EXAMINING THE RELATIONSHIP BETWEEN THE RASCH AND 2-PL MODELS OF IRT IN SELECTING ITEMS IN A CONSTRUCTED TEST FOR CURRICULUM DEVELOPMENT

Abstract

Test developments are critical in the measurement of learning and
education attainments. Its precision and accuracy is a major task
in an academic setting in Nigeria and beyond. The scales
employed for achievement tests are often based on classical test
theory (CTT) approach with a drawback of the variability of
results in different samples of the same population or from the
same pool of items in some Africa countries like Nigeria. This
work explores the method of the Rasch model and the 2-PL model
of Item Response Theory (IRT) to examine the relationship
between the models used on a constructed Mathematics Aptitude
Test (MAT) items. A 120 items instrument with a reliability value
of 0.86 was developed by the researcher. Mean Square (MNSQ)
and ZSTD of fitness of Winsteps and Two-Parameter Model (2-
PL) of Bilog-Mg3 were used to investigate how well the
Mathematics fit the Models. Eventually, thirty-three (33) items
whose parameters are known scaled through the Rasch model and
were confirmed to measure the same construct (unidimensionality) while 13 items are significant but do not fit into
the 2-PL model at p< 0.05. Rasch shows that only 33 items fit into
the model while the 2-PL model shows that 107 items fit into its
model. This shows that a great disparity occurs between Rasch
and 2-PL model. These items are banked for curriculum
development purposes. This paper concludes that the Rasch model
is preferred to the 2-PL model and therefore recommends its usage
in curriculum development in Nigerian schools and Examination
bodies in Africa and beyond.