How to support self-determined learners? (Part 2)

In the first episode about my personal learning experiences, I have reported that rigid curricula are not adequate to support experienced self-determined learners. This is not reduced to experienced academic learner but I believe one has to enlarge the above proposition to all adult learners who have work experiences and want to improve their knowledge and skills in their respective field of expertise.

Like me, these people want to use their present competencies to build on and do not want to be treated like blank slates. To support this kind of learners we need more flexibility. Isn't it strange that in the age of individualism we still use in education cohort learning like in the mechanic assembling line?

How to support self-determined learners?

To put my critical remarks about my personal learning experiences into affirmatory action some lesson learned for teachers may be drawn:

• Provide students as soon as possible with those meta-skills that are necessary to ask (the right) questions and to find support for their problems. This task was done in my Coursera experience quite well. There is for the R data science course on Coursera a special lesson Getting Help, where students learn how und where to get help. 
• Preparing lessons ask you the following questions:
  • What examples are representative for the intended learning outcomes?
  • What kind of meta-information are to provide so that student can advance on their own?
• Preparing exercises ask you the following questions:
  • What kind of basic knowledge should the exercises cover in order to acquire general skills to solve many problems of the same category?
  • How to ensure variability and flexibility in our task assignments?

How to ensure individualization?

How can we assure flexibility and variation with limited teaching resources? As a matter of fact, this question was one important motivation for me to start learning R and to look into data science. I want to explore the potential of learning analytics for individualized learning. But here, in this blog entry,  I will not talk about learning analytics but limit my suggestions to general teaching strategies.

In the Coursera MOOC on R Programming we had mainly three types of assignments:

• Using swirl: The swirlify R package provides a comprehensive toolbox for swirl instructors. Students can install the swirl R package with the standard <install.packages("swirl")> command to get a learning environment inside R and programmed in R. Although this seems to me a very interesting approach, the solution is focussed narrowly on one specific subject area: R programming. Furthermore, it was not a mandatory part of the course and had therefore not much relevance in the course design.  I will talk about different strategies for learning R in other episodes and will here discuss problems on a more general level which are independent of the subject area.
• Multiple Choice Test: This may seem boring, but actually, these tests are in the MOOC I attended pretty well done, very motivating and provided much room for flexibility. Students are provided with a real data set and get some specific questions to solve, like ranking hospitals according to certain characteristics. Students are free to find their own solution. In the test, we were asked questions like "What hospital in a specific state is on a specific rank according to a specific characteristic?" You couldn't answer this kind of questions without correct solutions. The different answer possibilities didn't give you a clue for the right answer. They did only help you to judge your solution: If your answer did not appear in the multiple answers possibilities then you knew that your programming solution was not correct and you could start to look for the error. This is a nice example how assignments could provide a new learning opportunity!– But still: This kind of assignment do not provide flexibility according to the background knowledge and working experiences of the learners. For that case, there had to be different data sets, for me, for instance, a data set from an educational subjet-matter, maybe data about schools to mention a concrete example.But how could we as teachers do the grading with so many different feasible tasks? We cannot judge as many different assignments as different interested students we do have in the class. Right, we – as teachers – have not the necessary resources, but how about the (self-determined) learner themselves? 
• Peer assignment: I have heard already many positive things about peer assignment in MOOCs were a multitude of people are enrolled. But the actual implementation as I had experienced it, was very disappointing. Instead to provide more adaptivity these peer assignments even restricted flexibility: I had to wait for other learners to finish their course work so that the would need my assignment as a peer. During this waiting time, my booked (and paid!) learning time was not stopped but continued to run. (The new business model in Coursera is not payment for finishing the course, but payment per month. I will discuss this model in another episode.)Another disappointment for me was, that all peer assignments I have experienced so far, were just formal judgments in the manner of: "Did the peer provide a readme file? Did s/he comment their programming functions? " We were explicitly instructed not to ran their programs and to see if they worked but had instead just to evaluate if their appearance and structure seemed to be correct!

Individualization through different model solutions

Under the premise to support the self-determined learner I think we could improve the role of peer assignment. I imagine a three step solution:

1. Teachers provide students with a concrete solution based on a real data set. This is not additional work as one needs the solution of a concrete example for the multiple choice test anyway.
2. Teachers provide students with a more abstract solution, e.g. a formal pattern of a solution. "Pattern" in this context is a specific notion, a terminus technicus, borrowed from the field of pattern theory inspired by Christopher Alexander. I cannot go into details of this approach here, so some pointers will have to suffice:
   1. Helmut Leitner: Pattern Languages and Christopher Alexander: Introduction and Crash Course (Video) 
   2. Helmut Leitner: Pattern Theory: Introduction and Perspectives on the Tracks of Christopher Alexander (Book)
   3. PURPLSOC: In Pursuit of Pattern Language of Societal Change (Conference)
3. Students are engaged in groups to check the solution of their peers. They fill their specific parameters into the model solution run the program to see if it works.

It may seem that this solution only works within a formalized learning subject, like programming languages. But I believe that patterns do work everywhere, even in the Humanities. In not formalized cases, teachers would have to provide several different model solutions. In comparing diverse model solutions with the own outcome, students would not only get a better understand of the problem but could also see what different approaches there are possible. 

How to build up a pool of different model solutions?

I know that the critical factor to realize my suggestion is the necessary pool of model solutions. Even if these solutions have to be worked out just once, it would need a lot of time and effort to develop them.

But here again, we could rely on the self-determined learner. Why not design the learning assignments in a way that they result in suggestions for new model solutions? Currently, there is much tutorial work required to answer all the different questions in the fora accompanying MOOCs. Part of this effort could be used to evaluate if some solution from students could be taken as model solutions. Even if they would still need some improvements or enhancements, they would not only aggregate in a relatively short time into a pool of different model solutions but would also provide learner from different subject areas with more motivating examples using their own field of expertise. 

This approach is not new but used anyway in all the fora I have seen. Tutors write special posts where they try to clear up common misunderstandings in a more general form. They could use part of their time to revise solutions from students. I do not mean, that we as teachers have to do this challenging task immediately and completely in the first course. No, I am thinking to build up this pool gradually. With every new course, we should look for another model solution with a different approach and from a different subject area.

It would be also inspiring for students to get the chance that their work on the assignments could be upgraded to a model solution. Course after course we would enlarge our database of model solutions which we could offer not only for the next course but also as additional learning material for public use (e.g. as OER = Open Educational Resources). 

Teaching self-determined learners: Summary

The following list summarizes my suggested procedure.

Activities before the first run of the course:

1. Provide material which helps students to build up meta-skills for asking (the right) questions and to find support for their problems.
2. Design a prototypical example with a solution which is representative for the intended learning outcomes.
3. Develop a more abstract model solution (pattern), which is suitable to serve as a template for comparisons. 
4. Research material from different subject areas which can be used as specific instantiations (e.g. applications or implementations) of the intended learning outcome.
5. Prepare this material in a way that it can be used as a starting point for a variety of task assignments from which students can choose from. (In the case of R programming this could be different data sets from one of the many different open repositories addressing a variety of subject areas, e.g. health, population, education, environment, governmental administration, traffic, economy, politics etc.)

Activities during the course:

1. Offer assignments with materials from different disciplines.
2. Offer students the model solution and explain how to use it.
3. Form groups of students for peer assessment.
4. Look into those evaluations which did not agree with their assessment and check if a bad design of your assignment is the reason.
5. Look into some  (maybe 10 or so) of the correct judged solutions provided by students and evaluate if they could be used as additional model solutions.
6. Find two to five solutions which could be used (maybe with some improvements) to function as additional model solutions for the next course.

Activities after the course:

1. Revise and improve your model solution when necessary.
2. Enhance some solutions you found in step 6 above.
3. Add this new material to your assignment pool and continue with step 5 of the course preparation with the new cycle of activities.