alireza khoshgooyanfard; Mohammadreza Falsafinejad; noorali farrokhi
Abstract
Nonresponse is an inevitable challenge to large-scale studies and can result in wasting money, time and human resource involved in data collection and can also prevent the studies from obtaining their objects especially scores distribution. Imputation methods have thus been invented to estimate item ...
Read More
Nonresponse is an inevitable challenge to large-scale studies and can result in wasting money, time and human resource involved in data collection and can also prevent the studies from obtaining their objects especially scores distribution. Imputation methods have thus been invented to estimate item nonresponses in order to make inference from a completed data set. Using a simulation study on a real data set in the form of a multivariate experimental design, this paper evaluates the accuracy of three models including cumulative logit model, graded response model and explanatory item response model. The results show that the imputed values of all three models are acceptable under random nonresponse mechanism although the imputed values of the explanatory item response model are always more accurate than those of the other models. If nonrandom nonresponses are occurred, explanatory item response model has acceptable imputed values only at 5% nonresponse rate and the other models are not accurate at all. The results also show that it is more accurate to impute individual item nonresponses and then compute the total score instead of directly imputing the total score.
