What is Learning Analytics and what can it ever do for me?

Putting up definitional fences and looking for connections.
Putting up definitional fences and looking for connections

#DALMOOC’s week 1 competency 1.2 gave me an excuse explore some definitions.

As a scientist, the insistence on using the term “analytics” as opposed to “analysis” I found intriguing…The trusty Wikipedia explained that analytics is “the discovery and communication of meaningful patterns in data” and has its roots in business. It is something wider (but not necessarily deeper) than data analysis/statistics as I am used to. Much of it is focused on visualisation to support decision-making based on large and dynamic datasets – I imagine producing visually appealing and convincing powerpoint slides for your executive meeting would be one potential application…

I was glad to discover that there are some voices out there which share my concern over the seductive powers of attractive and simple images (and metrics) – here is a discussion of LA validity on the EU-funded LACE project website; and who has not heard about the issues with Purdue’s Course Signals retention predictions? Yet makers of tools such as Tableau (used in week 2 of this course) emphasise how little expertise one needs to use them to look at the data via the “visual windows”… The old statistical adage still holds – “garbage in – garbage out” (even though some evangelists might claim that in the era of big data statistics itself might be dead;). That’s enough of the precautionary rant…;)

I liked McNeill and co.’s choice of Cooper’s definition of analytics in their 2014 learning analytics paper (much of it based on CETIS LA review series):

Analytics is the process of developing actionable insights through problem definition and the application of statistical models and analysis against existing and/or simulated future data (my emphasis)

It includes the crucial step in looking at any data in applied contexts – simply asking yourself what you want to find out and change as a result of looking at it (the “problem definition”). And the “actionable insights” – a rather offensive management speak to my ears – but nonetheless doing something about it seems rather an essential step in closing any learning loop.

The, currently, official definition of Learning Analytics came out of an open online course in Learning and Knowledge Analytics 2011 and was presented at the 1st LAK conference (Clow, 2013):

“LA is the measurement, collection, analysis and reporting of data about learners and their contexts, for purposes of understanding and optimising learning and the environments in which it occurs.”

This is the definition used in the course – the definition we are encouraged to examine and redefine as this very freshly minted field is dynamically changing its shape.

Instantly I liked how the definition is open on two fronts (although that openness seems to be largely in the realm of aspirations than IRL practice, but is not surprising, given the definition’s origins):

1. It does not restrict data collection to the digital traces left by the student within Learning Management Systems/Virtual Learning Environments so it remains open to data from entirely non-digital contexts. Although in reality, the field really grew out of and, from what I can see, largely remains in the realm of big data generated by ‘clicks’ (whether it be VLEs or games or intelligent tutoring systems). The whole point really is that it relies on data collected effortlessly (or economically) – compared to traditional sources of educational data, such as exams. What really sent a chill down my spine is the idea fleetingly mentioned by George Siemens in his intro lecture for this week – extending the reach outside of the virtual spaces via wearable tech…. So as I participate in the course I will be looking out for examples of marrying the data from different sources. I will also look out for dangers/side effects of focusing on what can be measured rather than what should be measured. I feel that the latter may be enhanced by limiting a range of perspectives involved in development and application of LA (to LA experts and institutional administrators). So keeping an eye out for collaborative work on defining metrics/useful data between LA and educational experts/practitioners, and maybe even LDs is also on the list (Found one neat example already via LACE SOLAR flare UK meet, which nicely coincided with week 1 –  Patricia Charlton speaks for 3 min about the mc2 project starting at 4.40 min. The project helps educators to articulate many implicit measures used to judge student’s mathematical creativity. Wow – on so many levels!).

2. It is open to who is in control of data collection (or ownership) and use – institution vs the educator and the learner. I was going to say something clever here as well but it’s been so long since I started this post, it’s all gone out of my head (or just moved on). I found another quote from McNeill and co., what is relevant here:

Different institutional stakeholders may have very different motivations for employing analytics and, in some instances, the needs of one group of stakeholders, e.g. individual learners, may be in conflict with the needs of other stakeholders, e.g. managers and administrators.

It sort of touches on what I said under point 1 – need for collaboration within an institution when applying LA. But it also highlights the students as voices of importance in LA metric development and application. It is their data after all so they should be able to see it (and not just give permission for others to use it) and they are supposed to learn how to self-regulate their learning and all…Will I be able to see many IRL examples of such tools made available to students and individual lecturers (beyond the simple warning systems for failing students such as Course Signals)? There was a glimmer of hope for this from a couple of LACE SoLAR flare presentations. Geoffrey Bonin talked about Pericles project and how it is working to provide a recommendation system for open resources in students’ personal learning environments (at 1 min here). Or rather more radical, Marcus Elliot (Uni of Lincoln) working on a Weapons of Mass Education project to develop a student organiser app going beyond institution giving students access to the digested data and involving research project around student perceptions around learning analytics and what institutional and student-collected data they find most useful – data analytics with students not for students (at 25 min here).

(I found Doug Clow’s analysis of LA in UK HE re: institutional politics and power play in learning very helpful here and it was such a pleasant surprise to hear him speak in person at the LACE Solar flare event!)

The team’s perspective on the LA definition was presented in the weekly hangout (not surprisingly, everybody had their own flavour to add) – apologies for any transcription/interpretation errors:

  • Carolyn (the text miner of forums): Defined LA as different to other forms of Data Mining as focussing on the learning process and learner’s experiences. Highlighted the importance of correct representation of the data/type of data captured for the analysis to be meaningful in this context vs e.g. general social behaviours.
  • Dragan (social interaction/learning explorer): LA is something that helps us understand and optimise learning and is an extension (or perhaps replacement) of the existing things that are done in education and research, e.g. end of semester questionnaires no longer necessary as can see all ‘on the go’. Prediction of student success is one of the main focuses of LA but it is more subtle – aimed at personalising learning support for success of each individual.
  • Ryan (the hard-core data miner who came to the DM table waaay ahead of the first SOLAR meet in 2011, his seminal 2009 paper on EDM is here):  LA is figuring out what we can learn about learners, learning and settings they are learning in to try to make it better. LA is about beyond providing automated responses to students but LA also focuses on including stakeholders (students, teachers and others) in communication of the findings to them.

So – a lot of insistence on focus on learners and learning…implying that there are in fact some other analytics foci in education. I just HAD TO have a little peak at the literature around the history of this field to better understand the context and hence the focus of the LA itself (master of procrastination reporting for duty!).

Since I have gone beyond the word count that any sensible co-learner may be expected to read, I will use a couple of images which do illustrate key points rather more concisely.

Long and Siemens’ Penetrating the fog – analytics in learning and education provides a useful differentiation between learning analytics and academic analytics, the latter being closer to business intelligence at the insititutional level (this roughly maps onto the hierarchical definition of analytics in education by Buckingham and Shum in their UNESCO paper – PDF):

Learning Analytics vs Academic AnalyticsI appreciate the importance of such “territorial” subject definitions, especially in such an emerging field, with the potential of being kidnapped by educational economic efficiency agenda prevailing these days. However, having had an experience of running courses within HE institutions, I feel that student experience and learning can be equally impacted by BOTH, the institutional administration processes/policy decisions AND the quality of teaching,course design and content. So I believe that joined up thinking across analytics “solutions” at all the scales should really be the holy grail here (however unrealistic;). After all there is much overlap in how the same data can be used at the different scales already. For that reason I like the idea of unified concept of Educational Data Sciences, with 4 subfields, as proposed by Piety, Hickey and Bishop in Educational data sciences – framing emergent practices for analytics of learning organisations and systems (PDF). With one proviso – it is heavily US-focused, esp at >institution level. (NOTE that the authors consider Learning Analytics and Educational Data Mining to belong in a single bucket. My own grip on the distinction between the two is still quite shaky – perhaps discussion for another post)

educationaldataanalysistaxonomyI would not like to commit myself to a revised LA definition yet – I shall return to it at the end of the course (should I survive that long) to try to integrate all the tasty tidbits I collect on the way.

Hmm – what was the assignment for week 1 again? Ah – the LA software spreadsheet….oops better get onto adding some bits to that:)


Headline image source: Flickr by Pat Dalton under CC license.


Treating myself to the dual layers of #DALMOOC with EdX

Big data's brotherly gazeJust as I was thinking of getting back to the founts of MOOCy goodness last week, Twitterous serendipity occurred yet again and voila, I am now enrolled on the Data Analytics and Learning course from Texas University Arlington via EdX’s Honour Code route no less. The course proper does not commence until Monday (so still time to follow me in;) but we’ve already been treated to some induction materials over the weekend…(Available separately on G+ and YouTube).

My discerning MOOCer palate has been tempted this time for two reasons:

  • I find institutional/governmental  collection of vast amounts of personal data and their use of data analytics to “improve my experience” extremely creepy – in a Big Brother 1984 way. If you know me from PLN seminar, you know my tendency for such doomsday scenarios;). But I am also a (recovering)scientists and so it is almost impossible for me to refuse a chance to play with some numbers and new analytical tools.
  • The MOOC design itself is intriguing – very explicitly trying to combine the more usual, linear xMOOC paradigm with the more open, social-learning-focused cMOOC. I have sampled a range of courses aiming for a version of the latter, so again I could not resist having a taste here. Especially that these guys are trying out some new tools to facilitate social interaction within both models. Oh – and since the “social learning” aficionado (and, apparently, also a learning data analytics guru), George Siemens (@gsiemens) , is the lead here, we are guaranteed an interesting ride!

In the words of the man himself (there will no doubt be more words on the topic on his elearnspace blog here):

“I think that in the MOOC landscape we too prematurely settled on the instructional model that we have and we really want to take an opportunity with this course to ask a range of questions and experiment with different ways of making learning happen in different contexts. So we are experimenting with social learning, with different support structures and software…” (DALMOOC Induction video 1)

Shouldn’t forget George’s collaborators on the project:

  • Carolyn Rose of Carnegie Melon Institute (innovator in such funky topic as Automated Analysis of Collaborative Learning Processes and Dynamic Support for Collaborative Learning and the person behind Bazaar and Quick Helper support system implementation within the course’s structured EdX platform. Having had designed and supported collaborative work online before – this is certainly of interest to me:)
  • Dragan Gasevic (@dgasevic) of Athabasca University (into applying semantic web principles to elearning systems and a father to the newly minted “credentialing pipeline” Prosolo tool to be used in the social layer of the course)
  • Ryan Baker (@BakerEDMLab) of Columbia University (looking at data mining intersection with human-computer interaction, and seemingly particularly interested in student’s motivation, or rather lack thereof, e.g. “WTF behaviour”, in the interaction with elearning systems).

The induction materials so far have been heavily focused on showing us around the dual layer course model and the introduction to the learning tools expanding the usual EdX set of forums and videos.

DALMOOC Visual SyllabusCentral to this introduction is “visual syllabus” designed as an intuitive overview of the complex course design and an introduction to the less-traditional social learning layer of the course by Matt Crosslin (@grandeped) of LINK Research Lab. Great idea but perhaps needs a tweak or two:

  • Less emphasis on esthetics of the design and more on information – e.g. including header text alongside more informative images so that we actually do get an overview without having to roll over pics?
  • As Matt explains in the induction hangout, the aim is to particularly focus on introducing the non-traditional, unstructured, “social learning” layer of the course, getting away from assumptions of learner’s zero prior knowledge at the point of entry. Yet the learner’s progress through the overview is highly structured through the prominent numbering of the sections, therefore falling back onto the traditional paradigm and assumptions of linear progress through materials. Adding text headers, even alongside the numbers, would instantly change this first impression and allow for choice of point of entry, especially for those learners who have already dabbled in non-traditional courses before;)

A brief comment on the first impressions of the tools we are going to be using to support our learning:

  • In the structured/linear layer (the blue pill) we have a couple of tools, Quick Helper and Bazaar, which seem to target the usual problem with massive courses – the sheer volume of people and messages and lack of more intimate collaborative learning experience which in f2f sessions may be achieved by simply turning to your neighbour in the classroom. Both are using automation, text analysis and algorithmic approaches to facilitate interactions. My immediate reaction (and some of my colleagues I discussed this with) is that while the tools may indeed help those students with less sophisticated online collaboration skills to find support within the EdX system, they do nothing for their online networking skill development. A couple more specific issues with each. Quick Helper’s “help matching intervention” system of targeting your forum questions to specific students/helpers may result in undue workload for students algorithm deems “an expert” and resentment if help is not provided (perhaps allow people to opt out/into the helper role?). Bazaar allows for spontaneous creation of collaborative/discussion groups, discussion aided by “virtual agent”. As @gsiemens pointed out in the induction hangout – it is a bit like Chatroulette, but with less nudity…Well, the analogy pretty much says it all.
  • So – not that excited about the aids to the structured layer of the course but then, I usually tend to live outside the course VLEs anyway, in the red pill territory. The course’s expectation of setting up and using our own learning and networking spaces is more up my alley:) and I am a bit excited about using the prototype of “credentialing pipeline”, Prosolo, which is supposed to help us create, share and assess each others’ artifacts and form interest networks.

More on the course design etc. from the horses’ mouths:

Now – despite all this “constructive criticism” – I do look forward to taking part. No doubt I will be pleasantly surprised…off to try my hand at the educational Chatroulette:)

Image source: Diodoro under CC license