Multiple Imputation using SAS

1
Multiple Imputation using SAS

• Don Miller
• 812 Oswald Tower
• miller_at_pop.psu.edu
• 814-863-3155

2
Introduction

• Missing values occur often in research
  refused/dont know, attrition, skip patterns
• Dropping missing values may bias results
  (e.g. women and/or
  overweight tend to disclose their weight less
  often than others)
• Attempts are made to impute the data (fill
  in missing values)
• Single imputation (e.g. with the mean) is biased,
  doesnt give measure of uncertainty

3
Paris datasets

• Open Windows Explorer (or My Computer)
• Tools Map Network Drive
• Drive P
• Folder \\paris\sas_data
• For help help_at_pop.psu.edu
• Stat help stat-core_at_pop.psu.edu

4
Data Setup
5
Multiple Imputation Simple Procedure

• 1. Impute using PROC MI
• 2. Round off, if you want plausible values
• (caution this will bias your results)
• 3. Do analysis PROC REG, LOGISTIC, etc.
• using by _imputation_ in the procedure
• 4. Combine results using PROC MIANALYZE
• For categorical variables Construct binary dummy
  variables, throwing out reference category (e.g.
  race 1white, 2black, 3other becomes
  black, other variables)

6
PROC MI

• Typical syntax
  
• proc mi databmx outimpdat seed33155
• var bmxbmi bmxht bmxwt bmxarmc bmxarml
• run
• data 1 copy of data with missing values
• out 5 copies of data with imputed values (will
  be different across copies)
• seed random seed, you can keep same to
  reconstruct your results
• var Variables with missing values you need
  imputed, in model, and those that may be
  helpful with imputation

7
PROC MI Sample Output
8
PROC MI Sample Output
9
PROC MI Options

• nimpute5 imputations, default5 0
  gives missing patterns
• minimum0 0 0 0 set min max, sometimes
  maximum1 1 1 90 doesnt converge as well
• round1 1 1 0.01 round off option
• alpha0.05 confidence limits
• mu00.5 0.5 0.5 25 t test null hypothesis µµ0

10
PROC MI Statements

• em maxiter200 outemdata
  EM algorithm, MLE of missing data
• freq fweight
  weighs observations by
  frequency weight
• mcmc (options)
  modify imputation method
• class sex race
  specify categorical variables (dont need
  dummies) (new / experimental)

11
Output dataset
12
Regression

• Fit your model as if data had no missing values,
  using by _imputation_
• proc reg dataimpdat outestparmcov covout
• model bmxbmibmxht bmxwt bmxarmc bmxarml
• by _imputation_
• run
• Youll get nimpute (usually 5) sets of output
• Estimates, covariances, errors will be combined
  in MIANALYZE (R² is just mean)
• Need to generate parameter estimates and
  covariance data set (varies by procedure)

13
Parameter Est. Covariance Matrix

• proc logistic dataimpdat descending model
  bmxbmibmxht bmxwt bmxarmc bmxarml /covb
  by _imputation_ ods output
  ParameterEstimatesparmsdat CovBcovbdat
  run
• proc mixed dataimpdat model
  bmxbmibmxht bmxwt bmxarmc bmxarml /solution
  covb by _imputation_ ods output
  covparmsparmcov run

14
Parameter Est. Covariance Matrix

• proc genmod dataimpdat model
  bmxbmibmxht bmxwt bmxarmc bmxarml /covb
  by _imputation_ ods output
  ParameterEstimatesparmsdat CovBcovbdat
  run
• proc glm dataimpdat model bmxbmibmxht
  bmxwt bmxarmc bmxarml /inverse by
  _imputation_ ods output ParameterEstimatesp
  armsdat InvXPXxpxidat
  run

15
PROC MIANALYZE

• Syntax depends on what procedure you used in
  previous step
• proc mianalyze dataparmcov (or)
  proc mianalyze parmsparmsdat covbcovbdat
  (or) proc mianalyze parmsparmsdat
  xpxixpxidat
• (then type this)
• modeleffects intercept bmxht bmxwt bmxarmc
  bmxarml
• run
• Note the var statement is now modeleffects
• Note that the dependent variable is omitted

16
PROC MIANALYZE Output