Gain/loss framing moderates the VMPFC’s response to persuasive messages when behaviors have personal outcomes

Abstract

Activity within the ventral medial prefrontal cortex (VMPFC) during encoding of persuasive messages has been shown to predict message-consistent behaviors both within scanner samples and at the population level. This suggests that neuroimaging can aid in the development of better persuasive messages, but little is known about how the brain responds to different message features. Building on past findings, the current study found that gain-framed persuasive messages elicited more VMPFC activation than loss-framed messages, but only when messages addressed outcomes that would be experienced by participants directly. Participants also perceived gain-framed messages as more effective than loss-framed messages, and self-reported perceptions of message effectiveness were positively correlated with VMPFC activation. These results support theories that VMPFC activity during message encoding indexes perceptions of value and self-relevance and demonstrate that established theories of persuasion can improve the understanding of the neural correlates of persuasion.

Introduction

More than a decade of research has shown that activation of the ventral portion of the medial prefrontal cortex (VMPFC) in response to persuasive messages predicts message-consistent behaviors and other desired outcomes, accounting for variance in behavior that is unique from self-report measures of message effectiveness (Falk et al., 2010, 2011, 2012, 2015a, 2015b; Cooper et al., 2015; Vezich et al., 2017; Pandey et al., 2021). This relationship has been established at two different levels. At the participant level, individuals who show more VMPFC activation during message exposure are more likely to show desired behavior changes after message exposure (Falk et al., 2010). At the message level, messages that elicit more VMPFC activation in small scanner samples tend to be more successful at the population level (Falk et al., 2012). Researchers have suggested that increased VMPFC activation underlies perceptions of message value in relationship to one’s goals and motivations (Falk and Scholz, 2018), but little is known about the ways this process might be affected by features of the messages themselves (Falk et al., 2015a). This has made it difficult for scholars to connect neuroimaging research with existing theories of persuasion (Vezich et al., 2016) and develop evidence-based recommendations for message design.

To bridge this theoretical gap, the current study sought to build upon recent research into effects of a common message-level feature: gain/loss framing. Specifically, the current study follows a line of inquiry opened by Vezich et al., (2017), who examined VMPFC activation in response to messages that promoted sunscreen use. These researchers found that gain-framed messages (i.e. messages that emphasized potential benefits of using sunscreen) elicited more VMPFC activation than loss-framed messages (i.e. warnings about the risks of not using sunscreen), and that the strength of this gain/loss framing effect on VMPFC activation predicted actual sunscreen use even when controlling for behavioral intentions (Vezich et al., 2017). Importantly, this effect was only observed in responses to a ‘why’-framed subset of the messages (i.e. messages explaining why one should wear sunscreen). When researchers attempted to replicate this effect using functional near infrared spectroscopy, they observed a positive association between MPFC activation and sunscreen use but did not find evidence that MPFC activation was sensitive to the gain/loss framing of messages (Burns et al., 2018).

Given the prominence of gain/loss message framing in health communications research (Guenther et al., 2021), inconsistent findings within the original Vezich et al., (2017) study between ‘how’ and ‘why’ messages and the observation of null gain/loss framing effects in the attempted replication (Burns et al., 2018), there is a need to further examine how gain- and loss-framed messages differ in their effects on the VMPFC, a key region associated with effective messaging and behavior change (Falk and Scholz, 2018). Further, it remains unknown whether the gain/loss framing effect can be observed outside the context of messages about sunscreen use. Therefore, the current study aimed to examine whether VMPFC activity differed in response to gain- vs loss-framed messages that varied across two dimensions: the temporal orientation of described outcomes (present-oriented vs future-oriented) and the nature of those outcomes (personal vs prosocial).

Persuasion and VMPFC activation

Activation of various subregions of the MPFC in response to persuasive messages has been associated with desired outcomes across a variety of messaging contexts (for a review, see; Cacioppo et al., 2018). For example, research examining VMPFC activity during exposure to messages that presented expert information about the benefits of sunscreen use compared to rest found that participants who displayed increased VMPFC activation when processing the appeals were more likely to increase sunscreen use in the week after the scan (Falk et al., 2010). In addition, increased MPFC activity among smokers exposed to tailored cessation messages (compared to untailored messages) predicted abstention from cigarettes a full four months after the scan session (Chua et al., 2011). Similar effects have been observed among participants exposed to messages encouraging physical activity (Cooper et al., 2017) and to pictorial warning labels designed for cigarette packages (Riddle et al., 2016). In all cases, VMPFC activation predicted variance in outcomes of interest over and above the variance explained by self-report measures.

Researchers have also found that VMPFC activation predicts desired outcomes at the message level. For example, when Falk et al., (2012) examined responses to three sets of anti-smoking public service announcements, they found that a rank ordering of the campaign messages by average VMPFC activation aligned with a rank ordering of the call volume to a tobacco quitline in the months after each campaign was broadcast (driving smokers to call the quitline was the intent of all campaigns). Interestingly, this rank ordering was different from the order both participants and experts proposed when they were asked to rank the campaigns by likelihood of success—the message that participants and experts ranked as the least likely to succeed ultimately elicited the highest average level of VMPFC activation and the greatest number of calls to the quitline. Similarly, research examining the relationship between smoking cessation materials and email click-through rates found that materials that elicited increased VMPFC activation in an fMRI sample also had higher click-through rates when distributed at the population level through an email newsletter (Falk et al., 2015b). Once again, participants’ perceived message effectiveness (PME) ratings were not effective predictors of message success.

In sum, activity within the VMPFC in response to persuasive messages has been shown to predict both individuals’ tendencies to engage in message-consistent behaviors and the tendency of messages to elicit desired outcomes in real-world settings (Falk et al., 2010, 2012). Furthermore, the tendency of VMPFC activity to outperform self-report as a predictor suggests that this measure might capture some message effects that occur outside of participants’ conscious awareness.

Because no brain region is exclusively associated with any one cognitive process, it is difficult to infer an individual’s cognitive experience from activation alone (Poldrack, 2006). Having witnessed its activation in several similar structured contexts, however, Falk and Scholz (2018) have proposed that VMPFC activation elicited by persuasive messages indicates a process of subjective valuation, which they assert is strongly associated with perceptions of self-relevance when audiences process persuasive messages. In other words, they proposed that VMPFC activity elicited by persuasive messages indicates that audience members perceive the value of the message as being relevant to them personally. Similarly, Vezich et al., (2017) suggested this activity might signal a process of self-value integration, in which individuals identify the target behavior as valuable and choose to incorporate it into their self-concept. Assertions about the nature of this process remain hypothetical, but a growing body of work suggests that VMPFC activity is a reliable predictor of persuasive message success.

Prospect theory and gain/loss messages

First proposed in the field of behavioral economics, the framing postulate of prospect theory (Kahneman and Tversky, 1973; Tversky and Kahneman, 1989) asserts that people tend to be risk-averse when considering the benefits of an action but are willing to tolerate risk when motivated to avoid a negative outcome. In the context of health promotion, existing evidence suggests that loss-framed messages are more effective at promoting detection behaviors that might involve undesirable outcomes, such as cancer screening and testing for sexually transmitted infection (Rothman and Salovey, 1997; Rothman et al., 2020). On the other hand, messages promoting relatively low-risk behaviors such as increases in physical activity tend to be more effective when they emphasize potential gains (Rothman and Salovey, 1997; Rothman et al., 2020). Given that sunscreen use is a relatively low-risk behavior, Vezich et al., (2017) found that gain-framed messages on this topic tended to elicit increased VMPFC activation, which is consistent with prospect theory’s framing postulate. Still, it remains to be seen whether this effect might also be observed in different combinations of message features.

Message features

Whenever researchers observe an effect within an experimental sample, they must consider sources of heterogeneity among participants before generalizing that effect to a larger (and likely more diverse) population of people. Similarly, researchers hoping to identify generalizable message-level effects must account for the reality that persuasive messages can differ in many ways. At present, most research on neural processes associated with persuasion combines multiple message features, which could explain why average-level message effects have been identifiable, but have included positive, negative and null effects (O’Keefe and Hoeken, 2021). To explore how gain/loss framing effects might generalize to a more diverse set of relevant messages, we varied our sample of messages across two additional dimensions: the temporal orientation of the gain or loss outcome and the object of that outcome.

Temporal orientation of the outcome (present vs future)

When describing the likely outcomes of adopting or failing to adopt a certain behavior, message designers often must choose between immediate (present-oriented) or distal (future-oriented) outcomes. For example, sunscreen use can be associated with either present-oriented gains like social approval or future-oriented gains like younger-looking skin later in life. Similarly, failure to use sunscreen can be associated with both present-oriented losses like sunburn and future-oriented losses like an increased risk of skin cancer.

Past research has shown that changes in temporal orientation can impact the effectiveness of messages that advocate for both low-risk (Strathman et al., 1994) and detection (Kreuter et al., 2005) behaviors and that the direction of these effects depends on traits of a message’s audience. Therefore, the present study tested for gain/loss framing effects in messages that oriented outcomes in the present and in messages that oriented outcomes in the future.

Nature of the outcome (personal vs prosocial)

Another important difference in persuasive messages is the nature of the behavior advocated for. Messages often focus on behaviors that impact individuals personally—for example, the outcomes of improving diet or failing to apply sunscreen will be felt primarily by the person who does or does not engage in those behaviors—but this is not always the case. Some messages advocate instead for prosocial behaviors, for which risks are incurred by the individual but benefits are experienced by another or by the community at large. For example, Helme et al., (2020) showed that Appalachian residents perceived preventing accidental and intentional misuse of opioids by others in their household as a benefit to disposing of unused opioid medications instead of keeping them in the home. Additionally, potential organ donors perceive the opportunity to save a life as a benefit of donation (Siegel et al., 2010; Quick et al., 2014; Williamson et al., 2017).

The study of behavior change has attracted increasing interest in the health and environmental domains (Liang et al., 2018) which also offer prosocial benefits (e.g. organ donation) or a mix of personal and prosocial benefits (e.g. vaccination). However, the field has not engaged in much theorizing and direct comparison of personal and prosocial benefits of behaviors. Work in environmental psychology gives some insight into the unique challenges associated with prosocial behavior, including difficulties thinking beyond the self and drawing a connection between one’s own behavior choices and positive social impact (Gifford, 2011; Liang et al., 2018). Therefore, the context of relevant outcomes (i.e. whether they were framed as applying to the individual or to some external agent) is a dimension along which messages often vary. In the present study, this was explored by testing messages that advocated for two types of behaviors: physical exercise (personal) and pro-environmental behaviors (prosocial).

The current study

The current study seeks to expand on the finding of Vezich et al., (2017) that gain-framed messages tend to elicit increased activation in the VMPFC compared to loss-framed messages encouraging adoption of a low-risk behavior. We sought to test whether this effect could be generalized to a set of messages that varied across some common dimensions (present-/future-oriented, personal-/prosocial-focused). In addition, given that self-report ratings of PME have varied in how well they correspond to neural activity during message encoding (Falk et al., 2012, 2015a), we tested both the effects of gain/loss framing on PME and the associations between PME and VMPFC activity. We hypothesized that gain-framed messages would elicit increased activation of the relevant VMPFC subregion and higher levels of PME compared to loss-framed messages across all possible message conditions.

Methods

Participants

Forty-five participants were recruited through an online job board from a large Midwestern research university, aged 18–34 years old (M = 20.53, SD = 2.65; N_Female = 30). All participants were right-handed, did not suffer from claustrophobia, had normal (or corrected to normal) vision, were not taking any psychoactive medications and did not have any history of psychiatric or neurological disorders. Data from one participant were excluded from this analysis due to errors during data collection, leaving a final sample of 44 participants.

Given concerns about low statistical power in neuroimaging studies (Yarkoni, 2009), we used the software G*Power to test whether our sample size was sufficient to detect the effects observed in past research with statistical power at 0.90, α = 0.05 using one-sample t-tests. To observe an estimated effect size d = 1.22, as calculated using the six R-squared values in Burns et al., (2018), the recommended total sample size was eight participants. To detect an estimated average effect size d = 0.62, which we calculated using the nine Pearson’s r values in Vezich et al., (2017), the recommended total sample size was 29 participants. Thus, our sample of N = 44 was deemed sufficient to detect similar effects.

Procedure

‘After obtaining consent, participants completed a series of pre-scan questionnaires (unrelated to this study) and fMRI safety screening and were trained on fMRI tasks prior to the scanning session. Participants then underwent a one-hour scan session at the Brain Imaging Core at the Waisman Center at the University of Wisconsin-Madison that included a nine-minute persuasive message task, which was the first task participants completed. All components of this protocol were approved by the University of Wisconsin-Madison IRB (#2019-0580). Statistical data, stimuli and analysis codes have been made publicly available at https://doi.org/10.17605/OSF.IO/FWE8B. In addition, fMRI data are available upon request.

Persuasive message tasks

While undergoing fMRI, participants viewed a series of 44 randomly ordered persuasive messages in a rapid event-related design. Messages were constructed for this experiment and varied across three fully crossed conditions: each used either a gain or loss frame (gain/loss conditions), described the outcomes of adopting or failing to adopt the behavior in either the present or the future (present-oriented/future-oriented conditions) and advocated either for an increase in physical activity or adoption of pro-environmental behaviors (personal/prosocial conditions). Two examples are presented below (Figure 1).

Fig. 1.

Left: A loss-framed, long-term message focused on personal outcomes. Right: A gain-framed, present-oriented message focused on prosocial outcomes.

Participants viewed each message for a period of five seconds, then were asked ‘How effective is this message?’ to rate their PME on a scale of 1 to 4 using a four-button Current Designs response pad held in the participant’s right hand. Participants were given three seconds to enter a rating and then viewed a fixation cross for a period of 1.5 seconds before presentation of the next message.

Framing manipulation check

Manipulation of the gain/loss framing of stimuli in this study was performed by making changes to the text that accompanied each image, but each textual message was always paired with the same image. To ensure that any effects of gain/loss framing were driven by manipulation of the text instead of some feature of the images used, we conducted a follow-up online survey to a demographically similar sample of N = 238 participants recruited via Qualtrics Panels. For the online study, we altered the 22 personal messages by reversing the frame associated with each image. Thus, each image that was associated with a gain-framed textual message in the original study was associated with a loss-framed textual message in the survey, and vice versa. Participants were prompted to provide a PME rating to the prompt: ‘How effective is this message?’ Response options ranged from 1(Not at all effective) to 5(Extremely effective), similar to the fMRI study.

fMRI data acquisition and preprocessing

Structural and functional brain imaging was conducted using a 3 Tesla GE Discovery MR750 scanner. Head motion was minimized using foam padding on the head coil. One functional run was used for analysis (Repetition Time = 800 ms, TE = 20, flip angle = 60°, matrix size = 96 × 96, 54 axial slices, 3 mm thick; voxel size = 3.0 × 3.0 × 3.0), and a motion-corrected T1-weighted MPnRAGE acquisition with 1.0 mm isotropic spatial resolution was used as an anatomical underlay (Kecskemeti et al., 2018). Image preprocessing and analysis were performed using the afni_proc.py program within the Analysis for Functional Neuroimaging (AFNI) software package (Cox, 1996). Functional and anatomical runs were warped to align with the Montreal Neurological Institute template brain and smoothed with a 4 mm Gaussian kernel. To ensure only steady-state images were used in our analysis, we discarded the first 7 TRs (5600 ms).

To assess the effects of gain/loss framing in the full dataset and in response to four relevant subsets, we estimated first-level regression models for the gain and loss conditions across the full set of messages and for message subsets of interest, including present-oriented, future-oriented, personal and prosocial messages. Each model included instances of both gain- and loss-framed messages as regressors, as well as a regressor for the parametric modulation of neural signal by PME. Random effects, motion and nuisance regressors were also modeled using AFNI’s 3dDeconvolve command.

Region of interest analyses

We were interested in a single a priori region of interest: a portion of the VMPFC overlapping the region of interest (ROI) used by Vezich et al., (2017). Activation of that region in response to persuasive messages has been used to predict behavior in a variety of contexts (Falk et al., 2011, 2012, 2015a; Cooper et al., 2015). Like Vezich and colleagues, we focused specifically on the cluster that was found to predict behavior (Falk et al., 2010). Specifically, we constructed a 10 mm, 33-voxel sphere around the coordinate 0.60,-9 using the Montreal Neurological Institute template brain. From this ROI, we exported four different measures for our analyses: gain/loss framing effects, present/future orientation effects, personal/prosocial outcome effects and parametric modulation by PME (Figure 2).

Fig. 2.

The VMPFC was constructed using a 10 mm, 33-voxel sphere centered around the coordinates x = 0, y = 60, z = −9 using the Montreal Neurological Institute template brain.

We estimated gain/loss framing effects for each of our four message design categories: present personal, future personal, present prosocial and future prosocial. To do this, we performed a general linear test comparing the signal during exposure to gain-framed messages with activation during exposure to loss-framed messages for each category and then averaged coefficients for all voxels within each participant’s VMPFC ROI. We estimated present/future and pro-activity/pro-environmental effects in a similar manner, comparing present-oriented messages with future-oriented messages and personal with prosocial messages, respectively.

Statistical tests

Tests of gain/loss framing on VMPFC activity

First, we tested whether the VMPFC was sensitive to the gain/loss framing of messages by conducting a series of one-sample t-tests on our measure of the gain/loss framing effect within our ROI. We performed four independent tests—one for each of our message conditions: present personal messages, future personal messages, present prosocial messages and future prosocial messages.

Second, we tested whether the effects of gain/loss framing on VMPFC activity were different for different combinations of message features by testing the fit of a series of linear mixed effects regression models. Each model used our measure of gain/loss framing effects in the VMPFC as the outcome and included dummy-coded regressors for three of the four possible message conditions. To account for the nested nature of the data, each model also included a by-participant random intercept and a by-participant slope for each dummy-coded regressor. Three models were tested in all, allowing for pairwise comparisons of all message conditions.

Third, we tested whether our other manipulated message features had any effects on VMPFC by conducting one-sample t-tests on measures of present/future orientation effects and personal/prosocial effects.

Tests of effects of message conditions on PME

To test the effects of our message conditions on participants’ PME ratings, we fit a linear mixed effects model in which PME ratings were the outcome and our gain/loss, present/future and personal/prosocial conditions and their interactions were entered as fixed effects. To account for the nested nature of our data, we also included by-participant and by-stimulus random intercepts and random slopes for all fixed effects (models that included by-participant or by-stimulus random slopes for interaction effects failed to converge).

Test of relationship between VMPFC activation and PME

To test whether VMPFC activation during message encoding was associated with participants’ subsequent rating of the messages, we conducted a one-sample t-test on the average ROI-wide coefficients for our parametric modulation regressor.

Test of framing manipulation check

To confirm that gain/loss framing effects were not driven by the images used in our stimuli, we conducted a test on our independent survey sample in which the images used were reversed. We tested the fit of a linear mixed-effects model with fixed effects for gain/loss framing and present/future orientation, by-participant random slopes for each fixed effect and a by-participant random intercept.

Results

Effects of gain/loss framing on VMPFC activity

First, we tested whether the gain/loss framing of messages impacted the level of activation elicited within the VMPFC for each of our message categories. Results indicated that neural activity within the VMPFC was significantly higher in response to gain-framed messages than loss-framed messages for two of our four message categories: present personal messages (t(43) = 4.062, P <0.001, P_adjust < 0.001 95% confidence interval (95% CI) [0.343, 1.018]) and future personal messages (t(43) = 3.526, P = 0.001, P_adjust = 0.005, 95% CI [0.183,0.671]). We did not find evidence that gain/loss framing had any effect on VMPFC activity during encoding of either present prosocial (t(43) = 0.747, P = 0.357, P_adjust = 0.714, 95% CI [−0.203, 0.442]) or future prosocial messages (t(43) = 1.208, P = 0.459, P_adjust = 0.714, 95% CI [−0.013, 0.052]). All P values were adjusted for multiple comparisons using the Holm–Bonferroni method. Results of these tests are illustrated in Figure 3.

Fig. 3.

X-axis describes message category, Y-axis describes average VMPFC activity from the contrast gain-loss. Error bars represent 95% confidence intervals.

Second, we compared the strength of gain/loss framing effects across our message categories by testing a series of linear mixed effects models. These tests allowed us to compare each possible pair of categories. We found that the gain/loss framing effect in response to present personal messages was significantly stronger than that for future prosocial messages (β = 0.532, F(1,48.818) = 7.963, P = 0.007, P_adjust = 0.042) and was stronger than present prosocial messages, though this effect did not remain significant after our multiple comparison correction (β = 0.561, F(1,52.367) = 6.342, P = 0.015, P_adjust = 0.075). We did not find evidence that the strength of the gain/loss framing effect differed across any other pair of message conditions. Complete results are presented in Table 1.

Table 1.

Results of a series of linear mixed-effects models comparing the strength of gain/loss framing effects across message categories

Future personal	β = 0.254 SE = 0.181 F(1, 48.057) = 1.954 P = 0.169 Padjust = 0.399
Present prosocial	β = 0.561 SE = 0.223 F(1,52.367) = 6.342 P = 0.015* Padjust = 0.075	β = 0.307 SE = 0.201 F(1,52.124) = 2.335 P = 0.133 Padjust = 0.399
Future prosocial	β = 0.532 SE = 0.189 F(1,48.818) = 7.963 P = 0.007** Padjust = 0.042*	β = 0.279 SE = 0.163 F(1,46.520) = 2.941 P = 0.093 Padjust = 0.372	β = –0.028 SE = 0.192 F(1, 49.076) = 0.022 P = 0.883 Padjust = 0.883

Notes: Significant results indicate a significant difference in the strength of the gain/loss framing effect. Adjusted P values reflect correction for multiple comparisons using the Holm–Bonferroni method. * = P < .05, ** = P <.0.

Finally, we tested whether activity in the VMPFC was affected by either of our other message manipulations: present/future orientation or personal/prosocial outcomes. We did not observe significant differences in VMPFC activity when comparing present-oriented message exposures to future-oriented message exposures (t(43) = −0.538, P = 0.999, 95% CI [−0.024, 0.014]), and the difference we observed between messages with personal outcomes vs those with prosocial outcomes did not survive our multiple comparison correction (t(43) = 2.28, P = 0.027, P_adjust = 0.110, 95% CI [0.003, 0.050]).

Effects of gain/loss framing and other message features on PME

Next, we tested whether gain/loss framing and message features affected participants’ PME ratings. Tests of our mixed-effects model suggest that gain/loss framing did exert a significant effect on PME ratings, such that participants tended to offer higher PME scores for gain-framed messages than for loss-framed messages (β = 0.616, F(1,60.099) = 34.382, P < 0.001).

We also found that gain/loss framing interacted with our present/future condition (β = 0.309, F(11 762) = 4.535, P = 0.041), such that the difference in PME ratings between gain-framed and loss-framed messages was larger for messages with present-oriented outcomes compared to future-oriented outcomes. Gain/loss framing also interacted with our personal/prosocial condition (β = 0.432, F(1,30.244) = 8.756, P = 0.006) such that the difference in PME ratings between gain-framed and loss-framed messages was larger for personal messages than for prosocial messages. Compete results are presented in Table 2 and illustrated in Figure 4.

Table 2.

Results of a linear mixed-effects model testing the effects of message features on perceived message effectiveness (PME)

	β	F	Df	Df.resid	P
Gain/loss(GL)	0.614	34.382	1	60.099	<0.001
Present-oriented/future-oriented (PF)	0.005	0.004	1	33.580	0.948
Personal/prosocial(PP)	0.026	0.094	1	43.766	0.760
GL*PF	0.312	4.535	1	30.243	0.041
GL*PP	0.434	8.755	1	30.244	0.006
PP*PF	−0.069	0.223	1	30.253	0.640
GL*PP*PF	−0.595	4.128	1	30.251	0.051

Note: Significant results indicate that PME ratings differed significantly across message subsets.

Fig. 4.

X-axis describes message category. Y-axis describes mean PME ratings by framing condition. Error bars represent 95% confidence intervals.

Association between PME and VMPFC activity

We tested the coefficients of our parametric modulation regressor to determine whether VMPFC activity during message encoding was associated with the subsequent PME ratings. Results indicated that activation within the VMPFC was positively associated with PME ratings (t(43) = 2.592, P = 0.013, 95% CI [0.005, 0.042]). In other words, participants showed more VMPFC activity while encoding messages they subsequently rated as effective.

Framing manipulation check

Finally, we checked whether the effect of gain/loss framing was driven by the images paired with the manipulated text in our stimuli by testing the fit of a linear mixed effects model against an independent sample in which the images were reversed. Results revealed an effect of gain/loss framing in the same direction as the fMRI findings (B = 0.437, F(1,237.02) = 58.9, P < 0.001), suggesting that differences in PME and VMPFC activity were driven by manipulation of the text rather than attributes of the images.

Discussion

To better understand the role of message features in the neural processing of persuasion, this study tested whether VMPFC activity during message encoding varied depending on whether messages framed outcomes as gains or losses. Past work found that certain gain-framed messages elicited more VMPFC activity than similar loss-framed messages (Vezich et al., 2017), but it remained unclear whether this effect would generalize to messages with other combinations of message features. Our results suggest this effect can be observed in response to a diversity of messages but may depend on whether those messages address outcomes relevant to individual persons or to society at large. We also found that gain/loss framing influenced participants’ ratings of perceived message effectiveness and that these ratings were positively associated with activity in the VMPFC. Taken together, our results align with theorizing that VMPFC activity during persuasive processing indexes a process of self-value integration, but they may suggest that this process only involves or depends on some assessment of personal risk.

Gain/loss framing affects VMPFC activity in messages about personal outcomes

Using an ROI modeled after past work (Vezich et al., 2017) that encompasses a region associated with successful persuasion (Falk et al., 2010, 2012), we observed that certain gain-framed messages elicited more VMPFC activity than similar loss-framed messages. This effect emerged regardless of whether messages framed outcomes in the present or in the future, but we did not observe this effect in response to messages advocating behaviors with prosocial outcomes. Further, a comparison of gain/loss framing effects across message conditions found stronger effects in one personal outcome condition (present personal) than in both prosocial outcome conditions. These results clarify the findings of Vezich et al., (2017), suggesting the effect reported in their paper may be generalizable to a diverse population of messages despite a previous failure to replicate (Burns et al., 2018). Further, they suggest that established theories of persuasion like prospect theory (Kahneman and Tversky, 1973; Tversky and Kahneman, 1989) may offer useful frameworks for explaining the relationships between the features of persuasive messages and the neural processes they evoke.

Our finding that gain/loss framing effects could only be observed in response to messages with personal outcomes has two possible explanations. First, the prosocial outcome messages may have contained some design features that were not intentionally manipulated but nonetheless interfered with the gain/loss framing effect. For example, it is possible that prosocial messages were generally of lower quality than personal messages. Comparison of message PME scores suggests this is unlikely: overall PME ratings did not differ significantly across the present-oriented/future-oriented or personal/prosocial message conditions. Although PME and VMPFC activation are distinct measures, the absence of any differences in PME ratings across message conditions suggests differences in activation are likely not due to major differences in message quality.

Another explanation is that the VMPFC’s sensitivity to gain/loss framing of messages depends on whether those messages address personal vs prosocial outcomes. The messages used in this study shared important similarities with those tested by Vezich et al., (2017), who also observed gain/loss framing effects during encoding of some messages but not others (‘why’-framed messages but not ‘how’-framed messages). All messages used in this study were ‘why’-framed and advocated for behaviors that, like sunscreen use, are considered low risk (increasing physical activity and ‘going green’). These similarities are appropriate given the tenets of prospect theory, which proposes that people are more attentive to gains than losses when they do not consider themselves to be at risk but more attentive to losses when they do feel at risk. Thus, the current study manipulated two message dimensions that were not expected to affect risk perceptions: the temporal orientation of the outcomes and the objects of those outcomes. Results suggest that gain/loss framing affects VMPFC activation regardless of whether a message focuses on the near-term or long-term outcomes, but it does not appear to have an effect when messages describe gains or losses that do not apply to their audience directly.

This latter explanation aligns with theorizing that the VMPFC’s role in persuasion processing is to incorporate perceptions of message value with considerations about the self. Vezich et al., (2017) proposed that gain-framed messages elicited more activity because they made the personal value of a behavior more explicit, making participants more likely to reflect positively on the prospect of incorporating the behavior into their self-concept. We suggest that our observations captured a similar process, but that the absence of effects in prosocial conditions might indicate a slightly different form of self-related processing. Given that prospect theory suggests framing effects depend on perceptions of relative risks and rewards, we propose that our personal/prosocial manipulation impacted some component of these risk perceptions. The absence of differences in VMPFC activity across present/prosocial conditions suggests that this activity does not index only immediate self-relevance, but such relevance might be an important antecedent to value judgments about message content. In this case, the ‘self’ processes included in self-value integration might involve some consideration not just of self-concept but of the outcomes themselves. For example, participants might have formed mental simulations of messages’ outcomes and shown VMPFC activity only with those simulations involved themselves personally.

Gain/loss framing affects perceptions of message effectiveness

Gain/loss framing also exerted an effect on PME ratings such that participants tended to rate gain-framed messages as more effective than loss-framed messages. We did not observe any direct effects on PME by other message features, but these features interacted with gain/loss framing so that its effects on PME were stronger when messages presented present-oriented and personal outcomes. Notably, this aligns with our comparisons of gain/loss framing effects on VMPFC activity across message conditions, which found that framing effects were significantly stronger during encoding of present-oriented personal messages than future-oriented prosocial messages. Thus, we found that gain/loss framing was a uniquely important factor in perceptions about a message’s effectiveness and that the way these perceptions varied across our message conditions aligned in some ways with variance in the gain/loss framing effects on VMPFC activity.

The link between the VMPFC and PME is further evidenced by the test of our parametric modulation regressor, which found that participants exerted more VMPFC activation when encoding messages they subsequently rated as more effective. This aligns with other work that has found activity in this region can predict not only value judgments but also assessments of personal significance. For example, when D’Argembeau et al., (2012) asked participants undergoing fMRI to reflect on their personal traits, they observed greater VMPFC activation when participants reflected on traits they considered personally or emotionally significant. Similarly, Lin et al., (2016) tested a parametric modulation regressor like our own to find VMPFC activity during recall of autobiographical memories tracked with the degree to which those memories involved things that participants valued or liked. Activity in this region has also been shown to track with value judgments of less personal stimuli, including monetary rewards (Shapiro and Grafton, 2020), making it to be widely considered a domain-general index of value (Bartra et al., 2013). Thus, our finding that VMPFC activity during message encoding tracks with subsequent PME ratings is consistent with past research suggests this region is involved in value judgments.

Suggestions for future research

These findings suggest several promising opportunities for future research. First, this work could investigate the possibility that VMPFC activity during persuasion involves or depends on some assessment of self-relevant risks. This is one potential explanation for our finding that the VMPFC was only sensitive to the gain/loss framing of messages that addressed personal outcomes, and it aligns with some components of prospect theory. Perceptions of risk are an important consideration of prospect theory, which posits that people are more attentive to gains when they perceive risks as low and more attentive to losses when they perceive risks as high. Thus, future work could both test our suggestion that risk assessments are important to self-value integration processes and further test the explanatory power of prospect theory by experimentally manipulating the degree to which messages are likely to induce perceptions of risk. Second, future work could incorporate independent measures of message success, such as message-consistent behavior or population-level message performance. This could help to further establish that gain-framed messages are more effective and that differences in VMPFC activation across gain- vs loss-framed messages suggests differences in the real-world effectiveness of those messages. Finally, testing of messages advocating for a variety of prosocial behaviors would provide context for our findings that differences in the gain/loss framing of messages do not tend to elicit the expected differences in VMPFC activation for messages of this type.

Conclusion

The current study found that gain-framed messages advocating for a low-risk behavior with personal outcomes elicited increased VMPFC activity compared to loss-framed messages. Participants also tended to rate these messages as more effective. These results align with those of past work (Vezich et al., 2017), contributing to nascent efforts to bridge neuroscientific research with established theories of persuasion. This suggests a promising future for research of this kind, which may soon allow neuroscientists to offer empirically supported recommendations for message design.

Funding

This study was funded by the University of Wisconsin-Madison Fall Research Competition AAG9877 and was supported in part by a core grant to the Waisman Center from the National Institute of Child Health and Human Development (P50HD105353).

Conflict of interest

The authors declared that they had no conflict of interest with respect to their authorship or the publication of this article.