How to test yourself when you are learning on your own

A class does one thing a book cannot: on a date you did not choose, it shows you what you do not know. Teach yourself a language, a programming language or a certification syllabus and that check is gone. Nobody sets the test, and often no exam date forces the question.

What remains is your own feeling of progress, which is a poor measure. This guide explains how to test yourself when you are learning on your own by building the checks a teacher would have supplied: a test for every unit, a rule for what counts as known, spaced re-tests at widening gaps, and a record.

EdProcto's page for self-learners covers three of them: a quiz for each chapter, re-tests that come due, and a map of what is known. A notebook works too.

Why is it hard to test yourself when you are learning on your own?

It is hard because the only judge on hand is your own sense of what you know, and that sense tracks how the studying felt more closely than what you can do afterwards.

Sitzmann et al. (2010) pooled the research on self-assessment of knowledge in education and workplace training. Learners' ratings of their own knowledge were tied most strongly to motivation and satisfaction. The tie to measured learning was moderate, and stayed weaker than the tie to motivation even where learners had practised rating themselves and had feedback on it. Nearly a third of the evaluation studies examined had still read such ratings as evidence of learning.

So "this is going well" is largely a report on your mood. Bjork et al. (2013), reviewing how people manage their own learning, locate the error: people tend to hold a faulty picture of how memory works, and they judge progress by their performance at that moment and by how fluent the material feels.

Do beginners overrate themselves most?

Possibly, and the point is disputed. Kruger and Dunning (1999) ran four studies using tests of humour, grammar and logic. Participants in the bottom quarter scored at about the 12th percentile and placed themselves at about the 62nd. Improving their skill helped them see the limits of it.

Gignac and Zajenkowski (2020) argue that comparing quarters of a group is confounded by regression toward the mean and by the habit of rating oneself above average. With 929 adults, a self-rating of intelligence and a reasoning test, they did not find the pattern, and they conclude that the effect may be much smaller than reported.

Either way, you are a beginner again at the start of every unit, and the sensible course is to measure, not to estimate.

What should you test when there is no syllabus?

Test what you should be able to do at the end of each unit, and draw up the units yourself if nobody has. A syllabus is only a list of units, each with what a learner can do after it.

Use a book's chapters as units

If you are working through one book, its contents page is your syllabus. Treat each chapter as a unit and test it before you open the next. Split a long chapter at its main headings, so that a unit can be studied in two or three sittings and tested in about ten minutes. That size is a working convention, not a research finding.

For each unit, write two to five lines that begin "After this I can", taken from the chapter's headings, worked examples and exercises: "After this I can put regular verbs into the present tense." The test must cover those lines, which stops it drifting towards whatever was easiest to ask.

Make units when there is no book

• A certification: one exam objective, from the outline the exam body publishes, where it publishes one.
• A language course or app: one lesson, with its phrases and its grammar point.
• A programming tutorial: one construct, such as loops or functions.
• Videos or a podcast: one episode, tested from your notes or the transcript.

If you are reading around a subject with no single source, borrow the contents page of a standard textbook. It is a list of units drawn up by someone who knows the field.

How do you test a skill, and not only facts?

Do a small piece of the skill itself with the support taken away, then check the result against something other than your own opinion. Listening and reading code are things you do, and a question about them tests something else.

Every shape in the last column is in the EdProcto builder, and so is writing a program that is run against tests.

Listening: once, at full speed

In a conversation nothing is replayed, so play the clip once, write what you heard, and only then open the transcript. Sort the words you missed into those you did not know and those you know on the page but cannot yet catch by ear.

Code: read before you write

Lister et al. (2004) tested novice programmers from seven countries on predicting what a short piece of code would produce and on choosing the right completion for a nearly finished piece. Many were weak at both. So test reading on its own. Copying a tutorial's code as you follow along does not test it.

Before you run this, write down what it prints:

total = 0
for n in range(1, 6):
    if n % 2 == 0:
        total += n
print(total)

It prints 6. The wrong answers tell you more than the right one. An answer of 12 means you think range(1, 6) includes 6. An answer of 15 means you skipped the condition. An answer of 9 means you read the check for even numbers as a check for odd ones.

What counts as "known"?

A unit counts as known when you have answered it correctly from memory on several separate days, never on the strength of one good sitting. One correct answer is where most people stop, and it is the wrong place.

Karpicke (2009) ran four experiments in which people learned foreign-language vocabulary over rounds of study and testing. Given control of their own practice, many dropped the items they had recalled instead of testing them again, and their later retention was poor. Karpicke names the illusion: once you can recall a thing, you tend to believe it is learned.

Rawson and Dunlosky (2011) put numbers on the alternative. Across three experiments, 533 students learned concepts by recalling them and restudying, to between one and four correct recalls in a first session and over one to five later sessions, and were tested one to four months on. The authors' advice was three correct recalls in the first session, then relearning three times at widely spaced intervals. The material was concepts recalled as short answers, so for listening or code the figure of three is a starting point, not a tested rule.

A rule you can copy

1. Set the pass mark before the test. Eight out of ten will do. The research does not supply the number, so fix it before you see your score.
2. Wait before the first test. Nelson and Dunlosky (1991) found that predictions of later recall were poor when made straight after studying an item and very accurate when made a short while later. Finish the chapter, do something else, then test.
3. Answer in full, from memory. Type it, write it or say it aloud. "I knew that" earns no mark.
4. Say how sure you are before you check. Answers you were sure of and got wrong are the most useful lines in the test. EdProcto's multiple choice with confidence asks for that rating with every answer.
5. Bring each miss back until you have got it right three times in the sitting.
6. Call the unit "passed once". It becomes known after three passes on later days, and a failed re-test sends it back to "learning".

When should you test a unit again?

Test it again after a gap, lengthen the gap each time it passes, and shorten it when it fails. The right gap depends on how long you want the knowledge to last, and a self-learner usually wants it for good.

Cepeda et al. (2006) drew together 839 assessments of spaced practice from 184 articles. The gap between sessions that gave the best retention grew as the delay before the final test grew. In the copy of the paper its authors have deposited, where the final test came six months or more after study, gaps of at least a month did best. With no exam date your final test is indefinitely far off, which argues for gaps that grow to weeks and months.

Whether widening gaps beat equal ones is less settled than popular advice suggests. The same review found only a few studies comparing the two, and concluded that widening gaps either help or do about as well as fixed ones, with too little data to say more.

A starting schedule of our own: re-test two days after the first pass, then after a week, three weeks and two months, and every few months from there. A failed re-test drops the unit back one step. Re-tests need no sitting of their own, so put three or four questions from older units into each new unit's test.

What record should you keep?

Keep one line for each unit: the date and score of every test, its status, and the date the next test is due. Your memory of how a test went is no more reliable than the feeling it was meant to replace.

Kornell and Bjork (2008) had people learn painters' styles from paintings shown one artist at a time or mixed together. Mixing led to better recognition of each artist's unseen work, yet participants went on rating the grouped version as the better method, their own test results notwithstanding.

Once a week, take what is overdue before starting anything new. If the overdue list keeps growing, new units are arriving too fast: the record has told you what no feeling would.

The limit of a test you wrote yourself

A test you set can only ask about what you noticed, so whatever you skimmed past is missing from it. Some questions should come from outside: the chapter's own exercises, a past paper, a speaker of the language, the interpreter that runs your code, or questions written from the material by something other than you.

How does EdProcto do this for a self-learner?

EdProcto is the best tool for this job when you are learning from your own material, because it supplies the two things a lone learner cannot: questions they did not write, and a schedule that does not depend on how confident they feel.

Paste a chapter, an article, a transcript or your notes into the quiz maker on the home page, and Procto AI writes the questions from that text.

In study mode each answer is followed at once by the right answer and an explanation, the questions you missed come round again, and the quiz falls due again in a few days. Upload a whole book and you work through it chapter by chapter while EdProcto keeps a map of what is known: the record described above, kept for you.

Each of the templates in the skills table has a page saying how it is answered and marked. The kits are ready-made sets, among them a language, code and a certification.

The Free plan carries 20 questions a month written by Procto AI, about a self-test a week from your own chapter. The Learner plan is the one made for self-study: self-tests without limit, 100 written questions a month, and the whole book with its map. The pricing page has the detail. No tool learns the material for you or can promise a result.

Quick questions

Can I test myself without writing the questions?

Yes. Use the chapter's own exercises or a past paper, or paste the material into EdProcto: its page for self-learners says what you can paste.

How do I know when I have really learned something?

When you have answered it correctly from memory on several separate days. Rawson and Dunlosky (2011) advise three correct recalls in the first session and three later sessions spaced widely apart.

How do I test my listening when I teach myself a language?

Play a short clip once, type what you heard, then compare it with the transcript. EdProcto's listen once shape plays the clip a single time, and dictation asks you to write down exactly what you hear.

Do I need an exam date for self-testing to be worth it?

No. An exam date only says how long the knowledge must last. Without one, assume you want it for good. For how to run a single sitting, see how to make your own quiz from a chapter.

References

• Bjork, R. A., Dunlosky, J., & Kornell, N. (2013). Self-regulated learning: Beliefs, techniques, and illusions. Annual Review of Psychology, 64, 417–444. https://doi.org/10.1146/annurev-psych-113011-143823
• Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132(3), 354–380. https://doi.org/10.1037/0033-2909.132.3.354
• Gignac, G. E., & Zajenkowski, M. (2020). The Dunning-Kruger effect is (mostly) a statistical artefact: Valid approaches to testing the hypothesis with individual differences data. Intelligence, 80, Article 101449. https://doi.org/10.1016/j.intell.2020.101449
• Karpicke, J. D. (2009). Metacognitive control and strategy selection: Deciding to practice retrieval during learning. Journal of Experimental Psychology: General, 138(4), 469–486. https://doi.org/10.1037/a0017341
• Kornell, N., & Bjork, R. A. (2008). Learning concepts and categories: Is spacing the "enemy of induction"? Psychological Science, 19(6), 585–592. https://doi.org/10.1111/j.1467-9280.2008.02127.x
• Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one's own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology, 77(6), 1121–1134. https://doi.org/10.1037/0022-3514.77.6.1121
• Lister, R., Adams, E. S., Fitzgerald, S., Fone, W., Hamer, J., Lindholm, M., McCartney, R., Moström, J. E., Sanders, K., Seppälä, O., Simon, B., & Thomas, L. (2004). A multi-national study of reading and tracing skills in novice programmers. ACM SIGCSE Bulletin, 36(4), 119–150. https://doi.org/10.1145/1041624.1041673
• Nelson, T. O., & Dunlosky, J. (1991). When people's judgments of learning (JOLs) are extremely accurate at predicting subsequent recall: The "delayed-JOL effect". Psychological Science, 2(4), 267–271. https://doi.org/10.1111/j.1467-9280.1991.tb00147.x
• Rawson, K. A., & Dunlosky, J. (2011). Optimizing schedules of retrieval practice for durable and efficient learning: How much is enough? Journal of Experimental Psychology: General, 140(3), 283–302. https://doi.org/10.1037/a0023956
• Sitzmann, T., Ely, K., Brown, K. G., & Bauer, K. N. (2010). Self-assessment of knowledge: A cognitive learning or affective measure? Academy of Management Learning & Education, 9(2), 169–191. https://doi.org/10.5465/amle.9.2.zqr169