.- help for ^implogit^ (STB-45: sg92) .- Logistic Regression for data including multiple imputations ----------------------------------------------------------- ^implogit^ depvar varlist [weight] [^if^ exp] [^in^ range] ^,^ ^impvars(^# of vars^)^ ^impno(^# of imputations^)^ [ ^r^obust ^cl^uster^(^varname^)^ ^l^evel^(^#^)^ ^or^ ] ^depvar^ may not contain multiple imputation data. ^varlist^ may contain multiple imputation data, with the following conditions: variables containing imputed data must be the last variables in varlist; and imputed variables must have variables names following special conventions -- they must be fewer than 5 characters, and have ^_01^ - ^_xx^ appended to them, where xx is the number of imputations done. Even though an imputed variable may have many individual variables to represent it, include it only once in varlist. Example: the variable ^incom^, was imputed 5 times. It should be labled ^incom_01 incom_02 incom_03 incom_04^ ^incom_05^. In the command call, list only 1, say, ^incom_01^. The program will iterate through the numbers and take care of the rest. If your variables are not named properly, the program will not work! ^fweight^s and ^pweight^s are allowed. ^implogit^ does not share features with all estimation commands. Because of the external variance adjustments implicit in the corrections to the std. errs and the programmer's limited skill in matrix algebra, this program does not post a full variance-covariance estimate. Post estimation commands that rely solely on ^_b^ and ^_se^ are available. Any commands requiring the off diagonal elements of the vce will either not work or will be wrong. Try ^matrix get(vce)^, and you will see what is missing. ^implogit^, typed without argument, does not replay the previous results. Description ----------- ^implogit^ uses the Rubin (1987) corrections of coeficents and standard errors for logistic regressions with data that contain multiple imputations. Multiple imputation variables must be ordered in a specific way and named in a special fashion; see ^varlist^ above. ^implogit^ proceeds by performing k logistic regressions (where k is the number of imputations done), cycling through the different imputations in each regression. Results are saved, and, when done, coeficents are averaged and standard errors are corrected. Results are then reported. In most regards, ^implogit^ behaves as the standard stata ^logit^ command. The procedure reports unexponentiated coeficents and their corrected standard errors. Options ------- ^impvars(^no. of vars^)^ indicates the number of variables included that contain multiple imputations. They must be the last variables specified in varlist. If your model contains 2 variables for which data have been imputed, ^impvars(2)^ should be specfied, and they should be the last 2 variables in varlist. ^impvars^ default is (1). ^impno^ is the number imputations done for each imputed variable and thus the number of iterations of regressions (k) that will be required. If there is more than one variable with multiple imputations, they all must have the same number of imputations. ^robust^ specifies the Huber/White/sandwich estimator of variance is to be used in place of the traditional calculation; see ^[U] 26.10 Obtaining^ ^robust variance estimates^. ^robust^ combined with ^cluster()^ allows observations which are not independent within cluster. ^cluster(^varname^)^ specifies that the observations are independent across groups (clusters) but not necessarily within groups. varname specifies to which group each observation belongs; e.g., ^cluster(personid)^ in data with repeated observations on individuals. See ^[U] 26.10 Obtaining robust^ ^variance estimates^. ^cluster()^ can be used with ^pweight^s to produce estimates for unstratified cluster-sampled data, but see help @svylogit@ for a command especially designed for survey data. Specifying ^cluster()^ implies ^robust^. ^or^ reports odds ratios (exponentiated coeficents) and their standard errors, etc., rather than unexponentiated coeficents. Examples -------- . ^implogit comply educ incom_01, impvars(1) impno(5)^ . ^implogit comply educ incom_01 size_01, impvars(2) impno(5)^ . ^implogit comply educ size_01, impno(7) cluster(town) level(99)^ Saved Results ------------- All typical S_E_ global macros are available, as are _b and _se vectors. $S_E_ll, $S_E_l0 and $S_E_prs2 are based on the unadjusted results of the last of the k component logistic regressions, as the programmer is not aware of the consequence to log-liklihood from multiple imputation. Author ------ Christopher Paul RAND, Santa Monica, California cpaul@@rand.org Also see -------- STB: sg92 (STB-45)