To address these limitations, we collected data from the general public both before and after July 5, 2025. Our study’s first aim is to identify who believed the prophecy—that is, which individual-difference psychological variables predict belief. The second aim is to examine the cognitive reactions that emerged after the prophecy failed.

As potential predictors of belief, we focused on reflective thinking, scientific reasoning, and conspiracy mentality. Because the July 5 prophecy lacked a scientific basis, we hypothesized that lower levels of reflective thinking and scientific reasoning would predict stronger belief, which is consistent with evidence showing that such tendencies reduce resistance to unfounded claims (e.g., Yelbuz et al., 2022). Conversely, a stronger conspiracy mentality is associated with greater susceptibility to pseudoscientific information (e.g., Lewandowsky et al., 2013) and is therefore expected to predict stronger belief in the prophecy. Along with these considerations, we examined several cognitive biases that may have manifested after the prophecy failure. First, hindsight (Roese & Vohs, 2012) and retrospective biases—misremembering one’s prior level of worry as being lower than it actually was (Morgenstern & Barrett, 1974)—were assessed by directly comparing the participants’ reports before and after July 5. After learning that the prophecy had failed—that is, after updating their knowledge—individuals might reconstruct their memory of pre-prophecy beliefs and precautionary actions in light of the new information (Roese & Vohs, 2012). Another study reported similar results related to the pandemic—when people were asked one year later to recall their initial levels of interest and risk perception at the onset of COVID-19, they systematically underestimated those initial levels (Yamagata & Miura, 2023). Accordingly, we expected that they would tend to recall their prior belief in the prophecy and the extent of their precautionary actions before the prophecy as weaker than they actually were. We then analyzed the “moving the goal-post” effect—shifting the criteria or reinterpreting the evidence (Tsapos, 2023)—using both pre–post comparisons and models that included belief in the prophecy as a moderator. As Festinger et al. (1956) reported, when a believed prophecy fails, people may shift the criteria for what would count as fulfillment of the prophecy or attempt to reinterpret the prophecy. These responses can be understood as efforts to reduce the cognitive dissonance (Festinger, 1962) produced by disconfirmation. Accordingly, the stronger one’s prior belief, the greater the dissonance, and the more likely it is that such goal-post moving will occur. In the present study, cognitive dissonance is considered to be an underlying mechanism only for “goal-post moving,” whereas hindsight and retrospective biases are conceptualized as post-event reconstructive processes that do not require a dissonance-based account.

In the pre-July 5 survey conducted on July 3, the mean for belief in the prophecy was M = 2.18 (SD = 1.21), and only five participants scored 5 or higher. This indicates that the vast majority of the participants did not believe in the prophecy. Therefore, the subsequent analyses should be interpreted in light of this distribution—as relative comparisons within a sample in which those who strongly believed the prophecy were very few in number.

First, we calculated the correlations between belief in the prophecy (July 3) and CRT, SRS³3

Participants who failed the SRS check test (n = 42) were excluded from the dataset when calculating the correlation between SRS and belief in the prophecy.

Subsequently, to examine hindsight and retrospective biases, we ran paired-samples t-tests on July 3 (pre-July 5 survey) and July 10 (post-July 5 survey) ratings of belief in the prophecy and prophecy-related impact (Table 2). For hindsight bias, belief scores were significantly lower at post than at pre (t(317) = 6.41, p < .001, d = 0.36, 95% CI_diff[0.20, 0.38]), consistent with hindsight bias. Regarding retrospective bias, post-July 5 survey scores were significantly lower than pre-July 5 survey scores for anxiety, stockpiling, and schedule changing (anxiety: t(317) = 4.84, p < .001, d = 0.27, 95% CI_diff[0.16, 0.38]; stockpiling: t(317) = 4.97, p < .001, d = 0.28, 95% CI_diff[0.16, 0.38]; schedule changing: t(317) = 2.83, p = .005, d = 0.16, 95% CI_diff[0.05, 0.26]), whereas information sharing showed no pre–post difference (t(317) = 0.16, p = .872, d = 0.01, 95% CI_diff[-0.11, 0.12])⁵5

To examine the potential effect of repeatedly measuring belief in the prophecy within a short period, 330 participants from a separate sample were asked about their belief in the prophecy only on July 10. An independent-samples t-test comparing this group with the main dataset revealed no significant difference.

We examined the “moving the goal-post” effect by conducting linear mixed-effects models for the three prophecy fulfillment criteria items (casualties, location, and temporal window). For each item, survey time point (pre/post), belief in the prophecy (pre-July 5 survey), and their interaction were included as predictors. In all cases, the main effect of time point was not significant (casualties: B = 0.15, 95% CI [-0.05, 0.36], p = .140; location: B = -0.04, 95% CI [-0.10, 0.02], p = .236; temporal window: B = 0.05, 95% CI [-0.03, 0.13], p = .221). A significant interaction was observed for the temporal window criterion, whereas no significant interaction effects were found for the other variables. Simple slope analyses showed that, when belief in the prophecy was at +1SD, time point was significantly positively associated with the outcome (B = 0.16, 95% CI [0.05, 0.28], p = .005), whereas, when belief was at -1SD, time point was not significant (B = -0.06, 95% CI [-0.18, 0.05], p = .284) (Figure 1). These results suggest that the “moving the goal-post” effect was present to some extent.

This study aimed to examine who believed in the July 5, 2025, disaster prophecy and what cognitive responses emerged when it failed. Using pre- and post-July 5 surveys, we assessed predictors of belief in the prophecy—reflective thinking, scientific reasoning, and conspiracy mentality—and investigated the cognitive biases following its failure. Belief in the prophecy was negatively associated with reflective thinking and scientific reasoning and positively associated with conspiracy mentality. Although few participants strongly endorsed the prophecy, belief scores declined significantly after July 5, consistent with hindsight bias. Retrospective bias was likewise observed, broadly consistent with predictions. Among participants with relatively stronger pre-July 5 belief, a “moving the goal-post” effect was also observed.

First, notably, in contrast to the widespread public attention and commotion—reflected in the present survey by a recognition rate of approximately 89% (318 out of 359 respondents)—very few participants in our online sample reported strongly believing in the prophecy. Accordingly, the distribution of belief was strongly concentrated at the lower end, suggesting a floor effect. This absolute level of belief should, however, be regarded as descriptive of this specific sample rather than as a precise population estimate. Moreover, these data pertain only to Japanese respondents. In Hong Kong, for example, many people who believed in the prophecy reportedly cancelled their trips to Japan, resulting in a more than 30% decrease in tourist arrivals compared to previous years⁶6

https://www.japantimes.co.jp/news/2025/07/16/japan/tourism-growth-quake-rumor/

Most Japanese respondents did not strongly believe in the prophecy, although there were individual differences in the lingering suspicion that it might come true, which were explained by reflective thinking, scientific reasoning, and conspiracy mentality. Notably, the explanatory power of reflective thinking and scientific reasoning was particularly weak. Although previous research has consistently demonstrated that individuals with higher levels of reflective thinking and scientific reasoning are less likely to believe in poorly substantiated information (Yelbuz et al., 2022), the decision to accept or reject a given piece of information is not solely determined by logical or analytical thinking. Rather, it is shaped by motivated reasoning (Kahan, 2013) and individuals’ worldviews and belief systems (Ozono & Sakakibara, 2024). In the case of the present prophecy, it is plausible that individuals who habitually favor sensational, emotionally charged information, even when it is scientifically unfounded, were unable to fully dismiss the possibility that the prophecy might come true.

As expected, hindsight bias regarding the prophecy and retrospective bias were generally observed, but the “moving the goal-post” effect was only partially evident. The “moving the goal-post” effect emerged only for the temporal window criterion, possibly because the prophecy was strongly associated with the specific date of July 5, whereas its content regarding the number of casualties and the location of the disaster was less salient. Consequently, the latter criteria may have been vaguely defined from the outset, leaving no need to shift—or even no possibility of shifting—the goal-post. In contrast, the temporal window was clearly recognized; thus, once July 5 had passed, the prophecy would have been seen as having failed, creating the necessity to move the goal-post. These findings suggest that post-event cognitive biases may vary depending on the nature of the target being recalled and the aspects of the relevant event that were most salient to people.

The findings of this study are important in the following respects. First, a weakly substantiated prophecy capturing widespread public attention and triggering a societal commotion is rare. Taking advantage of this unique opportunity, we collected data demonstrating that psychological phenomena typically observed in highly unusual, socially constrained, or experimentally controlled environments can also be observed—at least to a considerable extent—in a natural setting. Second, people’s responses following the failure of the prophecy offer important insights for risk communication. As Norman et al. (2012) demonstrated, even after being deceived by misinformation, people may continue to believe similar information regardless of its source. Conversely, as the “cry-wolf” effect (LeClerc & Joslyn, 2015) suggests, once people have been misled by prior warnings, they may cease to trust subsequent alerts, even when those alerts are grounded in scientific evidence. Although the present study did not address this issue directly, future research should examine the distinct conditions under which experiencing hindsight and retrospective biases serve to increase or decrease risk perception. Moreover, although not initially anticipated, the fact that most people did not actually believe in the prophecy, despite the public attention and media coverage it received, provides an important reference point for understanding people’s attitudes toward similar future events.

However, this study has several limitations. First, the sample was limited to individuals who participated in an online survey. Such individuals may have more frequent exposure to online information than the general population and may have encountered more diverse information pertaining to the prophecy (including skeptical or corrective information), which could have introduced bias into the results and may partly explain why relatively few participants in our sample strongly believed in the prophecy.

Second, the age distribution of the sample was skewed toward older participants. Because the data were not collected evenly across younger and older age groups, generalizations should be made with caution.

Third, because the phenomenon under investigation occurred in Japan, the data were limited to Japanese participants. Whether the present findings would apply to similar phenomena in other countries or cultural contexts remains unclear.

Fourth, although this was an online survey, we did not implement specific procedures to detect or exclude responses generated by large language models (LLMs). Thus, we cannot completely rule out the possibility that a small portion of the data may include such AI-generated responses.

Fifth, although previous research has examined in detail the mechanisms and functions of hindsight and retrospective biases (e.g., Levine et al., 2009; Roese & Vohs, 2012; Schacter et al., 2024), we were not able to explain these biases observed in the study in terms of specific psychological processes or variables. Building on previous work, future studies should assess a wider range of psychological variables in order to clarify the processes that give rise to these biases. Finally, this study was not preregistered, which somewhat limits the strength of the confirmatory claims that can be made.