An Item Response Model for Understanding Item

Abstract

Survey research has been widely used in public opinion research in Ghana. Ghanaian researchers are happy about data richness and they are also concerned about data quality. In this paper Item Response Theory (IRT) has been used to identify the most appropriate IRT model for understanding item. The techniques are appropriate and practical. A questionnaire data on Ghana collected in the 5th wave of the World Values Survey was used for the analysis. The five categories of survey questions that are most difficult to answer by respondents were Life Related Questions, Value Related Questions, Political Related Questions, Income Related Questions and Democracy Related Questions. Missing or ‘don’t know’ responses were assigned a 0 score, and 1 was assigned to answered items. The data was analysed based on four IRT models namely, the constrained Rasch model, the unconstrained Rasch model, the two parameter logistic model, and the three parameter logistic model. These models were explored to determine the most appropriate model for the data. In this paper, the unconstrained Rasch model emerged as the best model for understanding item non-response. We found that, income related questions had the highest difficulty parameter, hence the most difficult category of survey questions to answer. It was also found that, if an individual does not answer a survey question or give a ‘don’t know’ answer, it is not only because of the question’s difficulty but also because the respondent doesn’t want to answer.