Sign up free

The first multiple-choice test, 1914 to 1915, and what it was for

What the papers of 1912 to 1929 say about the first multiple-choice test, the Army tests, the scoring machine and the early criticisms, set against the legend.

The EdProcto editorial teamWritten and checked against the research it cites. Last checked . How the guides are made

Guides · Writing questions that measure · · 13 min read

Who made the first multiple-choice test, and what was it for?

The first multiple-choice test is usually credited to Frederick J. Kelly's Kansas Silent Reading Test, sat by 9,252 Kansas children in May 1915 and published in 1916. It was built so that any marker reached the same mark in minutes. About a quarter of its exercises were choices among printed words, and Kelly claimed only that it measured getting meaning from print.

The first multiple-choice test is usually said to be Frederick J. Kelly's Kansas Silent Reading Test, dated 1914 or 1915. The papers of the time support a smaller claim and a more useful one: a five-minute reading test, sat in Kansas in May 1915, in which every exercise had one right answer and about a quarter were choices among printed words.

This guide sets what those papers say against what later histories repeat. Every source was read on 10 October 2026, the old ones in the Internet Archive's copies; where only a later history could be read, the sentence says so. The longer story is in the history of proof in education.

EdProcto publishes this guide, and its multiple choice template is the same shape a century on: the last section says how one is made and marked there.

Who made the first multiple-choice test, and what does the record support?

Frederick J. Kelly is the name later histories give, and the record supports part of the credit: Kelly (1916) published a test with choices among printed words that 9,252 Kansas children had sat in May 1915. It does not support the date 1914, a paper made only of choices, or the word "first".

The article sets out three tests, for grades 3 to 5, 6 to 8 and 9 to 12, each of sixteen exercises to be done in five minutes. Kansas school superintendents sent in several hundred exercises; about two hundred were tried and forty-eight kept. Each was to bear one reading and ask one thing, so that an answer was right or wrong and never part right. Each carried a value, taken from the time children needed for it, and a child's score was the sum of the values of the exercises done right.

What the 1916 article prints

By our count of the forty-eight exercises as printed, eleven give a row or list of words and ask the child to mark one or more: seven in the test for the youngest, three in the middle test and one in the test for the high school. The rest ask for a written word or number, or for a mark on a figure or in the passage. The phrase "multiple choice" appears nowhere in the article.

Kelly makes no claim to a first, and nor does any source of 1914 to 1929 that we read. The Army's own account says group methods had been tried in a few scattered studies before 1917 and names one, an unpublished group test by A. S. Otis of Stanford (Yoakum and Yerkes, 1920). Ruch (1929), in a textbook on the objective examination, cites Kelly for his study of marks and, in the archive's text, never names the Kansas test. We did not search for earlier tests, so "usually credited" is as far as we can go.

Where the retelling departs from the papers

What is repeated What the papers we read say
Kelly was "someone who actually sat down and wrote the first one", in 1914 (Davidson, 2011, extract read) Kelly (1916) dates the sitting to May 1915, signs from the University of Kansas and claims no first
He wrote his thesis at a Kansas college (Watters, 2015) The title page of Kelly (1914) reads Teachers College, Columbia University
The thesis argued that marking took too much of a teacher's time (Davidson, 2011) None of its four conclusions is about time, which is one of four criteria in the 1916 article

What problem was the Kansas test meant to solve?

Two problems: teachers' marks on the same paper disagreed far more than teachers believed, and marking took hours.

The studies of marker disagreement

  • English. Starch and Elliott (1912) printed facsimiles of two English examination papers and sent them to 200 high schools; 142 usable sets of marks came back. The range of marks on a single paper, they wrote, may be "as large as 35 or 40 points". The weaker paper had been passed by the pupil's own teacher, and 22 of the 142 teachers failed it.
  • Geometry. Starch and Elliott (1913b) tested the belief that mathematics is marked more exactly. The marks varied more than in English, and five teachers in the paper's own school gave it 70, 65, 60, 70 and 59. The range usually quoted, 28 to 92, we read in Ruch (1929); the article shows it as a chart.
  • History. Starch and Elliott (1913a) found marks from 43 to 92 among seventy schools that shared a pass mark. They concluded that the variability "is not a function of the subject but a function of the examiner", and proposed standard tests and scales as one remedy.

Kelly (1914) summarised the first two of these studies in his doctoral thesis and added his own. In the New York Regents' examinations, he found, the state's examiners changed a quarter of the teachers' marks by ten points or more. He then tried three existing scales on judges new to them: disagreement fell sharply in drawing, stayed about the same in handwriting and rose a little in composition. The thesis names the idea the Kansas test would use, a special test so made that "the rating of the results is wholly objective", but in the archive's text it describes no test with options to choose from.

For marker disagreement today, see moderation.

The time marking took

Time is the last of the four criteria in Kelly (1916): giving and scoring must not cost a teacher more than the result is worth. He reports under fifteen minutes to give the test and twenty-five to forty to score a class of thirty-five. The teacher crossed out the value of each wrong exercise and added the rest, with a printed key to the harder exercises. Monroe (1922) reports W. A. McCall's estimate of about three hours to set, give and mark an ordinary examination, and reckons a teacher's year at sixty hours or more.

How did the format spread from Kansas to the Army and to admissions tests?

Through schools first, then through the Army's group tests of 1917 and 1918, then into college admissions from 1926.

  1. Schools, 1915 to 1916. In January 1916 the journal about to print Kelly's article reported the test in wide use in Kansas and neighbouring states, with 12,000 copies ordered for Iowa's schools ("Notes and news," 1916).
  2. The Army, 1917. A committee of psychologists built a group examination, drawing on published tests and on Otis's unpublished one, and tried it on about 80,000 men in four camps that autumn (Yoakum and Yerkes, 1920).
  3. Examination alpha, 1918. Eight tests in about fifty minutes, for groups as large as five hundred, answered by underlining, crossing out or checking and scored with stencils laid over the page. The Army's book counts 1,726,966 men examined; a later history says more than 1.5 million (Jacobsen, n.d.).
  4. Teachers' own tests, 1922. Monroe (1922) told teachers how to write such exercises themselves: at least four answers to each, and the right one in a different place each time.
  5. Admissions, 1926. The first SAT was sat on 23 June 1926 by more than 8,000 candidates: 315 mostly multiple-choice questions in 97 minutes (Lamb, 2026). As Jacobsen (n.d.) reports, the College Board suspended its essay examinations in December 1941.

Was the Army's test multiple choice?

In part. Of alpha's eight tests, three offered three or four printed answers (practical judgment, analogies and information), two offered a pair (same or opposite, true or false), and three asked for something written or drawn: spoken directions followed with a pencil, arithmetic and number series.

The name came later. Monroe (1922) calls the form the recognition exercise and says it had been used in silent reading tests and in group intelligence tests. Ruch (1929) writes of multiple-response or multiple-choice types and counts them as familiar.

What did the scoring machine of the 1930s change?

It moved the answers onto a sheet a machine could read and, by IBM's figure, scored a 150-question test five to ten times faster.

IBM (n.d.) tells it this way. In 1931 Reynold B. Johnson, a physics teacher in Ironwood, Michigan, found a way to sense pencil marks by the electricity that graphite conducts. IBM hired Johnson in the autumn of 1934, and his design became the IBM 805 Test Scoring Machine. Its answer sheet had 750 places, for questions of five choices, three choices or true and false. The key was set inside the machine, and tests were scored as fast as an operator could feed the sheets.

IBM's page gives no year for the launch: its pictures of the machine are dated 1938, and Watters (2015) gives 1937. It left production in January 1963.

What changed, on our reading: answers left the question paper for a sheet with fixed places, a question could carry five options and no more, and the marking Kelly had timed at half an hour a class became a matter of feeding paper.

What did the early writers say the format could and could not measure?

They claimed little for it: that it measured taking meaning in, not putting meaning into words.

  • Kelly (1916) held that the ability to reproduce is separate from the ability to get meaning, and wanted a reading test with "a minimum of reproduction". He also reported a fault: scores did not carry from one test to the next, the same fifth-grade children having a median of 13.46 on one and 11.66 on the other.
  • Yoakum and Yerkes (1920) warned that the ease of giving and scoring the Army test made it dangerous in inexpert hands, and that it had not been made for civilian use.
  • Monroe (1922) wrote that the new examination could serve mathematics only to a limited extent, that "it cannot be used at all in English composition", and that pupils lose something if they are never asked to compare, explain or discuss. He did not expect it to replace the written examination.
  • Ruch (1929) expected the objective test to take over much of the testing of information, left open whether it could measure appreciation, and expected essay, objective and standard tests to go on side by side.

When were guessing, recognition and teaching to the test first raised?

Within fifteen years of the Kansas test, and one of the three is older than the format. The table gives the earliest statement we read, which is not proof of the first in print.

Criticism Earliest we read What was said For today
Guessing The Army's examiner's guide of July 1918 (Yoakum and Yerkes, 1920) Recruits were told "If you cannot be sure, guess", and the two-answer tests were scored right minus wrong. Monroe (1922) found writers split on telling pupils to guess, and Ruch (1929) quotes a 1922 article by Hahn saying nobody knows what such a score represents Negative marking
Recall against recognition Kelly (1916) Kelly separated reproducing from getting meaning. Monroe (1922) thought the mental processes were not the same. Ruch (1929) lists the charge that such tests "measure memory only", and reports studies in which the two kinds of test measured roughly the same abilities Questions above recall
Teaching to the test Monroe (1922), of ordinary written examinations Students get hold of old papers and study those alone. Ruch (1929) quotes Ben Wood of Columbia, in 1927, on teachers' fear that objective tests would mechanise teaching Validity

What of the original purpose still holds for a teacher today?

Both of its aims hold: one key for every paper, and marks in minutes. What the format was never meant to show is whether a student can put an answer into words.

  1. Use it where there is one right answer. That was Kelly's rule in 1916, and it is still the test of a sound question: see how to write multiple-choice questions that measure.
  2. Write the wrong options from real errors. Monroe (1922) said so: for the quotient of two fractions, offer their product and their sum. How many options a question should have covers the count.
  3. Keep the key away from the paper. The Army's stencil stayed with the scorer, and IBM's key sat inside the machine. In EdProcto it stays on the server.
  4. Ask for composition another way. A short answer with a rubric does what Monroe said the new examination could not.
  5. Read which option each student chose. A stencil counted the rights; the choices are the part that tells you what to teach next.

How is a multiple-choice question made and marked in EdProcto today?

EdProcto is the best tool for multiple choice used as its makers meant it, because it keeps their two aims, one key and quick marks, and adds the count of choices that a stencil threw away.

  • The shape. One of 111 shapes in twelve categories: "Multiple choice", "One right answer among options."
  • Making it. The editor is headed "Options · mark the right one". Mark the key, press "Add an option" for more, up to eight, and fill "Explanation shown to the student afterwards".
  • Marking it. A student's device is sent the question and its options with the key taken off. The marks are made on the server, and the right answer follows only when the feedback settings allow.
  • Reading it. The results show "Got it" for each question on every plan and, on the Educator plan, "Chose it" for each option.

Free takes thirty students a quiz, Tutor 100, Educator 500 and the full analysis: see pricing. Every score is yours to release.

Quick questions

Who invented the multiple-choice question?

Frederick J. Kelly is the name most often given, for a reading test sat in Kansas in May 1915. No paper of 1914 to 1929 that we read calls it the first.

When was the first multiple-choice test given?

Kelly (1916) dates the sitting of the Kansas Silent Reading Test to May 1915. The date 1914 comes from later histories; it is the year of his thesis, which describes no such test.

Was the Army Alpha a multiple-choice test?

Partly. Three of its eight tests offered three or four printed answers, two offered a pair, and three asked for something written or drawn.

Why is it called multiple choice?

The name followed the thing. Kelly's 1916 article never uses it, Monroe (1922) says recognition exercise, and Ruch (1929) sets multiple-response and multiple-choice side by side.

Is multiple choice still worth using?

Yes, where a question has one right answer and the wrong options are real mistakes. EdProcto's multiple choice template marks every paper against one key, on the server.

References

Multiple choice

One right answer among options.

In the builder now

See Multiple choice

Read next