Internet Learning Volume 1, Number 1, Fall 2012 | Page 28

Assessing Virtual Students’ Quiz Performance in Web Mediated Synchronous Instruction 27
facility with the virtual environment in the latter case. The final variable measured the
instructor’s experience with the technology, which could be expected to increase with use of the
webinar service in each successive semester. Such learning might have been enhanced by
reliance on a stable hardware configuration, but, as review of Table 1 shows, the technical
solution was in flux, although use of Elluminate as the webinar service was a constant
throughout the trial period.

Results
The statistical evidence of the model was provided through an ordered logistic regression,
“ordered logit,” which requires sequencing the dependent categorical variable. The ordered logit
model produces coefficients that estimate the factor’s impact on the likely value of the dependent
variable, all other factors being held equal. The validity of ordered logistic regression is also
based on the test of parallel lines, which relies on the assumption that the coefficient estimates do
not vary significantly depending on the level of the dependent variable. This assumption was
validated using a test of non-parallel lines. The result was failure to reject the null hypothesis
(parallel lines) at the p < .10 level, with a significance value of .185.
Table 4 contains the results of the ordered logistic regression. The performance of the
overall model is represented by χ2 of 40.222, which establishes a significant difference (p < .05)
between the model and the null hypothesis that all coefficients are zero. The amount of variation
accounted for by the model is shown using McFadden, Cox and Snell, and Nagelkerke statistics,
yielding values of .068, .081, and .038 respectively. Such values would be substandard levels for
multiple regression. Unfortunately, these are not comparable metrics to R2, which represents the
proportion of variation accounted for by a multiple regression model (Long, 1997, pp. 104-106).
Two of the independent variables, non-quiz GPA and attendance mode, had significant
associations with the dependent variable, quiz grade, which tended in opposite directions. The
negative sign of the non-quiz GPA estimate is interpreted to mean that the shift to an adjacent
grouping of average grades obtained on the financial assignments, for example from “A” (3.75 to
4) to “B+/A-” (3.25 to 3.74) or from “B+/A-” to “B” (2.75 to 3.24), was associ