help missing
-------------------------------------------------------------------------------
Title
[U] 12.2.1 Missing values
Description
Stata has 27 numeric missing values:
., the default, which is called the "system missing value" or sysmiss
and
.a, .b, .c, ..., .z, which are called the "extended missing values".
Numeric missing values are represented by large positive values. The
ordering is
all nonmissing numbers < . < .a < .b < ... < .z
Thus, the expression age > 60 is true if variable age is greater than 60
or missing.
To exclude missing values, ask whether the value is less than ".". For
instance,
. list if age > 60 & age < .
To specify missing values, ask whether the value is greater than or equal
to ".". For instance,
. list if age >=.
Stata has one string missing value, which is denoted by "" (blank).
Remarks
More details concerning missing values and their treatment in Stata are
provided under the following headings:
Overview
Expressions
Operators
Functions
Matrices
Useful commands
Value labels
Estimation commands
Technical note: Version 7 and earlier
Overview
1. Stata supports different types of numeric missing values that can be
used to specify different reasons that a value is unknown. The most
frequently used missing value ., referred to as sysmiss, is nearly
always generated by Stata when it cannot assign a specific value.
The 26 extended missing values .a, .b, ..., .z are available to users
requiring more elaborate tracking of missing values.
Empty strings are treated as missing values of type string.
2. Numeric missing values are represented by large positive values.
This means that an expression such as income > 100 evaluates to true
for missing values of the variable income, as well as to those that
are greater than 100. Also, the simple expression if varname
evaluates to true for all nonzero values of varname, including
missing values.
3. The ordering of missing values is
all nonmissing numbers < . < .a < .b < ... < .z
4. Most Stata statistical commands deal with missing values by
disregarding observations with one or more missing values (called
"listwise deletion" or "complete cases only").
Expressions
Expressions occur in many places in Stata (see [P] syntax and exp). For
example,
. generate newvarname = exp
evaluates the expression exp for each observation of the variable
newvarname. Observations of newvarname are set to missing if exp
evaluates to missing.
Expressions are also used to restrict a command's operation to a subset
of the observations. For instance,
. summarize varname if exp
summarizes varname by using all observations for which exp evaluates to
true (not zero), including observations that are missing.
Operators
The relational operators (see operators) interpret missing values as
large positive numbers (see above). All the following thus evaluate to
true
73 < . . == . .a == .a
.a != . .a < .b .a <= .b
whereas all the following evaluate to false
73 >= . . == .a . > .a
The numerical operators (+ etc) return missing if any of their arguments
are missing.
Functions
Stata has a few special functions for dealing with missing values:
missing() returns 1 (meaning true) if any of its arguments,
numeric or string, evaluates to missing and 0
(meaning false) otherwise.
mi() is a shorthand for missing().
matmissing(K) returns 1 (meaning true) if any elements of the
matrix K are missing and 0 (meaning false)
otherwise.
Some Stata functions interpret . in a special way. For instance, the
function inrange(x,a,b) returns 1 if x belongs in the interval [a,b].
This function interprets a==. as -infinity and b==. as +infinity. These
special interpretations are discussed in functions.
Other Stata functions return missing (.) if one or more of the arguments
are missing or invalid.
Matrices
Matrices may contain all types of missing values. The matrix operators
(see matrix operators)
- negate
' transpose
\ row join
, column join
+ add
- subtract
* multiply (including multiply by scalar)
/ division by scalar
# Kronecker product
generate missing values elementwise.
In the matrix product C=A*B, C[i,j] is missing if row i of A or column j
of B contain a missing value.
Matrix division by scalar C=A/b is not allowed if the scalar b is a
missing value. Otherwise, missing values in matrix A generate missing
values in C elementwise.
Like the list command, the matrix list command has a nodotz option to
display extended missing value .z as a blank string rather than as ".z".
Useful commands
-------------------------------------------------------------------------
mvencode transforms missing values into numeric values
mvdecode transforms numeric values into missing values
codebook provides extensive information about variables,
including the occurrence of simple and extended
missing values
egen, rownonmiss() number of valid observations in a varlist
egen, rowmiss() number of missing values in a varlist
recode recodes a variable, optionally into a new variable,
with special facilities to recode missing values.
mi multiple imputation of missing values
xtdescribe describes participation patterns in panel data
-------------------------------------------------------------------------
Value labels
It is possible to define value labels for the extended missing values .a
to .z, but not for sysmiss .. These value labels show up in the same way
as value labels for nonmissing values. See [D] label.
Estimation commands
Most Stata commands ignore observations that are missing in one or more
of the variables referred to in the command. For instance, the
regression command regress disregards all observations that have a
missing value for the dependent variable or missing values for any of the
independent variables. This method is known as "listwise deletion",
"complete cases only", etc. It is statistically appropriate only if the
missing values are "at random". In an if or weight expression to a
command, the expressions will be evaluated, and the missing values will
be processed using the operators and function() logic.
Stata commands that can treat multiple observations as being related to
one observational unit (e.g., observations from a panel in xt models,
episodes in st models) ignore specific observations from the "group",
namely, those that have missing values.
Technical note: Version 7 and earlier
Before Stata 8, Stata had only one missing value, the period (.). Thus,
you could test whether an expression or variable exp was missing with the
expression exp==.. Starting with Stata 8, this method is no longer
correct. exp==. now means that the expression exp equals a specific
missing value, namely, sysmiss .. exp==. returns false if exp equals
one of the extended missing-value types such as .a or .z. To test
whether exp is missing, i.e., equals either . or one of the extended
missing values, one should use the expression
exp >= .
or
missing(exp)
which can be abbreviated to
mi(exp)
To test that exp is not missing, use one of the forms
exp < .
!missing(exp)
!mi(exp)
An advantage of the last two forms is that the missing functions
missing() and mi() allow multiple (numeric or string) arguments to test
whether any of the arguments is missing.
Old programs and do-files will continue to work using the old method, as
long as the version is set to 7 or less. See [P] version.
Also see
Manual: [U] 12.2.1 Missing values,
[D] missing values
Help: [D] codebook, [D] egen, [D] functions, [MI] intro, [D] mvencode,
> [D] recode, [XT] xtdescribe,
[U] 13 Functions and expressions (expressions),
[U] 13 Functions and expressions (operators)
