MCQ Test Grader & Item Analysis

About the MCQ Test Grader & Item Analysis

Mark a multiple-choice test in seconds without an answer sheet machine. Type the answer key, paste the students’ answers from a spreadsheet (or open a CSV or Excel file), and choose the marking scheme: marks for a right answer, marks taken off for a wrong one (negative marking, as a number or as a fraction of each question’s marks), what a double mark counts as, partial marks for questions with several right options, and different marks per question if the test has sections.

You get every student’s score, percentage and rank, the class mean, median, standard deviation and score distribution, and an item analysis of every question: the difficulty index, the discrimination index between the top and bottom 27 % of the class, the point-biserial correlation, how often each option was chosen, notes on questions that need a second look, and the KR-20 reliability of the test. With a pass, download the results and the item analysis as CSV, or everything as a PDF report; without one, the free preview shows the class counts and average in full, with the first rows of the rank list and of the item analysis.

Common uses

• Class tests and mock exams marked from a spreadsheet of answers.
• Entrance-exam practice with negative marking, such as +4 / −1, or one third of the question’s marks.
• Reviewing a question bank: which questions are too easy, too hard or do not separate stronger from weaker students.
• Answers exported from an online form, once its answers are letters.

How each answer is marked

• Right: the question’s marks. Wrong: minus the deduction (0 without negative marking). Blank: 0.
• Several options marked on a question with one right answer count as wrong (with the deduction), or as zero if you choose that.
• Several right options — a key such as (AC) — need all of them. With partial marks on, marking only some of them (and no wrong one) earns that share of the marks: 1 of 2 right options earns half.
• Either is right — (A/C) — accepts A or C.
• * gives everyone the question’s marks (a question you have cancelled in the students’ favour); - leaves it out of the total and of the item analysis.
• Ranks are standard competition ranks: equal scores share a rank, and the next rank skips (1, 2, 2, 4).

The item analysis

For every marked question (not * or -), from all the students:

• Difficulty index p: the share of students who got it right, from 0 (nobody) to 1 (everyone). Questions with p of 0.90 or more are flagged as very easy, 0.20 or less as very hard.
• Discrimination index D = share right in the top 27 % of the class − share right in the bottom 27 %, ranked by total score — the upper and lower groups of T. L. Kelley (Journal of Educational Psychology). Each group has round(0.27 × N) students; ties at the edge of a group are taken in the order of your list.
• Quality, from Robert L. Ebel and David A. Frisbie’s guidelines (Essentials of Educational Measurement): 0.40 and above very good; 0.30 to 0.39 reasonably good; 0.20 to 0.29 marginal, usually needing improvement; below 0.20 poor, to be revised or dropped. A negative D usually means a wrong key or a misleading question.
• Point-biserial: the correlation between getting the question right and the score on the rest of the test (the question itself left out, so it does not inflate its own figure).
• Options: how many students chose each option. A wrong option nobody chose is not doing its job; one chosen more by the top group than the bottom group deserves a look.

KR-20 reliability

KR-20, from G. F. Kuder and M. W. Richardson (Psychometrika), estimates how consistently the questions measure the same thing: KR-20 = k / (k − 1) × (1 − Σ p·q / σ²), where k is the number of marked questions, p and q = 1 − p are each question’s shares right and wrong, and σ² is the variance of the students’ totals (the number right, with the population formula). It runs from about 0 to 1; higher means the total score is more dependable. Short tests and classes that are very alike give lower values. When partial marks make questions not simply right or wrong, the same formula with each question’s variance is Cronbach’s alpha, and the page says so.

The data it reads

• One answer per column: Asha, A, B, AC, , D — a blank cell or - is a blank answer, and AC means both were marked.
• All answers in one cell: Asha, ABAC-D or Asha ABAC-D, with (AC) for a double mark.
• Headings: a first row such as Roll, Name, Q1, Q2 …, First name, Last name, 1, 2, 3 … or the headings of a form export (Timestamp, Email address, the question texts) tells the name and roll-number columns from the answers; when the headings are question texts, the page says it has read that row as headings. Without headings, the cells before the answers are the roll number and the name, lined up with the other rows (so a name such as Ida is not read as marked options).
• Numbered options: tick “Options are numbered 1, 2, 3 …” when answers are digits instead of letters.
• Paste from Excel, Google Sheets or LibreOffice (tab-separated), type CSV, or open a .csv, .tsv, .txt, .xlsx, .xls or .ods file (its first sheet).

Limitations

• Up to 300 questions and 2,000 students, with options A to J.
• Answers must be letters (or digits 1–9 with the numbered option): an export from an online form that holds the option texts needs them turned into letters first.
• The item analysis needs enough students: groups and correlations need at least two or three, flags appear from five, “nobody chose” from ten, and the figures settle down with about 20 or more.
• Paper answer sheets cannot be read here — use the OMR Sheet Scanner for those.

Privacy

On this device: nothing is sent

Everything is worked out in your browser and nothing is uploaded. The students’ answers stay in this tab only and are gone when you close it; only the answer key, the marking settings and the test’s name are remembered in this browser, to save typing next time.