Imputation of missing item responses
imputeitems varlist [if] [,prefix(string) method(string) random max(#)]
Description
imputeitems imputes missing item responses by different ways : Item Mean Substitution (IMS), Person Mean Substitution (PMS), Corrected Item Mean Substiutution (CIM), Interitem Correlation Substitution (ICS), logistic model (LOG) and Worst Case (WORST). A random process can be added to several methods.
Options
prefix defines the prefix to use to name the imputted variables (this prefix is followed by the name of the initial variable). By default, this prefix is "imp".
method defines the method to impute missing data :
pms computes the proportion of positive response of each individual on non missing items, and impute a deterministic result (if p<.5 then 0, else 1),
ims computes the proportion of positive response to each items, and impute a deterministic result (if p<.5 then 0, else 1),
cim computes the proportion of positive response to each items, corrected by the ability of the individual and impute a deterministic result (if p<.5 then 0, else 1),
ics searchs for each item the more correlated item and replaces a missing data by the data of this more correlated item (if the other response is missing too, there is no imputation),
log explains the responses of each item by a logistic model where the independent variables are the responses to the others items. Only significant variables are rettained (5%). These methods impute a deterministic result (if p<.5 then 0, else 1) [log] to missing responses (if the response to an independant variable is missing, there is no imputation),
worst replaces the missing data by a 0.
random adds a random effect to the imputation process (available only with pms, ims, cim or log). In these cases, the imputed value is randomly drawed from a binomial distribution using the parameter p.
max allows imputing missing values only for individuals with a maximal number of missing values defined with this option.
By default, pms method is working.
Old names of methods (bip, {bii}, bic and bil continues to run. They actually correspond to the add of the random option to the pms, ims, cim and log methods.
Example
. imputeitems itemA* /*PMS method, IMP prefix*/
. imputeitems itemA*, prefix(cim) method(cim)
. imputeitems itemA*, method(log) random
Reference
Huisman M. (2000), Imputation of missing item responses: some simple techniques. Quality & Quantity, 34, 331-351.
Author
Jean-Benoit Hardouin, PhD, assistant professor EA 4275 "Biostatistics, Clinical Research and Subjective Measures in Health Sciences" University of Nantes - Faculty of Pharmaceutical Sciences 1, rue Gaston Veil - BP 53508 44035 Nantes Cedex 1 - FRANCE Email: jean-benoit.hardouin@univ-nantes.fr Websites AnaQol and FreeIRT