Imputation Models for Special Subpopulations in Large-scale Survey Assessments

Usama S. Ali, Frederic Robin

DOI: 10.64028/odbw982677

Keywords: missing data, latent regression models, literacy-related non-response

Abstract: Nonresponse in large-scale survey assessments can arise from factors such as language barriers, reading difficulties, or disabilities. Excluding these subpopulations may introduce bias into survey results. This study develops an imputation method for literacy-related nonresponse cases in the international adult survey (PIAAC). These cases completed a special background questionnaire—the doorstep interview—but did not proceed to the main cognitive assessment. Using such limited data from respondents across selected countries with varying proportions of such cases, we compared and evaluated multiple imputation models to improve proficiency estimation. The proposed approach provides a practical solution for enhancing inclusivity in educational measurement.

Submission Number: 29