The acceptability of behavioural interventions in financial decision-making
Policymakers are increasingly applying behavioural insights and instruments to facilitate, influence or steer the behaviour of the public. Behavioural interventions are recognised as particularly useful for financial policymakers, regulators and others interested in improving financial decision-making (Frydman and Camerer, Reference Frydman and Camerer2016; Lefevre and Chapman, Reference Lefevre and Chapman2017). This is because ‘Financial decisions are among the most important life-shaping decisions that people make.’ (Frydman and Camerer, Reference Frydman and Camerer2016, p. 661), yet there are many well-documented ways in which consumers deviate from optimal decision-making when managing their money (Frydman and Camerer, Reference Frydman and Camerer2016; Beshears et al., Reference Beshears, Choi, Laibson and Madrian2018).
It is also increasingly recognised that understanding people's attitudes towards policy interventions (behavioural or otherwise) is critical for their successful implementation (Page and Shapiro, Reference Page and Shapiro1983; Tannenbaum et al., Reference Tannenbaum, Fox and Rogers2017). In recent years, the number of studies investigating people's attitudes towards behavioural policy interventions has grown rapidly (Michaelsen, Reference Michaelsen2021). However, as we will discuss in the next section, these studies do not provide much insights into the acceptability of behavioural interventions in financial decision-making. With the present research, we aim to partially fill this gap in our understanding.
Overview of previous research
We started with searching for relevant research publications concerning the acceptability of behavioural interventions. We searched Web of Science and Google Scholar for English-language papers (published, working papers and pre-prints) with variations on the search terms acceptability, support, behavioural intervention and nudge. We also used the references and citations of relevant literature to identify other relevant studies.
We selected studies that measure support for, or acceptability of, interventions as a dependent variable, and that include at least one behavioural intervention. These studies range from surveys where people are asked to rate the support for or acceptability of a number of interventions, to experiments with scenarios or descriptions in which one or more aspects of the intervention are manipulated. Support and acceptability refer to people's responses to questions like ‘Do you find the described policy acceptable?’ (Hagman et al., Reference Hagman, Andersson, Västfjäll and Tinghög2015, p. 444) or ‘Would you support a policy of … ?’ (Jung and Mellers, Reference Jung and Mellers2016, p. 66). We identified 32 relevant studies. Details on all 32 studies can be found in Supplementary Table A1 in Appendix 1. Below, we report the main findings from this literature review.
What is known about the acceptability of behavioural interventions in financial decision-making?
None of the studies in our literature review focused exclusively on financial decision-making. In contrast, 11 focused exclusively on health-related behaviours (of which eight on healthy eating). When an example of a financial intervention was included in a study, this was most commonly a pension enrolment default. Thus, little is known about the acceptability of behavioural interventions in financial decision-making.
In the studies conducted so far, behavioural interventions commonly received higher support and acceptability scores than more intrusive and/or costly interventions like bans or taxes but lower support and acceptability scores than purely information- or education-based interventions. Djupegot and Hansen (Reference Djupegot and Hansen2019) noted that the high support for behavioural interventions found in many studies could very well be the result of the focus on health in general, and healthy eating in particular, rather than of high support for the interventions per se. They argue that healthy eating is a less controversial policy goal than many others because its benefits are widely accepted at both the individual and societal level. This argument is supported by Gold et al. (Reference Gold, Lin, Ashcroft and Osman2023), who compared acceptability across several domains and found that the acceptability of their financial intervention was significantly lower than the acceptability of various health interventions. As we will discuss below, our own data support this argument as well.
Agent, perceived effectiveness and acceptability of behavioural interventions
Our review of the literature revealed two important factors associated with the acceptability of interventions: the agent of the intervention (i.e., who is responsible for the intervention) and the perceived effectiveness of the intervention (cf. Banerjee et al., Reference Banerjee, Savani and Shreedhar2021).
Agent
Six studies (Bos et al., Reference Bos, Van Der Lans, Van Rijnsoever and Van Trijp2015; Tannenbaum et al., Reference Tannenbaum, Fox and Rogers2017; Arad and Rubinstein, Reference Arad and Rubinstein2018; Evers et al., Reference Evers, Marchiori, Junghans, Cremers and De Ridder2018; Bang et al., Reference Bang, Shu and Weber2020; Krisam et al., Reference Krisam, Maier, Janßen and Krisam2021) have identified an intervention's agent as one of the factors influencing acceptability. Whether and how to disclose the agent of an intervention is a relevant question for policymakers (cf. Tannenbaum et al., Reference Tannenbaum, Fox and Rogers2017). Five of these studies (Bos et al., Reference Bos, Van Der Lans, Van Rijnsoever and Van Trijp2015; Tannenbaum et al., Reference Tannenbaum, Fox and Rogers2017; Arad and Rubinstein, Reference Arad and Rubinstein2018; Bang et al., Reference Bang, Shu and Weber2020; Krisam et al., Reference Krisam, Maier, Janßen and Krisam2021) use an experimental design and can provide causal evidence on the link between agent and acceptability. The results are mixed. Most studies found lower support for interventions implemented by governments than by other agents like firms and experts (Bos et al., Reference Bos, Van Der Lans, Van Rijnsoever and Van Trijp2015; Arad and Rubinstein, Reference Arad and Rubinstein2018; Evers et al., Reference Evers, Marchiori, Junghans, Cremers and De Ridder2018, Krisam et al., Reference Krisam, Maier, Janßen and Krisam2021). Gold et al. (Reference Gold, Lin, Ashcroft and Osman2023) found similar effects, but they considered the effect size too small to be relevant. On the other hand, Bang et al. (Reference Bang, Shu and Weber2020) argued that the perceived motives of the agent implementing the policy are more important than the identity of the agent, and found that perceived non-profit motivations are more beneficial for acceptability than financial motivations.
Perceived effectiveness
The most consistent finding in studies on support for behavioural interventions is the positive association between the perceived effectiveness of policy interventions and their acceptability. Eleven of the 28 studies have identified perceived effectiveness as a (and commonly the) main factor associated with acceptability (Cornwell and Krantz, Reference Cornwell and Krantz2014; Bos et al., Reference Bos, Van Der Lans, Van Rijnsoever and Van Trijp2015; Jung and Mellers, Reference Jung and Mellers2016; Petrescu et al., Reference Petrescu, Hollands, Couturier, Ng and Marteau2016; Hall et al., Reference Hall, Marteau, Sunstein, Ribisl, Noar, Orlan and Brewer2018; Cadario and Chandon, Reference Cadario and Chandon2019; Djupegot and Hansen, Reference Djupegot and Hansen2019; Reynolds et al., Reference Reynolds, Archer, Pilling, Kenny, Hollands and Marteau2019; Bang et al., Reference Bang, Shu and Weber2020; Davidai and Shafir, Reference Davidai and Shafir2020; Gold et al., Reference Gold, Lin, Ashcroft and Osman2023).
In all cases, this evidence is correlational, and very few studies provide insight into the underlying mechanisms. For example, ample studies in communication research have documented a ‘Third Person Effect’ (hereafter TPE) (see, for a review, Perloff, Reference Perloff1999). Davison (Reference Davison1983) coined this term for the belief that others are more easily persuaded or manipulated than they are. In our literature review, we found only three studies examining this TPE and its relationship with acceptability. Bos et al. (Reference Bos, Van Der Lans, Van Rijnsoever and Van Trijp2015) and Bang et al. (Reference Bang, Shu and Weber2020) documented a large and significant TPE. Bos et al. (Reference Bos, Van Der Lans, Van Rijnsoever and Van Trijp2015) concluded that both perceived influence on oneself and perceived influence on others are positively associated with acceptability. Bang et al. (Reference Bang, Shu and Weber2020) found a positive association for both measures of perceived influence in Study 2, but in Study 1 only perceived influence on others predicted acceptability. Gold et al. (Reference Gold, Lin, Ashcroft and Osman2023) found a positive association between perceived influence and acceptability, as well as a TPE for four out of five tested behavioural interventions. In reporting the relationship between perceived influence and acceptability, they did not distinguish between perceived influence on themselves and perceived influence on others.
Hypotheses tested in the present study
The present study investigates the acceptability of several interventions in financial decision-making. In doing so, we took a closer look at the relationship between agent and acceptability because of the mixed results in previous studies and because of its practical relevance for practitioners. Specifically, we examined whether disclosing the agent as a financial company or as policymakers influenced acceptability. Given the importance of non-profit motivations in previous work (Bang et al., Reference Bang, Shu and Weber2020), we hypothesised that acceptability would be higher when the agents are policymakers compared to financial companies (Hypothesis 1).
Our second hypothesis focused on the main factor associated with support in previous studies: perceived effectiveness. Particularly, we distinguished between perceived effectiveness in influencing your own decisions (hereafter: perceived self-effectiveness) as opposed to the decisions of other people (hereafter: perceived other-effectiveness). In line with the literature on TPE, we expected that perceived self-effectiveness would be lower than perceived other-effectiveness (Hypothesis 2). We also hypothesised that the association between perceived self-effectiveness and acceptability would be positive (Hypothesis 3). This hypothesis was based on the findings of Bos et al. (Reference Bos, Van Der Lans, Van Rijnsoever and Van Trijp2015), Gold et al. (Reference Gold, Lin, Ashcroft and Osman2023) and Bang et al. (Reference Bang, Shu and Weber2020). Finally, we explored the relationship between perceived other-effectiveness and acceptability. Based on the same previous studies (Bos et al., Reference Bos, Van Der Lans, Van Rijnsoever and Van Trijp2015; Gold et al., Reference Gold, Lin, Ashcroft and Osman2023; Bang et al., Reference Bang, Shu and Weber2020), we expected to find a relationship but were agnostic about its direction (Hypothesis 4). We report how we determined our sample size, all data exclusions, all manipulations and all measures included in our study (Simmons et al., Reference Simmons, Nelson and Simonsohn2012).
Method
Participants
This study was run in the consumer panel of the Dutch Authority for the Financial Markets (AFM) in April 2018.Footnote 1 We used this panel because it provides insights into the opinions of consumers with a strong interest in financial decision-making. Panel members are often highly motivated to complete studies and are more opinionated about financial matters than the average Dutch person. In this regard, our sample is different from the convenience or nationally representative samples commonly used in studies on the acceptability of behavioural interventions. Compared with the general Dutch population, panel members are also relatively wealthy. The sample size was based on the number of available panel members. Participants who gave the same answer to every question about all tested interventions were removed from the analysis (n = 12).Footnote 2 In all cases, these participants replied “neutral” to all questions about the interventions. The remaining sample consisted of 684 participants (17.3% women, M age = 61.6, SD = 11.2).
Design
The study had a two-group between-subject design. In the policymaker condition, the descriptions of the interventions stated policymakers as the agents. In the financial companies condition, the descriptions stated financial companies as the agents. Hypotheses 2, 3 and 4 were tested with within-subject analyses.
Procedure
Participants first read an introductory text, stating that we were interested in people's opinions about various kinds of instruments to guide behaviour. Next, the participants were confronted with seven interventions in random order for each participant. We provided participants with a general description of each intervention and then described an application in the context of financial decision-making. We used the neutral term “instruments” throughout the study and refrained from using technical or specialist language like “nudges” or “behavioural interventions”. Table 1 provides all interventions as described and the financial applications used. These interventions were chosen based on our review of the literature and internal discussions at the AFM. They are all examples of commonly studied interventions with relevance for the Dutch financial market context and with varying levels of intrusiveness (Evers et al., Reference Evers, Marchiori, Junghans, Cremers and De Ridder2018).
Table 1. Overview of interventions used in the present research
After each example, participants were asked to indicate whether they thought this particular application was a good idea or not (1 = very bad idea, 5 = very good idea). This question was our single-item acceptability measure.
Next, participants rated the instrument as a general instrument for guiding choices. This procedure was closely based on Tannenbaum et al. (Reference Tannenbaum, Fox and Rogers2017) and included to investigate if respondents distinguish between judging the instrument and judging the underlying policy objective. We reminded participants that policymakers (in the policymaker condition) or financial companies (in the financial companies condition) can use the instrument for different purposes. Respondents then replied to four Likert-type items (1 = completely disagree, 5 = completely agree): ‘Use of this instrument by policymakers OR financial companies in guiding choices is acceptable/is manipulative/evokes feelings of resistance’, and ‘I experience use of this instrument by policymakers OR financial companies in guiding choices as a threat to freedom of choice’. Items 2, 3 and 4 were reverse scored, so that higher numbers indicate a more positive attitude. The responses were averaged to create an acceptability index (Cronbach's α's ranging from 0.89 to 0.92). This acceptability index correlated substantially with the single-item acceptability measure discussed above (r's = 0.65 to 0.77). We therefore focus on the analysis for the acceptability index in the Results section below. Results for the single-item acceptability measure can be found in Supplementary Appendix Table A2. They are very similar, with a few nuances that we discuss in the relevant paragraphs of our Results section.
We then asked respondents to indicate via two questions: how much influence the instrument would have on their own decisions and how much influence it would have on other people's decisions (1 = very little influence, 5 = very much influence). Respondents then moved on to the next instrument, which was followed by the same set of questions, until they had completed questions for all seven interventions. As a final question, we asked people to state their political preference (1 = very left of centre, 5 = very right of centre, 6 = I don't know/don't want to say). Gender and age were available for all panel members and added to the dataset.
Results
How high, or how low, is the acceptability of behavioural interventions in financial decision-making?
We explored the difference in perceived acceptability of the different interventions across the two agents of intervention (policymaker versus financial company; see Table 2). The means for the acceptability index ranged from 2.83 to 3.67. The information-based interventions received the highest support, whereas the social norms interventions received the lowest support.Footnote 3 Contrary to expectations, the average perceived acceptability in the policymaker condition is similar to the one in the financial company condition across all the interventions, with the exception of the default intervention, where acceptability is higher in the financial company condition. This difference (3.14 versus 3.32) is significant, but in our view too small to be practically meaningful. We formally test Hypothesis 1 using a regression model where we also account for other factors.
Table 2. Main descriptives of the acceptability index by the agent of intervention
Note: The acceptability index runs from 1 to 5 with higher numbers indicating more acceptability.
The TPE: Is perceived self-effectiveness smaller than perceived other-effectiveness?
In order to test Hypothesis 2, we used paired-sample t-tests on the differences between perceived self- and other-effectiveness. We used the Bonferroni correction to correct for the increased risk of a type-I error due to multiple comparisons. The adjusted threshold for significance is obtained by dividing our main threshold of 5% by 7 (the number of t-tests we perform), leading to a threshold for statistical significance of 0.71% for Hypothesis 2.
Table 3 presents the means and standard deviations for perceived self-effectiveness and perceived other-effectiveness across the seven interventions. We observed a clear pattern of responses. Consumers perceived all interventions to have, on average, a greater influence on others' decisions than on their own decisions, and paired-sample t-tests confirm the difference is statistically significant for all interventions, which provides support for Hypothesis 2.
Table 3. Ratings of perceived self-effectiveness and perceived other-effectiveness
Note. Perceived self- [other-]effectiveness is the answer to the question how much influence each intervention would have on their own [other people's] decisions (1 = very little influence, 5 = very much influence).
Which factors are associated with the acceptability of behavioural interventions in financial decision-making?
In order to test Hypotheses 1, 3 and 4, we regressed the perceived acceptability index of each intervention on the experimental condition for the agent of the intervention together with perceived self-effectiveness and perceived other-effectiveness. We controlled for age and gender. In several previous studies (e.g., Reisch and Sunstein, Reference Reisch and Sunstein2016, see Supplementary Appendix Table A1), being female was associated with higher acceptability scores, whereas the association between age and acceptability was inconsistent (e.g., Sunstein et al., Reference Sunstein, Reisch and Kaiser2018a, Reference Sunstein, Reisch and Rauber2018b, see Supplementary Appendix Table 1). We find inconsistent associations between age, gender and acceptability. In addition, we examined the possibility that the participants’ political orientation is associated with the acceptability of an intervention; the addition of political orientation does not substantially change the results (see Supplementary Appendix Table A3).
Because each participant evaluated all seven interventions, we estimated a seemingly unrelated regression (SUR) model, which takes into account the correlations between the evaluations of interventions within participants. In other words, we estimated a system of linear multiple regression equations where the error terms are correlated across the equations (Greene, Reference Greene2002). This approach was necessary because the correlations between the error terms are substantial, as shown by a formal Breusch–Pagan test (χ 2(21) = 1020.25, p < 0.001). The results of the SUR model are shown in Table 4. We used an α value of 0.05 to interpret the results. Excluding the control variables does not change the conclusions.
Table 4. SUR model results for the seven interventions (main model, N = 684) (standard errors are reported in parentheses)
Hypothesis 1, which stated that the perceived acceptability of an intervention is higher in the policymaker condition than in the financial company condition, was rejected across all interventions except the default intervention. For our default intervention, acceptability was higher in the financial company condition.
Hypothesis 3, in which we expected a positive relationship between perceived self-effectiveness and acceptability, was supported for all the interventions. The coefficients across the interventions varied between 0.214 (for the intervention Active Choice) and 0.374 (for the intervention Social Norms).
Hypothesis 4, which stated that perceived other-effectiveness is associated with the acceptability of an intervention, was supported for three of the interventions: Knowledge and Information, Social Norms, and Presentation of Options. The relationship is positive for the Knowledge and Information intervention. In the model with the single-item acceptability measure (see Supplementary Appendix Table A2), there is an additional positive association between perceived other-effectiveness and acceptability for the Reminders intervention. The relationship between perceived other-effectiveness and acceptability is negative for the Social Norms and Presentation of Options interventions. For these two interventions, higher perceived other-effectiveness was associated with lower acceptability. These are also the interventions with the lowest average acceptability ratings (see Table 2).
Discussion
We examined the acceptability of seven different behavioural interventions in financial decision-making. In particular, we looked at the role of two previously established factors related to acceptability: the agent of the intervention (experimentally manipulated as policymaker or financial company) and the perceived effectiveness of the intervention (measured) in influencing decisions.
The acceptability of behavioural interventions in financial decision-making
We find that behavioural interventions in financial decision-making received mean acceptability index ratings from 2.83 to 3.67 on a 5-point scale. This appears to be somewhat lower than is typically found in studies on the acceptability of behavioural policy interventions in other domains (specifically health), where support for the most popular (information-based) intervention is often close to the scale maximum. We interpret these results as a tentative confirmation of the domain-specific nature of acceptability (as previously argued by Gold et al., Reference Gold, Lin, Ashcroft and Osman2023). This implies that academics and practitioners should be hesitant to generalise findings about acceptability from one domain to the other.
Similar to previous studies, the information-based interventions received the highest support. But perhaps, surprisingly, given the results from previous studies, support for defaults and active choice – interventions considered to be intrusive (Evers et al., Reference Evers, Marchiori, Junghans, Cremers and De Ridder2018) – was above the scale mid-point.
The role of agents
Contrary to our expectations, the experimental manipulation (disclosing policymakers or financial companies as the agents of the intervention) did not impact acceptability ratings in any practically meaningful way. One possible explanation is the composition of our sample. Specifically, their above-average familiarity with the financial sector might mean our sample trusts financial companies more than the average Dutch citizen. However, people who trust financial companies more, also tend to have higher trust in supervisory authorities (van der Cruijsen et al., Reference van der Cruijsen, de Haan and Roerink2021). Therefore, we believe a more likely explanation is that trust in both financial firms and policymakers in The Netherlands was at very low levels at the time of data collection (and still is at the time of writing) (Central Bureau for Statistics, 2023). It is not necessarily intuitive for people working in government, or in government-adjacent organisations, that citizens question or distrust their motives, and we believe a main policy implication of our findings is that policymakers should make sure to clearly explain the intentions behind a specific policy intervention. For researchers, this implies that it is important to include direct measures of trust and/or perceived motivations when studying the relationship between agent and acceptability.
Perceived effectiveness and acceptability
As in previous studies, the factor most strongly associated with acceptability was the perceived effectiveness of the intervention. We distinguished between perceived self-effectiveness and perceived other-effectiveness. As hypothesised, we documented a TPE for all our interventions. We also found a strong and positive association between perceived effectiveness and acceptability for all interventions – but only for the perceived effectiveness of own decisions. The relationship between perceived influence and others' decisions was inconsistent. In our main model, the correlation between perceived other-effectiveness and acceptability was not significant for the majority of our interventions. For the Knowledge and Information intervention, we found a positive (but weak) association, whereas we found a strong negative association between perceived other-effectiveness and acceptability for two interventions: Social Norms and the Presentation of Options. This finding stands in marked contrast to other studies that made the distinction between self- and other-effectiveness and found positive relationships for both (Bos et al., Reference Bos, Van Der Lans, Van Rijnsoever and Van Trijp2015; Bang et al., Reference Bang, Shu and Weber2020). Moreover, we note that these two interventions have the lowest acceptability scores. Our study design does not allow for any claims about the exact nature of this association. We suggest the distinction between perceived self-effectiveness and perceived other-effectiveness, and their relationship with acceptability, as a topic for further research. In the meantime, communication departments should take into account that the phrasing used to describe policy interventions (e.g., aimed at ‘you’ or ‘other people’) could impact its persuasiveness (cf. Cornwell and Krantz, Reference Cornwell and Krantz2014).
Conclusion
To summarise, in the present study, we find that the acceptability of behavioural interventions in financial decision-making is not as high as the acceptability levels found in previous studies, which focussed mostly on healthy eating. More generally, the results demonstrate that acceptability appears to be at least partly domain-specific. The results also highlight that we have only just begun understanding the potential drivers of acceptability. In contrast to many previous studies, but in line with Gold et al. (Reference Gold, Lin, Ashcroft and Osman2023), the agent of the intervention is not associated with acceptability in any practically meaningful way. While the perceived effectiveness of the intervention is – as in previous studies – the main factor associated with acceptability, we find that the association between perceived effectiveness and acceptability differs between perceived self-effectiveness and perceived other-effectiveness. An important implication of these heterogeneous findings is that academics as well as practitioners need to be cautious when aiming to apply insights about acceptability and its drivers from one domain to another.
Supplementary Material
To view supplementary material for this article, please visit https://doi.org/ 10.1017/bpp.2024.10.
Acknowledgements
We want to thank the Dutch AFM for access to the Consumer Panel. Views expressed in this paper are those of the authors and do not necessarily reflect the position of the AFM. All errors are our own. We also want to thank our (former) colleagues at the AFM and at the Vrije Universiteit Amsterdam for helpful feedback and discussions. This research did not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors.
Competing interest
We have no known competing interests to disclose.
Open access
