Emotional priming for sustainable consumption? Differential effects on valuation of labelled chocolate

Our randomized controlled trial includes four information treatments about the coca sector and no control group. The format of the treatments are short videos which consist of several animated slides. We designed the videos in line with formats commonly used on social media like Instagram. Our four treatments are distinct because they differ in sustainability objective as well as emotional prime.

2022-01-16

2022-01-30

The most important outcome to test the stated expectations is the consumers’ marginal willingness-to-pay (MWTP) for each one of the varying sustainability attributes in a discrete choice experiment. The MWTP corresponds to the marginal rate of substitution (MRS) between one attribute and price.

The implicit price, MWTP of each attribute (or better MWTP for no certification or no claim to a certification or a claim), is calculated as the MRS with respect to price. The MWTP of each attribute is the ratio of the coefficient of each attribute and the coefficient of price. Its unit is Euro. We assume a linear utility function and no heterogeneity in MWTP for different characteristics of the observations. We use the Krinsky-Robb parametric bootstrapping method to calculate standard errors and 95% confidence intervals.

We consider five other outcomes. Two are related to the MWTP measure: utility and predicted unconditional and conditional market shares for each of 16 chocolate bars that are used in the discrete choice experiment.
We also consider other outcomes: the willingness-to-purchase, recall attention frequency of certified and uncertified sustainability claims on products, and emotions triggered by the treatments.

Utility is captured as the probability of consumers to choose a certain alternative. It is estimated using a multinomial logit model for each treatment seperately.
The calculation of the unconditional and conditional market shares is based on the coefficient estimates of the multinomial logit models. We fix the price at the median of the amount participants reported to spend on chocolate in the last week across treatments. We consider the market shares of each of the 16 types of chocolate bars and the no purchasing option for the four video treatments separately. We exclude the no-purchasing option for conditional market shares. We calculate the market shares for each treatment by substituting the estimated utility coefficients into the probability equation.
A supplementary outcome is the willingness-to-purchase, a factor based on the self-reported likelihood or willingness of purchasing of one of the following four types of chocolate each: either with Fairtrade certification, Rainforest alliance certification, claim regarding social responsibility, or claim regarding environmental sustainability. We adapt the measure of self-stated likelihood or willingness to purchase Fairtrade chocolate in the future from Hansen et al. (see attached document). We also adjust the measure to a seven-point Likert-Scale. We choose to have an option of indifference despite a possible framing effect to ensure that we allow for all possible preferences of the participants. We will perform principal component analysis in order to narrow down the statement batteries to the core of the concept.
For the outcome of recall attention frequency we integrate a 14-day recall question by asking, how often they have paid attention to a number of claims or certifications in the last 14 days whilst shopping, e.g. the Nutri-Score and GMO-free labels. We include certifications and claims not related to our study so that we do not prime participants towards the certifications of interest. The categorical outcomes include seven levels of frequency.
Lastly, we capture how participants feel after watching the videos at the end of the questionnaire in the first wave and in a similar manner in the second wave. For that purpose, we employ the Discrete Emotions Questionnaire by Harmon-Jones et al. that measures anger, disgust, fear, anxiety, sadness, desire, relaxation and happiness via four items each. We will perform principal component analysis to establish continuous factors of eight groups of emotions that are based on each four components. Based on the findings in our pilot study, we will reduce the number of emotions asked for in the final survey.

To answer whether emotional priming in information provision increases consumers’ valuation of sustainable foods, we collect panel data of German consumers via online surveys. We randomly assign four information treatments that differ in sustainability aspect or addition of emotional primes (pictures). We employ a discrete choice experiment and answers related to self-reported consumption behavior to measure the consumers’ valuation of chocolate bars.

The data collection will be conducted by the panel provider respondi between January and February 2022. We conduct a pre-test with the same quota and the same panel provider in January 2022 for about one week, the first survey wave will take place in January 2022 for about two weeks, and the second survey wave - about four weeks later - will take place in February for again about two weeks.
For the discrete choice experiment, we offer choices with four varying sustainability attributes at two levels and four levels of the attribute price for a 100-g bar of chocolate with unknown brand and flavor. The sustainability attributes include the presence of the label Fairtrade, the label Rainforest Alliance, a claim related to social responsibility, or a claim related to environmental sustainability.
Since we are only interested in the main effects of the sustainability attributes, we use a fractional factorial design. Due to the large number of choice sets, we assign the participants randomly to one of 32 blocks of choice sets that are fixed groups of choice sets. To counteract order effects, we vary the order of the choice sets and the order of the choices within each set. The participants are each exposed to eight choice sets in which they are asked to make a choice between three chocolate bars. Respondents have the option to indicate that they would not buy any of the chocolate bars as a fourth option.

The participants will be randomly selected among the pool of available participants in a panel provider’s respondent pool. Invitations to participate will be sent out according to the quota settings for age, gender, income and education in order to match the population in Germany in these characteristics.
Once the participants have agreed to take part in the survey and match the quota, their assignment to the treatment occurs randomly. The software UniPark for designing the online questionnaire will randomize the treatments, choice set blocks, and choice set order.

The randomization of the treatment is on the individual level.

No

N/A

3,500 participants or 28,000 (8 times 3,500) observed choices

875 participants in each of the four treatment arms

As our focus lies on the outcomes of the discrete choice experiment, we only conduct power calculations for the utility outcomes using a multinomial logit and assuming linear factors of all attributes. Using the approach by de Bekker-Grob et al. that meets the requirements of the complexity of discrete choice experiments, we estimated the minimum sample size required for each treatment group. We assumed a 95%-significance level \alpha, a statistical power level (1-β) of 80% and a multinomial logit model with only linear effects. We selected the previously mentioned chosen discrete choice design with 32 blocks, 256 total choice sets, 4 alternatives per choice set, and priors of either 0.01 or 0.05 of estimated attribute effects for the sample size calculations. As we do not have any results from pilot studies yet, we used a fixed belief about parameter values that ranged from 0.05 to 0.01 for all attribute effects. The sample size ranges from 42 for parameter priors at 0.05 to 1088 for parameter priors at 0.01. In case of the parameters around 0.05 our total sample of 3,500 participants would be large enough to accommodate the required sample sizes for each of the four treatment groups. However, if all parameters are as small as 0.01, we would have to consider to reduce the number of treatments to achieve an adequate sample size. In our pilot study we will elicit real parameter effect sizes to be able to calculate the required sample sizes and prior to that discrete choice designs more accurately.