Taking an exam #

Your instructor sends you a link. That link carries the whole exam – the dataset recipe and the questions – so there is nothing to download, no account to create, and nothing to upload. Everything runs in your browser, and your work stays on your machine.

Quick start #

1. Open the link. A dialog asks you to enter your name.
2. Type your name exactly as your instructor expects it. Your name is what builds your dataset, so a different spelling is a different dataset.
3. Your data loads and the Exam panel appears at the right of the screen.
4. For each question: run the analysis it names, read the number off the result card, and type it into the panel. Then press Check {n} answers at the foot of the panel – one press checks every answer you have typed.
5. When enough questions have passed, export your report from the Export menu.

Why your numbers differ from everyone else’s: your name is turned into a random seed, and the data is drawn from that seed. Every student gets a dataset with the same structure but different values, so which variables come out significant is genuinely yours. Copying a classmate’s answers cannot work, and neither can looking them up – but your own data is perfectly reproducible, so your instructor can regenerate it from your name and check your report against it.

The exam panel #

The panel floats over the app so you can keep working while it is open.

• The header shows Exam and your score as passed / total. Click it to collapse the panel out of the way; click again to bring it back.
• The body lists every question with its own input.
• The footer shows the export gate (see exporting your report), the Check {n} answers button and a Restore the exam dataset button.

The panel sits underneath dialogs, so the Cases and Variables modals stay reachable while it is open.

Answering #

Each question has one input, matched to what it asks for.

An unanswered question reads — not answered —. A question that builds on an earlier one stays empty until you answer that one, and says so: Answer the earlier question this one builds on first.

Why a “pick one of the above” question is not graded: questions like pick one of the scales you just listed exist so the questions after them run on your choice. There is no right answer to the pick itself – the grading happens on what you say about the variable you picked.

Checking your answers #

Press Check {n} answers and every answered question not yet marked is marked at once: Correct, or Not the app’s number – read your output again. The button counts what it will check – Check 1 answer, or Nothing to check when every answer is marked. A table is marked cell by cell, so you can see exactly which row is wrong.

• Retries are unlimited and are not counted. A marked answer, a failed one included, is checked again once you change it; nothing records how many times you checked, and there is no penalty for checking early.
• A mark you have earned is kept. Once a question passes it stays passed, even if you change settings afterwards.
• A set is marked as a whole. If a checkbox question fails, one of your ticks is wrong – go back to your output and read the table again rather than guessing which one.
• Changing an answer clears what depends on it. If you change a pick, the questions built on it lose their answers and marks, because they were about a different variable.

Questions that need R run for a second or two behind the loading overlay – the panel says Checking… and the screen locks so a check can never collide with an analysis you started. Cancel stops it.

Type what the app shows #

Answers are compared against the app’s own formatting of the true value, under your precision settings. So you type what is on your screen and nothing else has to match: if you work at two decimals, you are graded at two decimals; if you work in significant figures, you are graded in significant figures.

• Below the p-value floor the app prints < 0.001, and that is an accepted answer – type it as shown.
• A comma decimal separator is accepted.
• Counts and other exact values are matched exactly.

A p-value on the boundary is yours to resolve. If a question turns on whether p is below 0.05 and your screen shows exactly 0.05, raise the precision in Settings and read it again. The grader follows your precision setting, so the extra digit is real, not a trick.

Settings that change your answers #

The exam grades you on the analysis you ran, so the settings you are working under are part of the answer. These are the ones that move a graded number:

• Multiple comparison adjustment – if you switch to Holm, you are graded on Holm.
• Significance level and the precision settings.
• Missing data handling – on listwise deletion your cards are computed on fewer rows, and so is the answer key.
• The reproducibility seed, for factor questions. Parallel analysis is graded at its default of 100 draws, whatever count the factor view is set to, so leave that field as it is.

Some settings can make a question impossible rather than merely different, and the panel tells you which:

• Turn p-value display on in Settings to answer this – p-values are hidden entirely under the display format setting none.
• Set p-value display to exact values in Settings to answer this – the category format prints a band rather than a number.
• Set a random seed in Settings to answer this – a few questions draw random numbers, and with no seed set they would give a different result every time they are checked.

The settings each answer was checked under are recorded with the mark and printed on your exported report, so your instructor can reproduce exactly what you did.

Transforming and filtering data #

Some questions ask you to change the data first – filter to one group in the Cases dialog, reverse-score an item, build a composite. Do it exactly as the question words it.

Your changes never corrupt the answer key: grading always runs against a pristine copy of your original dataset, kept aside when the exam opened. What matters is that your screen shows the right thing – a filter you forgot to clear is the usual reason a later answer stops matching.

If your data gets into a state you cannot untangle, press Restore the exam dataset. It redraws your original data from your name and discards every transformation and filter you have added; your answers and marks are kept. Reloading the exam link does the same thing.

See data transformation for what the transformation rules do.

Exporting your report #

The panel footer tells you where you stand:

• Export unlocked / Ready to export – you have passed enough questions.
• Export locked – n of m questions must be checked / Export locked – n of m questions must pass – the export is blocked until enough questions are checked, then until enough have passed.
• n of m questions should be checked before you export / n of m questions should pass before you export – you will be asked to confirm, but you can go ahead.

Your instructor chooses which of these applies. When you export to Word, the document carries an Exam answer sheet appendix listing every question, your final answer and its mark – including the questions that are not graded, so the picks your other answers were built on are visible too. A provenance line in the page header and a short integrity hash tie the document to your dataset.

Write your interpretations into the document as the prose questions ask. The panel only certifies that you ran the analysis correctly; the reasoning is what your instructor reads.

Common pitfalls #

A leftover filter. The most common cause of a correct-looking answer that will not pass is a Cases filter left on from an earlier question. Check what is filtered before you blame the number.

Typing the value you expected instead of the one on screen. The grader compares against the app’s own output. Round-tripping a number through your own arithmetic – converting a percentage, re-rounding – is what breaks an otherwise correct answer.

Reading the wrong percentage. A variable with missing values shows both a percentage of all rows and a valid percentage of the rows that have data. The question names which one it wants; read it again if your answer is close but not equal.

Reading the wrong column in a reliability table. Item-total correlations are printed in more than one form. Questions about reverse-keyed items mean the corrected item-total correlation, which is the one the module’s own warning is based on.

Assuming the exam is the same as a classmate’s. It is not – same questions, different numbers. If a question asks which variables differ significantly, your answer will not be theirs.