How to detect cheating when students take online quizzes or exams from home?

Question

I am currently teaching a basic R programming course with about 80 students. Under normal circumstances, a core course assessment component is a programming exam, which is done in the computer lab using the lab computers. I will arrange for the Internet to be cut off, and we will have exam invigilators who are patrolling the exam venue to detect or dissuade students from cheating. At the end of the exam, the Internet is restored, and students submit an R script (basically programming code) using the course learning management system.

This semester, due to the novel coronavirus disease (COVID-19), the university where I work has replaced all on-campus teaching activities with online teaching for the entire semester!

• With regards to teaching, the faculty have been asked to teach our classes online using Zoom, which gives students the freedom to take classes wherever they like.
• With regards to exams, the faculty have been asked to administer their course exams online, with students completing the exams online from home. (What??!!)

I am very concerned about students taking exams online from home, because I am worried that some or many of the students may use this as an opportunity to cheat on the exam — whether it is collaborating with their classmates, or even asking another student to take the exam in their place.

My questions:

1. Is there any way to detect if students are cheating (e.g., sharing answers), when students are doing an online quiz or exam at home?
2. Is there a way to dissuade students from cheating?

Response to comments:

• How much help do you have for these 80 students?

  I do have one PhD student who is a grader. She is supposed to spend about 6 hours a week to help with the course.

Answer 1

I also teach a data and programming course, though I mainly use Python.

Is it possible for you to shift grading away from an exam and toward something else? A project, for example? Before going back to academia I spent years working as a policy researcher, and exam conditions simply aren't a realistic test of real-life coding abilities. For that matter, coding with the internet turned off is also not remotely what real work looks like either.

In my class I like to give a larger project and let them have a few weeks to work on it, but I could also imagine a more simple project that they only have a few days, or even hours, to do. Real data work often comes with time constraints, but it does not come with a lack of internet access in exam-like conditions. Just scale the scope of the project based on how long you want them to have to work on it.

There is still a huge advantage to knowing the material, rather than having to look it up. You can also grade on the quality of code and correctness of results, and allow for creativity with open-ended questions like "summarize this data with two plots". This way each student is not only tested their knowledge of R, they're tested on their actual ability to use R in a realistic environment.

If you, unfortunately, are required to give an exam, I know there is software available for proctoring online tests. The sort of thing that utilizes their webcams to make sure they don't leave during the test, or consult other devices, or open other web pages. I've not used them though, so I couldn't make a recommendation. I imagine they all require university subscriptions.

Answer 2

Make the question non-amenable to Googling

Many of colleagues used to do what you mentioned: Having exams in labs and cutting off internet/network access to isolate their terminals. All with good intentions of not cheating.

A different approach that I implemented was rethinking what I want to test them on and use that as the guiding source and create the problems from scratch. That way, even if they were to Google the question it’s unlikely to find any answer. However, if they are to search for something specific like “Breadth First Search in Java” that is probably okay IMHO. The idea is for them to understand/decipher the problem such that they can correctly infer to use that algorithm. If the implementation language is Java, then they’re free to look up certain implementation details (assuming they’ve had past assignments/practice in some form).

This takes away the focus from the act of “coding the algorithm” to “solving a problem using the algorithm”. The latter is more copy-proof IMHO. Even though they may “copy” the algorithm implementation, they’ll still have to adapt it to the problem at hand and that itself can be a lot of work.

For those open to this “overhead” that’s the choice they make. For those who’re comfortable implementing it from scratch, they won’t rely on this approach. It also takes away the pressure of memorizing/debugging in a very time critical environment and gets them to focus on the act of solving the problem with the intended algorithm.

I’ve had very good success with this approach. It’s more work to come up with these problems and you may run out of them. For that a good strategy is to ask the students to come up with some good problems for the next year and they can earn bonus credit for good submissions - also works like a charm ;)

There are online tools that check for “code similarity” to infer if two people collaborated/shared code. One is MOSS that we used. There are surely others that can help with detecting “cheating/sharing” with great success.

Example:

I have a city map with ‘travel time’ for each bus route between 2 stops. Write a program that asks the user for starting and ending stop and returns the fastest way to get there:

Input/setup:
A -> B: 5
B -> C: 10
C -> D: 1
A -> C: 3

Example IO:
Start-stop?: A
End-stop?: D

Computing...

A -> C -> D: 4

Testing on the ability to find the shortest path - whether they use BFS or Dijkstra is up to them. This is not easy to Google for and even if they do, it’ll take a while to translate to the above problem.

You can timebox the implementation time if you’re using services like Code Signal or the like.

Answer 3

There are tons of open source software or scripts or libraries written in R.

Give your students a randomly chosen such project (not a too small one), with the task of explaining what that software does in written English (perhaps with some math formula in LaTeX or KaTeX), and to extend or customize or execute that software (or library) for a problem of their choice. In other words, ask them for a code review, and if additional time is available, to add one extra functionality or feature (described in written English) to that software.

Such tasks are currently hard to be entirely automated. They might use the Internet, but it won't help them that much....

You could expect cooperation, by giving to each small group of students one additional (but different for every group) feature to add. Notice that such team cooperation is close to industrial practice.

I did teach at University a few years ago. One of the tasks I gave students was to download sash and extend it to add one additional builtin (described in French) in two hours. Access to the internet does not help much in practice. The time was limited, of course.

Today, I am working (in my free time) on RefPerSys. By experience, it is difficult to make motivated persons (which I never met in real life) reviewing some of my git commit-s. I sent them the URL of many papers describing them (e.g. the ObjVLisp model). I am writing emails, submitting papers, draft reports, and slides quite often. We all are very motivated. Even with that, getting code reviews and code git commit-s is difficult. And today, such a task cannot be realistically automatized.

Answer 4

I come at this as a former student and as a TA who assisted with exams.

At my college there was one class I took that allowed for students to actively write and submit code via live submission for tests in person. They did not cut off the internet for us, but we were timed, and the requirements for the examination were strict enough that if you did not know the material well, searching for the answers would not get you very good results.

They did this by making sure that you were using a data set that was hand constructed for that test. It happened to have a test in R as well. This data set was unique, and as such, you could not find the answers anywhere online.

As a TA I also know, submissions were automatically compared to each other by the submission system in a way that it could be determined which ones were close matches, so that it would become obvious when 2 people shared answers. These would be flagged for the professor to review.

Altogether, I do not know how effective this system was, as it was fairly new at this point. I was in my senior year, and this was the first time they had tried it, having had paper programming tests only previously.

To summarize, what they tried was to develop an online submission system that uses Closest Strings Matching to determine percentage score matching between student submission, and to develop unique secret problems for each test that cannot be found online in order to have a digital system that can be used in person or online, as well as having a strict but fair time limit to take the test.

As for the success of that system, only time will tell. The school I graduated from is UCF.

Answer 5

Disallowing internet searches for a programming course seems a bit outdated these days. I would change the exam so that it relies less on memory and more on technique.

I've had pre-interview technical tests which allowed internet searches, but were hard enough that if you had to constantly look up stuff, you'd never complete it.

Main problem with online exams is that a test taker could just have someone else do the exam for them, even with a webcam pointed at them from their laptop (just have the accomplice use a separate keyboard and monitor).

Answer 6

For exams such as AZ-300, Microsoft use a combination of a secured browser and chat using the webcam of the computer being used. The person taking the exam has to scan the webcam around the room to show no other machines are on, and is observed while taking the test. They have dedicated software to ensure that no other browser sessions are open, but most screen sharing software with recording could be used for an invigilator to check that, once the webcam has confirmed there's only the one screen.

Answer 7

Ask the center for teaching and learning (or whoever at the university is in change of arranging faculty workshops on teaching - all the places I have taught have one) what the options for online proctoring are. When I have taught similar courses, there is software which can be used to proctor online exams. It basically consists of a browser lockdown, recording the screen, and access to the webcam, which records if the student behaves suspiciously while sitting there. Yes, this is pretty intense scrutiny, but it seems to work for our faculty who give exams for online courses.

Answer 8

As this applies to the entire university, your case may actually be more amenable to a technological solution than a course where the majority of exam answers are in the lecture notes that the students might have right next to them on paper as they take the exam (so even a secured browser session wouldn't help). The university is likely to be working on a way of handling grades under the circumstances, but it may be too little, too late for you to make adjustments yourself, or it may invalidate those adjustments.

Answer 9

From a student's perspective: Don't do time crunched tests that don't really help anyone. Do a project or smaller quizzes. Especially in online classes, tests are the most agonizing task that create SO MUCH anxiety because there is so much that can go wrong that is out of our control. Most students also do better on a honor policy, to trust most students will honorably do their work without the need to spy on them. They will have much less pressure and will learn and retain more. You can design the course in such a way that they are forced to keep up with instruction and that they pay attention to instruction and practice but take the pressure off of them. A little mercy and grace and leniency will mean a lot more than an iron fist and I can say with 100% certainty, your students will retain much more this way.