MOOC measurement problems reveal systemic evaluation challenges

Sadly, it’s not particularly surprising that it took a proclamation by researchers from prominent institutions (Harvard and MIT) to get the media’s attention to what should have been obvious all along. That they don’t have alternative metrics handy highlights the difficulties of assessment in the absence of high-quality data both inside and outside the system. Inside the system, designers of online courses are still figuring out how to assess knowledge and learning quickly and effectively. Outside the system, would-be analysts lack information on how students (graduates and drop-outs alike) make use of what they learned– or not. Measuring long-term retention and far transfer will continue to pose a problem for evaluating educational experiences as they become more modularized and unbundled, unless systems emerge for integrating outcome data across experiences and over time. In economic terms, it exemplifies the need to internalize the externalities to the system.

Advertisements

On the realistic use of teaching machines

From the perspective that all publicity is good publicity, the continued hype-and-backlash cycle in media representations of educational technology is helping to fuel interest in its potential use.  However, misleading representations, even artistic or satirical, can skew the discourse away from realistic discussions of the true capacity and constraints of the technology and its appropriate use. We need honest appraisals of strengths and weaknesses to inform our judgment of what to do, and what not to do, when incorporating teaching machines into learning environments.

Adam Bessie and Arthur King’s cartoon depiction of the Automated Teaching Machine convey dire warnings about the evils of technology based on several common misconceptions regarding its use. One presents a false dichotomy between machine and teacher, portraying the goal of technology as replacing teachers through automation. While certain low-level tasks like marking multiple-choice questions can be automated, other aspects of teaching cannot. Even while advocating for greater use of automated assessment, I note that it is best used in conjunction with human judgment and interaction. Technology should augment what teachers can do, not replace it.

A second misconception is that educational programs are just Skinner machines that reinforce stimulus-response links. The very premise of cognitive science, and thus the foundation of modern cognitive tutors, is the need to go beyond observable behaviors to draw inferences about internal mental representations and processes. Adaptations to student performance are based on judgments about internal states, including not just knowledge but also motivation and affect.

A third misconception is that human presence corresponds to the quality of teaching and learning taking place. What matters is the quality of the interaction, between student and teacher, between student and peer, and between student and content. Human presence is a necessary precondition for human interaction, but it is neither a guarantee nor a perfect correlate of productive human interaction for learning.

Educational technology definitely needs critique, especially in the face of its possible widespread adoption. But those critiques should be based on the realities of its actual use and potential. How should the boundaries between human-human and human-computer interaction be navigated so that the activities mutually support each other? What kinds of representations and recommendations help teachers make effective use of assessment data? These are the kinds of questions we need to tackle in service of improving education.

Expensive assessment

One metric for evaluating automated scoring is to compare it against human scoring. For some domains and test formats (e.g., multiple-choice items on factual knowledge), automation has an accepted advantage in objectivity and reliability, although whether such questions assess meaningful understanding is often debated. With more open-ended domains and designs, human reading is typically considered superior, allowing room for individual nuance to shine through and get recognized.

Yet this exposé of some professional scorers’ experience reveals how even that cherished human judgment can get distorted and devalued. Here, narrow rubrics, mandated consistency, and expectations of bell curves valued sameness over subtlety and efficiency over reflection. In essence, such simplistic algorithms resulted in reverse-engineering cookie-cutter essays that all had to fit one of their six categories, differing details be damned.

Individual algorithms and procedures for assessing tests need to be improved so that they can make better use of a broader base of information. So does a system which relies so heavily on particular assessments that the impact of their weaknesses can get magnified so greatly. Teachers and schools collect a wealth of assessment data all the time; better mechanisms for aggregating and analyzing these data can extract more informational value from them and decrease the disproportionate weight on testing factories. When designed well, algorithms and automated tools for assessment can enhance human judgment rather than reducing it to an arbitrary bin-sorting exercise.

Beating cheating

Between cheating to learn and learning to cheat, current discourse on academic dishonesty upends the “if you can’t beat ’em, join ’em” approach.

From Peter Nonacs, UCLA professor teaching Behavioral Ecology:

Tests are really just measures of how the Education Game is proceeding. Professors test to measure their success at teaching, and students take tests in order to get a good grade.  Might these goals be maximized simultaneously? What if I let the students write their own rules for the test-taking game?  Allow them to do everything we would normally call cheating?

And in a new MOOC titled “Understanding Cheating in Online Courses,” taught by Bernard Bull at Concordia University Wisconsin:

The start of the course will cover the basic vocabulary and different types of cheating. The course will then move into discussing the differences between online and face-to-face learning, and the philosophy and psychology behind academic integrity. One unit will examine the best practices to minimize cheating.

Cheating crops up whenever there is a mismatch between effort and reward, something which happens often in our current educational system. Assigning unequal rewards to equal efforts biases attention toward the inflated reward, motivating cheating. Assigning equal rewards to unequal efforts favors the lesser effort, enabling cheating. The greater the disparities, the greater the likelihood of cheating.

Thus, one potential avenue for reducing cheating would be to better align the reward to the effort, to link the evaluation of outputs more closely to the actual inputs. High-stakes tests separate them by exaggerating the influence of a single, limited snapshot. In contrast, continuous, passive assessment brings them closer by examining a much broader range of work over time, collected in authentic learning contexts rather than artificial testing situations. Education then becomes a series of honest learning experiences, rather than an arbitrary system to game.

In an era where students learn what gets assessed, the answer may be to assess everything.

What should we assess?

Some thoughts on what tests should measure, from Justin Minkel:

Harvard education scholar Tony Wagner was quoted in a recent op-ed piece by Thomas Friedman on what we should be measuring instead: “Because knowledge is available on every Internet-connected device, what you know matters far less than what you can do with what you know. The capacity to innovate—the ability to solve problems creatively or bring new possibilities to life—and skills like critical thinking, communication and collaboration are far more important than academic knowledge.”

Can we measure these things that matter? I think we can. It’s harder to measure critical thinking and innovation than it is to measure basic skills. Harder but not impossible.

His suggestions:

For starters, we need to make sure that tests students take meets [sic] three basic criteria:

1. They must measure individual student growth.

2. Questions must be differentiated, so the test captures what students below and above grade-level know and still need to learn.

3. The tests must measures [sic] what matters: critical thinking, ingenuity, collaboration, and real-world problem-solving.

Measuring individual growth and providing differentiated questions are obvious design goals for personalized assessment. The third remains a challenge for assessment design all around.

Automating assessment: How, when, and why?

EdX, the most prominent nonprofit MOOC provider, plans to use and share automated software to grade and give feedback on student essays. On the heels of this announcement come legitimate skepticism about how well computers actually grade student work (i.e., “Can this be done?”) and understandable concern whether this is a worthwhile direction for education to proceed (i.e., “Should this be done?”). Recasting these two questions in terms of how, when, and why to apply automated assessment yields a more critical framework for finding the right balance between machine-intelligent and human-intelligent assessment.

How?

When evaluating the limitations of artificial intelligence, I find it helpful to ask whether they can be classified as issues with data or algorithms. In some cases, available data simply weren’t included in the model, while in others, such data may be prohibitively difficult or expensive to capture. The algorithms contain the details of how data get transformed into predictions and recommendations. They codify what gets weighted more heavily, which factors are assumed to influence each other, and how much.

Limitations of data: Train on broader set of sample student work as inputs

Todd Pettigrew describes some familiar examples of how student work might appear to merit one grade on the surface but another for content:

it is quite common to see essays that are superficially strong — good grammar, rich vocabulary — but lack any real insight… Similarly some very strong essays—with striking originality and deep insight—have a surprising number of technical errors that would likely lead a computer algorithm to conclude it was bad.

This highlights the need to train the model on these edge cases to distinguish between style and substance, and to ensure that it does not false-alarm on spurious features. Elijah Mayfield points out that a training set of only 100 hand-graded essays is inadequate; this is just one example of the kind of information such a small sample could fail to capture adequately.

Limitations of data: Include data from beyond the assignment and the course

Another relevant concern is whether the essay simply paraphrased an idea from another source, or if it included an original contribution. Again from Todd Pettigrew:

the computer cannot possibly know how the students answers have related to what was done elsewhere in the course. Did a student’s answer present an original idea? Or did it just rehash what the prof said in class?

Including other information presented in the course would allow the model to recognize low-level rehashing; adding information from external sources could help situate the essay’s ideas relative to other ideas. A compendium of previously-expressed ideas could also be labeled as normative (consistent with the target concepts to be learned) or non-normative (such as common misconceptions), to better approximate the distance between the “new” idea and “old-but-useful” ideas or “old-but-not-so-useful” ideas. But confirming whether that potentially new idea is a worthwhile insight, a personal digression, or a flawed claim is probably still best left to the human expert, until we have better models for evaluating innovation.

Limitations of data: Optimizing along the wrong output parameters

Scores that were dashed off by harried, overworked graders provide a poor standard for training an AI system. More fundamentally, the essay grade itself is not the goal; it is only a proxy for what we believe the goals of education should be. Robust assessment relies on multiple measures collected over time, across contexts, and corroborated by different raters. If we value long-term retention, transfer, and future learning potential, then our assessment metrics and models should include those.

I recognize that researchers are simply using the data that are most readily available and that have the most face value. My own work sought to predict end-of-course grades as a preliminary proof of concept because that’s the information we consistently have and use, and our society (perhaps grudgingly) accepts that. Ideally, I would prefer different assessment data. In pointing the direction in which we ought to go with such innovations, ultimately we should identify better data (through educators, assessment experts, and learning scientists), make them readily available (through policymakers and data architects), and demand their incorporation in the algorithms and tools used (through data analysts, machine learning specialists, and developers).

Limitations of algorithms: Model for meaning

Predictive or not, features such as essay length, sophistication of vocabulary, sentence complexity, and use of punctuation are typically not the most critical determinants of essay quality. What we care about is content, which demands modeling the conceptual domain. Hierarchical topic models can map the relative conceptual sophistication of an essay, tracking the depth and novelty of a student’s writing. While a simple semantic “bag-of-words” model ignores word order and proximity, a purely syntactic model accepts grammatical gibberish. A combined semantic-syntactic model can capture not just word co-occurrence patterns, but higher-order relations between words and concepts, as evident in sentence and document structure. Compared to the approaches earning such public rebuke now, more sophisticated algorithms exist, although they need more testing on better data.

When?

The question here is which parts of the assessment process are best kept “personalized,” and which parts are best made “adaptive.”

Rapid, automated feedback is useful only if the information can actually be used productively before the manual feedback would have arrived. For a student whose self-assessment is wide of the mark, an immediate grade can offer reassurance or a kick in the pants. For others, it may enable doing just enough to get by. Idealist instructors might shudder at the notion, but students juggling competing demands on their time might welcome the guidance. How well students can make sense of the feedback will depend on its specificity, understandability, actionability, and perceived cost-benefit calculus, all open questions in need of further iteration.

For instructors, rapid feedback can provide a snapshot of aggregate patterns that might otherwise take them hours, days, or longer to develop. Beyond simply highlighting averages which an expert instructor could already have predicted, such snapshots could cluster similar essays that should be read together to ensure consistency of grading. They could flag unusual ideas in individual essays or unexpected patterns across multiple essays for closer attention. Student work could be aggregated for pattern analysis within an individual class, across the history of each student, or across multiple instances of the same class over time. Some forms of contextualization may be overwhelming or even undesirably biasing, while others can promote greater fairness and enable deeper analysis. Determining which information is most worthwhile for an instructor to know during the grading process is thus another important open question.

Both cases explicitly acknowledge the role of the person (either the student or the instructor) in considering how they may interact with the information given to them. An adaptive system would provide first-pass feedback for each user to integrate, but the instructor would still retain responsibility for evaluating student work and developing more sophisticated feedback on it, with both instructor and student continuing the conversation from there— the personalized component.

Essay-grading itself may not be the best application of this technology. It may be more aptly framed as a late-stage writing coach for the student, or an early-stage grading assistant for the instructor. It may be more useful when applied to a larger body of a given student’s work than to an individual assignment. Or it may be more effective for both student and instructor when applied to an online discussion, outlining emerging trends and concerns, highlighting glaring gaps, helping hasty writers revise before submitting, and alerting facilitators when and where to intervene. While scaling up assessment is an acknowledged “pain point” throughout the educational enterprise, automation may fulfill only some of those needs, with other innovations taking over the rest.

Why?

The purpose of technology should be to augment the human experience, not to replace or shortchange it, and education– especially writing– is fundamentally about connecting to other people. Many of the objections to automated essay grading reflect these beliefs, even if not explicitly stated. People question whether automation can capture something which goes far beyond that essay alone: not just the student’s longer learning trajectory, but the sense of a conversation between two people that extends over time, the participation in a meaningful interpersonal relationship. Whether our current instantiation of higher education currently meets this ideal is not the point. Rather, in a world where we can design technology to meet goals of our own choosing, and in which good design is a time-consuming and labor-intensive process, we should align those expensive technologies with worthwhile goals.

By these standards, any assessment of writing which robs the student and author of these extended conversations fundamentally fails. Jane Robbins claims that students need “the guidance of experts with depth and breadth in the field at hand”, a teacher who can also be “mentor, coach, prodder, supervisor.” As the students on the Brown Daily Herald’s editorial board argue:

an evaluation of an essay by a professor is just as important, if not more, to a student’s scholarship and writing. The ability to sit down and discuss the particularities of an essay with another well-informed and logical human is an essential part of the essay writing experience.

Coupled with arguments that other types of (machine-graded) assessment are better for evaluating content knowledge or even low-level critical thinking, these arguments beg the question: Why try to automate assessment of writing at all? After all, much of what I have advocated here is simply accelerating the assessment process, not truly automating it.

The most basic reason is simply that writing instruction is important, and students need ongoing practice and feedback to continue improving. To the extent that any assessment feedback can be effectively automated, it can help support this goal. That raises two additional questions: How deep must feedback be for a writing exercise to be worthwhile? More controversial, can writing that never sees a real audience still serve a legitimate pedagogical purpose?

Considering the benefits not just of actively retrieving and generating information, but also of organizing one’s thoughts into coherent expression, I would argue that some writing exercises can facilitate learning even without an audience. Less clear is how far that collection of “some” stretches, or what the specific parameters are which demand feedback from an expert human. Likely factors include more complex assignments, more extreme work quality, weaker feelings of student belonging, longer intervals between receiving human feedback, less sophisticated automated feedback, and more nuanced expert feedback. Better articulating these limits and anticipating what we can gain and lose will help guide future development and application of automated assessment.

Personalized and adaptive assessment: Placing the stakes

Compared to personalized and adaptive learning, the distinction between personalized and adaptive assessment is less contested, but perhaps also less widely discussed. As before, my definition will hinge upon the role of human decisionmaking in distinguishing between adaptive (machine-intelligent) and personalized (machine- and human-intelligent) assessment.

Most understand adaptive assessment in the context of computerbased adaptive testing, which adapts the parameters of the test to student performance in real time. Acing a series of questions may earn more challenging questions, while fumbling on several may elicit easier questions or increased assistance.

In a slightly different perspective on adaptive assessment, the BLE Group suggests that formative and summative (benchmark) assessments “measure whether an academic standard has been learned,” while adaptive assessments “measure growth and identify where the student is on the learning-ladder continuum,” and diagnostic assessments “determine missing skills and remediate them.” I see these as overlapping rather than distinct categories. Adaptive testing is already widespread in summative assessment. Further, questions may be adapted to discriminate between students or to measure mastery of key concepts and skills at any point along an expected learning trajectory, whether a prerequisite or endpoint.

How personalization goes beyond adaptivity in assessment is in explicitly incorporating the decisions of the persons involved in the process, the primary stakeholders being the learner, the instructor-grader, and the external audience interpreting the grade. I will focus on just these three roles as the simplest case, although other configurations may include separate roles for instructor and grader, peer grading, and group assessment.

Learners differ in the goals they have for their education, both in what they hope to learn and in what they will be expected to present as documentation of that learning. Some careers require a particular degree or certificate, while others may solicit work portfolios or barrage candidates with tricky interview questions. Some students seek a general liberal-arts education rather than job-specific training, while others may simply want to broaden their knowledge without regard for the mark received for it. The initial choices are left up to the learner, and the information sought is determined by the audience for the assessment (i.e., the employer, certifying organization, or society). Thus, tailoring what gets assessed and how results are presented around those expectations would entail personalization rather than adaptivity.

How to present assessment information to learners and instructors may also vary depending on their preferences and abilities for interpreting and responding to such information. In some cases, these factors may be adapted based on evaluations of their actual behaviors (e.g., a learner who disengages after seeing comparisons against peers). In other cases, users may have access to better information or more sophisticated responses than the adaptive system, and an appropriately personalized system would allow them to choose their action based on that information. Examples include a learner getting distracted upon trying to interpret very-frequent feedback (with the system failing to distinguish loss of focus from intent, productive concentration), or an instructor recognizing when personal contact would help. Again, personalization builds in opportunities for human intervention to take over when the adaptive system is less suited for the task.

While these distinctions may not seem that significant, highlighting them here enables more precision in examining the criticisms of personalized and adaptive learning. Many limitations apply specifically to adaptive learning systems that do not leave enough room for individual choice or personal interaction. Adopting a fairly broad view allows us to focus on the possibilities and constraints regarding where these developments can go, not just the shortcomings of what some particular instantiations have done so far.