Abstract
D.Phil.
This study was done in order to investigate the presence and functioning of construct, item,
and response bias across gender, ethnic, and language groups in a personality
questionnaire. The Basic Traits Inventory (Taylor & De Bruin, 2006) was used as the
personality assessment in this study, and is a South African-developed measure of the Big
Five personality factors. This study made use of both traditional methods based on
classical test theory and Rasch analysis from the item response theory genre. Comparison
groups based on gender, ethnicity, and home language were specified for the analyses. The
sample consisted of 6,112 students from a database of studies done using the Basic Traits
Inventory. There were 2,080 men and 3,104 women in the sample, of which 1,240 were
Black students and 1,139 were White students. The language groups were composed of
English-speaking (n = 1,739), Afrikaans-speaking (n = 1,648), and Indigenous African
language-speaking (n = 1,483) students. Some students did not indicate biographic details.
The reliability of the Basic Traits Inventory was evaluated using both Cronbach’s alpha
reliability coefficient and the person separation index (PSI) from the Rasch analysis. Both
methods revealed similar indices of internal consistency. For the Big Five factors of the
Basic Traits Inventory, the reliability estimates were similar across methods, and deemed
satisfactory for the Extraversion (B = 0.90; PSI = 0.89), Neuroticism (B = 0.94; PSI =
0.93), Conscientiousness (B = 0.94; PSI = 0.92), Openness to Experience (B = 0.88; PSI =
0.85), and Agreeableness (B = 0.88; PSI = 0.86) scales. Three facet scales, namely
Openness to Values, Straightforwardness, and Modesty, showed consistently lower than
acceptable Cronbach alpha values across the comparison groups, indicating that scores on
these facets should be interpreted with caution. From the Rasch analysis of each of the
factors of the Basic Traits Inventory, it emerged that 35 of the 180 items showed some
evidence of misfit, and specifically underfit. Of the 35 misfitting items, only 10 items
showed signs of extreme underfit.
iv
There was very little evidence for item bias across all groups on each of the five factors of
the Basic Traits Inventory. For the gender groups, there were three items with DIF contrast
values larger than 0.5 logits across all five factors. For the ethnicity groups, eight of the
items showed DIF contrast values larger than 0.5 logits. Only three items met the criteria
for item bias in the language groups. Items O2 and O23 were judged to show item bias in
both the ethnicity and language groups, and should be removed from future versions of the
Basic Traits Inventory.