Consumers’ Willingness to Adapt to Shifting Fish Availability Due to Climate Change

Abstract

Rising ocean temperatures driven by climate change are impacting the distribution of fish stocks. In the Northeastern United States, fish scientists predict that well-known local species will shift further north and will be replaced by lesser-known southern species in the local waters. It is unclear whether New England seafood consumers will accept these unfamiliar species when they enter the market, posing a threat to the resiliency of fishing communities. This paper investigates how New England seafood consumers might respond to a shifting supply of seafood by conducting an online stated preference survey. The choice experiment leveraged in the survey revealed that, compared to Atlantic Cod, consumers are willing to pay less for the unfamiliar fish species. However, significant heterogeneity was detected in the consumers’ preferences for purchasing these species. We find the varying degree of willingness to pay being affected by factors such as the type of venues they purchase seafood from and whether they fish recreationally. Our results suggest there will be a challenge in marketing these species, although with proper marketing strategies and coordination among the industry, these challenges may be reduced.

1. Introduction

One of the major challenges confronting the commercial fishing industry is the shifting distribution of fish stocks driven by climate change. In the Northeastern U.S., this redistribution has been documented through both modeling and empirical studies [1], with northern species like Atlantic Cod moving out of the region while southern species like Atlantic Croaker becoming more prevalent [2,3]. These shifts have also been corroborated by interviews with southern New England fishermen, who have observed similar trends [4]. These distributional changes reflect the combined influence of climatic shifts alongside other factors such as fishing pressure, management regimes, habitat conditions, and broader oceanographic variability, which often interact in shaping species movement and abundance [5]. While the commercial fishing industry has the potential to adapt to this evolving landscape, much hinges on whether consumers will accept and demand these unfamiliar, newly abundant species.

There is a growing body of research investigating consumers’ perception and behavior towards unfamiliar seafood in the U.S. and elsewhere, but the findings remain inconsistent. For instance, 61% of Italian and 58% of Turkish consumers report purchasing unfamiliar seafood occasionally, yet only 22% do so in Croatia [6]. In Portugal, 43% of surveyed respondents expressed interest in unfamiliar species, but 44% admitted to rarely trying them [7]. In the U.S., studies have uncovered significant consumer resistance in New England [8] and Georgia [9].

A better understanding of consumer hesitation towards unfamiliar species is a critical first step toward devising strategies to stimulate demand for such species. Studies have highlighted several key barriers, such as uncertainty about the texture, flavor, and preparation of these species [6,7,8,9]. These findings suggest that unfamiliar species with culinary traits similar to familiar ones may be more easily accepted by consumers. Additionally, research suggests that a broader diversity of seafood consumption and purchasing from alternative food networks, such as independent seafood markets, farmers’ markets, and community-supported fisheries (CSFs), correlates with a greater likelihood of trying unfamiliar species [8,9,10,11]. Raising consumer awareness through information about the conservation and sustainability benefits of these species also plays a key role in increasing their willingness to pay [12]. That said, pricing studies using choice experiments show that consumers generally assign a lower value to unfamiliar species compared to familiar ones [8,12]. This price gap varies with consumers who have previously purchased unfamiliar species, showing a higher willingness to pay. Providing information such as eco-labels can also increase consumer interest in these species [12].

A substantial benefit of stimulating demand for unfamiliar seafood species is the promotion of fisheries diversification, which provides advantages to both the fishing industry and marine ecosystems [13]. Fisheries diversification has been identified as a strategy for sustainable fisheries management [14], as it has the potential to help mitigate overfishing of high-demand species and create economic opportunities for fishers by expanding market options [8,15]. By broadening consumer preferences to include a greater variety of locally abundant but underutilized species, fisheries diversification has the potential to alleviate pressure on overfished stocks, and harvest species that were once considered bycatch or of little market value can be integrated into commercial markets. However, achieving successful diversification depends on consumer willingness to purchase and integrate these lesser-known species into their diets [16].

This study focuses on how New England seafood consumers might respond to a set of unfamiliar species that are becoming more common in local markets as the distribution of species changes. By examining consumer preferences, this study helps inform how shifting supplies can be integrated into seafood markets through a demand-oriented approach. The need to diversify seafood markets in the region has become increasingly urgent as warming waters alter species distributions and traditional stocks. Reference [15] highlights several underutilized finfish species emerging in the Northeastern U.S. that hold potential for local market development, yet consumer unfamiliarity remains a major barrier to their adoption. Building on prior work examining consumer preferences for underutilized species and marketing strategies in the region [8], this study focuses specifically on New England consumers to evaluate how local attitudes and purchasing habits influence willingness to adopt climate-driven species.

Based on [1], we selected 14 such species for this study. These 14 species are black sea bass (Centropristis striata), blue crab (Callinectes sapidus), butterfish (Peprilus triacanthus), croaker (Micropogonias undulatus), smooth dogfish (Mustelus canis), summer flounder (Paralichthys dentatus), northern kingfish (Menticirrhus saxatilis), northern pufferfish (Sphoeroides maculatus), red drum (Sciaenops ocellatus), scup/porgy (Stenotomus chrysops), shortfin and longfin squid (aggregated as a single species) (Illex illecebrosus/Doryteuthis pealeii), Spanish mackerel (Scomberomorus maculatus), triggerfish (Balistes capriscus), and silver hake/whiting (Merluccius bilinearis). To assess the market potential of these species and identify key entry points, this study leverages an online survey using a discrete choice experiment. The study aimed to identify distinct consumer segments with the highest potential for successful market entry and adoption of these species. To inform this segmentation, the analysis (1) assessed New England consumers’ awareness and purchasing habits regarding the 14 seafood species under investigation and (2) evaluated consumers’ willingness to pay (WTP) for a subset of those 14 species (see Section 3.1). These objectives are evaluated within a Random Utility framework, allowing for empirical testing of how consumers trade off species familiarity, origin, and purchasing context when making seafood choices. By quantifying the relative importance of each attribute and estimating willingness to pay (WTP), the objectives are empirically studied through the discrete choice econometric framework.

The research findings offer empirical insight into the heterogeneity in seafood purchasing behavior, along with the socio-demographic factors that influence these choices. By examining the marginal effects of key consumer segments, our findings contribute to the growing body of knowledge on consumer acceptance of unfamiliar seafood. Ultimately, this research provides insights that could inform policies and industry strategies to help the New England commercial fishing sector simultaneously address the challenges of shifting seafood composition driven by climate change, foster an overall greater acceptance of unfamiliar fish species, and enhance fisheries diversification, which leads to more resilience and sustainable fisheries.

2. Empirical Application

2.1. Survey Design and Implementation

An online survey platform, Qualtrics, was used to assess New England consumers’ awareness, purchasing behavior, and value of unfamiliar seafood species. After two pilot tests to refine the design and ensure clarity, the final version was distributed via Amazon Mechanical Turk (MTurk) in the Fall of 2022. The survey, which can be found in the online supplement, took respondents approximately 12 to 15 min to complete. The survey was approved by the University of Rhode Island Institutional Review Board (IRB 1920-071). We obtained informed consent from all respondents in the study.

The survey was divided into five sections. The first section introduced the survey, including eligibility screening questions (participants eligible to take the survey must be 18 years of age, consume seafood, and be a resident of New England) and a consent form. The key eligibility criteria included restricting the study sample to New England residents, as these states encompass the coastline that runs along the ocean, which has experienced a greater temperature shift compared to other regions, and is the location where the unfamiliar fish species are predicted to become more abundantly available. Further restricting the sample to seafood consumers is essential to determine the true market for the species of interest, as the study seeks to determine the preferences and consumption of seafood. In the second section of the survey, participants were asked directly about their awareness, purchasing behaviors, and attitudes towards the 14 unfamiliar seafood species under investigation. This section was strategically placed preceding the discrete choice experiment to ensure that all respondents had a clear understanding of the unfamiliar fish species of interest, as this section showed images of each respective fish species. The second section of the survey also asked about respondents’ actual purchasing behavior of these unfamiliar fish species, helping ground respondents in their real-world behavior. The third section was the discrete choice experiment, described in detail in the methodology section, which included a series of hypothetical choice questions aimed to solicit willingness to pay while mitigating the hypothetical bias. Section 4 explored general seafood shopping behaviors and perceptions, and Section 5 collected demographic information.

2.2. Data Quality Maintenance

A number of studies affirm that MTurk can yield samples that are representative of the overall population [17,18,19] and that are comparable to other sampling collection methods [17]. However, it is not completely immune to data quality concerns, such as unrepresented samples [20]. A robust set of data quality mitigation techniques was deployed before, during, and after the survey to address specific MTurk data quality concerns as well as overall best practices when conducting an online survey. Specifically, only participants with high approval ratings (HITs > 90%) were able to take the online survey [20,21]. The HIT Approval Rating represents the proportion of completed tasks that are approved by Requesters. Higher HIT rates can help requesters target workers who have a record of higher quality performance performing tasks on MTurk. Detailed information regarding the survey, compensation, and expected completion time was provided upfront [22,23].

In the survey, a CAPTCHA Verification question was utilized as well as a VPN/VPS warning, which notifies participants that the survey uses a protocol to check that respondents are from within the U.S. and are not using a Virtual Private Server or Virtual Private Network [24]. A cheap talk script was included to warn the participants that the quality of their answers would be checked. Throughout the survey, there were two attention check questions and two short-answer questions embedded to identify bots and unengaged participants. For instance, one short-answer question asked respondents to list fish species that they regularly purchase, with specific instructions to format the response in lower case letters and separate species by commas—an approach aimed at flagging automated responses. Lastly, we checked the data quality after the completed survey was submitted by checking the mean completion time to eliminate “speeders” [24] and identifying duplicate submissions by reviewing associated IP addresses [25].

2.3. Final Data Set

The final data set comprised 503 completed survey responses that passed the quality check described in the previous section. The majority of participants reside in Massachusetts (39.4%), and slightly more than a quarter (26.5%) live in Connecticut. This distribution mirrors the actual population size shares of these two states in New England. The sample was also predominantly clustered in metropolitan areas, such as Boston, MA, and Providence, RI (Figure 1).

In terms of demographics, the survey sample closely matched the overall distribution of the New England population (

Table 1. Demographic statistics.

Variables	Survey Sample	New England
Race %
White	81.85	70.0
Black or African American	7.66	6
American Indian	1.41	0
Asian	6.05	5
Hispanic or Latino	7.22	13
Female %	54.4	51.0
Age (Median)	36	41.4
Education %
Less than HS	0.2	8
HS or Equivalent	12.85	25
Some College or Associates	25.9	23
Bachelors	38.38	24
Graduate or professional	22.22	20
Median Household Income	$62,500	$92,017
Political Views %
Republican	13.74	N/A
Democrat	46.67	N/A
Independent	33.33	N/A
Prefer not to Answer/Other	6.26	N/A
Married	51.7%	49.0%

Notes: N = 503. New England data is sourced from census data reported for the year 2023. The survey median income is the average of the median income bracket $50,000 to $74,999.

). However, the sample tended to be slightly more White, better educated, and had a lower household income than the regional average, patterns commonly observed in MTurk samples [20]. Demographic comparisons were made using data from the 2023 New England Division census reporter [26].

3. Methodology

3.1. Choice Experiment Design

Our main strategy in assessing consumers’ demand for unfamiliar seafood species is to estimate their WTP using the discrete choice experiment (DCE) method, a workhorse tool for measuring WTP [27]. Our choice experiment included four attributes: fish species, origin (local/non-local), store type, and price (

Table 2. Choice experiment attributes and associated levels.

Attributes	Levels
Fish species	Scup (porgy/porgies)
	Fluke (summer flounder)
	Triggerfish
	Crocker
	Atlantic Cod
Origin	Local (please use your definition of local)
	Non-local
Store type	Large chain grocery store (e.g., Shaw’s/Walmart)
	Local grocery store (e.g., family-owned, not a large chain)
	Farmers’ market
	Seafood shop
Price ($ per pound)	From $10.49 to $18.49 in $2.00 intervals (total of five levels)

Note: Price represents an ordered continuous variable ($10.49–$18.49 in $2 increments), whereas Fish species, Origin, and Store type are nominal categorical attributes.

).

The four fish species chosen for the experiment were selected from the larger group of 14 unfamiliar seafood species predicted to increase in abundance in the New England waters due to rising ocean temperatures. This was done primarily to reduce the dimension of the choice experiment design to a more manageable size, as well as to include species with varying unfamiliarity to New England consumers, and to have both white and non-white fish filets. Atlantic Cod, one of the most popular types of fish consumed in the U.S. [28], was included as a benchmark species for comparison. In the choice sets, the fish species were presented in the filet option, accompanied by both their respective names and images to enhance the realism of the hypothetical scenarios (Figure 2).

The store type and locality attributes, along with their respective levels, were selected based on the relevant literature [29] and to create a realistic purchasing scenario. The payment vehicle was the price per pound of fish, and its range was carefully selected to reflect the known prices of Fluke and Cod ($10.49 to $18.49) at the time the study was conducted.

To incorporate all levels in every attribute in a manageable DCE module, we employed the D-efficient design using Stata command D-create [30]. The design included 20 choice sets divided into five blocks. Each respondent was randomly assigned to just one block to minimize survey fatigue. Each choice set contained two seafood filet options with a third ‘opt out’ option.

Given that the design of a choice experiment can influence results, careful attention was paid to the information provided to participants. Clear and concise instructions were given before the experiment began, explaining the hypothetical nature of the task while encouraging respondents to make decisions as if they were actually purchasing seafood in New England. To address potential concerns of hypothetical bias, an ex-ante approach was employed [31], urging participants to consider their choices seriously, even though no real money or fish filets were involved.

3.2. Econometric Model for Estimating WTP

Choice experiments are based on two established theories; one is the random utility theory, which is based on the rational individual assumption, their decision will lead them to obtain the highest level of utility when choosing among the available option sets [32]. Another is the Lancastrian consumer theory [33], which states that a good is a bundle of attributes that consumers derive utility from, and hence, it is possible for an analyst to uncover shadow prices that economic actors are willing to pay for each attribute or characteristic of a certain good. In this context, WTP is defined as the maximum monetary amount an individual is willing to forego for a marginal improvement in an attribute while maintaining constant utility—that is, the marginal rate of substitution between the attribute and price. These advantages made the choice experiment an ideal method for estimating WTP for food products and beyond.

Our theoretical framework follows the standard model in the literature. We begin by defining a random utility model where the indirect utility function of individual i selecting the option j at a choice occasion t is written as:

$$U_{ijt}=V_{ijt}+{ϵ}_{ijt} i=1,\dots ,N; j=1,\dots ,J; t=1,\dots ,T.$$

The deterministic component of the above utility model, $V_{ijt}$, is assumed to take a linear functional form $V_{ijt}={\mathsf{\beta }}_i^{\prime }{\mathbf{X}}_{ijt}, i=1,\dots ,N; j=1,\dots ,J; t=1,\dots ,T.$ ${\mathsf{\beta }}_i$ is a vector of individual-specific coefficients to be estimated and ${\mathbf{X}}_{ijt}$ is a vector of observed attributes relating to individual i and alternative j on choice occasion t. The density of ${\mathsf{\beta }}_i$ is denoted as $f\left(\mathsf{\beta }\right|\mathsf{\theta })$, where $\mathsf{\theta }$ are the parameters of the distribution. Finally, the second term in the utility model, ${ϵ}_{ijt}$, is the random term assumed to be independently and identically distributed (iid) extreme value [34].

Conditional on knowing ${\mathsf{\beta }}_i$, the probability of individual i choosing alternative j on choice occasion t, denoted as $s_{it}$, can be written as:

$$\mathrm{Pr}\left(s_{it}|{\mathsf{\beta }}_i\right)={\displaystyle \frac{\mathrm{e}\mathrm{x}\mathrm{p}\left({\mathsf{\beta }}_i^{\prime }{\mathbf{X}}_{i,s_{it},t}\right)}{{\sum }_{j=i}^J\mathrm{e}\mathrm{x}\mathrm{p}({\mathsf{\beta }}_i^{\prime }{\mathbf{X}}_{ijt})}},$$

which is the conditional logit formula [32]. Using this expression, the probability of the observed sequence of choices conditional on knowing ${\mathsf{\beta }}_i$ is:

$$\mathrm{Pr}\left(s_i|{\mathsf{\beta }}_i\right)=\prod _{t=1}^T\left[{\displaystyle \frac{\mathrm{e}\mathrm{x}\mathrm{p}\left({\mathsf{\beta }}_i^{\prime }{\mathbf{X}}_{i,s_{it},t}\right)}{{\sum }_{j=i}^J\mathrm{e}\mathrm{x}\mathrm{p}({\mathsf{\beta }}_i^{\prime }{\mathbf{X}}_{ijt})}}\right].$$

Integrating the above equation over the distribution of ${\mathsf{\beta }}_i$ yields the unconditional probability of the observed sequence of choices

$$\mathrm{Pr}\left(s_i|\mathsf{\theta }\right)=\int \mathrm{Pr}\left(s_i|{\mathsf{\beta }}_i\right)f\left(\mathsf{\beta }|\mathsf{\theta }\right)d\mathsf{\beta }.$$

The log-likelihood for the model can be written based on the above equation as:

$$LL\left(\mathsf{\theta }\right)=\sum _{i=1}^N\mathrm{l}\mathrm{n}[\mathrm{Pr}\left(s_i|\mathsf{\theta }\right)].$$

The equation $LL\left(\mathsf{\theta }\right)$ cannot be solved analytically, and as such it is approximated using simulation methods [35]. In this study, we used the Stata command -mixlogit- to do the simulation and estimate the vector of parameters ${\mathsf{\beta }}_i$.

The specification of $LL\left(\mathsf{\theta }\right)$ is general in the sense that it allows fitting models with both individual- and alternative-specific covariates. Additionally, this random parameter logit model allows (a) heterogeneous preferences among individuals; (b) unrestricted substitution patterns; and (c) correlations in unobserved factors over time. For these reasons, we decided to employ the random parameter mixed logit model (RPL) for this study.

Our base model, henceforth called Model 1, is shown below with a dichotomous dependent variable $Y_{ijt}$ denoting whether the alternative j was chosen by individual i on choice occasion t ($Y_{itj}=1)$ or not:

$$Y_{ijt}= {ASC}_{ijt}+{\beta }_i{Price}_{jt}+\sum _{k=1}^4{\gamma }_{ik}Specie{s}_{kjt}+\sum _{l=1}^3{\delta }_{il}Stor{e}_{ljt}+{\lambda }_i{Local}_{jt}+ {ϵ}_{ijt }$$

(1)

where $Specie{s}_k=${scup, croaker, triggerfish, fluke} and $Stor{e}_l=$ {local grocery store, farmers’ market, seafood shop}, and the ASC is the alternate specific constant controlling for the opt-out option. All covariates, with the exception of Price, are dichotomous dummy variables. We employed a dummy coding scheme, with base reference categories being Atlantic Cod for Species, large chain grocery store for Store, and non-local seafood for Local. We did not prescribe what local seafood means, but rather asked respondents to use their own subjective definition of local seafood. This decision was based on (a) the scope of this study is not how consumers perceive ‘local’, and (b) it was more important for respondents to perceive the seafood as local when it is stated as such in DCE. The opt-out ASC was coded as 1 for the opt-out alternative and 0 otherwise. While price was treated as a continuous variable. Following the literature, all coefficients, with the exception of ${\beta }_i$ associated with Price, were specified to be normally distributed, while ${\beta }_i$ was specified to be log-normally distributed, hence, the WTP of an attribute is computed as [36,37,38,39]

$${WTP}_{\mathsf{\Omega }}={\displaystyle \frac{\mathsf{\Omega }}{\exp \left(\beta +{\sigma }^2/2\right)}}$$

(2)

where $\mathsf{\Omega }$ is one of the estimated parameters of interest, i.e., $\mathsf{\Omega }\in \{{\gamma }_1,\dots ,{\gamma }_4, {\delta }_1, \dots , {\delta }_3, \lambda \}$. Here, Ω denotes the population-level mean of the random parameter distribution, whereas β_i, γ_ik, δ_il, and λ_i in Equation (1) represent individual-specific draws from these distributions. It is important to note that if a log-normal distribution is assumed, then the interpretation of the estimated parameters (β) and standard deviation (σ) is not straightforward; the mean can then be calculated as seen in the denominator of Equation (2) [40].

Model 1 is extended in two ways, each one intended to investigate a distinct theme. The first extension, Model 2, includes two two-way interaction terms to investigate how the choice preference for each fish species is affected by its locality and when purchased at a farmers’ market (FM) [11]. The specification of Model 2 is:

$$\begin{gathered} Y_{ijt}= {ASC}_{ijt}+{\beta }_i{Price}_{jt}+\sum _{k=1}^4{\gamma }_{ik}Specie{s}_{kjt}+\sum _{l=1}^3{\delta }_{il}Stor{e}_{ljt} +{\lambda }_i{Local}_{jt} \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+\sum _{k=1}^4{\eta }_{ik}Specie{s}_{kjt}\times {FM}_{ijt}+\sum _{k=1}^4{\mu }_{ik}Specie{s}_{kjt}\times Loca{l}_{jt}+ {\epsilon }_{ijt}, \end{gathered}$$

(3)

where ${\gamma }_k$ and $\lambda$ represents the main effects and ${\mu }_k$ captures the interaction effect. It is well documented in the literature that there can be issues associated with the effects and misinterpretation of the coefficient on the interaction term in nonlinear models such as logit [41,42]. In order to account for this in the WTP calculations, point estimates, significance levels, and confidence intervals were constructed for the nonlinear combinations of estimators using the delta method.

The WTP for species k that is sourced locally can be computed as seen in Equation (4) [37,43,44]. To calculate WTP for species k purchased at the farmers’ market, the numerator in Equation (4) will become ${\gamma }_k+{\delta }_{l=FM}+{\eta }_k$ and the denominator will remain unchanged.

$$WT{P}_{k,Local=1}={\displaystyle \frac{({\gamma }_k+\lambda +{\mu }_k)}{\exp \left(\beta +{\sigma }^2/2\right)}}$$

(4)

The second extension, Model 3, includes interactions between the fish species and a set of control variables of interest to investigate whether the estimated WTP will differ across these controls. The control variables include whether the respondent previously purchased the unfamiliar fish species included in the model (yes/no) (C1); the number of unfamiliar species purchased previously (C2); whether the respondent purchased from alternative food networks (C3); and whether the respondent is a recreational fisher (yes/no) (C4). A separate estimation was run for each control variable:

$$\begin{gathered} Y_{ijt}|C_z= {ASC}_{ijt}+{\beta }_i{Price}_{jt}+ \sum _{k=1}^4{\gamma }_{ik}Specie{s}_{kjt}+\sum _{l=1}^3{\delta }_{il}Stor{e}_{ljt}+{\lambda }_i{Local}_{jt} \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+{\nu }_i{C}_{z }\times {ASC}_{ijt}+\sum _{k=1}^4{\omega }_{ik}Specie{s}_{kjt}\times C_z+ {\epsilon }_{ijt}; C_Z=\left\{C1, C2, C3, C4\right\}. \end{gathered}$$

(5)

The resulting WTP of species k given the control variable $C_Z$ is computed as [45];

$$WT{P}_k={\displaystyle \frac{({\gamma }_k+{\omega }_k)}{\exp \left(\beta +{\sigma }^2/2\right)}}.$$

(6)

As described in the next section, we found statistically significant heterogeneity in our estimated RPL coefficients, suggesting that individual preference heterogeneity does exist. To investigate the nature of identified heterogeneity, we applied the latent class model (LCM) using the choice experiment data following [46].

4. Results

This section will first present the seafood purchasing behavior of our survey respondents for the 14 unfamiliar seafood species. The analyses include descriptive analysis to describe the overall trends, followed by a more detailed investigation using an ordered logistic regression. Then, the results of regressions using the choice experiment data will be presented. The interesting patterns regarding the purchasing behavior of the unfamiliar seafood species will be integrated into our inferences of the choice experiment results.

4.1. Purchasing Behavior of the Unfamiliar Seafood Species

No one will be surprised to find that the share of survey respondents who have previously purchased the 14 unfamiliar species is very low across all species (orange bars in Figure 3). One possible explanation for this trend is that consumers simply do not know, or are not aware, of these species. However, we find that many respondents are highly aware of some of these 14 species (blue bars in Figure 3), including blue crab (78.9%), flounder (65.1%), and black seabass (64.1%). Even the least-known species, such as red drum and triggerfish, approximately 30% of respondents knew about them.

Another interesting finding regarding the purchasing behavior of the 14 unfamiliar seafood species is that the species purchased most regularly were neither the ones they were most aware of nor the ones with the highest rate of being purchased. Among the species respondents had previously purchased, the most frequently bought on a weekly or bi-weekly basis were butterfish (25.6%), triggerfish (23.8%), red drum (18.2%), and Croaker (18.2%). Conversely, species like flounder (9.2%), silver hake (10%), and blue crab (11.5%) were purchased far less frequently (Figure 4). This divergence underscores a gap between the species awareness and regular purchasing behavior of the unfamiliar seafood species.

We also identified the motivations behind why respondents initially purchased unfamiliar seafood species. For 10 of the 14 species surveyed (black sea bass, blue crab, Atlantic Croaker, dogfish, summer flounder, pufferfish, red drum, squid, and triggerfish), the most frequently cited reason for the initial purchase was a recommendation from an employee at the seafood counter. This finding suggests that a relatively straightforward and effective strategy for expanding the market for these lesser-known species could be to enhance the provision of information at the point of purchase, particularly through informed and targeted recommendations from seafood counter employees. Overall, nearly half of the sample (48.6%) reported never having purchased any of the 14 unfamiliar fish species, while just under a third (29.1%) had purchased one to two of these species. The remaining 22.3% of respondents had purchased three or more unfamiliar species.

To better understand the factors influencing consumers’ propensity to try unfamiliar seafood, we ran an ordered logistic regression examining the relationship between consumer behaviors and characteristics and the total number of unfamiliar species purchased. The dependent variable is the number of unfamiliar species purchased in the past, categorized into none = 0, one or two species = 1, and three or more species = 2. The explanatory variables included were variables capturing seafood consumption behaviors, involvement in the seafood industry, and general demographic controls.

The analysis revealed several statistically significant predictors of purchasing a higher number of unfamiliar species (

Table 3. Likelihood of purchasing unfamiliar seafood species.

Variable	Coefficient
Recreational fisher (yes = 1)	0.648 *** (0.222)
Employed in seafood industry (yes = 1)	0.771 ** (0.343)
Seafood consumption frequency (couple of times per year)
Once per month	0.399 (0.449)
Couple of times per month	0.864 ** (0.37)
1 to 2 times per week	1.082 *** (0.378)
More than 2 times per week	1.013 * (0.53)
Whole fish frequency (never)
Not often	0.649 *** (0.225)
Often	0.574 * (0.318)
Purchasing location (yes = 1)
Large grocery store	−0.812 ** (0.331)
Wholesale	0.495 ** (0.229)
Local grocery	0.195 (0.195)
Alternate Networks	0.514 ** (0.223)
Restaurant	0.134 (0.203)
Seafood Shop	0.42 ** (0.205)
Convenience store	0.685 (0.463)
White	−4.21 * (0.256)
Cut point 1	1.271 (0.76)
Cut point 2	2.93 (0.77)
Number of observations (N)	487
Pseudo r-squared	0.153
Chi-square	155.746
Akaike information criteria (AIC)	926.933
Bayesian information criteria (BIC)	1065.145

***, **, * denote statistical significance at 1%, 5%, and 10% levels, respectively.

). These included being a recreational fisher, having previous or current employment in the seafood industry, consuming seafood more frequently, consuming whole fish more often, and purchasing seafood from wholesale outlets, seafood shops, or alternative networks, all of which statistically significantly increase the odds ratios of purchasing a higher level of unfamiliar seafood species. To ensure the reliability of the order logistic results, the proportional odds assumption, also known as the parallel regression assumption, must be satisfied. Using the Stata Commands model, we confirmed that this assumption does hold for our data. These findings are intuitive: individuals with closer ties to seafood, whether through work, recreational activities, or dietary habits, are more inclined to try unfamiliar species.

These insights provide actionable strategies for expanding the market for unfamiliar seafood. Specifically, targeting consumers at venues like seafood shops, farmers’ markets, and direct-from-dock sales may prove effective. Additionally, focusing on specific consumer subsets, such as recreational anglers, could offer a more impactful and straightforward approach to increasing demand for these lesser-known species.

4.1.1. Choice Experiment Results: RPL Models

The robust analysis of the choice experiment revealed informative findings from both the RPL and LCM regressions conducted. To discuss these results, the sample size and model specification must be made clear. The final sample for the choice experiment consisted of 7455 observations, derived from 498 participants, each of whom evaluated all five choice sets, of which each presented two filet options and one opt-out option. To estimate the RPL models, the data was analyzed via Stata, while the LCM was executed using Nlogit. Modeling specifications include variables following a normal distribution, except price in the RPL, which was specified to be log-normally distributed. This specification ensures that the price, theoretically expected to be strictly negative, does not produce unrealistic positive value estimates that could arise from a normal distribution [34,47]. To accomplish this, the price variable must be multiplied by a negative one, enforcing the expected coefficient estimate to be strictly positive [37]. In alignment with economic theory, the price coefficient is negative and significant across all RPL models and LCMs conducted. Henceforth, as the price increases, the probability of the fish filet being selected decreases as the associated utility of the product decreases, all else being equal.

The baseline model, or Model 1 as specified in Equation (1), reveals that consumers experience a decline in utility when purchasing unfamiliar seafood species compared to the popular Atlantic Cod (

Table 4. Random parameters logit regression results: Model 1.

	Model 1 (Base)
Variables	Coefficient
ASC	−6.69 *** (0.7)
Fish species (base: Atlantic Cod)
Scup	−4.65 *** (0.52)
Croaker	−4.46 *** (0.5)
Fluke	−1.13 *** (0.22)
Trigger	−3.82 *** (0.37)
Local (base: not local)	1.2 *** (0.17)
Purchasing location (base: large grocery store)
Local grocery	0.39 ** (0.19)
Seafood shop	0.4 * (0.21)
Farmers’ market (FM)	0.58 *** (0.19)
Price	−1.58 *** (0.16)
Observations (N)	7455
Log likelihood	−1931.06

Note: All parameters are modeled as random parameters and normally distributed except for Price, which is log-normally distributed. Standard errors in parentheses, bold standard errors indicate insignificant standard deviation, implying homogeneous preferences. ***, **, * denote statistical significance at 1%, 5%, and 10% levels, respectively.

). This is evident by the negative and statistically significant coefficients for all unfamiliar fish species analyzed in the choice experiment. This implies that consumers need to be compensated, hence a willingness to accept (WTA), between a high of $20.16 for Scup and a low of $4.89 for Fluke to maintain a constant level of utility when purchasing these unfamiliar species compared to Atlantic Cod (

Table 5. WTP estimation for RPL Models.

	Model 1 (Base)	Model 2 (Local)	Model 2 (Farmers’ Market)	Model 3.1 Purchased Unfamiliar Species Before (Yes/No)	Model 3.2 Total Number of Species Bought Before	Model 3.3 Purchase Seafood at Alternative Networks	Model 3.4 Recreational Angler
Scup
Main effect	−20.16 ***	−24.67 ***	−24.67 ***	−20.97 ***	−24.12 ***	−24.38 ***	−23.8 ***
Combined effect		−24.27 ***	−21.95 ***	−7.4 *	−19.13 ***	−11.08 ***	−16.4 ***
Croaker
Main effect	−19.35 ***	−23.01 ***	−23.01 ***	−19.91 ***	−20.93 ***	−21.76 ***	−20.66 ***
Combined effect		−20.61 ***	−13.83 ***	−7.15 *	−18.84 ***	−12.69 ***	−16.62 ***
Fluke
Main effect	−4.89 ***	−9.4	−9.4	−5.12***	−6.41 ***	−6 ***	−5.67 ***
Combined effect		−0.96	5.04	−3.23 **	−4.57 ***	−1.77	−3.31 ***
Trigger
Main effect	−16.54 ***	−22.43 ***	−22.43 ***	−16.46 ***	−18.95 ***	−18.38 ***	−18.13 ***
Combined effect		−17.89 ***	−22.06 ***	−10.99 ***	−15.95 ***	−11.98 ***	−14.62 ***

Notes: All WTP values are calculated via the delta method. Formulas for calculating the WTP values are Model 1: ${WTP}_{\mathsf{\Omega }}={\displaystyle \frac{\mathsf{\Omega }}{\exp \left(\beta +{\sigma }^2/2\right)}} ,$ Model 2: $WT{P}_{k,Local=1}={\displaystyle \frac{({\gamma }_k+\lambda +{\mu }_k)}{\exp \left(\beta +{\sigma }^2/2\right).}}$, Model 3: $WT{P}_k={\displaystyle \frac{({\gamma }_k+{\omega }_k)}{\exp \left(\beta +{\sigma }^2/2\right)}}.$ ***, **, * denote statistical significance at 1%, 5%, and 10% levels, respectively.

). This finding aligns with other pricing studies utilizing choice experiments, which report that consumers’ WTP is significantly less for unfamiliar seafood species than for more familiar and popular ones [8,12].

One of the key advantages of the RPL model is its ability to account for preference heterogeneity [46]. Notably, the standard deviations for all unfamiliar fish species are statistically significant in the base model, suggesting the presence of preference heterogeneity. This implies that different consumers have varying preferences when it comes to purchasing unfamiliar fish filets. While the RPL model can detect unobserved heterogeneity, it cannot identify which individuals or groups have different preferences impacting their purchasing decisions. In light of this, Models 2 and 3 were analyzed, which included the addition of varying interaction terms to Model 1. As cautioned by [41], the interpretation of interaction terms in nonlinear models such as logits is not straightforward, and the significance of these interactions can be misleading. Researchers cannot simply evaluate results by looking at the sign, magnitude, or statistical significance of the coefficient on the interaction term when the model is nonlinear [41].

To address these concerns, the researchers calculated WTP point estimates, significance, and confidence intervals for the main effects and the combined effects captured by the interaction utilizing the delta method (Appendix A 

Table A4. WTP estimates model.

	WTP	95% CI
Fish Species
Scup	−20.16 ***	(−24.85, −15.48)
Croaker	−19.35 ***	(−23.55, −15.15)
Fluke	−4.89 ***	(−6.89, −2.90)
Trigger	−16.54 ***	(−20.09, −13.00)

Notes: All WTP values are calculated via the delta method. *** denotes statistical significance at the 1% level.

,

Table A5. WTP estimates Model 2.

	WTP	95% CI
Fish Species
Scup	−24.67 ***	(−32.69, −16.67)
Scup + Local + (Local × Scup)	−24.27 ***	(−33.69, −14.84)
Scup + FM + (Scup × FM)	−21.95 ***	(−30.47, −13.43)
Croaker	−23.01 ***	(−31.64, −14.39)
Croakers + Local + (Local × Croaker)	−20.61 ***	(−27.75, −13.46)
Croaker + FM + (Croakers × FM)	−13.83 ***	(−20.59, −7.06)
Fluke	−9.40 ***	(−15.23, −3.56)
Fluke + Local + (Local × Fluke)	−0.96	(−5.41, 3.50)
Fluke + FM + (Fluke × FM)	5.04	(−6.72, 16.8)
Tigger	−22.43 ***	(−30.27, −14.60)
Tigger + Local + (Local × Tigger)	−17.89 ***	(−25.10, −10.68)
Trigger + FM + (Tigger × FM)	−22.06 ***	(−30.38, −13.75)

Notes: All WTP values are calculated via the delta method. *** denotes statistical significance at the 1% level.

,

Table A6. WTP estimates Model 3.1.

	WTP	95% CI
Fish Species
Scup	−20.97 ***	(−25.68, −16.26)
Scup + (Pervious × Scup)	−7.40 *	(−14.24, −0.58)
Croaker	−19.91 ***	(−24.07, −15.75)
Croaker + (Pervious × Croaker)	−7.15 *	(−13.85, −0.45)
Fluke	−5.12 ***	(−7.21, −3.03)
Fluke + (Pervious × Fluke)	−3.23 **	(−6.53, 0.05)
Trigger	−16.46 ***	(−19.92, −12.30)
Trigger + (Pervious × Fluke)	−10.99 ***	(−19.10, −2.88)

Notes: All WTP values are calculated via the delta method. ***, **, * denote statistical significance at 1%, 5%, and 10% levels, respectively.

,

Table A7. WTP estimates Model 3.2.

	WTP	95% CI
Fish Species
Scup	−24.12 ***	(−29.92, −18.33)
Scup + (Unfamiliar Species × Scup)	−19.13 ***	(−23.64, −14.61)
Croaker	−20.93 ***	(−25.83, −16.04)
Croaker + (Unfamiliar Species × Croaker)	−18.84 ***	(−23.02, −14.66)
Fluke	−6.41 ***	(−8.94, −3.89)
Fluke + (Unfamiliar Species × Fluke)	−4.57 ***	(−6.66, −2.48)
Trigger	−18.95 ***	(−23.24, −14.66)
Trigger + (Unfamiliar Species × Trigger)	−15.95 ***	(−19.40, −12.5)

Notes: All WTP values are calculated via the delta method. *** denotes statistical significance at the 1% level.

,

Table A8. WTP estimates Model 3.3.

	WTP	95% CI
Fish Species
Scup	−24.38 ***	(−30.03, −18.73)
Scup + (Alternative Location × Scup)	−11.08 ***	(−15.61, −6.55)
Croaker	−21.76 ***	(−26.53, −16.99)
Croaker + (Alternative Location × Croaker)	−12.69 ***	(−17.07, −8.31)
Fluke	−6.00 ***	(−8.35, −3.65)
Fluke + (Alternative Location × Fluke)	−1.77	(−5.07, 1.53)
Trigger	−18.38 ***	(−22.37, −14.40)
Trigger + (Alternative Location × Trigger)	−11.98 ***	(−15.70, −8.27)

Notes: All WTP values are calculated via the delta method. *** denotes statistical significance at the 1% level.

and

Table A9. WTP estimates Model 3.4.

	WTP	95% CI
Fish Species
Scup	−23.80 ***	(−29.68, −17.92)
Scup + (Angler × Scup)	−16.40 ***	(−21.48, −11.30)
Croaker	−20.66 ***	(−25.32, −16.01)
Croaker + (Angler × Croaker)	−16.62 ***	(−21.83, −11.42)
Fluke	−5.67 ***	(−8.02, −3.32)
Fluke + (Angler × Fluke)	−3.31 *	(−6.56, −0.05)
Trigger	−18.13 ***	(−22.22, −14.05)
Trigger + (Angler × Trigger)	−14.62 ***	(−18.85, −10.39)

Notes: All WTP values are calculated via the delta method. *** and * denote statistical significance at 1% and 10% levels, respectively.

). Therefore, the results discussed for the model leveraging interaction effects to identify preference heterogeneity will focus on the stimulated WTP estimates (Table 5) that account for the nonlinear combinations of estimators; all regression output is located in Appendix A (

Table A1. Random parameter logit regression results: Models 1 and 2.

	Model 1 (Base)	Model 2 (Local Interaction)
Variables	Coefficient	Coefficient
ASC	−6.69 *** (0.7)	−10.03*** (1.82)
Fish species (base: Atlantic Cod)
Scup	−4.65 *** (0.52)	−7.19 *** (1.41)
Croaker	−4.46 *** (0.5)	−6.7 *** (1.24)
Fluke	−1.13 *** (0.22)	−2.74 *** (0.79)
Trigger	−3.82 *** (0.37)	−6.53 *** (1.24)
Local (base: not local)	1.2 *** (0.17)	1.81 *** (0.58)
Purchasing location (base: large grocery store)
Local grocery	0.39 ** (0.19)	0.66 ** (0.34)
Seafood shop	0.4 * (0.21)	0.84 ** (0.37)
Farmers’ market (FM)	0.58 *** (0.19)	0.02 (0.55)
Interaction terms
Local × Scup		−1.69 * (0.92)
Local × Croaker		−1.1 (0.82)
Local × Fluke		0.65 (0.78)
Local × Trigger		−0.48 (0.74)
FM × Scup		0.774 (0.97)
FM × Croaker		2.66 ** (1.08)
FM × Fluke		4.19 ** (2.03)
FM × Trigger		0.089 (0.9)
Price	−1.58 *** (0.16)	−1.31 *** (0.236)
Observations (N)	7455	7455
Log likelihood	−1931.06	−1912.11

Note: All parameters are modeled as random parameters and normally distributed except for Price, which is log-normally distributed. Standard errors in parentheses, bold standard errors indicate an insignificant standard deviation, implying homogeneous preferences. ***, **, * denote statistical significance at 1%, 5%, and 10% levels, respectively.

,

Table A2. Random parameter regression results: Models 3.1 and 3.2.

	Model 3.1 Purchased Unfamiliar Species Before (Yes/No)	Model 3.2 Total Number of Species Bought Before
Variable	Coefficient	Coefficient
ASC	−7.02 *** (0.79)	−6.48 *** (0.71)
Fish species (base: Atlantic Cod)
Scup	−5.16 *** (0.6)	−5.53 *** (0.64)
Croaker	−4.9 *** (0.58)	−4.8 *** (0.57)
Fluke	−1.26 *** (0.26)	−1.47 *** (0.28)
Trigger	−4.05 *** (0.44)	−4.34 *** (0.48)
Local (base: not local)	1.26 *** (0.18)	1.18 *** (0.17)
Purchase venue (base: chain grocery store)
Local grocery store	0.43 ** (0.2)	0.45 ** (0.2)
Seafood shop	0.42 * (0.22)	0.47 ** (0.21)
Farmers’ market	0.65 *** (0.21)	0.64 *** (0.2)
Interaction Controls (Previously Purchased)
Previously purchased Scup × ASC	−0.85 (0.98)
Previously purchased Croaker × ASC	0.31 (1.03)
Previously purchased Fluke × ASC	0.14 (0.45)
Previously purchased Trigger × ASC	−3.93 *** (1.77)
Previously purchased Scup × Scup	3.34 *** (0.93)
Previously purchased Croaker × Croaker	3.14 *** (0.9)
Previously purchased Fluke × Fluke	0.46 (0.44)
Previously purchased Trigger × Trigger	1.35 (1.01)
Total species purchased control × ASC		−0.33 (0.29)
Total species × Scup		1.14 *** (0.35)
Total species × Croaker		0.48 (0.32)
Total species × Fluke		0.42 ** (0.25)
Total species × Trigger		0.69 *** (0.26)
Price	−1.5 *** (0.16)	−1.58 *** (0.16)
Observations (N)	7455	7455
Log likelihood	−1903.32	−1913.78

Notes: All parameters are modeled as random parameters and normally distributed except for Price, which is log-normally distributed. Standard errors in parentheses, bold standard errors indicate an insignificant standard deviation, implying homogeneous preferences. ***, **, * denote statistical significance at 1%, 5%, and 10% levels, respectively.

and

Table A3. Random parameter regression results: Models 3.3 and 3.4.

	Model 3.3 Purchase Seafood at Alternative Networks	Model 3.4 Recreational Angler
Variable	Coefficient	Coefficient
ASC	−6.94 *** (0.83)	−6.58 *** (0.75)
Fish species (base: Atlantic Cod)
Scup	−5.86 *** (0.71)	−5.67 *** (0.7)
Croaker	−5.23 *** (0.64)	−4.93 *** (0.57)
Fluke	−1.44 *** (0.27)	−1.35 *** (0.27)
Trigger	−4.42 *** (0.49)	−4.32 *** (0.47)
Local (base: not local)	1.26 *** (0.2)	1.28 *** (0.19)
Purchase venue (base: chain grocery store)
Local grocery store	0.4 ** (0.2)	0.43 ** (0.2)
Seafood shop	0.43 ** (0.22)	0.48 ** (0.22)
Farmers’ market	0.61 *** (0.21)	0.68 *** (0.22)
Interaction terms
Alternative networks × ASC	0.19 (0.47)
Alternative networks × Scup	3.2 *** (0.68)
Alternative networks × Croaker	2.18 *** (0.58)
Alternative networks × Fluke	1.02 ** (0.47)
Alternative networks × Trigger	1.54 *** (0.46)
Recreational angler × ASC		−1.16 ** (0.52)
Recreational angler × Scup		1.77 *** (0.6)
Recreational angler × Croaker		0.96 (0.61)
Recreational angler × Fluke		0.56 (0.45)
Recreational angler × Trigger		0.84 (0.49)
Price	−1.58 *** (0.17)	−1.53 (0.16)
Observations (N)	7455	7425
Log likelihood	−1903.83	−1913.78

Notes: All parameters are modeled as random parameters and normally distributed except for Price, which is log-normally distributed. Standard errors in parentheses, bold standard errors indicate an insignificant standard deviation, implying homogeneous preferences. *** and ** denote statistical significance at 1%, and 5% levels, respectively. Alternative networks include purchasing seafood at any of the following locations: farmers’ markets, community-supported fisheries, or direct from docks.

).

Model 2 WTA results revealed that the combinations of local-sourced and farmers’ market (FM) contribute to the reduction in WTA towards the unfamiliar species. It was found that WTA for all three species, except Fluke, varied based on whether they were marked as local or not and the venue from which these species were purchased (Table 5). The overall combined marginal WTA for local Scup, Croaker, and Trigger reduced by $0.40, $2.40, and $4.54, respectively, compared to the non-local options of these species. Both the main effect WTA for non-local Fluke and the interaction effect for local Fluke were statistically insignificant, suggesting consumers’ WTP for both local and non-local Fluke are similar to that of Atlantic Cod.

When Scup, Croaker, and Trigger were purchased at the farmers’ market, the combined marginal WTA declined by $2.72, $9.18, and $0.37, respectively. This finding suggests a strategy to expand the market for unfamiliar species by offering them at farmers’ markets, as consumers show a higher acceptance of unfamiliar seafood species in such alternative networks. Similar trends have been documented by other studies [8,9,10,11].

Next, Model 3 was analyzed to investigate other sources of consumer preference heterogeneity. First, we analyzed for consumers who had reported purchasing a particular unfamiliar fish species (C1, Model 3.1) (Appendix A Table A2). By including a dummy variable indicating if a consumer had reported previously purchasing a particular unfamiliar fish species, and interacting the dummy variable with the respective fish species, the WTA calculations (Table 5) revealed that previous experience purchasing the species significantly reduced unwillingness to pay across all four fish. WTA decreased by $13.57 for Scup, $12.76 for Croaker, $1.89 for Fluke, and $5.50 for Trigger, indicating that familiarity may play a role in consumer acceptance of unfamiliar species. This is evidenced by the significantly lower WTA estimates from the sum of the main effect, Species = {scup, croaker, fluke, trigger}, and the interaction effect C1 (previously purchased) × Species. This result provides insights into another potential avenue for expanding the market for unfamiliar fish species. The disutility of purchasing these now familiar but less popular fish species declines drastically once consumers have purchased them previously.

We ran a separate regression where Species variables were interacted with C2 (number of unfamiliar species purchased previously, Model 3.2, Appendix A Table A2). The combined effect of the main and interaction terms was significant for all four species. Compared to consumers who had not purchased unfamiliar fish before, those with more experience showed a reduced WTA by $4.99 for Scup, $2.09 for Croaker, $1.84 for Fluke, and $3.00 for triggerfish per pound. This result suggests purchasing a broader range of unfamiliar fish species may reduce consumer dissatisfaction when purchasing these species compared to more known and popular seafood species, with similar findings being reported by [8].

Next, we analyzed whether the preference heterogeneity among seafood consumers might stem from whether consumers reported purchasing seafood from alternative networks (C3, Model 3.3) or anglers (C4, Model 3.4) (Appendix A Table A3). Recall that our previous ordered logit regression analysis also suggested that these could be the source of heterogeneity (Table 3). Alternative networks capture seafood consumers who purchase seafood from any of the following locations: farmers’ markets, direct from docks, or community-supported fisheries. For three of the four unfamiliar fish species, the disutility associated with purchasing these species, WTA, declined when bought by consumers who source seafood from alternative networks (Table 5). Scup had the largest decline by $13.30 per pound, while Croaker decreased by $9.07, triggerfish by $6.40, and Fluke was insignificant. These results suggest that consumers who purchase seafood from alternative networks may have different preferences towards purchasing unfamiliar species.

As for incorporating anglers (C4, Model 3.4 Appendix A Table A3), our findings revealed that the disutility and WTA from purchasing all four species were significantly reduced when purchased by an angler (Table 5). Scup had the largest decline by $7.40 per pound, while Croaker decreased by $4.04, Fluke by $2.36, and triggerfish by $3.69. These findings highlight that recreational anglers may have different preferences compared to non-anglers regarding purchasing unfamiliar seafood species.

The data analysis to uncover the source of preference heterogeneity by including interaction terms indicated that there might be multiple other unknown factors influencing consumers’ decision-making processes regarding purchasing unfamiliar fish filets, as many of the coefficients’ standard deviations remain significant (Appendix A Table A1, Table A2 and Table A3). Given this inclination, conducting an LCM is an appropriate option.

4.1.2. Choice Experiment Results: LCM

LCM assumes that individuals can be sorted into different clusters or classes based on observed and latent attributes, capturing subgroup heterogeneity. The number of classes is endogenous; in practice, a researcher sets the number of classes and runs the LCM. The optimal number of classes is determined based on the overall fit of the model, such as information criteria (e.g., C-AIC and McFadden R²). Following Refs. [48,49], who showed that solely relying on information criteria can lead to intractable parameter estimates, we also considered theoretical insights (negative price coefficient), the smallest class size, and model parsimony (four classes or less). We ran an LCM for class sizes varying from two to eight and determined that the three-class model was the best-performing model overall (

Table 6. Latent class selection criteria.

Number of Classes	C-AIC	Log Likelihood	McFadden R2	Smallest Class Size
2	3929.6	−1943.79	0.288	0.36
3	3852.7	−1894.36	0.31	0.11
4	3829.3	−1871.62	0.31	0.046
5	3802.8	−1847.38	0.32	0.028
6	--	--	--	--
7	3801.5	−1824.74	0.33	0.025
8	--	--	--	--

).

The LCM results did not particularly divulge any new findings but solidified and supported already documented findings from the previous RPL models (

Table 7. Latent class model regression results.

	LCM Coeff.			WTP ($)
	Class 1	Class 2	Class 3	Class 1	Class 2	Class 3
ASC	−3.94 ***	−5.31 ***	−1.45	−43.78	−44.25	−9.67
Fish Species (Atlantic Cod)
Croaker	−0.76 ***	−4.38 ***	−2.48 ***	−8.44	−36.50	−16.53
Scup	−0.49 **	−4.99 ***	−2.97 ***	−5.44	−41.58	−19.80
Fluke	−0.05	−1.85 ***	−0.97 ***	0.00	−15.42	−6.47
Trigger	−0.70 ***	−3.88 ***	−3.53 ***	−7.78	−32.33	−23.53
Purchasing location (base: large grocery store)
Farmers’ market	0.28 **	0.26	0.93 **	3.11	0.00	6.20
Seafood shop	0.42 ***	−0.02	0.39	4.67	0.00	0.00
Small grocery	0.18	0.22	0.65	0.00	0.00	0.00
Local (base: not local)	0.65 ***	0.38 **	0.77 **	7.22	3.17	5.13
Class Membership
Constant	−0.02	0.12		0.00	1.00
Seafood employment	0.06	−1.52		0.00	0.00
Angler	1.32 **	0.84		14.67	0.00
Seafood consumption	0.18	0.47 ***		0.00	0.00
Whole fish consumption	0.26	−0.01		0.00	0.00
Unfamiliar fish species	0.24	0.20		0.00	0.00
Alternative networks	0.35	−1.11 **		0.00	−9.25
Price	−0.09 ***	−0.12 ***	−0.15 **
Log—likelihood		−1873.80
McFadden-R2		0.31
CAIC		3835.6
Latent class probability	0.456	0.408	0.136
Observations (N)		7455

Note: All parameters are modeled as random parameters. All parameters are normally distributed, including Price. Hence, WTP with Price normally distributed is the negative ratio of the coefficient of interest divided by the coefficient on Price. Standard errors in parentheses. All variables are dichotomous except for the Price variable. *** and ** denote statistical significance at 1% and 5% levels, respectively.

). The majority of the sample (45.6%) had the highest probability of belonging to Class 1, deemed as the “species-adaptive anglers”; followed by Class 2 (40.8%), the “conservative consumers”; and finally Class 3, “local-preference buyers”, had the smallest portion of the sample (13.6%). The species-adaptive anglers of Class 1 were the least deterred from purchasing unfamiliar seafood species, as indicated by the smaller negative coefficients on the fish species level and further by the lower WTA values for these species. Class 1 did not have a utility difference between purchasing Fluke and Atlantic Cod, as seen by the insignificant coefficient on the Fluke variable. Members of Class 1 showed a preference towards the venue where they purchased their seafood, as seen by the only class to have positive significant coefficients on both venue locations, seafood shop, and farmers’ market. As suggested by the class name, species adaptive anglers, members of Class 1 were more likely to be recreational anglers. This is inferred from the significant and positive coefficient on recreational anglers, one of the various controls added to the utility function to capture individual characteristics influencing class membership. For comparison, Class 3 parameters were normalized to zero. This provides more empirical evidence that recreational anglers may be an effective target market for growing the demand for unfamiliar seafood species.

The LCM provides further evidence that consumers who purchase seafood from alternative networks are less deterred by unfamiliar seafood and may be a good entry point for these species. This is empirically seen by Class 2, the conservative consumers being the most deterred from purchasing unfamiliar fish species, requiring significantly higher compensation than Class 3 to purchase unfamiliar fish filets compared to Atlantic Cod. Class 2, compared to Class 3, is significantly less likely to purchase seafood from alternative networks, such as farmers’ markets, farm stands, and direct from the dock, as seen by the large and significant negative coefficient on the alternative networks variable for Class 2. Class 3, local buyer preference, is not only more likely to purchase seafood from alternative networks but also has the highest preference for buying the fish filet options marked as local or from the farmers’ market, as seen by the largest WTP and a positive and significant coefficient on both the local and farmers’ market variables. The LCM provides compelling evidence that helps corroborate the findings revealed in early RPL regressions that most consumers have a disutility of purchasing unfamiliar seafood species compared to the popular Atlantic Cod. However, these preferences are not constant among consumers, as seen by the existence of preference heterogeneity and variations based on seafood consumers’ characteristics and behaviors. Identifying different consumer segments aids in sound empirical recommendations on potential key points of entry to expand the markets for unfamiliar seafood predicted to become more abundantly available in the region.

5. Policy Implications

To support commercial fisheries faced with the changing fish species landed due to climate change, targeted policies can play a role in facilitating consumer adoption of newly abundant yet unfamiliar seafood species. As ocean temperatures rise and species distributions change, traditional fisheries reliant on a narrow set of high-demand species, such as Atlantic Cod, may face increasing instability. One possible key policy approach to addressing this shift is promoting local consumption of the underutilized species, helping to align market demand with the changing availability of seafood resources. Fisheries diversification has been recognized as a pillar of sustainable fisheries management, reducing the reliance on a narrow set of commercially harvested fish, improving ecological balance, and enhancing the long-term economic stability of fishing communities. By expanding consumer preferences to include a broader range of seafood options, diversification could not only provide fishers with new revenue streams but also potentially mitigate risks in fish stock composition shifts due to climate change.

This study has shown that there may be some major challenges, where consumers generally exhibit a disutility when purchasing an unfamiliar species compared to purchasing a mainstream species such as Atlantic Cod. This insight could be taken into consideration when determining if and what policies or marketing campaigns could be enacted to help influence consumer demand for these species. Ultimately, in efforts to maintain a sustainable commercial fishing industry, action should be required to help the market embrace these newly arriving, unfamiliar species.

The study identified several potential marketing entry points to help design effective campaigns and policies. A significant proportion of the unfamiliar seafood purchased was driven by recommendations from seafood counter employees. This suggests a possible, straightforward, and cost-effective method to alter consumer behavior at the point of sale. Related research indicates that consumers are often hesitant to purchase unfamiliar seafood due to barriers such as price, unknown flavor, and lack of knowledge on preparation [6,7,8,9]. It can be hypothesized that employee recommendations may include information on taste and preparation, thereby helping mitigate the hesitation instincts. The likely fear of unfamiliarity, potentially driven by unknown flavors or a lack of knowledge on preparation, was further identified by consumers who have previously purchased an unfamiliar fish species as being more willing to pay for all species. This likely implies that initial exposure to unfamiliar fish reduces the loss of utility associated with their purchase, highlighting the fact that these newly emerging fish could find a potential market if consumers are initially exposed to them. Marketing campaigns and policies could leverage this by offering free or discounted samples, cooking instructions, or flavor notes in efforts to help consumers overcome initial hesitations and aid in the reduction of discernment around unknown taste, flavor, and how to prepare the fish.

The study also highlights other key market entry points with specific consumer segments. Preferences for unfamiliar fish species varied by point of purchase and among consumers who typically source their seafood from alternative venues. This aligns with existing studies suggesting that expanding markets for unfamiliar fish species could target consumers at farmers’ markets and community-supported fisheries (CSFs) [9,10,50]. Consumers who shop at local markets and CSFs tend to be more environmentally motivated and open to underutilized seafood species, partly because these venues provide greater exposure to local and diverse fish options than mainstream retail channels [50]. This finding could be considered when crafting marketing campaigns or policies that may provide support in expanding markets for these unfamiliar species. Efforts could be put forth to focus on breaking the barriers of entry by focusing on selling at alternative points of purchase, such as the farmers’ market, farm stands, direct from the dock, or community-supported fisheries, and targeting consumers who normally purchase from these venues.

Another potential marketing entry point is the recreational angler community, as anglers were identified as having a higher willingness to accept unfamiliar fish, suggesting they are more willing to try and purchase them, compared to non-anglers. This information could be leveraged by policymakers to help introduce these species to the public by strategically targeting recreational anglers. Given the significant number of recreational anglers in New England, this has the potential to be an impactful strategy. State-level policymakers could disseminate information to anglers through online posts on licensing pages, emails, or notices at popular fishing locations. This possible low-cost approach could effectively inform and encourage a community with distinct preferences to purchase unfamiliar fish species.

To address climate-driven shifts in species abundance and promote fisheries resilience through species diversification, policy interventions, and targeted marketing strategies appear to be essential. Effective programs may rely on establishing market entry points for underutilized and unfamiliar fish species predicted to become more abundant to promote consumer adoption. Incorporating consumer attitudes, behaviors, and perceptions is an important step when trying to introduce these species to the right markets. This study provides valuable guidance for policymakers and fishery stakeholders seeking to enhance acceptance and market integration of these emerging seafood options by identifying key consumer segments and motivations behind purchasing unfamiliar seafood species under investigation.

These recommendations could be explored in future research to address limitations and expand on the present study. Future research should address the limitation of hypothetical bias inherent in stated preference methods by incorporating actual purchase experiments or in-person taste trials. Research could also explore the effectiveness of specific marketing interventions—such as in-store tastings or targeting specific consumer segments—in real-world retail environments to better understand how to shift consumer behavior and reduce the disutility associated with unfamiliar species. A limitation of this study is its reliance on a regional sample collected via MTurk, which may restrict the generalizability of findings. Future research could expand the sample size and include diverse sampling platforms to enable state-by-state analysis across the New England region. This would not only enhance representativeness but also allow for a deeper exploration of geographic and demographic-based preference heterogeneity, offering more targeted insights for policy and marketing interventions at the state level.