After the experimental treatments (the control group received no prior information), we assessed belief in free will regarding people with a drug addiction. We therefore used seven adopted items (see Table 3) of the Free Will and Determinism (FAD-Plus) instrument (Paulhus and Carey, 2011), which is an enhancement of the preliminary FAD-4 version. As described above, the items were translated to German and back-translated to English, followed by a cognitive pretest. Participants were asked to think about people addicted to either alcohol or cocaine and under the influence of the respective substance and to rate these items on a 6-point scale ranging from “strongly disagree” [0] to “strongly agree” [5].

We used principal component factor analysis with oblique oblimin rotation (to allow the factors to correlate) to identify the dimensionality of the scale. The Kaiser-Meyer-Olkin Measure of 0.75 for the alcohol-subsample as well as for the cocaine-subsample indicated a good suitability of the data for structure detection. Based on this analysis, a two-factor solution was developed: for both subsamples, the items of the originally proposed one factor-solution were separated into a factor focusing on responsibility (FW_{RESPONSIBILITY;} items 1, 2, 3, with an eigenvalue above 2.83 for the alcohol subsample and 2.83 for the cocaine subsample) and one focusing on volition (FW_VOLITION; items 4, 5, 6, and 7, with an eigenvalue above 1.15 for the alcohol subsample and 1.33 for the cocaine subsample). Due to a satisfying internal consistency of the items within each of these factors (Cronbach‘s α for FW_{RESPONSIBILITY}: 0.78 for the alcohol subsample and 0.77 for the cocaine subsample; and for FW_VOLITION: 0.60 for the alcohol subsample and 0.67 for the cocaine subsample), we continued our analysis with these two factors.

The authors of the scale already mentioned that this scale assesses “assumptions about autonomy” and “declarations that people are responsible for their actions” (Paulhus and Carey, 2011, p. 97). The duality of the scale is also reflected in their remark that “free will beliefs are consistent with an internal locus of control but also include moral responsibility” (Paulhus and Carey, 2011, p. 99). Both factors are assumed to be conditions for free will (Lavazza and Inglese, 2015). For the following analysis, regression factor scores were used for each factor (the score 0 indicates an average attributed responsibility or volition, and 1 is the standard deviation), because usually some items are more important than others when explaining a certain construct. By using factor scores instead of unweighted sum scores, the different impacts of each item was accounted for (DiStefano et al., 2009). Descriptive results for this and the following two instruments are shown in Table 1. When describing the results we refer to the two factors as FW_{RESPONSIBILITY} and FW_VOLITION, while we use the generic concept of free will to refer to literature that has not differentiated both factors.

Respondents described their overall knowledge about neuroscience by responding to the following item: “My knowledge of neuroscience in general is…” with an 11-point scale ranging from “very low” [0] to “very high” [10].

An adopted version of the ultra-rapid screening for substance-use disorders (ASSIST-LITE) (Ali et al., 2013) was used to assess respondents' SUD of the two investigated substances, alcohol and cocaine. Based on this screening, respondents were either grouped as no alcohol-SUD (respectively no cocaine-SUD) [0] or as having a tendency toward an alcohol-SUD (respectively a cocaine-SUD) [1].

Furthermore, SUD among peers was assessed by asking whether the respondents know anyone who is addicted to the two substances under investigation (cf. Sorsdahl et al., 2012). Respondents were either grouped as not knowing anyone with an alcohol-SUD (respectively a cocaine-SUD) [0] or as knowing peers with an alcohol-SUD (respectively a cocaine-SUD) [1].

As shown in Table 4, no statistically significant main effects for neuroscience knowledge and the experimental treatment were found on the perceived FW_VOLITION of people with addiction to alcohol. However, we also saw that neuroscience-knowledge had an effect in the control group (see Model 2 in Table 5). Specifically, increased neuroscience-knowledge slightly augmented (p = 0.036) the attributed FW_VOLITION of people with addiction to alcohol (see ascending dotted gray line in Panel B in Figure 2). However, increasing neuroscience-knowledge resulted in an opposite pattern in the neuroimage-only-treatment: here, increasing neuroscience-knowledge led to slightly lower perceived FW_VOLITION of people with addiction to alcohol (p = 0.035)⁵.

No statistically significant overall effect for neuroscience-knowledge was found in the model on FW_VOLITION of people with addiction to cocaine without the interaction effects—as is visible in Table 4. Nonetheless, Model 4 in Table 5 (see also Panel D in Figure 2) show that, in the control group, increasing neuroscience-knowledge resulted in a higher attribution of FW_VOLITION to people with addiction to cocaine (p < 0.001). This effect of neuroscience-knowledge significantly differed for the other three experimental treatments, i.e., for the text-only-treatment (p = 0.001) and the image-only-treatment (p = 0.007) the respective lines were almost parallel to the x-axis, indicating no moderating effect of neuroscience-knowledge, and for the text and neuroimage-treatment an increase in neuroscience-knowledge slightly decreased the attributed FW_VOLITION of people with addiction to cocaine (p < 0.001).

In Table 4, we reported a negative main effect of neuroscience knowledge on the attributed FW_{RESPONSIBILITY} of people with an addiction to cocaine. As Model 3 in Table 5 and Panel C in Figure 2 show, by analyzing the interaction between neuroscience-knowledge and the experimental treatments, we found that increasing neuroscience-knowledge reduced FW_{RESPONSIBILITY} only for the neuroimage-only-treatment (p = 0.023). This interaction effect significantly differed from the interaction effect between neuroscience-knowledge and text-only-treatment (confirmed by a post-estimation Wald test, p = 0.036): the results show that no statistically significant effect of neuroscience-knowledge was found for the text-only-treatment. However, a post-estimation Wald test also indicated that for those respondents with the lowest value of neuroscience-knowledge, the attributed FW_{RESPONSIBILITY} was significantly higher in the neuroimage-only-treatment compared to the text-only-treatment (p = 0.030).

One possible interpretation is that neuroscience discourse and neuroimages simply do not carry the effects that both proponents and opponents of the brain-disease model of addiction have claimed. The literature on the seductive allure of neuroimaging, which offers a specific experimental context where the impact of neuroscience information has been investigated, seems to be pointing in that direction (Schweitzer and Saks, 2011; Schweitzer et al., 2011, 2013; Greene and Cahill, 2012; Gruber and Dickerson, 2012; Farah and Hook, 2013; Michael et al., 2013). Two reviews (Farah and Hook, 2013; Michael et al., 2013) and several other studies have now failed to replicate the original findings of McCabe and Castel (Gruber and Dickerson, 2012; Farah and Hook, 2013). One study has suggested that, based on the use of different types of neuroimaging information (e.g., inflated brain and whole brain images were more convincing), the perceived complexity of a neuroimage, rather than its familiarity or its resemblance to a real brain, could contribute to swaying beliefs about scientific explanations (Keehner et al., 2011). However, this study did not include a control group (e.g., no image or no brain image) and did not assess the initial effect reported by McCabe and Castel. In addition, a few studies have examined the impact of neuroimaging evidence on jurors, but the results are divided with some showing effects (Gurley and Marcus, 2008; McCabe et al., 2011; Ikeda et al., 2013) and others not (Schweitzer and Saks, 2011; Schweitzer et al., 2011; Greene and Cahill, 2012).

However, in contrast to the literature on the effects of neuroimages, the literature on the seductive allure of textual neuroscience explanations (Weisberg et al., 2008, 2015; Michael et al., 2013; Rhodes et al., 2014; Scurich and Shniderman, 2014; Fernandez-Duque et al., 2015; Rhodes, 2015) has so far evidenced a more robust effect, consistent with the original description of the phenomenon of neuro-realism in textual (print media) forms (Racine et al., 2005). Weisberg et al. have recently found that the length of the explanation could modulate the effects of the presence of textual neuroscience explanations, although this did not fully explain the effect. In contrast, the level of complexity (amount of jargon) in a textual neuroscience explanation does not appear to change the general effect (Weisberg et al., 2015). Interestingly, sample differences between undergraduate students and MTurk workers were noted, with the students generally attributing lower scores for neuroscience explanations, perhaps because the educational setting and its emphasis on critical thinking could be at stake (Weisberg et al., 2015). However, in one of the experiments in this study, students in psychology seemed particularly swayed by neuroscience explanations when the explanations were bad (Weisberg et al., 2015). The replication study by Fernandez-Duque et al. (2015), which included both textual and neuroimaging information found only effects for textual information, suggesting that the effect of neuroscience is conceptual (i.e., textual information on brain research best explains mental phenomena) and not pictorial (i.e., based on the representation of the brain per se). In contrast to this body of positive results, we found very limited statistically significant effects of neuroimaging in combination with the text (see results for research question 3).

Finally, consistent with our findings, recent studies on the impact of the brain disease model of addiction have not found strong effects of this model on attributions of stigma and blame, which should be reduced if attribution of free will is diminished by neuroscience information (Meurk et al., 2014b,c; Sattler et al., 2017). Alongside our own results, these recent findings suggest that the previous debate on the effects of the brain disease model of addiction on free will has potentially been overdone, at least in terms of the actual effects of neuroscience discourse on attribution of free will. The debate may have reflected the strong stances of the authors (Hall et al., 2015; Racine et al., 2015) and not necessarily of the general public, which does not seem to be swayed by neuroscience information.

Another possibility is that the effects of neuroscience information interact with pre-existing beliefs about the phenomenon at hand such as beliefs about the biological basis of addiction and its relationship to belief in free will. Scurich and Shniderman have found effects of motivated reasoning: there is a seductive allure of textual neuroscience information when the information confirms prior beliefs (Scurich and Shniderman, 2014). This finding is distinct from studies about motivated reasoning on the study of attributions of free will (Clark et al., 2014). Likewise, in the context of addiction, an individual already adhering to a brain-disease model of addiction could find neuroscience information more convincing while someone not adhering to a brain-disease model could find neuroscience information less compelling, and even repulsive. Since we did not survey pre-existing beliefs about the biological aspects of addiction, we cannot address such a possibility directly. However, other recent studies (Meurk et al., 2014b,c) have suggested that the general public does not strictly adhere to a brain-disease model of addiction. Accordingly, neuroscience discourse in the context of addiction may not have the effects that were initially predicted on stigma reduction (Dackis and O'Brien, 2005). It is possible that the understanding of addiction as brain disease and even as a disease may be a simplification and a form of reductionism not representative of public opinion, at least in Australia where that study was conducted (Meurk et al., 2014c). In sum, the existence of motivated reasoning cannot be ruled out.

In spite of the unlikeliness of a general phenomenon of seductive allure of neuroscience explanations, our study—and several others—leaves the possibility that higher-stake situations in real-world settings could elicit such a seductive allure effect. For example, the textual information we used, although found in a credible source, may not have sufficiently emphasized the biological basis of addiction to elicit effects. One could retort that the text was a piece of naturally occurring public discourse and that, even though more hyperbolic discourse could have greater effects, hyperboles could be easily criticized for their strong and artificial overtones. It is possible that neuroscience information, if it is emphasized or plays a more significant rhetorical role (e.g., to validate a discourse or provide additional confirmatory evidence) could play a more consequential role than found in our study. This would be consistent with the apparently significant influence that neuroimaging can have on patients and on clinical practices. For example, the use of imaging for back pain has increased dramatically (Chou et al., 2011), despite its debated clinical legitimacy, thus suggesting a significant influence of imaging on clinical practices. However, the discourse in which such imaging evidence is embedded (e.g., helps locate and visualize the pain) (Rhodes et al., 1999) could have considerable impact because it fits in a narrative where evidence is sought to confirm one's initial suspicion about the locus of pain (i.e., confirmation bias or “motivated reasoning”) (Scurich and Shniderman, 2014). Accordingly, in such a context, neuroimaging evidence could help to convince patients to side with medical opinions in favor of surgery or, in the context of addiction, neuroimaging could support certain types of biologically-grounded treatments for addiction to the detriment of more socially-grounded approaches (Hall, 2006; Dingel et al., 2011; Hall et al., 2015). We can thus speculate that in real-world contexts in which neuroimaging evidence is introduced, neuroscience information could play a greater role than found in more hypothetical settings. Furthermore, the fact that we partially found different effects for the substances at stake (cocaine and alcohol) could be an indication that the effects of a brain disease model of addiction need to be examined more specifically and with greater attention to different substance types (Buchman et al., 2010; Meurk et al., 2014a; Carter et al., 2016) or more generally to other phenomena that people can get addicted to and what these addictions imply.

In the future, it will be important to assess the effects of neuroscience information in naturally occurring discourse such as in the discourse of healthcare providers and its impact on patients or application in marketing approaches (e.g., promotional videos about addiction treatment on consumers using neuroimaging) in real social settings. For example, if the stakes of believing in neuroimaging are higher and aligned with other interests (e.g., seeking a diagnosis or insurance claims for addiction treatment), then perhaps they could play a larger role in attributions of free will than for those who have no such vested interest. In other words, the prior interests that one has in believing the credibility of neuroscience information may shape attitudes more than the information itself.

Much like our findings about the greater impact of respondent characteristics than neuroscience explanations on attribution of free will in addiction, the literature on the brain disease model of addiction suggests that respondent characteristics shape attitudes toward addiction to a greater extent than adherence to a brain disease model of addiction (Sattler et al., 2017). For example, women and older respondents have been identified as believing more that addiction is a disease (Meurk et al., 2014c). Being older and having more education (>15 years) was associated with less support for coerced treatment, and being older also predicted a lack of support for punishment by imprisonment (Meurk et al., 2014b). However, these last attitudes were not predicted by beliefs that addiction was a disease or a brain disease. Accordingly, like in our study, basic respondent characteristics appeared to be stronger predictors of attitudes toward addiction than adherence to the brain disease model in these studies (Meurk et al., 2014b,c).

Interestingly, a broader literature on attitudes toward stigma and blame in addiction also suggests that respondent characteristics play a role. We readily recognize the gap between the literature on belief in free will and the literature on stigma, but it is important to keep in mind that belief in free will, as operationalized in the FAD-Plus scale, measures responsibility, whereas a common measure of stigma, the attribution questionnaire (AQ), has for one of its explicit dimensions “blame,” i.e., that “people are responsible for and can control their mental illness” (Corrigan et al., 2003; Corrigan, 2012). In fact, there are possible parallels between findings based on the AQ (measuring stigma) and the FAD-Plus scale (measuring belief in free will). For example, we found that certain characteristics have been associated with less attribution of FW_{RESPONSIBILITY}, such as being female (for cocaine addiction), more educated (for cocaine and alcohol), more self-reported knowledge of neuroscience (for cocaine and alcohol), reported knowledge of a SUD (for alcohol), have also been reported to lower blame toward people with an addiction (Sattler et al., 2017). As our study and the literature on the brain disease model of addiction and stigma associated with addiction suggest, factors such as respondent characteristics may have important effects on belief in free will. It is also possible that the effects of respondent characteristics were generated by the fact that we explored belief in free will in the specific context of addiction. Yet, at the same time, this would likely mean that other specific contexts (e.g., investigating the effects of belief in free will in everyday moral behavior, in the determination of criminal responsibility, in health behavior, etc.) could carry with them a series of context-sensitive beliefs and assumptions that are socially embedded or topic-specific, and where respondent characteristics play a role. Perhaps these characteristics could even have different roles in these different contexts. In this light, the interaction effects we found between belief in free will and self-reported knowledge about neuroscience (discussed below) could be a telling example of this potential social embeddedness of the seductive allure of neuroscience explanations.

In the context of research on the influence of neuroscience information, knowledge about neuroscience has been reported to have important effects in some studies while other respondent characteristics have typically not been found to lead to differences. For example, McCabe and Castel reported that the effects of neuroimaging evidence on the ranking of scientific explanations was particularly noteworthy for novice individuals but not for experts (McCabe and Castel, 2008). However, a replication study comprising 10 experiments (Michael et al., 2013) did not find differences regarding the effects of neuroimages related to education (and also age). Our analysis of interactions between the experimental treatments and the level of neuroscience-knowledge yields a complex picture that partially challenges the relationship found by McCabe and Castel but, more importantly, yields more specific questions to tackle.

Overall, our results suggest that increased neuroscience-knowledge results in lower attribution of FW_{RESPONSIBILITY} for people addicted to cocaine in the neuroimage-only-treatment. However, according to McCabe and Castel's study, one would perhaps expect neuroimaging to decrease attribution of FW_{RESPONSIBILITY} in novices but not in experts. However, we found that FW_{RESPONSIBILITY} was decreased in those who are more knowledgeable. This comparison assumes that neuroscience information undermines belief in free will in the context of addiction as suggested by Vohs and Baumeister (Vohs and Baumeister, 2009) and assumes also that what we measured as the level of neuroscience-knowledge can be mapped to concepts of novice and expert as deployed by McCabe and Castel. Indeed, most of our more knowledgeable participants remained nevertheless novices, see Table 1.

Despite these caveats, one possible interpretation of our findings is that greater emphasis on the biological underpinnings of addiction may lead to the belief, in the eyes of those with more neuroscience knowledge, that the person has less control over the addiction as predicted by proponents of the brain-disease model of addiction (Dackis and O'Brien, 2005). However, this effect is also the basis of the worries captured by critiques of the brain-disease model of addiction: namely that neuroscience information about addiction can actually exacerbate blame and stigma because the person with an addiction is seen as less able to take care of him/herself (Hammer et al., 2013; Szott, 2015). As a result of such beliefs, the person with an addiction can be considered passive and powerless, and relinquish his/her decision-making capacity to others such as healthcare professionals or state authorities (Gartner et al., 2012; Racine et al., 2015). For example, a qualitative study found that the brain disease model is integrated into compassionate care but nevertheless risks downplaying the autonomy of those with an addiction (Szott, 2015).

A similar interpretation could help make sense of our findings about FW_VOLITION. We found that increasing neuroscience-knowledge resulted in increasing attribution of FW_VOLITION (alcohol and cocaine) in the control group which perhaps reflects other findings suggesting that higher education is associated with increased beliefs in volition and the ability for self-determination or the sheer valuing of autonomy (Ryan and Deci, 2000; Say et al., 2006; Zizzo et al., 2017). However, we found no such trend in the other treatments. Results for these other treatments suggest that some of them (only consistently shown in the neuroimage-only-treatment) partially undermine higher attribution of FW_VOLITION associated with increased neuroscience-knowledge. Perhaps neuroscience information induces a deterministic view that diminishes attribution of FW_VOLITION in those with greater knowledge of neuroscience just like it reduces attribution of responsibility. These effects would fit in the broader context where both academic (Saigle et al., 2017) and public discussions (Racine et al., 2017) of neuroscience research about volition tend to be casted in deterministic overtones such that the actual existence of free will is seriously questioned.

In sum, these interaction effects between self-reported knowledge of neuroscience and belief in free will are intriguing even if a comparison with previously published results is difficult because of the different measures used to capture constructs such as level of education or level of neuroscience knowledge (e.g., self-reported knowledge of neuroscience in our study, level of advancement in a graduate program for McCabe and Castel). Nonetheless, these interactions suggest that the study of a phenomenon like the effects of neuroscience explanations is complex and needs to carefully integrate information about the context in which the prompting information is introduced (e.g., addiction, substance, or another context) and about the person considering this information (e.g., based on respondent characteristics such as neuroscience knowledge). Our observations also support the need for new, more refined scales to clearly tease apart the constructs of volition and responsibility, such as the Free Will Inventory which specifically recognizes this problem and proposes a cleaner measure of free will per se (separated from moral responsibility) as well as the ability to examine the relationships between these important constructs (Nadelhoffer et al., 2014).