Skip to main content
Category: Effects
Type: Later inferences that still use a cause after the reader can report that the cause was corrected. The counted comparisons are a scripted warehouse fire, not a belief rating that rose
Origin: Hollyn M. Johnson and Colleen M. Seifert, “Sources of the continued influence effect: When discredited information in memory affects later inferences,” Journal of Experimental Psychology: Learning, Memory, and Cognition 20(6), 1994, pages 1420–1436. That PDF was not opened. The title line is the one in Seifert’s 2002 reference list. The counted text is Colleen M. Seifert, “The continued influence of misinformation in memory: What makes a correction effective?”, Psychology of Learning and Motivation 41, 2002, pages 265–292, DOI 10.1016/S0079-7421(02)80009-3. The author’s PDF was read. The warehouse script is the one printed in that PDF. A. L. Wilkes and M. Leatherbarrow, Quarterly Journal of Experimental Psychology 40A, 1988, pages 361–387, was not opened
Also known as: Continued influence of misinformation, the phrase in the 2002 chapter title. It is not the Backfire Effect as a rise in belief, and it is not the Illusory Truth Effect as a higher truth rating after a repeat
Quick Answer — Continued Influence Effect is the later use of a cause that a correction has already withdrawn, by a reader who can still report the correction. Johnson and Seifert named it in 1994. Seifert’s 2002 chapter, which was read, says 90 percent of people who saw the correction still referred to the paint cans and gas cylinders, against 20 percent of people who never saw that report. Around 90 percent accurately recalled the correction. A bare negation did not match the group that never heard the cause. A later cause that covered the same fire features did.

​
What is Continued Influence Effect?

Continued Influence Effect is the use of a retracted cause in a later inference, after the reader can still say what the correction was.
Yet despite these factors, the misinformation continues to influence later judgments after correction, a phenomenon called the “continued influence effect”
That sentence is Seifert (2002). The name in parentheses is Johnson and Seifert (1994). The 1994 paper was not opened. The four factors in the sentence just before it are these. The report and the correction are connected. Both are believable on their face. The correction marks the first report as incorrect. At the time of the questions, people can still recall both. A useful picture is a group chat about smoke in a kitchen. Someone writes that the oven held a pan of oil. A later message says the oven was empty. Asked what was corrected, people name the empty oven. Asked why the smoke was oily, some still name the pan. The picture is not a rate. The rate on this page belongs to one scripted stationery-warehouse fire. Illusory Truth Effect is a higher truth rating after a plausible line is repeated. Continued Influence Effect is use of a line after a correction the person can report. Sleeper Effect is a later rise in agreement with a message a weak source first held down. This page has no delayed agreement score.

​
Continued Influence Effect in 3 Depths

  • Beginner: If you can say “the closet was empty” and still answer “burning paint” when asked about the fumes, that pair is the claim. It is not yet a percent.
  • Practitioner: When you withdraw a cause, write a replacement that covers the same features the false cause covered. In this chapter, “the closet was empty” did not match the group that never heard about the cans. A later arson cause that could account for heat, fumes, and explosions did.
  • Advanced: The 90 percent and the 20 percent are Seifert’s summary of people who read this fire script. The chapter does not print the sample size. A newsroom, a classroom, or a household that copies 90 percent has left the script.

​
Origin

The name is the 1994 paper. Seifert’s chapter is the co-author’s later account of that work and of the studies around it. Wilkes and Leatherbarrow (1988) supplied the news-message method. Their paper was not opened. As Seifert describes it, people read on-the-scene reports of a warehouse fire. One message said cans of oil paint and gas cylinders had been stored where the fire started. Later messages added oily smoke, sheets of flame, explosions, and toxic fumes. Near the end, a message said the storage room had held no paint and no gas cylinders. Seifert writes that people who saw the correction were much more likely to mention those materials than people who never saw the first report, and that they did not differ statistically, on the number of answers involving the discredited materials, from people who saw the report and no correction. Johnson and Seifert (1994) moved the correction. One group read it in message 6, immediately after the cans. One group read it in message 11 of 13, after the fire’s features. Seifert reports no improvement from the early correction. Both groups still used the cans and cylinders as possible causes of explosions and fumes. The chapter reads that result as evidence that the inferences were not only old conclusions left unedited. People drew them when the questions arrived, after they already had the correction. Wilkes and Reynolds (1999) are cited as a replication of that timing result. Their paper was not opened.
As a whole, these studies suggest the direct negation of information is not successful at “removing” misinformation from memory. Instead, people benefit from corrections that either replace the misinformation with a better alternative, or at the least, explain how the contradiction came to be.
That sentence is the chapter’s close. The studies it gathers are Wilkes and Leatherbarrow (1988), Johnson and Seifert (1994), Wilkes and Reynolds (1999), Johnson and Seifert (1999), and Bush, Johnson, and Seifert (1994). The 1999 book chapter, the 1999 journal paper, and the 1994 conference paper were not opened. One sentence in the 2002 chapter dates Wilkes and Reynolds to 1996. The reference list prints 1999. This page uses 1999.

​
Key Points

Four splits keep a recalled correction, a mentioned can, and a replaced cause from collapsing into one slogan.
1

The correction can be recalled and still unused

Seifert prints protocols in which the same person writes “the paint and gas cylinders” as the cause of the fumes, and then writes that the second police message said the storage room was empty. Across the board, 90 percent of subjects who saw the correction referred to gas cylinders and paint cans. Twenty percent of control subjects, who never saw that report, did so. Around 90 percent accurately recalled the correction. Dropping the people who did not recall it did not change the results. Asked directly, fewer than 20 percent said the materials were responsible for the fire. The references show up when the question needs a cause, such as why the fire was intense. None of the answers treated the correction as a mistake or as intentionally misleading. The 1994 PDF was not opened, so this page does not add a sample size or a test statistic.
2

An early correction was not the fix

The editing account says people draw the inference before the correction and then cannot find it to revise. The early-correction group had the empty-closet message before the oily smoke, the explosions, and the fumes. Seifert reports no improvement over the late correction. The chapter’s reading is that the cause was used at question time. A speeded check in Johnson and Seifert (1998), as the chapter summarizes it, points the same way. A hockey player was said to play for Boston, then corrected to New York. People were faster to accept New York than a filler city, Philadelphia, and just as fast to accept Boston. The 1998 paper was not opened, so no reaction time is printed here.
3

Mentioning the words was not enough

Availability Heuristic is the neighbor the chapter tests and does not keep as the whole account. One version mentioned paint and gas only at another store that reported the fire. Another group never read the cans, then saw the words “gas” and “paint” in an unrelated task. Seifert says neither version produced the fire explanations. People used the materials when the story had offered them as a possible cause. After the correction, another group was asked to invent four or five causes of a fire of this type. They named arson, wiring, lightning, bombs, cigarettes, and gas leaks. Inventing that list did not lower the use of the cans on the inference questions. The cause had to be in the account, and it had to cover the features the questions asked about.
4

A covering replacement did what a bare negation did not

A later message offered arson: gasoline-soaked rags near the paper bales, and emptied steel drums. Seifert calls this the first comparison in which a group that saw a correction did as well as a group that never saw the cans, on using those cans as causes. The correction group still mentioned the cans. They were less likely to offer them as causes, and more likely to offer the arson materials. A follow-up varied the replacement. Whether the materials were typical of a stationery warehouse had no main effect. Replacements that could account for fumes and explosions reduced use of the cans. When coverage was low, the more plausible replacement was taken up more often. When coverage was high, plausibility did little. Bush, Johnson, and Seifert (1994), as the chapter reports that conference paper, compared kinds of correction. Undisputed reference produced more volatile-materials inferences than simple or enhanced negation. Those negations produced more than explanations of why the first report existed. Those explanations produced more than a story that never mentioned the cans. The chapter calls each of those steps significant. The explanations against simple negation alone were only marginally significant.

​
Applications

These four moves follow splits the 2002 chapter actually reports. They are not a claim that a rewritten correction changes a newsroom, a classroom, a court, or a household by 90 percent, 20 percent, 8 percent, or 23 percent.

Ask the fact and the cause apart

The fact question is what the second police message said. The cause question is why the fumes, the heat, or the explosions happened. In this chapter, people could answer the first and still use the cans on the second. A check that only asks “did you see the correction” has not yet run the inference.

Replace the cause that covered the features

“The closet was empty” states the negation. It does not say what produced oily smoke, sheets of flame, and explosions. The arson materials were the first correction Seifert reports as matching the never-heard group on causal use of the cans. People still mentioned the cans. A replacement is not a deletion from memory.

Do not stop at a list the reader invents

Asking people to generate other fire causes, after they had read the correction, did not reduce references to the cans. The alternatives that worked were written into the story and could cover the same features. A brainstorm in the margin is the condition the chapter says did not help.

Say why the first report was filed

One correction said the closet was empty because a trucker’s strike had blocked an expected delivery of paint and gas. Another said the closet held coffee and soda, not paint and gas. Seifert reports fewer inferences than after a bare negation, and calls that contrast marginal. Those readers still made more inferences than people who never heard the cans. Their extra references were more often hedged. An explanation of the contradiction is not the same size of change as a covering replacement.

​
Case Study

The account in the chapter’s appendix is a script, not a fire that was counted in a city. Message 1 is an alarm at 8:58 p.m. on January 25 at a wholesale stationery warehouse: offices, a display room, and a storage hall. Message 2 sends an engine at 9:00 p.m. Message 4, at 4:00 a.m. on January 26, has the fire captain attribute the start to a short circuit in the wiring of a closet off the storage hall. Message 5, at 4:30 a.m., is Police Investigator Lucas. In the misinformation version, cans of oil paint and pressurized gas cylinders had been in the closet. In the control, the closet was empty. The next messages are the features a cause would have to cover. Firefighters report thick oily smoke, sheets of flame, and intense heat. Explosions occur. No fatalities are in the script. Two firefighters go to the hospital after breathing toxic fumes. A small fire on the same premises six months earlier was put out by the workers. At 10:00 a.m. the owner estimates damage in the hundreds of thousands of dollars, and the premises were insured. Message 11, at 10:40 a.m., is the correction: the closet that had reportedly held paint and gas cylinders had actually been empty. The control’s second police message does not repeat the closet. It says the two firefighters had been released. Message 12 lists paper bales, envelopes, scissors, pencils, other school supplies, and photocopying machines. At 11:30 a.m. the captain says the fire is out and the hall is gutted. Seifert’s description of the 1988 procedure places a ten-minute distractor before the questions. The 1988 paper was not opened, so this page does not treat ten minutes as a time that was re-checked in the 1994 file. The questions split. Fact items ask what business the firm was in, what was in the hall, and how the fire was thought to have started. Inference items ask why the fire spread, what could have caused the toxic fumes, why it was intense, and what could have caused the explosions. Two items check the manipulation: what the second police message said, and whether the reader noticed a correction. The indicator is Seifert’s summary of people who saw the correction. Ninety percent referred to the gas cylinders and the paint cans somewhere in their answers. Twenty percent of people who never received that report did so. Around 90 percent accurately recalled the correction, and removing those who did not recall it left the pattern in place. Fewer than 20 percent, asked directly, said the materials were responsible for the fire. The lesson inside this script is that a reported empty closet and a later causal use of the cans can sit on the same sheet. The boundary is the same chapter. It does not print how many people produced the 90 percent. It says the studies are variations on one story. A second warehouse, a real alarm, or a different year is not this appendix.

​
Boundaries and Failure Modes

Continued Influence Effect, as this chapter measures it, is not a rate of belief and not a rate of forgetting. People could recall the correction. The gap is between that recall and the causes they offered for fumes, heat, and explosions. Two uses overshoot the script. One exports 90 percent, 20 percent, or “fewer than 20 percent” onto a newsroom, a class, a court, or a household. The chapter never prints the denominator. The other treats the arson replacement as proof that a correction deletes the first cause. Seifert says those readers still mentioned the cans. The chapter states its own limit. The findings are variations of a single account, and generality to other materials is left for later work. Wilkes (1988), as Seifert summarizes that chapter, found more later use when the corrected detail was central, and less when it was not. That chapter was not opened. Wilkes and Reynolds (1999), again as Seifert summarizes them, found that one unexplained causal feature was enough. The number of links was not the key. Their paper was not opened. A correction of a side detail, with no feature left hanging, is outside the result this page counts. Bare negation has a second, narrower miss inside the same script. In the correction-type comparison, 8 percent of people who saw a correction named the volatile materials as “the cause” of the fire, against 5 percent of the no-reference control. That is the direct-cause question. It is not the 90 percent, which counts any reference. The chapter then says only 23 percent failed to make at least one direct, uncontroverted reference to the materials. It does not print the count behind 23 percent, so this page does not assign it to one bar on the figure. The figure’s bar heights are not printed as numbers. The chapter also separates this effect from two older tasks. Directed forgetting of unrelated words is not this script. Post-event misinformation, in the Loftus studies the chapter cites, usually changes a peripheral detail and leaves people unaware of the conflict. Those papers were not opened. The contrast Seifert draws is the one to keep: here the correction is direct, people can report it, and the cause was central.

​
Common Misconceptions

Three readings turn a recalled correction, a mentioned can, or the word “backfire” into a different score.
The chapter’s manipulation check is the second police message. Around 90 percent accurately recalled the correction, and the inference pattern remained after the others were dropped. Fewer than 20 percent, asked in a direct way, blamed the materials for the fire. The 90 percent is references, often on questions that ask for a cause. Forgetting the correction is the wrong description of a person who writes both the empty closet and the burning paint.
The chapter tested that reading and did not get the fire explanations from it. Paint and gas at a neighboring store, and the words “gas” and “paint” in a side task, were not enough. A list of causes the reader invented after the correction was not enough either. The materials had to be offered in the story as a cause of this fire. Availability Heuristic is about judging from what comes to mind. It does not, on these pages, replace the causal-gap account.
Backfire Effect is a correction that leaves the false belief stronger. This chapter does not print a belief rating that rose past the undisputed story, in which the cans were never withdrawn. People could state the retraction. Seifert does use “backfired” for one wording: adding that no paint or gas had ever been stored on the premises produced more inferences than the simple empty-closet sentence. That is a comparison among corrections, not a measured increase above an uncorrected belief. Confirmation Bias would keep a prior view and collect fitting evidence. These readers had just met the story. Hindsight Bias is a finished outcome later called predictable. Predictability is not what the inference questions scored.
These six pages sit beside Continued Influence Effect. Each one asks a different question.

Backfire Effect

A correction that strengthens the false belief. Continued Influence Effect is later use of a cause the person can still report as corrected.

Illusory Truth Effect

A higher truth rating after a plausible statement is repeated. This page has no repetition rating. The correction is one message, and people can recall it.

Sleeper Effect

A later rise in agreement with a message a weak source first held down. The fire questions were not a delayed attitude score.

Availability Heuristic

Judging from what comes to mind easily. In this chapter, mentioning paint and gas off to the side did not produce the causal answers.

Confirmation Bias

Searching for, reading, and recalling what fits a belief already held. The warehouse readers were not selected for a prior view of the fire.

Hindsight Bias

Treating a known outcome as having been predictable. A cause offered after the fumes are known is a different question from whether the outcome looks foreseeable.

​
One-Line Takeaway

When you retract a cause, write the replacement that covers the same features, and do not treat 90 percent of one fire script as the rate for every correction.