Effect of outliers on standard characteristics according to item response theory

Karrar Ghalib Jader Altaweel


The current research aims at the effect of extreme values on the standard
characteristics according to the paragraph response theory . To achieve this goal ,
the researcher followed precise scientific steps . The researcher adopted a test
Abstract inference prepared by the educational institution (pmteducation) in
(2019), and followed the procedures for preparing it according to the paragraph
response theory (the two-parameter model), as the researcher verified the
translation validity procedures for the test.
And after completing the translation validity procedures , and to verify the
descriptive validity of the test, which consists of: (55) paragraphs , The researcher
presented the test to a group of arbitrators specialized in psychological sciences,
measurement and evaluation, their number is (14) arbitrators , and no paragraph
of the test was modified or excluded because it obtained a percentage of agreement
(100%), and thus the descriptive validity of the test was confirmed.
To verify the clarity of the test instructions and paragraphs The test was
applied to a sample of (10 0. ). (A male and a female student chosen at random
from the middle school students - for the fourth and fifth grades) scientific and
literary) And it became clear from this experience that the instructions and
paragraphs of the test are clear and understandable.
For the purpose of statistical analysis of the paragraphs of the abstract
inference test and extracting its standard characteristics , the test was applied to
a sample consisting of (1000 A male and female student from the fourth and fifth
grades (scientific and literary) of Najaf Governorate, were chosen. b , by random
sample method , and thus the researcher obtained a file containing (1000)
participants in the case of (extreme values), and (903) participants in the case of
(absence of extreme values), i.e. after deleting the (97) extreme values , (34) ofwhich are a high extreme value and (63) a minimum extreme value for the purpose
of measuring the effect of those values on the standard characteristics (honesty,
difficulty, discrimination, stability) according to the traditional measurement theory
and paragraph response theory.
The test was prepared according to the paragraph response theory,
specifically (the two-parameter model), which contains (55) paragraphs with three
3 alternatives and some paragraphs with four 4 alternatives in its initial form, and
after verifying the factor analysis, one paragraph 1 of the sequence (3) of the test
paragraphs And two paragraphs of the sequence (25, 43) were tested for fit to
square Ka2, and the number of test items became (52) in its final form and adopted
in the final statistical analysis and for the two cases (the presence of extreme
values) and (the absence of extreme values) according to the paragraph response
The statistical analysis according to For the paragraph response theory , it
was done through the researcher’s dependence on the two-teacher model as one
of the paragraph response theory models in analyzing the paragraphs of the
abstract inference test, as calculated by the Bilog -mg3 program The coefficients
of (honesty, difficulty, discrimination, stability) of the paragraphs were calculated,
and the extent of the influence of extreme values on them was measured, once
(with the presence of extreme values) for (1000) participants, and another (without
the presence of extreme values) for (903) participants.
To achieve the model assumptions, the researcher followed the following:
To verify the one-dimensional assumption, the researcher subjected the test
to factor analysis using the basic components method, as one meaningful factor
was obtained for the test. 1), and the adoption of a saturation percentage (0.30)
to accept the saturation of each paragraph of the tests with the general factor ,
and (3) paragraphs of the sequence (3, 11, 15) were deleted in the case of (the
absence of extreme values) because the saturation of the paragraphs was less than
( 0.30), and the second indicator is the relationship of the paragraph's degree with
the total degree to verify one-dimensionality.
In addition to matching paragraphs to a model Two-parameter is
evidence that the items measure a one-dimensional trait , based on the value
of the chi-square at the level of significance (0.05 ) And as calculated by the
program, and accordingly it was done exclusion A paragraph introduced by
the program, and two (2) because it is a function according to the value of
the chi-square, and the analysis was re-analyzed again and the paragraphs
whose difficulty exceeded (- 2.5) to (+ 2.5), as well as the paragraphs whose
discrimination coefficient exceeded (5, 0) to me (5, 2 And its number was (3)
paragraphs The validity of the test was verified through (descriptive honesty,
functional honesty), and the reliability of the test, and in light of the results
of the current research, the researcher reached some conclusions,
recommendations and suggestions