Introduction

At present, China is the largest grape-producing country in the world and its grape production reached 14.314 million tons by the end of 2020 (National Bureau of Statistics of China [NBSC], 2021). The Chinese grape industry has made great strides because of progress in cultivation techniques and diversity of grape varieties (Jiang et al., 2018). However, the high yield of grapes faces some challenges in economic consumption. First, table grapes are not resistant to storage. Under the impact of abundance of substitute fruits, the sales volume of grapes still has room for growth in the consumption market of table grapes in China. Second, the concentrated listing of table grapes has led to the phenomenon of periodic oversupply, thereby leading to fluctuations in grape prices (Mu et al., 2019). Therefore, it is particularly important to systematically analyze the demand of consumption market and reasonably arrange production according to changes in market demands.

China has a large population. Consumers with different sociodemographic characteristics, purchasing behavior, and perceived values are likely to have different preferences for various attributes of grapes. According to the relevant theories of behavioral economics, consumer behavior and preferences are not always rational and are influenced by many factors. Exploring the factors influencing consumer preferences helps to promote sustainable consumption. Understanding the preferences of table grape consumers and studying the factors that affect their preferences provide breeding workers with reference breeding directions (Si-Hong et al., 2020), and enable operators of the table grape industry to accurately predict consumer requirements and formulate appropriate marketing strategies (Qian et al., 2018; Zhou et al., 2015). In order to determine consumer preferences, some studies were conducted on their preferences of table grapes in China; however, preference predictions were not made (Mu et al., 2019; Yao et al., 2020; Zhou et al., 2016). The table grape market in China is facing a situation where grape varieties are highly similar, coupled with the impact of alternative products, resulting in fewer choices for consumers. If consumer demands are not predicted in time, their purchasing power is reduced greatly. Therefore, in view of the competitive environment of China’s table grape consumption market, it is of great practical significance to build consumer preference prediction models on the basis of understanding consumer preferences.

The attributes of fruit have an important impact on purchasing behavior. The attributes of table grapes mainly include intrinsic attributes and extrinsic attributes. Intrinsic attributes, such as size, shape, color, aroma type, aroma degree, texture, taste, seedless degree, skin powder, peeling degree, etc., are usually inherent in grapes itself and cannot be changed (Mu et al. 2019). The extrinsic attributes mainly refer to the attributes affected by external factors, such as price, packaging, place of origin, etc. At present, most studies believe that the intrinsic attributes of grapes mainly affect purchasing behavior. Therefore, these studies researched the intrinsic attribute preferences of Chinese table grape consumers. For example, a study conducted on the table grape consumption market in Jiangsu Province investigated consumer preferences and their consumption characteristics of table grapes (Yao et al., 2020), but this survey faced a problem of uneven sample distribution. Another study conducted on the table grape market in Beijing summarized the preferences of most Chinese consumers, but the survey had a small overall sample size of 315 (Zhou et al., 2015). In these two studies, the intrinsic attributes of table grapes were incomplete, lacking the analysis of consumer preferences for skin powder and peeling degree as well as the analysis of factors affecting consumer preferences for each attribute, so it was impossible to predict specific consumer preferences. Moreover, the above two surveys were conducted in a certain region; hence, the results could not represent the consumption preferences of entire Chinese consumers.

The quality of grapes cultivated under expert guidance has improved (Lohitnavy et al., 2010; Piazzolla et al., 2016; Poudel et al., 2022); however, the traits cultivated by scholars failed to represent consumer preferences. Therefore, some researchers explored their preferences for table grapes from the perspective of consumers (Mu et al., 2016; Zamzami et al., 2020). However, as yet, little research is done on the prediction of consumer preferences for table grape attributes; hence, we referred to the methods used to study the influencing factors of consumer preferences for attributes or purchasing intention of other fruits.

Many studies used statistical methods to analyze consumer product attribute preferences and its influencing factors. Logistic regression model is mostly used in predicting purchasing intention of consumers because of its simple and practical advantages. Ma et al. (2023) used logistic regression to explore the effects of dietary knowledge and age on food preferences. The study determined that in China, the proportion of people aged more than 18 years with healthier food preferences was highest. The lower the level of dietary knowledge, the more likely they were to prefer non-healthy foods (Ma et al. 2023).

Romadhon et al. (2021) used logistic regression to analyze the influence of individual characteristics, such as age, marital status, education, and gender, on the preference of Nigerian consumers for attributes of rice, such as color, head rice, flavor, aroma, soft texture, durability, and its whiteness. Massaglia et al. (2023) used logistic regression to explore the influence of social demographic characteristics of Italian consumers on chocolate attributes, such as type, brand, and label information. Chu et al. (2020) used multivariate logistic regression to predict consumer preferences for wine attributes. Mu et al. (2022) also used logistic regression model for recommending consumers’ wine attributes. In conclusion, logistic regression has an important significance for predicting consumer preferences regarding attributes of table grapes.

The above analysis demonstrates that, based on the framework theory in behavioral economics, people’s behavioral biases or preferences have a certain impact on economic decision-making and market operations. Therefore, the present study first conducted a comprehensive survey on the consumption market of table grapes in China to obtain overall preferences of consumers; it then explored the factors affecting their preferences from the perspective of table grape consumers, and finally constructed the preference prediction models of table grapes. Our results would enhance the understanding of consumer preferences and purchasing behavior of table grapes, and provide reference and guidance for domestic and foreign grape producers and marketers.

Materials and methods

Design of questionnaire

Based on the analysis of previous literature, we designed a questionnaire for this survey, discussed the problems in the questionnaire with experts in the field of table grapes, and revised the questionnaire repeatedly. The questionnaire consisted of four parts. The first part investigated the participants’ preferences for various intrinsic attributes of grapes, mainly including visual attributes, such as size, shape, and color, and value attributes, such as aroma type, aroma degree, texture, taste, seedless degree, skin powder, and peeling degree. In order to explore the factors that affect consumers’ intrinsic attribute preferences for table grapes, subsequent parts investigated the purchasing behavior, perceived value factors, and sociodemographic characteristics of participants. Among these, the perceived value factors referred to the importance that consumers attached to the intrinsic and extrinsic attributes of table grapes, which was consumers’ evaluation of the product (Cronin Jr. et al., 2000). These questions were designed with five-point Likert scale with the following specifies: 1 means “very unimportant,” 2 means “unimportant,” 3 means “medium,” 4 means “important,” and 5 means “very important.” The influencing factors and prediction attributes used in the prediction models constructed in this study are shown in Figure 1.

Survey

The survey was conducted from June 2022 to July 2022. We selected and trained 28 students from China Agricultural University as investigators. The training content included survey requirements, sampling methods, and interview skills. After training, investigators were directed to select participants by random sampling in their home province. The participants were required to have good independent thinking and understanding to ensure that they were serious in answering the survey, and that the survey results were more effective. The investigators were selected from different provinces and autonomous regions of China so that the survey results obtained by us were representative of China’s geographical regions. The participants belonged to 32 provinces and autonomous regions of China, and the sample distribution of each region confirmed the population distribution of the seventh national population census of 2020. A total of 4678 questionnaires were distributed in this survey. After eliminating incomplete fillings, short answering time, and repeated questionnaires, 4324 valid questionnaires were obtained, with a high recovery rate of 92.43%.

Data collection

Cronbach’s α was used to test reliability of the questionnaire. When the reliability was between 0.70 and 0.90, it indicated consistent variables within the scale (Tavakol and Dennick, 2011); when it exceeded 0.9, it indicated that the variables within the scale had high reliability. This paper used SPSS 27.0 to test reliability of the scale data in the sample, and the value was between 0.915 and 0.70, indicating that our survey data had a good consistency.

Table 1 shows the sociodemographic characteristics of all participants. A large number of studies demonstrated that the sociodemographic characteristics of consumers, such as gender, age, education, occupation, marital status, and personal disposable income, had an important impact on consumers’ purchasing behavior and preferences (Del Carmen et al., 2020; Okpiaifo et al., 2020; Park et al., 2021; Saeed et al., 2020). In our survey results, women (58.81%) were more than men (41.19%). Most of the participants were aged between 18 and 25 years and had bachelor’s degrees or above. Nearly one-third of the participants were company employees. More participants were unmarried, had three persons at home, had a personal disposable income <2000 renminbi (RMB, China’s currency), and were urbanites. This demonstrated that the survey participants were mainly young or middle-aged, well-educated consumers with medium income.

Analysis of influencing factors

In order to construct prediction models of consumers’ attribute preferences for table grapes, this paper used χ² analysis of contingency table to explore the influence and correlation of consumers’ sociodemographic characteristics, purchasing behavior, and perceived value factors to select independent variables of prediction models. χ² test was a hypothesis test based on χ² distribution. First, it was assumed that each influencing factor was not related to attribute preferences of fresh grapes, and then calculated deviation between observed value and theoretical value, namely χ² value. If the result of χ² test was significant and the χ² value was large, it tended to reject the original hypothesis, indicating that this influencing factor was related to the corresponding attribute preferences of table grapes.

Prediction model

Logistic regression is often used for model prediction (Pucci et al., 2019). The dependent variables used in this study were binary variables, so this paper used binary logistic regression to construct consumer preference prediction models (Allison 2012). First, each option of independent variables was assigned, as shown in Table 2.

The binary logistic regression model is shown in Equation (1), where y represents the attribute value selected by the consumer, and P represents the proba- bility that the consumer prefers y. y In Equation (1), y is a linear combination of each independent variable, x_i (i = 1,2,…, n) . In Equation (2), β₀ represents the con- stant term β_i, representing the regression coefficient of independent variable, xi, and n is the number of influ- encing factors. Another common linear form of binary logistic model can be obtained by logit transformation of P, as shown in Equation (3).

Results and discussion

Consumers’ intrinsic attribute preferences for table grapes

The attributes of table grapes mainly include intrinsic attributes and extrinsic attributes. The intrinsic attributes referred to in this paper are inherent and unchangeable attributes of grapes (Mu et al. 2019), including size, shape, color, aroma type, aroma degree, texture, taste, seedless degree, skin powder, and peeling degree. Among these, size, shape, and color are the most important visual attributes for consumers (Predieri et al., 2021), while the attributes related to taste, such as type and degree of aroma, texture, taste, seedless degree, skin powder, and peeling degree, determine the value experience of grape consumption (Zhou et al., 2015). This paper analyzed the intrinsic attributes of table grapes that have attracted much attention. The statistical results of consumer preferences for various intrinsic attributes of table grapes are summarized in Table 3.

Our results showed that in terms of berry size, the participants preferred medium berries of about 2 cm in diameter (68.20%), although large berries of about 2.5 cm in diameter (52.64%) also had great market potential. In terms of berry shape, the participants preferred spherical or near-spherical shape (63.44%), although more than half of the participants were also interested in oval berries, partly because consumers had limited knowledge of elongated finger and cylindrical grapes. In terms of berry color, the participants preferred purple-red (57.08%) and purple-black (53.01%), although green (45.40%) berries also occupied a large consumption market. The common aroma types of table grapes included strawberry aroma, rose aroma and non-aroma types (Mu et al. 2019), among which strawberry aroma type (45.31%) was the most popular one. For aroma degree, the participants preferred light aroma (47.13%). Among all texture types, soft flesh and juicy (55.23%) was the most popular type for the participants. Among taste types, the participants preferred sweet (89.98%) grapes. With the development of breeding technology, people are able to have seedlessness of most grape varieties (Akkurt et al., 2019). Consistent with the global table grape consumer preferences, most Chinese consumers preferred seedless grapes (73.68%). Skin powder is a layer of powdery substance attached to the skin of grapes; it protects the berries during growth period, reduces diseases and pests (Fan et al., 1999; Yin et al., 2011), and has no adverse reactions on human consumption. However, many consumers believe that skin powder is a pesticide residue, which leads to a few people liking grapes’ skin powder. Most consumers preferred thin skin powder (38.60%). For the peeling degree, most participants preferred grapes that were peeled easily (64.55%).

Prediction models of consumers’ intrinsic attribute preferences for table grapes

This paper used χ² analysis to explore the influence of consumers’ sociodemographic characteristics, purchasing behavior, and perceived value factors for consumer preferences. Then we selected the factors related to the attribute preferences of table grapes as independent variables and grape attributes as dependent variables, and used binary logistic regression algorithm to construct and test the prediction models of consumers’ intrinsic attribute preferences for table grapes.

Factors influencing attribute preference of table grapes

This paper used χ² analysis of contingency table to explore the influence and correlation of consumers’ sociodemographic characteristics, purchasing behavior, and perceived value factors on the attribute preferences of table grapes to select independent variables of prediction models. We collated the χ² analysis results of all influencing factors and the attributes of table grapes as shown in Table 4. The χ² values marked with superscript asterisks^* indicated that the original hypothesis was rejected (p < 0.05), which means that this influencing factor had a significant relationship with the corresponding attribute of table grapes.

It is observed in Table 4 that the age factor is very important in sociodemographic characteristics and has a significant impact on all 10 grape attributes. Among the purchasing behavior factors, the annual consumption was related to the other nine grape attributes, except skin powder. Among the perceived value factors, the texture importance factor was related to the other nine grape attributes, except aroma type, and the freshness factor was related to the other nine grape attributes, except the aroma degree. Factors such as aroma type importance, preference for grapes, packaging, quality, and variety were related to eight grape attributes. The mentioned 10 factors were proved as important influencing factors for consumers’ purchasing behavior (Chu et al., 2020; Kleih and Sparke, 2021; Marques et al., 2021; Siegrist et al., 2013; Szczesniak, 2002; Taylor et al., 2019; Uribe et al., 2020; Verbeke, 2015; Wang et al., 2021a).

Construction of prediction models

In this paper, binary logistic regression of SPSS 27.0 was used to construct the prediction models of consumers’ intrinsic attribute preferences for table grapes in combination with the previous selection results of influencing factors. The complete parameter estimation results of each attribute are summarized in Table A1 given in Appendix. Parameter estimation results reflected the fitting degree of the model, and effectiveness of the influencing factors. Significance value <0.05 demonstrated that the influencing factor had a significant impact on the attribute preference of table grapes. The influence results of the 10 most important factors of table grape attribute preferences are summarized in Table 5.

It is observed in Table 5 that compared to elderly consumers, young consumers had higher acceptance of grapes with rich aroma, crunchy texture, sour with sweetness, seedlessness, thin skin powder, and peeling difficulty. They preferred grapes with diverse sizes and shapes, which indicated that young people pursued grape quality, gave attention to visual effects, and were more inclusive regarding different grapes varieties. This type of consumer group had a strong ability to accept new grape varieties.

Consumers who often ate grapes had a higher acceptance of difficult-to-peel grapes, and they had more fixed preferences for the shape and color of grapes. Therefore, traditional grape varieties, such as Summer Black, Sunshine Rose, Red Earth, or their preferred varieties, could be recommended for such a consumer group.

Aroma-conscious consumers obviously preferred grapes with rose aroma, rich aroma, and soft flesh and juicy varieties. Texture-conscious consumers preferred grapes with light aroma, crunchy texture, thin skin powder, and easy-to-peel varieties; their preference for the shape of grapes was more diversified, which indicated that such consumers gave more attention to the taste of grapes. Consumers with a higher liking for grapes preferred grapes with rich aroma and sweet taste, and they had diversified requirements for the size, shape, and color of grapes; therefore, some new or unpopular grape varieties were favored for this type of consumer group.

Freshness-conscious consumers preferred easy-to-peel grapes. Packaging-conscious consumers preferred difficult-to-peel grape varieties. Easy-to-peel grapes often had a soft taste, while difficult-to-peel grapes often had a crunchy taste.

Consumers who valued packaging were potentially motivated to purchase grapes as gifts. While using grapes as gifts, such consumers usually chose difficult-to-peal grape varieties with crunchy taste. Therefore, consumers who valued packaging preferred difficult-to-peel grapes. Quality-conscious consumers preferred grapes with thin skin powder. This was because skin powder could lead to a bad eating experience, and some consumers felt that skin powder was like a dust, and even high-quality consumers did not accept grapes with skin powder.

Variety-conscious consumers preferred difficult-to-peel grapes, and their preference for shape of grapes was relatively simple. Brand-conscious consumers preferred grapes with light aroma, sweet and sour taste, and their preference for grape color was relatively simple. This indicated that consumers who valued grape sources had fixed preferences, and it was not suitable to recommend new grape varieties to them.

After constructing the prediction models, we compared the prediction results of each model to the actual situation, and tested the prediction performance of all models. The prediction results for each attribute are shown in Table 6. It is observed in the table that the binary logistic regression application performed well in the preference prediction of intrinsic attributes of table grapes, and the prediction results for taste, seedless degree, and peeling degree of table grapes were excellent, with more than 80% prediction accuracy. Although the prediction results for shape diversity, color diversity and aroma type of table grapes were low, they were very close to 60%.

Based on χ²-test results, this paper used binary logistic regression to predict intrinsic attribute preferences of Chinese table grape consumers. The probability of consumer preferences for each attribute value of table grapes was obtained by using the models constructed in this paper, which not only provided references for table grape breeders and marketers in China’s table grape consumption market but also had certain references significant to the prediction of consumers’ attribute preferences of other fruits.

Conclusion

Based on the results of our national survey on table grape consumers, this paper established an in-depth discussion on the attribute preferences of Chinese consumers of table grapes. Chinese table grape consumers tended to choose medium berries, which could be related to the influence of the doctrine of mean on Chinese people (Cui et al., 2022; Feng et al., 2011; Zhou et al., 2015). Too large or too small grapes were not very popular with Chinese consumers. Spherical and oval grapes were more common in the market, so most consumers preferred these two shapes, although many consumers also preferred elongated finger-shape and cylindrical grapes. Moreover, many consumers had no knowledge about these two shapes of grapes and buy these shapes of grapes out of curiosity so they had a great potential in the market. As far as the color of table grapes is concerned, the grapes with lighter colors were not of much concern for consumers. The common green, purple-red, and purple-black grapes in the market were more popular with consumers. In terms of aroma of table grapes, light aroma and rose aroma were more popular with consumers. In terms of taste of table grapes, consumers preferred soft flesh and juicy, sweet taste, seedlessness, thin skin powder, and easy-to-peel grape varieties, which were more palatable and had great development space.

In this study, we constructed prediction models of consumers’ attribute preferences for table grapes in China after studying the influencing factors of intrinsic attribute preferences of consumers. First, we used analysis to screen the factors that affected consumer preferences for different attributes. Our results showed that age was the most important factor, which had a significant impact on all 10 grape attributes. Other sociodemographic characteristics had different degrees of influence on these attribute preferences. This was because consumers of different age groups had different dietary habits, which greatly affected consumer preferences for intrinsic attributes, such as texture, taste, and seedless degree.

Among the purchasing behavior factors, annual consumption was more important, which was related to other nine grape attributes, except skin powder. This showed that the consumption of table grapes greatly affected consumers’ taste and preference for grapes. In other words, the higher the consumption of table grapes, the more obvious the preference for each attribute of grapes.

Perceived value factors mainly referred to the degree of consumers’ attention to various attributes of table grapes. Different consumers valued different factors while purchasing grapes. Therefore, perceived value factors were important to affect consumer preferences. Among them, consumers’ attention to texture, freshness, brand, aroma type, preference for grapes, packaging, quality and variety were relatively important factors. According to these sociodemographic characteristics, purchasing behavior, and perceived value factors, this paper used binary logistic regression to construct prediction models. Most models predicted consumer preferences for some intrinsic attributes of table grapes with a high accuracy of 80%, which was used to predict consumer preferences for taste, seedless degree, and peeling degree of table grapes. This indicated that we easily and accurately determined consumer preferences for the aforementioned grape attributes, and recommended varieties that satisfied their preferences based on the predicted results of these attributes. However, the predictive results of logistic regression for shape diversity, color diversity, and aroma type attributes were not very ideal. On the one hand, it indicated that these attributes were difficult to be predicted accurately, and on the other hand, it indicated that these attributes were not as important as taste, seedless degree, and peeling degree of table grapes when recommending varieties to consumers.

In the future work, our approach may need the following improvements:

Finally, our question design only focused on the attributes of table grapes, not being specific to certain varieties. There could be deviations between survey results and actual preferences, so designing of the questionnaire can be optimized.