---
url: "https://questionstar.com/blog/survey-and-witness-interview/"
title: "Surveys and witness interviews: two fields, one question"
description: "What surveys and witness interviews share: one word in a question changes answers and memories. The experiments, the 1983 meeting, and where the parallel ends."
lastmod: "2026-11-13"
type: "blog"
---

# Surveys and witness interviews: two fields, one question

*What do a survey and a witness interview have in common, and why do the two fields know so little about each other?*

What surveys and witness interviews share: one word in a question changes answers and memories. The experiments, the 1983 meeting, and where the parallel ends.

> **A question before we start.** It comes from an American survey published in 1941. Please answer it in your head, yes or no:
> 
> This question was published in 1941 in two versions. Please read only one of them for now.
> 
> **Version A:** "Do you think that the United States should forbid public speeches against democracy?"
> 
> **Version B:** "Do you think that the United States should allow public speeches against democracy?"
> 
> Remember your answer. We'll come back to it.
> 
> *Wording as quoted in Plous 1993, pp. 68-69.*

Seattle, November 1980. Six months earlier, on May 18, the volcano Mount St. Helens erupted, and most people in the American Northwest still know when that was. Over three days, from November 17 to 19, 144 people are asked to take a short survey: people in and around Seattle and airline passengers, among them the people waiting for a flight that fog has delayed by an hour. The topic is crime. Has anyone stolen something from you? Has anyone tried to rob you? Has anyone beaten you up or attacked you? Some hear the questions with the words "In the last 6 months", the others with the words "Since the first major eruption of Mt. St. Helens" (Loftus and Marburger 1983, p. 116).

Both wordings cover the same period. Yet in the first group 7 of 76 respondents report at least one crime, which is 9.2%. In the second group it is 1 of 68, or 1.5% (ibid.). A follow-up question points to one explanation: of the seven, two say the crime actually happened longer ago, and three are no longer sure whether it happened before or after the eruption (ibid.). "Six months" is a period without an edge, and older experiences slip into it easily. The volcano gives the period an edge that is easy to find in memory.

The idea came from Elizabeth Loftus, a psychologist better known for her experiments with eyewitnesses. Here she is trying out a better question for a large American survey of crime victims (ibid., pp. 114-115). That leads to a bigger question:

*What do a survey and a witness interview have in common, and why do the two fields know so little about each other?*

In short: survey research and research on witness interviewing have the same task. Both want to learn something from people that can be found only in their heads, and in both cases the question helps shape the answer. Both fields picked up ideas from memory research for this, partly from the same people. Different methods grew out of those ideas. Since then the two have run side by side: the studies this essay draws on show only a few connections.

We'll take it step by step. You'll find out:

- [how a single word shifts an answer](#one-word), for witnesses and for survey respondents alike,
- [when the two fields met](#st-michaels) and why [two different methods carry almost the same name](#one-name) today,
- [what both have found out about the same problems](#same-problems),
- [where the transfer from the witness interview to the survey ends](#limits).

The idea for this essay came from a reader. He wrote to me that surveys and witness interviews have a lot in common, and that it is a pity the two fields know so little about each other. The thought made sense to me, and I went and read up on it. What I found surprised me.

## One word is enough

In 1974 Elizabeth Loftus and John Palmer show 45 students seven short films of traffic accidents, each between 5 and 30 seconds long. After each film the viewers fill in a questionnaire. One of the questions reads: "About how fast were the cars going when they hit each other?" But only nine people get the version with "hit". The others get versions with four other verbs, with nine people assigned to each verb (Loftus and Palmer 1974, p. 586).

| Verb in the question | Mean estimate in miles per hour | In km/h |
|------|------|------|
| smashed | 40.8 | 66 |
| collided | 39.3 | 63 |
| bumped | 38.1 | 61 |
| hit | 34.0 | 55 |
| contacted | 31.8 | 51 |

*The same films, the same question, a different verb. Values from Loftus and Palmer 1974, Table 1, p. 586. The km/h values are converted and rounded.*

Between "contacted" and "smashed" the mean estimates differ by nine miles per hour, about 15 km/h, for the same films.

One detail is usually left out. For four of the films the true speed was known: 20, 30 and twice 40 miles per hour. The mean estimates were 37.7, 36.2, 39.7 and 36.1 (ibid.). So whether a car was doing 20 or 40 made hardly any difference to the estimate. The verb in the question made a bigger one.

Could people remember correctly but let the word guide their speed estimates? A second experiment tests that. This time 150 students watch a film in which the accident itself lasts four seconds. Of these, 50 are asked about the speed with "smashed", 50 with "hit", and 50 are not asked about the speed at all. A week later they all come back and answer a new question: "Did you see any broken glass?" There was none in the film. Sixteen of the 50 in the "smashed" group answer yes, compared with 7 of the 50 in the "hit" group and 6 of the 50 who were never asked about the speed (ibid., p. 587).

Now back to the question you answered at the beginning.

> **Both versions stood at the top: A with "forbid", B with "allow".**
> 
> Version A, "forbid": of those who gave an opinion at the time, 54% said yes. So they wanted to forbid such speeches.
> 
> Version B, "allow": of those who gave an opinion, 75% said no. So they did not want to allow such speeches.

Your own single answer can't show this. The difference appears only when many people each get one of the two versions.

Forbidding and not allowing come to the same thing. Yet there are 21 percentage points between the two versions (Rugg 1941, cited in Holleman et al. 2016, p. 3). If you count the undecided as well, 46% of all respondents wanted to forbid and 62% did not want to allow (cited in Plous 1993, pp. 68-69). Two versions of one question, assigned at random to two groups: this is still how you test what a wording does today. The article on [A/B tests in surveys](/blog/survey-experiments/) explains how.

There is an important difference between the survey and the accident film. With the film there is a truth: you can check whether any glass broke. With the question about the speeches there is none. Nobody can say which version measures the "right" opinion. All we know for sure is that the answer depends on the word.

How can one word do so much? The obvious picture of memory is a photo album. You open the page and the picture is there, a little yellowed perhaps, but the same as it was then. If that were true, a question couldn't change the picture. It would only decide which page we open.

The experiments fit a different picture better. A memory is like a dish that is cooked fresh every time. Only the ingredients are kept: individual impressions, a feel for how things unfolded, and knowledge of how such things usually go. When someone asks, the dish is cooked from them, along with everything else that happens to be in the kitchen. That includes the question itself. The word "smashed" is an ingredient: it brings force with it, and broken glass goes with force. A person who cooks again a week later may find the broken glass already in the pantry and can no longer tell whether it came from the film or from the question.

Put plainly: remembering means reconstructing. What a person reports can be put together from what they perceived and from what was added later, including by questions. This is how Loftus and Palmer read their results. Over time, they write, it becomes impossible to tell which of the two sources a detail came from: "All we have is one 'memory'" (1974, p. 588). The kitchen picture has a limit, though. A cook knows what goes into the pot. A person who remembers doesn't notice.

> 💡 **How solid is this?**
> 
> In the first experiment of 1974 only nine students got each verb, and they watched films, not real accidents. In the second experiment, even after "smashed", 34 of 50 said they had seen no glass. The wording changed the share of yes answers, but most people in that group still said no. According to Rolf Zwaan, there has still been no large direct replication using modern methods (Zwaan 2025). The difference between "forbid" and "allow", by contrast, has been retested many times. It varies a great deal from question to question. Sometimes it is absent, and sometimes it reverses (Holleman et al. 2016, p. 3).

## Answers about something that wasn't there

The broken glass is not a one-off. In 1978 Loftus and two colleagues show 195 students a series of 30 slides: a red Datsun drives up to an intersection, and then an accident happens. One half sees a stop sign at the intersection, the other half a yield sign. Afterward everyone answers 20 questions. One of them asks in passing whether another car passed the red Datsun "while it was stopped at the stop sign". Other viewers read the same question with a yield sign. So for half of the viewers the question names the wrong sign. Twenty minutes later everyone has to pick, out of two slides, the one they saw. Those whose question named the sign they had seen pick the correct slide in 75% of cases. Those whose question named the other sign do so in 41% (Loftus et al. 1978, pp. 21-22).

Nobody claimed that a stop sign stood there. The question took it for granted and asked about something else.

The experiment did not show that this also holds for questionnaires. That is an inference. But it is a natural one, because there too almost every question takes something for granted. "How satisfied were you with the advice you received?" assumes that you received advice. "When did you last use our app?" assumes that you have used it. A person who never got any advice first has to contradict the question. Whether everyone does that, the slide experiment did not test. It only suggests being careful with such assumptions.

## Summer 1983: a week in St. Michaels

In the early 1980s a few people noticed that memory research and survey research have the same problem. From June 15 to 21, 1983, two groups that otherwise have little to do with each other meet in St. Michaels, Maryland: cognitive psychologists such as the memory researcher Endel Tulving, and survey methodologists such as Norman Bradburn, Howard Schuman and Roger Tourangeau. The host is the Committee on National Statistics of the U.S. National Research Council (Jabine et al. 1984, pp. v and 1). The list of participants doesn't clearly identify anyone as a researcher on witness interviewing. Elizabeth Loftus with her eyewitness experiments is not on the list either, although she is involved in the project (ibid., pp. v-vi and 25).

The organizers have come up with a nice way to start. Before the seminar almost all of the participants are interviewed themselves, as ordinary respondents in the large American health survey, by interviewers from the Census Bureau. Two of these interviews are filmed and shown at the seminar (ibid., pp. 150-151). The experts are meant to find out for themselves what it is like to be asked how often they saw a doctor in the past twelve months.

The seminar report carries the subtitle "Building a Bridge Between Disciplines". Its diagnosis: until a few years earlier the two disciplines seemed to have had "little contact" (ibid., p. 1).

Within the project Loftus suggests having respondents think aloud while they answer. She tries it out in five conversations, with questions from real surveys. Even these few conversations show something. One participant first answers the general question about visits to the doctor, the dentist and the hospital. Asked later about the dentist alone, the same person remembers two more visits (ibid., pp. 61-62).

This suggestion grew into a method that is a standard tool today: cognitive interviewing, which is used to test a questionnaire before it goes out to respondents (Beatty and Willis 2007, p. 289). You ask test participants the questions and have them tell you how they understand each question and how they arrive at their answer. How to run such conversations is covered in the pretest article, in the section [Cognitive interviews](/blog/questionnaire-pretest/#cognitive-interviews).

The timeline shows what came before and what followed. The next section tells the story of the lower track.

## Almost one name, two methods

At the same time, in Los Angeles, something with almost the same name takes shape. In 1984 the psychologists Edward Geiselman and Ronald Fisher and their colleagues present an interviewing technique for witnesses in a police science journal and call it the "cognitive interview" (Geiselman et al. 1984, cited in Memon et al. 2010, p. 366).

A year later it is put to the test. In that study 89 students watch police training films, among them a bank robbery and a store robbery. About 48 hours later they are interviewed by 17 experienced investigators, in one of three ways: in the usual way, under hypnosis, or with the cognitive interview (Geiselman et al. 1985, pp. 403-404). The cognitive interview consists of four plain instructions. Put yourself back in the situation. Report everything, even what seems unimportant to you. Describe the events in a different order as well. Describe them from a different perspective (ibid., p. 404).

With the usual interview the witnesses report 29.4 correct details on average, with the cognitive interview 41.2. Hypnosis comes in between at 38.0, so in this experiment it does no better than the four instructions. The number of incorrect details is similar: 6.1 with the usual interview and 7.3 with the cognitive interview (ibid., p. 405).

The two similar names refer to two different methods. Fisher and Geiselman were not in St. Michaels (Jabine et al. 1984, pp. v-vi). And the goals differ. The police cognitive interview is meant to help the witness remember. Cognitive interviewing in survey research tests the question, not the person. What the two share are ideas from memory research: the 1985 paper already refers to Loftus's work on how the wording of a question changes what a witness says (Geiselman et al. 1985, p. 401).

The two names differ only in their endings, and a footnote shows that survey researchers know about the other method. In 2007 Paul Beatty and Gordon Willis sum up the survey method and call it "cognitive interviewing". In their very first note they make clear that it is not the "cognitive interview" for eyewitnesses (2007, p. 287). In everyday use the ending often gets lost: a single interview that tests a questionnaire is called a cognitive interview too, and our own pretest article says so as well. In some languages the names are identical. In German, for example, both methods go by exactly the same name.

> 💡 **How solid is this?**
> 
> The 1985 experiment had 89 students as witnesses, and the advantage showed up in only two of the four films (Geiselman et al. 1985, p. 410). A later review that pooled 46 papers with 2,887 participants finds a large increase in correct details and a small increase in incorrect ones (effect sizes d = 1.20 and d = 0.24). The percentage of reported details that were correct differed by less than one percentage point on average in the 19 studies that report it (Memon et al. 2010, pp. 352-354). So you learn more. The percentage that is correct stays almost the same on average. Studies from real investigations were left out, because there it can't be reliably checked which details are true (ibid., p. 353).

## What both fields know about the same problems

Put the findings side by side and you see how often both fields faced the same difficulty and studied it with their own goals in mind. You already know the first two rows of the table. The next sections deal with the others.

**One problem, two fields**

| Problem | In surveys | In witness interviews |
|------|------|------|
| A word in the question shifts the answer | "forbid" 54%, "not allow" 75% (among respondents with an opinion) | "smashed" 40.8 miles per hour, "hit" 34.0 |
| The question changes what people report about their experiences | "In the last 6 months" 9.2% victims, "Since the first major eruption" 1.5% | wrong sign in the question: 41% pick correctly instead of 75% |
| Ask openly or offer answers | rule of thumb for answer categories: at most 10% of answers in "Other" | with children: answers to open prompts are three to five times as informative |
| How much freedom does the person asking get? | fixed wording: 28% correct answers in difficult cases, 87% when interviewers may explain terms | with a fixed protocol investigators ask at least three times as many open questions |
| Testing the instrument beforehand | cognitive interviewing in rounds of 5 to 15 interviews | questionnaire for witnesses from 2009: tested for clarity over several rounds |

*Sources, row by row: Rugg 1941, cited in Holleman et al. 2016, p. 3, and Loftus and Palmer 1974, p. 586. Loftus and Marburger 1983, p. 116, and Loftus et al. 1978, p. 22. Malhotra and Birks 2007, pp. 484-485 (see [Open-ended questions](/blog/open-ended-questions/)), and Lamb et al. 2007. Schober and Conrad 1997, figures from the 1996 working paper, and Lamb et al. 2007. Beatty and Willis 2007, p. 295, and Gabbert et al. 2009.*

## The bridge from the other side

In the decades after 1983, witness interviewing moves toward the survey, and survey research starts an argument that points the other way. Nothing in the sources says that the two fields coordinated this.

Witness interviewing gets rules. In England and Wales a fixed procedure for police interviews is created in 1992 (Milne et al. 2019, p. 3 of the manuscript version). For conversations with children, a group of researchers develops a protocol that is built almost like a questionnaire: first the ground rules and some time to get acquainted, then the free account, and only at the very end questions with answer options. With this protocol investigators ask at least three times as many open questions as before. That matters, because children give three to five times as much information after an open prompt as after a focused question, and what they report is more likely to be accurate. The insight is not new. It is more than a hundred years old: around 1910 the German psychologist William Stern found, with children and adolescents, that "free recall produced the fewest errors" and misleading questions the most (cited in Ceci and Bruck 1993, p. 406). Even with the protocol, though, about a quarter of the details still come from questions with answer options or from suggestive questions (Lamb et al. 2007).

Survey research, meanwhile, argues about the opposite: whether its fixed wording is too strict. In 1997 Michael Schober and Frederick Conrad have 41 people answer questions from official surveys. The answers are about invented cases on paper, such as floor plans and receipts. That way it is known which answer is correct. Some cases are simple, others are tricky. A tricky one looks like this: the question is "How many half bathrooms are there in this house?", and the floor plan shows a room with only a single bathroom fixture. By the official definition the room doesn't count, because the definition requires two fixtures, for example a toilet and a sink. Without this definition it is unclear whether the room counts. When the interviewers only read out the fixed wording, about 97% of the answers are correct in simple cases, but only about 28% in tricky ones. When they are allowed to explain what a term means, 87% are correct in the tricky cases too. The price: a typical interview takes 11.5 minutes instead of 3.4 (medians, Schober and Conrad 1997, figures from the 1996 working paper).

So witness interviewing gets firmer rules, and survey research asks itself whether its own rules are too firm. It is not quite a mirror image. On one side practice has changed. On the other there is an argument and an experiment. And different things are fixed. The investigators' protocol sets the order and the kind of questions and leaves the wording free. The questionnaire fixes the wording.

## Where the transfer ends

So far it sounds as if survey researchers only had to copy from the forensic psychologists. In three places that doesn't work.

**First, it depends on what is being asked.** The aids used in witness interviews are aids to remembering. They fit where there is something to remember. In one sentence: where a question asks about something people experienced, memory aids can help in surveys too. Where there is nothing to look up, what remains from the experiments in this essay is mainly care in the choice of words.

| What is being asked | Example | Is there a truth the answer could be checked against? | What can be transferred from witness interviewing |
|------|------|------|------|
| Facts, behavior and knowledge | How often did you see a doctor in the past twelve months? What does your phone plan include? | Yes, in principle | Memory aids: a fixed time anchor like the volcano, ask in general first, then follow up specifically |
| Attitudes | How satisfied are you with your manager? | No | The warning: the word shapes the answer. Whether memory aids help here too, the experiments described don't show |

The evidence in this essay comes from questions about things people experienced: crimes, visits to the doctor, an accident. It seems likely that the same holds for knowledge questions, because they have a correct answer too. But this essay offers no separate evidence for it. For knowledge questions there is also a rule from survey research itself: [offer "don't know"](/blog/rating-scale/#dont-know).

The second row is the most important one. The large review of the cognitive interview had to exclude all studies from real investigations, because there it can't be checked what is true (Memon et al. 2010, p. 353). With attitude questions that is the permanent state of affairs. Satisfaction can't be looked up the way a doctor's appointment can.

**Second, it depends on how many people are asked.** An investigator has one witness, or three, and every detail counts. A survey has a thousand respondents and is interested in the share or the average. If one person errs upward and another downward, this largely evens out over many respondents. But that is true only of errors without a direction. In Seattle two respondents reported crimes that lay further back than the question allowed. Errors like these can distort a share even with a thousand respondents. The same goes for the shift a word causes: all thousand read the same word, and if it pushes the answers in one direction on average, that shift stays. The large number doesn't make the result more correct then. It only makes it look more reliable.

**Third, a witness is not a suspect.** All the examples so far are about people who are asked to report what they experienced and whom nobody accuses of anything. Where someone is supposed to confess, different rules apply. For questionnaires, the following experiment mainly offers a warning: a yes given under pressure need not be true. In 1996 Saul Kassin and Katherine Kiechel have students type letters on a computer. Beforehand the students are warned not to touch the ALT key, because the program would crash. After a minute the computer stops, and the experimenter accuses them of having pressed the key. None of them did. Even so, 69% of the 75 participants sign a confession. When the typing also has to be fast and a supposed witness says she saw the participant press the key, all 17 sign (Kassin and Kiechel 1996, pp. 126-127).

> 💡 **How solid is this?**
> 
> There were 75 students, and the "offense" is an accidental keystroke. The only threatened consequence of confessing was a phone call from the lead researcher. The authors themselves say it remains open how far this can be transferred to real suspects (Kassin and Kiechel 1996, p. 127). The experiment shows that pressure and false evidence can produce agreement. It does not show how often false confessions occur.

That leaves the hope that one could at least tell whether someone is telling the truth. A review that pooled 206 papers arrives at 54% correct judgments of "lie" or "truth" (Bond and DePaulo 2006, p. 214). With equal numbers of true and false statements, a person tossing a coin would be right 50% of the time. So people are hardly better at spotting a lie than they would be by guessing.

In court an expert's impression is not enough either. One example comes from Germany. Since 1999 the country's Federal Court of Justice has required experts who assess the credibility of a witness statement to take a step that will look familiar to statisticians: the expert starts by assuming that the statement is untrue, and the court itself calls this the null hypothesis ("sog. Nullhypothese" in the German original). Only when this assumption no longer fits the facts is it rejected (BGH, judgment of July 30, 1999, 1 StR 618/98).

And a person who answers honestly can still be wrong. What remains is the work on the question.

## Two banks, half a bridge

The bridge of St. Michaels was started from both banks and never finished. There are a few connections. In 1989 a report from the U.S. health statistics agency explicitly adopts techniques from the police cognitive interview (Means et al. 1989, summary), and the footnote of 2007 shows that the two sides know of each other. But there is a counterexample too. In 2009 researchers on witness interviewing develop a booklet that witnesses fill in themselves, tested for clarity over several rounds. That is a questionnaire. The paper's reference list does not contain Tourangeau, Schwarz, Dillman, Belli or Sudman, five well-known names from research on questionnaires (Gabbert et al. 2009). That is no proof of a lack of exchange, but it is a sign.

One side looks for the truth about a single incident, the other for numbers about many people. At the core both do something similar: they ask someone to cook something from what is in their pantry. The question is always one of the ingredients. You can't leave it out. You can only choose it with care. A good question helps decide what we learn at all and how far we can trust what we have learned.

So was the reader right? Half right. The two fields did know each other, and for a short time even closely. But the acquaintance never turned into a collaboration.

If you notice something similar: [Write to me.](mailto:editors@questionstar.com)

*Dr. Paul Marx*

## Further reading

- [Pretest: cognitive interviews](/blog/questionnaire-pretest/#cognitive-interviews): how to ask test participants what they understand your question to mean.
- [Writing survey questions](/blog/writing-good-survey-questions/): how to word questions without suggesting an answer.
- [Open-ended questions](/blog/open-ended-questions/): when to ask without a list of answers and how to analyze what comes back.
- [Social desirability](/blog/social-desirability-bias/): why respondents sometimes give the answer that sounds good, and what can reduce it.
- [Order effects](/blog/order-effects-in-surveys/): how an earlier question can change the answer to a later one.

## Frequently asked questions

### What is a cognitive interview?

The term is used for two methods. In survey research it is an interview used to test survey questions: test participants say how they understand a question and how they arrive at their answer. The method as a whole is called cognitive interviewing (Beatty and Willis 2007, p. 287). In witness interviewing it is a technique that helps witnesses remember (Geiselman et al. 1985, p. 404).

### Is the police cognitive interview the same as the one used in a pretest?

No. The police cognitive interview is meant to improve a witness's recall. The cognitive interview in a pretest tests a questionnaire. Both emerged from memory research in the mid-1980s, but they have different originators (Beatty and Willis 2007, p. 287).

### Can you tell whether someone is lying?

Hardly. In a review that pooled 206 papers, people without aids or training made 54% correct judgments on average when they had to tell true statements from false ones. Guessing would be expected to yield 50% (Bond and DePaulo 2006, p. 214).

### What can surveys learn from witness interviews?

Mainly for questions about facts and behavior: tie a time period to an event whose date respondents know well (Loftus and Marburger 1983, p. 116), and put nothing into the question that respondents have not said themselves. For attitude questions the warning remains that a single word can shift the answer.

## Sources

Beatty, P. C. and Willis, G. B. (2007). Research Synthesis: The Practice of Cognitive Interviewing. Public Opinion Quarterly, 71(2), 287-311. DOI 10.1093/poq/nfm006.

BGH (Bundesgerichtshof, German Federal Court of Justice) (1999). Judgment of July 30, 1999, 1 StR 618/98, BGHSt 45, 164. In German. Full text: [hrr-strafrecht.de](https://hrr-strafrecht.de/hrr/1/98/1-618-98.pdf).

Bond, C. F. and DePaulo, B. M. (2006). Accuracy of Deception Judgments. Personality and Social Psychology Review, 10(3), 214-234. DOI 10.1207/s15327957pspr1003\_2.

Ceci, S. J. and Bruck, M. (1993). Suggestibility of the Child Witness: A Historical Review and Synthesis. Psychological Bulletin, 113(3), 403-439. DOI 10.1037/0033-2909.113.3.403.

Gabbert, F., Hope, L. and Fisher, R. P. (2009). Protecting Eyewitness Evidence: Examining the Efficacy of a Self-Administered Interview Tool. Law and Human Behavior, 33(4), 298-307. DOI 10.1007/s10979-008-9146-8. Checked against the accepted manuscript: [research.gold.ac.uk/id/eprint/18840](https://research.gold.ac.uk/id/eprint/18840/).

Geiselman, R. E., Fisher, R. P., Firstenberg, I., Hutton, L. A., Sullivan, S. J., Avetissian, I. V. and Prosk, A. L. (1984). Enhancement of Eyewitness Memory: An Empirical Evaluation of the Cognitive Interview. Journal of Police Science and Administration, 12, 74-80. Cited in Memon et al. 2010, p. 366.

Geiselman, R. E., Fisher, R. P., MacKinnon, D. P. and Holland, H. L. (1985). Eyewitness Memory Enhancement in the Police Interview: Cognitive Retrieval Mnemonics Versus Hypnosis. Journal of Applied Psychology, 70(2), 401-412. DOI 10.1037/0021-9010.70.2.401.

Holleman, B., Kamoen, N., Krouwel, A., van de Pol, J. and de Vreese, C. (2016). Positive vs. Negative: The Impact of Question Polarity in Voting Advice Applications. PLoS ONE, 11(10), e0164184. DOI [10.1371/journal.pone.0164184](https://doi.org/10.1371/journal.pone.0164184). Page number refers to the PDF version.

Jabine, T. B., Straf, M. L., Tanur, J. M. and Tourangeau, R. (Eds.) (1984). Cognitive Aspects of Survey Methodology: Building a Bridge Between Disciplines. Report of the Advanced Research Seminar on Cognitive Aspects of Survey Methodology. Washington, DC: National Academy Press. DOI 10.17226/930.

Kassin, S. M. and Kiechel, K. L. (1996). The Social Psychology of False Confessions: Compliance, Internalization, and Confabulation. Psychological Science, 7(3), 125-128. DOI 10.1111/j.1467-9280.1996.tb00344.x.

Lamb, M. E., Orbach, Y., Hershkowitz, I., Esplin, P. W. and Horowitz, D. (2007). A structured forensic interview protocol improves the quality and informativeness of investigative interviews with children: A review of research using the NICHD Investigative Interview Protocol. Child Abuse & Neglect, 31(11-12), 1201-1231. DOI 10.1016/j.chiabu.2007.03.021. Checked against the freely accessible manuscript version (PubMed Central, PMC2180422), which does not carry the journal's page numbers.

Loftus, E. F. and Marburger, W. (1983). Since the eruption of Mt. St. Helens, has anyone beaten you up? Improving the accuracy of retrospective reports with landmark events. Memory & Cognition, 11(2), 114-120. DOI 10.3758/BF03213465.

Loftus, E. F., Miller, D. G. and Burns, H. J. (1978). Semantic integration of verbal information into a visual memory. Journal of Experimental Psychology: Human Learning and Memory, 4(1), 19-31. DOI 10.1037/0278-7393.4.1.19.

Loftus, E. F. and Palmer, J. C. (1974). Reconstruction of automobile destruction: An example of the interaction between language and memory. Journal of Verbal Learning and Verbal Behavior, 13(5), 585-589. DOI 10.1016/S0022-5371(74)80011-3.

Malhotra, N. K. and Birks, D. F. (2007). Marketing Research. An Applied Approach (3rd European ed.). Harlow: Pearson Education.

Means, B., Nigam, A., Zarrow, M., Loftus, E. F. and Donaldson, M. S. (1989). Autobiographical Memory for Health-Related Events. Vital and Health Statistics, Series 6, No. 2. Hyattsville, MD: National Center for Health Statistics. [stacks.cdc.gov/view/cdc/6145](https://stacks.cdc.gov/view/cdc/6145).

Memon, A., Meissner, C. A. and Fraser, J. (2010). The Cognitive Interview: A Meta-Analytic Review and Study Space Analysis of the Past 25 Years. Psychology, Public Policy, and Law, 16(4), 340-372. DOI 10.1037/a0020518.

Milne, R., Griffiths, A., Clarke, C. and Dando, C. J. (2019). The cognitive interview: a tiered approach in the real world. In: Evidence-based Investigative Interviewing: Applying Cognitive Principles. London: Routledge. Page number refers to the accepted manuscript: [core.ac.uk/download/199180173.pdf](https://core.ac.uk/download/199180173.pdf).

Plous, S. (1993). The Psychology of Judgment and Decision Making. New York: McGraw-Hill. Chapter 6, pp. 64-76.

Rugg, D. (1941). Experiments in Wording Questions: II. Public Opinion Quarterly, 5(1), 91-92. DOI 10.1086/265467. Cited in Holleman et al. 2016, p. 3 (54% and 75% of those with an opinion) and Plous 1993, pp. 68-69 (46% and 62% of all respondents). The wording of the two versions is quoted as printed in Plous 1993, pp. 68-69. The year 1941 is the year of publication.

Schober, M. F. and Conrad, F. G. (1997). Does Conversational Interviewing Reduce Survey Measurement Error? Public Opinion Quarterly, 61(4), 576-602. DOI 10.1086/297818. Figures from the 1996 working paper of the Bureau of Labor Statistics: [bls.gov](https://www.bls.gov/osmr/research-papers/1996/pdf/st960090.pdf).

Zwaan, R. (2025). Memory, Misinformation, and the Need to Replicate. Blog post, September 8, 2025. [rolfzwaan.substack.com](https://rolfzwaan.substack.com/p/memory-misinformation-and-the-need).

Author: Dr. Paul Marx
Dr. Paul Marx is the founder and Managing Director of QUESTIONSTAR and was a professor of marketing at the University of Siegen.

*Author: Dr. Paul Marx — see [About](/about).*