This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Human Factors, is properly cited. The complete bibliographic information, a link to the original publication on http://humanfactors.jmir.org, as well as this copyright and license information must be included.
Numerical anchoring occurs when exposure to a numeric quantity influences a person’s subsequent judgment involving other quantities. This could be applicable to the evaluation of pain, where exposure to an unrelated number before the evaluation of pain could influence pain ratings.
This study aimed to determine whether exposure to a random numeric anchor influences subsequent pain intensity ratings of a hypothetical patient.
In this study, 385 participants read a vignette describing a patient with chronic pain before being randomly assigned to one of four groups. Groups 1 and 2 spun an 11-wedge number wheel (0-10), which was, unbeknown to the participants, programmed to stop on a high number (8) or a low number (2), respectively. Group 3 spun a similar letter wheel (A-K), which was programmed to stop on either the letter C or I (control 1). Group 4 did not spin a wheel (control 2). Participants were then asked to rate the patient’s pain intensity using a 0 to 10 numeric rating scale.
The high-number group rated the patient’s pain (median 8, IQR 2) significantly higher than the letter wheel control (median 7, IQR 2;
Pain ratings were influenced by prior exposure to a random number with no relevant information about the patient’s pain, indicating anchoring had occurred. However, contrary to the traditional definition of anchoring where anchoring occurs even when participants are unaware of the anchor’s influence, in this study, the anchoring effect was seen only in participants who believed that the anchor had influenced them. This suggests that anchoring effects could potentially occur among health care providers tasked with evaluating a patient’s pain and should be evaluated further.
Health care providers are often required to assess and treat pain; however, it is recognized that health care provider ratings of a patient’s pain intensity may be biased and inaccurate [
One rarely studied situational factor that appears to contribute to biased health care provider ratings of a patient’s pain intensity has been termed
Not all anchors are unrelated to the subsequent decision-making process. Anchoring effects have also been studied in relation to pain but to a much lesser extent. Riva et al [
Pain-related numeric anchors appear to influence a health care provider’s perception of the patient’s pain [
This study tested four hypotheses. The study’s primary hypothesis (H1) was that participants who were exposed to a random numerical anchor would be influenced by that anchor, with the median pain rating of participants who were exposed to a high numerical anchor being significantly higher than the median pain rating of those who were exposed to a low numerical anchor. The second hypothesis (H2) was that the two groups of control participants who were not exposed to a numerical anchor would not differ in their initial pain intensity ratings. The third hypothesis (H3) was that participants who were originally not exposed to a numerical anchor would instead anchor to their original pain ratings when asked to rerate the patient’s pain, even if they were subsequently exposed to a high anchor. The fourth hypothesis (H4) was that participants who were exposed to a numerical anchor would deny that the anchor influenced their subsequent pain rating and that pain ratings would not differ between those who reported vs those who denied being influenced.
A total of 516 participants were recruited through Mechanical Turk (MTurk, version May 2018; Seattle, Washington), a Web-based study recruitment website that has millions of users worldwide who participate in Human Intelligence Tasks in exchange for money [
Flow chart of participant recruitment.
This study was reviewed and approved by the York University Research Ethics Board (Human Participants Review Committee certificate #e2018-017). Participants were recruited through MTurk, where the study was entitled “Answer a psychology survey about pain.” Participants would also see the brief description: “Complete psychological questions and complete a small task on the computer.” MTurk users were compensated US $1 to participate in the study, which took approximately 20 min to complete. We had no restrictions on the location or prior approval rating of the MTurk users. In addition to the survey responses, we also recorded the MTurk user’s internet protocol addresses to eliminate participants who may have attempted to complete the task more than once using multiple MTurk accounts. At the end of the survey, participants received a random code which they subsequently submitted to MTurk to receive payment and confirm that they had completed the survey.
The study was administered using Qualtrics software (version May 2018, Provo, Utah), a Web-based survey management system. Participants were directed to the Qualtrics website, where they first provided informed consent to participate. Each page of the survey consisted of a single question. Participants were unable to return to previous questions after completing an item to maintain the validity of the anchoring process. Participants began the survey by completing demographic questions, including questions regarding their history of pain. We included open-ended questions for participants who endorsed experiencing chronic pain to detail their pain history, which also served as an internal validation question to ensure consistency in participants’ responses. Participants who were inconsistent in their responses were removed from the analysis (“inappropriate responses” in
Steve lives in a modest house on a quiet, tree-lined street very close to a major highway. Last year, as Steve was driving to work one morning, he was involved in a serious collision that nearly cost him his life. He spent months in the hospital and underwent multiple surgeries to repair his leg which was shattered in the crash. After many more months of physical rehabilitation, Steve is left with chronic leg pain and requires a cane to walk especially when the pain acts up. Steve sees his physical therapist once a week for treatment and despite the increased pain he has after each session, he feels the therapy is helping.
Virtual spinning wheels comprised 11 wedges, were each created using Adobe Flash (version 2018, Adobe, Seattle, Washington) animation for the purposes of this study. Unbeknown to the participants, these virtual spinning wheels were programmed online to stop at a predetermined value. Participants in group 1 (n=102) and group 2 (n=93) spun a virtual wheel containing the numbers 0 to 10, which was programmed to stop on either a high number (8) or a low number (2), respectively. To control for viewing numeric values, participants in group 3 (n
Spinning wheels used for group 1 (A), group 2 (B), and group 3 (C and D).
Immediately after spinning the wheel, participants in groups 1 and 2 were asked to recall the number they saw on the wheel and to indicate if they thought the number was higher, lower, or equal to the intensity of pain that the patient in the vignette experiences on a typical day. Participants in group 3 were only asked to recall the letter they saw on the wheel spin. Participants in groups 1, 2, and 3 were then asked to estimate the patient’s pain intensity on a typical day using a numeric rating scale (NRS), ranging from 0 (
Participants in group 4 were asked to provide an NRS pain rating immediately after reading the vignette. On providing a pain rating, participants in group 4, who initially did not spin a wheel, were asked to reread the vignette, spin the high-anchor wheel (set to stop on the number 8), and rerate the patient’s pain. This was done to determine whether participants in group 4 would anchor to their own original pain rating or if they would be influenced by the numerical anchor.
After completing the experimental task, all participants completed the Pain Catastrophizing Scale (PCS) questionnaire and the Hospital Anxiety and Depression Scale (HADS) questionnaire, as previous studies have indicated that both pain catastrophizing and anxiety or depression can influence pain ratings [
HADS measures symptoms of anxiety and depression and has been widely used among both clinical and nonclinical populations [
PCS measures the extent to which an individual experiences pain-related catastrophic thinking, including how much they think and worry about pain, magnify the amount of pain experienced, and feel helpless toward painful experiences. It consists of 13 items, each rated on a 5-point Likert scale, with scores ranging from 0 to 52. Scores above 30 are considered to be clinically relevant for catastrophizing [
Sample size estimation using G*Power (version 3.1.9.4; University of Düsseldorf, Germany) [
Data analyses were conducted with a significance level of .05. Chi-square tests of independence were conducted to determine any significant demographic group differences. A Kruskal-Wallis test was used to determine whether the groups differed in age.
H1 was analyzed using a nonparametric Kruskal-Wallis test, as initial screening of the data revealed a non-normal distribution, necessitating a nonparametric approach to data analysis (see the Results section). The medians of the four groups were compared to determine whether the high and low numerical groups (groups 1 and 2) significantly differed and to determine whether the median pain ratings of groups 3 and 4 were higher than the median pain ratings of group 2 and lower than group 1.
H2, stating that the two control groups (groups 3 and 4) would not significantly differ from one another, was analyzed using a Kruskal-Wallis test.
H3, stating that participants in group 4 would anchor to their original pain ratings rather than be influenced by the high numerical anchor, was analyzed using a Friedman test.
H4, stating that the median pain ratings between participants who believed they had been influenced and participants who believed they had not been influenced by the numerical anchor would not differ, was first analyzed using a chi-square test of independence to determine whether the proportion of participants being influenced by the anchor differed by group. A Kruskal-Wallis test was used to determine if pain intensity ratings were significantly different across groups for those participants who reported they had not been influenced by the anchor and those who felt they had been influenced by the anchor.
Chi-square tests of independence did not demonstrate significant differences between groups in gender, ethnicity, education, or pain history (see
A visual inspection of the histograms shown in
Numeric rating scale pain intensity scores for the four groups.
Pain intensity ratings | Wheel 8 (n=102) | Wheel 2 (n=92) | Letter wheel (n=102) | Control (n=87) |
Pain intensity rating, median (IQR) | 8 (2) | 6 (2) | 7 (2) | 7 (2) |
Pain intensity rating after spinning the wheel (group 4 only), median (IQR) | N/Aa | N/A | N/A | 7 (2) |
aNot applicable.
Frequency (percent) of pain intensity ratings for the four groups.
Pain intensity rating (0-10) | Wheel 8 (n=102), n (%) | Wheel 2 (n=92), n (%) | Letter wheel (n=102), n (%) | No wheel (n=87), n (%) |
0 | 0 (0.0) | 0 (0) | 0 (0.0) | 0 (0) |
1 | 0 (0.0) | 0 (0) | 0 (0.0) | 0 (0) |
2 | 0 (0.0) | 3 (3) | 0 (0.0) | 0 (0) |
3 | 2 (2.0) | 6 (7) | 4 (4.0) | 1 (1) |
4 | 6 (6.0) | 12 (13) | 8 (8.0) | 5 (6) |
5 | 15 (15.0) | 9 (10) | 13 (13.0) | 7 (8) |
6 | 12 (12.0) | 25 (27) | 20 (20.0) | 18 (21) |
7 | 12 (12.0) | 16 (17) | 27 (27.0) | 26 (30) |
8 | 40 (39.0) | 15 (16) | 18 (18.0) | 20 (33) |
9 | 9 (9.0) | 3 (3) | 10 (10.0) | 8 (9) |
10 | 6 (6.0) | 3 (3) | 2 (2.0) | 2 (2) |
Boxplots of pain intensity ratings for groups 1-4.
Kruskal-Wallis tests showed a significant difference between the mean ranks of at least one pair of groups in their pain intensity ratings (
Significant differences were not observed in pain ratings between groups 3 and 4 (
A Friedman test indicated that there were no significant differences in pain ratings for group 4 between time 1, initially after reading the vignette (meanrank 1.55), and time 2, after rereading the vignette and spinning the high-anchor wheel (meanrank 1.45; χ21=3.2;
A chi-square test of independence demonstrated that there were significant differences between groups in the proportion of participants who believed that their pain intensity rating of the patient had been influenced by the number they spun (χ23=11.0
In particular, participants in group 1 were significantly more likely to believe that they had been influenced by the anchor, whereas participants in group 3 were significantly more likely to believe that they had not been influenced by the anchor. In group 1, 35.3% (36/102) of participants endorsed being influenced in comparison with 20% (19/93) of participants in group 2, 16.7% (17/102) of participants in group 3, and 22% (19/87) of participants in group 4 after these participants had spun the high-anchor wheel.
Participants’ perceptions of whether they were influenced by the anchor that they were exposed to.
Influence | Group 1, (n=102), n (%) | Group 2, (n=92), n (%) | Group 3, (n=102), n (%) | Group 4, (n=87), n (%) | Chi-square ( |
|
Yes | 36 (35.2) | 19 (21) | 17 (16.7) | 19 (22) | 11.0 (3) | .01a |
No | 66 (64.7) | 74 (80) | 85 (83.3) | 67 (77) | N/Ab | N/A |
aSignificance was at an alpha level of .05.
bNot applicable.
A Kruskal-Wallis test indicated that among participants who indicated that they had not been influenced by the anchor, there were no significant differences between groups in pain intensity ratings (
This study examined whether prior exposure to a pain-unrelated, random numerical anchor would influence the participants’ ratings of a hypothetical patient’s pain intensity. This was done by asking participants to read a vignette depicting a hypothetical patient with chronic pain, before asking the participants to spin a wheel, which was programmed to land on a high numerical anchor (8), a low numerical anchor (2), or a letter (C or I). A fourth group served as a control condition and did not spin a wheel initially before rating the patient’s pain intensity but was later asked to spin the high-anchor wheel and rerate the patient’s pain.
The findings supported the main hypothesis in that exposure to a numerical anchor influenced the participants’ estimations of a hypothetical patient’s pain intensity. Participants who spun a high numerical anchor estimated that the hypothetical patient experienced a much higher pain intensity than did the other three groups. In addition, participants in the low numerical anchor condition had the lowest estimation of pain intensity for the hypothetical patient. Importantly, H2 was supported, as there was no difference in pain intensity ratings between participants who spun a wheel containing a letter and the control group that did not spin a wheel, indicating that the spinning of the wheel itself had no effect on pain intensity ratings. These results are in line with studies that have also used a spinning wheel or similar devices to anchor their participants to a random numerical anchor [
The third hypothesis was supported in that participants who were originally not exposed to an anchor anchored to their original pain rating when asked to rerate the patient’s pain, even when subsequently exposed to the high anchor. Participants did not adjust their second pain rating when asked to rerate the patient’s pain. This was expected, given the results from the study by Riva et al [
The fourth hypothesis, that the median pain ratings in each group would not differ between those who did and did not believe they had been influenced by the anchor, was unsupported. In the high-anchor group, those participants who believed they had been influenced had a significantly higher median pain rating than those who did not believe they had been influenced. Similarly, in the low-anchor group, those participants who believed they had been influenced had a significantly lower median pain rating than those who did not believe they had been influenced. Although the majority of participants in all four groups indicated that they had not been influenced by the anchor, participants who spun a high-anchor wheel were also more likely than any other group to indicate that they had been influenced by the anchor. This may relate to the abovementioned suggestion. The vignette may have depicted a higher pain rating, and after spinning the wheel and rating the patient’s pain as higher, the participants may then have inferred that they must have been influenced. This has been discussed later in detail.
The results also demonstrate that participants who acknowledged the anchor’s influence on their pain rating were, in fact, influenced. Among participants who reported that they had been influenced by the anchor, the results were very similar to the overall study findings in that participants who spun a high-anchor wheel rated the patient’s pain as being more intense than all other groups. In contrast, the median pain ratings for all four groups were not significantly different among those participants who indicated that they had not been influenced by the anchor. In other words, the anchoring effect was seen only in participants who reported being aware of the anchor’s influence on their decision making. These results deviate from previous studies that have examined the role of influence on anchoring effects. Although only one study has looked directly at whether participants believed they had been influenced or not [
The effect of influence that was seen in the three anchoring conditions was also seen in participants in group 4, who initially did not spin a wheel. After rereading the vignette, the participants were asked to spin the high-anchor wheel and rerate the patient’s pain. Overall, the participants did not change their pain intensity rating after rerating the pain, which was expected. Riva et al [
By taking into consideration the entire sample, the results suggest that anchoring has occurred. However, when considering the effect of influence, anchoring only appears to have occurred in those who reported that they had been influenced. These findings are contradictory to the traditional definition of anchoring, where anchoring is conceived as an implicit cognitive process and is thought to occur regardless of the participant’s awareness of the anchor’s influence on their subsequent decisions.
The effect of influence rarely has been studied in anchoring. Given the traditional anchoring template as designed by Tversky and Kahneman [
It is possible that these influence effects seen across groups are because of a confirmatory search mechanism, as proposed by Chapman and Johnson [
This study has a number of limitations that are important to consider. Given that the study was completed online, it is possible that participants were not able to fully attend to the vignette, the wheel, or the subsequent questions. As a result, the anchoring effects and influence effects seen may be instead due to the fact that the participants had very recently been exposed to a number rather than the true anchoring effects, ie, if participants were not attending fully, they may have rated the participant’s pain according to the numerical anchor they were exposed to simply because of the availability of the anchor in their memory rather than because that is the pain intensity rating they believe the patient experiences or because of anchoring effects. These same participants might subsequently indicate that they had been influenced by the anchor, as their response was based on the number they had been exposed to. Previous studies have demonstrated that data collected through MTurk are as reliable as data collected in a laboratory setting, with the exception of attention paid to the study itself [
A second limitation is that this study has no pilot data on the vignette that was used to give a description of the hypothetical patient. As a result, it is unknown what the patient’s baseline pain intensity would be rated as. This information would help to ensure that the vignette itself was not a confounding variable. For instance, if the vignette was shown to depict a pain intensity that is higher without the presence of a numerical anchor, it is possible that the influence effect that was seen in the high-anchor group may have been because of participants inferring that they had been influenced, given the pain intensity rating that they had given.
Finally, this study is limited by the fact that it is one of the first anchoring studies to look at the effect of influence on anchoring effects. As such, the questions regarding influence had not been previously tested and may not have been valid or may have unwittingly created biased responses.
Despite the abovementioned limitations, the study also has a number of strengths. First, with a relatively large sample size of participants who were recruited globally, it is likely that the data are not only reliable but also cross-culturally validated. Participants were diverse in their age, education, ethnicity, and pain history, which also helps to ensure that the data are valid and generalizable. Although participant characteristics are often unreported in studies that use crowdsourcing such as MTurk [
Second, this study is strengthened by the presence of two control conditions. In this way, both the effect of spinning a wheel and the effect of having the wheel land on a number could be controlled. This helps to ensure that the anchoring effects seen are, in fact, because of anchoring effects, as opposed to being because of a confounding variable.
Finally, this study is one of the only studies to have looked at the effect of influence and found that anchoring effects were contingent upon the participant’s belief that they had been influenced. Anchoring research has been very robust and well established, but there has been very little research on the effect of influence on anchoring and what these findings mean for the definition of anchoring itself. This study’s results may help to better understand anchoring effects as a whole as well as its underlying cognitive pathways.
Future studies should attempt to clarify the role of influence on numerical anchoring. Namely, attempts should be made to replicate anchoring studies while also considering the participant’s perception of influence. It may be that the current definition of anchoring is not suitable if the effects of influence are reliably seen across studies, given that the current definition implies that participants are not aware of the anchor’s influence on their judgment. Future studies should also expand on this research about how random numerical anchoring might affect the pain response. It would be interesting to determine whether these same random numerical anchors would affect a participant’s judgment of their own pain experience in both acute and chronic pain patients. Future studies may also look at how numerical anchoring may be evident in the health care context in relation to how random numerical anchors may influence a health care provider’s judgment and treatment of a chronic pain patient’s experience.
The results of this study are consistent with previous studies of numerical anchoring. Exposure to a high numerical anchor influenced participants’ subsequent rating of a hypothetical patient’s pain to be higher, whereas exposure to a low numerical anchor influenced participants to rate the patient’s pain as lower. However, although the majority of participants across groups did not believe they were influenced by the anchor, the anchoring effect was seen only in participants who did indicate that the anchor had influenced their judgments. Further research is necessary to determine the role of influence on anchoring effects and the applicability of anchoring effects in the health care context.
Anchoring questions.
Demographic information for the four groups.
Frequency distributions of pain intensity ratings for groups 1-4.
Hospital Anxiety and Depression Scale
Mechanical Turk
numeric rating scale
Pain Catastrophizing Scale
JK is supported by a Canadian Institutes of Health Research Canada Research Chair in Health Psychology at York University. The authors thank M Gail Rudakewich of Synapse Visuals for creating the Adobe Flash animation spinning wheels used in this study.
None declared.