Internet Learning Volume 6, Number 2, Fall 2017/Winter 2018 | Page 9
Internet Learning Journal
ly findings. Swan and Bloemer (2013)
discuss the growth of informal blending
at a single institution and use descriptive
statistics to assess its efficacy.
Their research showed a positive correlation
between blending curriculum
and higher average course loads as well
as early completion for some groups
of students. They also showed that, for
students who seem to prefer on-ground
instruction, successful outcomes were
lower when they blended.
A much deeper analysis was recently
published in the journal Online
Learning. Using the Predictive Analytics
Reporting (PAR) data set consisting
of 656,000 student records, James,
Swan, and Daston (2016) were able to
use logistic regression to assess the odds
ratio of first-to-second year retention
by institution type while controlling for
a specific set of confounding variables
for students who were (a) fully online;
(b) fully on-ground; and (c) blended.
The results clearly and consistently indicated
that blending curriculum is
correlated with higher overall retention
rates than for fully on-ground or fully
online curricula. The effect was more
pronounced for students in community
colleges. Little difference was seen
between all on-ground student and
all-online students.
Methodology
The work of James, Swan, and
Daston provides an intriguing
opportunity to verify those
results using a different data set and
different research method and to further
explore some potential contributing
factors. This study uses the multiple
data sets and predictive models in
production from the company Civitas
Learning across a number of different
institutions of higher education. The institutions
that have deployed their predictive
analytics infrastructure range
across all sectors of U.S. higher education
to include large R1 institutions,
access-oriented 4-year universities,
community colleges, private liberal arts
institutions, as well as proprietary institutions.
Ranging from enrollments of
2,500 to well over 80,000, these schools
represent a good mix of urban and rural
and have a variety of technical systems
(i.e. student information systems (SISs)
and learning management systems
(LMS)). The online enrollments at these
institutions also vary widely, with some
being 100% online and some essentially
offering no online courses. Within
this mix of institutions, we are able to
find a set of institutions that represent
4-year, 2-year, and proprietary sectors
and which had significant populations
of students who are taking courses exclusively
on-ground, exclusively online,
and who mix modalities.
The Civitas Learning predictive
analytics process ingests large data
sets from the SIS and from the LMS at
each institution, which are federated,
segmented, and clustered. This transformed
data set is then used to produce
a set of predictive models that have
been shown to be highly accurate for
predicting any given student’s likelihood
to persist. Persistence here is defined
as re-enrolling in the next term or
successfully completing their program
of study at their enrolled institution in
8