Thanks for your *very* helpful reply.
________________________________ From: Juned Siddique <[email protected]> To: [email protected] Sent: Friday, June 17, 2011 5:54 PM Subject: Re: Univariate imputation Hi Paul, I don't know of any ABB routines in SAS or Stata. I think you could do it in SAS using the SURVEYSELECT procedure to draw a simple random sample with replacement of size nobs. Then use it again on this new sample to draw another SRS with replacement of size nmis. You could do something similar in Stata using the bsample (bootstrap sample) command. It is worth noting that you do not have to assume that data are ignorable with ABB imputation. For example, if you think that missing values tend to be larger than observed values, you could sample with probability proportional to size in the first stage. See my paper with Tom Belin for some variations on this idea. http://www.sciencedirect.com/science/article/pii/S0167947308003915 For bivariate data, both SAS and Stata have predictive mean matching (PMM) hotdeck procedures. The SAS Macro MIDAS also does PMM with a nonignorable ABB: http://www.jstatsoft.org/v29/i09 -Juned -----Original Message----- From: Impute -- Imputations in Data Analysis [mailto:[email protected]] On Behalf Of Paul von Hippel Sent: Wednesday, June 08, 2011 6:37 PM To: [email protected] Subject: Re: Univariate imputation Great suggestion. Is there a convenient ABB routine in SAS or Stata? A bivariate version -- X complete, Y incomplete -- would also be welcome.... Sent from my iPhone. Please excuse my brevity. On Jun 6, 2011, at 4:48 PM, Juned Siddique <[email protected]> wrote: > Hi. > > I agree with Dave. An approximate Bayesian bootstrap (ABB) (Rubin > 1987, page 124; Rubin and Schenker, 1986) will work in this situation. > > -Juned Siddique > > -----Original Message----- > From: Impute -- Imputations in Data Analysis > [mailto:[email protected]] On Behalf Of David > Judkins > Sent: Monday, June 06, 2011 12:22 PM > To: [email protected] > Subject: Re: Univariate imputation > > Any hotdeck procedure would also do the job. Solas can implement > this. The advantage being not having to make any parametric assumptions. > > http://www.statistical-solutions-software.com/products-page/solas-for- > missin > g-data-analysis/ > > --Dave Judkins > > -----Original Message----- > From: Impute -- Imputations in Data Analysis > [mailto:[email protected]] On Behalf Of Raghunathan, > Trivellore > Sent: Monday, June 06, 2011 11:21 AM > To: [email protected] > Subject: Re: Univariate imputation > > I believe that IVEware will use the normal posterior distribution with > constant mean, variance if the variableis continuous, > binomial/constant logit model for binary, Poisson constant mean for > count, multinomial for categorical etc. It should simply use the > regression model with intercept term. You should specify the iterations to be 1. > > (Though, we have never tried for one variable data set!) > > Raghu > ________________________________________ > From: Impute -- Imputations in Data Analysis > [[email protected]] On Behalf Of Paul von Hippel > [[email protected]] > Sent: Monday, June 06, 2011 10:26 AM > To: [email protected] > Subject: Re: Univariate imputation > > To clarify: I have N cases of a single iid variable Y. n<N cases are > missing the Y value at random. There are no Xs to predict Y, and that > rules out most imputation software, including IVEware. What software is available? > > In practice imputation is useless in this situation, but it's > important from a theoretical point of view. > > Sent from my iPhone. > Please excuse my brevity. > > On Jun 6, 2011, at 4:56 AM, William Winkler > <[email protected]> > wrote: > >> univariate imputation - there are additional methods besides the >> below two >> >> http://www.isr.umich.edu/src/smp/ive/ >> >> http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.96.3441 - >> documentation for slice software - need another url for the software >> >> >> >> >> >> From: Paul von Hippel <[email protected]> >> >> To: [email protected] >> >> Date: 06/06/2011 05:18 AM >> >> Subject: Univariate imputation >> >> Sent by: Impute -- Imputations in Data Analysis > <[email protected]> >> >> >> >> >> >> >> It is common in simulations to impute data with only one variable, >> but most imputation software assumes that there are at least two >> variables. Is software available that carries out univariate imputation?
