Abstract
The “truth sandwich” correction format, in which false information is bookended by factual information, has frequently been presented as an optimal method for correcting misinformation. Despite recurring recommendations, there is little empirical evidence for enhanced benefits. In two pre-registered experiments (total N = 1046), we evaluated the effectiveness of the truth sandwich correction format against a “bottom-loaded” refutation format, in which the misinformation is first presented prior to factual statements. In Experiment 1, participants first rated belief in cancer misinformation. The misinformation was then corrected using the truth sandwich, corrected using a bottom-loaded refutation, or left uncorrected (control). Participants subsequently rerated their belief in the claims. Experiment 2 replicated and extended Experiment 1 by including a two-week test delay. We found that both correction formats were highly effective. However, there was no evidence that the truth sandwich format enhanced the effectiveness of corrections either immediately after reading or after a two-week delay period, with Bayesian analyses providing consistent evidence for a null effect of correction format. We repeated our analyses isolated to participants who endorsed complementary and alternative medicines, given this subgroup is particularly likely to believe cancer misinformation. We again found no evidence for any superiority of the truth sandwich correction format. These findings suggest that clear and detailed corrections can be powerfully effective against misinformation regardless of format, and advocacy for the truth sandwich correction above other simpler formats are unwarranted.
There are many ways to correct misinformation. One commonly cited method that has received popular attention and expert advocacy is the truth sandwich, which leads with a fact, followed by the misinformation, and ends with a secondary fact. The truth sandwich was first mentioned by George Lakoff during an episode of the CNN podcast “Reliable Sources” (2018). It has since been popularized by the media (for example, CNN, 2018; Illing, 2019; Peter Clark, 2020) and is often cited as a particularly beneficial method to correct misinformation (Lewandowsky et al., 2020). However, the truth sandwich correction format has not received substantial empirical examination. It thus seems premature to continue advocating for truth sandwiches as a powerful tool for reporters and information disseminators, given the need for empirical evaluation assessing the relative effectiveness of the truth sandwich as compared to other corrective formats.
The rationale behind the truth sandwich’s efficacy
The rationale behind the truth sandwich structure’s efficacy is based on the notion that by beginning with accurate information, any subsequent misinformation should be understood in the context of the factual frame (Sullivan, 2018).1 This strategy avoids foregrounding the inaccuracy and/or engaging a misinformed framing of the ideas, and may also better prepare people to tag the misinformation as false during encoding (Ecker et al., 2010). There is evidence that supports the need for care in any initial presentation. For example, when people form impressions of others the first trait presented has a considerable influence on judgments (Sullivan, 2018), and can prove resistant to refutation (Rapp & Kendeou, 2007, 2009). The truth sandwich also places an accuracy prior to and after an inaccurate message, exploiting the benefits of both primacy and recency effects. Information presented first has been shown to be encoded more strongly (Farrell & Lewandowsky, 2002) and to receive more rehearsal (Tan & Ward, 2008) than information following it. Recent information is also found to be readily available in memory and easily retrieved (Baddeley & Hitch, 1993). Therefore, based on results associated with framing, primacy, and recency effects, it seems reasonable to hypothesize that truth sandwiches might be particularly effective at reducing misunderstandings.
Lakoff (CNN, 2018) also invoked ironic process theory as a reason for why truth sandwiches are particularly effective. This theory explains why individuals find themselves thinking even more about content when attempting to suppress it, than if they are not trying to ignore the information (a classic example here is “don’t think of a white bear”; Wegner et al., 1987; Lakoff, 2014). Bookending misinformation with accurate information can presumably help prevent any rumination of inaccurate messages given the extended, repeated focus on correct information. However, it is unclear whether this theory is appropriate for considering how people process corrected misinformation. Theoretical accounts differ as to the exact way beliefs are updated (see Sanderson & Ecker, 2020 for a review), yet few place emphasis on suppression as the primary mechanism. This means that while ironic process theory could offer intriguing considerations with respect to the challenges of ignoring ideas, it may not be necessary to invoke suppression when accounting for the processes that drive the updating of previously encountered inaccuracies and conditions that support them.
Truth sandwich versus refutation structures
Attention to the truth sandwich format is perhaps most surprising given that there is a much larger and validated body of work demonstrating the benefits of refutation structures.
Refutations typically involve three components: presentation of the misinformation, identification of the misinformation as incorrect, and subsequent presentation of corrective information and explanations (i.e., the refutation; Tippett, 2010). The clearest difference between a refutation and the truth sandwich formats is that the misinformation is presented to participants first in a refutation, but second in a truth sandwich. Refutations have been shown to consistently support conceptual change (Schroeder & Kucera, 2022), which requires updating people’s mental models of how concepts interconnect (Chi et al., 1994; diSessa & Sherin, 1998; Vosniadou, 1994). Examinations of refutations have primarily focused on science education with benefits observed across biology, chemistry, and physics (Potvin et al., 2020; Aleknavičiūtė et al., 2023; Pacaci et al., 2024; but see Donovan et al., 2018).
Danielson et al. (2024)’s meta-analysis of misconception research is compelling evidence of refutation benefits. The authors selected 71 articles that described 76 studies and 294 effect sizes, finding a consistent and significant advantage of refutation texts over non-refutation texts. This benefit showed little variation in effects across different designs, groups of individuals, settings, topics, text features, or measures. One explanation for these benefits has been offered in Kendeou and O’Brien’s (2014) Knowledge Revision Components (KReC) framework. The authors argue that the reason for refutation benefits is that presenting misinformation immediately followed by a correction facilitates co-activation of the ideas and allows for the detection and resolution of potential discrepancies. As such, this theoretical framework allows for hypothesizing that the refutation format could actually be more effective than presenting facts prior to misinformation (i.e., in the truth sandwich format). There has been some empirical support for this in the literature: Vraga et al. (2020) found that climate misperceptions were reduced when participants read information offered in a misinformation-first approach (in which the fact was presented subsequently) than in a fact-first approach (in which the misinformation was presented subsequently).
Empirical examination of correction formats
In contrast to the large body of work on refutation structures and texts, few studies have explicitly examined the effects of truth sandwich exposures. König (2023) examined the efficacy of the truth sandwich for misinformation about nutrition and diet. They found that the format effectively reduced agreement with the misinformation in comparison to a control condition that offered no correction. However, the project did not compare outcomes to other correction formats. Kotz et al. (2023) conducted a comparative examination of the truth sandwich format and “bottom-loaded refutations” (i.e., misinformation followed by two facts), about COVID-19, vitamin C, and GMO foods. The authors reported no difference between the two corrective formats with respect to participants’ agreement with the misinformation immediately after reading the corrections. Finally, Tulin et al. (2024) compared the truth sandwich with a series of correction formats that all started with a false claim, and a no-correction control. They found that the truth sandwich did not reduce belief in the misinformation relative to a no-correction control. Indeed, only one of the four correction formats assessed significantly reduced belief in the false claim.2 It’s important to note that the study only tested a single claim (specifically, that “the order in which you eat a meal affects your weight”), and the truth sandwich and other corrections were not matched for content. Thus, it is unclear whether the lack of truth sandwich benefits was the result of the correction format, the correction content, or other factors unrelated to the correction itself (e.g., measurement error; Swire-Thompson et al., 2022). Nevertheless, the emerging findings have not been particularly promising or consistent, calling into question any additional benefit of the truth sandwich format as compared to other correction methods and structures (Anderson et al., 2019).
To date, studies have primarily focused on correction effectiveness immediately following correction exposure without examining effects following delays. This is important because belief updating following a correction is rarely sustained over time, a phenomenon called belief regression (Swire-Thompson et al., 2023). Belief regression occurs with all correction formats, including refutational texts (Zengilowski et al., 2021). If any enhanced effectiveness of the truth sandwich format is due to improved encoding of the facts relative to the misinformation in memory, it is possible that a comparative advantage of the format may only reveal itself over time. However, there is currently no evidence of truth sandwich superiority even after delays. Pillai et al. (2023) examined whether wording mattered for political fact checks, testing whether memory errors could be minimized by placing evaluations before a claim was stated (e.g., “No, X did not do Y, as A claims”), rather than after (e.g., “A claims X did Y. No, this is false”). They found that effects were equivalent regardless of order, even after a three-week period. Furthermore, Swire-Thompson et al. (2021) investigated whether the presentation order of misinformation and facts within corrections influenced belief updating, generally finding no difference whether misinformation was presented first (e.g. only older people need the flu vaccine - myth) or a fact was presented first (the flu vaccine is beneficial to all age groups etc.). Across four experiments there was only one instance in which presentation order made any difference, with a small effect size after a three-week retention interval. In this case, placing the misinformation first was more effective than the fact-first approach, running counter to what is expected if the truth sandwich were particularly effective. Converging findings suggest the particularities of a correction format may have minimal impact.
In sum, the lack of clear benefits of truth sandwiches over and above other formats runs counter to advice promoting truth sandwiches as a powerful tool for information disseminators. Investigations of truth sandwich benefits have not directly compared different correction structures, carefully controlled the included content, or attempted to replicate any effects. As such, there are a host of remaining questions and issues with respect to the potential benefits of truth sandwiches, including a comparison to other techniques for correcting misinformation such as refutation structures. This is important given that recent work has started to question whether an emphasis should be placed on particular correction formats over and above the utility of just providing corrections when possible and relevant.
The current study
The current study examined the relative effectiveness of the truth sandwich structure against a bottom-loaded refutation which states the inaccurate claim followed by factual statements. We chose to present participants with misinformation about cancer as it is a topic that could cause real-world harm (Swire-Thompson & Johnson, 2023). This topic also enabled us to repeat our analyses specifically with participants who endorse complementary and alternative medicines. People who endorse complementary and alternative medicines are more likely to (1) believe in the misinformation initially (Scherer et al., 2021), and (2) be more resistant to belief change once interventions are presented (Swire-Thompson et al., 2024). People could be more resistant to belief change if they regard the correction as not aligning their worldview (Ecker & Ang, 2019), or if they do not believe the correction comes from a credible source (Vraga & Bode, 2018). Focusing on individuals who endorsed complementary and alternative medicines allowed us to test whether the truth sandwich format might be especially useful for participants who would most benefit from corrections. Experiment 1 tested the effects of exposures to truth sandwich and bottom-loaded refutation structures immediately after reading the corrections, and Experiment 2 replicated and extended this test to include a two-week delay.
Experiment 1
In Experiment 1 we presented participants with cancer misinformation that was (a) corrected in a truth sandwich structure (i.e., fact, misinformation, fact), (b) corrected in a bottom-loaded refutation structure with the same number of corrections (i.e., misinformation, fact, fact), or (c) uncorrected (i.e., control). We measured participants’ beliefs in cancer claims both before and after reading corrections and also asked “inference questions”. Inference questions gain insight into whether people rely on the misinformation when reasoning. They are beneficial as people can continue to rely on misinformation when making inferences despite updating their belief as assessed with direct belief measures (see Ecker et al., 2011).
Methods
Design
Experiment 1 was a 2 × 3 within-subjects design with factors pre/post (pre-test vs. post-test) and correction condition (truth sandwich structure vs. bottom-loaded refutation structure vs no correction control). The primary dependent variables were (1) belief in false cancer claims and (2) claim-congruent inferential reasoning (measured only at the post-test timepoint).
Transparency and Openness
Experiment 1 was pre-registered at https://osf.io/twf4p. We predicted that belief in misinformation would reduce for both the truth sandwich and bottom-loaded correction formats in comparison to the uncorrected control. Given the lack of consistent and clear evidence for truth sandwich benefits as reported in the available literature, we predicted that the truth sandwich would be equivalently beneficial to the bottom-loaded refutation, with no additional advantage in its use. We also predicted an identical pattern of results as measured with the inference questions. Finally, we predicted that these patterns would replicate even with a subgroup of participants who endorsed complementary and alternative medicines. Please see Supplement A for whether pre-registered hypotheses were supported or rejected.
Participants
Participants were a representative sample (using sex, age, and race) of 518 people from the US over the age of 18 recruited from Prolific. Our pre-registered exclusion criteria omitted participants who completed the task too quickly (i.e., outliers with regards to completion time; N = 0), and participants who indicated that their data should be excluded due to lack of effort (N = 1). Our final sample included 517 people, with an age range between 19 and 86 years (Mage = 45.50, SDage = 16.15). The sample consisted of 245 males, 262 females, and ten individuals who were non-binary, preferred to self-describe their gender, or who chose not to disclose their gender. Complete demographic breakdowns for Experiments 1 and 2 are available in the Appendix.3 Participants were paid $3 for the 15 minute task.
Stimuli
Claims.
Twenty-four statements about cancer (12 true, 12 false) were used as the target stimuli. A pilot study was conducted to select these statements. One-hundred-and-fifteen participants from Prolific (16 males, 93 females, six individuals who were either non-binary or chose not to disclose their gender; Mage = 32.20, SDage = 12.19, age range = 18 – 72) rated 29 false cancer statements and 27 true cancer statements on believability, familiarity, and how much they had considered the item previously. To calculate the test-retest reliability of the items, 93 participants returned one week later to re-rate their belief in each item. Our goal was to select false items that were sufficiently believable to avoid floor effects, with true items having similar mean ratings to the false items. False items were selected with believability between 3.5 and 7.5 (M = 4.73) and true items between 2.5 and 7 (M = 5.12). True and false items did not significantly differ with respect to pilot participants’ belief ratings (p = .523). See Supplement B for item level information.
Corrections.
All corrections had three sections, which were simply re-ordered to make either a truth sandwich or bottom-loaded refutation. One section repeated the claim with a “false tag” (e.g., X causes cancer - this is false). Depending on condition, this section was either presented first (refutation) or second (truth sandwich). The other two sections contained relevant factual information that provided additional evidence against the target false claim but did not directly reference the claim. Two out of the three sections mentioned reputable sources (e.g., the National Cancer Institute), such that each truth sandwich or bottom-loaded refutation referenced two such sources. An example correction in the bottom-loaded refutation and truth sandwich format is presented in Table 1. A full list of corrections is available in Supplement C.
Table 1.
Example bottom-loaded refutation and truth sandwich for the claim “ingredients in red wine have been shown to prevent both colon and prostate cancer.” Note that the control condition did not provide any information to participants. Participants simply rated their belief in the claim then re-rated their belief in the claim at the end of the study.
| Bottom-loaded refutation | Truth sandwich |
|---|---|
| It has been purported that drinking red wine can prevent both colon and prostate cancer: this is false, according to the National Cancer Institute. The National Toxicology Program of the US Department of Health and Human Service advises that alcohol consumption can cause several types of cancer, including breast, throat, mouth, liver, and colon. The evidence indicates that the more alcohol a person drinks—particularly the more alcohol a person drinks regularly over time—the higher their risk of developing an alcohol-associated cancer. |
The National Toxicology Program of the US Department of Health and Human Service advises that alcohol consumption can cause several types of cancer, including breast, throat, mouth, liver, and colon. It has been purported that drinking red wine can prevent both colon and prostate cancer: this is false, according to the National Cancer Institute. The evidence indicates that the more alcohol a person drinks—particularly the more alcohol a person drinks regularly over time—the higher their risk of developing an alcohol-associated cancer. |
Affirmations.
We also presented “affirmations” to the factual information. These were structured in a similar manner to the corrections, although in this case all three sections provided support for the initial claim. In other words, each affirmation either presented the fact and associated “true tag” immediately, or in the middle (in line with the truth sandwich). It is important to note that affirmations to factual items were only included to allow for variation in participant experience, so that participants did not only receive false information with associated corrections. A full list of affirmations is available in Supplement C.
Outcome measures.
Belief was measured directly on an 11-point scale from 0 (“Definitely False”) to 10 (“Definitely True”). There were also 24 inference questions (one per item), included to assess participant reliance on the true and false information when reasoning. For example, the inference question for the false claim “Microwaving plastic can release cancer-causing toxins into food” was “How likely are you to avoid microwaving FDA approved plastic due to the cancer-causing toxins it releases?” (0, Extremely unlikely - 10, Extremely likely). See Supplement C for full list of inference questions.
Procedure
Participants first read a Northeastern University approved consent form (#19–04–90). They were then presented with the 24 statements about cancer (12 true and 12 false) in a randomized order, and were required to rate their belief in each statement. Following the presentation of all statements, participants were presented with corrections (for false items) and affirmations (for true items), again in a randomized order. All participants received four corrections using a truth sandwich format, four corrections using a bottom-loaded refutation, which left four false claims uncorrected as a control condition. Correction condition and statement pairing were counterbalanced in a Latin Square design across participants such that each claim was presented in all correction conditions (see Table 2). Participants were not explicitly told which items they got right or wrong, but were simply shown the corrections and affirmations one at a time in a randomized order. For the facts, affirmation type and statement pairing were similarly counterbalanced across participants in an identical fashion to the misinformation.
Table 2.
Presentation versions used across both Experiment 1 and 2. Versions (V1-V3) counterbalanced assignment of misinformation to correction conditions (bottom-loaded refutation, truth sandwich, and no correction control). Assignment of participants to the presentation version was randomized, with the constraint that approximately a third of participants were randomized to each version. Affirmations to facts were counterbalanced in an equivalent manner.
| V1 | V2 | V3 | |
|---|---|---|---|
|
Claim group 1
Misinformation 1–4 |
Bottom-loaded Refutation | No correction control | Truth Sandwich |
|
Claim group 2
Misinformation 5–8 |
Truth Sandwich | Bottom -loaded Refutation | No correction control |
|
Claim group 3
Misinformation 9–12 |
No correction control | Truth Sandwich | Bottom-loaded Refutation |
Following the presentation of all corrections and affirmations, participants were required to rerate their belief in all 24 statements, and to respond to 24 inferential reasoning questions. At the end of the study, participants completed demographic questions as well as the Complementary and Alternative Medicine Beliefs Inventory (CAMBI; Bishop et al., 2005). The CAMBI includes 17 items and three subscales: participation in treatment, natural treatment, and holistic health. Each subscale has acceptable reliability, with Cronbach’s alphas .68, .75, and .73, respectively (Grzywacz et al., 2012). Upon completion, participants were asked whether their data should be discarded due to lack of effort, and were provided with a debriefing that outlined corrective information for all of the claims that were presented.
Sample size justification
Our main analysis is a 2 × 3 within-subjects ANOVA with within subjects factors pre/post and correction condition (truth sandwich vs. bottom-loaded refutation vs. no correction). The smallest effect size from a previous similar study was ηp2 = .02 (f = .14; an interaction effect from Swire-Thompson et al., 2021), which indicated that refutations were more effective than were fact-first corrections over time. A power analysis conducted in G*Power3 (Faul et al., 2009) recommended a sample size of 168 participants (for effect size f = .14 with ɑ = .05, power = .95 and a moderate correlation between repeated measures, r =.50). We boosted this by three times (N = 500) given that we wished to conduct subgroup analyses on the top tertile of the complementary and alternative medicine scores.
Results
All data analyses and visualizations were performed in R version 4.3.2 (R core team, 2023). ANOVAs were conducted in afex (Singmann et al., 2024), follow-up contrasts were run using emmeans (Lenth, 2023), and Bayesian analyses were conducted using BayesFactor (Morey & Rouder, 2024). Figures were created using ggplot2 (Wickham, 2016) and cowplot (Wilke, 2020). Data, R script, and Rmarkdown files are available at https://osf.io/shf63/. We only report results of the misinformation items, given that fact items were presented only so that participants did not receive exclusively false items. For completeness, results for fact affirmations are available in Supplement E. Please note that all post-hoc t-tests in this manuscript are two-tailed and corrected for multiple comparisons, with Tukey’s adjusted p-value reported.
Belief in misinformation
To examine the differences in belief updating between conditions, we conducted a 2 × 3 within-subjects ANOVA with within subjects factors pre/post (pre vs. post-test) and correction condition (no-correction control vs. bottom-loaded refutation vs. truth sandwich). There was a pre/post by correction condition interaction, F(1.99, 1025.14) = 567.59; p < .001, MSE = 1.72, ηp2 = 0.52, showing that belief reduction depended on condition. Compared to the control, both the bottom–loaded refutation and truth sandwich reduced belief in the misinformation, t(516) = 28.59, p < .001; t(516) = 29.61, p < .001 respectively. However, as predicted, belief did not differ as a function of whether participants received bottom–loaded refutations or truth sandwiches (p = .546; see Figure 1, panel A). Rerunning the ANOVA with only the truth sandwich and bottom–loaded refutation (i.e. excluding the no–correction control) returned a non–significant effect of correction condition, F(1, 516) = 0.22, p = .643, MSE = 2.23, ηp2 < .01, and non–significant pre/post × condition interaction; F(1, 516) = 2.78, p = .096, MSE = 1.57, ηp2 < .01.
Figure 1.

(a) Belief in misinformation across correction conditions pre- and post-test (b) inference ratings across conditions. All error bars represent 95% confidence intervals; Violins and jitter provide distributional information.
Given that we hypothesized a null, we additionally assessed correction condition effectiveness using a 2 × 2 within-subjects Bayesian ANOVA as Bayes factors (BFs) can quantify relative evidence favoring the null hypothesis (BF01). A BF between 1 and 3 is anecdotal evidence, 3–10 is moderate evidence, 10–30 is strong evidence, 30–100 is very strong evidence, and a BF >100 constitutes extreme evidence (Wagenmakers et al., 2018). This analysis returned strong evidence for a null effect of correction condition (BF01 = 19.68), and moderate evidence for the null for the pre/post × correction condition interaction (BF01 = 6.76). Furthermore, a paired sample t-test with the hypothesis that the truth sandwich would have lower belief (than the bottom-loaded refutation) showed very strong support for the null (BF0- = 52.80).
We next repeated the above analysis specifically with the subset of participants who endorsed complementary and alternative medicines. We excluded the holistic treatment subscale of the CAMBI as it does not correlate with pre-test misinformation belief (in line with prior research, see Swire-Thompson et al., 2024). Our new composite only included the natural health and holistic health subscales, which we call the CAMBI-NH. When analyses were isolated to only participants who scored in the top tertile of the CAMBI-NH, all findings replicated, indicating benefits of both correction formats but no differential effects (see Supplement D for full details).
Inference questions
To assess if reliance on misinformation varied across correction conditions for inference questions, we conducted a one-way ANOVA. There was a significant main effect of condition, F(1.93, 997.37) = 216.89; p < .001, MSE = 2.73, ηp2 = .30, with inference ratings in both the refutation and truth sandwich conditions lower than in the control, t(516) = 16.65, p < .001; t(516) = 17.84, p < .001 respectively. As predicted, there was no significant difference between the refutation and truth sandwich correction conditions (p = .556), see Figure 1 panel B. This was confirmed with a Bayesian paired sample t–test, with the hypothesis that the truth sandwich is more effective than the refutation showing very strong evidence for the null (BF0+ = 39.20). The observed pattern of results was again identical when analyses were isolated to only participants in the top tertile of CAMBI-NH scores (see Supplement D).
In sum, corrections were beneficial in Experiment 1, reducing beliefs in and inferences related to false claims, as compared to a no correction condition. However, we found no evidence that the truth sandwich format enhanced the effectiveness of corrections relative to a bottom-loaded refutation, with both correction formats being equally effective. This was true for belief in misinformation, inference scores, and even with participants who specifically endorsed complementary and alternative medicines. Given similarity in the results obtained across the two experiments, we defer our discussion of the findings until after presentation of Experiment 2.
Experiment 2
Experiment 2 sought to replicate Experiment 1 and also included a two week retention interval. A delay between the correction and post-test was included because comparative advantages of the truth sandwich may only reveal itself over time (Swire-Thompson et al., 2021; Zengilowski et al., 2021).
Methods
Design
Experiment 2 was a 2 × 2 × 3 between-within subjects design with between subjects factors retention group (immediate vs. delayed) and within subjects factors of pre/post (pre-test vs. post-test) and correction condition (truth sandwich vs. bottom-loaded refutation vs. no correction). The primary dependent variables were belief in false cancer claims, and claim-congruent inferential reasoning (measured only at the post-test timepoint).
Transparency and Openness
Experiment 2 was pre-registered at https://osf.io/c954z. As in Experiment 1, we predicted that (a) both correction formats would reduce belief as compared to the control, and (b) the truth sandwich would be beneficially equivalent to the bottom-loaded refutation. We further predicted that the above would be true in both immediate and delayed conditions, based largely on findings that have focused specifically on immediate effects (Pillai et al., 2023; Swire-Thompson et al., 2021), and even with participants who endorsed complementary and alternative medicines (i.e. scored high on the CAMBI-NH). Finally, we predicted that there would also be evidence of belief regression, whereby corrections would be less effective in the delayed than the immediate condition for both correction formats. Please see Supplement A for an overview on whether pre-registered hypotheses were supported or rejected.
Participants
We recruited a representative sample of 529 US participants over the age of 18 from Prolific. Our exclusion criteria (i.e., participants identified as completing the task too quickly, and participants who indicated their data should be excluded due to low effort) did not result in omitting any participants. Thus, our final sample included 529 people ranging in age from 18 to 85 (Mage = 46.27, SDage = 15.42). The sample consisted of 250 males, 274 females, and 5 individuals who preferred to self-describe or chose not to disclose their gender. Participants in the immediate condition were paid $3 for a 15 minute task, and participants in the delayed condition were paid an additional $2 for completing the two-week delayed test.
Procedure and Stimuli
The procedure and stimuli were identical to Experiment 1 with the additional inclusion of a delayed group. For the delayed group, the procedure and stimuli were identical to the immediate group except that a two-week delay was introduced between reading the claims, corrections, and affirmations, and the post-test.
Sample size justification
Our main analysis was a 2 × 2 × 3 between-within ANOVA with between subjects factors retention group (immediate vs. delayed), and within subjects factors of pre/post (pre vs. post correction) and correction condition (truth sandwich vs. bottom-loaded refutation vs. no correction) on the belief in misinformation. The smallest effect size from a previous similar study was ηp2 = .02 (Swire-Thompson et al., 2022), which showed that refutations were more effective than were fact-first corrections after a delay. A power analysis conducted in G*Power3 (Faul et al., 2009) recommended a sample size of 86 participants (for effect size f = .14 with ɑ = .05, power = .95 and a moderate correlation between repeated measures, r =.50). However, we boosted this by three times (N = 258) given that we wished to conduct subgroup analyses on the top tertile of the CAMBI scores. Given we were also predicting a null finding, we doubled this (N = 516) to boost the sensitivity of the study, and to help account for expected attrition in the delayed group.
Results
Belief in misinformation
To assess the difference in belief updating across conditions and retention groups, we first conducted a 2 × 2 × 3 between-within-subjects ANOVA with between subjects factor retention group (immediate vs. delayed) and within-subjects factors pre/post (pre vs. post-test) and correction condition (truth sandwich vs. bottom-loaded refutation vs. no correction control); see Figure 2 for descriptives. There was a significant three-way interaction, F(1.98, 1042.35) = 32.59, p < .001, MSE = 1.65, ηp2 = .06, indicating that the impact of the correction condition on belief varied across retention groups. Belief in corrected misinformation was significantly lower in the immediate condition than in the delayed condition (reflecting belief regression) for both the refutation and truth sandwich conditions, t(527) = −8.89, p < .001; t(527) = −10.39, p < .001 respectively. Despite reduced efficacy of the corrections over time, both the refutation and truth sandwich conditions nevertheless led to beneficial reductions in false-claim beliefs relative to the control condition both in the immediate condition, t(527) = 21.81, p < .001; t(527) = 21.39, p < .001, and after the two-week delay, t(527) = 11.24, p < .001; t(527) = 9.57, p < .001. As predicted, there was no significant difference in post-correction belief across the refutation and truth sandwich conditions either immediately or after a two-week delay (ps ≥ .206).
Figure 2.

(a) Belief in misinformation across correction conditions and pre- and post-test (b) Inference ratings across correction conditions. Left panels show the immediate time points and right panels show results after a 2-week delay. Error bars represent 95% confidence intervals; Violins and jitter provide distributional information.
Analyses excluding the control condition (i.e., a 2 × 2 × 2 between-within subjects ANOVA) confirmed there was no main effect of correction condition (bottom-loaded refutation vs. truth sandwich; p = .669), correction condition × pre/post-test interaction (p = .445), or three-way interaction (p = .991). Furthermore, conducting a 2 × 2 × 2 analysis as a Bayesian ANOVA provided strong evidence for a null effect of correction condition (BF01 = 20.64), and moderate evidence for the null for the three-way interaction (BF01 = 9.56). As in Experiment 1, the observed pattern of results was also equivalent when analyses were isolated specifically to participants who endorsed complementary and alternative medicines (i.e., participants in the top tertile of CAMBI-NH scores; see Supplement D).
Inference questions
To examine the impact of correction formats on the inference ratings, we next conducted a 2 × 3 between-within subjects ANOVA; see Figure 2 panel B for mean inference ratings. There was a significant correction condition by retention group interaction, F(1.95, 1029.84) = 14.57, p < .001, MSE = 2.67, ηp2 = .03, showing that the efficacy of the corrections depended upon the retention interval. As expected, both the bottom-loaded and truth sandwich corrections were less effective in the delayed than in the immediate condition, t(527) = −5.85, p < .001; t(527) = −5.08, p < .001. Despite the corrections being less effective after the delay, inference ratings were significantly lower than the control in both correction conditions immediately, t(527) = 11.90, p < .001; t(527) = 9.34, p < .001, and following the two-week delay, t(527) = 4.02, p < .001; t(527) = 3.08, p = .026. There were again no significant differences between the two correction conditions for either retrieval interval (ps ≥ .760). The ANOVA was repeated excluding the control, and returned a non-significant main effect (p = .190) and non-significant interaction (p = .604). This was confirmed with a Bayesian ANOVA which provided strong evidence for no difference in inference ratings between correction conditions (BF01 = 20.44).
We repeated analyses focusing specifically on participants who scored in the top tertile of CAMBI-NH and found a comparable pattern of results for the primary research question (i.e., no evidence of a differential effectiveness of the bottom-loaded and truth sandwich correction formats). Furthermore, compared to the control, both the bottom-loaded refutation and truth sandwich conditions were associated with reduced reliance on misinformation in the immediate post-test, t(187) = 8.04, p < .001; t(187) = 5.06, p < .001 respectively. However, we note results did diverge from the complete dataset in one instance: There were no differences between the control and correction conditions after the two-week delay (ps ≥ .752; see Supplement D).
General Discussion
This study aimed to empirically evaluate the efficacy of the truth sandwich correction format in comparison to a bottom-loaded refutation. Experiments 1 and 2 consistently showed that corrections were beneficial in general, leading participants to show reduced misinformation belief and inferences related to the misinformation. However, the truth sandwich format was not superior to the bottom-loaded refutation. This finding was true regardless of whether participants responded immediately after reading the corrections or after a two-week delay. It was also obtained for participants who specifically endorsed complementary and alternative medicines. Bayesian analyses consistently provided evidence in favor of a null effect, with both formats associated with equivalent belief updating. Indeed, we found very strong evidence against the hypothesis that the truth sandwich is more effective than the bottom-loaded refutation at reducing misinformation belief. While truth sandwiches have been routinely recommended by the media and some researchers (CNN, 2018; Illing, 2019; Lewandowsky et al., 2020; Peter Clark, 2020), the current findings suggest that the use of the truth sandwich format does not improve correction effectiveness over and above corrections that lead with misinformation.
The current results are consistent with both König (2023) who found that the truth sandwich format reduced belief in false information relative to a control, and Kotz et al. (2023) who found no difference between the truth sandwich format and bottom-loaded refutations when tested immediately after the corrections were presented. These findings suggest corrections can be effective regardless of which frame is activated first (i.e., misinformation or facts), and that being informed that misinformation is false is likely different than the suppression of undesirable thoughts (Wegner et al., 1987). The results contrast with those of Tulin et al. (2024) who found no difference between the truth sandwich and control. Reasons for this discrepancy could include that their corrections were overall not very effective in any format (i.e., only one of the four tested correction formats actually resulted in significantly reduced belief), each correction format had different content, or other factors unrelated to the corrections themselves such as measurement error.
We have focused on the lack of additional benefits for the truth sandwiches, but it is also worth considering that the bottom-loaded refutation was no more effective than the truth sandwich. This may be surprising given a substantial body of work demonstrating the efficacy of refutational texts. One possible reason for the lack of additional benefits for the bottom-loaded refutation is that both the bottom-loaded refutations and the truth sandwich always presented misinformation prior to at least one piece of factual information. In other words, both formats utilized best practices for helping to ensure co-activation would occur (Kendeou & O’Brien, 2014), and it may still be better to present the misinformation immediately prior to the fact (see Ecker et al., 2017). Nonetheless, the current findings contribute to a growing body of literature suggesting that corrections containing the necessary components are highly useful for fostering belief change, irrespective of their format (Swire-Thompson et al., 2021, 2023; Pillai et al., 2023).
Although the current study provides clear evidence against advocations for particular correction structures, we note several limitations and avenues for future work. Firstly, the current study focused on short format, health-based misinformation, and future research may benefit from replicating the current findings with other misinformation content, correction types, or contexts. Correction structures can appear in a variety of forms, including in single texts (as in the current project), across multiple statements from different authors as in social media (e.g., Mason & Rapp, 2025), or as spoken descriptions. It is important to examine correction formats under these different circumstances. Future directions should also include conducting correction format studies in more naturalistic settings such as in social media contexts and in fact-checking domains. Only one study, to our knowledge, has measured the use of the truth sandwich in the real world. Kenix and Manickam (2021) examined corrections to inaccurate claims made by President Trump in four newspapers published between 2017 and 2018, and found that only one article (0.2%) used the truth sandwich format. Of course, further studies are necessary to determine whether this is indeed true across news dissemination outlets or other real world settings. Should real world analyses of correction format be a challenge due to ethical limitations or due to media companies being resistant to academic collaborations, external validity can be improved by using social media simulators (see Butler et al., 2024).
This study has clear practical applications for education, fact checkers, health practitioners, journalists, science writers, and policy makers. The results indicate that it may not matter what order corrective information is placed in as long as it is included (and attended to). Again, there is some evidence that the truth sandwich has not been taken up by the media (Kenix & Manickam, 2021), and our study suggests that this may not be a missed opportunity. One reason for the lack of uptake is that truth sandwiches can often be challenging to construct, which can lead to lengthy or convoluted corrections that are difficult to read or interpret. It can be extremely valuable for fact-checkers and practitioners to know what not to spend time, resources, and effort focusing on (Drobnic Holan, 2023), especially if particular correction formats (e.g., refutation structures) might be easier to prepare and more straightforward to present. Our study is also important because recommendations touting the efficacy of the truth sandwich by reputable sources and anecdotal instances where the format is used remains high. Most recently, in October 2024, there were claims about the truth sandwich’s efficacy in New Scientist magazine (New Scientist, 2024; Barras, 2024), and the format appears to have been used by CBS during the recent 2024 vice-presidential debate to correct climate misinformation (Fisher, 2024).
In sum, the current project fills a critical research gap, showing that while the truth sandwich format has been promoted as a powerful tool, its actual impact on reducing misinformation is no greater than that of simpler refutation formats. Making corrections clear and coherent is likely far more important than ensuring they buttress or bookend misinformation in any particular way. To assist fact-checkers and communicators in the future, researchers should strive for robust evidence prior to making firm recommendations regarding best practices for misinformation correction.
Supplementary Material
Educational Impact And Implications Statement.
This study examined whether the recommended “truth sandwich” correction format (fact, misinformation, fact) is more effective than a standard refutation (misinformation, fact, fact) at reducing belief in misinformation. This is a critical and timely issue, as the truth sandwich correction format is often promoted by theorists, the media, and educational resources as an optimal method for correcting misinformation. To date, these recommendations have not necessarily been based on empirical evidence. We found that the truth sandwich correction is no better than a standard refutation correction format, both over time and in people who are particularly likely to believe the initial misinformation. These findings suggest that advocating for one format over another is not necessary; rather, clear and detailed corrections can be extremely effective against misinformation in a variety of formats. These findings also highlight the importance of ensuring recommendations offered to educators, policymakers, and fact-checkers are evidence-based.
Funding:
This work was funded by the National Institutes of Health National Cancer Institute, awarded to Briony Swire-Thompson “Cancer misinformation on social media and its correction” (Grant R00CA248720).
Appendix
Table A1.
Race and Gender Demographics for Experiment 1
| Race | Female | Male | Other/Undisclosed | Total |
|---|---|---|---|---|
| Asian | 14 | 14 | 3 | 31 |
| Black or African American | 36 | 24 | 0 | 60 |
| White | 203 | 196 | 7 | 406 |
| More than One Race | 5 | 6 | 0 | 11 |
| Unknown or Not Reported/Other | 4 | 5 | 0 | 9 |
| Total | 262 | 245 | 10 | 517 |
Table A2.
Race and Gender Demographics for Experiment 2
| Race | Immediate | Delayed | Total | ||||
|---|---|---|---|---|---|---|---|
| Female | Male | Other/Undisclosed | Female | Male | Other/Undisclosed | ||
| Asian | 12 | 5 | 0 | 7 | 8 | 0 | 32 |
| Black or African American | 24 | 16 | 0 | 16 | 18 | 0 | 74 |
| White | 115 | 99 | 2 | 92 | 94 | 2 | 404 |
| More than One Race | 3 | 3 | 0 | 1 | 3 | 0 | 10 |
| Unknown or Not Reported/Other | 2 | 2 | 1 | 2 | 2 | 0 | 9 |
| Total | 156 | 125 | 3 | 118 | 125 | 2 | 529 |
Footnotes
Conflicts of interest: All authors declare no conflicts of interest.
Ethics: This research received approval from Northeastern University (#19–04-90). BST and DNR designed the study, BST recruited participants, BST and LB conducted the analyses, LB created the figures, BST drafted the manuscript, and all authors provided critical revisions. All authors have approved this manuscript.
It is important to note that “sandwiching” as a communication practice is older than the truth sandwich’s appearance in 2018. For instance, the feedback sandwich, in which a corrective statement about a behavior is sandwiched between two positive statements, has long been touted as the best way to provide critique (James & Shephard, 2001). However, the efficacy of this strategy has also been associated with mixed results (Henley & DiGennaro Reed, 2015; Procházka et al., 2020).
The effective correction format presented the misinformation as “mostly false” (i.e., “The order in which you eat a meal has at most an indirect effect on your weight”) rather than “false” (i.e., “The order in which you eat a meal has no effect on your weight”), without including a label that explicitly mentioned a clear verdict. In this study, belief was a composite of perceived credibility, accuracy, reliability, bias, and completeness of the false health claim.
Note that participants were asked “what is your gender” and presented with the options male, female, prefer to self describe, and prefer not to say, aligning with NIH recommendations at the time. We note this does not conform to APA guidelines that recommend the terms “man” and “women”. Additionally, we did not ask about the race or ethnicity of the participants beyond the Prolific representativeness survey.
References
- Aleknavičiūtė V, Lehtinen E, & Södervik I (2023). Thirty years of conceptual change research in biology–A review and meta-analysis of intervention studies. Educational Research Review, 100556. [Google Scholar]
- Anderson ER, Horton W, & Rapp D (2019). Hungry for the truth: Evaluating the utility of “truth sandwiches” as refutations [Conference presentation]. Annual meeting of the society for text & discourse. [Google Scholar]
- Baddeley AD, & Hitch G (1993). The recency effect: Implicit learning with explicit retrieval?. Memory & Cognition, 21, 146–155. 10.3758/BF03202726 [DOI] [PubMed] [Google Scholar]
- Barras C (2024) The archaeologist fighting claims about an advanced lost civilisation. New Scientist. https://www.newscientist.com/article/mg26435130-400-the-archaeologist-fighting-claims-about-an-advanced-lost-civilisation/ [Google Scholar]
- Bishop FL, Yardley L, & Lewith G (2005). Developing a measure of treatment beliefs: The complementary and alternative medicine beliefs inventory. Complementary Therapies in Medicine, 13(2), 144–149. 10.1016/j.ctim.2005.01.005 [DOI] [PubMed] [Google Scholar]
- Butler LH, Lamont P, Wan DLY, Prike T, Nasim M, Walker B, Fay N & Ecker UKH (2024). The (Mis) Information Game: A social media simulator. Behavior Research Methods, 1–22. 10.3758/s13428-023-02153-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- Chi MTH, Slotta JD, & De Leeuw N (1994). From things to processes: A theory of conceptual change for learning science concepts. Learning and Instruction, 4, 27–43. 10.1016/0959-4752(94)90017-5 [DOI] [Google Scholar]
- CNN. (2018, June 17). How to make a ‘truth sandwich’. CNN. https://www.cnn.com/videos/tv/2018/06/17/how-to-make-a-truth-sandwich-rs.cnn [Google Scholar]
- Danielson RW, Jacobson NG, Patall EA, Sinatra GM, Adesope OO, Kennedy AA, … & Sunday OJ (2024). The effectiveness of refutation text in confronting scientific misconceptions: A meta-analysis. Educational Psychologist, 1–25. 10.1080/00461520.2024.2365628 [DOI] [Google Scholar]
- diSessa AA, & Sherin BL (1998). What changes in conceptual change? International Journal of Science Education, 20, 1155–1191. 10.1080/0950069980201002 [DOI] [Google Scholar]
- Donovan AM, Zhan J, & Rapp DN (2018). Supporting historical understandings with refutation texts. Contemporary Educational Psychology, 54, 1–11. 10.1016/j.cedpsych.2018.04.002 [DOI] [Google Scholar]
- Drobnic Holan A (2023). Fact-checkers applaud academic research on fact-checking but offer wide range of opinions on which research is most valuable. Poynter. https://www.poynter.org/fact-checking/2023/fact-checkers-applaud-academic-research-on-fact-checking-but-offer-wide-range-of-opinions-on-which-research-is-most-valuable/ [Google Scholar]
- Ecker UK, & Ang LC (2019). Political attitudes and the processing of misinformation corrections. Political Psychology, 40(2), 241–260. [Google Scholar]
- Ecker UKH, Hogan JL, & Lewandowsky S (2017). Reminders and repetition of misinformation: Helping or hindering its retraction?. Journal of applied research in memory and cognition, 6(2), 185–192. 10.1016/j.jarmac.2017.01.014 [DOI] [Google Scholar]
- Ecker UKH, Lewandowsky S, Swire B, & Chang D (2011). Correcting false information in memory: Manipulating the strength of misinformation encoding and its retraction. Psychonomic bulletin & review, 18, 570–578. 10.3758/s13423-011-0065-1 [DOI] [PubMed] [Google Scholar]
- Ecker UKH, Lewandowsky S, & Tang DT (2010). Explicit warnings reduce but do not eliminate the continued influence of misinformation. Memory & cognition, 38, 1087–1100. 10.3758/MC.38.8.1087 [DOI] [PubMed] [Google Scholar]
- Farrell S, & Lewandowsky S (2002). An endogenous distributed model of ordering in serial recall. Psychonomic bulletin & review, 9(1), 59–79. 10.3758/BF03196257 [DOI] [PubMed] [Google Scholar]
- Faul F, Erdfelder E, Buchner A, & Lang A-G (2009). Statistical power analyses using G*Power 3.1: Tests for correlation and regression analyses. Behavior Research Methods, 41(4), 1149–1160. 10.3758/BRM.41.4.1149 [DOI] [PubMed] [Google Scholar]
- Fisher (2024). Three ways the CBS vice presidential debate’s climate segment stood out. Media Matters for America. https://www.mediamatters.org/cbs/three-ways-cbs-vice-presidential-debates-climate-segment-stood-out [Google Scholar]
- Grzywacz JG, Neiberg R, Quandt SA, Lang W, Bell RA, & Arcury TA (2012). Measuring differential beliefs in complementary therapy research: An exploration of the Complementary and Alternative Medicine Beliefs Inventory (CAMBI). Complementary therapies in medicine, 20(1–2), 54–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Henley AJ, & DiGennaro Reed FD (2015). Should you order the feedback sandwich? Efficacy of feedback sequence and timing. Journal of Organizational Behavior Management, 35(3–4), 321–335. [Google Scholar]
- Illing S (2019) How the media should respond to Trump’s lies: State of the Union edition, Vox, available from https://www.vox.com/2018/11/15/18047360/trump-state-of-the-union-speech-2019-george-lakoff
- James T, & Shephard D (2001). Presenting Magically. Crown House Publishing. [Google Scholar]
- Kendeou P, & O’Brien EJ (2014). The knowledge revision components (KReC) framework: Processes and mechanisms. In Rapp D & Braasch J (Eds.), Processing inaccurate information (pp. 353–378). MIT Press. [Google Scholar]
- Kenix LJ, & Manickam J (2021). A Missed Opportunity? President Trump, the Truth Sandwich, and News Coverage Across an Ideological Spectrum. Media Watch, 12(2), 177–196. 10.15655/mw_2021_v12i2_160145 [DOI] [Google Scholar]
- König LM (2023). Debunking nutrition myths: An experimental test of the ‘truth sandwich’ text format. British journal of health psychology, 28(4), 1000–1010. 10.1111/bjhp.12665 [DOI] [PubMed] [Google Scholar]
- Kotz J, Giese H, & König LM (2023). How to debunk misinformation? An experimental online study investigating text structures and headline formats. British Journal of Health Psychology, 28(4), 1097–1112. 10.1111/bjhp.12670 [DOI] [PubMed] [Google Scholar]
- Lakoff G (2014). The all new don’t think of an elephant!: Know your values and frame the debate. Chelsea Green Publishing. [Google Scholar]
- Lenth R (2023). emmeans: Estimated Marginal Means, aka Least-Squares Means (R package version 1.8.9). Retrieved from https://CRAN.R-project.org/package=emmeans.
- Lewandowsky S, Cook J, Ecker UKH, Albarracín D, Amazeen MA, Kendeou P … Zaragoza MS (2020). The Debunking Handbook 2020. 10.17910/b7.1182 [DOI] [Google Scholar]
- Mason CE, & Rapp DN (2025). Can corrective social media replies lead to refutational benefits? Refutation text structure grounded in discourse interactions. [Manuscript submitted for publication]
- Morey R & Rouder J (2024). BayesFactor: Computation of Bayes Factors for Common Designs (R package version 0.9.12-4.7). Retrieved from https://CRAN.R-project.org/package=BayesFactor>.
- New Scientist (2024). We’re homing in on the best ways to tackle misinformation https://www.newscientist.com/article/mg26435132-500-were-homing-in-on-the-best-ways-to-tackle-misinformation/
- Pacaci C, Ustun U, & Ozdemir OF (2024). Effectiveness of conceptual change strategies in science education: A meta-analysis. Journal of Research in Science Teaching, 61(6), 1263–1325. [Google Scholar]
- Peter Clark R (2020, August 18). How to serve up a tasty ‘truth sandwich?’. Poynter. from https://www.poynter.org/reporting-editing/2020/how-to-serve-up-a-tasty-truth-sandwich/ [Google Scholar]
- Pillai RM, Brown-Schmidt S, & Fazio LK (2023). Does wording matter? Examining the effect of phrasing on memory for negated political fact checks. Journal of Applied Research in Memory and Cognition, 12(1), 48–58. 10.1037/mac0000022 [DOI] [Google Scholar]
- Potvin P, Nenciovici L, Malenfant-Robichaud G, Thibault F, Sy O, Mahhou MA, … & Chastenay P (2020). Models of conceptual change in science learning: Establishing an exhaustive inventory based on support given by articles published in major journals. Studies in Science Education, 56(2), 157–211. [Google Scholar]
- Procházka J, Ovcari M, & Durinik M (2020). Sandwich feedback: The empirical evidence of its effectiveness. Learning and Motivation, 71, 101649. [Google Scholar]
- R Core Team. (2023). R: A language and environment for statistical computing. https://www.R-project.org/.
- Rapp DN, & Kendeou P (2007). Revising what readers know: Updating text representations during narrative comprehension. Memory & cognition, 35(8), 2019–2032. 10.3758/BF03192934 [DOI] [PubMed] [Google Scholar]
- Rapp DN, & Kendeou P (2009). Noticing and revising discrepancies as texts unfold. Discourse Processes, 46(1), 1–24. 10.1080/01638530802629141 [DOI] [Google Scholar]
- Sanderson JA, & Ecker UKH (2020). The challenge of misinformation and ways to reduce its impact. In Handbook of learning from multiple representations and perspectives. Routledge. [Google Scholar]
- Scherer LD, McPhetres J, Pennycook G, Kempe A, Allen LA, Knoepke CE, Tate CE, & Matlock DD (2021). Who is susceptible to online health misinformation? A test of four psychosocial hypotheses. Health Psychology, 40, 274–284. 10.1037/hea0000978 [DOI] [PubMed] [Google Scholar]
- Schroeder NL, & Kucera AC (2022). Refutation text facilitates learning: A meta-analysis of between-subjects experiments. Educational Psychology Review, 34(2), 957–987. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Singmann H, Bolker B, Westfall J, Aust F & Ben-Shachar M (2024). afex: Analysis of Factorial Experiments (R package version 1.3-1). Retrieved from https://CRAN.R-project.org/package=afex.
- Sullivan M (2018, June 17). Perspective | Instead of Trump’s propaganda, how about a nice “truth sandwich”?. Washington Post. https://www.washingtonpost.com/lifestyle/style/instead-of-trumps-propaganda-how-about-a-nice-truth-sandwich/2018/06/15/80df8c36-70af-11e8-bf86-a2351b5ece99_story.html [Google Scholar]
- Swire-Thompson B, Butler LH & Rapp D (2024, November 19). Efficacy of the “truth sandwich” in the correction of misinformation. Retrieved from https://osf.io/shf63/ [DOI] [PMC free article] [PubMed]
- Swire-Thompson B, Cook J, Butler LH, Sanderson JA, Lewandowsky S, & Ecker UK (2021). Correction format has a limited role when debunking misinformation. Cognitive Research: Principles and Implications, 6(1), 83. 10.1186/s41235-021-00346-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Swire-Thompson B, Dobbs M, Thomas A, & DeGutis J (2023). Memory failure predicts belief regression after the correction of misinformation. Cognition, 230, 105276. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Swire-Thompson B, & Johnson S (2023). Cancer: A model topic for misinformation researchers. Current Opinion in Psychology, 101775. 10.1016/j.copsyc.2023.101775 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Swire-Thompson B & Rapp D (2023, September 16). Experiment 2: The efficacy of the “truth sandwich” in the correction of misinformation over time. https://osf.io/twf4p/ [DOI] [PMC free article] [PubMed]
- Swire-Thompson B & Rapp D (2023, May 17). The efficacy of the “truth sandwich” in the correction of misinformation. https://osf.io/c954z/ [DOI] [PMC free article] [PubMed]
- Swire-Thompson B, Miklaucic N, Wihbey JP, Lazer D, & DeGutis J (2022). The backfire effect after correcting misinformation is strongly associated with reliability. Journal of Experimental Psychology: General, 151(7), 1655. 10.1037/xge0001131 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Tan L, & Ward G (2008). Rehearsal in immediate serial recall. Psychonomic bulletin & review, 15(3), 535–542. 10.3758/pbr.15.3.535 [DOI] [PubMed] [Google Scholar]
- Tippett CD (2010). Refutation text in science education: A review of two decades of research. International journal of science and mathematics education, 8, 951–970. [Google Scholar]
- Tulin M, Hameleers M, de Vreese C, Opgenhaffen M, & Wouters F (2024). Beyond belief correction: Effects of the truth sandwich on perceptions of fact-checkers and verification intentions. Journalism Practice, 1–20. 10.1080/17512786.2024.2311311 [DOI] [Google Scholar]
- Vosniadou S (1994). Capturing and modeling the process of conceptual change. Learning and Instruction, 4, 45–69. 10.1016/0959-4752(94)90018-3 [DOI] [Google Scholar]
- Vraga EK, Kim SC, Cook J, & Bode L (2020). Testing the effectiveness of correction placement and type on Instagram. The International Journal of Press/Politics, 25(4), 632–652. 10.1177/1940161220919082 [DOI] [Google Scholar]
- Vraga EK, & Bode L (2018). I do not believe you: How providing a source corrects health misperceptions across social media platforms. Information, Communication & Society, 21(10), 1337–1353. [Google Scholar]
- Wagenmakers E-J, Love J, Marsman M, Jamil T, Ly A, Verhagen J, et al. (2018). Bayesian inference for psychology. Part II: Example applications with JASP. Psychonomic Bulletin & Review, 25, 58–76. 10.3758/s13423-017-1323-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wegner DM, Schneider DJ, Carter SR, & White TL (1987). Paradoxical effects of thought suppression. Journal of personality and social psychology, 53(1), 5. 10.1037//0022-3514.53.1.5 [DOI] [PubMed] [Google Scholar]
- Wickham H (2016). ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag; New York. [Google Scholar]
- Wilke C (2020). cowplot: Streamlined plot theme and plot annotations for ‘ggplot2’ (R package version 1.1.1). Retrieved from https://CRAN.R-project.org/package=cowplot
- Zengilowski A, Schuetze BA, Nash BL, & Schallert DL (2021). A critical review of the refutation text literature: Methodological confounds, theoretical problems, and possible solutions. Educational Psychologist, 56(3), 175–195. 10.1080/00461520.2020.1861948. [DOI] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
