{smcl}
{* 01nov2006}{...}
{hline}
help for {hi:micombine}{right:Patrick Royston}
{hline}
{title:Estimation of regression models with multiply imputed samples}
{p 8 17 2}
{cmd:micombine}
{{it:supported_regression_cmd} | {it:other_regression_cmd}}
[{it:yvar}]
[{it:covarlist}]
[{it:other_stuff]}
[{cmd:if} {it:exp}]
[{cmd:in} {it:range}]
[{it:weight}]
[
{cmd:,}
{cmd:br}
{cmdab:nocons:tant}
{cmdab:det:ail}
{cmdab:ef:orm}|{{cmdab:ef:orm(}{it:string}{cmd:)}}
{cmdab:g:enxb(}{it:newvarname}{cmd:)}
{cmdab:imp:id(}{it:varname}{cmd:)}
{cmdab:inf:gain}
{cmd:lrr}
{cmdab:nowar:ning}
{cmdab:obs:id(}{it:varname}{cmd:)}
{cmd:svy}[{cmd:(}{it:svy_options}{cmd:)}]
{it:regression_cmd_options}
]
{p 4 4 2}
where
{p 8 8 2}
{it:supported_regression_cmd}s are
{help clogit},
{help cnreg},
{help glm},
{help logistic},
{help logit},
{help mlogit},
{help ologit},
{help oprobit},
{help poisson},
{help probit},
{help qreg},
{help regress},
{help rreg},
{help stcox},
{help streg},
or
{help xtgee}, and {it:other_regression_cmd} is any other Stata regression command
(see Remarks).
{p 4 4 2}
{cmd:micombine} shares a subset of the features of all estimation commands
(see help {help estimates}); see {it:Remarks}.
{p 4 4 2}
All weight types supported by {it:regression_cmd} are allowed; see help
{help weights}.
{title:Description}
{p 4 4 2}
{cmd:micombine} estimates the parameters of a regression model whose
type is determined by {it:supported_regression_cmd} or {it:other_regression_cmd}.
Parameter estimates are combined
across several replicates obtained previously by multiple imputation,
e.g. by using {help ice} to create a file of imputed data.
See {it:Remarks} for a brief account of how {cmd:micombine} combines
the estimates and obtains standard errors.
{title:Options}
{p 4 8 2}
{cmd:br} calculates degrees of freedom and tests of significance for each predictor
according to the formulae (3)-(5) of Barnard & Rubin (1999).
After estimation, the required degrees of freedom are stored in a matrix
(column vector) {cmd:e(nutilde)}. Note that if {cmd:test}
is used after {cmd:micombine} for significance testing of regression
coefficients, such tests assume that the degrees of freedom are
equal to the number of observations minus the number of parameters
estimated, not those given in {cmd:e(nutilde)}.
{p 4 8 2}
{cmd:detail} gives details of the regression model for each imputation.
{p 4 8 2}
{cmd:eform(}{it:string}{cmd:)} indicates that the exponentiated
form of the coefficients is to be output and reporting of the constant is to
be suppressed; {it:string} is used to label the exponentiated coefficients.
{p 4 8 2}
{cmd:eform} indicates that the exponentiated
form of the coefficients is to be output and reporting of the constant is to
be suppressed; the exponentiated coefficients are labelled {cmd:exp(b)}.
{p 4 8 2}
{cmd:genxb(}{it:newvarname}{cmd:)} creates {it:newvarname} to hold the
linear predictor from each regression model, averaged over all the
imputations.
{p 4 8 2}
{cmd:impid(}{it:varname}{cmd:)} specifies that {it:varname} is the variable
identifying the imputations. The number of imputations is determined as
the number of unique values of {it:varname}. All observations for which
{it:varname} takes the value zero are ignored in the analysis.
Default {it:varname}: {cmd:_mj}.
{p 4 8 2}
{cmd:infgain} reports the percentage increase in information and sample
size due to the use of multiple imputation. The information gain is
the percent increase in Wald chisquare for the entire model, comparing
the Wald chisquare for the model on the original data (complete case
analysis) with that using the variance-covariance matrix of the
parameters estimated using Rubin's rules. With a bad imputation model
the information increase could be negative.
{p 4 8 2}
{cmd:lrr} specifies that the Li-Raghunathan-Rubin (LRR) robust estimate of the
variance-covariance matrix of the regression coefficients be used.
{p 4 8 2}
{cmd:noconstant} suppresses the regression constant in all regressions.
{p 4 8 2}
{cmd:nowarning} suppresses the warning message about the use
of {it:other_regression_cmd}s (see {it:Remarks}).
{p 4 8 2}
{cmd:obsid(}{it:varname}{cmd:)} is provided to allow {cmd:micombine} to analyse
datasets created by programs other than {cmd:ice}. {it:varname} specifies the name
of a variable holding the "observation ID", i.e. the sequence number of each
observation in a given imputation. The number of observations should
be identical between imputations, as should the order of the observations.
{it:varname} should run 1,...,N for imputation 1, 1,...,N for imputation 2, and
so on. {cmd:ice} automatically stores the information with the data, so this
option is not required. Default {it:varname}: {cmd:_mi}.
{p 4 8 2}
[Stata 9] {cmd:svy}[{cmd:(}{it:svy_options}{cmd:)}] performs survey regression.
The prefix {cmd:svy:} is placed before
{it:regression_cmd}. If {it:svy_options} is supplied then {cmd:, }
{it:svy_options} is placed after {cmd:svy} and before the colon.
The data must be {cmd:svyset} before
this option is used. This must be done before {cmd:ice} is used
to impute missing values. That the data have been {cmd:svyset} is
inherited by the file of imputations created by {cmd:ice}.
{p 4 8 2}
[Stata 8] {cmd:svy} performs survey regression.
The prefix {cmd:svy} is placed before
{it:regression_cmd}, so that for example {cmd:micombine regress ..., svy}
is interpreted as {cmd:micombine svyregress ...}. Options for survey
regression are included as options to {cmd:micombine}.
The data must be {cmd:svyset} before
the {cmd:svy} option is used. This must be done before {cmd:ice} is used
to impute missing values. That the data have been {cmd:svyset} is
inherited by the file of imputations created by {cmd:ice}.
{p 4 8 2}
{it:regression_cmd_options} may be any of the options appropriate to
{it:regression_cmd}.
{title:Remarks}
{p 4 4 2}
Details of statistical inference from multiple imputed datasets are nicely described
in a recent Stata Journal article by John Carlin and colleagues (Carlin et al, 2003).
Here, with due acknowledgment to John, I give an edited version of Section 2 of his article.
{p 4 4 2}
A simple method of combining estimates from several models was derived by Rubin (1987).
Suppose initially that primary interest lies in estimating a scalar quantity, Q.
Here, Q is a regression coefficient, for example, the log hazard ratio in a
proportional hazards model. Suppose that we have imputed m complete datasets
using an appropriate model. In each dataset, standard complete-data methods
are used to obtain an estimate of Q with an associated standard error.
Let Q_j and V_j denote the point estimate and variance respectively from the jth
(j = 1, 2, ... , m) dataset. The point estimate Q^ of Q from multiple imputation
is simply the arithmetic mean of Q_1,...,Q_m.
{p 4 4 2}
Obtaining a valid standard error for this estimate of Q^ requires combining information
on within-imputation and between-imputation variation. The latter is important in
reflecting uncertainty due to variability between imputation samples. First,
a within-imputation variance component, W, is obtained as the mean over the m
imputations of the complete-data variance-covariance matrices, V_1,....,V_m.
Second, a between-imputation variance
component, B, is calculated as the sum of squares of Q_1,....,Q_m about Q^,
divided by m-1. In summary,
{p 8 12 2}
Q^ = (Q_1 + ... + Q_m)/m
{p 8 12 2}
W = (V_1 + ... + V_m)/m
{p 8 12 2}
B = ((Q_1 - Q^)^2 + ... + (Q_m - Q^)^2)/(m - 1)
{p 4 4 2}
The (total) variance T of Q^ is given by
{p 8 12 2}
T = W + B * (1 + 1/m)
{p 4 4 2}
Rubin (1987) showed that (Q - Q^)/sqrt(T) is distributed approximately
as Student's t on nu degrees of freedom, where
{p 8 12 2}
nu = (m - 1) * (1 + W /(B * (1 + 1/m)))^2
{p 4 4 2}
The (1 + 1/m) term in these expressions indicates that it is not necessary to
a create large number of imputed datasets, particularly when B is much smaller
than W. The condition will be satisfied unless there is much missing data and
the parameter estimates within each dataset are very precise.
{title:Available regression commands}
{p 4 4 2}
{cmd:micombine} has been tested with the commands listed
under {it:supported_regression_cmd} at the beginning of this
help file. {cmd:micombine} {it:may} work satisfactorily
with {it:other_regression_cmd}s, but this cannot
be guaranteed. This facility is provided so that the
researcher familiar with a particular Stata command has a
fighting chance of obtaining correct MI estimates and standard errors.
{hi:HOWEVER, THE AUTHOR DISCLAIMS RESPONSIBILITY FOR}
{hi:THE CORRECTNESS OF RESULTS ARISING FROM USE OF AN} {it:other_regression_cmd}.
Note that {it:other_stuff} in the syntax diagram
is code that may be required by
some {it:other_regression_cmd}s, for example {cmd:ivreg} wants
{cmd:(}{it:varlist2}{cmd: = }{it:varlist_iv}{cmd:)}.
{cmd:micombine} parses for the
occurrence of an opening parenthesis. There may be other
syntaxes that are not accommodated by this approach; if so,
please contact the author with details.
{title:Post-estimation prediction}
{p 4 4 2}
The {cmd:predict} command {it:may} work as you expect after {cmd:micombine},
but this feature should be
treated with caution. {cmd:micombine} stores the quantities needed by
{cmd:predict} at the last execution of the regression command, that is at the
final imputation, but prediction following some regression commands has
non-standard features that are hard to emulate accurately.
Known issues are as follows:
{p 8 12 2}
1. After {cmd:micombine mlogit}: {cmd:predict} may require that the outcome
levels are known as 0, 1, 2, ... , so it may be necessary to drop the
score label for the outcome variable, if such a label is defined.
This is KNOWN to be a problem using {cmd:mfx} following {cmd:micombine mlogit}.
For example, {cmd:mfx compute, predict(outcome(0))} will work only if
the lowest level of the outcome is 0, and is not labelled.
{title:Sample size}
{p 4 4 2}
The sample size reported by {cmd:micombine} is the number of observations found
when fitting the model in the first imputation (i.e. by default: for {hi:_mj==1}).
It may happen that the sample size varies between imputations,
for example, when the effect of an {cmd:if} or {cmd:in} filter differs between
imputations, or when a weighting scheme effectively removes different
observations in different imputations. The resulting parameter estimates
and their SEs are believed still to be approximately valid in this situation.
The program alerts you to the occurrence of variable sample size, but no action
need be taken by you. Post-estimation commands {cmd:test} and {cmd:testparm}
use the sample size found when the model is fitted in the final imputed dataset.
{title:Examples}
{p 4 8 2}{cmd:. ice y x1 x2 x3 using imp, m(10) genmiss(m_)}{p_end}
{p 4 8 2}{cmd:. use imp, clear}{p_end}
{p 4 8 2}{cmd:. micombine regress y x1 x2 x3}{p_end}
{p 4 8 2}{cmd:. stset time, failure(cens)}{p_end}
{p 4 8 2}{cmd:. micombine stcox x1 x2 x3, genxb(index)}{p_end}
{p 4 8 2}{cmd:. test x2==1}{p_end}
{p 4 8 2}{cmd:. testparm x1 x2}{p_end}
{p 4 8 2}{cmd:. micombine regress y x1 x2 x3, svy(subpop(if sex==1))}{p_end}
{title:Author}
{p 4 4 2}
Patrick Royston, MRC Clinical Trials Unit, London.
pr@ctu.mrc.ac.uk
{title:References}
{p 4 8 2}
Barnard, J. and D. B. Rubin. 1999. Small-sample degrees of freedom with
multiple imputation. {it:Biometrika} {cmd:86}: 948-955.
{p 4 8 2}
Carlin, J. B., N. Li, P. Greenwood, and C. Coffey. 2003.
Tools for analyzing multiple imputed datasets. {it:Stata Journal} {cmd:3(3)}: 226-244.
{p 4 8 2}
Carlin, J. B., N. Li, P. Greenwood, and C. Coffey. 2003.
Tools for analyzing multiple imputed datasets. {it:Stata Journal} {cmd:3(3)}: 226-244.
{p 4 8 2}
Li, K., T. Raghunathan, and D. Rubin. 1991. Large sample significance levels from
multiply-imputed data using moment-based statistics and an F reference distribution.
{it:Journal of the American Statistical Association} {cmd:86}: 1065-1073.
{p 4 8 2}
Royston P. 2004. Multiple imputation of missing values.
{it:Stata Journal} {cmd:4(3)}:227-241.
{p 4 8 2}
Royston P. 2005. Multiple imputation of missing values: update.
{it:Stata Journal} {cmd:5(2)}:188-201.
{p 4 8 2}
Royston P. 2006. Multiple imputation of missing values: update of ice.
{it:Stata Journal} in preparation.
{p 4 8 2}
Rubin, D. 1987. Multiple Imputation for Nonresponse in Surveys. New York: John Wiley &
Sons.
{p 4 8 2}
Schafer, J. 1997. Analysis of Incomplete Multivariate Data. London: Chapman & Hall.
{p 4 8 2}
van Buuren, S., H. C. Boshuizen and D. L. Knook. 1999. Multiple imputation of
missing blood pressure covariates in survival analysis.
{it:Statistics in Medicine} {cmd:18}:681-694. (Also see http://www.multiple-imputation.com.)
{title:Also see}
{p 4 13 2}
Online: help for {help ice}.