Interventions that inﬂuence animal-product consumption: A meta-review

Transitioning toward plant-based diets can alleviate health and sustainability challenges. However, research on interventions that inﬂuence animal-product consumption remains fragmented and inaccessible to researchers and practitioners. We conducted an overview of systematic reviews, also known as a meta-review. We searched ﬁve databases for reviews that examined interventions that inﬂuence (increase or decrease) the consumption of animal products. We quantitatively summarised results using individual studies’ directions of eﬀect because reviews rarely reported eﬀect sizes of primary studies. Eighteen reviews met inclusion criteria, 12 of which examined interventions intended to decrease animal-product consumption and 6 of which examined interventions intended to increase animal-product consumption. In total, only two reviews conducted quantitative meta-analyses. Across all reviews, vote counting indicated that providing information on the environmental impact of meat consump- tion may reduce consumption, with 10 of 11 estimates suggesting reduced consumption (91%, 95% CI [62.3%, 98.4%]; p = .012). Providing information on the health consequences, emphasising social norms, and reducing meat portion sizes also appeared promising, albeit with more limited evidence. Reviews examining interventions that decreased consumption predominately focused on meat (10/12 reviews). Future reviews should conduct quantitative syntheses where appropriate and examine interventions that inﬂuence the consumption of animal products other than meat.


Introduction
There is growing consensus that a transition toward a primarily plant-based diet would benefit public health, food security, the conservation of biodiversity, the climate, and animal welfare ( Aiking, 2011 ; for ruminant animals such as cattle ( Clark and Tilman, 2017 ). Further, it is widely accepted that factory farming causes severe, extensive, and potentially unethical animal suffering ( Broom, 2007 ;Bryant, 2019 ;Proctor et al., 2013 ;Wagner et al., 2015 ), and there is substantial public support for a ban on factory farming in countries such as the USA ( Norwood and Murray, 2018 ;Reese, 2017 ). The EAT-lancet commission report calls for a "Great Food Transformation " and a paradigm shift in our food systems ( Willett et al., 2019 ). Adoption of planetary health diets that optimise health and environmental sustainability, while reducing suffering, will require interventions that stimulate a range of actions from both individuals and organisations.
Yet, encouraging people to substitute plant-based foods for animalproduct foods in their diets is difficult and changing the food habits of millions of people has been identified as a key research area requiring more attention ( Béné et al. 2020 ). Demand for animal products worldwide is increasing ( FAO, 2017 ), a trend that is projected to continue as the world population grows, affluence increases ( Aiking, 2011 ), and more countries (especially those with large populations such as China and India) adopt a Western-style diet ( FAO, 2017 ;Slingo et al., 2005 ). Diets can also be difficult to change because driving forces -such as taste preferences, social context, familiarity, habit, and cultural tradition -are complex, interacting, and sometimes immutable ( Sanchez-Sabate and Sabaté, 2019 ;Valli et al., 2019 ). For example, reasons for low willingness to reduce meat consumption include a lack of cooking skills, lack of information, enjoyment of dishes rich in animal products, and the belief that meat is essential to a healthy diet ( Graça et al., 2019 ;Valli et al., 2019 ).
There is increasing research on how to reduce animal-product consumption. Literature has identified several personal, socio-cultural, and environmental factors that influence the consumption of animal products and there is a growing focus on interventions -actions taken by individuals, businesses, or governments -which incorporate these factors (e.g., Bianchi et al., 2018b;Graça et al., 2019 ;Harguess et al., 2020 ;Taufik et al., 2019 ). Existing reviews draw on disciplines such as environmental sustainability (e.g., Wynes et al., 2018 ) and health promotion (e.g., Valli et al., 2019 ). Compared with primary studies, systematic reviews often provide stronger indications of what interventions work and when they work ( Higgins et al., 2019 ), and offer insights into the reproducibility of key findings ( Stanley et al., 2018 ). However, reviews typically have a narrow focus, such as interventions addressing consumption of only one type of animal product (e.g., meat but not eggs, or vice versa) or consumption in only one context (e.g., supermarkets; Hartmann-Boyce et al., 2018 ). In cases like these, where multiple systematic reviews are available regarding a range of interventions affecting the same outcome, collating and comparing those reviews allows for researchers, practitioners, and policymakers to identify and act upon the most robust evidence ( Becker and Oxman, 2011 ).
To provide a parsimonious and accessible synthesis of the available evidence in a single source, we conduct a systematic review of systematic reviews (hereafter 'meta-review') of interventions that influence (increase or decrease) intended or actual consumption of animal products (a.k.a. an "umbrella review " or "overview of reviews "; Becker and Oxman, 2011 ;Grant and Booth, 2009 ;Higgins et al., 2019 ;Khangura et al., 2012 ;World Health Organisation, 2017 ). Metareviews are particularly effective for making review-level evidence useful and accessible for researchers, practitioners, and policymakers ( Khangura et al., 2012 ;World Health Organisation, 2017 ). The approach enables evidence to be summarised across a broader range of research fields and perspectives than would be practical for a systematic review focused on primary studies ( Becker and Oxman, 2011 ). We include factors found to increase animal-product consumption (e.g., subsidising chicken farmers) because of the potential that the same mechanism could be targeted to decrease consumption (e.g., removing subsidies).

Protocol and registration
Our methods were based on best-practice guidelines for conducting meta-reviews (Johnson & Hennessy, 2019;Hennessy et al., 2019). The protocol was registered in advance using the Open Science Framework (bit.ly/OSF-meta-review).

Eligibility criteria
The inclusion and exclusion criteria are presented in Table 1 .

Search strategy
To broaden the reach and utility of the review, our expert advisory board of health and animal welfare researchers (DR, MW, ST, MM, KZ, AH, CB, DM, PP, DW, JP) and practitioners (JH) assessed the proposed search terms, inclusion and exclusion criteria, and database sources. Our search strategy comprised two stages; in stage 1 we searched both peer reviewed and grey literature to identify relevant articles. Stage 2 involved forwards and backwards searching to identify any additional eligible articles. The two stages are described below.

Database and grey literature searching
A search strategy was developed in collaboration with an academic reference librarian. Five databases -Medline, Scopus, PsycINFO (Ovid), Web of Science, and Proquest Dissertations and Theses -were searched on 21 February, 2020 for peer-reviewed articles published from 1990, due to very few systematic reviews being conducted before then ( Smith et al., 2011 ). Terms used for identifying the type of study were informed by a comprehensive typology of review methods ( Grant and Booth, 2009 ) and included: meta-anal * , "systematic review ", meta-regress * , "rapid review ", "scoping review ", "state-of-the-art review ", and "mapping review ". Terms used to indicate interventions included: influenc * , nudg * , "behavio * r change ", educat * , encourag * , and interven * . Terms targeting animal-product consumption as outcomes included: "animal product * ", animal-based, vegan * , vegetarian * , meat, and plant-based. An example search string is provided in our preregistered protocol (bit.ly/OSF-meta-review). No terms, subject areas, or languages were excluded. Only papers which were found electronically were included, given that this approach expedites the review without influencing conclusions ( Ganann et al., 2010 ). No papers were excluded based on this requirement, as all reviews found were available electronically.
To avoid missing relevant research, and to help mitigate publication bias, we searched the grey literature for eligible reviews using the process outlined by Stansfield and colleagues (2016). This involved using the team's knowledge of relevant resources to generate a list of websites to search (e.g., Food Climate Research Network and Animal Charity Evaluators) and recording details such as the date searched, the pathways followed, any search terms used, and relevant records found. These details were recorded in two spreadsheets (a summary of the search; Supplementary file 1, and the data extraction; Supplementary file 2) by the 11 authors involved in this process (AD, AR, AS, EG, HA, JL, KW, NF, PS, SC, TH).

Forward and backward citation searching
Following the completion of searches and full-text screening, forwards and backwards searching was undertaken using Scopus ( Hinde and Spackman, 2015 ). Backwards searching involved reviewing all citations of articles included in Stage 1. The forwards search involved searching for all articles that cited an included article. These articles were then screened following the process outlined below. Following completion of all full-text screening, included articles were circulated to the expert advisory board to solicit any relevant omissions.

Table 1
Inclusion and exclusion criteria.

Inclusion criteria Exclusion criteria
Types of reviews • Systematic reviews published in academic journals or grey literature (with or without quantitative meta-analyses) • Similar reviews including reproducible search and filtering strategies ( Grant & Booth, 2009 ) • Non-systematic reviews or meta-analyses; primary research papers; theory papers; narrative reviews Types of participants • Any population of participants Types of interventions • Reviews assessing interventions intended to influence (increase or decrease) consumption or purchase of animal products • Reviews including non-interventional studies (e.g., observational research) or irrelevant intervention studies (e.g., examining another food group) were eligible if they also included relevant interventional studies • No exclusion criteria regarding source of intervention (e.g., restaurants), method of persuasion (e.g., flyers), or communication channel (e.g., face-to-face) • Reviews solely exploring factors associated with consumption (e.g., personality) were excluded Types of outcome measures • Reviews measuring intentions or behaviours regarding animal-product consumption. Measurements of intentions included (but not limited to) purchasing behaviour; choosing animal products vs. non-animal products in discrete choice experiments; actual or intended consumption • Reviews could concern dietary products from any animal(s) • Excluded outcomes regarding non-dietary use of animal products (e.g., leather) and reviews of plant-based alternatives (e.g., meat substitutes, tofu, soy milk) that did not measure consumption of animal products • Given our focus on immediate drivers of behaviour, we excluded reviews which focused solely on knowledge of or affective responses to animal products without assessing consumption • Willingness to pay for animal products was originally included in the protocol as an acceptable proxy for intended consumption but was excluded following advice from our advisory board around difficulty of interpretation

Screening and selection of reviews
Fifteen reviewers were involved in screening (NF, RG, KW, PS, AS, AD, JL, AM, HA, AR, SC, TH, EG, LF, MZ). Titles and abstracts of records were independently screened by two reviewers against the inclusion and exclusion criteria. An article progressed to full-text screening if either or both reviewers voted to include it. During full-text screening, two reviewers independently evaluated inclusion criteria in detail. Disagreements at the full-text screening stage resulted in the decision being made by a senior member of the team, not involved in the initial decision (EG, MZ, AS, or PS).

Data extraction, synthesis, and quality appraisal
Fifteen reviewers were involved in data extraction (NF, SC, EG, KW, HA, RG, AM, AD, TH, AR, LF, JL, MZ, PS, AS). A data extraction spreadsheet was developed to capture the following information: review question(s); methods; number of studies and participants; key findings; effect size estimates; key conclusions, and limitations. This spreadsheet was refined after pilot-testing and extractor calibration on three randomly selected articles. Data extraction was conducted in duplicate, with one author (EG) resolving inconsistencies in extraction. A context mapping spreadsheet was also developed to categorise contextual variations in the interventions identified by reviews ( Slattery et al., 2020 ). This captured i) the source(s) of the intervention(s) that reviews examined, ii) the method(s) of intervention(s), iii) the channel(s) transmitting the intervention(s), iv) the receiver(s) of the intervention(s), v) the outcomes(s) of the intervention(s) and vi) the data collection techniques used. Further detail on context mapping is provided in Supplementary file 3.
We planned to convert effect sizes from reviews to a common metric and conduct meta-meta-analyses, however we needed to deviate from our protocol because so few reviews reported effect sizes of primary studies. Instead, we undertook vote counting based on direction of effect -an acceptable statistical synthesis method for when meta-analysis of effect estimates is not possible and consistent effect measures or data are not reported across studies . We performed vote counting by assessing, for each review, the percentage of individual study estimates whose signs suggested reduced rather than increased consumption in the intervention group, regardless of the effect sizes or statistical significance. Statistical significance was not considered because underpowered studies in vote counting can produce misleading conclusions ( McKenzie and Brennan, 2020 ). As recommended, confidence intervals for these percentages of estimates suggesting reduced consumption were calculated using Wilson interval methods ( Brown et al., 2001 ), and we tested the null hypothesis that only 50% of estimates suggested reduced consumption using a two-tailed binomial test . Vote counting based on direction of effect was not implemented to replace a meta-analysis as it neglects factors like sample size and effect size ( Borenstein et al., 2009 ). Considering this paper's focus on interventions that are intended to reduce animal-product consumption, vote counting was not undertaken with interventions intended to increase consumption. Instead, we provided a qualitative synthesis of those results.
Quality assessment of all included articles was conducted in duplicate by two trained authors (KW, TH) using the Assessing the Methodological Quality of Systematic Reviews checklist 2 (AMSTAR 2; Shea et al. 2017 ). Systematic reviews vary widely in quality, so it is important to critically evaluate the methods and reporting, and their potential impact on the findings ( Shea et al., 2017 ). For assessing the quality of reviews, comparative studies have shown AMSTAR 2 to be more reliable and equally valid compared with other tools ( Lorenz et al., 2019 ). These assessments, in addition to all data and materials, are available on the Open Science Framework for transparency (bit.ly/OSF-meta-review). Fig. 1 is a PRISMA diagram illustrating the search results. After removing duplicates, we screened 11,989 articles in total: 11,666 articles from academic databases, 36 from the grey literature search, 283 from forwards and backwards searching, and 4 from the expert advisory board. Of these articles, we assessed the full text of 72. A total of 18 articles met the eligibility criteria and all were included. Table 2 provides a summary of included reviews. The 18 reviews synthesised the literature on interventions influencing animal product consumption in various ways. Five posed a research question regarding     Note. K = Total number of studies included in the review. N = Total number of participants included in the review. specific types of interventions or contexts (e.g., the effect of interventions in grocery stores). The remaining 13 conducted a broad search for any interventions relating to one or more specific animal product(s). Ten reviews examined interventions targeting meat consumption, five focused on dairy, and three focused on animal products in general. In total, only two reviews conducted meta-analyses and pooled estimates of effect sizes ( Nisa et al., 2019 ;Rouf et al., 2018 ). The remaining 16 reviews used qualitative methods of synthesis, and rarely reported the effect sizes of their included primary studies. Of the 18 reviews, 12 examined interventions intended to decrease animal-product consumption, and six examined interventions intended to increase animal-product consumption. Where available, all pooled effect sizes reported by each review are described in Table 2 . The contexts covered by the included reviews are summarised in Supplementary file 3.

Quality assessment
We assessed the quality of included reviews using the criteria set out by AMSTAR 2 (see Supplementary file 4). While these are stringent, gold-standard processes for conducting a comprehensive and reliable systematic review, not all criteria have been shown to decrease bias in conclusions. For example, in many cases, large meta-analyses may to some extent mitigate publication bias without explicitly correcting for it ( Mathur and VanderWeele, 2020b ). Nevertheless, all are 'good practice' and conclusions should be drawn in light of the quality of the reviews. Fewer than half of the reviews (7/18) conceptualised their search in terms of the specific participants, interventions, comparisons, and outcomes they sought to identify. Only one provided an adequately detailed prospective registration with justification for modifications ( Valli et al., 2019 ). In most reviews (13/18), there was no rationale for the type of studies eligible for inclusion (e.g., randomised trials only, randomised and non-randomised trials, all designs). The AMSTAR2 criteria for a 'comprehensive search' are stringent but most reviews (10/18) failed to meet even the 'partial' criteria (i.e., > 2 databases, keywords provided, restrictions justified). More than half conducted study selection (11/18) and data extraction (11/18) in duplicate. Only one listed the excluded studies with reasons ( Valli et al., 2019 ). Most described included studies in either full detail (6/18) or some detail (6/18), but well-described meta-analyses were rare (2/18; Nisa et al., 2019 ;Rouf et al., 2018 ). Nevertheless, many reviews attempted to explain systematic patterns that may have explained heterogeneity in study findings (8/18). Most reviews failed to assess risk of bias (11/18) and most did not use risk judgements in assessing the conclusions drawn from their included studies (14/18). More information on reviews' assessments of the quality of their included studies and their study designs is included in our extraction table in Supplementary file 4. Only one assessed whether risk of bias influenced pooled effect size estimates ( Nisa et al., 2019 ). Most reviews declared any conflicts of interest and had strategies for mitigating biases (12/18), but we did not find any that reported on funding of included studies.
In general, higher quality reviews (AMSTAR score ≥ 5) did not appear to differ from lower quality reviews (AMSTAR score < 5) in their focus. One exception was that three of the higher quality reviews focused on dairy and calcium consumption in young adults compared to only one in the lower quality reviews. Table 3 summarises the vote counting results from the relevant studies within the 12 reviews that assessed interventions intended to decrease animal-product consumption and the implications of those findings. Relevant data from studies that were extracted from the reviews is available in Supplementary file 5. Two types of interventions appeared particularly promising, in that (i) a high proportion of estimates suggested reduced consumption; and (ii) there was a moderately large number of relevant studies (at least 10). First, regarding interventions that provided information on the environmental impact of eating meat, 10 of 11 studies (91%, 95% CI [62.3%, 98.4%]; p = .012) had estimates suggesting reduced consumption. Second, regarding interventions that provided information on the health consequences of meat consumption, 8 of 10 studies (80%, 95% CI [49%, 94.3%]; p = .11) had estimates suggesting reduced consumption. Emphasising social norms was also among the more reliable interventions (4/4 estimates suggesting reducing consumption, 100%, 95% CI [51%, 100%]; p = .125), as was reducing meat portion sizes (4/4; 100%, 95% CI [51%, 100%]; p = .125). Reviews only cited one unique study for several interventions, including images of cows before slaughter, emphasising the social consequences of eating meat, informing consumers that the consumption of meat is associated with social dominance, and providing default plant-based meals.

Qualitative synthesis of interventions intended to increase animal-product consumption
Six reviews discussed the evidence for interventions intended to increase animal-product consumption or related intentions. All six reviews focused on dairy consumption. Providing dairy products or tastings was a reliable intervention for increasing consumption ( Hartmann-Boyce et al., 2018 ). Although the most prevalent behaviour change technique regarding dairy consumption was general nutrition education, this did not reliably increase consumption ( Hendrie et al., 2013 ;Jung et al., 2016 ;Rouf et al., 2018 ). One of the two meta-analyses identified in our search found that educational interventions increased dairy consumption ( d = .31, 95% CI [.11, .50], k = 4), but the high risk of bias in included studies led reviewers to have low confidence in this estimate ( Rouf et al., 2018 ). There is mixed evidence for changes to food placement or signage ( Hartmann-Boyce et al., 2018 ), providing recipes and brochures ( Hartmann-Boyce et al., 2018 ), and whether increasing dairy or calcium consumption alone is more effective than targeting general health behaviour ( Hendrie et al., 2013 ;Jung et al., 2016 ;Marquez et al., 2015 ).

Discussion
This paper is, to our knowledge, the first meta-review to synthesise this research area and provide an accessible synthesis of interventions that reduce animal-product consumption. To simplify the breakdown of these results, we will use sections suggested by the Individual, Social, Material (ISM) Tool ( Darnton and Horne, 2013 ). We describe how animal-product consumption can be addressed in three contexts: the individual (factors held by an individual affecting the choices they make), the social (factors relating to social groups), and the material (factors relating to the wider environment). Following this breakdown, we discuss what we can learn from interventions that increase animal-product consumption, limitations of included reviews, and limitations of our metareview.

The individual context
The individual context includes an individual's skills, knowledge, values, attitudes, and evaluations ( Darnton and Horne, 2013 ). Interventions targeting the individual included providing information about the consequences of animal-product consumption, assisting with goal setting and implementation intentions, personalised messaging, individual lifestyle counselling, and emphasising animal welfare.

Providing information on consumption consequences is a promising intervention
Interventions that provide information typically involve presenting the environmental, health, and animal welfare arguments for reducing animal-product consumption, or a combination of these.
Emphasising environmental c onsequences. Interventions that provided information about the negative environmental consequences of meat consumption had estimates that consistently suggested reduced meat consumption. This aligns with retrospective research suggesting that 4-19% of consumers who report reducing intake were motivated by environmental concerns ( Sanchez-Sabate and Sabaté, 2019 ). This tactic also appears to target a gap in education -consumers tend to be unaware of the environmental impact of the production of meat ( Hartmann and Siegrist, 2017 ; Sanchez-Sabate and Sabaté, 2019 ). Individuals consider meat reduction to be one of the least effective methods for alleviating climate change when compared to other options (such as driving cars less), despite shifting to a plant-based diet being one of the highest impact actions that can be taken by an individual to reduce emissions ( de Boer et al., 2013 ;Wellesley et al., 2015 ;Wynes et al., 2018 ).
When presenting information on the environmental consequences of animal-product consumption, several factors should be considered. First, this intervention may be especially effective when considering particular populations such as university students ( Sanchez-Sabate and Sabaté, 2019 ). Second, the intervention effect may be moderated by prior belief -environmental messages may change meat consumption amongst individuals who already hold negative attitudes towards meat consumption, but not among those who believe it is healthy and climate friendly ( Vainio et al., 2018 ). Third, framing environmental messages so they appeal to emotions and values can be especially effective ( Sanchez-Sabate and Sabaté, 2019 ). Environmental appeals are often presented in a detached and unemotional manner ( Sanchez-Sabate and Sabaté, 2019 ). Framing animal-product consumption as an environmental issue can be "a bridge too far " (p. 28) for consumers if no direct personal relevance is emphasised ( de Boer et al., 2013 ), especially considering that negative consequences are often long-term ( Aschemann-Witzel, 2015 ), complicated, vague, and lacking urgency ( de Boer et al., 2013 ;Wellesley et al., 2015 ). Further research is needed to ascertain how this argument can be emotively assimilated in an informational intervention ( de Boer et al., 2013 ;Wellesley et al., 2015 ).
Emphasising health c onsequences. We found some supporting evidence for interventions that emphasise health consequences. When informed about undesirable health consequences, most omnivores report low willingness to change their meat consumption ( Valli et al., 2019 ). Resistance to change can be generated by beliefs that meat is healthy and necessary ( Graça et al., 2019 ;Valli et al., 2019 ), that the consequences of consumption are trivial, and the belief that individuals have already reduced consumption in the past ( Valli et al., 2019 ). Yet, interventions can successfully influence the intentions of older consumers or those with pre-existing negative perceptions of meat consumption ( Graça et al., 2019 ). Health concerns also appear to be a primary motivation for many vegetarians ( Valli et al., 2019 ) and are the most common motive for non-vegetarians to consider plant-based diets ( Hopwood et al., 2020 ). Therefore, health concerns may be a driver among the health literate, Promising strategies include: -Labelling meat options as "meat" instead of "standard" or "normal" in cafeterias -Referring to "beef" and "pork" dishes as "cow" and "pig" Description of meat alternatives One review 1 0 cited 3 unique studies. Majority of results were not in favour of the intervention (1/3; 33.34%, 95% CI [6.1%, 79.2%]; p = 1).
Mixed results for strategies including: -Changing name of meat-free meals to more appealing alternatives -Highlighting a plant-based meal as the "Chef's recommendation" ( continued on next page ) Mixed results for strategies including: -Tailoring messaging based on the receivers state of change, animal-product intake levels, or personality Individual lifestyle counselling One review 2 cited 8 unique studies. Majority of results were in favour of the intervention (6/8; 75%; 95% CI [40.9%, 92.9%]; p = .29).
Promising strategies include: -Providing individualised supporting material such as information on barriers to change, feedback, and support to prompt behaviour change Reducing meat portion sizes Three reviews 4 7 1 0 cited 4 unique studies conducted in lab and field (restaurant, stores) settings. Results were consistently in favour of the intervention (4/4; 100%; 95% CI [51%, 100%]; p = .125).
Promising strategies include: -Supermarkets offering the option of smaller meat portion sizes -Restaurants reducing meat portion sizes but maintaining dish volume by increasing vegetable servings Note.  (2018) but a barrier for those who believe that plant-based diets are inadequate ( Dibb & Fitzpatrick, 2014 ). Emphasising a nimal w elfare c onsequences. Although Bryant and Barnett (2018) note that "the most commonly perceived benefit of cultured meat was in terms of animal welfare " (p. 16), emphasising animal welfare had relatively little supporting evidence compared to environmental or health messages, partly reflecting the smaller number of studies examining animal welfare interventions. There is some supporting evidence behind implicitly emphasising animal welfare through presenting images of cows heading to slaughter, meat with the head attached, or animals next to recipes.
Providing information on multiple consequences. The effect of educating individuals on multiple consequences of eating meat, rather than targeting a specific framing, is unclear with estimates being in mixed directions. Integrating several food-related values (e.g., health and environmental) into the approach, rather than presenting it as an isolated issue (e.g., only emphasising environmental considerations), may help to mitigate resistance generated by particular approaches ( de Boer et al., 2013 ). At the other extreme, providing excessive information, as opposed to tailored information, may sometimes have a backfiring effect and increase consumption, potentially through text and images triggering unconscious cravings ( Klöckner and Ofstad, 2017 ).

Personalised messaging
We found mixed support for tailoring interventions to the receiver (e.g., animal-product intake levels, personality, self-schema, or values). Research suggests that the same intervention is not always as effective for all individuals and populations. For example, although some consumer groups are more affected by health concerns, it should not be the go-to approach in all circumstances ( Veul, 2018 ). Younger consumers, people belonging to lower socioeconomic status groups, and those who follow unhealthy diets may be less sensitive to this approach ( Pribis et al., 2010 ;Veul, 2018 ). Further, people who are 'meat believers' (those who are convinced that meat is essential for a healthy diet) may respond negatively to the cognitive dissonance invoked by health appeals ( Veul, 2018 ). Similarly, suggestions that meat-free meals are a potential option for addressing environmental issues can trigger detrimental responses in climate-change skeptics ( de Boer et al., 2013 ). Therefore, those advocating for a plant-based diet may consider not only what they are saying, and how they are saying it, but who they are saying it to.

Individualised lifestyle counselling
Estimates were fairly consistently in favour of individualised lifestyle counselling as an intervention. Although these interventions were often successful in reducing red or processed meat consumption, they seemed to involve substantial investment to conduct (e.g., telephone counselling with health advisors and providing tailored supporting material such as information on barriers to change, feedback, and support to prompt behaviour change). Therefore, it may be difficult to scale this intervention and to disentangle the driving factors of behaviour change from all the components involved.

Goal setting and self-monitoring
This meta-review found estimates consistently in favour of goal setting and self-monitoring interventions. Digital notifications reminding individuals to monitor their red or processed meat consumption appear to be promising ( Carfora et al., 2017 ), but more research is needed to solidify these findings. Further, creating implementation intentions (e.g., creating an intention to consume a meat-alternative in a specific circumstance) may be a reliable tool for reducing meat consumption ( Loy et al., 2016 ;Rees et al., 2018 ).

Implicitly highlighting animal suffering
There is some evidence to suggest that different methods of implicitly highlighting animal suffering may reduce intended meat consumption. These interventions, which typically involve reminding consumers of the animal origins of meat (e.g., with cute pictures, photos of meat with the head attached, or images of cows heading to slaughter), often aim to induce emotionally charged states such as empathy, disgust, and cognitive dissonance ( Harguess et al., 2020 ). Confronting consumers with their contradicting desires to consume meat but not harm animals (the "meat paradox "; Loughnan et al., 2010 ) can prompt them to resolve this discomfort by reducing consumption ( Tian et al., 2016 ). Care must be taken when presenting animal welfare appeals to ensure that confronting consumers with their cognitive dissonance is not harmful and does not induce defense mechanisms that maintain -or even increase -animal-product consumption ( Veul, 2018 ). Further research is needed on how to help individuals who consume animal products overcome defence mechanisms that are employed to reduce discomfort such as avoidance and denial of animal pain ( Rothgerber, 2014 ).

The social context
The social context includes the norms, institutions, roles, relationships, and leaders that an individual is exposed to ( Darnton and Horne, 2013 ). Interventions targeting the social context that were discussed by included reviews included emphasising norms and connotations of meat consumption.

Emphasising social norms
There is some evidence that portraying positive social norms regarding plant-based diets can reduce meat consumption (e.g., through popular TV shows discussing vegetarianism; Byrd-Bredbenner et al., 2010 ). Although vegetarians and vegans represent a minority of the population in Western countries ( Leitzmann, 2014 ;Sanchez-Sabate and Sabaté, 2019 ), and these diets may differ substantially from current norms ( Béné et al. 2020 ), communicating dynamic norms may reduce consumption ( Sparkman and Walton, 2017 ). This involves emphasising how the amount of people following plant-based diets is growing, which prompts pre-conformity due to believing that meat consumption will be lower in future and reducing consumption matters to others. Given that perceptions of social expectations and norms may influence the likelihood of following more plant-based diets ( Wyker and Davison, 2010 ), reference to these could be incorporated when relevant and encouraging.

Social implications of meat consumption
Two studies were cited by included reviews that assessed how being informed of potential social consequences of meat consumption affected intentions. Cordts and colleagues (2014) found that conveying that meat consumption may harm an individual's personal image (e.g., through reduced popularity and cooperative skills) reduced intentions to eat meat. However, Allen and Baines (2002) found no evidence of an effect of manipulating the symbolic meaning of meat (by telling participants that more socially dominant individuals eat more meat).

The material context
The material context includes the infrastructure, rules and regulations, technologies, and objects that surround an individual ( Darnton and Horne, 2013 ). Interventions targeting the material context that included reviews discussed were nudges, provision of meat alternatives, description of meat or meat alternatives, and economic interventions.

Nudges
Nudges -aspects of choice architecture which alter behaviour in a predictable way without forbidding options or changing economic incentives ( Thaler and Sunstein, 2009 ) -can be used to prompt dietary behaviour change ( Bucher et al., 2016 ). Several reviews attest to the popularity and impact of nudges to reduce animal-product consumption (e.g., Bianchi et al., 2018 ;Harguess et al., 2020 ;Taufik et al., 2019 ;Veul, 2018 ). Our meta-review found estimates supporting interventions which reduce meat portion sizes (e.g., Reinders et al., 2017 ;Rolls et al., 2010 ) and involve default meat-free menus (e.g., Campbell-Arvai et al., 2014 ). These strategies may be effective as they do not require the consumer to actively and voluntarily change their behaviour, or even realise the need for behaviour change ( Veul, 2018 ). This has been supported in previous research on healthy eating choices -as eating is mostly habitual, and is vulnerable to self-regulation failures, changing the food environment is often more successful than strategies which attempt to directly change what people think or feel ( Duckworth et al., 2016 ;Wansink and Chandon, 2014 ).

Provision of meat alternatives
Another aspect of changing the environment is providing appealing alternatives. We found some evidence of an effect of providing people with meat alternatives to try (e.g., mycoprotein products). This intervention may be promising -especially considering that the enjoyment derived from eating meat is a key motivator behind meat consumption and that a lack of food alternatives is a reported barrier to reduction ( Valli et al., 2019 ). Alternatives such as cultivated meat may be particularly poised to address these barriers, yet our meta-review found no reviews that focused on how the provision of cultivated meat would impact the consumption of traditional meat. Although achieving mimicry and efficiency presents a significant challenge for the growing field before it can become accepted, industrialised, and economically feasible ( Post, 2012 ), research is needed to examine consumer willingness to substitute cultivated meat for traditional meat.

Description of meat and meat alternatives
There is some evidence that there is an effect of changing the description of meat meals, but mixed evidence on altering the description of plant-based meals. Highlighting the animal origins of a meal (e.g., referring to "beef " and "pork " dishes as "cow " and "pig "; Kunst and Hohle, 2016 ), and pushing back on implicit suggestions of meat-eating norms ( Stewart et al., 2016 ), show some promise. Yet, the outcome of changing the name of meat-free meals to more appealing alternatives, or of highlighting a plant-based meal as the "Chef's recommendation", is unclear ( Bacon and Krpan, 2018 ).

Pricing interventions
No estimates we found demonstrated that financial incentives decreased animal-product consumption. Neither providing financial incentives for healthy purchases nor changing the price of different meat serving portions produced results in the desired direction.

Techniques to increase dairy intake can also provide insights into how to decrease it
The most prevalent behaviour change technique assessed by individual studies in this space was nutrition education, but those interventions did not robustly increase dairy consumption ( Hendrie et al., 2013 ;Jung et al., 2016 ;Rouf et al., 2018 ). Techniques that have been found to increase dairy consumption include delivering interventions in a single environment (e.g., school), delivering interventions at the group level (i.e., not personalised), and involving taste exposure (e.g., providing dairy foods). Given the efficiency and cost-effectiveness of these interventions, it suggests group-setting interventions such as school milk programs could also involve plant-based alternatives (e.g., soy milk). We speculate that offering free tastings of plant-based substitutes might also reduce the demand for animal products, given the effectiveness of taste exposure and increasing familiarity with products ( Hartmann-Boyce et al., 2018 ). Together, the evidence from these reviews summarise what interventions increase dairy consumption and may guide us toward strategies to promote plant-based alternatives.

Limitations of included reviews and directions for future research
While well-conducted systematic reviews of rigorous studies can provide robust evidence ( Higgins et al., 2019 ), the reviews we identified failed to meet a number of recommendations that generally lead to robust, causal findings ( Shea et al., 2017 ). None of the reviews used best-practice methods for assessing quality of primary studies, assessed whether the quality of the included studies influenced their conclusions, or explored the funding in the included studies, and few reviews assessed publication bias.
In general, future reviews would benefit from improved methodology to provide more robust findings. First, many reviews did not implement transparent reporting. As a result, using even a simple reporting checklist like PRISMA Page et al., 2020 ) would increase the transparency of future reviews. Second, most studies did not conduct a quantitative synthesis. Quantitative meta-analyses can be conducted across heterogeneous participants, interventions, comparisons, and outcomes. Using analysis methods that quantify evidence strength across these measured sources of heterogeneity as well as unmeasured sources could help identify the settings in which these interventions may be effective ( Hedges et al., 2020 ;Mathur and VanderWeele, 2019 ). Meta-analyses are possible -indeed recommended -when there are even just two comparable studies ( Ioannidis et al., 2008 ), as was often the case in our included reviews. Adding a meta-analysis allows for a more robust exploration of heterogeneity and a better quantification of publication bias ( Pustejovsky and Rodgers, 2019 ).
Most reviews focused on reducing meat consumption, and few addressed other animal products, like eggs or dairy. The reviews which examined dairy focused on interventions to increase consumption. One review conducted a broad search of behavioural interventions to encourage more plant-based diets generally ( Taufik et al., 2019 ). Yet, all studies found were focused on meat, except one which examined dairy. This suggests that there is either a lack of primary studies, or that they are not detected by the search strategies in the reviews.
Most of the literature examined focused on Western, educated, industrialized, rich and democratic (WEIRD) populations. This may limit the generalisability of the findings due to differences in population perceptions and norms, as well as different needs. For example, increased consumption of meat in Africa may be necessary considering micronutrient deficiencies (e.g., zinc, iron, etc.) or financial stability (e.g., owning a cow or goat).
Context mapping revealed several limitations of included reviews. Firstly, they did not compare differences in how the source of information influences intervention outcomes. Future research could determine to what extent people are more persuaded by similar people or whether messages attributed to government or health organisations are more effective than animal activists. Several reviews compared the effectiveness of different channels, primarily in examining dairy consumption, and focused on comparing levels of personalisation (e.g., ( Hendrie et al., 2013 ;Marquez et al., 2015 ). Given the growth in social media use and influence, future research should explore how mediated communication (e.g., Facebook adverts) perform versus face to face, and whether the scale and cost benefits of mediated communication outweigh the presumed downside of less interpersonal connection. Reviews also did not assess if outcome types or study methodology were associated with effect sizes (e.g., if self-reported data led to larger estimates than purchase data, or if non-randomised studies reported smaller estimates than randomised studies). Findings on the intention-behaviour gap ( Faries, 2016 ), and differences between field and lab studies ( Dubois et al., 2020 ), suggest that there may be value in future research which addresses both of these questions.
The included studies themselves are also fraught with a series of limitations. Many relied on self-report measures and intended consumption rather than observable behaviours and actual consumption ( Harguess et al., 2020 ). This means that some of the findings may be driven by socially desirable responding. Similarly, the reviews that did rate the risk of bias in the included studies often found a preponderance of high-risk designs that are susceptible to bias (e.g., Valli et al., 2019 ). Primary studies on this topic would therefore also benefit from improved methodology and reporting. Specifically, a large number of criteria for risk of bias require better reporting or prospective registration of methods ( Higgins et al., 2019 ). If primary studies consult research quality information sources such as the EQUATOR network and use the standardised reporting checklist relevant for their design, then the field can increase transparency, reproducibility, and validity of their conclusions ( Simera et al., 2010 ).
Given randomised trials of dietary change interventions are often difficult to conduct, the field could still benefit from conducting quasi-experiments rather than correlational studies, more robust techniques for finding causal evidence from observational data (e.g., propensity score matching, IPTW, structural causal models, or directed acyclic graphs; Pearl, 2009 ), and meta-analysis methods that characterise the robustness of findings to potential unmeasured confounding ( Mathur and VanderWeele, 2020a ).
There are a variety of interventions that have been identified in correlational studies but have not been experimentally tested or systematically reviewed ( Hartmann and Siegrist, 2017 ). Behaviour change interventions that could promote sustainable diets and the adoption of more plant-based meals have been suggested by the World Resources Institute ( Attwood et al., 2020 ) and the Behavioural Insights Team (2020). Although these reports did not meet our inclusion criteria due to their unsystematic searches, they provide an indication of interventions that could be assessed. Exploring interventions that are known to be effective with other dietary behaviours (e.g., healthy eating) may also be a promising avenue.
Shah (2020) draws on best practice in health and global development research to argue that there could be considerable value in developing: i) a theory of change for influencing animal-product consumption, ii) the cost effectiveness of different interventions to influence consumption of animal products, iii) an information model to prioritise research, and iv) sector wide evidence grading standards. Future researchers and reviewers could consider implementing many of these suggestions.

Limitations of our overview of reviews
The findings of our meta-review are contingent on the quality of the included reviews and the quality of the studies included in those reviews, discussed above. Moreover, this review is limited to reviews with a systematic, reproducible method. Evidence regarding other interventions may be available in the form of non-systematic reviews, and primary studies not included in systematic reviews. For example, a review by Animal Charity Evaluators (2017) found that leafleting does not have a significant effect on reducing animal product consumption. Yet, as their review lacked a systematic, reproducible search strategy, it was not included. Similarly, the Playbook for Guiding Diners Toward Plant-Rich Dishes in Food Service provides 23 behaviour change interventions that could be implemented in the food service sector to encourage diners to select more plant-rich dishes ( Attwood et al., 2020 ). However, this report was excluded as the review results are not explicitly discussed (but are used to inform their recommendations).
There are also limitations regarding our quality appraisal and intervention categorisation. AMSTAR2 was developed to assess the quality of systematic reviews of interventions ( Shea et al., 2017 ). Although most of our included reviews were focused on interventions, some had a broader focus on exploring variables (e.g., Sanchez-Sabate and Sabate, 2019 ; Valli et al., 2019 ). Other quality appraisal tools may have been better suited to evaluating those few reviews. We also note that we did not use an established behaviour change technique taxonomy (e.g., Michie et al. 2013 ;Teixeira et al. 2020 ) to categorise interventions found in included reviews. Although this would have been ideal, we judged that many included interventions were not well described by those taxonomies; for example, information about health and environ-mental consequences might be classified as the same behaviour change technique under established frameworks but are important to separate for this review.

Conclusion
Reducing animal-product consumption could provide numerous benefits to society, from helping to address environmental destruction to reducing animal suffering to optimising health. Despite limitations in the quality of reviews available, there is some promising evidence that informing consumers about environmental consequences reduces meat consumption. Discussing health impacts, implicitly emphasising animal welfare, encouraging goal-setting and self-monitoring, and conveying positive social norms around plant-based diets are also promising strategies, albeit with a more limited evidence base. Future reviews should conduct quantitative syntheses where appropriate and examine interventions that influence animal-product consumption other than meat. Investigating and adopting interventions to reduce animalproduct consumption is one important factor in safeguarding planetary health ( Willett et al., 2019 ).

Declaration of Competing Interest
None.