Page 52 - Special Topic Session (STS) - Volume 4
P. 52

STS560 James Houran et al.
                  related than other procedures. Similarly, Rynes and Connerley (1993) found
                  that  the  interview  was  perceived  to  possess  high  job-relatedness.  Schuler
                  (1993) suggested that selection methods which are perceived as controllable
                  by the candidate, obvious in purpose, providing task relevant information, and
                  offering  a  means  of  feedback  are  considered  the  most  socially-valid  or
                  acceptable. We propose that the validity and effectiveness of the combined
                  approach in Figure 1 can be bolstered by empirically and procedurally aligning
                  assessments and interviews to work collectively as a seamless and integrated
                  process.  Using  insights  gained  from  psychometrics  and  data-mining,  we
                  outline  one  solution  that  uses  empirical  insights  gleaned  from  test-takers’
                  responses. This solution has a long (> 15 years) track record of actual usage in
                  the area of Human Resources (HR) testing. The approach relies heavily on the
                  statistical  machinery  provided  by  Item  Response  Theory  (IRT),  and  Rasch
                  scaling in particular (see van der Linden & Hambleton, 1997).

                  2.  An IRT Approach
                      The  Rasch  (1960/1980)  latent-trait  model  uses  the  simplest  IRT  where
                  items are described solely by their difficulty, at least for binary items. In the
                  case of rating scales, the structure of the rating scale, as described by transition
                  from one rating to another, has to be taken into account as well. The Rasch
                  rating scale model is probabilistic and revolves around the log odds of Pijk, i.e.,
                  the probability that item i will receive from person j the rating k. In particular,

                            Figure 2: Category structure of hypothetical item (see text).
























                                             
                                                        log (  ) =  −  − { }                                       (1)
                                          (−1)      

                  In Equation 1,




                                                                      41 | I S I   W S C   2 0 1 9
   47   48   49   50   51   52   53   54   55   56   57