Abstract
This chapter introduces the use of basic time-to-event analysis (a variation of “survival analysis”) to identify time-series patterns from learning management system (LMS) data portal datasets to enable empirical-based theorizing and interpretation. This approach addresses questions such as How long does it usually take before a particular event occurs? What time patterns may be seen in empirical data? What sorts of analysis and decision making can be understood from the time patterns? This chapter uses multiple datasets—related to assignment submittals and their time to grading, learner enrollments and the updates to those enrollments, and group membership and how long groups last, and other data—to demonstrate this process.
TopIntroduction
In observational research, there are some basic givens—such as time and space—because most human events occur in both. “Survival analysis” and its variations (such as time-to-event analysis, event history analysis, reliability analysis, failure time models, and duration analysis) use time as a core element. To simplify, how much time passes before…
The prior are over-simplifications, of course. But they capture something of the basic concept. Time, of course, is both a real phenomenon and a construct. In terms of research, time may be conceptualized as continuous (occurring on a continuum) or as discrete (broken into specific and defined intervals).
Survival analysis measures the time until a defined “event” (a defined occurrence) occurs. In its earlier incarnations, “survival analysis” measured the time until non-survival or death, in the medical science realm. The survival rates between various cohorts were compared, such as those who received a particular intervention and a control group that did not. Or the cohorts could be divided up based on demographic features. Such analyses were also applied in the actuarial sciences in a predictive way based on mortality rates (“hazard functions” in survival analysis), to understand the likelihood of survival for any individual at a particular time based on various descriptive features of their lives. In time series data, the “hazard rates” for non-survival in physical beings and objects are generally thought to increase with time and age (and these are observed as non-decreasing data points, with some plateaus in which no events of non-survival occur).
Beyond the applications in biomedicine and healthcare and actuarial sciences, survival analysis techniques have been applied to understand when inanimate objects may degrade and fail, in fields like engineering, manufacturing, safety, emergency preparedness, and others. In organizational psychology, survival analysis has been applied to the length of people’s tenure in professional roles and their likelihood of moving on and when.
This work explores using a basic time-to-event analysis on three data tables from the Canvas Learning Management System data portal at Kansas State University. The data is from the LMS which was initially stood up in test version in Fall 2013, and the cumulative data was downloaded in late March 2018. The three sets are submission_dim for assignment submissions, enrollment_dim for enrollment data, and group_membership_dim for group memberships. These are not N = all except for one of the sets because these, in total, would involve millions of rows of data, which do not show well on some of the line graphs and which are highly demanding on computational processing. These three are used to show some variations on the application of time-to-event analysis to online educational data for some fresh insights. The software used here include IBM’s Statistical Package for the Social Sciences (SPSS) and Microsoft Excel 2016. To open the .gz files from the LMS data portal, 7Zip was used to extract the necessary files.
TopReview Of The Literature
Researchers have also suggested the power of using survival analysis on educational data to study event timing and duration (Singer & Willett, Summer 1993). Time-to-event analysis applies fairly well to online learning data because of the precision of time data from online learning platforms and the relevance of such insights to understanding and decision-making. While in other cases, the researcher needs to define the objective observable details needed to define an event’s occurrence, the data from a learning management system (LMS) data portal comes generally pre-defined in the structured datasets [Canvas LMS Data Portal Data Dictionary (Ver. 2.0.0), 2018]. “Big data” from LMS data portals may offer a range of insights about the technology, the people using it, and time-based insights at both high levels and at close-in views (Hai-Jew, Spring/Summer 2017). As yet, however, this statistical analysis method has not caught on widely.
Key Terms in this Chapter
Lost to Follow-Up: A term used to explain data censoring.
Predictive: Forecasting a forthcoming event, anticipating the frequencies of particular data.
Left-Censoring: The data lost to observation because the event occurred prior to the start of the observational research period.
Exploratory: An investigative approach to data to see what patterns and insights might be there albeit without necessarily any basic hypothesis.
Unit of Time (“Spell”): The time interval used in the statistical analysis (or time-to-event analysis).
Onset: The beginning of an event or incident.
Censored Data: Data that is not captured during observational research; data lost to observation (and therefore resulting in an incomplete record).
Unobserved Interval: An unknown period of time or time interval (lost to observation).
Discrete Time: The understanding of time as particular and defined intervals.
Survival Analysis: A statistical technique that involves the measuring of time until a target event.
Continuous Time: The understanding (or depiction or representation) of time as occurring on a continuum.
Event: An objectively observable and recorded incident.
Observed Data: Records and facts that are objectively viewable, often used in evidence-based and/or observational research.
Line Chart: A data visualization that portrays data with straight line segments.
Right-Censoring: The data lost to observation because the event occurred (or not) after the end of the observational research period.
Stepwise: A serious of steps or stages, not continuous.