1. Introduction
The phenomenon of agricultural protectionism has been extensively investigated by agricultural economists, and research suggests that a majority of United States (US) consumers favor measures that protect domestic producers from international competition (Beckman and Countryman, Reference Beckman and Countryman2021; Moon and Pino, Reference Moon and Pino2018; Tong, Pham, and Ulubaşoğlu, Reference Tong, Pham and Ulubaşoğlu2019). Concurrently, US policymakers have consistently expressed concerns regarding the demand for domestically produced food. In response to these concerns, they have progressively implemented policies designed to address these challenges (Christiaensen, Rutledge, and Taylor, Reference Christiaensen, Rutledge and Taylor2021; Xiang, Malik, and Nielsen, Reference Xiang, Malik and Nielsen2020). In January 2021, the Biden Administration issued an executive order that would further tighten federal procurement regulations to stimulate domestic production and encourage government purchases of domestic goods (Lynch, Stein, and Parker, Reference Lynch, Stein and Parker2021). However, the impacts of this “Buy American-produced Goods” program on consumer demand remain ambiguous.
In this study, we examine consumer reactions to informational nudges aimed at promoting local food products by conveying the main message behind the “Buy American-produced Goods” program. Additionally, we investigate how promotional marketing message influences consumers’ responsiveness to the informational nudge and the differential impact of this message across political partisanship. The responsiveness of consumers towards this type of informational messaging was measured by their willingness to pay (WTP) for a US-produced attribute.
Message framing has proven highly effective in influencing public attitudes toward multiple issues (Bolsen, Druckman, and Cook, Reference Bolsen, Druckman and Cook2014; Goodwin et al., Reference Goodwin, Raffin, Jeffrey and Smith2018; Pratto and Lemieux, Reference Pratto and Lemieux2001). Framing has been used in many scientific disciplines, such as economics, sociology, political science, and psychology (Guenther, Gaertner, and Zeitz, Reference Guenther, Gaertner and Zeitz2021). Framing involves highlighting specific contents of perceived reality in a message, making this content more prominent while pushing other aspects into the background (Entman, Reference Entman1993). Researchers have applied message framing to shape how issues are understood and interpreted by subjects, and to influence how individuals perceive problems, causes, moral implications, and potential solutions (Entman, Reference Entman1993).
The “information framing effect” on public opinion and consumer decisions has been demonstrated in several previous studies (Ahlert, Breyer, and Schwettmann, Reference Ahlert, Breyer and Schwettmann2016; Aschemann-Witzel and Grunert, Reference Aschemann-Witzel and Grunert2015; Chang and Lee, Reference Chang and Lee2009; van Esch, Cui, and Jain, Reference Esch, (Gina) Cui and Jain2021). The basic premise of the information-framing effect is that consumer perceptions of an event vary depending on the rhetoric used to convey the subject matter of the event. Subtle changes in message framing can have a considerable impact on how consumers perceive and react to the content, ultimately affecting their decision-making process in various aspects (Chang and Lee, Reference Chang and Lee2009; Frey and Meier, Reference Frey and Meier2004; van Esch, Cui, and Jain, Reference Esch, (Gina) Cui and Jain2021; Bolsen, Druckman, and Cook, Reference Bolsen, Druckman and Cook2014; Cui, Kim, and Kim, Reference Cui, Kim and Kim2021). This implies that US consumer preferences and behaviors surrounding the purchase of domestically produced foods could be influenced by the way a message is presented, and the same message can have a significantly different impact on decisions and attitudes based on the framing and the way the information is conveyed. Given the importance of context in the political sphere, it is crucial to investigate how framing influences consumers’ reactions to political messages intended to promote higher preferences for local food products.
US consumers have demonstrated their advocacy for a variety of domestic commodities (Cantillo, Martín, and Román, Reference Cantillo, Martín and Román2020; Moon and Pino, Reference Moon and Pino2018; Stampa, Schipmann-Schwarze, and Hamm, Reference Stampa, Schipmann-Schwarze and Hamm2020). At the same time, previous studies have shown that citizens might back or reject policies contrary to their usual stance, predominantly if such a stance is endorsed by their political party. This is largely attributed to a phenomenon known as partisan-motivated reasoning, where people interpret information based on their party affiliation (Baum and Groeling, Reference Baum and Groeling2009; Bolsen, Druckman, and Cook, Reference Bolsen, Druckman and Cook2014). Indeed, partisan cues, such as political leadership, have been shown to significantly impact public opinions and even change previously held views (Baum and Groeling, Reference Baum and Groeling2009; Bolsen, Druckman, and Cook, Reference Bolsen, Druckman and Cook2014; Goren, Federico, and Kittilson, Reference Goren, Federico and Kittilson2009; Satherley et al., Reference Satherley, Yogeeswaran, Osborne and Sibley2018). When individuals identify with a specific political party (or the policies espoused therein), they are more likely to follow the cues provided by their party’s leaders, aligning their opinions and policy preferences with those of the leadership. Given this increasing polarization, it would be unsurprising that consumer opinions toward certain policies can be volatile depending on political affiliations. It is therefore crucial to consider the influence of partisanship on individuals’ attitudes, especially when investigating the effectiveness of political messaging in promoting certain behaviors, in our case, higher WTP for US-produced foods.
This study holds significant implications for both policymakers and marketers seeking to influence consumer behavior and demand towards domestically produced food products. Our study investigated consumer preferences for US-produced foods (in our study, tomatoes) and demonstrated how WTP for various product attributes varies by political affiliations. Our research also emphasizes the importance of employing appropriate testing methods to assess differences in WTP estimates in choice experiments (CEs). Specifically, we used Complete Combinatorial test, t-test, Mann–Whitney test, and Bootstrap test to measure the effects of political framing. Our results provide insights for policymakers and marketers that for more effective communication with the public, especially in turbulent times, policymakers should consider adopting more nuanced and comprehensive communication strategies.
2. Survey and method
2.1. Survey instrument
A national survey was conducted between March and April 2021 via Qualtrics to collect US residents’ preferences for domestically produced crops, grocery shopping behaviors, and perceptions toward US-produced goods and services. The sample used in this study was targeted to be representative of the US population census statistics for gender, education, race, and income. We selected tomatoes as the focal product in our study for their significant role in US agriculture. Tomatoes rank the second most consumed vegetable in the US (Wu, Yu, and Pehrsson, Reference Wu, Yu and Pehrsson2022), with over half of the domestic consumption relying on imports (Kassas et al., Reference Kassas, Cao, Gao, House and Guan2023). This underscores the importance of exploring US consumer preferences for tomatoes with varying origins. A sample of 1,062 valid responses was collected. Eligible respondents were adults (18 years or older) who consumed tomatoes in the past three months and shared 50% or more of the grocery shopping responsibilities in their households. Attention-check questions were added to ensure high-quality responses. We also collected information on consumers’ political characteristics, including voting and political identification. Therefore, the interaction effect between political message frames and partisanship on respondents’ attitudes toward the US-produced attribute can be analyzed.
2.2. Information treatment
Respondents in this study were randomly assigned to one of three groups (a control and two treatments) to analyze the effects of promotional marketing message on preferences for domestically produced tomatoes. Unlike respondents in the control, those in treatments 1 and 2 were presented with messaging aimed at promoting US domestic goods and services. The message’s substance remained consistent across both treatments, mainly emphasizing governmental guidelines endorsing domestically produced goods and services. However, the treatments diverged in their framing of the presented information. The text for Treatment 1 (hereafter “Framed Information”) was cast as originating from President Biden, while Treatment 2 (hereafter “General Information”) was generalized as stemming from the US government at large. The content of the message in each treatment is presented in Table 1.
Table 1. Wording of the treatments
2.3. Choice experiment design
In order to emulate the decision-making process that consumers encounter when making purchases, we employed a discrete CE that presented participants with a sequence of choice situations. CEs are based on consumer demand theory, assuming that the consumers gain utility from a product’s attributes (Lancaster, Reference Lancaster1966). They have been widely used in consumer behavior research due to their well-recognized advantages (Lai, Widmar, and Bir, Reference Lai, Widmar and Bir2020; Lusk and Schroeder, Reference Lusk and Schroeder2004). We used SAS Macro to create an orthogonal fractional factorial design with a D-efficiency of 94% (Kuhfeld, Reference Kuhfeld2012).
Four attributes (Price, US, H2-A, Organic) were chosen to capture consumers’ preferences. The “Price” attribute, with six levels derived from a price series by the USDA (United States Department of Agriculture) Agricultural Marketing Service, represents the cost of purchasing the product. The “US” attribute is an indicator that the tomatoes were produced in the US, aligning with political initiatives’ objectives. The “H2-A” attribute signifies that all migrant workers employed on the farm producing these tomatoes were certified as legally hired through the US Department of Homeland Security’s H-2A visa program. The “Organic” attribute confirms whether the USDA has certified the product as organic. Detailed descriptions and classifications of all attributes are displayed in Table 2.
Table 2. Attributes and levels for the choice experiment
Given our attributes and attribute levels, a total of 18 choice sets were created and divided evenly into three blocks. Participants in each treatment were randomized across the blocks. Before subjects answered the CE, we showed them the detailed definitions and levels of each attribute and gave them an example choice set to help them better understand and respond. In a typical choice set, two different kinds of tomatoes and one opt-out option were presented. Participants were asked to make their selection to indicate which kind of tomatoes they would prefer to buy. The opt-out option was included to avoid forcing a choice of one of the tomatoes. Further, a cheap talk script was presented to subjects prior to starting the CE to reduce hypothetical bias (Lusk, Reference Lusk2003). Figure 1 provides an example of a choice set in the CE.
Figure 1. Sample choice task.
2.4. Theoretical framework and modeling willingness to pay
Data was analyzed with a Random Parameter Logit (RPL) model to estimate consumers’ WTP for US-produced tomatoes. In this case, the RPL model is superior to the standard logit model in multiple aspects (Train, Reference Train2009). The most significant advantage of the RPL model is its ability to capture unobserved heterogeneity in preferences across individuals. Further, RPL relaxes the assumption of the independence from irrelevant alternatives property that the odds of choosing one alternative over another are not affected by the introduction or removal of other alternatives, making it more flexible and realistic.
The principles of random utility theory model individuals as agents who aim to maximize their own expected utility in regard to the available choice options (Lancaster, Reference Lancaster1966; Morey, Rowe, and Watson, Reference Morey, Rowe and Watson1993). When individuals lack complete information, individual utility is regarded as a random variable (Kitamura and Stoye, Reference Kitamura and Stoye2018). The random utility theory posits that the utility derived from selecting option i at a specific time t from a set of all potential choices C for individual n is equal to the utility determined by the fixed attributes of that option (V it ), coupled with a random utility component, ε nit . This random element is presumed to be independent and identically distributed across all options and choice scenarios. Therefore, we can represent the utility function as:
$$U_{nit}=V_{nit}+ \varepsilon _{nit},$$
where V nit = β n ′X njt , β n represents a vector of coefficients that vary over n individuals, and X nit represents observed variables associated with alternative i for individual n during choice occasion t. In a given situation t, it’s assumed that an economic agent n will choose option i that offers the highest achievable level of utility. Hence, if the utility of picking option i at time t surpasses the utility of choosing any other option j at the same time (denoted as: U nit > U njt , ∀i ≠ j), then the probability of consumer n opting for alternative i can be described as:
$$P_{njt}=P\left(\varepsilon _{nit}-\varepsilon _{njt}\gt V_{njt}-V_{nit};\forall i\neq j,\forall j\epsilon C\right).$$
The deterministic component can be written as a function of product characteristics as follows:
$$V_{nit}=\beta _{1}x_{nit}+\beta _{2}x_{nit}+\ldots +\beta _{k}x_{nit},$$
where x nit represents a vector of characteristics present in option i, and the β values correspond to the parameters linked to the attributes of option i.
We then can model the deterministic component (V it ) as a combination of product attributes as follows:
$$V_{nit}=\beta _{1}{Price}_{nit}+\beta _{2}US_{nit}+\beta _{3}H2A_{nit}+\beta _{4}{Organic}_{nit}+\beta _{5}{OptOut}_{nit}.$$
Effects coding is chosen for analyzing categorical variables (i.e., attributes) in this study because it provides a balanced and nuanced view of each category’s impact. Unlike dummy coding, which uses 0 and 1 with one category as the reference (where 0 indicates the absence and 1 indicates the presence of a category), effects coding uses −1, 0, and 1. In effects coding, −1 represents a category’s deviation below the mean, 1 represents a deviation above the mean, and 0 indicates the category is not present. This approach helps researchers understand the effect of each category relative to the average, ensuring all categories are equally represented. The inclusion of the −2 factor in the WTP calculation accurately represents the categorical variable’s impact, offering more reliable insights. Bech and Gyrd-Hansen (Reference Bech and Gyrd-Hansen2005) provide a comprehensive discussion on advantages of effects coding in choice experiments, demonstrating its application and benefits. Based on our choice of effects coding, the mean WTP is calculated as follows:
$$WTP_{m}=-2*\left({\beta _{{attribute}} \over \beta _{{price}}}\right),$$
where β attribute is the coefficient of the non-price attribute, andβ price is the price coefficient. Additionally, we utilize the Krinsky and Robb method to generate 1,000 WTP values for the US attribute across each treatment in order to examine differences in mean WTP (i.e., the treatment effects). We then construct 95% confidence intervals for the WTP estimates pertaining to all attributes of our participants (Krinsky and Robb, Reference Krinsky and Robb1986).
3. Results
3.1. Descriptive statistics and treatment balance
A summary of respondent characteristics is presented in Table 3, along with US population characteristics for comparison.Footnote 1 As shown, our sample was representative of the population in gender, age, and income. About half the sample (49.9%) were female, and the majority (86.4%) identified as Caucasian/White. Age was distributed evenly across categories, while highest educational attainment leaned slightly towards a bachelor’s degree (30.5%). Total household income was distributed across various income brackets, with relatively higher fraction of respondents reporting income between $75,000 to $99,999 (16.3%) and $150,000 to $199,999 (16.3%) than other income categories.
Table 3. Summary statistics for demographic variables
Sample summary statistics are broken down by treatment in Table 4. To ensure balance between our treatments, a Kruskal-Wallis test was performed to test for differences in characteristics across treatments. This test is a robust non-parametric approach, ideal for evaluating the disparities among multiple samples originating from an identical distribution (Kruskal and Wallis, Reference Kruskal and Wallis1952). Its application in our study was crucial to assess the distinctiveness of characteristics across various treatments, ensuring a balanced experimental design. As shown by the p-values in Table 4, we do not observe any statistically significant differences in characteristics across treatment groups, which is a necessary criterion for obtaining unbiased estimates of average treatment effects.
Table 4. Balance of treatment table
3.2. Random parameter logit model estimates
Results from the RPL models are reported in Table 5. Four different regressions were estimated using data from the full sample, control group, Framed Information treatment, and General Information treatment, respectively. The estimated coefficients across all RPL models are highly significant. More precisely, the coefficients for Price and Opt-Out maintain a consistent negative value across all models, as anticipated. This suggests that consumers experience disutility as prices escalate or when they do not make a choice from the available options. Furthermore, the coefficient estimates for US, H2-A, and Organic collectively indicate that consumers are willing to pay positive premiums for each of these attributes. In addition, the standard deviation estimates are generally statistically significant, indicating heterogeneity in consumer preferences for the attributes included in the CE.
Table 5. Random parameter logit estimations
Notes: The level of significance is indicated by asterisks, with three asterisks representing a p-value less than 0.001, two asterisks representing a p-value less than 0.01, and one asterisk representing a p-value less than 0.05.
The WTP estimates for all attributes were calculated and reported in Table 6, along with corresponding 95% confidence intervals. Focusing on the US attribute, the mean WTP for each group is also displayed in the bar chart in Figure 2. Our full sample analysis reveals a mean WTP for the US attribute of $1.44 per pound of tomatoes. We find suggestive evidence of differences in WTP for US attribute across treatments. The General Information group exhibits the largest WTP ($1.54) and the Framed Information group exhibits the lowest WTP ($1.35). These figures are slightly higher and lower, respectively, compared to the control group’s WTP, which stands at $1.45 per pound of tomatoes.
Figure 2. Willingness to pay for the US attribute by treatment groups. Notes: The error bars presented are 95% confidence intervals.
Table 6. Derived mean willingness to pay (WTP) (in US dollars) estimates from base model
Notes: The 95% confidence intervals for each attribute’s WTP are presented in parentheses.
To evaluate treatment effects, we generated 1,000 bootstrapped WTPs for the US attribute within each treatment group by using Monte Carlo simulation technique of Krinsky and Robb (Krinsky and Robb, Reference Krinsky and Robb1986). Past research has employed various methodologies, such as the Complete Combinatorial Method, the Bootstrap test, the Mann–Whitney test, and the t-test, to examine the differences in the mean values of WTP obtained through bootstrapping or other resampling techniques (Gao, House, and Bi, Reference Gao, House and Bi2016; Lai, Widmar, and Bir, Reference Lai, Widmar and Bir2020). Therefore, we subsequently compared the bootstrapped WTP for the U.S. attribute between the control group and each of the two treatment groups using the four different tests previously adopted in the literature. The results, as shown in Table 7, were intriguing as the p-values derived from these different tests yielded varying outcomes. Specifically, the Complete Combinatorial test and the Bootstrap test indicated an absence of treatment effects, while the treatment effects were found highly statistically significant for both treatments under the Mann–Whitney test and t-test. However, the test results from the t-test and Mann–Whitney test are notably influenced by the number of bootstraps used. More simulations of WTP from the Krinsky and Robb method will lead to a lower standard error in the t-test and Mann–Whitney test, and thus further inflate Type I error. This can inadvertently boost the significance of both tests by increasing the bootstrapped WTPs. Therefore, it is important to caution against the use of the Mann–Whitney test and t-test as they may yield false positive results in such circumstances. Conversely, the Complete Combinatorial Method has been widely used by agricultural and applied economists for its simplicity and effectiveness as a computational approach for assessing differences between independent empirical distributions estimated via resampling (Poe, Giraud, and Loomis, Reference Poe, Giraud and Loomis2005). More importantly, due to the Complete Combinatorial Method’s non-parametric nature, the bootstrapped sample size (number of simulations) has no direct impact on the test score of this test. Rather, it simply counts the number of successes and is divided by the total count. Thus, the likelihood of the null hypothesis being rejected remains unaffectedFootnote 2 . The Bootstrap test, on the other hand, can be a potentially acceptable alternative to the Complete Combinatorial test since it leads to the same findings and conclusions. We also carried out estimations incorporating two interaction terms, namely “US * Framed Information” and “US * General Information,” where results can be found in Table A3 in Appendix. The coefficients for these interaction terms proved to be insignificant, which is consistent with the results obtained from the Complete Combinatorial test and the Bootstrap test.
Table 7. P-values for hypothesis tests of treatment effects on the US attribute
3.3. Subsample analysis
The political nature of our information treatments warrants investigation of differences in treatment effects based on respondents’ political affiliation (Carney et al., Reference Carney, Jost, Gosling and Potter2008; Jung and Mittal, Reference Jung and Mittal2020). In this light, it is reasonable to think that an individual’s political identity might interfere with the way they react to the political framing in the information treatments used in this study. We estimated RPL models separately for Democrats and Republicans in a subsample analysis. Three models were estimated for each political group, one for each treatment, and the results are reported in Table 8. The significance and magnitude of both the Price and Opt-Out attributes reaffirm the disutility experienced by both Democrats and Republicans from a price increase and inability to choose one of the alternatives, respectively. Notably, the coefficients for the US attribute are highly significant, suggesting that consumers, irrespective of their political affiliation, have a positive willingness to pay for domestically produced tomatoes. However, there are discernible differences when it comes to other attributes. Specifically, while Republicans typically exhibit a significantly positive premium for the H2-A certification (with the exception of respondents in the General Information group), such significant coefficients are absent among Democrats. This suggests that Democrat respondents are generally irresponsive to the H2-A certification. However, this could be due to an assumption that these workers are legal US citizens or residents. On the other hand, the coefficients for the Organic attribute reveal that Democrats, on average, are inclined to pay a premium for organic products, while Republican subjects are not.
Table 8. Random parameter logit estimations for different political parties
Notes: The level of significance is indicated by asterisks, with three asterisks representing a p-value less than 0.001, two asterisks representing a p-value less than 0.01, and one asterisk representing a p-value less than 0.05.
Mean WTP estimates for the US attribute, along with 95% confidence intervals, were calculated by political group and treatment (see Table 9). For the subsample of democrat respondents, we observe that the control group, who did not receive any information, exhibited a WTP of $2.00. The General Information group has the highest WTP at $2.07, while surprisingly, the Framed Information group has the lowest WTP ($1.71). In the subsample of republican respondents, we observe the average WTP for the US attribute was $1.26 for the control group. Respondents in the General Information and Framed Information treatments have a WTP of $1.53 and $1.32 for the US attribute, respectively. Bar graphs of the WTP for each subgroup are presented in Figure 3. Comparing across partisan groups, we find that on average, Democrats exhibit higher WTPs for the US attribute in all treatment groups compared to Republicans. However, as demonstrated in Table 10, looking at statistical significance, we find similar results to our full sample analysis. Namely, results from the Complete Combinatorial test and the Bootstrap test reveal no statistically significant differences in WTP for US attribute across any treatment for either partisan group. On the other hand, results from the Mann–Whitney test and t-test show high significance. This underscores the importance of choosing the appropriate statistical test when comparing WTP across different RPL models to avoid incorrect conclusions. To solidify the reliability of our results, we conducted additional robustness checks to examine individuals’ heterogeneous preferences, namely Latent Class Models (LCMs) and Quantile Regression (Koenker, Reference Koenker2005; Louviere, Hensher, and Swait, Reference Louviere, Hensher and Swait2000).Footnote 3 By comparing LCMs with different numbers of classes (i.e., 2–9), we determined that an LCM with three classes is most suitable. However, the regression results, as demonstrated in Table A5 in the Appendix, do not show significant outcomes for the interaction terms in any of the latent classes. Moreover, the results of regressing individuals’ WTP for the US attribute do not reveal any significance for either treatment variables in any of the quantiles (see Table A6). These supplementary analyses (reported in Appendix) corroborated our main findings, reaffirming the lack of statistical significance of the treatment effects.
Figure 3. Willingness to pay for the US attribute of democrats and republicans. Notes: The error bars presented are confidence intervals.
Table 9. Derived mean willingness to pay (WTP) (in US dollars) estimates from subsample analysis
Notes: The 95% confidence intervals for each attribute’s WTP are presented in parentheses. The last row shows the POE test p-values for each treatment compared to the control group.
Table 10. p-values for hypothesis tests of treatment effects on the US attribute from subsample analysis
4. Discussion
This study sought to understand how message framing and political affiliation influence consumer preferences and decisions with regard to US-produced foods. While it may initially appear that the General Information group has the highest mean WTP for the US attribute, followed by the control group, then the Framed Information group, appropriate statistical tests (i.e., the Complete Combinatorial test) revealed statistically insignificant differences in WTP for the US attribute between treatments. This is in contrast to other statistical tests (e.g., Mann–Whitney test and t-test), which have been used in a similar context in previous literature, but that are not suitable for this type of analysis. Relying on the results of such tests can lead to false conclusions regarding the significance of treatment effects (or significance of differences in WTP estimates across models). It is therefore critical to choose the statistical tests carefully when comparing parameter estimates between different random parameter logit models.
Results of full sample analysis, as well as sub-analysis across political partisanship, suggest that government interventions in the domestic market, at least in the form of political and promotional marketing message, do not have a significant influence on consumption of products such as tomatoes. The results observed in this study could be attributed to various factors and may not be universally applicable across different circumstances. In particular, the results of this study are bound by the context and timing surrounding the study period. The government announcement was made in January 2021, which was a tumultuous time following the 2020 Presidential elections and the challenges posed by the COVID-19 pandemic. The literature suggests that social instabilities, such as those mentioned above, could have an influence on consumers’ attitudes and decisions (Ji and Lee, Reference Ji and Lee2021; Kassas and Nayga, Reference Kassas and Nayga2021; Micah et al., Reference Micah, Cogswell, Cunningham, Ezoe, Harle, Maddison and Darrah2021; Skoufias, Reference Skoufias2003; Winsor, Reference Winsor1995). It is possible that government interventions could be more effective during a more stable period or when implemented in formats other than framed information in a real-world setting.
Moreover, the results of our study have certain limitations. First, while our sample aligns with U.S. census data in terms of gender and income representation, significant differences still exist in race and education. Specifically, our sample contains a higher proportion of white respondents, resulting in underrepresentation of all other racial and ethnic groups. The sample also appears to be more educated compared to US census data, with a greater proportion of respondents holding a college degree or higher. Further, the findings may only be applicable to domestically grown tomatoes. Certain products may inherently elicit less variation in consumers’ response based on political framing. Since the government announcement does not explicitly focus on agricultural products, the effectiveness of government intervention on other goods and services, such as automobiles, electronics, and clothing industries, might vary. The results may not necessarily apply to other domestic food products either. Fourth, the potential ambiguity in the wording or design of the information treatments may lead respondents to associate the General Information about the US Government with the current administration (President Biden), regardless of the specific framing intended. The treatments may not have been sufficiently distinct to differentiate political message framing effectively. Last but not the least, conducted during a period of heightened political tension or specific socio-political events, the general public’s receptiveness to political messaging may have been affected.
5. Conclusion
In this study, we employed a CE to investigate US consumers’ WTP for domestically produced tomatoes. Two treatment groups received general information and framed information prior to entering the experiment, respectively. This study sheds light on the complex relationship between message framing, political affiliation, and consumer preferences for domestically produced foods. Results suggest that consumers are willing to pay a premium for domestically produced tomatoes, while disparity exists in their WTP for H2-A and Organic attributes. In particular, Democrat respondents demonstrate a strong preference for Organic products, while Republicans do not show the same inclination. On the other hand, Republicans prioritize the legal employment of immigrant workers on farms, a concern that isn’t as prevalent among Democrats.
Our findings indicate that the type of promotional marketing message examined in this study did not significantly impact consumers’ preferences for local tomatoes. Specifically, results suggest that consumer responsiveness is unaffected by wording the message as originating from a broad political body vs. a specific politician (i.e., the US president). Our study also underscores the importance of using the appropriate testing methods to evaluate variations in bootstrapped WTP. Certain tests, like the Mann–Whitney test and t-test, which have been used in past studies, can potentially give misleading indications of positive effects and lead to incorrect conclusions. On the other hand, the Bootstrap test can be a suitable alternative to the Complete Combinatorial test as it leads to the same conclusions.
Neither of the treatments resonates with the Democratic or Republican respondents, underscoring the need for policymakers to explore more targeted and customized methods. These methods should be specifically tailored to resonate with distinct segments within political groups, especially during relatively tumultuous and challenging times. The findings also imply that traditional methods of information dissemination may not be as impactful. Public communicators should explore and test alternative strategies. Future research could focus on different categories of domestically produced products, alternative formats of framed information, or distinct time periods to further investigate the effectiveness of this type of government intervention or promotional marketing messages.
Data availability statement
The data that support the findings of this study are available from the corresponding author, Jianhui Liu, upon reasonable request.
Author contribution
Conceptualization, J.L.(1), B.K., and J.L.(2) ; Data Curation, J.L.(1), B.K., and J.L.(2) ; Formal Analysis, J.L.(1), B.K., and J.L.(2); Investigation, J.L.(1), B.K., and J.L.(2); Methodology, J.L.(1), B.K., and J.L.(2); Project Administration, J.L.(1), B.K., and J.L.(2); Supervision, B.K. and J.L.(2); Writing – Original Draft, J.L.(1); Writing – Review & Editing, J.L.(1), B.K., and J.L.(2).
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing interests
Jianhui Liu, Bachir Kassas, and John Lai declare none.
Appendix
Table A1. Illustrative examples of t-test outcomes between control and treatment groups
Notes: In the t-test, each simulated WTP in the control group and treatment is compared once. The number of simulations directly impacts the standard error. An increase in number of simulations results in a decreased standard error, which in turn raise the test score. This escalation directly influences the likelihood of rejecting the null hypothesis (the treatment effects).
Table A2. Illustrative examples of Complete Combinatorial Test outcomes between control and treatment groups
Notes: In the Complete Combinatorial Test, the number of simulations does not affect the outcome. This test involves comparing each simulation in the control group with all bootstrapped WTPs in the treatments. It simply counts the number of successes (i.e., the number of instances where WTPcontrol is less than WTP treatment) and divided by the total count.
Table A3. Random parameter logit estimation with interaction terms
Notes: The level of significance is indicated by asterisks, with three asterisks representing a p-value less than 0.001, two asterisks representing a p-value less than 0.01, and one asterisk representing a p-value less than 0.05.
The insignificant coefficients of two interaction terms (ie., “US * Framed Info” and “US * General Info”) indicate that the treatment effects are ineffective to consumers’ WTP for the US attribute, which is consistent with the results obtained from the Complete Combinatorial test and the Bootstrap test.
Table A4. Log-likelihood, AIC, and BIC values for a LCM with different number of classes
Table A5. Regression results for LCM with sociodemographic variables
Notes: The level of significance is indicated by asterisks, with three asterisks representing a p-value less than 0.001, two asterisks representing a p-value less than 0.01, and one asterisk representing a p-value less than 0.05.
The insignificant coefficients of two interaction terms (ie., “US * Framed Info” and “US * General Info”) indicate that the treatment effects are ineffective to consumers’ WTP for the US attribute in all three classes.
Table A6. Regression results for Quantile Regression including treatment variables
Notes: The level of significance is indicated by asterisks, with three asterisks representing a p-value less than 0.001, two asterisks representing a p-value less than 0.01, and one asterisk representing a p-value less than 0.05.
The insignificant coefficients of two information treatments (ie., Framed Info” and “General Info”) indicate that the treatment effects are ineffective to consumers’ WTP for the US attribute in any of the quantiles.
Open access
