As a former SAS user converting to Stata, I needed a reference for statistical methods commonly used in clinical research.
These methods may also apply to other fields. This is an early version missing many important functions.

Descriptive Statistics

• initial analysis might be su, inspect
• summarize (su) – mean, stddev, min max; add ,detail for more info
• tabstat, stat() by() – get summary statistics
  • note fsum is an optional plugin that is a better version of this
• codebook – an alternative to summarize
• describe – variable name, label, type (describe using to look at a file)
• inspect (ins) – graph, number obs, num missing
• list (li) – print data; add “in 1/20” for first 20 obs
• swilk – Shapiro-Wilk test for normality (if significant, not normal)
• pnorm – normal probability plot

• table – table of summary statistics, use contents() with this.

  • show vitd25 by sex:

    table male, contents(min vitd25 mean vitd25 max vitd25)
    

• assert – test if something is true;

  assert maxweight >= weight
  

• tab – frequency table

• tab2 – two way table

Optional plugins

• I use fsum. To get this plugin, use:

  ssc install fsum
  

• find to top 10 plugins

  ssc hot
  

• distinct: counts distinct values of a variable

• vallist: lists distinct values of a variable (or can just use tabulate)

Commands

• display – give the value of something (or calculate)
• help – give info about a command
• rename – change a variable’s name
• use – load a file
• order – reorder variables (order var1, after(var2))

Variable creation

• use gen to create new variables

  gen obese = weight > 30 if weight != .
  

• use replace to update the values of an existing variable

  replace var2 = 1 if var1 > 1
  

• use xtile to create quantiles

  xtile newvar = oldvar, nq(3) /* create tertiles */
  

• can specify cutpoints using a variable name:

  egen writecat = cut(write), at(30,40,50,60,70)
  

• specify a counter based on the number of observations for an ID

  gsort id -eventdate
  by id: gen eventnum = _n
  

• convert strings to dates

  gen realdate  = date(datestring,"MDY") /* for 4-digit years */
  gen realdate  = date(datestring,"MD20Y") /* for 2-digit years */
  format realdate %td /* so that its readable */
  

Dummy Variables

• (generate dummy variables for each level of vargroup, e.g. var1, var2)

  tabulate vargroup, gen (var)
  

Missing Values

To copy values from the prior row for a given variable if values are missing:

    replace var = var[_n-1] if var1==.

It is critical that the data is sorted correctly.

Copy these variables only if the id matches:

    replace var = var[_n-1] if var==. & id==id[_n-1]

Math functions

• round(var, increment)

  round(var1, 0.01)
  

• Change variable type

  • convert to string:

    tostring zip, generate (zip5) format(%05.0f)
    

  • use “replace” instead of generate to replace the original variable

  • convert from string: destring

    destring notastring, replace
    

String functions

• strpos(first,second): returns the position of second in first, or 0 if it’s not included

  Example: strpos("example","amp") would equal 3
  

• subinstr(source,find,with,n) searches the source string, finds n occurrences of the second string and replaces them iwth the third string. If n is ommitted, all instances are replaced.

  Example: replace var1 = subinstr(var1,"exmpl","example",1)
  

Tips

• Use a wildcard to match a bunch of variables (e.g. su vit* to summarize any variable beginning with vitamin)
• use bysort [variable]: function – to do analysis by group
• if is the equivalent of SAS’s WHERE

Statistical analysis

Basic statistics

• ttest – two sample ttest

  ttest vitd25, by(male)
  

• ranksum – Wilcoxon rank-sum test (nonparametric)

  ranksum vitd25, by(male)
  

• median – turns continuous variable into binary based on median and performs chi-squared test

  median vitd25, by(male)
  

• pwcorr – Pearson correlation

  pwcorr var1 var2, obs sig
  

• spearman – Spearman correlation

  spearman var1 var2, pw stats(p obs rho)
  

• tab, chi2 – Chi-squared analysis

   tab var1 var2, chi2
  

Regression

• Stepwise:

  stepwise pr(0.10) pe(0.05) forward: regress var1 var2 var3 var4
  

• store estimates:

  /* store estimates in var model1 */
  estimates store model1 
  
  
  /* create a table comparing models */
  estimates table model1 model2 model3, p b(%9.2f)
  

• reg – linear regression

• logit – logistic regression with coefficients
  • add odds ratios using or option
• logistic – logistic regression with odds ratios

Mixed models

    xtmixed varlist || groupvar:

Shortcuts

Foreach loops

• foreach variable of varlist var1 var2 var3 {
  spearman `variable’ var2;
  }

Graphical Analysis

Scatterplots

• sc (variables), mcolor(color) msize(size) title(title)

• With regression line:

  graph twoway (sc var1 var2) (lfit var1 var2)
  

• use lfitci to add confidence intervals

• Use third variable as categorical:

  twoway (sc bmi dbpugml), by(gender, total)
  

Histograms

    histogram bmi, by(male, col(1))

Boxplots

    graph box var1, over(var2)

Bar graphs

    graph bar var1, over(var2)

Formatting

• replace numeric values with text (e.g. in graphs)

  label define labelname 1 "yes" 2 "no"
  

• assign this naming scheme to a variable

  label values var1 labelname
  

• replace a variable name with a label

  label variable var1 "Variable One"