Production process labels

Mandatory labels for production process attributes exist for processes that are viewed favorably by consumers (e.g., organic) and for processes that are viewed as less favorable by consumers (e.g., genetic engineering). Furthermore, depending on the nature of the mandatory label, the label can communicate properties for all of the ingredients in a food product (e.g., the USDA Organic Seal), but in other cases, the label communicates the presence of a trait for some, but not all, ingredients in a processed food product (e.g., genetic engineering). In the specific case of labels for genetically engineered food products in the United States, the mandate makes exemptions for genetically modified ingredients that have been highly refined.Footnote ² Most label mandates for production process attributes do not require disclosure of the attribute across the ingredients used in the final product. For more favorable process attributes such as those communicated by the USDA Organic Seal, many food manufacturers will voluntarily list the ingredients that carry the attribute; this may provide a halo effect and further boost consumer demand even though the presence of the USDA Organic Seal requires that all ingredients be organic.Footnote ³

There is substantial evidence that consumers discount genetically engineered food items compared to conventional analogs, and more recent work has shown that the same is true for gene-edited foods (Ortega, Lin and Ward Reference Ortega, Lin and Ward2022; Caputo, Lusk and Kilders Reference Caputo, Lusk and Kilders2020). Recent work by Caputo, Lusk and Kilders (Reference Caputo, Lusk and Kilders2020) indicates that consumers’ willingness to pay for products with gene-edited ingredients is lower than that for products with genetically engineered ingredients, yet they also found that there was a very low awareness among consumers regarding gene editing compared to genetical engineering. Our research shares some similarities with Caputo, Lusk, and Kilders (Reference Caputo, Lusk and Kilders2020), but we also provide some extensions to their work studying ingredient-specific effects and the impact of product claims. Caputo, Lusk, and Kilders (Reference Caputo, Lusk and Kilders2020) consider consumer response to fresh and processed foods that use a gene-edited ingredient (as we do), but they do not consider the response across different ingredients for a given food product. They also examine the role of claims associated with foods produced using gene editing (e.g., environmental claims), but do not examine the implications of different types of claims (e.g., environmental claims versus health claims).

There is also evidence that although consumers discount gene-edited products compared to conventional items, the discount is not as significant as it is for genetically engineered products (Marette, Disdier and Beghin Reference Marette, Disdier and Beghin2021; Ortega, Lin and Ward Reference Ortega, Lin and Ward2022; Muringai, Fan and Goddard Reference Muringai, Fan and Goddard2020). Findings from meta-analyses of numerous studies indicate that, in general, consumers are willing to pay premiums to avoid genetically engineered foods (Lusk et al. Reference Lusk, Jamal, Kurlander, Roucan and Taulman2005; Clare, Dominic, and David Reference Clare, Dominic and David2006; Dannenberg Reference Dannenberg2009). The extent of consumer aversion to genetically engineered foods varies among products and across consumers, and this variation depends largely on the direct benefits perceived by consumers, their trust in government and regulatory agencies, and their level of knowledge of each breeding technique (Muringai, Fan and Goddard Reference Muringai, Fan and Goddard2020; Ortega, Lin and Ward Reference Ortega, Lin and Ward2022; Caputo, Lusk and Kilders Reference Caputo, Lusk and Kilders2020).

It is much less common for food manufacturers to list details of the ingredients when a less favorable attribute is the subject of the label mandate. For example, when a food product carries the label mandate for bioengineering (a genetically engineered product), the ingredient list does not disclose which ingredients are genetically engineered. Furthermore, exemptions for certain highly refined ingredients provide food manufacturers with additional opportunities not to disclose ingredients derived from genetically modified crops. In light of this, it is plausible that this lack of clarity over which ingredients are genetically modified, or derived initially from genetically modified crops, could be confusing for some consumers and may be a source of ambiguity. Additionally, full disclosure of the genetic composition of all ingredients may dampen overall demand for the food product, especially if consumers assume that the label mandate applies to all ingredients in a manufactured food product. Such ambiguity effects would warrant a re-examination of how label mandates are implemented, and how consumers would respond to alternative label mandates that did not allow for ingredient exemptions.

Product labels

In addition to a labeling mandate that specifies a production process attribute, many other sources of information displayed on food packaging can affect consumers’ food choice decisions. These include a wide range of claims about the product (notably regarding health, but also claims related to the environment and product value), and certain claims are also subject to regulation in some countries (Lalor and Wall Reference Lalor and Wall2011). Research analyzing consumer demand for genetically modified (both genetically engineered and gene-edited) potatoes among Canadian consumers found that acceptance depended on the stated product benefits, the type of breeding technology used, and the brand (Muringai, Fan and Goddard Reference Muringai, Fan and Goddard2020). Their results indicate a higher willingness to pay for genetically modified potatoes in the presence of a claim about health attributes (such as reduced acrylamide produced when potatoes are fried) and improvements in environmental attributes (relative to genetically modified potatoes without these benefits). Colson and Huffman (Reference Colson and Huffman2011) found that consumers were willing to pay more for vegetables with intragenic modifications when accompanied by claims about increased nutrient content compared to conventional products. Other research finds that consumer acceptance of genetically modified foods increases when there are claims highlighting specific direct benefits for consumers, such as improved health (De Steur, et al. Reference De Steur, Buysse, Feng and Gellynck2013; González, Johnson and Qaim Reference González, Johnson and Qaim2009), lower levels of acrylamide content (Harkness and Areal Reference Harkness and Areal2018; McFadden and Huffman Reference McFadden and Huffman2017), improvements in taste (Loureiro and Bugbee Reference Loureiro and Bugbee2005), mitigation of world food shortages (Moon and Balasubramanian Reference Moon and Balasubramanian2003), and an increase in environmental amenities (Frewer, Howard and Shepherd Reference Frewer, Howard and Shepherd1996).

The use of claims on food packages is a topic that has been controversial (Nocella and Kennedy Reference Nocella and Kennedy2012). Food manufacturers have a long history of using claims to affect consumer decision-making, and as a result, there continue to be discussions on the role of public policy in governing claims for food products (Roe, Levy and Derby Reference Roe, Levy and Derby1999). Research shows that consumers are more accepting of genetic modifications when these food products carry claims or are otherwise linked to information that states implicit benefits for consumers or the environment compared to information about agronomical benefits accruing mainly to producers (e.g., pest resistance). Here we extend this literature to focus on the role of health and environmental claims made by food manufacturers that could plausibly be associated with gene editing. The health-related claim follows earlier work and highlights the potential for gene-edited foods to deliver human health benefits. The environmental claim focuses on the potential for gene-edited crops to combat food loss and waste. Our environmental claim is motivated by some real-world promotional messages that explicitly link genetically engineered apples and potatoes to reductions in food loss and waste. Specifically, promotion messages released with the commercialization of the Arctic® apple noted that it has the capacity to reduce food waste (see https://arcticapples.com/about-us/). In addition, messages by Simplot released with the commercialization of the Innate® potato highlighted its ability to combat food loss (see https://www.innatepotatoes.com).

Motivation for the selected food products used in our survey

Gene editing has been applied to a number of crops, including camelina sativa, casava, flaxseed, grapefruit, maize, mushrooms, oranges, potatoes, rice, soybeans, tomatoes, and wheat (Zhang et al. Reference Zhang, Massel, Godwin and Gao2018). Given the expansion of interest in using gene editing by plant breeding programs and the number of gene-edited crops that are being developed and commercialized (Menz et al. Reference Menz, Modrzejewski, Hartung, Wilhelm and Sprink2020), there is a need for more research that examines consumer acceptance of gene editing. Our survey asked subjects to consider gene-edited ingredients in one of two manufactured food products: potato chips and apple pie.

We chose these manufactured products because (i) they both include an unrefined fruit or vegetable crop as the primary ingredient, (ii) they both include vegetable oil which is typically considered to be a highly refined ingredient derived from genetically modified crops but exempt from the U.S. label mandate, and (iii) apples and potatoes are two crops that have witnessed commercialization of genetically engineered varieties.

More research is needed to carefully understand the market potential for crops produced with this technology, as well as the effect that label mandates may have on consumers of manufactured products that use gene-edited ingredients. In particular, we seek to better understand how consumers respond to mandates that disclose the presence of genetically modified ingredients (unrefined ingredients and highly refined ingredients derived from genetically modified crops). We are also interested in food product claims and the potential framing effects; our survey includes treatments to study consumer response to product claims that report the benefits of gene editing to human health and the environment.

A description of the survey

The data used in the analysis are from a longitudinal survey that asked a panel of consumers a wide range of questions related to food purchase and food consumption decisions.Footnote ⁴ The survey contained five sequential waves; wave 1 (n = 4,446) was distributed in mid-March 2020, wave 2 (n = 2,579) in late-March 2020, wave 3 (n = 1,613) in early-April 2020, wave 4 (n = 1,389) in late-April 2020, and wave 5 (n = 2,002) in October 2020Footnote ⁵ . Each wave contained a core set of recurring questions about food expenditures and food acquisition decisions (for more details see Ellison et al. Reference Ellison, McFadden, Rickard and Wilson2021), and then each wave included a unique set of supplementary survey questions. In wave 1, the supplementary questions elicited demographic details and food purchasing habits among the respondents. In waves 2 and 5 of the survey, we included unique questions to evaluate consumers’ responses to gene-edited food products. In wave 2, we included three questions to elicit consumers’ likelihood to purchase potato chips with and without gene-edited ingredients and with and without claims touting potential benefits from the gene-edited product. In wave 5, we replicated the supplementary questions used in wave 2, but here we used apple pie as the food product being studied.Footnote ⁶

Table 1 shows some descriptive statistics for the subjects from wave 2.Footnote ⁷ Reporting of the data from the U.S. Census Bureau (2020) does not align exactly with the variables we collected, which are shown in Table 1. However, there is substantial overlap between the summary statistics for our sample in wave 2 and the average values reported by the U.S. Census Bureau (2020) for gender, income, and region. Our sample is less racially diverse, older, and more educated. As shown in the Appendix, results from additional analyses of the data from wave 2 that include demographic variables as controls do not significantly affect our main results.

Table 1. Demographic characteristics for subjects in wave 2 ^a

^a There is substantial overlap between the summary statistics for our sample in wave 2 and the average values reported by the U.S. Census Bureau (2020) for gender, income, and region. Our sample is slightly less racially diverse, older, and more educated.

We pre-tested the survey with 10 respondents and hosted a focus group discussion to identify potential areas of confusion.Footnote ⁸ Our University’s Institutional Review Board granted ethical approval for the involvement of human subjects in our study (reference number blinded for review). All subjects in the survey gave informed consent via the statement: “I am aware that my responses are confidential, and I agree to participate in this survey”, where an affirmative reply was required to enter the survey; subjects were also able to withdraw from the survey at any time without providing a reason. In addition, we included a “cheap talk” screen at the start of the survey, and a quality check question within the survey, in an effort to reduce any hypothetical bias issues.

Survey questions

Figure 1 outlines the organizational flow for the survey questions included in waves 2 and 5. Each subject was asked, in three subsequent questions, to state their likelihood to purchase a food product on a 5-point Likert scale (where 1 = “Extremely Unlikely”, …, 5 = “Extremely Likely”). In wave 2 the food product is potato chips, and in wave 5 the product is apple pie. We used generic product names (i.e., no brand names) and package sizes to describe the products. We did not show product images to mitigate the confounding effects of brand-level preferences amongst the participants.

Figure 1. Between-subject survey design.

Note: The Manufactured product is either potato chips (sample sizes denoted with superscript p) or apple pie (sample sizes denoted with superscript a). The unrefined ingredient is potatoes for potato chips, and it is apples for apple pie; the highly refined ingredient is vegetable oil for both manufactured food products.

The first question (Q1) asks subjects about their likelihood to purchase the product with no gene-edited ingredients labeled. Data from wave 2 yielded 2,579 usable responses for Q1, and data from wave 5 yielded 2,002 usable responses from Q1.Footnote ⁹ In the second question (Q2), we ask subjects about their likelihood to purchase the same product shown in Q1 but with an ingredient labeled as gene-edited. Here approximately half of the subject pool is randomly selected to see the unrefined ingredient (potatoes or apples) labeled as gene-edited, and the other half sees the refined ingredient (vegetable oil) labeled as gene-edited. In the final question (Q3), we randomly split the samples in Q2 and asked subjects their likelihood to purchase the same product with the same gene-edited ingredient, but we included a claim that explains the potential benefits of the gene-edited ingredient. The third question that randomly divides the original sample into four groups creates the between-subject design in our survey. The order of the questions was not randomized but the subjects were randomly placed into the sub-questions for Q2 and Q3. We also performed a mean comparison of demographics across the treatment groups for wave 2, and the results indicate that there are no significant differences in terms of demographics across the four treatment groups.

In Figure 2, we provide an example of the three sequential questions that we asked subjects in one specific treatment group. The figure outlines the within-subject design of the survey. Here the subject is in wave 2 and asked about their likelihood to purchase potato chips; for Q2 they are randomly assigned to the group that is presented a label stating that the unrefined ingredient (potatoes) is gene-edited. Figure 2 also shows the third question (Q3) that subjects see in this specific case, which also includes information that the gene-edited potatoes have an environmental benefit due to their potential to reduce food waste.

Figure 2. An example of the within-subject survey questions.

Although not shown in Figure 2, subjects were also provided with the following details as part of the explanation about gene editing. These details were provided to subjects before they were asked to provide their response to the likelihood question in Q2. Below is the text shown to a subject (just prior to Q2) that was randomly placed in the group with gene editing applied to the unrefined ingredient.

Furthermore, prior to answering their likelihood to purchase response to Q3, subjects were given details about gene editing and the specific treatment. Below is the text shown to a subject (just prior to Q3) that was randomly placed in the group with gene editing applied to the unrefined ingredient and the treatment that discusses the potential environmental benefits of gene editing.Footnote ¹⁰

Consumer acceptance of gene editing across ingredients

Our design allows us to consider the effects of gene editing on consumers’ LTP across ingredients. This question is motivated by the policy debate surrounding the exemption of certain ingredients from the genetically engineered label mandate in the United States. In the following notation, we use subscript i to denote a respondent, j for an ingredient, k for a product claim, and l for a food product. The between-ingredient differences in LTP for a food product were determined by using the difference in Q2 and Q1 (each measured on a 5-point scale) as the dependent variable in a linear regression model that is specified in equation (1). As a reminder, and as described in Figure 1, Q1 asked LTP for a conventional version of the product and Q2 asked LTP for the same product with a gene-edited ingredient.

(1) $${\left( {{\rm{Q}}2 - {\rm{Q}}1} \right)_{il}} = {\delta _{0l}} + {\delta _{1l}}Oi{l_i} + {\varepsilon _{il}}$$

where ${\left( {{\rm{Q}}2 - {\rm{Q}}1} \right)_{il}}$ is the difference in LTP between Q2 and Q1 for respondent i and food product l. $Oi{l_i}$ is an indicator variable equal to one if respondent i was randomized to a group making choices for oil products, and ${\delta _{1l}}$ estimates the ingredient effects relative to the unrefined ingredient (potatoes in the case of potato chips and apples in the case of the apple pie), and ${\varepsilon _{il}}$ is a normally distributed error term with a mean of 0.

The results in Table 4 show no statistically significant ingredient effects in either food product. This is an important result as it highlights that consumer LTP in a food product decreases regardless of whether the modification is for unrefined or highly refined ingredients. This finding suggests that consumers may behave differently in the marketplace in the absence of ingredient exemptions that currently exist within label mandates.

Table 4. Estimated coefficients from the linear regression models to determine between-ingredient differences

Note: *** denotes a p-values < 0.01. Clustered standard errors are reported in parentheses. The baseline variable is the unrefined ingredient (Potato or Apple).

On the role of product claims

Next, we examine the consumer response to product claims that seek to communicate the potential benefits of gene editing. Within-group differences for LTP across the three questions were determined by estimating an ordered logistic regression model for each group. The LTP responses from the three questions were stacked, and the models can be specified by:

(2) $$\Pr (LTP_{ijkl}^*) = {\beta _{1jkl}}Q2 + {\beta _{2jkl}}Q3 + {\varepsilon _{il}}$$

where $LTP_{ijkl}^*$ is the latent score LTP for respondent i, ingredient j, claim k, and food product l. $Q2$ and $3$ are indicator variables equal to one for the associated LTP responses, ${\beta _{1jkl}}$ and ${\beta _{2jkl}}$ are estimated coefficients that determine whether LTP for Q2 and Q3 were different from LTP for Q1, and ${\varepsilon _{il}}$ is a logistically-distributed error term. Wald tests were conducted post-estimation to determine differences between Q3 and Q2 ( ${\beta _{2jkl}} - {\beta _{1jkl}}$ ).

Results in Table 5 show that LTP was less in Q2 and Q3 than in Q1 for all groups and ingredients for both food products. Additionally, the Wald test results show that information about the benefits of gene editing provided to respondents in Q3 increased the LTP relative to that from Q2. The magnitude of the Wald statistics in Table 5 highlights that there is a greater consumer response to the treatments focusing on health benefits. Therefore, in the next model, we examine the between-group differences in LTP for a food product. These were determined by using the difference in LTP in Q3 and Q2 as a dependent variable in a linear regression model that can be specified by:

(3) $$\eqalign{{\left( {{\rm{Q}}3 - {\rm{Q}}2} \right)_{il}} = & {\ \gamma _{0l}} + {\gamma _{1l}}Oil\_Healt{h_i} + {\gamma _{2l}}Product\_FoodWast{e_i} \cr & + {\gamma _{3l}}Product\_Healt{h_i} + {\mu _{il}}}$$

Table 5. Estimated coefficients from the ordered logistic regression models to determine within-group differences

Note: ***denotes a p-value <0.01. Clustered standard errors are reported in parentheses. The baseline variable is the LTP for the Conventional Product collected in Q1.

where ${\left( {{\rm{Q}}3 - {\rm{Q}}2} \right)_{il}}$ is the difference in LTP between Q3 and Q2 for respondent i and food product l. $Oil\_Healt{h_i}$ , $Product\_FoodWast{e_i}$ , and $Product\_Healt{h_i}$ are indicator variables equal to one if respondent i was randomized to the associated groups, and ${\gamma _{1l}}$ , ${\gamma _{2l}}$ , and ${\gamma _{3l}}$ estimate group differences relative to the base group ( $Oil\_FoodWast{e_i}$ ), and ${\mu _{il}}$ is a normally distributed error term with a mean of 0. Again, Wald tests were conducted post-estimation to determine differences between groups with estimated coefficients (e.g., ${\gamma _{1l}} - {\gamma _{2l}}$ ).

Results in Table 6 show the differences in LTP from Q2 to Q3 for groups relative to the base treatment where subjects were presented with gene-edited oil and information that it had the potential benefit of reducing food waste. For the subjects in groups that received information about the health benefits of gene editing (for potato chips or apple pie) relative to base groups, the results show a positive and statistically significant effect on the LTP. Wald test shows that, for potato chips, the treatment focusing on health benefits for gene-edited potatoes also had a statistically significant increase in LTP relative to the group exposed to the treatment describing the food waste benefits for gene-edited potatoes. From the Wald tests for the apple pie treatments, there was no strong statistical evidence of other group differences not observed in the linear regression model results in Table 5. The results in Table 6 have important industry implications as they suggest that consumers do respond to product claims and that the claims about health benefits carry greater weight than claims about environmental benefits from gene editing.

Table 6. Estimated coefficients from the linear regression models to determine between-group differences

Note: ***, **, and *denote p-values <0.01, <0.05, and <0.10 respectively. Clustered standard errors are reported in parentheses. The baseline variable is the refined ingredient (Oil) and the Food Waste claim treatment in Q3.

Our results align with, and extend, earlier work in this space. We find consumers responded more favorably to claims linking gene-edited foods to health benefits (versus claims related to food waste and the environment). Muringai, Fan, and Goddard (Reference Muringai, Fan and Goddard2020) also found that consumers are willing to pay more for gene-edited potatoes with health-related benefits (reduced acrylamide produced when potatoes are fried) compared with gene-edited potatoes that are resistant to pesticides. Overall, we find that U.S. consumers had lower LTP food with gene-edited ingredients, and this aligns with recent work, also focusing on apples, by Marette, Disdier, and Beghin (Reference Marette, Disdier and Beghin2021). In addition, Caputo, Lusk, and Kilders (Reference Caputo, Lusk and Kilders2020) also found a rebound effect in consumers’ willingness to pay for gene-edited foods after information highlighting potential benefits of gene editing was shared. Furthermore, Caputo, Lusk, and Kilders (Reference Caputo, Lusk and Kilders2020) find that demand for gene-edited foods varies across products and level of processing, but we add to research in this arena as we focus on differences across ingredients.

Additional results that include demographic variables

In the Appendix, we include three tables to showcase how our results are impacted by including demographic variables in the specification. In each case, we use demographic details for subjects in Wave 2 and provide a side-by-side comparison of the original results without control variables included (on the left) and with control variables included (on the right).

The first column in Table A1 replicates the results from Table 4 for Wave 2 on the left, and the second column shows the results from a specification that includes control variables for sex, race, income, age, education, SNAP usage, food pantry usage, and presence of children in the household. The results across the two columns are qualitatively similar. Table A2 compares the results from Table 5 for Wave 2 to a model that includes the same set of control variables, and Table A3 does the same exercise for the results in Table 6. The results in tables A1, A2, and A3 suggest that adding control variables for analysis of Wave 2 data does not alter the general thrust of our findings described above. Since we do not have demographic details for the subjects in Wave 5, and because we find that the results from Wave 2 are not significantly affected when control variables are included, we consider the results in Tables 4, 5, and 6 to be our main results.