejected unless you specify
the replace option. Only use the replace option if you really want to overwrite data.

Enter data with EpiData

To enter data I recommend EpiData, available for free from www.epidata.dk. This easy-to-
use program has all the facilities needed. Further information in appendix 2.

Enter data as commands or in a do-file [R] input

Very small data sets. Define the variables with the input command and enter the values.
Finish with end. It can be done interactively from the command line or in a do-file. See
more examples in section 14.7 (Graph examples).
. input case expos pop
0 0 100
0 1 30
1 0 23
1 1 21
. end

You may also enter data directly in Stata's data window (not recommended; see section 2 and
[GSW] 6, 9).

16
Reading ASCII data [U] 24

Reading tab- or comma-separated data [R] insheet

In tab-separated data the tabulator character, here displayed as <T>, separates the values. A
tab-separated ASCII file is created e.g. if you save an Excel worksheet as a text (.txt) file. If
row 1 is variable names, Stata will find out and use them. In this and the following examples
the value of type in observation 2 is missing.
id <T> type <T> sold <T> price
1 <T> 2 <T> 47 <T> 51.23
2 <T> <T> 793 <T> 199.70

You may read a tab-separated ASCII file with variable names in row 1 by the command:
insheet using c:\dokumenter\p1\a.txt , tab

In comma-separated data a comma separates each value:

1, 2, 47, 51.23
2, , 793, 199.70

If you have a comma-separated file without variable names in row 1 the command is:
insheet id type sold price using c:\dokumenter\p1\a.txt , comma
insheet assumes that all data belonging to one observation are in one line.

Reading freefield data [R] infile (free format)

In freefield data commas or blanks separate each value:
1 2 47 51.23
2 . 793 199.70

If you have freefield data the command is

infile id type sold price using c:\dokumenter\p1\a.txt
infile does not assume that data belonging to one observation are in one line, and the
following data are the same as the data above:
1 2 47 51.23 2 . 793 199.70

Reading fixed format data [R] infix; [R] infile (fixed format)
In fixed format data the information on each variable is determined by the position in the line.
The blank type in observation 2 will be read as missing.
1 2 47 51.23
2 793 199.70

infix id 1 type 2-3 sold 4-7 price 8-14 using c:\dokumenter\p1\a.txt

Fixed format data can also be read by infile; to do this a dictionary file must be created,
specifying variable names and positions etc. See [R] infile (fixed format).

17
9. Documentation commands [GSW] 8

Stata does not need documentation commands; you need the documentation yourself. The
output becomes more legible, and the risk of errors when interpreting the output is reduced.

Data set label [U] 15.6.1; [R] label

You can give a short description of your data, to be displayed every time you open (use) data.
label data "Fertility data Denmark 1997-99. ver 2.5, 19.9.2002"

It is wise to include the creation date, to ensure that you analyse the most recent version.

Variable labels [U] 15.6.2; [R] label

You can attach an explanatory text to a variable name.
label variable q6 "Ever itchy skin rash?"

Use informative labels, but make them short; they are sometimes abbreviated in output.

Value labels [U] 15.6.3; [R] label

You can attach an explanatory text to each code for a variable. This is a two-step procedure.
First define the label (double quotes around text with embedded blanks):
label define sexlbl 1 male 2 female 9 "sex unknown"

Next associate the label sexlbl with the variable sex:

label values sex sexlbl

Use informative labels, but make them short; value labels are often abbreviated to 12
characters in output.
Most often you will use the same name for the variable and its label:
label define sex 1 male 2 female
label values sex sex
but the separate definition of the label enables you to reuse it:
label define yesno 1 yes 2 no
label values q1 yesno
label values q2 yesno

If you want to correct a label definition or add new labels, use the modify option:
label define sexlbl 9 "unknown sex" , modify
adds the label for code 9 to the existing label definition.
In output Stata unfortunately displays either the codes or the value labels, and you often need
to see them both, to avoid mistakes. You may solve this by including the codes in the labels;
this happens automatically with:
numlabel _all , add

18
See label definitions
See the value label definitions by:
label list or
labelbook

See the variable label definitions by:

describe

See a full codebook by:

codebook

Notes [R] notes

You may add notes to your data set:
note: 19.9.2000. Corrections made after proof-reading
and to single variables:
note age: 20.9.2000. Ages > 120 and < 0 recoded to missing

The notes are kept in the data set and can be seen by:
notes
Notes are cumulative; old notes are not discarded (and that is nice)

19
10. Modifying data
Don't misinterpret the title of this section: Never modify your original data, but add
modifications by generating new variables from the original data. Not documenting
modifications may lead to serious trouble. Therefore modifications:
• should always be made with a do-file with a name reflecting what it does:
gen.alfa2.do generates alfa2.dta.
• The first command in the do-file reads data (eg. use, infix).
• The last command saves the modified data set with a new name (save).
• The do-file should be 'clean', ie. not include commands irrelevant to the modifications.
See examples of modifying do-files in section 16 and in Take good care of your data.

10.1. Calculations
Operators in expressions [GSW] 12; [U] 16.2

Arithmetic Relational Logical

^ power > greater than ! not
* multiplication < less than ~ not
/ division >= > or equal | or
+ addition <= < or equal & and
- subtraction == equal
!= not equal
~= not equal

Arithmetic operators
generate alcohol = beers+wines+spirits
generate bmi = weight/(height^2)

The precedence order of arithmetic operators are as shown in the table; power before
multiplication and division, before addition and subtraction. Control the order by parentheses;
however the parentheses in the last command were not necessary since power takes
precedence over division – but they didn't harm either.

Relational and logical operators

replace salary = . if age<16 // salary set to missing
summarize age if sex==1
list sex age weight height if sex==1 & age <= 25
keep if sex==1 | age <= 25

Logical expressions can be true or false; a value of 0 means false, any other value (including
missing values) means true. This means that with sex coded 1 for males and 0 for females
and no missing values, the second command could have been written as:
summarize age if sex

20
generate; replace [R] generate
Generate a new variable by:
generate bmi=weight/(height^2)

If the target variable (bmi) already exists in the data set, use replace:
replace bmi=weight/(height^2)

Do a conditional calculation (males only):

generate mbmi=1.1*bmi if sex==1

Besides the standard operators there are a number of functions: [R] functions
generate y=abs(x) absolute value |x|
gen y=exp(x) exponential, ex
gen y=ln(x) natural logarithm
gen y=log10(x) base 10 logarithm
gen y=sqrt(x) square root
gen y=int(x) integer part of x. int(5.8) = 5
gen y=round(x) nearest integer. round(5.8) = 6
gen y=round(x, 0.25) round(5.8, 0.25) = 5.75
gen y=mod(x1,x2) modulus; the remainder after dividing x1 by x2
gen y=max(x1,...xn) maximum value of arguments
gen y=min(x1,...xn) minimum value of arguments
gen y=sum(x) cumulative sum across observations, from first to current obs.
gen y=_n _n is the observation number
gen y=_N _N is the number of observations in the data set

egen [R] egen

egen (extensions to generate) gives some more useful functions. It may be confusing that
functions working differently with generate and egen have the same names, so take
care. Here are some examples:

Generating the same value for all observations

egen meanage=mean(age) Mean age across observations
by sex: egen meange=mean(age) Mean age across observation, for each sex
egen sumage=sum(age) Sum of age across all observations (unlike
generate's sum)
egen maxage=max(age) Maximum value of age across observations
egen minage=min(age) Minimum value af age across observations

Generating individual values for each observation (each row)

egen minq=rmin(q1-q17) Min. value of q1-q17 for this observation
egen maxq=rmax(q1-q17) Max. value of q1-q17 for this observation
egen meanq=rmean(q1-q17) Mean value of q1-q17 for this observation
egen sumq=rsum(q1-q17) Sum of q1-q17 for this observation
egen valq=robs(q1-q17) Number of non-missing q1-q17 for this observation

21
recode [R] recode
Changes a variable's values, e.g. for grouping a continuous variable into few groups. The
inverted sequence ensures that age 55.00 (at the birthday) goes to category 4:
recode age (55/max=4)(35/55=3)(15/35=2)(min/15=1) , generate(agegr)

Value labels for the new variable may be included at once:

recode age (55/max=4 "55+")(35/55=3 "35-54")(15/35=2 "15-34") ///
(min/15=1 "-14") , generate(agegr)

Very important: The generate option creates a new variable with the recoded
information; without generate the original information in age will be destroyed.

Other examples:
recode expos (2=0) Leave other values unchanged
recode expos (2=0) , gen(exp2) Values not recoded transferred unchanged
recode expos (2=0) if sex==1 Values not recoded (sex != 1) set to missing
recode expos (2=0) if sex==1 , copy Values not recoded (sex != 1) unchanged
recode expos (2=0)(1=1)(else=.) Recode remaining values to missing (.)
recode expos (missing=9) Recode any missing (., .a, .b etc.) to 9

Another way to recode continuous data into groups: [R] egen

egen agegrp=cut(age) , at (0 5(10)85 120)

age agegrp
0 ≤ age < 5 0
5 ≤ age < 15 5
15 ≤ age < 25 15
.. ..
85 ≤ age < 120 85

for
Enables you with few command lines to repeat a command. To do the modulus 11 test for
Danish CPR numbers (see section 15.3) first multiply the digits by 4,3,2,7,6,5,4,3,2,1; next
sum these products; finally check whether the sum can be divided by 11. The CPR numbers
were split into 10 one-digit numbers c1-c10:
generate test = 0
for C in varlist c1-c10 \ X in numlist 4/2 7/1 : ///
replace test = test + C*X
replace test = mod(test,11) // Remainder after division by 11
list id cpr test if test !=0

C and X are stand-in variables (names to be chosen by yourself; note the use of capital
letters to distinguish from existing variables), to be sequentially substituted by the elements in
the corresponding list. Each list must be declared by type; there are four types: newlist
(list of new variables), varlist list of existing variables, numlist (list of numbers),
anylist (list of words).

for is not documented in the manuals any more. The foreach and forvalues
commands partially replace it, but they don't handle parallel lists as shown above. See section
15.7.
22
10.2. Selections

Selecting observations [GSW] 13; [R] drop

You may remove observations from the data in memory by:
keep if sex == 1 or, with the same effect:
drop if sex != 1

You may restrict the data in memory to the first 10 observations:

keep in 1/10

A selection may be requested already when opening a data set:

use c:\dokumenter\p1\a.dta if sex == 1

Observations dropped can only be returned to memory with a new use command. However,
preserve and restore (documented in [P]) let you obtain a temporary selection:
preserve // preserve a copy of the data currently in memory
keep if sex == 1
calculations
analyses
restore // reload the preserved dataset

Selecting variables [GSW] 13; [R] drop

You may remove variables from the data in memory by:
keep sex age-weight and by:
drop sex age-weight

A selection may be requested already when opening a data set:

use sex age-weight using c:\dokumenter\p1\a.dta

Sampling [R] sample

Keep a 10% random sample of the observations:
sample 10

To obtain a sample of exactly 57 observations:

sample 57 , count

10.3. Renaming and reordering variables

Renaming variables [R] rename

rename koen sex
The variable name koen is changed to sex. Contents and labels are unchanged.

Reordering variables [R] order

To change the sequence of variables specify:
order id age-weight

The new sequence of variables will be as defined. Any variables not mentioned will follow
after the variables mentioned.
23
10.4. Sorting data
sort [R] sort, [R] gsort
To sort your data according to mpg (primary key) and weight (secondary key):
sort mpg weight

sort only sorts in ascending order; gsort is more flexible, but slower. To sort by mpg
(ascending) and weight (descending) the command is:
gsort mpg –weight

10.5. Numbering observations [U] 16.7

The variable age in the third observation can be referred to as age[3]. The principle is:
First observation age[1]
Last observation age[_N]
Current observation age[_n]
Previous (lag) observation age[_n-1]
Next (lead) observation age[_n+1]
Observation 27 age[27]

From a patient register you have information about hospital admissions, one or more per
person, identified by cpr and admdate (admission date). You want to construct the
following variables: obsno (observation number), persno (internal person ID), admno
(admission number), admtot (patient's total number of admissions).
. use c:\dokumenter\proj1\alfa1.dta
. sort cpr admdate
. gen obsno=_n // _n is the observation number
. by cpr: gen admno=_n // _n is obs. number within each cpr
. by cpr: gen admtot=_N // _N is total obs. within each cpr
. sort admno cpr // all admno==1 first
. gen persno=_n if admno==1 // give each person number if admno==1
. sort obsno // original sort order
. replace persno=persno[_n-1] if persno >=. // replace missing persno
. save c:\dokumenter\proj1\alfa2.dta
. list cpr admdate obsno persno admno admtot in 1/7
cpr admdate obsno persno admno admtot
1. 0605401234 01.05.1970 1 1 1 3
2. 0605401234 06.05.1970 2 1 2 3
3. 0605401234 06.05.1971 3 1 3 3
4. 0705401234 01.01.1970 4 2 1 1
5. 0705401235 01.01.1970 5 3 1 1
6. 0805402345 01.01.1970 6 4 1 2
7. 0805402345 10.01.1970 7 4 2 2
. summarize persno // number of persons (max persno)
. anycommand if admno==1 // first admissions
. anycommand if admno==admtot // last admissions
. tab1 admtot if admno==1 // distribution of n of admissions

You may also create a keyfile linking cpr and persno, and remove cpr from your
analysis file. See example 16b in Take good care of your data.

24
10.6. Combining files [U] 25

Appending files [R] append

To combine the information from two files with the same variables, but different persons:
// c:\dokumenter\proj1\gen.filab.do
use c:\dokumenter\proj1\fila.dta , clear
append using c:\dokumenter\proj1\filb.dta
save c:\dokumenter\proj1\filab.dta

Merging files [R] merge

To combine the information from two files with different information about the same persons:
// c:\dokumenter\proj1\gen.filab.do
use c:\dokumenter\proj1\fila.dta , clear
merge lbnr using c:\dokumenter\proj1\filb.dta
save c:\dokumenter\proj1\filab.dta

Both files must be sorted beforehand by the matching key (lbnr in the example above), and
the matching key must have the same name in both data sets. Apart from the matching key the
variable names should be different. Below A and B symbolize the variable set in the input
files, and numbers represent the matching key. Missing information is shown by . (period):
fila filb filab _merge
1A 1B 1AB 3
2A 2A. 1
3B 3.B 2
4A1 4B 4A1B 3
4A2 4A2B 3

Stata creates the variable _merge which takes the value 1 if only data set 1 (fila)
contributes, 2 if only data set 2 (filb) contributes, and 3 if both sets contribute. Check for
mismatches by:
tab1 _merge
list lbnr _merge if _merge < 3

For lbnr 4 there were two observations in fila, but only one in filb. The result was two
observations with the information from filb assigned to both of them. This enables to
distribute information eg. about doctors to each of their patients – if that is what you desire.
But what if the duplicate lbnr 4 was an error? To check for duplicate id's before merging,
sort and compare with the previous observation:
sort lbnr
list lbnr if lbnr==lbnr[_n-1]

Another way to check for and list observations with duplicate id's is:
duplicates report lbnr
duplicates list lbnr

merge is a lot more flexible than described here; see [R] merge.

25
10.7. Reshaping data
collapse [R] collapse
You want to create an aggregated data set, not with the characteristics of each individual, but
of groups of individuals. One situation might be to characterize physicians by number of
patient contacts, another to make a reduced data set for Poisson regression (see section 13):
. // gen.stcollaps.cancer2.do
. use c:\dokumenter\proj1\stsplit.cancer2.dta , clear
. collapse (sum) risktime died , by(agegr drug)
. save c:\dokumenter\proj1\stcollaps.cancer2.dta
. summarize
Variable | Obs Mean Std. Dev. Min Max
-------------+-----------------------------------------------------
drug | 15 2 .8451543 1 3
agegr | 15 55 7.319251 45 65
risktime | 15 4.133334 3.01067 .3581161 10.88906
died | 15 2.066667 2.374467 0 8

reshape [R] reshape

E.g. with repeated measurements some analyses require a 'wide', some a 'long' data structure:
Long structure Wide structure
id time sbp id sbp1 sbp2 sbp3 sbp4
1 1 140 1 140 120 130 135
1 2 120 2 155 etc.
1 3 130
1 4 135
2 1 155
etc.

Switch between structures by:

reshape wide sbp , i(id) j(time) // Restructures long to wide
reshape long sbp , i(id) j(time) // Restructures wide to long

xpose [R] xpose

You may transpose observations and variables, i.e. let observations become variables and
variables become observations. This may be useful e.g. for restructuring data for a graph:
xpose , clear // clear is not optional

expand [R] expand, contract

You may duplicate observations according to a weighting variable. Look at the example in
section 7 about weighting observations. You can obtain the same result by expanding the four
observations to 130:
expand pop
tabulate expos case , chi

contract does the opposite of expand.

26
11. Description and analysis
This section gives information on the simpler statistical commands with examples of output.

summarize [R] summarize

summarize gives an overview of variables. It is useful for an initial screening of the data,
especially the Obs column giving the number of non-missing observations, and the Min and
Max columns.
. summarize
Variable | Obs Mean Std. Dev. Min Max
---------+-----------------------------------------------------
id | 37 19 10.82436 1 37
type | 37 1.891892 .9656254 1 4
price | 35 46.58 16.3041 11.95 78.95
rating | 35 2.514286 .9194445 1 4

Obtain detailed information on the distribution of selected variables by the detail option:
summarize price , detail

Find the nice modification summvl, also displaying variable labels, by: findit summvl.
. summvl
Variable Obs Mean Std.Dev Min Max Label
-------------------------------------------------------------------------
id 37 19 10.8244 1 37 identification number
type 37 1.89189 .965625 1 4 type of wine
price 35 46.58 16.3041 11.95 78.95 price per 75 cl bottle
rating 35 2.51429 .919444 1 4 quality rating

list [GSW] 11; [R] list

Case listings are useful to examine data, to check the result of calculations, and to locate
errors. The following lists sex-age for the first 10 observations. Codes rather than value labels
are displayed.
list sex-age in 1/10 , nolabel

Stata's listing facilities are clumsy when you want to list many variables simultaneously. Find
and install the useful alternative slist by: findit slist.

11.1. Categorical data

tab1 (simple frequency tables) and tab2 (crosstables) are both described in [R]
tabulate.

tab1
tab1 gives one-way tables (frequency tables) for one or more variables:

27
 . tab1 rating type (table for type not shown)
-> tabulation of rating
quality |
rating | Freq. Percent Cum.
--------------+-----------------------------------
1. poor | 6 17.14 17.14
2. acceptable | 9 25.71 42.86
3. good | 16 45.71 88.57
4. excellent | 4 11.43 100.00
--------------+-----------------------------------
Total | 35 100.00

tab2 [R] tabulate

tab2 with two variables give a two-way table (crosstable). In the following you see the use
of three options. column requests percentage distributions by column; chi2 a χ2 test, and
exact a Fisher's exact test:
. tab2 rating type , column chi2 exact
-> tabulation of rating by type
quality | type of wine
rating | 1 red 2 white 3 rosé 4 undeter | Total
--------------+--------------------------------------------+----------
1. poor | 4 1 1 0 | 6
| 25.00 9.09 20.00 0.00 | 17.14
--------------+--------------------------------------------+----------
(Part of output omitted)
--------------+--------------------------------------------+----------
4. excellent | 2 1 0 1 | 4
| 12.50 9.09 0.00 33.33 | 11.43
--------------+--------------------------------------------+----------
Total | 16 11 5 3 | 35
| 100.00 100.00 100.00 100.00 | 100.00
Pearson chi2(9) = 6.9131 Pr = 0.646
Fisher's exact = 0.653

You can request a three-way table (a two-way table for each value of nation) with:
bysort nation: tabulate rating type

The command:
tab2 rating type nation
gives three two-way tables, one for each of the combinations of the variables. But beware:
you can easily produce a huge number of tables.
[P] foreach
Imagine that you want 10 two-way tables: each of the variables q1-q10 by sex. With
tabulate you must issue 10 commands to obtain the result desired. If you call tab2 with
11 variables you get 55 two-way tables: all possible combinations of the 11 variables. The
foreach command (see section 15.7) lets you circumvent the problem:
foreach Q of varlist q1-q10 {
tabulate `Q' sex
}

The local macro Q is a stand-in for q1 to q10, and the commands generate 10 commands:
tabulate q1 sex
tabulate q2 sex etc.

28
tabi [R] tabulate
tabi is an 'immediate' command (see section 15.5) enabling you to analyse a table without
first creating a data set. Just enter the cell contents, delimiting the rows by \ (backslash):
. tabi 10 20 \ 17 9 , chi exact
| col
row | 1 2 | Total
-----------+----------------------+----------
1 | 10 20 | 30
2 | 17 9 | 26
-----------+----------------------+----------
Total | 27 29 | 56
Pearson chi2(1) = 5.7308 Pr = 0.017
Fisher's exact = 0.031
1-sided Fisher's exact = 0.016

epitab [ST] epitab

The commands in the epitab family perform stratified analysis. Here I show cc.
. cc case exposed , by(age) woolf
Maternal age | OR [95% Conf. Interval] M-H Weight
----------------+-------------------------------------------------
<35 | 3.394231 .9048403 12.73242 .7965957 (Woolf)
35+ | 5.733333 .5016418 65.52706 .1578947 (Woolf)
----------------+-------------------------------------------------
Crude | 3.501529 1.110362 11.04208 (Woolf)
M-H combined | 3.781172 1.18734 12.04142
------------------------------------------------------------------
Test of homogeneity (M-H) chi2(1) = 0.14 Pr>chi2 = 0.7105
Test that combined OR = 1:
Mantel-Haenszel chi2(1) = 5.81
Pr>chi2 = 0.0159

All procedures perform stratified analysis (Mantel-Haenszel). cc gives odds ratios for each
stratum and the Mantel-Haenszel estimate of the common odds ratio. The test of homogeneity
tests whether the odds ratio estimates could reflect a common odds ratio.

Command Measure of association Immediate command

ir Incidence rate ratio, incidence rate difference iri
cs Cohort studies: Risk ratio, risk difference csi
cc Case-control studies: Odds ratio cci
tabodds Odds ratio, several exposure levels. Trend test
mhodds Odds ratio, several exposure levels. Trend test
mcc Odds ratio (matched case-control data) mcci

If you want to stratify by more than one variable, the following command is useful:
egen racesex=group(race sex)
cc case exposed , by(racesex)

The immediate commands do not perform stratified analysis; an example with cci. Just
enter the four cells (a b c d) of the 2×2 table:
cci 10 20 17 9 , woolf

29
11.2. Continuous variables
oneway [R] oneway
compares means between two or more groups (analysis of variance):
oneway price type [ , tabulate noanova]

. oneway price type , tabulate

type of | Summary of price per 75 cl bottle
wine | Mean Std. Dev. Freq.
------------+------------------------------------
1. red | 48.15 12.650239 15
2. white | 42.590909 20.016952 11
3. rosé | 43.45 17.375268 6
4. undeterm | 59.616666 15.821924 3
------------+------------------------------------
Total | 46.58 16.304104 35
Analysis of Variance
Source SS df MS F Prob > F
------------------------------------------------------------------------
Between groups 780.660217 3 260.220072 0.98 0.4162
Within groups 8257.34954 31 266.366114
------------------------------------------------------------------------
Total 9038.00975 34 265.823816
Bartlett's test for equal variances: chi2(3) = 2.3311 Prob>chi2 = 0.507

The table, but not the test, could also be obtained by;
tabulate type , summarize(price) [R] tabsum

anova [R] anova

Similar to oneway, but handles a lot of complex situations.

tabstat [R] tabstat

tabstat is a flexible tool for displaying several types of tables, but includes no tests. To
obtain a number of descriptive statistics; here the mean, standard deviation, and 25, 50, and
75 percentiles:
. tabstat price mpg weight , stat(n mean sd p25 p50 p75) col(stat) format(%8.2f)
variable | N mean sd p25 p50 p75
-------------+------------------------------------------------------------
price | 74.00 6165.26 2949.50 4195.00 5006.50 6342.00
mpg | 74.00 21.30 5.79 18.00 20.00 25.00
weight | 74.00 3019.46 777.19 2240.00 3190.00 3600.00
--------------------------------------------------------------------------

The col(stat) option let the statistics form the columns; without it the statistics would
have formed the rows. The format() option lets you decide the display format. The
statistics are:
n, mean, sum, min, max, range, sd, var, cv (coefficient of variation), semean, skew
(skewness), kurt (kurtosis), p1, p5, p10, p25, p50 (or median), p75, p90, p95, p99, q
(quartiles: p25, p50, p75), and iqr (interquartile range).

30
ttest [R] ttest
T-test for comparison of means for continuous normally distributed variables:
ttest bmi , by(sex) Standard t-test, equal variances assumed
ttest bmi , by(sex) unequal Unequal variances (see sdtest)
ttest prebmi==postbmi Paired comparison of two variables
ttest prebmi==postbmi , unpaired Unpaired comparison of two variables
ttest bmidiff==0 One-sample t-test
ttesti 32 1.35 .27 50 1.77 .33 Immediate command. Input n, mean
n1 m1 sd1 n2 m2 sd2 and SD for each group

Distribution diagnostics
Diagnostic plots: [R] diagplots
pnorm bmi Normal distribution (P-P plot)
qnorm bmi Normal distribution (Q-Q plot)
Formal test for normal distribution: [R] swilk
swilk bmi Test for normal distribution
Test for equal variances: [R] sdtest
sdtest bmi , by(sex) Compare SD between two groups
sdtest prebmi==postbmi Compare two variables
Bartlett's test for equal variances is displayed by oneway, see above.

Non-parametric tests
For an overview of tests available, in the Viewer window command line enter:
search nonparametric

Here you see e.g.

kwallis Kruskall-Wallis equality of populations rank test
signrank Sign, rank, and median tests (Wilcoxon, Mann-Whitney)
Another way to look for nonparametric tests is via the menu system:
Statistics ► Summaries, tables & tests ► Nonparametric tests

31
12. Regression analysis
Performing regression analysis with Stata is easy. Defining regression models that give sense
is more complex. Especially consider:
• If you look for causes, make sure your model is meaningful. Don't include independent
variables that represent steps in the causal pathway; it may create more confounding
than it prevents. Automatic selection procedures are available in Stata (see [R] sw), but
they may seduce the user to non-thinking. I will not describe them.
• If your hypothesis is non-causal and you only look for predictors, logical requirements
are more relaxed. But make sure you really are looking at predictors, not consequences
of the outcome.
• Take care with closely associated independent variables, e.g. education and social class.
Including both may obscure more than illuminate.

12.1. Linear regression

regress [R] regress
A standard linear regression with bmi as the dependent variable:
regress bmi sex age

xi: [R] xi
The xi: prefix handles categorical variables in regression models. From a five-level
categorical variable xi: generates four indicator variables; in the regression model they are
referred to by the i. prefix to the original variable name:
xi: regress bmi sex i.agegrp
You may also use xi: to include interaction terms:
xi: regress bmi age i.sex i.treat i.treat*i.sex

By default the first (lowest) category will be omitted, i.e. be the reference group. You may,
before the analysis, select agegrp 3 to be the reference by defining a 'characteristic':
char agegrp[omit] 3

predict [R] predict

After a regression analysis you may generate predicted values from the regression
coefficients, and this may be used for studying residuals:
regress bmi sex age
predict pbmi
generate rbmi = bmi-pbmi
scatter rbmi pbmi or use rvfplot, see below

Regression diagnostics [R] Regression diagnostics

The chapter is very instructive. Get a residual plot with a horizontal reference line by:
rvfplot , yline(0)

32
12.2. Logistic regression
logistic [R] logistic
A standard logistic regression with ck as the dependent variable:
logistic ck sex smoke speed alc
The dependent variable (ck) must be coded 0/1 (no/yes). If the independent variables are also
coded 0/1 the interpretation of odds ratios is straightforward, otherwise the odds ratios must
be interpreted per unit change in the independent variable.
The xi: prefix applies as described in section 12.1:
xi: logistic ck sex i.agegrp i.smoke
xi: logistic ck i.sex i.agegrp i.smoke i.sex*i.smoke

After running logistic, use predict as described in section 12.1:

predict

After running logistic obtain Hosmer-Lemeshow's goodness-of-fit test with 10 groups:

lfit , group(10)

After running logistic obtain a classification table, including sensitivity and specificity
with a cut-off point of your choice:
lstat , cutoff(0.3)

Repeat lstat with varying cut-off points or, smarter, use lsens to see sensitivity and
specificity with varying cutoff points:
lsens

See a ROC curve:

lroc

33
13. Survival analysis and related issues
st [ST] manual
The st family of commands includes a number of facilities, described in the Survival
Analysis manual [ST]. Here I describe the stset and stsplit commands and give a few
examples. The data is cancer1.dta, a modification of the cancer.dta sample data
accompanying Stata.
The observation starts at randomization (agein), the data set includes these variables:
. summvl // summvl is a summarize displaying variable labels.
// Get it by: findit summvl
Variable Obs Mean Std.Dev Min Max Label
------------------------------------------------------------------------
lbnr 48 24.5 14 1 48 Patient ID
drug 48 1.875 .841099 1 3 Drug type (1=placebo)
drug01 48 .583333 .498224 0 1 Drug: placebo or active
agein 48 56.398 5.6763 47.0955 67.8746 Age at randomization
ageout 48 57.6896 5.45418 49.0122 68.8737 Age at death or cens.
risktime 48 1.29167 .854691 .083333 3.25 Years to death or cens.
died 48 .645833 .483321 0 1 1 if patient died

stset [ST] stset

stset declares the data in memory to be survival time (st) data. I create two versions: In
st.cancer1.dta time simply is risktime, age not taken into consideration. In
st.cancer2.dta time at risk is defined by age at entry (agein) and exit (ageout)
enabling to study and control for the effect of age.

Simple analysis – age not included

stset data with risktime as the time-of-exit variable:
. // c:\dokumenter\proj1\gen.st.cancer1.do
. use c:\dokumenter\proj1\cancer1.dta , clear
. stset risktime , failure(died==1) id(lbnr)
id: lbnr
failure event: died == 1
obs. time interval: (risktime[_n-1], risktime]
exit on or before: failure
-------------------------------------------------------------------------
48 total obs.
0 exclusions
-------------------------------------------------------------------------
48 obs. remaining, representing
48 subjects
31 failures in single failure-per-subject data
62 total analysis time at risk, at risk from t = 0
earliest observed entry t = 0
last observed exit t = 3.25

34
 . summarize
Variable | Obs Mean Std. Dev. Min Max
-------------+-----------------------------------------------------
lbnr | 48 24.5 14 1 48
.... |
risktime | 48 1.291667 .8546908 .0833333 3.25
died | 48 .6458333 .4833211 0 1
_st | 48 1 0 1 1
_d | 48 .6458333 .4833211 0 1
_t | 48 1.291667 .8546908 .0833333 3.25
_t0 | 48 0 0 0 0
. save c:\dokumenter\proj1\st.cancer1.dta

Four new variables were created, and the st'ed data set is prepared for a number of incidence
rate and survival analyses:
_st 1 if the observation includes valid survival time information, otherwise 0
_d 1 if the event occurred, otherwise 0 (censoring)
_t time or age at observation end (here: risktime)
_t0 time or age at observation start (here: 0)

Summary of time at risk and incidence rates [ST] stptime

stptime , by(drug) per(1000) dd(4) // rates x 1000, 4 decimals
stptime , at(0(1)5) by(drug) // 1-year intervals

A table of the survivor function: [ST] sts list

sts list , by(drug) compare at(0(0.5)5) // ½ year intervals

The corresponding graph: [ST] sts graph

sts graph , by(drug)
To obtain a cumulative incidence (1–S) graph displaying the interval 0-0.25 at the y-axis:
sts graph , by(drug) failure ylabel(0(0.05)0.25)

A logrank test comparing two or more groups: [ST] sts test

sts test drug

Cox proportional hazards regression analysis: [ST] stcox

stcox drug01 // drug dichotomized
xi: stcox i.drug // 3 drugs

35
Including age in the analysis
stset data with ageout as the time-of-exit variable, agein as the time-of-entry
variable:
. // c:\dokumenter\proj1\gen.st.cancer2.do
. use c:\dokumenter\proj1\cancer1.dta , clear
. stset ageout , enter(time agein) failure(died==1) id(lbnr)
. summarize
Variable | Obs Mean Std. Dev. Min Max
-------------+-----------------------------------------------------
lbnr | 48 24.5 14 1 48
... |
_st | 48 1 0 1 1
_d | 48 .6458333 .4833211 0 1
_t | 48 57.61966 5.444583 49.87939 68.70284
_t0 | 48 56.328 5.659862 47.97637 67.73915
. save c:\dokumenter\proj1\st.cancer2.dta

The sts and stcox analyses as shown above now must be interpreted as age-adjusted
(delayed entry analysis). Summary of time at risk and age-specific incidence rates:
stptime , at(45(5)70) by(drug) // 5 year age intervals

stsplit [ST] stsplit

To look at the influence of age at incidence or survival, stsplit the data, expanding each
observation to an observation for each age interval:
. // c:\dokumenter\proj1\gen.stsplit.cancer2.do
. use c:\dokumenter\proj1\st.cancer2.dta , clear
. stsplit agegr , at(45(5)70)
. summarize
Variable | Obs Mean Std. Dev. Min Max
-------------+-----------------------------------------------------
lbnr | 61 26.37705 14.03586 1 48
drug | 61 1.967213 .8557105 1 3
drug01 | 61 .6229508 .4886694 0 1
agein | 61 55.58628 5.651005 47.09552 67.87458
ageout | 61 56.87054 5.563221 49.01218 68.8737
risktime | 61 1.412568 .8856398 .0833333 3.25
died | 48 .6458333 .4833211 0 1
_st | 61 1 0 1 1
_d | 61 .5081967 .5040817 0 1
_t | 61 56.87054 5.563221 49.01218 68.8737
_t0 | 61 55.85415 5.610502 47.09552 67.87458
agegr | 61 54.01639 5.832357 45 65
. save c:\dokumenter\proj1\stsplit.cancer2.dta

The data now has 61 observations with events and risktime distributed to the proper age
intervals. Describe risktime etc. by:
bysort drug: stsum , by(agegr)

36
poisson [R] poisson
The stsplit.cancer2.dta data set above can be used for Poisson regression with a little
more preparation. died and risktime must be replaced as shown. You also may collapse
the file to a table with one observation for each age group and drug (see section 10.7):
. // c:\dokumenter\proj1\gen.stcollaps.cancer2.do
. use c:\dokumenter\proj1\stsplit.cancer2.dta , clear
. replace died = _d
. replace risktime = _t - _t0
. summarize
Variable | Obs Mean Std. Dev. Min Max
-------------+-----------------------------------------------------
.... |
risktime | 61 1.016394 .7343081 .0833321 2.75
died | 61 .5081967 .5040817 0 1
_st | 61 1 0 1 1
_d | 61 .5081967 .5040817 0 1
_t | 61 56.87054 5.563221 49.01218 68.8737
_t0 | 61 55.85415 5.610502 47.09552 67.87458
agegr | 61 54.01639 5.832357 45 65
. collapse (sum) risktime died , by(agegr drug)
. summarize
Variable | Obs Mean Std. Dev. Min Max
-------------+-----------------------------------------------------
drug | 15 2 .8451543 1 3
agegr | 15 55 7.319251 45 65
risktime | 15 4.133334 3.01067 .3581161 10.88906
died | 15 2.066667 2.374467 0 8
. save c:\dokumenter\proj1\stcollaps.cancer2.dta

These data are ready for a Poisson regression:

. xi: poisson died i.drug i.agegr if risktime>0 , exposure(risktime) irr
i.drug _Idrug_1-3 (naturally coded; _Idrug_1 omitted)
i.agegr _Iagegr_45-65 (naturally coded; _Iagegr_45 omitted)
Poisson regression Number of obs = 15
LR chi2(6) = 29.20
Prob > chi2 = 0.0001
Log likelihood = -19.558255 Pseudo R2 = 0.4274
-------------------------------------------------------------------------
died | IRR Std. Err. z P>|z| [95% Conf. Interval]
-------------+-----------------------------------------------------------
_Idrug_2 | .2125451 .1044107 -3.15 0.002 .0811534 .5566669
_Idrug_3 | .1434259 .068503 -4.07 0.000 .0562441 .3657449
_Iagegr_50 | 2.427286 2.576403 0.84 0.403 .3031284 19.43637
_Iagegr_55 | 3.892978 4.067407 1.30 0.193 .5022751 30.17327
_Iagegr_60 | 6.20448 6.644274 1.70 0.088 .7606201 50.61077
_Iagegr_65 | 11.05612 12.48911 2.13 0.033 1.208027 101.1879
risktime |(exposure)
-------------------------------------------------------------------------

After running poisson, test goodness-of-fit by:

poisgof

37
14. Graphs
14.1. Introduction
The purpose of this section is to help you understand the fundamentals of Stata 8 graphs, and
to enable you to create and modify them.
With Stata's dialogs you can easily define a graph. Once you made your choices, press
[Submit] rather than [OK]; this gives the opportunity to modify your choices after having
looked at the result.
This was the easy part. For the purpose of analysis you can do most things with the dialogs.
Look at the illustrations in this section to get some ideas of the types and names of graphs. At
www.ats.ucla.edu/stat/stata/Library/GraphExamples/default.htm you find a number of graph
examples with the commands used.
The following more complex stuff illustrates how to make graphs ready for publication.
–o–
Stata can produce high-quality graphs, suited for publication. However, the first edition of the
Graphics manual is complicated to use, to say the least; don't feel inferior if you get lost in the
maze while looking up information. The on-line help works better, once you understand the
general principles.
Use the Graphics manual to see examples of graphs, but skip the style and options specifi-
cations unless you are very dedicated.
The style of the graphs presented here is different from the manual style; I attempted to hit a
leaner mainstream style used in most scientific journals. The graphs are based upon my
schemes lean1 with a framed plot area and no gridlines and lean2 with no frame but
with gridlines. Find the schemes used by findit lean schemes. See more on this issue
under Schemes, section 14.9.
You will meet some critical remarks in this section. However:
• Stata's graphics is a very versatile system; you can create almost whatever you want,
except (fortunately) 3-D effects.
• The Stata people are very open to criticism and suggestions, and the users' input no
doubt will give inspiration to improved design, accessibility and documentation.

38
14.2. The anatomy of graphs
Figure 1 shows the most important elements of a graph. The graph area is the entire figure,
including everything, while the plot area is the central part, defined by the axes.

Title: Figure 1
Subtitle: The anatomy of a graph
40

Plot area
30
Y-axis title

20
Legend
first plot
second plot
10
2,000 3,000 4,000 5,000
X-axis title
Note: This is the outer region or background

A graph consists of several elements: Title, legend, axes, and one or more plots, e.g. two
scatterplots within the same plot area; Figure 1 includes two scatterplots.
Below is the command that generated Figure 1 (except the dashed outer frame). The elements
of the command will be explained later.
sysuse auto.dta // open auto.dta accompanying Stata
set scheme lean1
twoway (scatter mpg weight if foreign==0) ///
(scatter mpg weight if foreign==1) ///
, ///
title("Title: Figure 1") ///
subtitle("Subtitle: The anatomy of a graph") ///
ytitle("Y-axis title") xtitle("X-axis title") ///
note("Note: This is the outer region or background") ///
legend(title("Legend") , ///
label(1 "first plot") label(2 "second plot")) ///
text(35 3400 "Plot area")

14.3. The anatomy of graph commands

The overall syntax of graph commands is:
graph-command (plot-command , plot-options) (plot-command , plot-options) , graph-options

This is the syntax style generated by the dialogs, and I will stick to it.
Unfortunately the Graphics manual frequently uses another, less transparent style:
graph-command plot-command , plot-options | | plot-command , plot-options | | , graph-options

Clue: Put a | | where the standard syntax has a ) parenthesis closing a plot specification.

39
When letting the dialog generate a simple scatterplot command, the result is like this:
twoway (scatter mpg weight)
twoway defines the graph type; scatter defines a plot in the graph. You could enter the
same in the command window, but Stata also understands this short version:
scatter mpg weight

The variable list (e.g. mpg weight) in most graph commands may have one or more
dependent (y-) variables, and one independent (x-) variable, which comes last.
Graph commands may have options; as in other Stata commands a comma precedes the
options. title() is an option to the twoway graph command:
twoway (scatter mpg weight) , title("74 car makes")

Plot specifications may have options. msymbol() is an option to scatter; it is located

within the parentheses delimiting the plot specification. msymbol() lets you select the
marker symbol (a hollow circle) to be used in the scatterplot:
twoway (scatter mpg weight , msymbol(Oh))

Options may have sub-options. size() is a sub-option to the title() option; here it lets
the title text size be 80% of the default size:
twoway (scatter mpg weight) , title("74 car makes" , size(*0.8))

Warning: Options don't tolerate a space between the option keyword and the parenthesis,
like the following (□ denotes a blank character):
title□("74 car makes")
The error message may be confusing, e.g. 'Unmatched quotes' or 'Option not allowed'.

Advice: Graph commands tend to include a lot of nested parentheses, and you may make
errors (I often do). In the Do-file editor, place the cursor after an opening parenthesis and
enter [Ctrl]+B, to see the balancing closing parenthesis. In NoteTab you can use [Ctrl]+M
(match) in the same way.

14.4. Axis options

Axis lengths
Unfortunately axis lengths cannot be controlled directly, only the entire graph size. By trial
and error you may then obtain the desired axis lengths. To make a graph 3 by 4 inches:
twoway (scatter mpg weight) , ysize(3) xsize(4)

You can, however, determine the aspect ratio of the plot area (the y/x axis ratio) by the
aspect() option. To obtain a square plot area:
twoway (scatter mpg weight) , ysize(3) xsize(4) aspect(1)

40
Ticks, labels and gridlines
Stata sets reasonable ticks and labels at the axes; you may also define them yourself. The
following command sets a tick and a label for every 20 years at the x-axis, minor ticks divide
each major interval in two. The y-axis has a log scale; tick marks are defined.
twoway (line incidence year) , ///
xlabel(1900(20)2000) xmtick(##2) ///
yscale(log) ylabel(1 2 5 10 20 50 100)

You may define maximum and minimum values at the axes:

... , yscale(log range(0.8 150))

If you use the s2color, s2mono or lean2 scheme, the default is horizontal gridlines and
no vertical gridlines. To drop horizontal and include vertical gridlines (hardly a good idea in
this case):
... , xlabel( , grid) ylabel(1 2 5 10 20 50 100 , nogrid)

If you want to display decimal commas rather than periods, give the Stata command:
set dp comma

Plotregion margin
By default twoway graphs include a margin between the extreme plot values and the axes, to
avoid symbols touching axes. If you want a zero margin – as in the twoway line plot,
section 14.7 – include:
... , plotregion(margin(zero))

14.5. Placing graph elements

The placement of graph elements, e.g. the legend, is defined by location relative to the plot
area (ring position) and a direction (clock position). The placement of elements in Figure 1
was determined by the scheme applied (see section 14.9); the placements were:
Element Ring position Clock position Position can
ring() pos() be modified
Plot area 0 ... No
Y-axis title 1 9 No
X-axis title 1 6 No
Subtitle 6 12 Yes
Title 7 12 Yes
Legend 3 4 Yes
Note 4 7 Yes

The twoway line plot, section 14.7, illustrates an alternative placement of the legend:
... , legend(label(1 "Males") label(2 "Females") ring(0) pos(8))

41
A text block is placed in the plot area by giving its y and x coordinates; place(c) (the
default) means that the coordinates apply to the center of the text block; place(se) that
they apply to the block's southeast corner. See example in Figure 1 and the twoway line
plot, section 14.7:
... , text(90 69 "1999-2000")

14.6. Appearance of markers, lines, etc.

Options and their arguments for defining the appearance of lines, bars and markers:
Lines Markers
Element Color Pattern Width Symbol Size
Legend etc. color()
fill fcolor()
outline lcolor() lpattern() lwidth()
Bars, areas bcolor()
fill bfcolor()
outline blcolor() blpattern() blwidth()
Markers mcolor() msymbol() msize()
fill mfcolor()
outline mlcolor() mlwidth()
Connecting
lines clcolor() clpattern() clwidth()

Arguments: none blank none i invisible *0.7 70%

l or solid *1.3 O circle of default
Grayscale: - or dash 130% of D diamond
black _ or longdash S square
default
gs0 (black) T triangle
.. shortdash p point
dot
gs16 (white) dash_dot
+ plus
white X cross
Formula, e.g. Small:
"-." o d s t x
"--.."
Hollow: Oh

Marker symbols
Markers are defined by symbol: msymbol(), outline colour: mlcolor(), fill colour
mfcolor() and size msize(). To define a hollow circle:
twoway (scatter mpg weight , msymbol(Oh))

A hollow circle (Oh) is transparent. Obtain a circle with a non-transparent white fill by:
twoway (scatter mpg weight , msymbol(O) mfcolor(white))

Connecting lines
The twoway line and twoway connected examples, section 14.7, use connecting
lines; here the clpattern() and clwidth() options apply:
twoway (line m1840-k1999 age , clpattern( - l - l – l ))

42
The default connect-style is a straight line. Obtain a step-curve like a Kaplan-Meier plot by:
twoway (line cum time , connect(J))

Bars
Bar graphs (twoway bar) and range plots use bar outlines; here the blpattern() and
blwidth() options apply. The colour of the bar fill is defined by the bfcolor() option:
... , bar(1, bfcolor(gs9)) bar(2, bfcolor(gs14))

14.7. Examples
On the following pages you find illustrations of some important graph types, including the
commands that generated the graphs. The appearance is different from the manual's graphs; it
was determined by my schemes lean1 and lean2, described in section 14.9.

For each graph you see the do-file that made it, including the data for the graph or a use
command. I suggest letting do-files generating graphs always start with a gph. prefix, for
easy identification.
In the illustrations I reduced the graph size by the xsize() and ysize() options. This,
however, leads to too small text and symbols, and I enlarged them by the scale() option.

twoway graphs have continuous x- and y-axes. Many plot-types fit in twoway graphs;
exceptions are graph bar, graph box and graph pie.

histogram

80

60
N of children

40

20

0
1000 1500 2000 2500 3000 3500 4000 4500 5000
Birthweight, grams

A histogram depicts the distribution of a continuous variable. The y-axis may reflect a count
(frequency), a density or a percentage; the corresponding normal curve may be overlaid.
Histograms are documented in [R] histogram and in [G] graph twoway histogram.

43
// c:\dokumenter\...\gph.birthweight.do
use "C:\dokumenter\...\newborns.dta" , clear
set scheme lean2
histogram bweight ///
, ///
frequency ///
normal ///
start(750) width(250) ///
xlabel(1000(500)5000) ///
xmticks(##2) ///
xtitle("Birthweight, grams") ///
ytitle("N of children") ///
plotregion(margin(b=0)) ///
xsize(4) ysize(2.3) scale(1.4)

graph bar

10
Prevalence (per cent)

2 Males
Females
0
16-24 25-44 45-66 67-79 80+
Age

// c:\dokumenter\...\gph.diabetes prevalence.do
clear
input str5 age m f
16-24 .9 .2
25-44 .8 .8
45-66 3.8 2.9
67-79 8.2 5.4
80+ 9.1 7.2
end
set scheme lean2
graph bar m f ///
, ///
over(age) ///
b1title("Age") ///
ytitle("Prevalence (per cent)") ///
legend( label(1 "Males") label(2 "Females") ) ///
xsize(4) ysize(2.3) scale(1.4)

For some reason the xtitle() option is not valid for bar graphs. To generate an x-axis title
you may, however, use b1title() instead.

Bar fill colours are assigned automatically according to the scheme. This option would
generate a very dark fill for females:
... , bar(2 , bfcolor(gs3))

44
In bar graphs the x-axis is categorical, the y-axis continuous. In the example variables m and
f defined the heights of the bars, but actually graph bar used the default mean function,
as if the command were (with one observation per bar the result is the same):
graph bar (mean) m f , over(age)

With the auto.dta data you could generate bars for the number of domestic and foreign
cars by:
graph bar (count) mpg , over(foreign)
Actually what is counted is the number of non-missing values of mpg.

Bar graphs are documented in [G] graph bar and [G] graph twoway bar.

twoway scatter

40
Mileage (mpg)

30

20

Domestic
Foreign
10
2,000 3,000 4,000 5,000
Weight (lbs.)

// c:\dokumenter\...\gph.mpg_weight.do
clear
sysuse auto
set scheme lean2
twoway ///
(scatter mpg weight if foreign==0, msymbol(Oh)) ///
(scatter mpg weight if foreign==1, msymbol(O)) ///
, ///
legend(label(1 "Domestic") label(2 "Foreign")) ///
xsize(4) ysize(2.3) scale(1.4)

Twoway graphs have continuous x- and y-axes; scatter is the "mother" of twoway graphs.
A graph with one plot has no legend; this one with two plots has. The default legend texts
often need to be replaced by short, distinct texts, like here. Since the xtitle() and
ytitle() options were not specified, Stata used the variable labels as axis titles.

msymbol may be abbreviated to ms; I chose to avoid abbreviations for readability; it is

much more important that commands are easy to read than easy to write. In this case the
msymbol() options were not necessary since (Oh) and (O) are the default first symbols
under the lean schemes.
The graph displays the same data as the initial Figure 1, this time using the scheme lean2:
The plot area has horizontal grid-lines, but no frame.

45
twoway line
100
1999-2000
80 1901-05
Per cent surviving

60
1840-49
40

20
Females
Males
0
0 20 40 60 80 100
Age

// c:\dokumenter\...\gph.DKsurvival.do
use c:\dokumenter\...\DKsurvival.dta , clear
sort age // Data must be sorted by the x-axis variable
list in 1/3, clean // List to show the data structure
age m1840 k1840 m1901 k1901 m1999 k1999
1. 0 100.00 100.00 100.00 100.00 100.00 100.00
2. 1 84.47 86.76 86.93 89.59 99.16 99.37
3. 2 80.58 83.11 85.22 87.89 99.08 99.32
set scheme lean1
twoway ///
(line m1840-k1999 age , clpattern( - l - l – l )) ///
, ///
plotregion(margin(zero)) ///
xtitle("Age") ///
ytitle("Per cent surviving") ///
legend(label(1 "Males") label(2 "Females") order(2 1) ///
ring(0) pos(8)) ///
text(91 72 "1999-2000") ///
text(77 48 "1901-05") ///
text(49 40 "1840-49") ///
xsize(3.3) ysize(2.3) scale(1.4)

A line plot is a variation of scatterplot without markers, but with connecting lines. This graph
includes six line plots, required by one plot-specification with six y- and one x-variable. The
clpattern() option defines the six connected-line patterns.

Make sure data are sorted according to the x-axis variable; otherwise the result is nonsense.
The example shows how to include text in a graph and how to position the legend within the
plot area (see section 14.5 on placement of graph elements).
Twoway graphs by default include "empty" space between the axes and the extreme plot
values. The graph option plotregion(margin(zero)) lets the plot start right at the axes.

46
twoway connected; twoway rcap

100

80
Mean score

60

40 Observed
Expected
95% CI
20
PF RP BP GH VT SF RE MH
SF-36 subscale

// c:\dokumenter\...\gph.SF36a.do
clear
input scale n obs sd norm
1 139 60.81 27.35 70.77
2 139 37.65 42.06 62.01
...
8 139 73.06 21.54 79.99
end
generate se=sd/sqrt(n)
generate ci1=obs+1.96*se
generate ci2=obs-1.96*se
label define scale 1 "PF" 2 "RP" 3 "BP" 4 "GH" 5 "VT" 6 "SF" 7 "RE" 8 "MH"
label values scale scale
set scheme lean1
twoway ///
(connected obs scale , msymbol(O) clpattern(l)) ///
(connected norm scale , msymbol(O) mfcolor(white) clpattern(-)) ///
(rcap ci1 ci2 scale) ///
, ///
ytitle("Mean score") ///
xtitle("SF-36 subscale") ///
xlabel(1(1)8 , valuelabel noticks) ///
xscale(range(0.5 8.5)) ///
legend(label(1 "Observed") label(2 "Expected") label(3 "95% CI")) ///
xsize(4) ysize(2.3) scale(1.4)

This graph includes three plots; two connected and one rcap. In twoway plots both axes
are continuous, so you could not have a categorical variable (PF, RP etc.) at the x-axis.
Solution: use a numerical variable and use value labels to indicate the meaning. This graph
style is frequently used to present SF-36 results, although connecting lines may be illogical
when displaying eight qualitatively different scales.
xscale(range(0.5 8.5)) increased the distance between plot symbols and plot margin.

rcap does not calculate confidence intervals for you; it is up to you to provide two y- and
one x-value for each confidence interval. rspike would have plotted intervals without caps.

47
twoway rspike

Cross-sectional study

// c:\dokumenter\...\gph.length_bias.do
clear
set obs 20
gen x=_n
gen y1=x
gen y2=y1+2
replace y2=y1+8 if mod(x,2)==0
set scheme lean2
twoway (rspike y1 y2 x , horizontal blwidth(*1.5)) ///
, ///
yscale(off) ylabel(, nogrid) ytitle("") ///
xlabel(none) xtitle("Cross-sectional study") ///
xline(14.5) ///
xsize(3.7) ysize(2.3) scale(1.4)

The purpose of this graph is to illustrate length bias: a cross-sectional (prevalence) study may
mislead you. Cases with short duration (due to successful treatment or high case fatality) are
underrepresented in a cross-sectional sample.
rspike is in the twoway r* family: range plots, like rcap shown before; this time it is
horizontal.
In range plots and droplines (next page) the lines technically are bar outlines, and options are
blcolor(), blpattern() etc.; hence the blwidth(*1.5) to make the spikes wider
than the default.
It is easy to create one or more reference lines; use xline() and yline().

48
twoway dropline

20
Patient number

15

10

5
Deaths
Censorings
1
0 1 2 3 4 5 6 7
Years after diagnosis

// c:\dokumenter\...\gph.obstime.do
use "c:\dokumenter\...\cohort1.dta" , clear
list patient time died in 1/5 , clean
patient time died
1. 1 0.578 1
2. 2 0.867 1
3. 3 1.235 1
4. 4 1.374 0
5. 5 1.437 1
set scheme lean2
twoway ///
(dropline time patient if died==1, horizontal msymbol(S)) ///
(dropline time patient if died==0, horizontal ///
msymbol(S) mfcolor(white)) ///
, ///
plotregion(margin(zero)) ///
ytitle("Patient number") ///
yscale(range(0 22)) ///
ylabel(1 5 10 15 20 , nogrid) ///
xtitle("Years after diagnosis") ///
xlabel(0(1)7) ///
legend(label(1 "Deaths") label(2 "Censorings") ring(0)) ///
xsize(3.7) ysize(2.5) scale(1.3)

In a dropline plot a line 'drops' from a marker perpendicularly to the x- or y-axis. Droplines
technically are bar outlines, like range plots, and their appearance is controlled by
blpattern(), blcolor() and blwidth().

The marker for censorings is a square with white fill, not a hollow square, to avoid the
dropline to be visible within the marker.

49
twoway function

-3.00 -1.96 -1.00 0.00 1.00 1.96 3.00

Standard deviations from mean

// c:\dokumenter\...\gph.normal.do
set scheme lean2
twoway ///
(function y=normden(x) , range(-3.5 3.5) ///
droplines(-1.96 -1 0 1 1.96)) ///
, ///
plotregion(margin(zero)) ///
yscale(off) ylabel(, nogrid) ///
xlabel(-3 -1.96 -1 0 1 1.96 3 , format(%4.2f)) ///
xtitle("Standard deviations from mean") ///
xsize(3) ysize(2.3) scale(1.4)

twoway function gives you the opportunity to visualize any mathematical function. The
result has no relation to the actual data in memory. The range() option is necessary; it
defines the x-axis range.
Other examples:
An identity line, to be overlaid in a scatterplot comparing two measurements:
twoway ///
(scatter sbp2 sbp1) ///
(function y=x , range(sbp1))

A parabola:
twoway (function y=x^2 , range(-2 2))

50
graph matrix

graph matrix
10 20 30 40 150 200 250

15,000

Price 10,000

5,000
40

30 Mileage
20 (mpg)

10
5,000
4,000
Weight
(lbs.) 3,000
2,000
250

200 Length
(in.)
150
5,000 10,000 15,000 2,0003,0004,0005,000

// c:\dokumenter\...\gph.matrix.do
sysuse auto , clear
set scheme lean1
graph matrix price mpg weight length ///
, ///
title(graph matrix) ///
mlwidth(*0.7) ///
xsize(5) ysize(4)

Matrix scatterplots are useful for analysis, but are infrequently used for publication.
The lean1 and lean2 schemes by default use a small hollow circle as marker in matrix
scatterplots. Here mlwidth(*0.7) made the marker outline thinner.

The upper right cells are redundant rotated images of the lower left cells; omit them by:
graph matrix price mpg weight length , half

51
14.8. Saving, displaying and printing graphs

Save a graph
The active graph can be saved as a .gph file:
graph save "c:\dokumenter\...\DKsurvival.gph" [, asis replace]
The asis option saves a 'frozen' graph, it is displayed as is, regardless of scheme settings.
Without this option you save a 'live' graph: you may display it again, maybe using a different
scheme or modifying its size. The manual states that you may edit it, but that is not the case.
My firm recommendation:
Rarely save graph files; always save a do-file for each graph with a name that tells what it
does, e.g. gph.DKsurvival.do. Let all graph-defining do-files start with a gph. prefix,
for easy identification. The do-file documents what you did, you can edit it to modify the
graph, and you can modify the do-file to create another graph. Remember to include the data
or a use command reading the data used. This advice also applies when you initially
defined the graph command with a graph dialog.

Open a saved graph

A saved graph may be displayed by:
graph use "c:\dokumenter\...\DKsurvival.gph"

Display and print a graph

Re-display the current graph by:
graph display [, scale(1.2) ysize(3) xsize(5) scheme(lean2)]

The scale() option is useful to increase marker and text size, e.g. for a slide show.
xsize() and ysize() modify the size of the graph area (arguments in inches), and
scheme() lets you display a graph under a different scheme – but that sometimes fails.

Copying and printing 'smooth' coloured or gray areas sometimes give poor results, and a
raster-pattern is preferable. This is a printer, not a Stata issue; in this respect modern printers
are worse than older. At my old HP LaserJet 1100 printer the LaserJet III printing mode
translates gray areas to raster-patterns, copying and printing nicely. You may need to
experiment.
If you in the future don't want Stata's logo being printed on each graph:
graph set print logo off

Copy a graph to a document

Copy-and-paste a graph to another document using the standard [Cltr]+C and [Ctrl]+V proce-
dure. The graph will be transferred as a metafile; there are two types, which you may select
via the Graph Preferences:
Prefs ► Graph Preferences ► Clipboard

52
Select Enhanced Metafile (EMF) or Windows Metafile (WMF); which one works best depends
on your system and printer; take a critical look at the results.

Submitting graphs to journals etc.

The requirements of journals vary, but the best results are probably obtained by Windows
metafiles (.wmf, .emf) and Encapsulated PostScript (.eps). When saving a graph, select that
format. You may open an eps-file by Adobe Acrobat Reader to ensure that the result is
satisfactory. Formats like Portable Network Graphics (.png) and TIFF (.tif) give
unsatisfactory results.
For updated information on exporting graphs, see:
whelp graph_export

14.9. Schemes: Default appearance

A scheme is a collection of default styles; the Graphics Manual uses a scheme close to
s2mono, while the Base Reference Manual uses s1mono. s2mono is itself a modification
of the s2color scheme.

This note uses the schemes lean1 and lean2; they are modifications to s1mono and
s2mono. Most scientific journals use a lean graph style – or at least require that graphs
submitted are lean and black-and-white. If you are interested, download and install both
schemes (use the command findit lean schemes).3

The difference between the two is that lean1 has a framed plot area, but no gridlines, while
the opposite is the case for lean2. Section 14.7 includes examples using both schemes.

To select a scheme:
set scheme lean2

To make Stata remember lean1 as your default scheme:

set scheme lean1 , permanently

To create a scheme with your own preferences use Stata's do-file editor or another text editor
to enter the options you want in your own personal scheme (e.g. myscheme) and save it as
c:\ado\personal\myscheme.scheme. Scheme terminology differs from graph command
terminology; documentation is forthcoming.3

3
Juul S. Lean mainstream schemes for Stata 8 graphics. The Stata Journal 2003; 3: 295-301.
53
15. Miscellaneous
15.1. Memory considerations [U] 7
In Intercooled Stata a data set can have a maximum of 2,000 variables (Stata/SE: 32,000).
Stata keeps the entire data set in memory, and the number of observations is limited by the
memory allocated. The memory must be allocated before you open (use) a data set.

As described in section 1.2 the initial default memory is defined in the Stata icon. To change
this to 15 MB, right-click the icon, select Properties, and change the path field text to:
c:\stata\wstata.exe /m15.

If the memory allocated is insufficient you get the message:

no room to add more observations
You may increase the current memory to 25 MB by:
clear // You can't change memory size with data in memory
set memory 25m
You can see the amount of used and free memory by:
memory

compress [R] compress

To reduce the physical size of your data set – and the memory requirements – Stata can
determine the fewest bytes needed for each variable (see section 6.2), and you can safely:
compress
and save the data again (save... , replace). This may reduce the memory need by 80%.

Handling huge data sets

If you are regularly handling huge data sets you may consider:
• to use compress to reduce memory requirements
• to increase your computer's RAM (not expensive)
For instance with 512 Mb RAM you could set memory 400m for Stata, and this would fit
very large data sets, e.g. a million observations with 100 variables. Stata/SE can handle up to
32,000 variables, but memory restrictions otherwise don't differ from Intercooled Stata. SAS
or SPSS might be alternatives.
You might not be able to create the entire Stata data set because of its hugeness. Try to read
one part of the data, compress and save, read the next part of the data, compress and save,
etc., and finally combine (append) the partial data sets (see section 10.6):
// c:\dokumenter\p1\gen.aa.do
infix id 1 type 2 sold 4-7 using c:\dokumenter\p1\aa.txt in 1/10000
compress
save c:\tmp\aa1.dta
infix id 1 type 2 sold 4-7 using c:\dokumenter\p1\aa.txt in 10001/20000
compress
append using c:\tmp\aa1.dta
save c:\dokumenter\p1\aa.dta

54
15.2. String variables [U] 15.4; [U] 26
Throughout this text I have demonstrated the use of numeric variables, but Stata also handles
string (text) variables. It is almost always easier and more flexible to use numeric variables,
but sometimes you might need string variables. String values must be enclosed in quotes:
replace ph=45 if nation == "Danish"
"Danish", "danish", and "DANISH" are different string values.

A string can include any character, also numbers; however number strings are not interpreted
by their numeric value, just as a sequence of characters. Strings are sorted in dictionary
sequence, however all uppercase letters come before lowercase; numbers come before letters.
This principle is also applies to relations: "12" < "2" < "A" < "AA" < "Z" < "a".

String formats [U] 15.5.5

%10s displays a 10 character string, right-justified; %-10s displays it left-justified.

Reading string variables into Stata

In the commands reading ASCII data (see section 8) the default data type is numeric. String
variables should be defined in the input command. str5 means a 5 character text string:
infix id 1-4 str5 icd10 5-9 using c:\dokumenter\p1\a.txt

Generating new string variables

The first time a string variable is defined it must be declared by its length (str10):
generate str10 nation = "Danish" if ph==45
replace nation = "Swedish" if ph==46

Conversion between string and numeric variables

Number strings to numbers
If a CPR number is recorded in cprstr (type string), no calculations can be performed.
Conversion to a numeric variable cprnum can be obtained by:
generate double cprnum = real(cprstr)
format cprnum %10.0f
cprnum is a 10 digit number and must be declared double for sufficient precision (see
section 6.2). Another option is destring (it automatically declares cprnum double):
destring cprstr , generate(cprnum)

Non-number strings to numbers

If a string variable sex is coded as eg. "M" and "F", convert to a numeric variable
gender (with the original string codes as value labels) by:
encode sex , generate(gender) [R] encode

Display the meaning of the numeric codes by:

label list gender

55
Numbers to strings
You want the numeric variable cprnum converted to a string variable cprstr:
generate str10 cprstr = string(cprnum , "%10.0f")

String manipulations [U] 16.3.5; [GSW] 12

Strings can be combined by +:
generate str5 svar5 = svar3 + svar2

You may isolate part of a string variable by the substr function. The arguments are: source
string, start position, length. In the following a3 will be characters 2 to 4 of strvar:
generate str3 a3 = substr(strvar,2,3)

You may substitute characters within a string. In an example above the string variable
cprstr was created from the numeric variable cprnum. However, for persons with a
leading 0 in the CPR number the string will start with a blank, not a 0. This can be
remedied by:
replace cprstr = subinstr(cprstr," ","0",1)

The upper function converts lower case to upper case characters; the lower function does
the opposite. Imagine that ICD-10 codes had been entered inconsistently, the same code
somtimes as E10.1, sometimes as e10.1. These are different strings, and you want them to
be the same (E10.1):
replace icd10 = upper(icd10)

Handling complex strings, eg. ICD-10 codes [U] 26.4

In the ICD-10 classification of diseases all codes are a combination of letters and numbers
(e.g. E10.1 for insulin demanding diabetes with ketoacidosis). This is probably convenient for
the person coding diagnoses (an extremely important consideration), but for the data handling
it is quite inconvenient. I suggest to split a 5 character ICD-10 string variable (scode) into a
one character string variable (scode1) and a four digit numeric variable (ncode4):
generate str1 scode1 = substr(scode,1,1)
generate ncode4 = real(substr(scode,2,4))
format ncode4 %4.1f

What did we obtain? Two variables: the string variable scode1 with 26 values (A to Z) and
a numeric variable ncode4 (0.0-99.9). Now identify diabetes (E10.0-E14.9) by:
generate diab=0
replace diab=1 if scode1=="E" & ncode4>=10 & ncode4<15

If you received ASCII data, the same result could have been obtained by letting eg. the infix
command read the same data twice as different types:
infix id 1-4 str5 scode 5-9 str1 scode1 5 ncode2 6-9 ///
using c:\dokumenter\...\list1.txt

The diabetics could also have been identified this way:

replace diab=1 if scode>="E10" & scode<"E15"

56
15.3. Dates. Danish CPR numbers

Date variables [U] 15.5.2

Dates are numeric variables; the internal value is the number of days since 1 Jan 1960; dates
before that are negative.

Date formats [U] 15.5.3; [U] 27.2.3

Format specifications start with %d. Specifying %d only is equivalent to %dDlCY
displaying a date as 28sep2000. To display this date as 28.09.2000 specify the format
%dD.N.CY (D for day, N for numeric month, C for century, Y for two-digit year). Example:
format bdate %dD.N.CY

Reading date variables [U] 27.2.1

A date may be input as three variables: day, month, year and next transformed to a date
variable:
infix bd 1-2 bm 3-4 by 5-8 using c:\dokumenter\p1\datefile.txt
generate bdate = mdy(bm,bd,by) // sequence must be m d y
format bdate %dD.N.CY

Another option is to enter the date as a string (sbdate) and translate it to a date variable:
infix str10 sbdate 1-10 using c:\dokumenter\p1\datefile.txt
generate bdate = date(sbdate,"dmy") // "dmy" defines sequence
format bdate %dD.N.CY
The date function 'understands' most input formats: 17jan2001, 17/1/2001,
17.1.2001, 17 01 2001, but not 17012001. However todate, a user-written function,
handles this situation; find and download it by: findit todate.

In general: enter 4-digit years to avoid ambiguity on the century.

Calculations with dates [U] 27.2.4

To express the length of a time interval in years you must:
generate opage = (opdate-bdate)/365.25

You may extract day, month and year from a date variable (bdate):
generate bday = day(bdate)
gen bmonth = month(bdate)
gen byear = year(bdate)

On Danish CPR numbers: extracting key information

Sometimes you get date information as a CPR number in an ASCII file. You can read the the
CPR number as one string variable and the date information from the same columns:
infix str10 cprstr 1-10 bday 1-2 bmon 3-4 byear 5-6 ///
control 7-10 using c:\dokumenter\p1\datefile.txt
generate bdate = mdy(bmon,bday,byear)

57
Or you can extract key information from a CPR number read as one string variable (cprstr):
generate bday = real(substr(cprstr,1,2))
gen bmon = real(substr(cprstr,3,2))
gen byear = real(substr(cprstr,5,2))
gen control = real(substr(cprstr,7,4))
gen pos7 = real(substr(cprstr,7,1)) // to find century

Before creating bdate you must decide the century of birth; see the rules below:
generate century = 19
replace century = 20 if pos7 >= 4 & byear <= 36
replace century = 18 if pos7 >= 5 & pos7 <= 8 & byear >= 58
replace byear = 100*century + byear
generate bdate = mdy(bmon,bday,byear)

The information on sex can be extracted from control; the mod function calculates the
remainder after division by 2 (male=1, female=0):
generate sex = mod(control,2)

Century information in Danish CPR numbers

The 7th digit (the first control digit) informs on the century of birth:
Pos. 5-6 (year of birth)
Pos. 7 00-36 37-57 58-99
0-3 19xx 19xx 19xx
4, 9 20xx 19xx 19xx
5-8 20xx not used 18xx
Source: www.cpr.dk

Validation of Danish CPR numbers

To do the modulus 11 test for Danish CPR numbers first multiply the digits by 4, 3, 2, 7, 6, 5,
4, 3, 2, 1; next sum these products; finally check whether the sum can be divided by 11.
Assume that the CPR numbers were split into 10 one-digit numbers c1-c10. Explanation of
for : see section 7.
generate test=0
for C in varlist c1-c10 \ X in numlist 4/2 7/1 : ///
replace test=test+C*X
replace test=mod(test,11) // Remainder after division by 11
list id cpr test if test !=0

To extract c1-c10 from the string cprstr:

for C in newlist c1-c10 \ X in numlist 1/10 : ///
gen C=real(substr(cprstr,X,1))

To extract c1-c10 already when reading data:

infix str10 cprstr 1-10 c1-c10 1-10 using c:\...\dfile.txt

I developed an ado-file (cprcheck.ado) that extracts birth date and sex information and checks
the validity of a CPR number. Find and download it by:
findit cprcheck

58
15.4. Random samples, simulations

Random number functions [R] Functions

Stata can create 'pseudo-random' numbers:
gen y=uniform() Uniformly distributed in the interval 0-1
gen y=invnorm(uniform()) Normal distribution, mean=0, SD=1
gen y=10+2*invnorm(uniform()) Normal distribution, mean=10, SD=2
If you run the same command twice it will yield different numbers. If you need to reproduce
the same series of 'random' numbers, initialize the seed (a large integer used for the initial
calculations):
set seed 654321

Random samples and randomization

You may use sample to select a random sample of your data set:
sample 10 Selects an approximately 10 percent random sample.
sample 53 , count Selects exactly 53 observations at random.
You may assign observations randomly to two treatments:
generate y=uniform()
generate treat=1
replace treat=2 if y>0.5

And you may sort your observations in random sequence:

generate y=uniform()
sort y

Generating artifical data sets

You may use set obs to create empty observations. The following sequence defines a file
with 10,000 observations, used to study the behaviour of the difference (dif) between two
measurements (x1, x2), given information about components of variance (sdwithin,
sdbetw).
set obs 10000
generate sdbetw = 20
generate sdwithin = 10
generate sdtotal = sqrt(sdbetw^2 + sdwithin^2)
generate x0 = 50 + sdbetw*invnorm(uniform())
generate x1 = x0 + sdwithin*invnorm(uniform())
generate x2 = x0 + sdwithin*invnorm(uniform())
generate dif = x2 - x1
summarize

See another example in section 14.7 (the twoway rspike graph)

Advanced simulations [R] simulate

With simulate you may set up quite complex Monte Carlo simulations.

59
15.5. Immediate commands [U] 22
An 'immediate' command requires tabular or aggregated input; data in memory are not
affected. The immediate commands tabi, cci, csi, iri and ttesti are mentioned in
section 11, and sampsi (sample size estimation) in section 15.6.

Confidence intervals [R] ci

The general command ci and the 'immediate command cii calculate confidence intervals.
I here show the use of cii:

Normal distribution: cii 372 37.58 16.51

N mean SD

Binomial distribution: cii 153 40

N events

Poisson distribution: cii 247.1 40 , poisson

time events

Stata as a pocket calculator [R] display

The display command gives the opportunity to perform calculations not affecting the data
in memory (_pi is a Stata constant):
. display 2*_pi*7
43.982297

You may include an explanatory text:

. display "The circumference of a circle with radius 7 is " 2*_pi*7
The circumference of a circle with radius 7 is 43.982297

60
15.6. Sample size and study power
sampsi [R] sampsi
Sample size and study power estimation are pre-study activities: What are the consequences
of different decisions and assumptions for sample size and study power?

You must make these decisions:

• The desired significance level (α). Default: 0.05.
• The minimum relevant contrast – expressed as study group means or proportions.
• Sample size estimation: The desired power (1–β). Default: 0.90.
• Power estimation: Sample sizes.
And with comparison of means you must make an assumption:
• The assumed standard deviation in each sample.

Here are short examples for the four main scenarios:

Comparison of: Sample size estimation Power estimation
Proportions sampsi 0.4 0.5 sampsi 0.4 0.5 , n(60)

Means sampsi 50 60 , sd(8) sampsi 50 60 , sd(8) n(60)

Further options are available:

Sample size Power
Situation Option
prop. mean prop. mean
Significance level; default: 0.05 alpha(0.01) + + + +
Power; default: 0.90 power(0.95) + +
Unequal sample sizes; ratio=n2/n1 ratio(2) + +
Unequal sample sizes n1(40) n2(80) + +
Unequal SDs sd1(6) sd2(9) + +

Example: Sample size estimation for comparison of means, unequal SDs and sample sizes:
sampsi 50 60 , sd1(14) sd2(10) ratio(2)

sampsi also handles trials with repeated measurements, see [R] sampsi.

61
15.7. ado-files [U] 20-21, [P] (Programming manual)
An ado-file is a program. Most users will never write programs themselves, but just use
existing programs. If you are a freak, read more in the User's Guide ([U] 20-21) and the
programming manual [P]. Save user-written programs in c:\ado\personal. To see the
locations of all ado-files issue the command sysdir.

The simplest form of an .ado file is a single command or a do-file with a leading program
define command and a terminating end command. There must be a new line after the
terminating end.

Here is an example to demonstrate that creating your own commands is not that impossible.

datetime displays date and time

program define datetime
// c:\ado\personal\datetime.ado. Displays date and time.
display " $S_DATE $S_TIME "
end

Just enter datetime in the command window, and the date and time is displayed:
. datetime
9 Feb 2003 16:54:15

Two ado-files useful for the interaction between Stata and NoteTab are shown in appendix 3.

foreach and forvalues [P] foreach; forvalues

These commands are documented in the programming manual, and in the online help (whelp
foreach). Also see the FAQ www.stata.com/support/faqs/data/foreach.html. They enable
you to repeat a command for a number of variables or values. The commands can be used not
only in ado-files, but also in do-files and even interactively. Look at the sequence in section
11.1:
foreach Q of varlist q1-q10 {
tabulate `Q' sex
}

Q is a local macro (see [U] 21.3); foreach defines it as a stand-in for the variables q1 to
q10, and the sequence generates ten tabulate commands. The local macro is in effect
only within the braces {} which must be placed as shown.

When referring to the local macro Q it must be enclosed in single quotes: `Q'. In the
manuals single quotes are shown differently; but the opening quote is ` (accent grave), and
the ending quote the simple '.

62
15.8. Exchange of data with other programs
Beware: Translation between programs may go wrong, and you should check carefully eg. by
comparing the output from SPSS' DESCRIPTIVES and Stata's summarize. Especially
compare the number of valid values for each variable and take care with missing values and
date variables.

StatTransfer [U] 24.4

StatTransfer is a reasonably priced program (purchase: see Appendix 1) that translates
between a number of statistical packages, including Stata. Variable names, and variable and
value labels are transferred too. StatTransfer 7 understands Stata 8 files, but StatTransfer 6
does not. To create a Stata 7 data set for conversion by StatTransfer 6:
saveold c:\dokumenter\proj1\alfa.dta

Transferring data to Excel and other spreadsheets [R] outsheet

Many statistical packages read Excel data. To create a tab-separated file (see section 8)
readable by Excel:
outsheet [varlist] using c:\dokumenter\proj1\alfa.txt , nolabel
In Excel open the file as a text file and follow the instructions. Variable names, but no labels
are transferred.
[R] outfile
If you want the data written to a comma-separated ASCII file the command is:
outfile [varlist] using c:\...\alfa.txt , nolabel comma

Reading Excel data [R] insheet

Many packages can create Excel data, and probably all can create text files similar to those
created by Stata's outsheet command. From Excel save the file as a tab-separated text file
(see section 8). Stata reads it by:
insheet using c:\dokumenter\p1\a.txt , tab

15.9. For old SPSS users

SPSS and Stata have similarities and differences. Among the differences are:
• While SPSS can define any numeric code as a missing value, Stata's user-defined
missing values are special codes; see section 6.3.
• Stata's missing values are high-end numbers. This may complicate conditions; see
section 6.3.
• While SPSS executes all transformation commands up to a procedure command one
case at a time, Stata performs each command for the entire data set before proceeding to
the next command. This leads to different behaviour when combining selections (keep
if; drop if) with observation numbers ([_n]).

63
Frequently used SPSS commands and the similar Stata commands
SPSS command Similar Stata command
Data in and out
DATA LIST infile; infix; insheet
GET FILE use
SAVE OUTFILE save

Documentation commands etc.

VARIABLE LABELS label variable
VALUE LABELS label define followed by
label values
FORMAT sex (F1.0). format sex %1.0f
MISSING VALUES Missing values are special; see section 6.3.
COMMENT; * * or //
DOCUMENT note
DISPLAY DICTIONARY describe; codebook

Calculations
COMPUTE generate; replace; egen
IF (sex=1) y=2. generate y=2 if sex==1
RECODE a (5 thru 9=5) INTO agr. recode a (5/9=5) , generate(agr)
DO REPEAT ... END REPEAT for; foreach; forvalues
SELECT IF keep if; drop if
TEMPORARY. command if sex==1
SELECT IF (sex=1).
SAMPLE 0.1. sample 10
SPLIT FILE by...:
WEIGHT Weights can be included in most commands; see
section 7.
Analysis
DESCRIPTIVES summarize
FREQUENCIES tabulate; tab1
CROSSTABS tabulate; tab2
MEANS bmi BY agegrp. oneway bmi agegrp , tabulate
T-TEST ttest
LIST list
WRITE outfile; outsheet

Advanced
SORT CASES BY sort
AGGREGATE collapse
ADD FILES append
MATCH FILES merge

64
16. Do-file examples
Here follow short examples of do-files doing typical things. Find more examples in Take
good care of your data. All major work should be done with do-files rather than by entering
single commands because:
1. The do-file serves as documentation for what you did.
2. If you discover an error, you can easily correct the do-file and re-run it.
3. You are certain that commands are executed in the sequence intended.
Example 1 generates the first Stata version of the data, and example 2 generates a modified
version. I call both do-files vital in the sense that they document modifications to the data.
Such do-files are part of the documentation and they should be stored safely. Safe storage also
means safe retrieval, and they should have names telling what they do. My principle is this:
In example 1 gen.wine.do generates wine.dta. In example 2 gen.visit12a.do
generates visit12a.dta. This is different from example 3 where no new data are
generated, only output. This do-file is not vital in the same sense as example 1 and 2, and it
should not have the gen. prefix (the Never Cry Wolf principle).

As mentioned in section 3 I prefer to use NoteTab rather than the do-file editor for creating
do-files. The last command in a do-file must be terminated by a carriage return; otherwise
Stata cannot 'see' the command.
Example 1. gen.wine.do generates Stata data set wine.dta from ASCII file
// gen.wine.do creates wine.dta 13.5.2001
infix id 1-3 type 4 price 5-10 rating 11 ///
using c:\dokumenter\wines\wine.txt
// Add variable labels
label variable id "Identification number"
lab var type "Type of wine"
lab var price "Price per 75 cl bottle"
lab var rating "Quality rating"
// Add value labels
label define type 1 "red" 2 "white" 3 "rosé" 4 "undetermined"
label values type type
lab def rating 1 "poor" 2 "acceptable" 3 "good" 4 "excellent"
lab val rating rating
// Add data set label
label data "wine.dta created from wine.txt, 13.5.2001"
save c:\dokumenter\wines\wine.dta

Example 2. gen.visit12a.do generates visit12a.dta from visit12.dta

// gen.visit12a.do generates visit12a.dta with new variables.
use c:\dokumenter\proj1\visit12.dta, clear
// Calculate hrqol: quality of life score.
egen hrqol=rsum(q1-q10)
label variable hrqol "Quality of life score"
label data "Visit12a.dta created by gen.visit12a.do, 02.01.2001"
save c:\dokumenter\proj1\visit12a.dta

65
Example 3. Analyse Stata data
// winedes.do Descriptive analysis of the wine data 14.5.2001

use c:\dokumenter\wines\wine.dta
describe
codebook
summarize
tab1 type rating
tabulate type rating , chi2 exact
oneway price rating , tabulate

Note that in example 1 and 2 the structure was:

1. Read data (infix, use)
2. Calculation and documentation commands
3. Save data (save)
In these examples a new generation of the data was created; changes were documented with a
do-file with a gen. prefix.

In example 3 the structure was:

1. Read data (use)
2. Analysis commands
No new data generation was created, and the gen. prefix does not belong to such do-files.

Example 4. Elaborate profile.do

// c:\ado\personal\profile.do executes automatically when opening Stata.
// Write session start time in time.txt.
set obs 1
gen time="******* Session started: `c(current_date)' `c(current_time)'"
outfile time using c:\tmp\time.txt, noquote replace
clear
// Copy session start time to the cmdlog (cmdlog.txt) and open it.
// ! means that a DOS command follows.
! copy /b c:\tmp\cmdlog.txt + c:\tmp\time.txt c:\tmp\cmdlog.txt /y
cmdlog using c:\tmp\cmdlog.txt , append
// Open the log (stata.log).
set logtype text
log using c:\tmp\stata.log , replace

Compared to the profile.do suggested in section 1.2, this version adds a time stamp to
the command log file (cmdlog.txt). This means better possibilities to reconstruct previous
work.

66
Appendix 1

Purchasing Stata and manuals

To most users the following manuals will suffice:
[GSW] Getting Started manual
[U] User's Guide
[R] Base Reference Manual (four volumes)
but I wouldn't recommend less to anybody. This booklet does not intend to replace the
manuals, but hopefully it can serve as a guide.
If you work with epidemiological studies and survival analysis, you also need:
[ST] Survival Analysis and Epidemiological Tables
If you want to decide exactly how your graphs should look, you need:
[G] Graphics manual
However, the Graphics manual is a difficult companion; it took quite some time for me to
understand where to look for what.
If you want to write your own programs (ado-files), the User's Guide helps you some of the
way, but you may need:
[P] Programming manual

The Scandinavian sales agent for Stata and StatTransfer is Metrika (www.metrika.se).
Students and employees at University of Aarhus and Aarhus University Hospital can purchase
Stata at a special discount rate. Other educational institutions may have similar arrangements.

Various local information concerning Stata and other software may be found at:
www.biostat.au.dk/teaching/software.

67
Appendix 2

EpiData 3.0 www.epidata.dk

EpiData is an easy-to-use program for entering data. It has the facilities needed, but nothing
superfluous. Data entered can be saved as EpiInfo, Excel, DBase, SAS, SPSS and Stata files.
EpiData with documentation is available for free from www.epidata.dk.

EpiData files
If your dataset has the name first, you will work with three files:
first.qes is the definition file where you define variable names and entry fields.
first.rec is the data file in EpiInfo 6 format.
first.chk is the checkfile defining variable labels, legal values and conditional jumps.

Suggested options
Before starting for the first time, set general preferences (File < Options). I recommend:

[Show dataform] [Create datafile]

Font: Courier New bold 10pt. IMPORTANT:
Background: White First word in question is fieldname
Field colour: Light blue Lowercase
Active field: Highlighted, yellow
Entry field style: Flat with border
Line height: 1

Working with EpiData

EpiData's toolbar guides you through the process:
[Define data] ≡ [Make datafile] ≡ [Add checks] ≡ [Enter data] ≡ [Document] ≡ [Export data]

[Define Data]: You get the EpiData editor where you define variable names, labels, and
formats. If the name of your dataset is first, save the definition file as first.qes:
FIRST.QES My first try with EpiData.

entrdate Date entered <today-dmy>

lbnr Questionnaire number ####
init Initials ___
sex Sex # (1 male 2 female)
npreg Number of pregnancies ##
=========================================================
Page 2
bdate Date of birth <dd/mm/yyyy>
occup Occupation ## (see coding instruction OCCUP)

• The first word is the variable name, the following text becomes the variable label.
• ## indicates a two-digit numeric field,
• ##.# a four-digit numeric field with one decimal.
• ___ a three character string variable,
• <dd/mm/yyyy> a date,
• <today-dmy> an automatic variable: the date of entering the observation.
• Text not preceding a field definition ("1 male 2 female"; "======="; "Page 2")
are instructions etc. while entering data.

68
Variable names can have up to 8 characters a-z (but not æøå) and 0-9; they must start
with a letter. Avoid special characters, also avoid _ (underscore). If you use Stata for
analysis remember that Stata is case-sensitive (always use lowercase variable names).

[Make Datafile]: Save the empty data file first.rec.

[Add checks]: You do not have to write the actual code yourself, but may use the menu
system. The information is stored in a checkfile (first.chk) which is structured as below.
* FIRST.CHK Good idea to include the checkfile name as a comment.
LABELBLOCK
LABEL sexlbl Create the label definition sexlbl. You might give it the
1 Male
2 Female name sex – but e.g. a label definition n0y1 (0 No; 1 Yes)
END might define a common label for many variables
END
sex
COMMENT LEGAL USE sexlbl Use the sexlbl label definition for sex. Other entries than
JUMPS 1, 2, and nothing will be rejected.
1 bdate If you enter 1 for sex, you will jump to the variable
END
END
bdate; se the menu below.

The meaning of the Menu dialog box is not obvious at first sight, and I will explain a little:

first.chk Checkfile name

sex ▼ Select the variable
Sex of respondent
Number
Variable label and data type displayed
Range, Legal 1, 2, 9 Define possible values. A range e.g. as: 0-10, 99
Jumps 1>bdate Jump to bdate if sex is 1
Must enter No ▼ Skipping the field may be prevented
Repeat No ▼ Same value in all records (eg operator ID)
Value label sexlbl ▼ + [▼] Select among existing label definitions
[+] Define new value labels
Save Edit Close Save Save variable definitions
Edit Edit variable definitions

[Enter data]: You see a data entry form as you defined it; it is straightforward. With the
options suggested the active field shifts colour to yellow, making it easy for you to see where
you are.
As an assurance against typing mistakes you may enter part or all of the data a second time in
a second file and compare the contents of file1 and file2.

[Document] lets you create a codebook, including variable and value labels and checking
rules. The codebook shown below displays structure only, to be compared with your primary
codebook; you also have the option to display information about the data entered.

[Export]: Finally you can export your data to a statistical analysis programme. The .rec file is
in EpiInfo 6 format, and EpiData creates dBase, Excel, Stata, SPSS and SAS files. Variable
and value labels are transferred to Stata, SAS and SPSS files, but not to spreadsheets.

69
Appendix 3

NoteTab Light www.notetab.com

Both the Results window and the Viewer window have limitations in the ability to handle
output, and you will benefit from a good text editor. I use NoteTab Light, available for free
from www.notetab.com. I find NoteTab superior to Stata's Do-file editor; however you cannot
execute a do-file directly from NoteTab as you can from the Do-file editor. The use is
straightforward, like a word processor. I recommend the following options:
View ► Printing Options
Margins Left 2 cm, Right 1 cm, Top 1 cm, Bottom 1 cm
Font Courier New 9 pt
Other
Page Numbers Top, right
Number format Page %d
Header Date + Title
Footer None
Date Filter "your name" dd.mm.yyyy hh:nn
When you finished, save the settings by clicking the [Save] button.
Some NoteTab versions put a .txt extension to every file when saving it. To prevent this:
View ► Options ► File Filters
Default Extension: (nothing)

Making NoteTab work with Stata

In the following I assume that profile.do (see section 1.2) defined c:\tmp\stata.log
as the full log, including both commands and results; it must be in simple text format, not
SMCL format.
Open c:\tmp\stata.log in NoteTab to view, edit and print results. In NoteTab each file
has it own tab; you need not close them at exit. If NoteTab was open while running Stata you
might not see the latest output, but don't worry, just select Edit ► Reload (or [Alt]+[E] [R]),
and you have access to the updated output. Or right-click the file's tab and select Reload.
I suggest that you create two ado-files (see section 15.7) to ease your work:

nt opens Stata's log in NoteTab

Enter nt in Stata's command line window, and the log (c:\tmp\stata.log) opens in
NoteTab. winexec executes a Windows command:
program define nt
// c:\ado\personal\nt.ado. Opens the Stata log file in NoteTab.
winexec "C:\programmer\NoteTab Light\NoteTab.exe" c:\tmp\stata.log
end

newlog discards old log and opens new log

program define newlog
// c:\ado\personal\newlog.ado. Discards old log and opens new log.
log close
log using c:\tmp\stata.log , replace
end

70
Index
Do-file editor ..........................6
A Do-files.........................7;10;65
H
ado-files ............................... 62 drop ....................................23 help...................................... 9
Aggregating data .................. 26 dropline (graphs) ...........49 histogram........................ 43
anova ................................. 30 Hosmer-Lemeshow test ....... 33
append ............................... 25
Arithmetic operators ............ 20
E
ASCII data ........................... 17 egen ....................................21
I
Axis labels (graphs) ............. 40 encode ...............................55 ICD-10 codes....................... 56
Axis options (graphs) ........... 40 Entering data ........................68 if qualifier ................... 14;20
EpiData.................................68 Immediate commands.......... 60
B epitab command family ...29 in qualifier ........................ 14
Error messages .......................9 infile............................... 17
Bar graph options................. 42 Excel.....................................63 infix ................................. 17
Bar graphs ............................ 44 expand ...............................26 input ................................. 16
Bartlett's test......................... 31 insheet ....................... 17;63
browse ................................. 6
by: prefix .......................... 15 F Installing Stata ....................... 3
bysort prefix ................... 15 File names ............................10
findit .................................9 K
C Fixed format data..................17
for.......................................22
Kaplan-Meier curve ............. 35
keep.................................... 23
Calculations.......................... 20 foreach .............................62 Kruskall-Wallis test ............. 31
cc......................................... 29 format ...............................12
char.................................... 32 Format, dates ........................57
ci, cii ............................... 60 Format, numeric data............12 L
clear ................................. 16 Format, strings......................55 label ................................. 18
codebook .......................... 19 forvalues ........................62 Labels .................................. 18
collapse .......................... 26 Freefield data........................17 lfit.................................... 33
Command line window ....... 5;7 function (graphs) ...........50 Line graphs .......................... 46
Command syntax ................. 13 Linear regression ................. 32
Comma-separated data......... 17
Comments ............................ 15
G list.................................... 27
Log files................................. 8
compress .......................... 54 generate...........................21 Logical operators ................. 20
Conditional commands ........ 14 Goodness-of fit test .........33;37 logistic .......................... 33
Confidence interval (graphs) 47 Graph area ............................39 Logistic regression............... 33
Confidence intervals ............ 60 graph bar ........................44 Logrank test ......................... 35
Connecting lines (graphs)42;47 Graph command syntax........39 Long command lines............ 15
Continuation lines ................ 15 graph display...............52 lroc.................................... 33
contract .......................... 26 graph matrix .................51 lsens ................................. 33
Cox regression ..................... 35 Graph options .......................40 lstat ................................. 33
CPR numbers ....................... 57 graph save ......................52
cprcheck .......................... 58 graph twoway
Customizing Stata .................. 3 connected....................47 M
graph twoway dropline Macro................................... 62
D .........................................49
graph twoway function
Mann-Whitney test .............. 31
Mantel-Haenszel analysis .... 29
Data entry............................. 68 .........................................50 Manuals ............................ 9;67
Data set label........................ 18 graph twoway line .....46 Markers (graphs).................. 42
Data window .......................... 6 graph twoway rcap .....47 Matrix scatterplot................. 51
Date formats......................... 57 graph twoway rspike 48 mdy function ...................... 57
date function...................... 57 graph twoway scatter Memory considerations ....... 54
Date variables....................... 57 .........................................45 merge ................................. 25
describe .......................... 19 graph use ........................52 Missing values ..................... 12
destring .......................... 55 Graphs ..................................38
display............................. 60 Grid lines (graphs)................40
do........................................... 7
71
 recode ...............................22 summvl............................... 27
N Reference line (graphs) ........48 Survival analysis.................. 34
newlog.ado (user program) regress .............................32 Syntax .................................. 13
......................................... 70 Regression analysis ..............32
Non-parametric tests ............ 31
Normal distribution .............. 31
Regression, Cox....................35
Regression, linear .................32
T
Notation in this booklet.......... 2 Regression, logistic ..............33 tab1.................................... 28
note.................................... 19 Regression, Poisson..............37 tab2.................................... 28
NoteTab Light...................... 70 Relational operators..............20 tabi.................................... 29
nt.ado (user program) ...... 70 rename ...............................23 table ................................. 30
Number lists ......................... 14 Reordering variables.............23 Tab-separated data ............... 17
Numbering observations ...... 24 replace .............................21 tabulate .......................... 27
Numeric formats .................. 12 reshape .............................26 Ticks (graphs) ...................... 40
Numeric ranges .................... 14 Results window ...................5;8 time.ado (user program) 62
Numeric variables ................ 11 Review window......................5 Transposing data.................. 26
numlabel .......................... 18 ROC curve............................33 T-test.................................... 31
run.........................................7 ttest, ttesti................. 31
O twoway connected ....... 47

Observations ........................ 11
S twoway dropline ......... 49
twoway function ......... 50
oneway ............................... 30 sample ..........................23;59 twoway graphs.................. 43
Open a graph ........................ 52 Sample size estimation .........61 twoway line................... 46
Operators.............................. 20 sampsi ...............................61
twoway rcap................... 47
Options................................. 13 save ....................................16
twoway rspike .............. 48
order ................................. 23 Saving graphs .......................52
twoway scatter............ 45
outfile............................. 63 Scatterplot, matrix ................51
Output .................................... 8 Scatterplots ...........................45
outsheet .......................... 63 Schemes (graphs) .................53 U
sdtest ...............................31
search .................................9 Updating Stata ....................... 3
P Selecting observations..........23
use ...................................... 16
Plot area ............................... 39 Selecting variables................23
pnorm ................................. 31 Simulations...........................59 V
poisgof............................. 37 slist..................................27
sort ....................................24 Value labels ......................... 18
poisson............................. 37
Spreadsheets .........................63 Variable labels ..................... 18
Poisson regression................ 37
SPSS and Stata .....................63 Variable lists ........................ 13
Power estimation.................. 61
st command family .............34 Variable names .................... 11
P-P plot ................................ 31
Variables.............................. 11
predict............................. 32 Stata manuals..........................9
Variables window .................. 5
profile.do................... 4;66 StatTransfer ..........................63
Variance homogeneity ......... 31
Programs .............................. 62 stcox..................................35
Viewer window................... 6;8
Purchasing Stata................... 67 stptime .............................35
Stratified analysis .................29
Q String formats .......................55 W
string function................56
Weighting observations ....... 14
qnorm ................................. 31 String variables.....................55
whelp ................................... 9
Q-Q plot ............................... 31 sts graph ........................35
Wilcoxon test....................... 31
Qualifiers.............................. 13 sts list...........................35
Windows in Stata................... 5
Quotes .................................. 15 sts test...........................35
stset..................................34
R stsplit .............................36 X
stsum..................................36 xi: prefix................................ 32
Random numbers ................. 59 Study power..........................61 xpose ................................. 26
Random samples ............. 23;59 substr function................56
real function .................... 55 summarize ........................27