{smcl} {hline} {hline} help for {hi:mi_twoway}{right:Jean-François HAMEL} {hline} {title:Two-way imputations: imputing missing item responses in questionnaires for computing scores} {p 8 14 2}{cmd:mi_twoway} {it:varlist}, [{cmdab:sc:orename}({it:newvarname}) {cmdab:rep:lace} {cmdab:add:}({it:#}) {cmdab:st:yle}({it:keyword}) {cmdab:clear} {cmdab:da} {cmdab:it:erate}({it:#})] {title:Description} {p 8 14 2}{cmd:mi_twoway} is an implementation of the multiple imputation procedure proposed by Van Ginkel for computing scores on questionnaires containing missing item responses.{p_end} {p 14 14 2}Two methods are available: imputations based on a fixed effects two-way ANOVA, and imputations generated using data augmentation based on a mixed effect two-way ANOVA (with a random person effect assumed to follow a Normal distribution and a fixed item effect.{p_end} {p 14 14 2}{cmd:mi_twoway} is fully compatible with the Sata {help mi} procedures. The data is {help mi_set:set} and {help mi_impute:imputed} using {cmd:mi_twoway}, but all the estimations using multiple imputations are performed using the standard mi {help mi_estimate:estimate} procedures. {p 4 8 2}{it:varlist} is the list of the variables containing the item responses of the questionnaire (with possible missing data). {title:Options} {p 4 14 2}{cmd:scorename} specifies the name of the new variable containing, for each individual, the value of the score computed as the sum of the item responses. {it:scorename} is missing for each individual with at least one item missing response. {p 4 14 2}{cmd:replace} allows to replace individual scores in existing variables {p 4 14 2}{cmd:add} specifies the number of imputations to add; required with no imputations {p 4 14 2}{cmd:style} specifies in which style should be recorded the data: wide, mlong, flong, or flongsep; see {help mi_styles:[MI] styles}. {p 4 14 2}{cmd:clear} allows performing new imputations by remouving the previous one. {p 4 14 2}{cmd:da} generates imputations using data augmentation based on a mixed effect two-way ANOVA (with a random person effect and a fixed item effect. By default, imputations are generated using a fixed effects two-way ANOVA. {p 4 14 2}{cmd:iterate} defines the number of iterations of the data augmentation algorithm. By default, this number is fixed to 10. {marker example}{...} {title:Example} {pstd} Simulation of the data (using {help simirt}): {cmd:. simirt, nbobs(200) dim(5) group(0.5) deltagroup(0.4) clear}{right:(1) } {pstd} Creating the missing data, with a non-response rate of 10%: . {cmd:set more off }{right:(2) } . {cmd:forvalues j=1/5{c -(}}{right:(3) } 2. {cmd:replace item`j'=. if runiform()<0.1}{right:(4) } 3. {cmd:{c )-}}{right:(5) } {pstd} Generating 10 multiple imputations using a fixed effects two-way ANOVA: . {cmd:mi_twoway item*, scorename(score) add(10) style(wide)}{right:(6) } {pstd} Modeling score depending on {it:group} covariate using multiple imputations estimates ({help mi_estimate}): . {cmd:mi estimate: regress score i.group}{right:(7) } {pstd} Changing the way to impute data, using data augmentation based on a mixed effect two-way ANOVA: . {cmd:mi_twoway item*, scorename(score) replace add(10) style(wide) da clear}{right:(8) } {pstd} Changing the style of the data from {it:wide} to {it:mlong} ({help mi_convert}): . {cmd:mi convert mlong}{right:(9) } {pstd} Removal of the multiple imputations ({help mi_set##unset:mi_unset}): . {cmd:mi extract 0, clear}{right:(10) } {marker ref}{...} {title:References} {p 4 8 2}Van Ginkel JR, Van der Ark LA, Sijtsma K & Vermunt JK. Two-way imputation: A Bayesian method for estimating missing scores in tests and questionnaires, and an accurate approximation. Computational Statistics & Data Analysis (2007) 51: pp. 4013-4027.