Page No.127-136
Debdulal Dutta Roy
Indian Statistical Institute, Kolkata
With rapid change in web technology, training method gradually shifts from manual
to web based computer adaptive training. Here, each trainee is not exposed to
same stimuli. Stimulus exposure depends upon individual differences in proficiency
level accounted by the computer. Therefore, usual validity estimating statistical
tools (ANOVA with repeated measurement) are not applicable. Aim of the current
research is to examine the criterion related validity of one web based computer
adaptive training program (Fast ForWord Language Gateway Edition Product).
The program is based on the adaptive training techniques such as frequency,
reward, intensity and motivation allowed for more rapid learning. It includes seven
modules. All modules were classified into two as sound (Circus sequence, old
McDonald flying farm, Phoneme identification) and Word (phonic match, phonic
words, language comprehension builder and block commander) exercises. This
internet based training program helps children in building oral language
comprehension and in other critical skills necessary for learning to read or
becoming a better reader. 25 randomly selected students of grades four and five
were trained with Fast ForWord program for more than 20 days. Results of each
individual trainee by treatments and by module in the form of percentage of success
were downloaded from the computer. In considering the typical data structure,
Box whisker plot analysis was made to compare the results across treatments or
trials. Several parameters of box plots (Outliers, box size, location of median in
the box, location of upper and lower whisker, location of hinges, and fluctuation
of box size after achieving the target) were identified as useful to estimate the
validity of training modules. It provides insight about different psychological changes
occurred during the different training programs. Box plot results suggest possible
ability contamination in changing performance level. Therefore, in order to control
contamination of ability on validity assessment, high and low ability groups were
identified based on initial training performance. Learning graph profiles of both
groups were compared finally. Results identified very poor validity in two modules
out of seven. Finally role of box whisker plot analysis on validity assessment was
discussed
.