Joint Choice of Fresh Food Purchase Channels and Terminal Delivery Service: A Background on Major Public Health Events

Abstract

The paper aims to analyze the consumer joint choice behavior on fresh food purchase channels and terminal delivery services during major public health events, with the purpose of revealing the underlying influencing factors and behavioral characteristics. First, based on random utility maximization theory, the cross-nested logit model is formulated, which takes into account the influence of socioeconomic attribute factors, service attribute factors, risk perception attribute factors and trust perception attribute factors. Second, a questionnaire survey is conducted, and the obtained data are used to estimate the model parameters and perform an elasticity analysis of the utility variables. The parameter estimation results demonstrate that in the context of major public health events, consumers consider adjusting their attitudes toward e-commerce platforms first when the utility variables are altered, and fresh food purchase channels are easily replaced for consumers who choose unmanned equipment home delivery. The elasticity analysis results suggest that consumers are more willing to buy fresh food through community group-buying channels, are more sensitive to the convenience of the purchase process and are less concerned with delivery time. Although person-to-person contact increases the risk of infection, consumers still prefer attended terminal delivery services. Furthermore, consumers least agree with the effectiveness of body temperature detection methods in public places but feel that an effective way to increase consumer trust in enterprises is to strengthen personnel protection measures.

1. Introduction

In recent years, the frequency of public health events has exhibited an upward trend [1], with several notable outbreaks occurring in just the past 25 years, including the 2002 SARS outbreak in China, the 2014 Ebola epidemic in West Africa, and the 2019 COVID-19 pandemic [2]. These events not only pose a threat to physical health but also have a profound impact on people’s consumption activities. Jung et al. researched how MERS influenced consumer expenditures in Korea, and found that customers altered their purchasing behaviors to reduce the risk of infection [3]. During a pandemic influenza outbreak, people will also tend to limit their shopping activities to essentials [4]. In particular, COVID-19 drastically altered consumer purchasing behavior, with more individuals choosing online channels over physical stores to buy fresh food due to health concerns [5,6]. While online shopping ensures that consumers can avoid crowded places during the fresh food purchasing process, they still need to obtain goods via terminal delivery services that have multiple contact points with consumers [7]. This leads consumers to not only consider fresh food purchase channels but also to evaluate various terminal delivery services to reduce their exposure to risk [8,9]. Obviously, major public health events have caused changes in the retail environment, reshaped consumers’ shopping habits [10,11], and brought new challenges to the retail industry. To respond effectively to changes in consumer demand and atypical consumption situations, retail enterprises must thoroughly comprehend consumer preferences regarding fresh food purchase channels and terminal delivery services. Especially in the special context of major public health events, consumers’ fresh food shopping behavior has gone beyond just buying and evolved into a behavior closely related to health [12]. Different fresh food purchase channels and terminal delivery service choices lead to different senses of security and satisfaction for consumers. Therefore, a deep understanding of the preferences of consumers in fresh food purchasing channels and terminal delivery services has emerged as a critical focal point for retail enterprises.

Scholars have focused on changes in consumer fresh food purchasing habits during major public health events [13,14]. Using survey data from countries around the world, they analyzed consumer purchasing behavior and its influencing factors during the COVID-19 pandemic [7,14,15]. In addition to demographic variables, the impacts of perceived value [16], online shopping satisfaction [17], public fear [14], consumer risk attitudes [18], consumer confidence [11] and personal financial status [19] are also significant. This part of the research focused on data from the early stage of the COVID-19 outbreak. Asgari et al. conducted a study using post-pandemic data, and the results revealed that some people maintained the shopping habits formed during the pandemic [20]. Verhoef et al. also pointed out that the impact of COVID-19 on consumer behavior will be permanent [21]. In addition to focusing on where to buy, scholars have focused on how to obtain fresh food. Truong et al. used survey data from 2021 to analyze consumer preferences and influencing factors for three modes, including online shopping, roadside pickup and in-store shopping [19], whereas Wang et al. studied contactless shopping and delivery in response to the COVID-19 pandemic [12]. Just like purchasing behaviors, the delivery habits consumers form will be routinely kept up after the epidemic [22]. The shifts in consumer preferences towards purchase channels and terminal delivery services during the COVID-19 pandemic signify the evolution of the retail industry, providing crucial guidance for retail enterprises to refine their omni-channel strategies and enhance their logistics services. More importantly, Wang et al. argue that there is a close interrelation between the terminal delivery service and the purchase channel, highlighting the necessity of integrating the two for a thorough and comprehensive analysis [22]. Betancourt et al. also noted that channel choices are influenced by delivery services [23], but their research lacks the subdivision of delivery service mode, so the analysis of the interaction between the two is insufficient. Xiao et al., Xi et al. and Shen et al. have focused on the relationship between terminal delivery services and online channels, yet unfortunately, they have overlooked offline channels [24,25,26]. This existing research gap will be meticulously addressed and bridged in the present study.

Fresh food is consumed every day, and its purchase frequency is much higher than other commodities. Coupled with its shorter shelf life compared to other foods and its vulnerability to spoiling easily in transportation, this makes consumers pay even closer attention to accessing fresh food during major public health events. Impacted by the COVID-19 pandemic, the purchasing channels for fresh food among consumers have undergone changes [27,28], with more people preferring to buy fresh food online. Lu et al. conducted a comparative analysis of online fresh food shopping behavior during normal and COVID-19 crisis periods [28], and Wang and Li focused on the influencing factors of online and offline fresh food purchasing [29]. These studies place more emphasis on online channels while neglecting the role of offline channels. Zhao et al. also recognized the high correlation between purchasing channels and terminal delivery services [30]. However, this study only addressed the link between online fresh food purchasing channels and self-pickup services at physical stores, without exploring the relationship between other options of terminal delivery services (such as home delivery) and fresh food purchasing channels. In particular, given the influence of major public health events, consumers’ choices regarding fresh food purchase channels and terminal delivery services display different preferences. To bridge this gap in the literature, this study examines consumer joint choice behavior regarding the fresh food acquisition process and focuses on the reality that fresh food purchase channels and terminal delivery services are diverse. The aim is to describe the behavioral characteristics exhibited by consumers in acquiring fresh food during such events.

The discrete choice model based on random utility maximization is the predominant analytical method utilized in the field of choice behavior research. This method can be used to judge which factors really affect the behavior, and how important they are. The multinomial logit model (MNL) is the simplest and most widely applied method [31], while it has the Independence of Irrelevant Alternatives (IIA) property which is inconsistent with reality. To reasonably explain that consumers’ choice of fresh food purchase channels and terminal delivery services is closely related, the most important thing is to properly deal with the correlation between alternatives. Nested logit (NL) models accept the correlation of alternatives, allowing the alternatives within each nest to be correlated, while the alternatives belonging to different nests are independent of each other. To a certain extent, an NL model relaxes the IIA assumptions, but it requires that the hierarchical structure of the selection problem be set accurately in advance. The cross-nested logit (CNL) model is different from the NL model in that it allows one alternative to belong to multiple subsets, and it can flexibly describe the correlation between different dimensional choices in multidimensional choice behavior [32]. Therefore, this study uses the CNL model to analyze consumers’ joint choice of fresh food purchase channels and terminal delivery services. On the basis of the sample data obtained from the questionnaire survey, parameter estimation and elasticity analysis of the model are conducted, and finally, management suggestions are proposed to support retail enterprises in optimizing fresh food sales service schemes. Compared with other relevant research, the differences in the contributions of this study are as follows: ① With a focus on the diversification of fresh food purchase channels and terminal delivery services, consumer joint choice behavior towards these two aspects is analyzed rather than single-choice behavior. ② A CNL model based on generalized extreme value (GEV) theory is constructed to consider the effects of socioeconomic factors, service factors, risk perception factors and trust perception factors. ③ An elasticity analysis is performed to explain the change in probability of each alternative, given one unit of change in a given utility variable.

To achieve the research goals, the remainder of this paper is organized as follows: In Section 2, fresh food purchase channels and terminal delivery services are explained. In Section 3, the methodology adopted is described, and the detailed formulations of the CNL model are proposed. In Section 4, the data used for this study are interpreted, and Section 5 presents the estimation results and discusses direct elasticities and cross-elasticities. Section 6 presents the managerial implications, and Section 7 outlines the conclusions.

2. Fresh Food Purchase Channels and Terminal Delivery Service

In order to minimize personnel gatherings, retail enterprises and logistics enterprises have put forward many innovative service schemes for obtaining fresh food driven by internet thinking and artificial intelligence. These service schemes meet the consumers’ needs for flexible, convenient, and safe access to fresh food during major public health events. The process of obtaining fresh food mainly includes two decision-making steps: One step is where to buy, which is the choice of purchase channels; the other step is how to deliver it home, which is the choice of terminal delivery services. Without any of the above steps, consumers cannot achieve their goal of obtaining fresh food. Different combinations of purchase channels and terminal delivery services form differentiated ways of obtaining fresh food.

Purchase channels are generally divided into online and offline channels, and many scholars have focused on the role played by online purchase channels during major public health events, such as how they facilitate the flow of materials during social isolation and ensure the supply of daily necessities for residents [33,34]. However, we cannot ignore the importance of offline channels, which still occupy an irreplaceable position in the minds of consumers due to their advantages such as intuitive experience and immediate service [1,34]. In order to comprehensively demonstrate consumer preferences in fresh food purchase channels, online and offline channels will be segmented. According to the consumer’s final fresh food shopping place, the diversified purchasing channels that integrate online and offline behaviors are summarized in four forms: traditional market channels (farmers’ markets and supermarkets), fresh food store channels, community group-buying channels and e-commerce platform channels [30,35,36]. For example, if consumers make purchases through the online platforms of retail supermarkets, the purchase channel they choose is defined as an e-commerce platform.

Terminal delivery service refers to a certain combination of delivery activities through which the physical transfer of goods from the last transit point to the final drop point can be accomplished. The delivery activities may be performed by logistics operators or consumers [37]. Terminal delivery services fall into two main categories: self-pickup and home delivery [38]. According to whether the services are provided by personnel, they are subdivided into four types of services: manned pick-up points, unattended self-pickup cabinets, home delivery by couriers and home delivery by unmanned equipment [39,40,41,42]. An attended terminal delivery service refers to the form of terminal delivery service provided to consumers by professionals specializing in delivery services or non-professionals who provide services on a part-time basis. An unattended terminal delivery service refers to the form of terminal delivery service provided to consumers by unmanned devices. Here, the situation where consumers purchase fresh food from farmers’ markets, supermarkets, or retail stores and take it home by themselves is categorized as a service type where consumers opt for a manned pick-up point.

The purchasing decision-making process of consumers varies due to individual preferences, needs, and external factors. Some consumers first decide on the purchase channel and then select the appropriate terminal delivery service offered by that channel. Others, in order to maintain social distance or for convenience and flexibility in pick-up, decide on the terminal delivery service first and then consider which purchase channel can provide that service. There is a clear correlation between fresh purchase channels and terminal delivery services. Although there are abundant research results on purchase channels and terminal delivery services, few studies put the two into a framework to analyze consumer behavior in the fresh food acquisition process. This research gap highlights the importance of gaining a deeper understanding of consumer joint choice behavior toward purchase channels and terminal delivery services, as well as how these joint choices influence consumer decisions in obtaining fresh food and the future development trends of the fresh food market.

3. Model Specification

3.1. Model Structure Description

Considering the correlation between fresh food purchase channels and terminal delivery services, a CNL model of joint choice behavior is constructed (as shown in Figure 1). The model choice set is comprised of two subsets, which are the fresh food purchase channel subset s and the terminal delivery service subset d. The selective limbs, which are the traditional market channel (s₁), the community group-buying channel (s₂), the fresh food store channel (s₃) and the e-commerce platform channel (s₄), contained in subset s and the home delivery by couriers (d₁), the home delivery by unmanned equipment (d₂), the manned pick-up point (d₃) and the unattended self-pickup cabinet (d₄) contained in subset d constitute the eight nests of the CNL model. The model’s final choice set C = c₁, c₂, …, c_l is the joint choice set of subsets d and s, which contains 16 alternatives. Details of the alternatives are presented in

Table 1. Details of alternatives.

Fresh Food Purchase Channel	Terminal Delivery Service	Alternative Name
Traditional market channel	home delivery by couriers	c1
	home delivery by unmanned equipment	c2
	manned pick-up point	c3
	unattended self-pickup cabinet	c4
Community group-buying channel	home delivery by couriers	c5
	home delivery by unmanned equipment	c6
	manned pick-up point	c7
	unattended self-pickup cabinet	c8
Fresh food store channel	home delivery by couriers	c9
	home delivery by unmanned equipment	c10
	manned pick-up point	c11
	unattended self-pickup cabinet	c12
E-commerce platform channel	home delivery by couriers	c13
	home delivery by unmanned equipment	c14
	manned pick-up point	c15
	unattended self-pickup cabinet	c16

. In addition to the solutions that have been widely used in reality, the 16 alternatives also include service methods that may appear in the future or methods that enterprises are trying, such as the service plan formed by the combination of different fresh food purchase channels and home delivery by unmanned equipment.

δ_i and γ_j refer to the dissimilarity parameters (δ_i, γ_j ϵ [0, 1]), which describe the correlation between nests sharing the same subset. This correlation decreases with increasing parameters.

Each alternative belongs to both a purchase channel nest and a terminal delivery service nest (as shown in Figure 1). The allocation parameter a_im (0 ≤ a_im ≤ 1) indicates the proportion of alternative i that belongs to nest m. The value of the allocation parameter corresponding to the selected limb in each subset is not equal to 0, and all of the allocation parameters of alternative i sum to 1 over the nests (${\displaystyle \sum _m{a}_{im}=1}$, m = 1, 2, ……8).

3.2. Utility Function and CHOICE Probability

According to random utility theory, decision-maker n will select an alternative s_i (I = 1, 2, …, n) if and only if the utility U_in provided by alternative s_i is the largest utility value; for instance, U_in > U_jn (j $\in$ C, $\forall i\ne j$).

$$U_{in}={\displaystyle \sum _{l=1}^L{\beta }_l{X}_{inl}}+{\epsilon }_{in}$$

(1)

In Equation (1), X_inl denotes the value of the lth utility variables that are associated with the alternative s_i selected by decision-maker n, and β_l is the unknown parameter that must be estimated. The variable ε_in is a random error item that captures all other factors unobserved by researchers. For convenience, the notation n for the decision-maker is omitted in the following expressions.

Assuming that the distribution for each ε_i is drawn from the standard Gumbel distribution, the following function (as shown in Equation (2)) represents a cumulative extreme value distribution. McFadden’s GEV theorem states that a probability function can be obtained when a generator function satisfies certain conditions for serving as a basis of the distribution of random utilities [43,44]. According to McFadden [43], the probability of alternative i being selected in the CNL can be calculated via Equation (3), which is a closed-form formula. The expression μ_m (0 < μ_m < 1) is a nest dissimilarity parameter, a_im ($0\le a_{im}\le 1$, $\forall i,m$) is an allocation parameter, and V_i = U_in − ${\epsilon }_i$.

$$F({\epsilon }_1,{\epsilon }_2,\cdots {\epsilon }_I)=\exp \left\{-{\displaystyle \sum _m\left({\displaystyle \sum _{i\in N_m}{(a_{im}{e}^{-{\epsilon }_i})}^{1/{\mu }_m}}\right)}\right\}$$

(2)

$$P_i={\displaystyle \sum _m{P}_m{P}_{i|m}}={\displaystyle \sum _m\left\{\frac{{\left[{\displaystyle \sum _{i\in N_m}{(a_{im}{e}^{V_i})}^{\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${\mu }_m$}\right.}}\right]}^{{\mu }_m}}{{\displaystyle \sum _m{\left[{\displaystyle \sum _{i\in N_m}{(a_{im}{e}^{V_i})}^{\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${\mu }_m$}\right.}}\right]}^{{\mu }_m}}}•\frac{{(a_{im}{e}^{V_i})}^{\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${\mu }_m$}\right.}}{{\displaystyle \sum _{i\in N_m}{(e^{V_i})}^{\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${\mu }_m$}\right.}}}\right\}}$$

(3)

The unknown parameters that must be estimated via Equation (3) include the dissimilarity parameter μ_m, the allocation parameter a_im, and the coefficient β_l in U_in.

3.3. Utility Variables

During major public health events, consumers’ choice behaviors regarding fresh food purchase channels and terminal delivery services are influenced by various factors. On the basis of the existing research results, this paper describes the utility variables (as shown in

Table 2. Utility variables and their descriptions.

Utility Variables		Description
Socioeconomic factors
Gender	Dummy variable	1: male; 2: female.
Age	Discrete variable	1: under 18 years old; 2: 18–59 years old; 3: over 59 years old.
Education	Discrete variable	Education level (1: high school and below; 2: associate’s degree; 3: bachelor’s degree; 4: master’s degree or above).
Service factors
Price	Discrete variable	Degree of concern about price (1: completely unconcerned; 2: unconcerned; 3: average; 4: concerned; 5: deeply concerned).
Touch	Discrete variable	Degree of concern about touching (1: completely unconcerned; 2: unconcerned; 3: average; 4: concerned; 5: deeply concerned).
Convenience_M	Discrete variable	The convenience of shopping at farmers’ markets and supermarkets (1: very inconvenient; 2: inconvenient; 3: average; 4: convenient; 5: very convenient).
Convenience_G	Discrete variable	The convenience of shopping via community group-buying (1: very inconvenient; 2: inconvenient; 3: average; 4: convenient; 5: very convenient).
Convenience_S	Discrete variable	The convenience of shopping at a fresh food store (1: very inconvenient; 2: inconvenient; 3: average; 4: convenient; 5: very convenient).
Convenience_E	Discrete variable	The convenience of shopping on an e-commerce platform (1: very inconvenient; 2: inconvenient; 3: average; 4: convenient; 5: very convenient).
Satisfaction_M	Discrete variable	Satisfaction with delivery service of supermarkets and farmers’ markets (1: completely dissatisfied; 2: dissatisfied; 3: average; 4: satisfied; 5: greatly satisfied).
Satisfaction_G	Discrete variable	Satisfaction with delivery service of community group-buying (1: completely dissatisfied; 2: dissatisfied; 3: average; 4: satisfied; 5: greatly satisfied).
Satisfaction_S	Discrete variable	Satisfaction with delivery service of fresh food store (1: completely dissatisfied; 2: dissatisfied; 3: average; 4: satisfied; 5: greatly satisfied).
Satisfaction_E	Discrete variable	Satisfaction with delivery service of e-commerce platform (1: completely dissatisfied; 2: dissatisfied; 3: average; 4: satisfied; 5: greatly satisfied).
Tm_d2dwaiting	Discrete variable	Waiting time for home delivery (1: very troubled; 2: troubled; 3: average; 4: untroubled; 5: completely untroubled).
Tm_pkupwaiting	Discrete variable	Waiting time for self-pickup (1: very troubled; 2: troubled; 3: average; 4: untroubled; 5: completely untroubled).
Risk perception factors
Rickscene	Discrete variable	The infection risk of on-site purchases (1: very low; 2: low; 3: average; 4: high; 5: very high).
Rickpackage	Discrete variable	The infection risk of touching packages (1: very low; 2: low; 3: average; 4: high; 5: very high).
Rickdevice	Discrete variable	The infection risk of touching delivery devices (1: very low; 2: low; 3: average; 4: high; 5: very high).
Rickworker	Discrete variable	The infection risk of contact with delivery staff (1: very low; 2: low; 3: average; 4: high; 5: very high).
Trust perception factors
Cont_temp	Discrete variable	Effectiveness of temperature detection in public places (1: completely ineffective; 2: ineffective; 3: neutral; 4: effective; 5: very effective)
Cont_report	Discrete variable	Effectiveness of detection report (1: completely ineffective; 2: ineffective; 3: neutral; 4: effective; 5: very effective)
Cont_safeguard	Discrete variable	Effectiveness of protective measures for delivery staff (1: completely ineffective; 2: ineffective; 3: neutral; 4: effective; 5: very effective)
Cont_degassing	Discrete variable	Effectiveness of regular disinfection of equipment (1: completely ineffective; 2: ineffective; 3: neutral; 4: effective; 5: very effective)
Tr_pdegassing	Discrete variable	Trust in package disinfection and sterilization (1: very untrusted; 2: untrusted; 3: neutral; 4: trusted; 5: very trusted)
Tr_health	Discrete variable	Trust in the health of delivery staff (1: very untrusted; 2: untrusted; 3: neutral; 4: trusted; 5: very trusted)
Tr_ddegassing	Discrete variable	Trust in device disinfection and sterilization (1: very untrusted; 2: untrusted; 3: neutral; 4: trusted; 5: very trusted)

) that affect consumers’ decision-making in four dimensions—socioeconomic attributes, service attributes, risk perception attributes and trust perception attributes—and uses them to construct the utility function U_in. Socioeconomic attribute variables, which include gender, age and education, describe different socio-economic characteristics of consumers [19,45,46,47]. The service attribute variables describe different service preferences of consumers, which mainly include price, actual touch, convenience, delivery service satisfaction and waiting time [27,37,48,49]. The risk perception attribute variables describe the consumer preference differences for infection risk in the fresh food acquisition process and include the infection risk of onsite purchases, the infection risk of touching packages, the infection risk of touching delivery equipment and the infection risk of delivery staff contact [18,22,50,51]. The trust perception attribute variables describe the difference in consumers’ trust preference for the accepted prevention and control measures, which include the perception of prevention and control measure effectiveness and the degree of trust in each of the measures [18,52,53]. Table 2 provides descriptions of the entire set of utility variables used in the modeling process.

4. Data

The data used in this study were obtained through a Revealed Preference and Stated Preference survey that was conducted as an online form beginning in March 2022. The questionnaire used in this survey was designed with consultation from relevant experts in the field. To guarantee high-quality data, the survey adopted a reward-based method and required that only one questionnaire be filled out by the same user or IP address. A total of 1595 consumer questionnaires were obtained, of which 1208 were valid after those with unqualified factors and missing data were excluded, thus resulting in a rate of 75.73%. Through statistical analysis of the valid questionnaires, the general descriptive statistics of consumer choice are listed in

Table 3. Profile of consumer choices in this survey.

No.	Alternatives		Quantity	Proportion	Subtotal	No.	Alternatives		Quantity	Proportion	Subtotal
1	Traditional market channel	home delivery by couriers	30	2.48%	23.83%	9	Fresh food store channel	home delivery by couriers	80	6.62%	16.06%
2		home delivery by unmanned equipment	15	1.24%		10		home delivery by unmanned equipment	21	1.74%
3		manned pick-up point	224	18.54%		11		manned pick-up point	79	6.54%
4		unattended self-pickup cabinet	19	1.57%		12		unattended self-pickup cabinet	14	1.16%
5	Community group-buying channel	home delivery by couriers	104	8.61%	39.15%	13	E-commerce platform channel	home delivery by couriers	153	12.75%	20.95%
6		home delivery by unmanned equipment	52	4.30%		14		home delivery by unmanned equipment	18	1.49%
7		manned pick-up point	257	21.27%		15		manned pick-up point	60	4.97%
8		unattended self-pickup cabinet	60	4.97%		16		unattended self-pickup cabinet	21	1.74%

.

According to the statistical results, the preference order of consumers for fresh food purchase channels during major public health events is community group buying, traditional markets, e-commerce platforms, and fresh food stores. Consumers who choose community group-buying channels and traditional market channels prefer to pick up at manned pick-up points (21.27% and 18.54% of the total number of respondents, respectively), whereas those who choose fresh food stores and e-commerce platforms are more likely to prefer home delivery by couriers (6.62% and 12.75% of the total number of respondents, respectively).

The data in Table 3 indicate that 30.46% of the respondents choose home delivery by couriers, 8.77% choose home delivery by unmanned equipment, 51.32% choose manned pick-up points, and 9.44% choose unattended self-pickup cabinets. Although there is close contact between consumers and service personnel in the two services of a manned pick-up point and home delivery by couriers, they are the preferred methods of consumers. Consumers are still less receptive to unattended self-pickup cabinets and home delivery by unmanned equipment (i.e., contactless delivery).

It can be seen from the above data that consumers’ choice of fresh food purchase channels and terminal delivery services during major public health events is a decision process that considers the consumers’ personal service needs. However, these data make it difficult to explain the correlation between the service schemes formed by fresh food purchase channels and terminal delivery services. Furthermore, the influence of utility variables on such correlations is not readily discernible from general statistical characteristics, so the following text utilizes the constructed CNL model to analyze the data from 1208 questionnaires exhibiting the aforementioned sample characteristics. The purpose is to reveal the law of consumers’ joint choice of fresh food purchase channels and terminal delivery services during major public health events.

5. Results and Analysis

5.1. Results

As depicted in Figure 1, the CNL model comprises 8 nests and 16 alternatives, producing 8 dissimilar parameters and 32 allocation parameters. The existence of many unknown parameters leads to an expensive estimation process and an overparameterized model. With reference to other studies [54,55], we constrained all nonzero allocation parameters to a value of one-half. When fixed allocation parameters are used, nested likelihood ratio tests cannot be used. However, the models can still be evaluated via Rho-square bar statistics, which consider the cost of a model in terms of the number of parameters. The CNL model’s parameter estimation was performed via the software Biogeme 2.6a and the maximum likelihood estimation method [56]. The estimation results are presented in

Table 4. Parameter estimation results.

Category	Parameter	Value	t-Stat
Socioeconomic attributes	B_Gender	0.337	1.24
	B_Age	−0.107	−0.18
	B_Education	0.544	2.30 **
Service attributes	B_Price	−0.0553	−0.30
	B_Touch	0.650	3.45 ***
	B_Convenience_M	1.15	4.39 ***
	B_Convenience_G	1.18	4.31 ***
	B_Convenience_S	0.654	2.87 ***
	B_Convenience_E	1.47	4.50 ***
	B_Satisfaction_M	1.16	4.18 ***
	B_Satisfaction_G	1.31	4.95 ***
	B_Satisfaction_S	1.17	3.91 ***
	B_Satisfaction_E	1.44	4.35 ***
	B_Tm_d2dwaiting	0.338	2.70 ***
	B_Tm_pkupwaiting	−0.182	−1.43
Risk perception attributes	B_Rickscene	−0.0461	−0.44
	B_Rickpackage	−0.406	−2.76 ***
	B_Rickdevice	−0.0803	−0.88
	B_Rickworker	−0.434	−7.04 ***
Trust perception attributes	B_Cont_temp	0.214	1.79 **
	B_Cont_report	−0.0820	−0.64
	B_Cont_safeguard	0.543	4.96 ***
	B_Cont_degassing	0.00583	0.06
	B_Tr_pdegassing	−0.224	−1.14
	B_Tr_health	0.272	2.08 **
	B_Tr_ddegassing	0.195	1.54
Dissimilarity parameter	MU_δ1	1.00	0.00
	MU_δ2	0.373	6.38 ***
	MU_δ3	0.372	4.92 ***
	MU_δ4	0.634	2.72 ***
	MU_γ1	0.389	5.37 ***
	MU_γ2	0.297	5.39 ***
	MU_γ3	0.400	5.17 ***
	MU_γ4	0.303	2.34 ***
Final log likelihood		−2455.690
Rho-square-bar for the init. model		0.267
Sample size		1208

*: significant at the 10% level; **: significant at the 5% level; ***: significant at the 1% level.

.

From the values presented in Table 4, except for the estimated parameters for gender, age, price, Tm_pkupwaiting, Rickscene, Rickdevice, Cont_report, Cont_degassing, Tr_pdegassing, and Tr_ddegassing, the other estimated parameters were statistically significant, which indicates that they have a significant effect on the consumer joint choice behavior of fresh food purchase channels and terminal delivery services during major public health events. The estimated parameters for Rickpackage and Rickworker are negative, indicating that the utility of alternatives decreases with increases in the infection risk of touching packages and contact with delivery staff. The estimated parameters for other utility variables are positive, indicating a positive effect on consumer joint choice behavior, a finding that is consistent with research expectations.

The dissimilarity parameters are significant, except for that of the traditional market channel nest. The magnitudes of these parameters reflect the substitutability among nests within the same subset. A comparison of the values of the dissimilarity parameters reveals that the value of MU_δ₄ is the highest, indicating that e-commerce platform channels exhibit high independence, which means that consumers first consider adjusting their attitudes toward e-commerce platforms when the shopping scenario changes. With respect to terminal delivery services, the value of MU_γ₂ is the smallest, thus indicating that each fresh food purchase channel is easily replaced for consumers who choose home delivery via unmanned equipment. Moreover, the value of MU_γ₃ is higher than that of other terminal delivery services, indicating that consumers are more willing to change their choice to manned pick-up points when the prevention and control requirements of major public health events change.

5.2. Elasticity Analysis

The elasticity is calculated via utility variables with significant parameter estimation results in the CNL model. The change in probability of each alternative given one unit of change in a given utility variable is analyzed. Direct elasticity (DE) is related to the inherent properties of alternative solutions, whereas cross-elasticity (CE) is related to the properties of competing alternative solutions. The direct elasticity is computed with the expression given by Equation (4), and the cross-elasticity is calculated via Equation (5).

$$E_{P_i-x_{il}}={\displaystyle \frac{\partial P_i}{\partial x_{il}}}\cdot {\displaystyle \frac{x_{il}}{P_i}}$$

(4)

$$E_{P_i-x_{jl}}={\displaystyle \frac{\partial P_i}{\partial x_{jl}}}\cdot {\displaystyle \frac{x_{jl}}{P_i}}$$

(5)

where $E_{P_i-x_{il}}$ is the direct elasticity of the probability of selecting alternative s_i with respect to a marginal change in a given attribute x_il and where $E_{P_i-x_{jl}}$ is the cross-elasticity of the probability of selecting alternative s_i with respect to a marginal change in the value of the lth attribute of alternative s_j.

Table 5. Elastic analysis results of the utility variables.

		Home Delivery by Couriers				Home Delivery by Unmanned Equipment				Manned Pick-Up Point				Unattended Self-Pickup Cabinet
		DE	90%CI	CE	90%CI	DE	90%CI	CE	90%CI	DE	90%CI	CE	90%CI	DE	90%CI	CE	90%CI
Traditional market channel	Education	--		1.298	(0.26, 2.51)	--		1.502	(0.28, 3.25)	--		1.01	(0.22, 1.87)	--		1.504	(0.28, 3.27)
	Touch	2.195	(1.01, 3.77)	--		2.39	(0.89, 4.73)	--		1.681	(0.76, 2.80)	--		2.395	(0.90, 4.76)	--
	Convenience_M	3.948	(1.99, 6.49)	--		4.277	(1.84, 8.30)	--		3.029	(1.59, 4.83)	--		4.293	(1.85, 8.36)	--
	Convenience_G	--		3.865	(1.83, 6.87)	--		4.246	(1.80, 8.52)	--		2.953	(1.45, 5.12)	--		4.247	(1.79, 8.58)
	Convenience_S	--		2.254	(0.72, 1.14)	--		2.472	(0.76, 4.93)	--		1.738	(0.60, 2.98)	--		2.478	(0.76, 4.97)
	Convenience_E	--		5.11	(2.89, 8.45)	--		5.652	(2.30, 11.4)	--		3.905	(2.17, 6.30)	--		5.652	(2.31, 11.44)
	Satisfaction_M	3.114	(1.26, 5.51)	--		3.263	(1.19, 6.57)	--		2.377	(1.03, 3.86)	--		3.291	(1.20, 6.68)	--
	Satisfaction_G	--		2.508	(1.23, 3.90)	--		2.778	(1.12, 5.11)	--		1.892	(0.93, 2.89)	--		2.771	(1.11, 5.12)
	Satisfaction_S	--		2.576	(1.23, 4.36)	--		2.942	(1.09, 5.81)	--		2.035	(0.99, 3.32)	--		2.946	(1.09, 5.84)
	Satisfaction_E	--		2.701	(1.23, 4.28)	--		3.263	(1.28, 6.03)	--		2.088	(0.96, 3.29)	--		3.253	(1.27, 6.03)
	Tm_d2dwaiting	1.034	(0.31, 1.96)	--		--		1.039	(0.30, 2.12)	0.788	(0.21, 1.59)	--		--		1.04	(0.30, 2.12)
	Rickpackage	--		−1.345	(−2.43, 0.45)	--		−1.548	(−3.42, −0.44)	--		−1.054	(−1.80, −0.35)	--		−1.551	(−3.44, −0.44)
	Rickworker	−0.894	(−1.13, −0.54)	--		--		−1.227	(−2.18, −0.57)	−0.791	(−1.29, −0.44)	--		--		−1.229	(−2.18, −0.57)
	Cont_temp	0.464	(0.03, 1.05)	--		0.503	(0.04, 1.43)	--		0.344	(0.03, 0.78)	--		0.506	(0.04, 1.44)	--
	Cont_safeguard	1.178	(0.59, 1.86)	--		--		1.149	(0.49, 2.05)	0.911	(0.44, 1.47)	--		--		1.15	(0.49, 2.05)
	Tr_health	0.906	(0.36, 1.87)	--		--		0.963	(0.27, 2.32)	0.692	(0.26, 1.46)	--		--		0.964	(0.28, 2.33)
Community group-buying channel	Education	1.215	(0.21, 2.36)	--		1.46	(0.24, 3.06)	--		1.205	(0.21, 2.31)	--		1.462	(0.24, 3.08)	--
	Touch	--		1.798	(0.79, 3.02)	--		2.137	(0.82, 3.83)	--		1.793	(0.79, 2.96)	--		2.138	(0.82, 3.84)
	Convenience_M	--		2.866	(1.41, 4.54)	--		3.364	(1.56, 5.76)	--		2.873	(1.45, 4.49)	--		3.363	(1.55, 5.77)
	Convenience_G	3.945	(1.73, 6.84)	--		4.726	(1.98, 8.86)	--		3.901	(1.77, 6.65)	--		4.738	(1.97, 8.91)	--
	Convenience_S	--		1.806	(0.53, 3.29)	--		2.138	(0.59, 4.18)	--		1.806	(0.56, 3.22)	--		2.138	(0.59, 4.20)
	Convenience_E	--		4.61	(2.60, 7.28)	--		5.422	(2.66, 9.49)	--		4.564	(2.62, 7.10)	--		5.424	(2.66, 9.52)
	Satisfaction_M	--		1.544	(0.65, 2.57)	--		1.792	(0.67, 3.38)	--		1.578	(0.69, 2.57)	--		1.788	(0.66, 3.38)
	Satisfaction_G	3.46	(1.54, 5.32)	--		4.19	(1.62, 7.01)	--		3.387	(1.56, 5.13)	--		4.21	(1.62, 7.07)	--
	Satisfaction_S	--		2.117	(0.98, 3.50)	--		2.53	(1.01, 4.59)	--		2.127	(1.02, 3.46)	--		2.529	(1.00, 4.60)
	Satisfaction_E	--		2.67	(1.15, 4.11)	--		3.168	(1.17, 5.44)	--		2.625	(1.16, 3.99)	--		3.164	(1.16, 5.45)
	Tm_d2dwaiting	0.918	(0.28, 1.57)	--		--		1.037	(0.32, 1.78)	0.911	(0.27, 1.56)	--		--		1.039	(0.33, 1.78)
	Rickpackage	--		−1.229	(−2.14, −0.40)	--		−1.471	(−2.95, −0.44)	--		−1.222	(−2.09, −0.40)	--		−1.472	(−2.96, −0.44)
	Rickworker	−0.85	(−1.15, −0.53)	--		--		−1.108	(−1.59, −0.63)	−0.858	(−1.17, −0.54)	--		--		−1.108	(−1.59, −0.63)
	Cont_temp	--		0.321	(0.02, 0.67)	--		0.382	(0.03, 0.89)	--		0.321	(0.02, 0.66)	--		0.382	(0.03, 0.89)
	Cont_safeguard	1.076	(0.54, 1.56)	--		--		1.193	(0.58, 1.82)	1.068	(0.54, 1.55)	--		--		1.196	(0.58, 1.82)
	Tr_health	0.796	(0.33, 1.49)	--		--		0.93	(0.33, 1.87)	0.791	(0.33, 1.48)	--		--		0.931	(0.33, 1.87)
Fresh food store channel	Education	--		1.404	(0.30, 2.67)	--		1.494	(0.27, 3.16)	--		1.281	(0.29, 2.40)	--		1.494	(0.27, 3.13)
	Touch	2.28	(0.95, 4.14)	--		2.398	(0.81, 4.90)	--		2.08	(0.89, 3.57)	--		2.402	(0.81, 4.93)	--
	Convenience_M	--		3.704	(1.83, 6.50)	--		3.865	(1.59, 7.70)	--		3.416	(1.75, 5.63)	--		3.868	(1.58, 7.75)
	Convenience_G	--		4.073	(1.92, 7.61)	--		4.241	(1.68, 8.74)	--		3.7	(1.84, 6.60)	--		4.242	(1.67, 8.80)
	Convenience_S	2.432	(0.67, 4.56)	--		2.561	(0.65, 5.38)	--		2.214	(0.68, 3.95)	--		2.569	(0.65, 5.43)	--
	Convenience_E	--		5.341	(2.73, 9.55)	--		5.58	(2.19, 11.52)	--		4.855	(2.55, 8.25)	--		5.575	(2.18, 11.57)
	Satisfaction_M	--		2.311	(1.05, 3.91)	--		2.422	(0.85, 4.68)	--		2.188	(1.04, 3.43)	--		2.425	(0.84, 4.71)
	Satisfaction_G	--		2.773	(1.35, 4.50)	--		2.896	(1.14, 5.42)	--		2.496	(1.26, 3.86)	--		2.891	(1.12, 5.44)
	Satisfaction_S	3.372	(1.42, 6.51)	--		3.626	(1.20, 7.88)	--		3.056	(1.37, 5.39)	--		3.648	(1.20, 7.98)	--
	Satisfaction_E	--		3.02	(1.35, 4.92)	--		3.205	(1.23, 6.03)	--		2.736	(1.27, 4.29)	--		3.191	(1.22, 6.04)
	Tm_d2dwaiting	1.068	(0.31, 2.24)	--		--		1.05	(0.30, 2.22)	0.971	(0.26, 2.06)	--		--		1.051	(0.30, 2.23)
	Rickpackage	--		−1.47	(−2.81, −0.49)	--		−1.58	(−3.49, −0.44)	--		−1.348	(−2.38, −0.45)	--		−1.583	(−3.51, −0.44)
	Rickworker	−1.004	(−1.58, −0.56)	--		--		−1.217	(−2.13, −0.56)	−1.015	(−1.66, −0.55)	--		--		−1.217	(−2.13, −0.56)
	Cont_temp	0.429	(0.03, 1.00)	--		0.458	(0.03, 1.22)	--		0.388	(0.03, 0.89)	--		0.459	(0.03, 1.22)	--
	Cont_safeguard	1.266	(0.60, 2.11)	--		--		1.214	(0.50, 2.19)	1.16	(0.56, 1.91)	--		--		1.217	(0.50, 2.20)
	Tr_health	0.943	(0.33, 2.01)	--		--		0.963	(0.27, 2.29)	0.858	(0.31, 1.89)	--		--		0.964	(0.27, 2.29)
E-commerce platform channel	Education	1.309	(0.22, 2.65)	--		1.54	(0.24, 3.68)	--		1.286	(0.23, 2.57)	--		1.544	(0.23, 3.72)	--
	Touch	--		1.849	(0.73, 3.17)	--		2.163	(0.70, 4.43)	--		1.833	(0.75, 3.08)	--		2.164	(0.69, 4.45)
	Convenience_M	--		3.027	(1.34, 5.33)	--		3.492	(1.38, 7.41)	--		3.02	(1.40, 5.20)	--		3.492	(1.37, 7.46)
	Convenience_G	--		3.504	(1.54, 6.21)	--		4.032	(1.26, 8.37)	--		3.443	(1.59, 6.03)	--		4.032	(1.55, 8.44)
	Convenience_S	--		1.888	(0.51, 3.40)	--		2.184	(0.54, 4.55)	--		1.876	(0.54, 3.31)	--		2.184	(0.54, 4.58)
	Convenience_E	5.481	(2.47, 9.86)	--		6.368	(2.29, 14.11)	--		5.379	(2.54, 9.49)	--		6.39	(2.28, 14.25)	--
	Satisfaction_M	--		1.638	(0.60, 2.93)	--		1.938	(0.63, 3.95)	--		1.673	(0.64, 2.90)	--		1.935	(0.62, 3.97)
	Satisfaction_G	--		2.478	(1.08, 4.00)	--		2.826	(1.04, 5.51)	--		2.4	(1.08, 3.79)	--		2.825	(1.03, 5.55)
	Satisfaction_S	--		2.205	(0.89, 4.22)	--		2.627	(0.86, 5.74)	--		2.199	(0.92, 4.08)	--		2.629	(0.86, 5.79)
	Satisfaction_E	4.165	(1.57, 7.16)	--		4.85	(1.65, 9.87)	--		4.054	(1.61, 6.78)	--		4.88	(1.64, 10.00)	--
	Tm_d2dwaiting	0.966	(0.25, 1.84)	--		--		1.03	(0.25, 2.19)	0.952	(0.24, 1.82)	--		--		1.032	(0.25, 2.20)
	Rickpackage	−1.242	(−2.41, 0.41)	--		−1.443	(−3.31, −0.40)	--		−1.226	(−2.32, −0.41)	--		−1.444	(−3.32, −0.40)	--
	Rickworker	−0.932	(−1.48, −0.47)	--		--		−1.235	(−2.36, −0.52)	0.94	(−1.49, −0.49)	--		--		−1.238	(−2.37, −0.51)
	Cont_temp	--		0.315	(0.02, 0.73)	--		0.378	(0.03, 1.08)	--		0.313	(0.03, 0.72)	--		0.377	(0.03, 1.08)
	Cont_safeguard	1.159	(0.54, 1.87)	--		--		1.207	(0.49, 2.26)	1.142	(0.54, 1.84)	--		--		1.211	(0.49, 2.27)
	Tr_health	0.827	(0.28, 1.81)	--		--		0.934	(0.24, 2.49)	0.815	(0.28, 1.78)	--		--		0.935	(0.24, 2.50)

lists the elasticity of a given utility variable on each alternative, with their respective 90% confidence intervals (90% CI).

In terms of socioeconomic attributes, the choice probabilities of alternatives reveal significant and different marginal effects with respect to education level. From the data, the following conclusions are drawn. During the duration of major public health events, the community group-buying channel is the least sensitive to education level, which means that people from all walks of life are more willing to shop through the community group-buying channel. With respect to terminal delivery services, education level significantly increases consumers’ probability of choosing an unattended self-pickup cabinet, while the change in the probability of choosing unattended self-pickup cabinet services under e-commerce platform channels is the greatest (elasticity value is 1.544). The probability of consumers choosing the manned pick-up point increases most insignificantly, and the change in the consumer choice probability of the manned pick-up point under traditional market channels is minimal (elasticity value is 1.010).

With respect to service attributes, fresh food store channels are the most affected by changes in touch. Given a 1% increase in touch, the choice probabilities of the four terminal delivery services in the fresh food store channel increase by 2.280%, 2.398%, 2.080% and 2.402%, respectively. In terminal delivery services, consumers who choose manned pick-up points are the least sensitive to touch, and those who choose unattended self-pickup cabinets are the most sensitive. Tm_d2dwaiting has the same effect as touch, but the choice probability of each alternative only changes slightly when Tm_d2dwaiting changes (elasticity value between 0.788 and 1.068). This suggests that consumers are less concerned about the waiting time for terminal delivery services during major public health events. Compared with touch and Tm_d2dwaiting, consumers are more concerned about convenience, with Convenience_E exhibiting the most significant effect on consumers’ choice of other purchase channels. Consumer satisfaction with terminal delivery services is significantly different from convenience. Consumer satisfaction with the terminal delivery service of a specific purchase channel has a significant effect on the channel itself. In addition, the elasticity values of satisfaction indicate that their impact on contactless terminal delivery service is greater than that on attended terminal delivery service. This reflects that an increase in consumer satisfaction with terminal delivery services significantly motivates customers to opt for contactless terminal delivery services.

In terms of risk perception attributes, Rickpackage has a stronger impact than Rickworker. From the perspective of the purchase channel, the fresh food store channel is the most sensitive to Rickpackage, and the community group-buying channel is the least sensitive to Rickworker. This finding indicates that the greater the infection risk that consumers perceive from touching the package, the less likely they are to choose to buy in fresh food stores. The greater the infection risk perceived from contact with delivery personnel, the greater the likelihood that consumers are to shop through community group buying. From the terminal delivery perspective, the contactless terminal delivery service is more sensitive to Rickpackage and Rickworker than the attended terminal delivery service is, and the elastic values of the two contactless terminal delivery services are similar. This reflects that even when consumers perceive increased infection risk, they are still willing to choose home delivery by couriers or manned pick-up points.

In terms of trust perception attributes, Cont_safeguard has the greatest impact, followed by Tr_health and Cont_temp. Cont_safeguard has the greatest impact on the fresh food store and the least impact on the manned pick-up point. The impact of Tr_health on purchase channels and terminal delivery services is almost the same as that of Cont_safeguard. The reason why Cont_safeguard and Tr_health have similar rules is probably that both variables reflect the issue of trust in delivery personnel. Cont_temp demonstrated the least effect, meaning that consumers were generally insensitive to temperature.

6. Discussion

Some management suggestions can be obtained in this section by discussing the influence of utility variables on consumer joint choice behavior of fresh food purchase channels and terminal delivery services. The specific contents of this paper are as follows.

In the context of major public health events, buying fresh food through community group-buying channels is a priority for many people. Therefore, fresh food retail enterprises should prioritize setting up community group-buying channels and ensure the quality of goods purchased by consumers through these channels while improving after-sales service capabilities [57]. They should also conduct an in-depth analysis of various consumer service demands toward community group-buying channels and fully tap into the consumption potential of consumers in these channels. The ultimate goal is to “stick” consumers via high-quality products and services, thereby enhancing the competitive advantage of fresh retail enterprises in community group-buying channels.

Despite the greater tolerance for service quality during the shopping process during major public health events, consumers still have a high demand for shopping convenience. Gruntkowski and Martinez obtained similar conclusions by assessing the impact of COVID-19 on online grocery shopping in Germany [49]. Hence, fresh retail enterprises should combine the advantages of different purchasing channels and effectively utilize various terminal delivery services, starting with aspects such as service time, service location, and service methods to improve the convenience of consumer shopping during major public health events. In particular, the convenience of contactless terminal delivery services should be considered, as it has the potential to significantly reduce the gathering and movement of personnel.

There remain obstacles in the promotion of contactless terminal distribution services [58]. To accelerate consumer acceptance of contactless terminal delivery services, enterprises should accurately identify their target audience during the promotion process and initially provide the service to consumers with higher levels of education [59]. Second, during the design process of contactless terminal delivery services, it is necessary to enhance the development of communication functions and increase investment in visualization, traceability and information resources to address the interactive challenges of contactless terminal delivery services. By optimizing the quality of noncontact communication for consumers and publicizing the operational procedures and potential advantages, enterprises can increase their satisfaction with terminal delivery services, thereby fostering the widespread adoption of contactless delivery services [59].

Despite the risk of infection, consumers still prefer and trust the terminal delivery services provided by personnel. In order to meet the consumer’s demand, enterprises need to take relevant measures to alleviate concerns and ensure consumer safety [53]. Strengthening the control of personnel protection measures and health, continuing to implement strict safety measures, and making consumers believe that enterprises regard their personal safety as the top priority are important means for alleviating consumers’ concerns about personal safety and strengthening consumer trust. These measures also serve as the “golden key” for enhancing an enterprise’s ability to resist risk.

7. Conclusions

On the basis of random utility maximization theory, this study constructs a joint choice model of fresh food purchase channels and terminal delivery service (CNL model) during major public health events. The model was analyzed via questionnaire survey data, and the main research conclusions are as follows.

According to the fitting results of the CNL model, the model exhibits satisfactory statistical characteristics, making it a suitable analytical tool for addressing multidimensional choice issues related to consumers’ joint selection behavior of fresh food purchase channels and terminal delivery services during major public health events. The parameter estimates also indicate that the model accurately reflects the effect of utility variables on consumer choice behavior. Among the eight dissimilarity parameter estimates, the dissimilarity parameter of the e-commerce platform channel is the largest, whereas the dissimilarity parameter of home delivery by unmanned equipment is the smallest. This finding indicates that during major public health events, the e-commerce platform channel exhibits high independence and low substitutability, whereas the opposite is true for home delivery via unmanned equipment; this also reflects the relatively low acceptance of the unmanned delivery format among consumers, as well as the crucial role that e-commerce platforms play in ensuring material supply.

The results of the elasticity analysis suggest that people are more willing to shop through community group-buying channels during major public health events, so enterprises can enhance their competitive advantages by guaranteeing the rights and interests of consumers within this channel. In the shopping process, delivery time has less influence on consumer behavior. Hence, businesses should focus mainly on the convenience of shopping, especially the convenience of the e-commerce platform channel, which affects consumers’ choice of other fresh food purchase channels the most. In terms of terminal delivery services, although human contact increases the risk of infection, consumers still prefer manned terminal delivery services over contactless options, indicating a relatively low acceptance of the latter. Since delivery service satisfaction has a significant effect on contactless terminal delivery service, improving the quality of contactless terminal delivery effectively stimulates its further promotion. Finally, an effective means to increase consumer trust in enterprises is to strengthen personnel protection measures.

There are also several limitations in this study. First, there are interactions among socioeconomic attribute variables, service attribute variables, risk perception attribute variables, and trust perception attribute variables that are not discussed in this paper. Second, under the influence of major public health events, consumer choices of fresh food purchase channels and terminal delivery services are also affected by external factors such as government policies, technical level, social approval and preferential promotions, which have not been covered. These limitations will be addressed and improved upon in subsequent research to make the findings more credible and more applicable to other situations.