Skip to main content
Category: Effects
Type: A faster correct decision that both letter strings are words, when the two words are common associates
Origin: David E. Meyer and Roger W. Schvaneveldt, Journal of Experimental Psychology 90(2), 1971, pages 227–234. The paper’s word is facilitation. It does not print Priming Effect
Also known as: Associative facilitation, the paper’s own word for this contrast
Quick Answer — The Priming Effect on this page is a faster correct decision that both letter strings are words when those words are common associates. David E. Meyer and Roger W. Schvaneveldt measured it in 1971. Yes responses were 85 milliseconds faster for associates than for the swapped pairs. A later claim, that words about old age make a person walk more slowly, is a different task. In a 2012 hallway with an automatic clock and no expectation planted in the experimenter, that walk did not slow.

​
What is the Priming Effect?

The Priming Effect, as counted here, is a faster correct decision that both letter strings are words when the two words are common associates than when they are not.
“Yes” responses averaged 85 ± 19 msec, faster for pairs of associated words than for pairs of unassociated words.
That sentence is Meyer and Schvaneveldt’s report of Experiment I. The two strings were on the screen together. Bread above butter is the kind of pair they mean. Bread above doctor is the comparison, made by swapping the second words. A hidden sentence about old age, followed by a walk down a hall, is not this contrast. Anchoring Effect is a number used as the start of an estimate. Primacy Effect is extra weight on the first item in a list. Neither one is a millisecond gap between two word pairs.

​
The Priming Effect in 3 Depths

  • Beginner: Two words that often go together are accepted as words faster than two words that do not. In the 1971 yes task the gap was 85 milliseconds, 855 against 940.
  • Practitioner: Write the related time, the unrelated time, and both error rates. In Experiment I the error rates did not differ. In Experiment II they did. A faster key is not yet a cleaner key.
  • Advanced: The 85 milliseconds and the 117 milliseconds are two tasks in one paper. A hallway time after old-age sentences is a third object. The 1971 table does not contain a walk.

​
Origin

The paper is David E. Meyer and Roger W. Schvaneveldt, “Facilitation in recognizing pairs of words: Evidence of a dependence between retrieval operations,” Journal of Experimental Psychology, volume 90, number 2 (1971), pages 227–234. The scanned PDF read here does not print a DOI. The publisher record is 10.1037/h0031564. The note at the end says the paper was received on 6 February 1971. Meyer’s byline is Bell Telephone Laboratories, Murray Hill, New Jersey. Schvaneveldt’s byline is the University of Colorado. A footnote says the work was begun at Bell Laboratories and at the State University of New York at Stony Brook. The paper’s own word for the result is facilitation. It does not print priming, prime, or Priming Effect. This page uses Priming Effect for the associate contrast they measured. It does not treat that name as a second result. Two tasks used the same idea. In Experiment I the person pressed yes if both strings were words, and no otherwise. In Experiment II the person pressed same if the two strings were both words or both nonwords, and different otherwise. The associate contrast keeps the required key fixed. Associated pairs and unassociated pairs both ask for yes, or both ask for same. The associated pairs, such as bread–butter and nurse–doctor, were taken from the Connecticut free-association norms the paper cites (Bousfield, Cohen, and Whitmarsh, 1961). Inside an associated pair, the lower word was the first or second most frequent associate of the upper word. Unassociated pairs, such as bread–doctor and nurse–butter, were made by swapping those response words. The lower word was then never the first or second associate of the upper word. The same pool of words therefore sits on both sides of the contrast. No person saw any string more than once. The median length was 5 letters, with a range from 3 to 7. The median frequency in the count the paper cites (Kučera and Francis, 1967) was 59 per million, with a range from 1,747 to less than 1. This page did not re-read the norms or the frequency book. Each experiment used 12 high school students, paid volunteers, run one at a time. The second group had not been in the first and was drawn from the same population. The paper does not name the school or the town. Experiment I lasted about 45 minutes. Experiment II lasted about an hour. The students started with 3inExperimentIand3 in Experiment I and 3.50 in Experiment II. The scanned penalty schedule is not clear enough to print. The discussion offers two ways the second decision might depend on the first. One is a spread of excitation to nearby memory locations. The other is time spent shifting from one location to another, as on a tape or a disk. The paper says the data do not test those accounts apart. It also says the associate effect does not, by itself, show that the meaning of the word was retrieved for the lexical decision. A location can be checked for “this is a word” without bringing the meaning up.

​
Key Points

Four readings keep each number on the contrast that produced it. The yes key is not the same key. The error rate is part of the result. A hallway is not in Table 1.
1

Use the swapped pair, not a new word list

The unassociated pairs were the associated pairs with the second words exchanged. Bread still appears. Butter still appears. They no longer appear together. A comparison that brings in easier words on one side has not repeated this control. The lower word of an associated pair was the first or second associate. The lower word of an unassociated pair was not.
2

Put the error rate next to the milliseconds

In Experiment I, yes responses were 855 milliseconds for associates and 940 for the swapped pairs. The difference is 85 ± 19 milliseconds, F(1, 11) = 20.6, p < .001. Errors were 6.3 percent and 8.7 percent. That error gap was not significant, F(1, 11) = 2.1, p > .10. Eleven of the 12 students showed an advantage larger than 30 milliseconds. The paper does not print the twelfth student’s gap. The errors tended to rise with the times, so the speed was not bought by missing more associates.
3

Keep the yes key apart from the same key

In Experiment II the same key averaged 1,055 milliseconds for associates and 1,172 for the swapped pairs. The difference is 117 ± 18 milliseconds, F(1, 11) = 42.6, p < .001. Errors were 2.4 percent and 8.7 percent, and that gap was significant, F(1, 11) = 16.6, p < .01. The associate effect did not differ significantly between the two tasks, F(1, 22) = 1.4, p > .20. That test is not a license to print 85 for both tasks. Yes responses to word pairs were also 216 ± 68 milliseconds faster than same responses, F(1, 22) = 10.2, p < .01. The 216 is a task gap. It is not the associate gap.
4

Leave the other column in Table 1 alone

No responses to a word above a nonword averaged 1,087 milliseconds. No responses to a nonword above a word averaged 904. The difference is 183 ± 14 milliseconds, F(1, 11) = 171.7, p < .001. The paper reads this as the person often stopping once the top string is not a word. It is a position contrast. It is not the 85 milliseconds. Availability Heuristic treats an easy example as a common event. A faster key in this table is not that judgment.

​
Applications

Write the pair, the key, and the error rate before the word priming is used for a decision outside this table.

Record both times and both error rates

For a word decision, write the associate time, the swapped-pair time, and the two error rates. Experiment I is 855 against 940, with errors 6.3 percent and 8.7 percent, and the error test is not significant. A note that only says “related words were faster” has dropped the test that keeps the speed from being a looser criterion.

Name the key before you quote 85 or 117

The 85 milliseconds is the yes key. The 117 milliseconds is the same key. The test of the difference between those two advantages is F(1, 22) = 1.4, p > .20. Quote the key you mean. A training slide that prints one of them as the size of priming has chosen a task and not said so.

Do not move the millisecond gap onto a poster

The 85 milliseconds is 12 high school students, two strings on one screen, and first or second associates. It is not a slogan, a package, or a price. Mere Exposure Effect is a change in liking after repetition. Framing Effect is the same fact in two wordings. A poster can be either of those. It is not Table 1.

Ask which clock a walking claim used

If the claim is that old-age words slowed a walk, ask whether the clock was a stopwatch or a beam, and whether the experimenter had been told to expect a slow walk. The 2012 automatic clock, with experimenters kept blind, did not show a slower walk. The 85 milliseconds does not fill that gap.

​
Case Study

Meyer and Schvaneveldt ran the yes-no task with 12 high school students. Each person saw 24 associated pairs and 24 unassociated pairs among 96 test trials, after practice. Half of the test trials required yes. A warning word, ready, was followed by a one-second fixation. The two strings then appeared in capitals, one above the other. The clock ran from the onset of the strings to the key. A correct yes used the right index finger. A no used the left. The person was told after each trial whether the key had been correct, and was told a block mean and an error total during a rest of about two minutes. Table 1 gives the mean correct times. Associates, yes: 855 milliseconds, errors 6.3 percent, a quarter of the trials. Swapped pairs, yes: 940 milliseconds, errors 8.7 percent, a quarter of the trials. The associate advantage is 85 ± 19 milliseconds, F(1, 11) = 20.6, p < .001. The degrees of freedom match 12 students. The error rates on those two rows did not differ, F(1, 11) = 2.1, p > .10. Eleven of the 12 students were faster by more than 30 milliseconds on the associates. The same table holds three no rows that are not the associate contrast. A word above a nonword took 1,087 milliseconds, with errors of 27.6 percent. A nonword above a word took 904 milliseconds, with errors of 7.8 percent. Two nonwords took 884 milliseconds, with errors of 2.6 percent. The 183-millisecond position gap is the paper’s evidence that a no often stops after the top string. This page does not call that gap priming. Experiment II gave the same test pairs to 12 new students and changed the rule. Same meant both words or both nonwords. Different meant one of each. Associates on the same key took 1,055 milliseconds, errors 2.4 percent. Swapped pairs took 1,172 milliseconds, errors 8.7 percent. The advantage is 117 ± 18 milliseconds, F(1, 11) = 42.6, p < .001, and the error advantage is significant as well. Pairs of nonwords on the same key took 1,357 milliseconds. That is 185 ± 29 milliseconds slower than the swapped word pairs, F(1, 11) = 40.7, p < .001. The abstract’s remark that same responses were slowest for nonwords is this row. It is not the associate gap. The lesson is two screens of letter strings, 12 students on each rule, an 85-millisecond yes advantage with no significant error gap, and a 117-millisecond same advantage with a significant error gap. The associate effect did not differ reliably between the rules. The 85 is not a person’s walk, and it is not the 183-millisecond position gap in the same table.

​
Boundaries and Failure Modes

The Priming Effect is the wrong summary in three retellings. The 85 milliseconds is treated as any later behavior. A null hallway is treated as proof that no earlier word ever matters. A stopwatch reversal is treated as the walk itself. The first boundary is the task. The 1971 contrast is two strings on one screen, a key for “both are words,” and pairs drawn from free-association norms. It is not a scrambled sentence, and it is not a corridor. The paper leaves the mechanism open. Spread of excitation and a shift between memory locations are both still available at the end of the discussion. A team that picks one of them as what the F test proved has gone past the paper. The second boundary is the walk. Stéphane Doyen, Olivier Klein, Cora-Lise Pichon, and Axel Cleeremans, PLoS ONE 7(1), 18 January 2012, article e29081, tried to repeat the walking result that John A. Bargh, Mark Chen, and Lara Burrows reported in 1996. This page did not re-read the 1996 paper. Doyen’s reference lists it as Journal of Personality and Social Psychology, volume 71, pages 230–244. Doyen says the combined samples of the two walking studies were 60 people, the clock was a stopwatch, and 1 of 19 people on page 237 was counted as aware. No 1996 walking time is printed here. Doyen’s Experiment 1 used 120 French-speaking Belgian undergraduates, ages 20 to 34, average 21.9. It was a required part of a practical course at the University of Brussels. Four student experimenters, ages 21 to 24, average 22.5, followed a script. Questionnaires sat in envelopes, so the experimenter was not supposed to know the condition. Two infrared beams, 9.75 meters apart, timed the walk. That distance is the one Doyen attributes to the 1996 hallway. The prime was 30 scrambled sentences with a word tied to old age. The other version replaced those words with neutral words. There was no time limit. Using the time to enter the room as a covariate, the time to leave was 6.27 seconds after the old-age sentences (standard deviation 2.15) and 6.39 seconds after the neutral sentences (standard deviation 1.11). F(1, 119) < 1, and eta-squared was .01. The paper says this is no significant difference. The prime mean is the smaller of the two, so the printed means are not a slower walk. This page does not turn 6.39 minus 6.27 into an effect. It also does not divide 120 into two cells the paper does not print. Four people in the prime condition, which the paper calls 6.66 percent, said the sentences were about old people. On a four-picture choice the prime group picked the older person above chance, chi-square (1) = 5.43, p = 0.023. Ninety-six percent said they could not link the sentences to the walk. A slider for whether their own walk had changed, scored as a deviation from the start of a 0-to-100 scale, was 1.7 against 2.68, t(1, 119) < 1. No experimenter reported an expectation. Experiment 2 recruited 50 new participants, ages 18 to 30, average 21.3, and 10 experimenters, ages 20 to 24, average 21.7. Each experimenter tested five participants. Half of the experimenters were told that a primed person would walk more slowly. Half were told the person would walk faster. The briefing lasted an hour. The first person each experimenter saw was a confederate instructed to walk as the experimenter had been led to expect. The condition was no longer hidden. This page does not subtract those first persons from the 50, and it does not rebuild the degrees of freedom. On the infrared clock, with experimenters who expected a slow walk, the prime mean was 6.95 seconds (standard deviation 0.36) and the neutral mean was 6.52 (standard deviation 0.31), F(1, 24) = 7.07, p = .014, eta-squared .228. With experimenters who expected a fast walk, the means were 6.07 (standard deviation 0.57) and 6.01 (standard deviation 0.39), F(1, 24) = .231, p = .636, eta-squared .011. The paper calls that second pair no difference. Across prime conditions, people tested by the slow-expectation experimenters took 6.75 seconds, and people tested by the fast-expectation experimenters took 6.04, F(1, 49) = 30.44, p < .001. The 6.75 is not the prime cell. The prime cells are 6.95 and 6.07. The stopwatch told a further story. Under the slow expectation it agreed in direction: 7.25 seconds (standard deviation 0.33) against 6.73 (standard deviation 0.32), F(1, 24) = 12.32, p = .002. Under the fast expectation it reversed: the prime mean was 5.8 seconds (standard deviation 0.73) and the neutral mean was 6.43 (standard deviation 0.4), F(1, 24) = 7.55, p = .012. The paper’s interaction for those stopwatch times is F(3, 46) = 18.82, p < .001. Three of those F values are printed with a decimal comma in the file read here, as 12,32, 7,55, and 18,82, and the 6.43 is printed as 6,43. This page uses periods. It does not relabel the parentheticals. The figure captions say the bars are standard errors of the mean, while the prose says SD. Experiment 1’s standard deviations are 2.15 and 1.11. The Experiment 2 parentheticals are much smaller. Both labels stay as printed. The paper says a stopwatch-only study would have reported a reverse effect, and it attributes that reversal to error in stopping the watch. The absolute gaps between the two clocks, in the fast-expectation condition, are 0.73 seconds for the prime group (standard deviation 0.38) and 0.41 seconds for the neutral group (standard deviation 0.24), F(1, 24) = 5.819, p = .023. The sentence that calls the neutral cell the less accurate one does not match a larger printed gap on the prime cell. This page prints the two gaps and does not choose the sentence over the means. The authors’ conclusion has several sentences that do not fold together. Experiment 1 did not reproduce a slower walk. In Experiment 2 the objective slowing appeared when the experimenter expected slowing, and it did not appear when the experimenter expected speeding. They say this is not only a self-fulfilling prophecy, because the fast expectation did not make the infrared walk faster. Self-Fulfilling Prophecy is that neighboring page. They also say the primes alone were not sufficient, and that the scrambled-sentence method should be studied before the walk is treated as established. The abstract says a walking-speed effect was obtained only when the experimenters believed the primed people would walk more slowly. Ninety-five percent of the Experiment 2 prime group said they saw no link between the sentences and the walk. In the slow-expectation condition their slider deviation was −7.3, against 2.01 for the neutral group, F(1, 24) = 12.43, p = .042. That −7.3 is a mark on a 0-to-100 slider. It is not 7.3 seconds of walking. The third boundary is the export. The 85 milliseconds, the 117 milliseconds, and the 6.27 seconds are not a shop, a clinic, a classroom lesson, or a poster. The 6.27 is one hallway after French scrambled sentences, with the entering time used as a covariate, and it is not a significant slowing. The 6.95 is a different hallway sample, and only the slow-expectation arm. A team that treats every earlier word as an 85-millisecond gain has not read Table 1’s other rows. A team that treats every earlier word as a slower walk has not read Experiment 1.

​
Common Misconceptions

Three retellings travel with the name. One sends old-age words straight into the legs. One treats 85 milliseconds as the size of any related cue. One treats the faster yes as proof that the meaning was looked up.
Doyen’s Experiment 1, with 120 students and infrared beams 9.75 meters apart, found 6.27 seconds after the old-age sentences and 6.39 seconds after the neutral sentences. The paper reports no significant difference. The prime mean is not the slower one. Experiment 2 found an infrared slowing, 6.95 against 6.52, when the experimenter had been told to expect it, and no infrared difference, 6.07 against 6.01, when the experimenter had been told to expect a faster walk. The stopwatch reversal under that second instruction is a different instrument. Bargh, Chen, and Burrows were not re-read, so their seconds are not here to add to these.
Meyer and Schvaneveldt write that the associate effect does not necessarily mean the decision retrieved the word’s meaning. Their account allows a check that a string is a word, using a memory location, without bringing the meaning up. They also leave the spread-of-excitation story and the location-shift story untested against each other. The significant F on the 85 milliseconds is a time difference. It is not a diagram of the memory.
The Priming Effect sits beside a number that starts an estimate, beside extra weight on a first item, beside liking that grows with repetition, beside a judgment that uses ease of recall, and beside a belief that helps produce the outcome someone expects.

Anchoring Effect

A number used as the start of a later estimate. The contrast here is two word strings and a key. It is not a starting number.

Primacy Effect

Extra weight on the first item in a series. Bread above butter is a relation between two words, not the first slot in a list.

Mere Exposure Effect

Liking that increases after repetition. The 1971 clock is a word decision. It does not score how much the pair was liked.

Availability Heuristic

An easy example treated as a common event. A faster yes to bread and butter is not a judgment about how often that pair occurs in the world.

Self-Fulfilling Prophecy

An expectation that helps bring about the outcome it names. Doyen’s infrared clock did not speed up when the experimenter expected a faster walk.

​
One-Line Takeaway

Before you say an earlier word changed the next response, write the two times, the two error rates, and whether the clock was a key or a hallway.