Internet Learning Volume 6, Number 2, Fall 2017/Winter 2018 | Page 10

Better Together: How Blending Course Modalities Impacts Student Persistence
the term being studied. Since the data
are specific by institution and de-identified,
we cannot track students who
transfer to another institution prior to
completing their credential.
The native SIS and LMS data,
once ingested and federated, are then
used to create an additional set of derived
variables. Each of these variables,
often well over 1,000, is then assessed
for predictiveness via combinatorial
feature optimization. The output
of that process determines which features
remain in the predictive models
(Kil & Shin, 1996). The models thus
built using the specific and unique set
of institutional data create an analytics
infrastructure which is then used
to understand specific factors—in
combination—and how those factors
predict student persistence. Understanding
both historical and predicted
persistence rates by population
sub-grouping creates the opportunity
for pursuing specific research questions.
It also creates the infrastructure
on which to initiate direct student outreach,
nudges, campaigns, policy or
curriculum adjustments, with the goal
of increasing student success. This process
is repeated for each institution using
the Civitas Learning product suite
from the ground up, such that each
institution’s models are uniquely based
on their data sources, data breadth,
institutional mission, mix of students,
available resources, and policies. A
thorough explanation of this process
can be found in Milliron, Malcom,
and Kil (2014), and in McIntosh and
Robinson (2016). With well over seven
million enrolled students and over
20 million student records across the
deployed set of Civitas Learning institutions,
representing all sectors of U.S.
higher education, we are able to use
this platform as a research opportunity.
This study investigates the differences
in persistence rates for students
across three different course-taking
behaviors: all classroom instruction
(i.e. on-ground), all online instruction,
and students who mix on-ground and
online courses. These three student
populations are examined across a set
of institutions, and their historical persistence
rates and top data features and
feature values are compared. The set
of institutions reviewed include four
selective 4-year institutions, four access-oriented
community colleges, and
three proprietary primarily online institutions.
This set of institutions was
selected for their likelihood to have
statistically significant numbers of students
representing each of the three
modalities of interest. For the primarily
online institutions included in this
study, on-ground courses were also
offered and the number of students
blending their curriculum was significant.
The analysis conducted examines
individual institutions’ populations
by course-taking behavior to understand
correlations between those behaviors
and the predicted persistence
risk. Those individual institutions’
persistence correlations are then compared
among the set of institutions to
highlight if the general findings hold
across different student populations—
understanding that fully online students
may look very different than
fully on-ground students at different
colleges and universities.
9