Pledging after nudging improves uptake of plant-based diets: a field experiment in a German university cafeteria

Meat-based diets are carbon-intensive and incompatible with Paris climate targets. Reducing meat consumption is essential to mitigate climate change. Behavioural nudges, which present structured choices to citizens, have been increasingly used to reduce meat demand. But they face ethical challenges and limits when scaling up. We test if encouraging people to reflect after nudging improves the effectiveness of a nudge. We design and administer a novel randomised controlled trial in a German university cafeteria, lasting for five weeks (N participants = 129,N meals = 645). In week 1, we measure baseline dietary behaviours. In week 2, we introduce a labelling nudge in the cafeteria. Subsequently, in weeks 3 & 4, we assign participants randomly to three experimental conditions: a control group that continues to receive the labelling nudge and two treatment groups that get the labelling nudge with an opportunity to reflect, either on the nudge (nudge+ 1) or their own preferences (nudge+ 2). All treatments are discontinued in week 5. In the pooled sample, controlling for period fixed effects, we find that the labelling nudge is not associated with meaningful changes in meat-demand over time. Nonetheless, being encouraged to reflect reduces meat-demand significantly compared to the nudge—the nudge+ 1 reduces chances of buying a meat-based item in the cafeteria by 5% (μ = −0.25, 95% CI = [−0.49,−0.36]) whereas the nudge+ 2 reduces it by 7% (μ = −0.35, 95% CI = [−0.61,−0.08]). These treatment effects attenuate when the interventions are discontinued. We recommend that combining reflection with nudging can improve the uptake of climate-friendly diets, at least in the short-term.


Introduction
Current meat-based diets are carbon-intensive [1] and incompatible with Paris Climate targets [2].Recent studies indicate that meat-consumption continues to increase globally [3].This is driven by low-and middleincome countries mostly, where population and income levels continue to grow.Contrarily, while meat demand is stabilising (or even slightly decreasing) in high-income countries, these countries still exhibit patterns of overconsumption.Consequently, notable reductions in meat demand is important to reduce the climate impact of our diets.In these contexts, adopting 'planetary health diets', that is diets rich in plant-based foods, is essential to mitigate climate change [4,5].So, how can we reduce meat demand effectively?
Behavioural nudges-which are minor modifications in how choices are presented [6]-are popular interventions to change people's diets [7].There is growing experimental evidence that nudges can facilitate climate-friendly 4 dietary shifts [8]-both in the field [9,10], as well as in online settings [11,12].Food nudges vary, ranging from informational labels [13,14], to social and personal norms [15], default options and micro-physical changes in the environment [16,17], notably amongst others.Meta-analyses of nudges regularly place food interventions as one of the most secure findings with healthy effect sizes [18], one of the few domains that survive corrections for publication bias in nudge studies [19].However, there are limits to nudging people to change their diets.Nudge 'effects may be individualised and meaningful impact may require multiple nudges' [20, p 203].This is because diets are complex, dynamic human behaviours which remain difficult to change.Usually nudges are not designed to change underlying human preferences nor engage people in the behaviour change process.Consequently, when nudges are retracted, behaviours often return back to status quo 5 , a finding that suggests we need to empower citizens to use nudges better [22].There is also the criticism that nudging meat behaviours without acknowledging people's dietary preferences can be ethically dubious [23,24] and controversial [25], as meat-consumption is a personal choice and often associated with individual human rights.
To address these shortcomings of nudges, a modified toolkit-the nudge+, has been proposed in the field of behavioural public policy [26,27].Contrary to nudge, nudge+ encourages people to think alongside nudging.Reflection can have manifold objectives and can be designed in different ways, such as on one's own preferences and goals, the design of the nudge, the end goals of the nudge or on the nudger itself [28].Such thinking enhances the uptake of socially beneficial behaviours, as steered by the nudge, on average, thereby improving overall policy effectiveness [12].Prior experimental studies in online hypothetical settings have shown that reflection combined with a nudge increases behavioural intentions for climate-friendly diets [12,14].In this paper, we extend these tests to a field setting with real dietary choices, comparing two nudge+ interventions to a standard labelling nudge.We choose to do this in a German university cafeteria.Germany presents us with an interesting testbed for our experiment, where meat consumption level is steadily decreasing-for example, meat demand reached an all time low in 2022 (since 1989)-but still features a high per-capita demand of 52 kilograms [29], making it a country with one of the highest intake of animal-source foods [30].Recent evidence also shows that German diets are currently higher in greenhouse gas emissions compared to the global average [31].We choose a public catering set-up, as recent studies have indicated that a 'nationwide implementation of recipe revision [in German public catering] according to scientific guidance-such as concrete target goals for the GHG emissions per serving-can save up 40% of GHG emissions' [32, p. 2287].
In this context, we designed and administered a randomised controlled trial over a period of five weeks in a university cafeteria.A total of 129 participants, mostly students 6 reported their actual individual meal choices via a QR code during this period.As such, we have N = 645 (averaged) meal observations in our sample.The first week served as a baseline period.In the second week, all participants were exposed to a labelling nudge, displayed in the cafeteria in the form of a green label for vegetarian items, on menus and near tills.In weeks 3 & 4, we continued displaying the labelling nudge but additionally assigned trial participants randomly into three experimental conditions: a control group, which received no additional treatment, and two treatment groups.Both treatment groups were sent bi-weekly surveys that differed in their content.In one treatment group ('nudge+ 1'), receivers were asked to reflect on the labelling nudge and report their level of support for it, whereas in the other ('nudge+ 2'), receivers were asked to reflect on their own (climate-friendly) dietary preferences via a pledge, and thereby report on ways to make their dietary habits more climate-friendly.In week 5, all interventions were discontinued and people's meal choices were recorded for follow-up.The trial was conducted in association with the university's catering department.In this paper, we report the average effect of these nudge and nudge+ interventions using state-of-art econometric models, notably regression analysis with time fixed effects and difference-in-differences.
We make two important contributions to a growing field of behavioural science applied to promote the uptake of plant-based diets.We design and administer the first field test of nudge+ and experimentally compare it to a labelling nudge.In particular, we validate previous findings made in online hypothetical settings that suggest adding reflection to a nudge improves intentions to reduce meat consumption compared to the standalone nudge.Second, our experimental trial runs over a period of five weeks, thereby enabling us to assess the effectiveness of these behavioural interventions over time.Our ability to measure these temporal effects, which is limited in most current one-off evaluations of behavioural (dietary) nudges, enables us to speak to current debates in behavioural sciences which point towards the decreasing effectiveness of nudges over time.The temporal inconsistency in the effectiveness of the nudge+ interventions, along with recent findings on habit formation [33], suggests that persistent dietary behaviours will likely need repeated nudges, which are combined with reflection, and take place over longer periods of time.These findings generate important insights for practitioners in the food sector and policy-makers who can promote climatefriendly diets by encouraging consumers to reflect transparently on behavioural nudges and their own dietary preferences.
The remaining sections of paper are organised as follows.Section 2 outlines our experimental design and research methodology.Then, we present our findings in section 3. Finally, section 4 discusses these findings, highlights important policy implications of our research, acknowledges research limitations, and sets out future research directions.

Experimental design
The randomised controlled trial was administered in a German university cafeteria.At the time of the trial, a total of 2,082 students were available for recruitment.To recruit participants from this pool, an invitation email with an embedded survey was sent to all available participants (see figure 6 in Online Appendix), one week prior to the start of the trial.The email was distributed via a centralised mailing system that was already in use by the catering services 7 .The invitation email informed participants about the purpose of the study and asked them for consent to participate.Consenting participants were then asked to report their email address, gender, and affiliation in the survey.They were also asked to report their purpose and frequency of canteen visits.Finally, to measure their people's prior knowledge about the environmental sustainability 8 of different diets 9 , participants were asked to report perceived sustainability levels corresponding to these diets using a Likert scale.Participation in the trial was incentivised-as a reward, three Amazon vouchers, each worth 100 EUR, were offered to be distributed randomly among consenting participants.The survey received a total of 227 individual responses (≈10% response rate) who consented to participate in the trial.A copy of the invitation email is shown in figure 6 in the online appendix.The experimental trial was approved by the university.
The experimental trial lasted for five weeks, taking place between 27th March and 28th April, 2023.Consenting participants were asked to report their daily meal choices in the cafeteria using an appropriate QR code 10 .These QR codes were available near the checkout tills and were also placed at the tables for convenience (see figure 10 in the Online appendix).The trial was broadly divided into four periods.
1. Period 1 (Baseline): In this period, lasting one week, participants only reported their meal choices in a business-as-usual scenario.
2. Period 2 (Nudge): In this period, lasting one week, all participants were exposed to a labelling nudge displayed in the cafeteria (see figure 7 in online appendix).We used a visual nudge, in the form of a green leaf, which was placed on the (a) blackboard displaying the vegetarian menus, (b) on the tills corresponding to vegetarian foods, and (c) on the salad bar.Similar eco-label nudges have been used in previous experimental studies based in cafeterias.The green colour was chosen for its cultural symbolism and relevance to plant-based diets.
A visual presentation of this nudge placement in the cafeteria is shown in 7 in the Online appendix.

Period 3 (Between-subjects):
In this period, lasting two weeks, we continued to display the labelling nudge in the cafeteria.In addition, we assigned all consenting participants randomly11 to one of three different experimental conditions.
• Control group: In this condition, participants did not receive any additional treatment beside the continuing labelling nudge.
• Nudge+ 1 group: In this condition, participants were sent bi-weekly emails encouraging them to reflect on the labelling nudge (see figure 8).More precisely, they were asked if they had observed the labelling nudge and if they liked the nudge or not.The email sent to the nudge+ 1 group is shown in 8 the Online appendix.
• Nudge+ 2 group: In this group, participants were also sent bi-weekly emails but instead encouraged to reflect on their own dietary preferences.More specifically, they were asked if they had observed the labelling nudge, prompted to pledge to a climate-friendly diet, and asked to report ways in which they could maintain such a diet.The email sent to the nudge+ 2 group is shown in 9 in the Online appendix (also see figure 9). 7Participants were more likely to respond to emails sent by the catering services.The email distribution system contained valid email addresses for 1,917 sample participants 8 In the survey, participants were presented with a Likert Matrix (1 meaning very unsustainable-5 meaning very sustainable) containing various dietary options, asking them to state whether they consider this food option sustainable. 9Vegetarian, Vegan, Dairy-free, Pescetarian, Daily meat consumption, and Limited meat consumption Participants continued to report their meal choices using the QR code for follow-up.
The experimental design is summarised in table 1 below.Set up this way, and in line with prior findings from the literature, we tested the following hypotheses: • H1-The labelling nudge will reduce meat-demand over time.This is a one-sided test.
• H2-Nudge+ 1 will reduce meat-demand compared to the labelling nudge.This is a one-sided test.
• H3-Nudge+ 2 will reduce meat-demand compared to the labelling nudge.This is a one-sided test.
We determined our optimal sample size requirements using an ex-ante power analysis conducted in G * Power 3.1 [34].The current study applied a nudge+ design that has been tested in the literature [12], and therefore we used a minimum detectable effect size based on the previously estimated large Cohen's d (= 0.83).Correcting a family-wise error rate of 0.05 for testing three hypotheses and using the conservative Bonferroni approach, we obtain a corrected α = 0.0167.Assuming power of 80 per cent, using a two-groups, twoindependent means t-test (one-sided), we estimated a minimum sample requirement of 32 individuals in each experimental condition.In total, we aimed to recruit a minimum of N = 96 individuals in order to be sufficiently powered to detect meaningful differences between the treatment (nudge+) and the control (nudge) groups.We did not assign the nudge randomly and therefore we did not estimate sample size requirements to infer its effectiveness causally.

Analysis
Our main outcome variable of interest is meat-demand.Our unit of analysis is the individual i.Using the QR code, a participant could report either a meat-based (= 1) item or a vegetarian (= 0) item, measured as a dummy variable.For each week, we calculated meat-demand, for an individual i, as a sum total of all meat-based items 12consumed over a week.Therefore, meatdemand is a continuous variable that ranges from 0-5, where 0 indicates that a participant did not consume any meat-based item in the cafeteria over a week's time and 5 indicates that the participant only consumed meat-based items during this period.
We use two time dummies interchangeably: Period j , where j indicates the period of the experimental trial, such that j ä [1-4], and week k , where k indicates the week of the experimental trial, such that k ä [1-5] as indicated in table 1 above.Further for period 3, we use two treatment dummies, nudge+ 1 and nudge+ 2, which would take the value 1 if a participant i was assigned to either of the nudge+ conditions.To check for randomisation, we deploy two covariates, namely gender i which takes a value 1 if the participant i is a female and 0 otherwise, and education, measured as a categorical variable to indicate different levels of education (undergraduate/postgraduate-taught/postgraduate-research) at the university.Figure 4 summarises our key variables used in the Online appendix.
We use linear regression models with robust standard errors to test our hypothesis.Specifically, we test hypothesis H1 using the pooled sample controlling for period fixed effects using equation (1) below.We report these findings in table 2 in section 3 below.
Period j 1 Next we test hypotheses H2 and H3 using observations from period 3 with an ordinary least-squares regression, controlling for covariates and week fixed effects, and with robust standard errors via equation (2) below.We report these findings in table 3 in section 3 below.
Our estimate β, measuring the effectiveness of the nudge in all periods (except period 1) compared to baseline (equation ( 1)), is correlational in nature (as we do not randomise the labelling nudge), whereas β 1 &β 2 , measuring the effectiveness of the nudge+ interventions compared to a control group (equation ( 2)) is causal.Further, for robustness, we use a difference-in-differences specification to evaluate hypotheses H2 and H3 over time, as shown below in equation (3).These findings are reported in table 4 in section 3 below.
We report Young's [35] corrected randomised-t p-values.All statistical analysis has been performed using Stata 17.1.

Results
In our sample of 227 participants, 134 were male and 93 were female, resulting in a 60/40 gender split.Our sample is relatively young and educated, being a student sample.94% of all participants belonged to the age group 18-24, while 5% of the sample was 25 years of age or higher.Further 99% of the sample consisted of students in full-time education, of which 86% was enrolled in an undergraduate program, 13% was studying for a postgraduate (taught) degree, and remaining 1% was engaged in doctoral research.48% of the sample reported visiting the canteen, on average, once a week for lunch, and an average participant visited the canteen roughly twice per week.Most participants reported vegan and vegetarian diets to be the most (environmentally) sustainable, with a mean Likert score of 4.04 and 3.94 (out of 5).The baseline survey results are shown in Figure 5 in the Online appendix.
After screening out participants who do not meet the inclusion criteria reported above, our final sample of consists of 129 participants (out of 227 registered participants) who reported their meals consistently over five weeks (see Online appendix).More details on data cleaning is available in the Online appendix.All results using this sample data are reported below.

Result 1-
The labelling nudge is not associated with any significant change in meat-demand over time Table 2 below reports our coefficients from a linear regression corresponding to equation (1) in section 2.2.First, we restrict this to the first two periods only, making it a simple before-after analysis.Comparing meat-demand in period 2 versus period 1, we do not find any significant correlation between meat-demand and the exposure to the nudge (which was introduced in period 2).While the nudge seems to reduce meat-demand (β = −0.15),this effect is not statistically significant (p = 0.216).Our finding implies that nudging people by exposing them to eco-labelling information, where vegetarian items are displayed with a green leaf, does not meaningfully reduce their demand for meat-based foods.
Next we repeat this analysis over the span of whole experimental trial (five weeks), comparing all other periods (except the baseline) to the baseline.We find that barring period 3, there is no meaningful difference in the average meat-demand of participants over a period compared to the baseline.Recall, in period 3, we introduced the nudge+ treatments randomly in addition to the labelling nudge.Therefore, this finding suggests that the significant reduction in meat-demand in period 3 (compared to the baseline) likely comes from these additional treatments and not the labelling nudge per se.We investigate this causally next.Figure 1 displays the average meat-demand over the different periods of the experimental trial, with 95% confidence intervals.

Result 2-The nudge+ interventions reduce meat-demand significantly compared to the labelling nudge in period 3
Using equation (2), we test the causal effect of the two nudge+ interventions versus the labelling nudge.We find that both the nudge+ interventions significantly reduces meat-demand in period 3 compared to the control group.Being asked to reflect on the nudge, for example, by reporting their level of support for the labelling nudge reduces meat-demand by 0.23 units (p = 0.062).In other words, the nudge+ 1 intervention reduces the chances of buying a meat-based item by ≈ 5%, compared to the labelling nudge.Similarly, being asked to reflect on one's own dietary preferences via a pledge (nudge+ 2) reduces meat-demand by 0.35 units (p = 0.011).This is equivalent to almost a 7% reduction in the chances of buying a meat-based item when exposed to the nudge+ 2 treatment, versus the standard labelling nudge.These results indicate that when people are made to think before being nudged, they change their meat-demand significantly.Table 3 reports these regression coefficients below.
Figure 2 plots these linear predictions compared to a female, control group participant.Our experiment is not sufficiently powered to detect any heterogeneous treatment effects between these covariates and the different experimental conditions.

Result 3-Treatment effects are temporally inconsistent and attenuate when interventions are discontinued
Next, we check whether these treatment effects of the nudge+ interventions are consistent over time using equation (3). Figure 3 plots the average treatment effect of being assigned to the nudge+ treatments compared to the standard labelling nudge.We do not see any meaningful differences in meat-demand between the treatment and the control groups in periods 1 & 2. The nudge+ treatments were introduced randomly in period 3 (week 3) only, therefore, prior to this, we should not expect any meaningful differences between the nudge+ and the control groups.Consequently, this finding reaffirms that there were no baseline differences between the different experimental conditions prior to the introduction of the nudge+ interventions randomly in week 3. Further, in this difference-in-differences specification, we find that the reduction in meat-demand, produced by the nudge+ 1 intervention over time, is no longer statistically significant.This implies that once we control for changes in people's demand over time, we do not find the nudge+ 1 intervention to produce meaningful difference in meat demand.However, we might be insufficiently powered to detect these effects and therefore these results should be interpreted carefully and validated externally.On the flip side, our observations indicate that the nudge+ 2 treatment demonstrates its effectiveness only during the second week of implementation, specifically in week 4.This implies that the nudge+ 2 intervention requires a certain period to manifest its efficacy in curbing meat demand.This observation aligns with our expectations, as the nudge+ 2 intervention hinges on individuals committing to environmentally friendly dietary choices, and the process of habit formation often necessitates some time before its impacts become apparent.However, its effectiveness is attenuated once the treatment is discontinued.These regression coefficients are reported below in table 4.

Discussion
In this paper, we present novel experimental evidence on the effectiveness of two nudge+ interventions, versus a labelling nudge, in reducing meat consumption in a German university cafeteria (N participants = 129, N meals = 645).In a pooled sample, using period fixed effects, we find that the labelling nudge is not meaningfully correlated with a reduction in meatdemand over time.However, when trial participants are randomly assigned to be encouraged to reflect, either on the labelling nudge (nudge+ 1) or their own dietary preferences (nudge+ 2), meat-demand is significantly reduced compared to the standard labelling nudge condition.On average, the nudge+ 1 reduces chances of buying a meat-based item by 5% (μ = −0.25,95% CI = [−0.49,−0.36])whereas the nudge+ 2 reduces by 7% (μ = −0.35,95% CI = [−0.61,0.08]).However, these effects are not consistently detected over time.Using a differencein-differences specification, we see that the effect of the nudge+ 2 is detectable only in its second week of application (week 4).Treatment effects of all interventions are attenuated once the interventions are retracted.
Our findings contribute to a growing literature in behavioural-environmental economics, particularly on the use of behavioural insights to reduce meat consumption, in online as well as field settings.Specifically, we make two important contributions.First, we conduct the first field test of a nudge+ intervention, versus a nudge, in reducing meat consumption with real behaviours in a cafeteria.We are able to validate prior findings from previous research, a stylised hypothetical online experiment suggesting that reflection before a nudge improves climate-friendly dietary intentions [12].The set-up and design of the 'nudge+ 2' treatment in our experimental trial matches that of the 'nudge+ (pledge+default)' treatment in the survey experiment, albeit with a different nudge.We show that these effect sizes are modestly comparable-our nudge+ 2 treatment has a small effect size, when compared to the labelling nudge (d = 0.197).This is slightly lower than that reported in the survey experiment (d = 0.334), which may reflect differences in the subject pool or the outcome measurement.
Second, our experimental design allows us to measure the effectiveness of these nudge+ interventions over time.A common criticism of most behavioural experiments relates to their one-off nature, where behaviours are not tracked over time.We overcome this shortcoming with our experimental design which lasts over four periods (five weeks), whereby we are able to assess if the treatment effects persist shortly after the treatments are retracted.We find that treatment effects are attenuated significantly such that we do not measure any meaningful differences in meat-demand between the different experimental conditions.This finding speaks to a growing base of evidence that suggests behavioural interventions lose their effectiveness over time, especially when nudges are discontinued.For example, in a recent meta-evaluation, DellaVigna and Linos [21] show that treatment effects attenuate by 0.7 percentage points with each passing day.This attenuation effect has been observed for different pro-social behaviours, namely charitable giving, financial savings, tax and other types of compliance, and voter turnout, with the exception of water conservation and energy demand [36].While we cannot test the mechanisms of this 'voltage drop' [37] in our case, it is likely caused by the short-lived nature of our nudge+ interventions.Dietary behaviours are complex and sticky, and, therefore, it is possible that behavioural interventions have to be applied for a longer time before being discontinued to induce sustained behavioural shifts.For example, as we show the benefits of reflection on one's dietary preferences, via the pledge, take longer to be realised after its introduction.This relates to evidence in the literature that new habits can take from a few months to nearly a year to form fully [33,38] and behavioural interventions should be repeated sufficiently over time to sustain positive behaviour change [38].
Our research design has some limitations, which must be carefully considered and overcome in future research.First, we observe attrition in our sample.Of the 227 participants who consented to participate in the trial, only 129 respondents actively participated over five weeks.Our attrition rate (of is not uncommon as has been documented in the literature previously-for example, Dumville and colleagues [39] reviewed 132 randomised controlled trials, of which they noted 54% of all trials suffered from some degree of attrition and reported 48% as the highest attrition rate.Despite this attrition, we are confident that our findings are robust to attrition bias.This is because we follow CONSORT guidelines [40] in reporting all screen outs over time and ensuring that our treatments (in weeks 3 and 4) are randomised effectively in the sample.We have reported our data cleaning strategy transparently in the Online appendix.Second, while our internal validity is not threatened, we acknowledge threats to our external validity and generalisability.We have inherent selection bias as 10% of the total available pool (N = 2,082) voluntarily self-selected into the experimental trial.We also have a highly educated, student sample based in a Western country.As such, our experimental findings must be generalised in a diverse population setting.We recommend follow-up studies using different regions, sex, or age groups.Third, we are unable to measure experimenter demand or social desirability effects.It is possible that reducing meat-demand induces a high degree of normative pressure on participants.Although any climate policy directed to reduce meat consumption is likely to suffer from this problem, we call for future research to explore the magnitude of these effects.To this we must add that only the nudge+ hypothesis is causally identified, whereas the nudge is not, meaning we cannot rule out any treatment effect as found in other studies.Finally, the intervention is relatively short at five weeks do we do not know what would have happened had there been a more sustained intervention, which could have increased the treatment effect or lead to disengagement.Larger and longer-lasting studies are needed.
Despite these limitations, we are confident that our research demonstrates potential for using reflection in a nudge (nudge+) as a new means of promoting climate-friendly dietary behaviours.By confirming prior findings from online hypothetical experiments [12,14], we show that nudge+ interventions can facilitate stronger behavioural shifts than standalone nudges, at least in the short-term.Our findings are policy relevant for practitioners in the food industry, such as cafeterias, hospitals and other public institutions which use food menus, and can use similar prompts to encourage thinking in individuals alongside nudging.The public catering sector is highly carbon-intensive, and the adoption of carbon-neutral menus can reduce the footprint of this sector by up to 40% [32].We recommend that food practitioners engage consumers in the process of behaviour change by enabling them to think through behaviour change policies.Policy-makers in central and local governments may promote the greater use of these kinds of communications through funding pilot schemes and encouraging interventions and evaluations, such as in staff canteens.

Online appendix Data cleaning and analysis
For the experiment, a total sample of 2082 students was available.To identify contactable participants, an exante survey was sent out one week before the start of the experiment to the email address: [redacted for privacy], which included 1917 students out of the student body.In the email, students were encouraged to answer the initial survey and utilize the QR code system in the university canteen to if their food choice was either vegetarian or meat-based.As a reward for their cooperation, three Amazon vouchers, each worth 100 EUR, were offered, distributed randomly amongst students.The survey received 227 individual responses.
A data management and data cleaning strategy was applied to ensure a comparable sample.This strategy evolved around a three-step data-cleaning approach, based on Van den Broeck and colleagues' recommendations [41], subdividing data cleaning and management into three main parts: (1) Screening, (2) Diagnosis, and (3) Editing.In the Screening period, excess data and doubles were assessed, and rules were defined for staying within the sample set.As the experiment incorporated within-and between subjects characteristics, the data already had to be cleaned during the experiment at the end of period 2. Therefore, the following two sets of exclusion criteria were defined: 1. Participants should have registered at least once a week within the first two weeks and have filled out the exante survey with the overall sample maintaining a gender split of 60/40.
2. Participants should be included in the data which was used for pt. 1 above, have registered in the canteen at least once a week over the course of the five weeks, and have reacted at least once a week to the surveys which were sent out as a treatment to nudge+ 1 and nudge+ 2 treatment groups with the overall sample maintaining a gender split of 60/40.
In the Diagnosis period, missing data was identified, depicted by 'n.a.' in the excel spreadsheet, indicating that students did not scan the QR code.Finally, in the Editing period, data was corrected as following.After removing the data that did not adhere to pt. 1 above, the observable sample consisted of 192 participants, who were subsequently assigned randomly to one of the three groups: control (nudge) and treatments (nudge+ 1 and nudge+ 2) randomly.The allocation took place via stratified or blocked randomization, using a 'Randbetweenfunction' in Microsoft Excel.Following pt. 2 above, the final observable sample consisted of 129 participants, subdivided into three test groups consisting of 64 participants, with 38 being male and 26 being female.This final sample mainly consisted of undergraduate students, with the majority being first-year students.
This final sample was used for all statistical evaluations to ensure comparability between the groups in period 3 and across time periods.As the study aimed to assess climate-friendly food options, taking vegetarian food choices was used as a uniform definition, as the vegetarian food option was considered the most climate-friendly in the ex-ante survey, with 82.46% of all participants considering this as either (environmentally) sustainable or very sustainable.The data cleaning and analysis process is summarised in table 5 below.

4. Period 4 (
Post-intervention): In this period, lasting a week, all nudge and nudge+ interventions ceased.

Figure 1 .
Figure 1.Average meat-demand over different trial periods.

Figure 2 .
Figure 2. Effects of linear prediction.Here reference category is a participant i in the control group, who is a female, with undergraduate degree and a low sustainability score for vegetarian food (= 1/2).

Figure 3 .
Figure 3. Effects of the nudge+ intervention (introduced randomly in Week 3) versus the labelling nudge for different weeks of the experimental trial.

Figure 4 .
Figure 4. List of variables and their coding as used in the data analysis.

Figure 8 .
Figure 8.The email survey that was sent to participants randomly in the nudge+ 1 treatment group in weeks 3 & 4.

Figure 9 .
Figure 9.The email survey that was sent to participants randomly in the nudge+ 2 treatment group in weeks 3 & 4.

Table 1 .
Summary of experimental design.

Table 5 .
Summary of data cleaning and analysis process.