Table 2 Summary of included studies

 Aschemann-Witzel (2010) [28]

Germany.

Repeated measures: non-hypothetical choice/purchase simulation. Conducted in a laboratory.

220 consumers.

Chi-squared test (proportion chosen carrying claim vs overall proportion not carrying claims).

 Aschemann-Witzel (2013) [29]

Germany.

Repeated measures: realistic purchase simulation. Conducted in a laboratory.

210 consumers.

One-sample T-tests: (proportion chosen carrying claim vs overall proportion not carrying claims).

 De Marchi (2016) [45]

USA.

Repeated measures: price (4 levels) x calories (3 levels) x health claim (with/without) x organic claim (with/without) x carbon trust logo (with/without). Online choice experiment.

173 primary food shoppers and consumers of yogurt.

Random parameter logit with an error component model.

 De-Magistris (2016) [36]

Spain.

Repeated measures: price (4 levels) x nutrient claim (absent, reduced fat claim, low salt claim). Setting unclear, conducted in-person, participants seated individually.

217 primary food shoppers.

Random Parameters Logit (RPL) model.

 Fernández-Polanco (2013) [37]

Spain.

Repeated measures: price (4 levels) x origin (2 levels) x harvest method (2 levels) x sustainability (2 levels) x health claim (2 levels) x safety (2 levels).

169 participants.

Heteroscedastic logit model.

 Gracia (2009) [38]

Spain.

Repeated measures: price (2 levels) x brand (2 levels) x nutritional information panel (2 levels x claim (2 levels).

400 food shoppers.

Logit model.

 Krystallis (2012) [42]

Greece.

Repeated measures: product type (2 levels) x claims (5 levels) x flavour (2 levels) x price (3 levels).

140 participants.

Heteroscedastic extreme value (HEV) model.

 Van Wezemael (2014) [54]

Belgium, France, the Netherlands, and the UK.

Mixed design: between groups (nutrition or health & nutrition claim exposure), within group (claim, no claim) x price (4 levels). Conducted online.

2400 beef consumers, 600 participants from; the Netherlands, Belgium, France, and the UK.

Multinomial logit (MNL) model, error component (EC) logit model.

 Ares (2010) [27]

Uruguay.

Repeated measures: type of yogurt (3 levels) x brand (3 levels) x price (3 levels) x claim (with/without).

104 yogurt consumers.

Multinomial logit model (MNL). MNL used to estimate part-worth utilities.

 Barreiro-Hurle (2010) [30]

Spain.

Repeated measures: price (4 levels) x nutrition labelformats (2 levels) x claims (1 nutrient comparison, 1 disease reduction).

800 participants, consumers of sausages and yoghurt.

Random Parameters Logit (RPL) model.

 Casini (2014) [33]

Italy.

Repeated measures: certification (4 levels) x site of production (4 levels) x health claim - (4 levels including no claim) x price (4 levels). Online survey.

260 Italian consumers.

Latent class choice model.

 Contini (2015) [35]

Denmark and Italy.

Repeated measures: price (4 levels) x origin/site of production (4 levels) x health claim (8 levels −3 relevant to Review).

2024 participants, 51% Denmark, 49% Italy.

Latent class model. Cluster analysis: 8-class model.

 Loose (2013) [44]

Australia.

Repeated measures: 8 attributes (levels ranging from 2 to 8): incl. Price (4 levels) and claims (3 levels). Conducted online.

1718 seafood consumers.

Scale adjusted latent class model.Aggregated multinominal logit model

 McLean (2012) [47]

New Zealand.

Repeated measures: 4 factorial design: brand (3 levels) x FOP label (3 levels) x claim (3 levels) x sodium content (2 levels). Screen-based.

500 participants with hypertension, 191 participants without hypertension.

Multinominal logit regression model

 Mohebalian (2012) [48]

USA.

Repeated measures: juice type (3 levels) x origin (3 levels) x health claim (2 levels) x price (continuous). Online survey.

508 participants.

Conditional logistic regression.

 Mohebalian (2013) [49]

USA.

Repeated measures: fruit type x price x product origin, x health claim. Online survey.

1043 participants. Study 1: 535 participants. Study 2: 508 participants.

Conditional logit regression.

 Kiesel (2013) [39]

USA.

Five differentiated labelling treatments over a period of four weeks in each of five supermarkets, targeting microwave popcorn products.

Supermarket details: five treatment stores.

Summary statistics and difference-in-differences.

 Roberto (2012) [52]

USA.

Randomised controlled experiment, between groups design (no label, Smart choices, a modified SC symbol with serving size). Conducted in a laboratory.

243 participants.

One-way ANOVA (continuous variables). Chi-squared tests (categorical outcomes).

 Belei (2012) [31]

The Netherlands.

Randomised controlled experiment, between groups design, 3 conditions (incl. With/without claim).

109 undergraduate students.

ANOVA.

 Carbonneau (2015) [32]

Canada.

Randomised controlled experiment, between groups design, 3 conditions (low fat, energy, no claim), take home meals.

160 women.

Mixed models for repeated measures used to compare impact of the experimental labelling groups on mean daily energy intake.

 Koenigstorfer (2013) [40]

Germany.

Study 2: 1 factorial experiment (with claim/without) but without being made aware of perceived serving size and not observed by interviewer, conducted in a University.

Study 2: 135 students.

ANOVA.

 Steenhuis (2010) [53]

The Netherlands.

Repeated measures: two conditions: with claim/without claim, 1 week washout period between. Conducted in a University.

31 female participants from the University community.

Paired sample t-tests.

 Wansink (2006) [14]

USA.

Study 1: Between groups design (with claim/without), conducted during a University open day.

Study 3: Between groups design (2 (regular versus low-fat label) × 3 (no serving label, “Contains 1 Serving” label, “Contains 2 Servings” label). Conducted in a cinema.

Study 1: 269 participants, students and their families visiting food science and human nutrition open day, aged 18 < .Study 3: 210 university staff, undergraduates, and graduate students.

ANCOVAs: consumption by label type (low fat versus regular).

 Ares (2008) [25]

Uruguay.

Repeated measures, factorial experimental design (4 × 4), resulting in a set of 16 food concepts.

104 participants.

ANOVA.

 Ares (2009) [26]

Uruguay.

Repeated measures: three categorical factors: type of functional ingredient (2 levels) x name of the ingredient (2 levels) x claim (3 levels - No claim, ‘Enhanced function’ claim, ‘Reduced disease risk’ claim).

83 participants.

ANOVA.

 Coleman (2014) [34]

UK.

Repeated measures, online survey.

122 volunteers.

ANOVA with a Bonferroni post-hoc test.

 Kozup (2003) [41]

USA.

Between subjects design: 2 (heart-healthy, no claim) ×3 (nutrition information level with control). Mail survey.

147 participants, primary shoppers of household.

Multivariate and univariate

 Lin (2015) [43]

Taiwan.

Between subjects design: randomly assigned to with or without claim.

300 students and office workers

ANOVA.

 Maubach (2014) [46]

New Zealand.

Repeated measures: 4 FOP summary indicators, ×3 nutrition profile levels, × 3 product claim levels (no claim, nutrient-content, health claim), ×4 flavours. Conducted online.

768 participants.

Odds ratio.

 Moon (2011) [50]

USA.

Between subjects design, randomly assigned to treatment: (1) FDA permitted health claims (2) same claim without FDA approval (3) no information. Online survey.

3456 participants.

Logistic regression, t-test.

 Orquin (2015) [51]

Denmark.

Between subjects design, realistic product photographs shown 1 at a time.

STUDY 3: 204 participants, recruited online.

Linear regression.