E-Commerce Revolution: How the Pandemic Reshaped the US Consumer Shopping Habits: A PACF and ARIMA Approach

Abstract

Accelerated digital transformations and the evolution of consumer behavior in recent years underscore the need for a systemic perspective in marketing analytics to better comprehend the complex interplay between technology, data, and the profound changes triggered by global events, such as the COVID-19 pandemic. The COVID-19 pandemic has catalyzed a massive shift toward digitalization and transformed e-commerce from an option to a necessity for both businesses and consumers. This paper analyzes the total store and non-store sales, as well as total e-commerce sales, of the US retail trade across six main business categories and nine subcategories from the first quarter of 2018 to the first quarter of 2024. The data was divided into three time spans, corresponding to pre-, during, and post-COVID-19 pandemic periods, to examine the changing behavior of US consumers over time for different business categories. The statistical and econometric methods employed are the partial autocorrelation function (PACF), autocorrelation function, autoregressive integrated moving average model, inferential statistics, and regression model. The results indicate that the pandemic significantly increased non-store retailer sales compared to the pre-pandemic period, underscoring the importance of e-commerce. When physical stores reopened, e-commerce sales did not decline to pre-pandemic levels. The PACF analysis showed seasonality and lagged correlations. Thus, the pandemic-induced buying behaviors of US consumers continue to influence current sales patterns. The pandemic was more than just a temporary disruption, which permanently changed the retail sector. Retailers that quickly adapted to online models gained a competitive edge, whereas US consumers became accustomed to the convenience and flexibility of e-commerce. The behavior of US consumers adapted not only in response to immediate needs during the pandemic but also led to longer-term shifts in spending patterns, with each category reacting uniquely based on product type and perceived necessity. The analysis of how the COVID-19 pandemic transformed consumer behavior in the US reveals several important implications for both consumers and trade policymakers. First, the long-lasting and structural shift toward e-commerce is confirmed, representing a fundamental change in the dynamics of demand and supply. For consumers, the convenience, flexibility, and accessibility of digital channels have moved beyond mere situational advantages to become a behavioral norm. This shift has empowered consumers by giving them greater access to price comparisons, more diverse options, and increased informational transparency. Additionally, the data shows the emergence of hybrid consumption models: essential goods are mainly purchased online, while purchases of branded clothing, electronics, furniture, luxury items, and similar products continue to favor the traditional retail experience.

1. Introduction

Starting with the 1970s when large companies implemented e-commerce for private communication and financial transactions, it evolved through four stages until today due to the development of at least six factor categories such as technology (Internet, mobile phones, digital payments, logistics, etc.), economic (globalization, competition, income, efficiency, etc.), social and cultural (consumer behavior, social media, etc.), legal (regulations by World Trade Organization, each country’s legislation, etc.), demographic (young people are adopting more easily to new technologies), infrastructure (data centers, cloud services, warehouse facilities, etc.), etc. [1,2,3,4].

The COVID-19 pandemic had a positive impact on e-commerce by increasing sales for many retail companies, including Amazon, eBay, Rakuten, Walmart, AliExpress, and Etsy. In the US, in the first part of the pandemic, some products recorded boosted sales, for instance disposable gloves (670%), bread machines (652%), cough and cold medicines (535%), soups (397%), dried grains and rice (386%), packaged foods (377%), fruit cups (326%), weight training (307%), milk and cream (279%), and dishwashing supplies (275%), whereas other products registered important drops, namely luggage and suitcases (−77%), briefcases (−77%), cameras (−64%), men’s swimwear (−64%), bridal clothing (−63%), men’s formal wear (−62%), women’s swimwear (−59%), rash guards (−59%), boy’s athletic shoes (−59%), and gym bags (−57%) [5,6]. As a consequence, consumer behavior moved to online shopping, as it provided a safer alternative during the pandemic. At the same time, retailers invested in digital marketing strategies to retain customers by highlighting the advantages of online shopping [7].

The research by Inoue and Hashimoto [8] outlines that in Japan, the pandemic generated changes in both consumers’ attitudes towards product pricing, variety, and delivery services, as well as their purchasing mechanisms. Some consumer categories increased their purchases of electronics and started using the package drop delivery system. Marketing strategies in the context of e-commerce are shaped by changes in consumer behavior in relation to social challenges. This aspect is emphasized by digitalization and is noticeable in electronic logistics and trade platforms, as well as in public–private partnerships for supporting digitalization in the sphere of goods circulation, an aspect studied for the Russian Federation in the paper by Panasenko et al. [9]. The authors mention transparency, data protection, technological solutions, and the implementation of modern strategies for strengthening relationships between economic actors as marketing strategies. In the maritime transport services sector, the digitalization of processes is combined with ambidextrous innovation, which involves balancing exploration and exploitation. The effect is to increase the resilience and competitive advantage of companies. This involves marketing strategies centered on technological innovation, which guarantees a dynamic response to post-pandemic changes [10,11]. Consumer behavior online has been transformed by the pandemic context, according to Gu et al. [12]. This records marketing strategies that target factors such as decision speed and purchase consistency. Thus, the emphasis is on personalizing e-marketing tools, while simultaneously optimizing online presence and building a loyal customer base through strengthened online relationships during periods of self-isolation. The research by Jadhav et al. [13] is conducted in the nutraceutical market, and the marketing strategies are focused on health awareness, consumer education, habit change, product innovation, and the regulation of uneven manifestations. García-Roldán et al. [14] also highlight the contribution of digital influencers to online communities in shaping pro-environmental attitudes through marketing strategies that leverage social influence to promote organic products. The fashion industry is examined in the paper by Liu et al. [15], in which pricing strategies are adapted to the utility of fashion and consumer behavior. In parallel, medical esthetic services, as described by Li et al. [16], utilize Magic Mirror to increase purchase intention by reducing perceived risk and increasing value, indicating marketing strategies based on predictive and personalized technologies. In summary, current strategic success depends on digital integration, understanding consumer behavior and needs, integrating modern technologies, and adaptive innovation.

To achieve the aims of this study, the following research questions are addressed through econometric and statistical methods, including partial autocorrelation function (PACF), autocorrelation function (ACF), autoregressive integrated moving average (ARIMA) modeling, regression analysis, and non-parametric tests:

• How did US consumer shopping habits evolve across pre-, during, and post-COVID-19 periods, as revealed by time series analyses of retail trade sales?
• How do PACF and ARIMA models capture seasonality, stationarity, and lagged correlations in e-commerce categories, and what insights do they provide for forecasting future trends?
• Are there statistically significant differences in e-commerce sales between pandemic phases, as identified through inferential tests, and how do these differences vary across retail categories and subcategories?

The paper contributions are linked to both theoretical and practical frameworks, as follows:

• From a theoretical standpoint, our research demonstrates how and in what direction the COVID-19 pandemic reshaped US consumer shopping habits, and therefore, all commerce theory must be rewritten according to the new trends and behaviors.
• From a practical perspective, based on secondary data, quarterly registrations, about traditional and e-commerce in the US by applying complex statistical and econometrical methods, we demonstrate that the PACF, ARIMA model (combined with other specific methods as inferential statistics, regression model, etc.) and some specific graphical representations could particularly emphasize the changing behavior over time for a different type of commerce under the influence of particular time and conditions.

The paper is structured as follows. The literature review is presented in Section 2. Section 3 describes the dataset and the methods used for analyzing data. The results are presented in Section 4. Section 5 focuses on the discussion and limitations, whereas Section 6 presents the conclusions and future work of the paper.

2. Literature Review

2.1. E-Commerce of Furniture and Home Furnishings

Customization is a trend in the furniture industry that allows consumers to adapt the products to their needs and preferences. Bumgardner and Nicholls [17] state that customization increases customer satisfaction and direct engagement between manufacturers and consumers, bypassing traditional retail channels. This approach has led to the development of innovative products, such as character-marked furniture. E-commerce has the customization feature advantage because consumers can visualize and modify products before buying [18]. The research by Le et al. [19] presents LumaAI, a system that uses augmented reality (AR) to ensure furniture conception and personalization according to consumers’ needs. It facilitates users in creating and interacting with 3D furniture models on consumers’ mobile devices, providing a comprehensive experience that includes quality assessment of these products. The continuous development of online platforms presents an important opportunity for furniture companies to expand into new markets. The research by Ojala et al. [20] outlines that, nowadays, it is easier to reach customers from different continents through e-commerce. The firms must consider local preferences and employ targeted marketing strategies to increase their market share. New technologies such as artificial intelligence (AI) and the Internet of Things (IoT) [21] are also used by furniture companies to meet customer demands for product customization and mass personalization, thereby ensuring economic efficiency [22].

The rise in second-hand furniture markets has gained popularity among consumers who value cost-effectiveness and environmental friendliness. The paper by Hristova [23] demonstrates that cost savings are an important motivation for consumers in this segment. This change in consumer behavior presents both issues and opportunities for traditional furniture companies as they compete with second-hand online platforms. Firms use social media platforms such as TikTok, Instagram, X, and Facebook to present their products to consumers. As noted by Petrache et al. [24], companies employ digital marketing strategies to interact with online consumers and influence their purchasing decisions. Yu et al. [25] analyzed the factors that affect consumers’ online furniture buying. The factors were grouped into three categories: personal, product, and service, and each category included 4, 6, and 6 subfactors, respectively. The results showed that the product category is the most important, followed by the personal and service categories. The top 3 most weighted subfactors of all 16 are revenue, price, and quality.

2.2. E-Commerce of Electronics and Appliances

Więcek-Janka et al. [26] examine how demographic factors, such as social status, gender, and age, influence Polish consumers’ purchase decisions in the online purchase process of household appliances. The results of the study showed that social status influences preferences for product card content across all product categories, gender considerably impacts content preferences for refrigerator product cards, social status affects the prioritization of traits like price, durability, reliability, and technical novelty during the purchasing process, and age correlates with emphasis on price, durability, and technical novelty [27]. Another study [28] focuses on stating the factors that affect consumers’ purchase intentions and behavior regarding home appliances. Thus, electronic word-of-mouth (eWOM) and social media marketing influence purchase intentions, whereas perceived risk and online convenience negatively impact acquisition intentions.

Guan and Lin [29] show that consumers consider energy savings to be an important factor in online purchasing of home appliances, and the level of attention consumers give to energy savings varies depending on their location and the type of home appliance. By analyzing over 29,000 consumers’ online reviews of refrigerators, Luo et al. [30] find that consumers’ purchase decisions are based on several features, including function, price, appearance, post-sales service, brand, quality, and volume. Svobodová and Rajchlová [31] examine e-commerce platforms that sell electronics to analyze their strategic approaches concerning customer shopping behavior. The findings show that most e-commerce companies use a balanced e-strategy that does not align with the rapid growth trend of the e-commerce sector. This study found fifteen factors that influence customers’ decisions when buying electronics through e-commerce. Thus, e-commerce platforms take into account these behavioral factors in creating their marketing strategies. The analysis reveals a disparity between the balanced e-strategy adopted by companies and their financial strategies, an issue that needs to be addressed and improved.

Corbos et al. [32] investigated the impact of online reviews on consumers’ purchase intentions for large household appliances, analyzing customer trust, review volume, and the COVID-19 pandemic. Hence, some customers trust online reviews, and the volume of online reviews has a positive impact on purchase intentions and companies’ sales performance. There is no significant relationship between the pandemic and consumers’ intentions to purchase large appliances, outlining the consistency of consumer decisions during the COVID-19 pandemic. Other authors [33] analyzed the factors and dimensions that shape consumers’ trust in e-commerce platforms in Romania, particularly when purchasing electronic and household appliances. Previous purchasing experiences and the actual quality of appliances influence consumer confidence and their final buying decision. To ensure consumer trust in e-commerce transactions, particularly in online payments, companies should focus on consumer data protection and maintain up-to-date information. Considering the numerous platforms available, a dissatisfied consumer who cannot find the necessary information or has limited accessibility to some product information is unlikely to return. To maintain existing users and attract new ones, the e-commerce platform must be continuously optimized according to consumer feedback.

2.3. E-Commerce of Building Materials and Garden Equipment

The e-commerce of building materials and garden equipment is experiencing an increasing trend as companies strive to adapt to consumers’ needs. Debadutta and Raja [34] researched the implementation of e-commerce to improve building materials procurement in South Africa for sustainable building. Even though e-commerce offers numerous advantages for buyers, including cost reduction and improved environmental protection, they face risks such as limited awareness, reluctance to adopt change, significant installation and operational costs, and online fraud. The authors recommend that an effective implementation of these findings should provide a viable strategy for tackling the challenges of material procurement in sustainable building construction.

When the COVID-19 pandemic struck in 2020 [35], building materials and cement retailers had to adapt quickly to the challenges posed by COVID-19 restrictions to maintain operations, meet consumer demands, and ensure safety. Therefore, the e-commerce of these products developed rapidly due to numerous advantages for consumers, such as 24 h accessibility to features and diverse product images, as well as cost transparency. Moreover, e-commerce has stimulated retailers to innovate and increase efficiency in construction supply chains [36]. Mulyawan et al. [37] developed an e-commerce website application to provide product recommendations for building materials using the FP-Growth algorithm. The algorithm analyzes stored transaction records by analyzing data patterns that appear most frequently. The application leverages the FP-Tree structure alongside the FP-Growth algorithm, which requires only two database scans. The first scan processes historical sales transaction data, while the second scan traces the FP tree to identify association patterns. By calculating item set linkages that exceed a predefined minimum support threshold, the system generates tailored product recommendations, which are then displayed on the e-commerce platform. Another approach is to use image processing-based methods that improve user experience, improve product searchability, and optimize business operations [38]. Considering the high number of people interested in growing and caring for plants in the Philippines, the paper by Blancaflor et al. [39] proposes a business-to-consumer (B2C) e-commerce platform that provides information about plants, teaches users how to care for them, and features a virtual marketplace for buying and selling plants. Information on plant diseases and their treatments can be added to provide a comprehensive platform for consumers, combining the e-commerce trade function with educational resources.

2.4. E-Commerce of Clothing and Clothing Accessories

The development of e-commerce for fashion products is a result of companies’ compliance with consumers’ needs regarding convenient online transactions, product return policies, and payment methods, among other factors [40]. E-commerce platforms tend to expand by including fashion products for new consumer categories, such as children, and by providing personalized shopping experiences [41]. Live-streaming e-commerce is an important strategy employed by fashion companies, in which products are presented in real-time, allowing consumers to interact directly with the designer or an influencer paid by the fashion company. This strategy focuses on generating a sense of urgency and excitement around purchases [42]. Companies continuously adapt their marketing strategies to maintain consumer interest and attract potential customers by creating a captivating shopping experience that replicates the in-store shopping experience.

New technologies assist consumers in the buying process by recommending products and improving the shopping experience. For instance, deep learning (DL) models assist consumers in finding fashion products that match their search criteria [43]. No less important is the cultural environment in which fashion e-commerce operates. Fashion companies adapt their products to cultural differences, as they target a global market [44]. Marketing strategies must be correlated with the consumers’ behavior for a specific region. For instance, the marketing strategy employed in the United Kingdom may not yield the same results in China, and companies must adapt their tools to account for cultural factors. The research by Giasi et al. [45] demonstrates that consumers’ online buying experience, rapid feedback from companies to consumer requests, and security measures all affect consumers’ satisfaction and loyalty in purchasing fashion products.

Social media has an important influence on consumer perceptions in the fashion industry. TikTok, Instagram, X, and Facebook are the main social platforms that the fashion industry uses directly or indirectly to promote its products and build brand loyalty. Social commerce (fusion of social media with e-commerce) is a tool that enables fashion companies to employ user-generated content and influencer partnerships to increase their visibility and credibility [46]. This strategy is targeted at younger consumers, who typically make purchasing decisions based on social media interactions [47,48]. E-commerce fashion companies can implement 3D body scanning, which is a step-forward technology that focuses on solving fit-related issues, the main frequent cause of consumer dissatisfaction [49]. 3D body scanning provides precise size guidance for clothing, which improves consumers’ online shopping experiences and decreases return rates, a main concern for most e-commerce firms. As consumers become more familiar with personalized purchasing, the demand for new technologies is expected to rise. Brand loyalty among millennial consumers has both advantages and disadvantages for e-commerce fashion companies, as they must cater to specific preferences and behaviors [50]. Marketing strategies used to build brand loyalty take into account that millennials prefer companies that align with their values and lifestyle choices. This requires interactive and customer-focused approaches that favor consumer input and feedback throughout the product development [51].

2.5. E-Commerce of General Merchandise

Triani et al. [52] state that e-commerce changed consumers’ behavior with retail companies, mainly for non-perishable products. Indonesia recorded the tenth-highest e-commerce sales volume worldwide in 2021, with a 23% increase from the previous year. This growth reflects a substantial shift, as e-commerce provides consumers with easy access to a global platform and new opportunities for companies to expand their market share. Zhao [53] highlights the challenges that Chinese general merchandise stores have faced in recent years in their e-commerce development. To maintain competitiveness and ensure growth in a rapidly changing market, general merchandise stores must employ new management practices and integrate their cultural heritage with modern internet tools.

Lin et al. [54] propose an algorithm that integrates four pre-implemented machine learning (ML) algorithms for location selection in general merchandise stores based on rational and scientific criteria to ensure efficiency. The new algorithm was successfully tested in Hangzhou, showing its ability to evaluate the suitability of alternative locations and identify optimal areas for new stores [55]. The paper by Ye et al. [56] outlines that companies that trade online should focus on predicting gross merchandise value. The market circumstances and consumers’ preferences are the fundamentals for this proposed indicator that measures the level of sales volumes [57].

2.6. E-Commerce of Food and Beverages

Duffy et al. [58] outline that demographic factors contribute to the transition from on-site to online grocery stores. For example, persons with higher incomes and education levels, especially women with children, prefer online grocery shopping more than other groups studied in the survey. In contrast, lower-income individuals were forced to use online grocery stores due to the pandemic restrictions. This shows e-commerce’s potential to ensure food access [59]. The COVID-19 pandemic indirectly imposed the development of online food buying due to the mandatory obligation to maintain social distancing. For instance, due to both current necessity and the fear of missing out in the medium term, the foods sold on online platforms by Italian retailers or independent producers recorded a high increase [60]. Similar trends have been spotted worldwide, mostly during lockdowns. The pandemic has altered buying habits and encouraged the adoption of online shopping across all product categories, including food and beverages [61]. In the food and beverage e-commerce sector, product quality and attractiveness are important for consumer satisfaction and repeat buying behavior. Studies indicate that businesses should create attractive offerings that align with consumers’ hedonic spending values, taking into account that consumer preferences are influenced by perceived quality and specificity [62,63]. Like any other product, the discount policy in food ordering applications has led to an increase in consumer satisfaction, highlighting the importance of pricing strategies [64]. Halal certification has become increasingly important in building consumer trust and enhancing the perceived quality of food and beverages in the sector. For small and medium-sized enterprises, holding a halal certification ensures an increase in consumer confidence and sales of online food purchases, as consumers perceive it as a low-risk option when buying online [65]. This applies in regions where halal food is deeply ingrained in the culture, as it meets consumer expectations and regulatory requirements.

Selling food and beverage products through e-commerce platforms has led firms to adjust their marketing strategies. Digital marketing has become an important tool for interacting with consumers, particularly during the pandemic when traditional marketing channels were unavailable [66]. Digital marketing has the advantages of reaching a wide consumer group at low costs and penetrating new, emerging food and beverage markets [67]. Nowadays, food and beverage companies use social media platforms to interact with consumers, where product experiences and recommendations are mutually shared. Thus, these platforms contribute to the creation of a community around one or more products, benefiting consumers, and are also used by companies to influence consumers’ purchasing decisions [68]. The impact of e-commerce on dietary quality has also been explored. For instance, in China, in some regions, where access to fresh food is limited due to different factors, e-commerce improves dietary quality by offering consumers with a wider variety of food options, thereby promoting healthier eating habits [69]. Qi et al. [70] observe that consumers increasingly use online platforms to purchase organic food products, which are viewed as healthier alternatives to conventional options. This trend represents a more complex behavior that some consumers aspire to emulate as part of a lifestyle that has long existed in society and is now gaining new followers due to increased awareness of the connection between nutrition and health.

2.7. E-Commerce of Health and Personal Care

Harris-Lagoudakis [71] states that online buying encourages healthier consumer choices. In health and personal care, having easy access to information about product ingredients and benefits guides purchasing decisions toward more nutritious options. Shopping from home saves time and allows consumers to make more informed choices without the pressure often present in physical retail settings [72]. The study by Zilong et al. [73] on the influence of e-commerce live broadcasts on consumers’ buying intentions for eye health products shows that consumers’ acquisition intentions were positively influenced by the perceived value and simplicity of use of the products, as well as the expertise and genuineness of the anchor’s language, whereas the anchor’s linguistic charm did not impact purchase intentions. The consumer’s trust is determined by the link between perceived convenience and buying intention, as well as partially by the relationship between perceived usefulness and purchase intention.

Another research [74] conducted in Malaysia outlines the relationships between TikTok’s influence, price fairness perceptions, health awareness, and purchase intentions for healthcare products. This study contributes to understanding consumer behavior in the context of social media-driven healthcare marketing. It offers valuable insights for leveraging TikTok as an effective platform for promoting healthcare products. Nguyen and Duong [75] analyzed the impact of influencers on the purchasing behavior of Vietnamese Generation Z consumers concerning cosmetics bought through e-commerce. The findings underlined that influencers’ credibility, appeal, and competence positively affect consumers. Sarjono et al. [76] found that eWOM, e-service quality (eSQ), and e-satisfaction (eS) impact consumers’ purchase intentions for cosmetics, and eWOM alters buying intention indirectly through eS. In contrast, eSQ also indirectly affects acquisition intention led by eS. Incorporating AI and AR technologies into personalized recommendations for cosmetics e-commerce platforms increases customer interest and purchase intentions. The main factors that shape the consumers’ attitudes and institutions are the buyer’s advantages provided by these technologies, such as accessibility, relevance, credibility, and innovation [77].

2.8. E-Commerce of Sporting Goods, Musical Instruments, and Books

Sporting goods include a wide range of products. Liu et al. [78] explore the impact of confirmatory psychology on consumers’ intention to purchase sports goods from an e-commerce platform. The results show that brand quality has a high influence on the intention to buy: consumers focus mainly on foreign brands; the higher the quality of the brand, the longer the product is used by consumers; it is a positive correlation between brand quality and purchase intention; the consumers with high confirmatory psychology trend to choose foreign sporting products to show consumer enthusiasm for consumption. Other researchers [79] analyzed the online shopping patterns of the United Kingdom’s consumers for sporting goods during and after the COVID-19 pandemic. Thus, the findings outline an increase in consumers’ online searches for fitness apps, home gyms, and fitness equipment during the pandemic period, indicating that people were proactively seeking ways to sustain their health and fitness at home despite the challenging circumstances.

Concerning musical instruments, Shaopeng et al. [80] investigated the factors that influence consumers’ decisions to purchase large musical instruments from e-commerce platforms. The results show that consumers’ perceived trust is independent of the perceived risk (due to transport details, after-sales service, etc.), and consumers’ behavior control has a moderate influence on decisions, as evidenced by the lower repurchase rate of these products. Xu [81] designed an e-commerce platform for trading second-hand musical instruments, allowing users to list their instruments for sale and purchase products from others. At the same time, the administrator oversees the management of users, products, and website information. Reviews from e-commerce platforms affect online shopping behavior. Kaur and Singh [82] note that consumers frequently read online reviews to inform their purchasing decisions, particularly for books. Positive reviews can build consumer trust and encourage purchases, while negative feedback can discourage potential buyers. The authors analyzed the number of reviews, star ratings of reviews, and sales of two book categories, and they concluded that the average star ratings influence consumers’ book choices and book sales. Wei et al. [83] propose a DL model for predicting click-to-conversion rates on book e-commerce platforms. The model integrates user attributes, product characteristics, and platform activity data to accurately forecast the likelihood of users converting their clicks on books into purchases. The model was successfully tested on a real e-commerce website.

2.9. E-Commerce of Miscellaneous Goods and Gasoline Stations

Miscellaneous goods include a wide range of products such as medical equipment, toys, jewelry, etc. Tomas et al. [84] evaluate the reliability of star ratings compared to sentiment analysis and identify the most suitable classification algorithm for analyzing customer online reviews for acquiring medical personal protective equipment. Additionally, the authors aim to determine the most effective classifier model based on performance. The results reveal a high similarity between star ratings and annotated reviews regarding sentiment and polarity classification per review [85]. COVID-19 has severely affected patients with chronic diseases’ access to essential medicines of utmost importance. During the pandemic, e-commerce platforms were used to buy needed drugs. The research by Han and Han [86] concluded that attitude, perceived social norms, and perceived self-efficacy appreciably influenced patients’ buying intentions. Of these factors, attitude had the greatest impact on acquisition intentions. Secondly, price value affected attitude, perceived social norms, perceived self-efficacy, and buying intention, with its strongest impact observed on attitude. Finally, price value greatly indirectly influences acquisition intentions through attitude, perceived social norms, and perceived self-efficacy.

Concerning toys, Halli et al. [87] analyzed caregivers’ perspectives on toy sustainability factors and studied their online purchasing behavior. They found that sustainable criteria such as challenging, enjoyable, and updating according to child age account for 72% of the purchasing decision, online ratings, price, and durability account for 23%, whereas esthetics and second-hand potential account for 5%. Moreover, the consumers struggled to grasp the material, its potential risks, or the toy’s sustainability ratings while reviewing the product on the e-commerce platform. The research by Hung and Chiu [88] investigated cross-cultural customer behavior toward the online visual merchandising of independently crafted designer toys. Firstly, designer toy photography was categorized into three elements: focal character, background, and dioramas. Customer behavior was evaluated using the AIDMA model, and the results demonstrate that adding backgrounds and dioramas increases customer preference for designer toys. Furthermore, cultural differences emerged, with East Asian customers showing a stronger preference for displays with backgrounds than Western customers.

Chen and Wang [89] surveyed the factors that affect consumers’ willingness to buy jewelry products through TikTok live streaming. Thus, consumers’ purchase intentions depend on live-stream features such as exposure, communication, and genuineness. However, perceived trust and perceived usefulness act as intermediaries between the TikTok live-stream features and consumers’ purchase intentions. Webrooming is a common practice among consumers. Zarinkamar et al. [90] studied the customers’ webrooming (WR) behavior for gold and jewelry. The findings show that the advantages of online searching have a positive and considerable influence on attitudes towards WR, behavioral control, and mental norms, which have a positive and significant impact on the intention to engage in WR behavior. In contrast, perceived online risk and anticipated regret have a negative and substantial effect on the behavioral intention to engage in WR. The oil and gas companies had to adapt to the e-commerce development and gasoline stations from many countries implemented fully automated stations, contactless refueling, mobile applications for locating stations, mobile payments, in-app convenience store purchases, electric vehicles (EVs) charging integration, etc. [91,92,93].

3. Materials and Methods

The dataset used for analysis is provided by the United States Census Bureau [94] and consists of retail trade sales, expressed in millions of USD, between the 1st quarter of 2018 and the 1st quarter of 2024, which include pre-, during, and post-pandemic periods. No missing values were identified for the analyzed time, except the suppressed data mentioned by the United States Census Bureau with the main explanation for this: is that the estimate does not meet publication standards because of high sampling variability (coefficient of variation is greater than 30%), poor response quality (total quantity response rate is less than 50%), or other concerns about the estimate’s quality. These missing values and suppressions were for two subcategories of FBME (FHF—6 quarters within 2020–2021, EA—5 quarters within 2018–2019, and 1 for FB). This technical problem was resolved by stationarizing the time series, and all the time series components (lags, unstandardized residuals, errors, seasonal factors, etc.) were replaced using linear interpolation, the method used by default by IBM SPSS Statistics v30 software. The structure of the dataset comprises six types of business categories:

• Total retail trade (TRT);
• Motor vehicle and parts (MVP);
• Furniture, building materials, and electronics (FBME);
• Clothing and general merchandise (CGM);
• All other (AO);
• Nonstore retailers (NSR).

Each category includes two records with total store and non-store sales (TSNSS) and total e-commerce (TEC),

Additionally, the subsequent categories from the dataset contain TEC for the component subcategories:

• FBME:

  ○

  Furniture and home furnishings (FHF);

  ○

  Electronics and appliances (EA);

  ○

  Building materials, garden equipment, and supplies (BMGES);

• CGM:

  ○

  Clothing and clothing accessories (CCA);

  ○

  General merchandise (GM);

• AO:

  ○

  Food and beverage (FB);

  ○

  Health and personal care (HPC);

  ○

  Sporting goods, hobbies, musical instruments, and books (SGHMIB);

  ○

  Miscellaneous including gasoline stations (MGS).

To test the normal distribution of the data, the one-sample Kolmogorov–Smirnov test was applied (

Table 1. Results of one-sample Kolmogorov–Smirnov test for normal distribution.

Null Hypothesis	Sig.a,b	Decision
TSNST of TRT distribution is normal with a mean of 1,547,999 and a standard deviation of 219,447.423.	0.200 c	Retain the null hypothesis.
TEC of TRT distribution is normal with a mean of 208,986 and a standard deviation of 62,670.002.	0.200 c	Retain the null hypothesis.
TSNST of MVP distribution is normal with a mean of 344,722 and a standard deviation of 44,812.723.	0.200 c	Retain the null hypothesis.
TEC of MVP distribution is normal with a mean of 12,331 and a standard deviation of 2557.195.	0.006	Reject the null hypothesis.
TSNST of FBME distribution is normal with a mean of 185,621 and as standard deviation of 26,484.213.	0.200 c	Retain the null hypothesis.
TEC of FBME is normal with a mean of 22,903 and a standard deviation of 6838.045.	0.002	Reject the null hypothesis.
TEC of FHF distribution is normal with a mean of 4007 and a standard deviation of 1034.854.	0.200 c	Retain the null hypothesis.
TEC of EA distribution is normal with a mean of 12,836 and a standard deviation of 2817.257.	0.060	Retain the null hypothesis.
TEC of BMGES distribution is normal with a mean of 7166 and a standard deviation of 2798.002.	0.005	Reject the null hypothesis.
TSNST of CGM distribution is normal with a mean of 297,223 and a standard deviation of 46,831.792.	0.189	Retain the null hypothesis.
TEC of CGM distribution is normal with a mean of 37,184 and a standard deviation of 13,427.570.	0.200 c	Retain the null hypothesis.
TEC of CCA distribution is normal with a mean of 14,946 and a standard deviation of 4387.663.	0.008	Reject the null hypothesis.
TEC of GM distribution is normal with a mean of 22,237 and a standard deviation of 9224.068.	0.200 c	Retain the null hypothesis.
TSNST of AO distribution is normal with a mean of 537,347 and a standard deviation of 67,287.733.	0.006	Reject the null hypothesis.
TEC of AO distribution is normal with a mean of 14,840 and a standard deviation of 4780.193.	<0.001	Reject the null hypothesis.
TEC of FB distribution is normal with a mean of 5801 and a standard deviation of 2705.310.	<0.001	Reject the null hypothesis.
TEC of HPS distribution is normal with a mean of 1988 and a standard deviation of 664.160.	0.200 c	Retain the null hypothesis.
TEC of SGHMIB distribution is normal with a mean of 3000 and a standard deviation of 897.134.	0.068	Retain the null hypothesis.
TEC of MGS distribution is normal with a mean of 4191 and a standard deviation of 685.983.	0.002	Reject the null hypothesis.
TSNST of NSR distribution is normal with a mean of 183,086 and a standard deviation of 48,914.977.	0.200 c	Retain the null hypothesis.
TEC of NSR distribution is normal with a mean of 121,728 and a standard deviation of 36,456.889.	0.200 c	Retain the null hypothesis.

^a The significance level is 0.050; ^b Lilliefors Corrected. Asymptotic significance is displayed; ^c This is a lower bound of the true significance.

), yielding results indicating both normal and non-normal distributions. The non-normal distribution was obtained for the following variables from the study, especially for e-commerce: (i) TEC of MVP, (ii) TEC of FBME, (iii) TEC of BMGES, (iv) TEC of CCA, (v) TSNSS of AO, (vi) TEC of AO, (vii) TEC of FB, and (viii) TEC of MGS. Their significance value is written in italics. Therefore, to compare whether there are statistically significant differences between data from pre-, during, and post-pandemic periods, the Kruskal–Wallis test is applied.

The primary research hypothesis of the paper is that the post-pandemic trend of increasing total e-commerce is correlated with the rise in e-commerce buying behavior during the COVID-19 pandemic. To demonstrate this and according to the above methodological approach, based on time series analysis, we used and analyzed the following:

• PACF charts were employed to determine whether the data from the time series are stationary or not and to formally check for unit roots. The SPSS software does not offer the possibility of directly applying the Dickey–Fuller test, the formal test for establishing the presence and nature of stationarity. However, the PACF is sufficient to determine the nature and presence of non-stationarity. By applying PACF, the standard errors are calculated too. Since data is quarterly, we opted for multiple quarter-type periods (more precisely 16 periods) as the maximum number of lags (9 quarters for pre-pandemic and 7 quarters for during the pandemic and we want to demonstrate that the buying behavior for the post-pandemic is correlated with the e-commerce buying behaviors during the pandemic).
• For non-stationary time series, a transformation of the data based on a differentiating correction was applied (as opposed to a log transformation). For the research data, we opted for differencing, as it proved more effective than the log format in SPSS, because the log scale flattens out the more pronounced patterns. The research data show some seasonality in sales for certain categories [95]. In addition, we use this transformation to build the ARIMA models.
• The Autocorrelations Function (ACF) charts were used to establish the ARIMA type of model for each type of commerce, namely (a) TSNSS and (b) TEC.

Since the data is a time series, the main problem is to establish that these time series are stationary. However, time series data often contradict the classical assumption that each non-stationary variable obeys the key assumption that “each variable is distributed randomly” [96]. This problem of stationarity is addressed through data transformation, both for econometrically applied methods and for graphical representations, with SPSS software providing this technical capability [97]. Additionally, to test for non-stationarity, we applied the PACF charts, which are sufficient for testing non-stationarity and Unit Roots [98]. We also used the ACF charts, together with the PACF charts, whereas ACF charts indicate the “integration-order” [99]. For example, for graphical representation based on sequence charts, we consider the advantage of opting for possible data transformation from time series using natural log transformation or difference in SPSS software [100]. Another important advantage of using sequence charts is that they facilitate a simplistic interpretation of the non-stationarity of time series, with the non-stationarity analysis in SPSS being akin to the Unit Root problem [101,102].

Many empirical studies have illustrated the effectiveness of statistical and econometric methods (such as ARIMA models, Box-Jenkins, neural network methods, etc.) in time series analysis [103,104,105,106]. There are several types of delayed effect models (lag, dynamic): unifactorial, autoregressive, mixed, with distributed lag [107]. From these models, only the autoregressive ones have been practically applied in this paper. Thus, we shall verify if the model is characterized as a mobile average to build an ARIMA model, which makes it possible to make forecasts and perform econometric analysis based on a given stationary time series [108].

To calculate the time series correlation coefficient at “k” lag, we form “n−k” pairs of observations: $\left(x_1,x_{1+k}\right),\left(x_2,x_{2+k}\right)\dots \left(x_{n-k},x_n\right)$. The pairs are arranged on columns applying Equation (1) [109,110].

$$r_k={\displaystyle \frac{{\sum }_{t=1}^{n-k}\left(x_t-\overline{x}\right)\left(x_{t+k}-\overline{x}\right)}{{\sum }_{t=1}^n{\left(x_t-\overline{x}\right)}^2}}$$

(1)

$$\overline{x}={\displaystyle \frac{{\sum }_{t=1}^n{x}_t}{N}}$$

(2)

where

N is the size of the time series;

k represents the lag, expressed by a number of units after which the effect of some impulses appears (a change in its behavior or of a cause).

The ARIMA is an aggregate that includes the repeated impulses with a lag of 1, 2, …, p time intervals, as well as the reactions expressed in average, at accidental deviations (u_t) from the linear evolution, manifested 1, 2, …, q time intervals ago [109]. The general form of the ARIMA model is computed with Equation (3) [111,112,113,114].

$$y_t=\overline{y}+a_0+a_1{y}_{t-1}+a_2{y}_{t-2}+\dots +a_p{y}_{t-p}-b_u{u}_{t-1}-\dots -b_q{u}_{t-q}+u_t$$

(3)

where

y_t is the stationarized economic variable, the trend being excluded from the data;

a, b are the parameters.

To analyze the time series, we use the variation graph of the autocorrelation coefficient of lag k according to k. The scores of the autocorrelation coefficient, along with the interpretation of the correlogram, provide the basis for identifying the ARIMA model. In this research, we follow all the stages of the ARIMA models [99,112,115,116,117,118], from model identification to model estimation and validation [118], finishing with formulating the final model and results interpretation.

Therefore, scatter plot matrices were used because this type of analysis is useful for identifying potential correlations between different sectors of retail trade, which can help understand how movements in one sector relate to movements in another. Identifying the strong positive or negative correlations could indicate that these sectors move together due to common influencing factors (like economic cycles or consumer trends), while weak correlations may indicate independence between sectors. This pair plot provides a high-level view, showing which retail trade sectors might have relationships worth exploring further with statistical analysis.

The heat maps are employed to indicate whether there is a trend of increasing/decreasing e-commerce from pre-pandemic levels through the pandemic and into the post-pandemic recovery phase. The detected patterns would suggest that e-commerce experienced growth during the pandemic (possibly from shifts in consumer behavior) and will accelerate further post-pandemic, reflecting an economic rebound and possible inflationary effects on retail values. This visual helps highlight the overall resilience and growth of retail trade across these significant periods, with especially strong performance observed in the post-pandemic phase.

In this paper, the most important part of the results of this research is based on graphical representations, as these are essential for time series analysis. For statistical and econometric analysis, the SPSS 29.0 (licensed) software was used, along with the GeoDa 1.22.0.10 version and Microsoft Excel software.

4. Results

In this section, the results are presented comparatively for TSNSS and TEC. The descriptive statistics for TSNSS and TEC are shown in

Table 2. Descriptive statistics.

TSNSS/TEC			TRT	MVP	FBME	CGM	AO	NSR
TSNSS	N	Valid	25	25	25	25	25	25
		Missing	0	0	0	0	0	0
	Mean		1,547,999.36	34,4721.88	185,620.52	297,223.28	537,347.32	183,086.36
	Median		1,540,531.00	341,507.00	185,875.00	303,170.00	505,910.00	186,214.00
	Std. Deviation		219,447.423	44,812.723	26,484.213	46,831.792	67,287.733	48,914.977
	Minimum		1,214,805	267,945	138,049	230,762	443,823	114,152
	Maximum		1,892,581	414,552	228,052	393,257	640,567	276,657
TEC	N	Valid	25	25	25	25	25	25
		Missing	0	0	0	0	0	0
	Mean		208,986.12	12,331.12	22,903.24	37,183.60	14,840.24	121,727.92
	Median		224,757.00	12,847.00	25,793.00	38,268.00	16,753.00	129,825.00
	Std. Deviation		62,670.002	2557.195	6838.045	13,427.570	4780.193	36,456.889
	Minimum		111,690	8172	12,727	16,394	7523	66,813
	Maximum		32,2862	15,840	33,469	62,779	21,869	19,2843

for the entire analyzed period (1st quarter of 2018–1st quarter of 2024).

Considering that data for TEC for some types of e-commerce are detailed,

Table 3. Descriptive statistics for specific types of e-commerce.

		FHF	EA	BMGES	CCA	GM	FB	HPS	SGHMIB	MGS
N	Valid	19	20	25	25	25	24	25	25	25
	Missing	6	5	0	0	0	1	0	0	0
Mean		4006.53	12,835.65	7166.40	14,946.28	22,237.32	5800.58	1987.76	3000.04	4191.48
Median		4196.00	12,592.50	8471.00	15,088.00	22,826.00	7225.50	1947.00	3131.00	4476.00
Std. Deviation		1034.854	2817.257	2798.002	4387.663	9224.068	2705.310	664.160	897.134	685.983
Minimum		2478	7385	2748	8489	7905	1417	1042	1703	3152
Maximum		5662	18,622	10,916	23,320	39,469	8868	3429	4601	5082

shows the descriptive statistics for these variables.

4.1. Analysis of E-Commerce Evolution in Pre-, During, and -Post-COVID-19 Pandemic

To analyze and detail the change in the structure of e-commerce for the categories from Table 2, graphical representations were made to emphasize these changes between three periods, pre-, during, and post-COVID-19 pandemic time, in which Q represents the quarter (from 1 to 4) of the year (Figure 1, Figure 2 and Figure 3).

For FBME (Figure 1), during the COVID-19 pandemic, there was an increase in TEC for EA, accompanied by a decrease in BMGES and FHF. Some changes in the post-pandemic period have a similar structure to those during the pandemic for all types of FBME, primarily aimed at increasing the appetite for e-commerce following the pandemic or due to health reasons.

For CGM (Figure 2), a slight upward trend is observed for CCA, from a pre-pandemic mean of 10,853 to 17,473 for the post-pandemic period, negatively correlated with the general trend for GM.

For TEC of AO, the structure of e-commerce recorded changes, especially for FB, with the main increase from an average of 20% to almost half (44%) for the 1st quarter of 2024. The TEC was maintained for HPS and SGHMIB and decreased by approximately 50% for MGS.

Regarding the changes in consumer behavior for both types of commerce, Figure 4 (TSNSS) and Figure 5 (TEC) present the correlation between the commercial branches through a scatter plot matrix, illustrating the relationships between the retail trade sectors examined in the study.

From Figure 4, the off-diagonal cells of the scatter plot indicate how values in these two categories may be related. The histograms along the diagonal show the distribution of values within each sector, with some sectors exhibiting more variation or spread than others. Additionally, the densely clustered points along a linear path in any cell indicate a positive or negative correlation between those two sectors.

The scatter plots involving TRT often have a more defined relationship with other categories, suggesting that TRT has a strong influence on individual sectors or vice versa. NSR also shows a moderate level of correlation with certain categories, such as TRT, but the points are more dispersed, implying a less direct correlation. Some categories, such as CGM and FBME, show closer clustering in their scatter plots, indicating potentially stronger relationships.

In Figure 5, the scatterplot matrix reveals some visual differences in TEC compared to Figure 4, which is dedicated to TSNSS. For some sectors, the points clustered around a linear trend suggest some correlation, while a random dispersion indicates weaker or no correlation. Comparing sectors like TRT with MVP shows a more defined pattern, suggesting a stronger relationship between these categories. The diagonal histograms indicate, for example, that CGM has a somewhat spread distribution, suggesting variability in this category. TRT with several other sectors shows some clustering of points, indicating that overall retail trade is likely influenced by or influencing these sectors. The scatter plot between NSR and TRT also suggests a relationship, but it appears to be more dispersed than, for example, MVP with TRT.

For better visualization of movement and increasing trends for e-commerce after the pandemic COVID-19 time, the heat maps were made for each commerce sector (Figure 6, Figure 7, Figure 8, Figure 9, Figure 10 and Figure 11) using the quarterly mean as an indicator (Y-axis) and segmented into three phases, pre-pandemic, during pandemic, and post-pandemic (X-axis), to provides context for understanding how retail trade evolved through these distinct phases. The color intensity represents the mean value of each indicator, with lighter shades indicating lower values and darker shades showing higher values. The legends on the right show the color mapping, where lighter red corresponds to mean values around the minimum value and darker red corresponds to maximum values.

The heat map from Figure 6 represents changes in TRT over time as follows: for pre-pandemic (Q1 2018–Q4 2019), TRT values are generally lower during this period, as indicated by the lighter shades of red and suggests a stable but lower baseline for TRT before the COVID-19 pandemic; during pandemic (Q1 2020–Q4 2021), as the pandemic begins in Q1 2020, there is a noticeable increase in color intensity, this trend suggests an upward shift in retail trade values during the pandemic period, which could reflect increased consumer spending in certain areas like online shopping, essential goods, or home-related products due to lockdowns and lifestyle adjustments; post-pandemic (Q1 2022–Q1 2024), there is a significant increase in the color intensity, with the post-pandemic phase showing the darkest shades of red. This indicates that TRT values have reached their highest levels during this period, likely due to the economic recovery, pent-up demand, and increased consumer spending as restrictions eased and the economy adapted.

In the case of MVP, different evolutions in the time of these sales are noticed, but with the same evolution for the intensity of e-commerce after the COVID-19 pandemic, and some observable seasonality for the 3rd quarter of each year, during summertime and holidays, including for the 3rd quarter of 2021 during pandemic time (Figure 7). It is also obvious that the increase in these sales post-pandemic is marked by the growth of e-commerce in this sector as well.

The TEC of FBME recorded an increase for the first quarter of the pandemic, followed by a similar rise in the post-pandemic (Figure 8).

In Figure 9, CCA registered lower e-commerce sales during the pre-pandemic but during the pandemic, it can be seen darker red colors appeared, indicating an increase in sales or demand for these goods, possibly due to changes in shopping behavior during the pandemic to prepare for remote work and changes in the style of CGM. For the post-pandemic period, the trend continues with darker shades, indicating that sales levels have been maintained or increased further. There is a noticeable increase in sales during the pandemic, with these elevated levels continuing into the post-pandemic period. The shift in color from lighter to darker shades during the pandemic suggests that consumer spending in this category rose significantly and sustained that increase post-pandemic.

Figure 10 shows that the TEC of AO in pre-pandemic times had low sales levels in these years, but during the pandemic, it underscores a substantial increase in sales for this category. In the post-pandemic period, the dark red color in Figure 10 indicates that the TEC of AO continued to rise. By considering the sales resilience of AO sectors and e-commerce categories compared to previous ones, AO sales increased during the pandemic and remained elevated afterward, possibly indicating a permanent shift in consumer behavior. As a key aspect of this sector, the sustained darker colors in AO indicate that this category may have experienced more robust growth or higher absolute sales levels.

In the case of NDR, the value scale ranges from 75,000 to 250,000, which is lower than the AO category but similar to CGM (Figure 11). This means that, while NSR (like online shopping) is a significant segment, it does not reach the same sales volume as a broad mix of physical and other goods. During the pandemic, the graph shifts quickly to darker shades, showing a notable surge and an increase in NSR. This likely reflects the rapid shift toward online shopping as physical stores faced restrictions and consumers turned to e-commerce for essential and non-essential goods. The rise in color intensity during this period underscores the growing importance of online retail. In the post-pandemic period, non-store retailer sales remain at a higher level than pre-pandemic, with consistently darker red shades. This persistence suggests that many consumers have continued shopping online even after restrictions were eased, indicating a lasting shift toward digital shopping habits.

The heat maps from Figure 6, Figure 7, Figure 8, Figure 9, Figure 10 and Figure 11 show, comparatively, how different product categories and e-commerce sectors responded similarly but at varying scales to the pandemic. For the AO category, Figure 10 also shows a stronger overall sales volume both before and after the pandemic, and this difference in magnitude highlights varying consumer priorities or spending capacities across different types of goods pre-, during, and post-pandemic. In all of the heat maps, there is a distinct shift to darker shades right at the beginning of the pandemic period. This marks a noticeable increase in demand across these e-commerce sectors, likely driven by changes in lifestyles and consumption patterns during lockdowns. However, the transition in the AO category is slightly more rapid, reflecting a surge in demand for these items as opposed to the more gradual shift seen in all the other categories.

These graphical representations sustained high levels post-pandemic, and all the categories maintained high sales levels post-pandemic, as indicated by the continuous dark red shading. This expresses that whatever drove the sales increase during the pandemic (such as increased home-based needs, digital shopping trends, or lifestyle shifts) had a lasting impact. For CGM, this might indicate ongoing changes in fashion or casual wear trends, while for AO, it could reflect a broader increase in demand for household or non-clothing essentials. Another important aspect emphasized by the heat maps is that for the majority of the e-commerce sectors, the color density shifts steadily from light to dark over several quarters. In contrast, AO shifts to darker colors more quickly, hinting at a fast adjustment in consumer spending for broader categories outside of clothing, potentially covering a wider range of essential goods that have become priority purchases.

4.2. Testing Similarities or Differences Among E-Commerce Sales in Pre-, During, and Post-Pandemic Periods

To compare the similarities and differences in values of data from pre-, during, and post-pandemic periods, the independent samples Kruskal–Wallis test with Bonferroni multiple comparisons was applied, separately for total store and non-store sales (

Table 4. Results of the Kruskal–Wallis test for TSNSS.

Null Hypothesis	Sig.	Decision
TSNSS of TRT distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis
TSNSS of MVP distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis
TSNSS of FBME distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis
TSNSS of CGM distribution is the same across categories of Pre/during/post COVID-19	0.003	Reject the null hypothesis
TSNSS of AO distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis
TSNSS of NSR distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis

) and for total e-commerce (

Table 5. The results of the Kruskal–Wallis test for TEC.

Null Hypothesis	Sig.	Decision
TEC of TRT distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of MVP distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of FBME distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of FHF distribution is the same across categories of Pre/during/post COVID-19	0.001	Reject the null hypothesis.
TEC of EA distribution is the same across categories of Pre/during/post COVID-19	0.010	Reject the null hypothesis.
TEC of BMGES distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of CGM distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of CCA distribution is the same across categories of Pre/during/post COVID-19	0.004	Reject the null hypothesis.
TEC of GM distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of AO distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of FB distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of HPS distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of SGHMIB distribution is the same across categories Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of MGS distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.
TEC of NSR distribution is the same across categories of Pre/during/post COVID-19	<0.001	Reject the null hypothesis.

).

The TSNSS of all six categories indicates the statistically significant distribution of values through these three periods (Table 4). The Bonferroni pairwise multiple comparisons reveal some notable differences, for instance, between pre- and post-pandemic periods (0.000) for TRT, MVP, CGM, AO, and NSR, and between pre- and during-pandemic periods (0.010) and between pre- and post-pandemic periods (0.000) for FBME.

For TEC, the results from Table 5 also point out the statistically significant distribution of values across these three periods, and the Bonferroni pairwise multiple comparisons highlight some particularities. Thus, differences between pre- and during the pandemic are recorded for TRT, MVP, FBME (including FHF, EA, and BMGES), CGM, AO (FB, HPS, SGHMIB, and MGS), and variation between pre- and post-pandemic time are registered for CCA and NSR.

To determine if the data from time series are or not stationary (the formal checking for unit roots), the PACF charts were applied, and Figure 12a–f present the results for total store and non-store sales, and in Figure 13a–f for total e-commerce.

The PACF charts from Figure 12a–f show some particularities for each commerce sector for TSNSS in the US for the period Q1 2018–Q1 2024:

• For TRT, the partial auto-correlation coefficients are statistically significant for lags 1 and 5, with all the rest of the PACF coefficients falling between the horizontal lines. Within a 95% confidence interval, these aspects also indicate the existence of non-stationarity in this time series and suggest a first-order differencing as the remedy. The first-lag partial autocorrelation and the 5th-lag autocorrelation are above the critical limit. The meaning of the first-lag and 5th-lag partial autocorrelations implies that the TRT at time “t” is correlated with its value at time “t−1” but also with time “t−5”.
• For the MVP, the partial auto-correlation coefficients are statistically significant for lag 1, with the rest of the PACF coefficients falling between the horizontal lines. Within a 95% confidence interval, these aspects also indicate the existence of non-stationarity in this time series and suggest that first-order differencing is the remedy.
• For FBME, the partial auto-correlation coefficients are statistically significant for lags 1, 2, 4 (at the limit), and 5, while all the other PACF coefficients fall between the horizontal lines, within a 95% confidence interval. These aspects also state the existence of non-stationarity for this time series and suggest first-order differencing as the remedy. The first-lag, the second-lag partial autocorrelation, and the 5th-lag autocorrelation are above the critical limit, the 4th being at the limit. The meaning of these lag values for partial auto-correlations implies that FBME at time “t” is correlated with its value at time “t−1” and time “t−2”, time “t−4”, and with time “t−5”.
• For CGM, the partial auto-correlation coefficients are statistically significant for lags 4 and 5, while all the rest of the PACF coefficients fall between the horizontal lines. With a 95% confidence interval, these aspects also indicate the existence of non-stationarity in this time series and suggest a 4th-order differencing as the remedy. The fourth-lag partial autocorrelation and the 5th-lag autocorrelation are above the critical limit. The meaningfulness of the 4th and 5th lags’ partial autocorrelations implies that the CGM at time “t” is correlated not only with its value at time “t−4” but also with time “t−5” for clothing and general merchandise.
• For AO, the partial auto-correlation coefficients are statistically significant for lags 1 and 5, with all the rest of the PACF coefficients falling between the horizontal lines. Within a 95% confidence interval, these aspects also indicate the existence of non-stationarity in this time series and suggest that first-order differencing is the remedy. The first-lag partial autocorrelation and the 5th-lag autocorrelation are above the critical limit. The meaningfulness of the first-lag and 5th-lag partial autocorrelations implies that AO sales at time “t” are correlated with its value at time “t−1” but also with time “t−5”.
• For NSR, the partial autocorrelation coefficients are statistically significant for lag 1, with the rest of the PACF coefficients lying between the horizontal lines. Within a 95% confidence interval, these aspects also reveal the existence of non-stationarity in this time series and suggest first-order differencing as the remedy.

The PACF charts for TEC are presented in Figure 13a–f.

The PACF charts from Figure 13a–f show some particularities for each commerce sector for total store and non-store sales in the US for the period quarter 2018 to Quarter 1 2024:

• For TRT, the partial auto-correlation coefficients are statistically significant for lags 1 and 5, with all the rest of the PACF coefficients falling between the horizontal lines. Within a 95% confidence interval, these aspects also demonstrate the existence of non-stationarity in this time series and suggest a first-order differencing as the remedy. The first-lag partial autocorrelation and the 5th-lag autocorrelation are above the critical limit. The meaning of the first-lag and 5th-lag partial autocorrelations implies that the TRT at time “t” is correlated with its value at time “t−1” but also with time “t−5”.
• For the MVP, the partial auto-correlation coefficients are statistically significant for lag 1, with the rest of the PACF coefficients falling between the horizontal lines. Within a 95% confidence interval, these aspects also indicate the existence of non-stationarity in this time series and suggest first-order differencing as the remedy.
• For FBME, the partial auto-correlation coefficients are statistically significant for lag 1 and at the limit for lag 5; all the rest of the PACF coefficients are between the horizontal lines. With a 95% confidence interval, these aspects also indicate the existence of non-stationarity in this time series and suggest that first-order differencing is the remedy. The first lag is above the critical limit, and the 5th lag has an autocorrelation at the limit. The meaning of these lag values for partial auto-correlations implies that FBME at the time “t” is correlated with its value at the time “t−1”.
• For CGM, the partial auto-correlation coefficients are statistically significant for lags 1, 4, and 5 (at the limit), while all the other PACF coefficients fall between the horizontal lines, within a 95% confidence interval. These aspects also denote the existence of non-stationarity for this time series and reveal first-order differencing as the remedy. The first-lag, the fourth-lag partial auto-correlation, and the 5th-lag auto-correlation are above the critical limit. The meaningfulness of the 1st, 4th, and 5th lags’ partial autocorrelations implies that the CGM at time “t” is correlated with its value at time “t−1” and “t−4”, but also with time “t−5” for clothing and general merchandise.
• For AO, the partial auto-correlation coefficients are statistically significant at lags 1 and 5 (at the limit), with all the rest of the PACF coefficients falling between the horizontal lines, within a 95% confidence interval. These aspects also indicate the existence of non-stationarity in this time series and suggest first-order differencing as the remedy. The first-lag partial autocorrelation is above the critical limit, and the 5th-lag autocorrelation is at the limit. The meaning of the first-lag partial autocorrelations denotes that AO sales at time “t” are correlated with their value at time “t−1” but also with time “t−5”.
• For NSR, the partial auto-correlation coefficients are statistically significant for lags 1 and 5, with all the rest of the PACF coefficients falling between the horizontal lines of a 95% confidence interval. These aspects also indicate the existence of non-stationarity in this time series and suggest that first-order differencing is the remedy.

For all types of e-commerce, descriptive statistics were computed for each period (pre-pandemic, during, and post-pandemic). The results are depicted in

Table 6. Descriptive statistics for all the types of e-commerce.

Types of E-Commerce	Pre- Pandemic	During Pandemic	During/Pre (%)	Post-Pandemic	Post/During (%)	Post/Pre (%)	[Post/Pre (%)—During/Pre (%)]/During/Pre (%)
0	1	2	3 = 2/1	4	5 = 4/2	6 = 4/1	7 = (6 − 3)/3
TRT	136,107.22 ± 21,275.280	230,756.29 ± 28,057.793	169.5	264,932.67 ± 28,780.271	114.8	194.6	14.8
MVP	9355.78 ± 656.498	131,85.57 ± 1706.253	140.9	14,641.89 ± 630.953	111.0	156.5	11.0
FBME	14,538.33 ± 1942.593	27,506.14 ± 3312.669	189.2	27,688.11 ± 2293.299	100.7	190.4	0.7
CGM	22,847.56 ± 6480.572	42,848.14 ± 8977.690	187.5	47,113.89 ± 8259.925	110.0	206.2	10.0
AO	8945.56 ± 1382.249	17,468.57 ± 1585.802	195.3	18,690.67 ± 1714.561	107.0	208.9	7.0
NSR	80,420.00 ± 11,667.197	129,747.86 ± 14,719.021	161.3	156,798.11 ± 18,141.907	120.8	195.0	20.8

.

The results from Table 6 (columns 3, 5, 6, and 7) show the comparative evolution of the types of e-commerce during the pandemic compared with pre-pandemic (in percentages), post-pandemic pandemic, post-pandemic compared with pre-pandemic and, in the last column, the net evolution of post-pandemic, where increasing trends are obvious compared to during pandemic when practically the increasing of e-commerce begins. Therefore:

• During the pandemic, compared with pre-pandemic, the biggest percentage increase goes to AO (195.3%), followed by FBME (189.2%), CGM (187.5%), TRT (169.5%), NSR (161.3%), and with the minimum for MVP with 140.9%;
• Post-pandemic, compared with during the pandemic, for all the types of e-commerce, the increase was registered with the biggest percent for NSR (120.8%), followed by TRT (114.8%), and the minimum for FBME (100.7%);
• Post-pandemic, compared with pre-pandemic time, the hierarchy is quite different from comparison during the pandemic with pre-pandemic, as follows: AO (208.9%), CGM (206.2%), NSR (195%), TRT (194.6%), FBME (190.4%), and the minimum increase for MVP with 156.5%.

The last column of Table 6 indicates that NSR (+20.8%) and TRT (+14.8%) recorded the largest increase, followed by MVP at 11%, CGM at 10%, AO at 7%, and FBME at 0.7%.

For all sub-types of e-commerce, descriptive statistics were computed for each period (pre-pandemic, during, and post-pandemic). The results are presented in

Table 7. Descriptive statistics for all the sub-types of e-commerce.

Sub-Types of E-Commerce	Pre- Pandemic	During Pandemic	During/Pre (%)	Post-Pandemic	Post/During (%)	Post/Pre (%)	[Post/Pre (%)—During/Pre (%)]/During/Pre (%)
0	1	2	3 = 2/1	4	5 = 4/2	6 = 4/1	7 = (6 − 3)/3
FHF	2844.71 ± 368.231	5038.67 ± 560.008	177.1	4566.11 ± 544.636	90.6	160.5	−9.4
EA	9011.50± 1817.399	14,174.71 ± 2571.558	157.3	13,493.78± 812.084	95.2	149.7	−4.8
BMGES	3688.89 ± 564.288	8472.29 ± 795.451	229.7	9628.22 ± 882.289	113.6	261.0	13.6
CCA	10,853.22 ± 2764.693	169,60.57 ± 3519.601	156.3	17,472.67 ± 3353.004	103.0	161.0	3.0
GM	11,994.33 ± 3752.069	25,887.57 ± 5476.962	215.8	29641.22 ± 5226.094	114.5	247.1	14.5
FB	2183.63 ± 922.325	7293.29 ± 496.574	334.0	7854.67 ± 577.312	107.7	359	7.7
HPS	1338.22 ± 298.615	2155.00 ± 450.125	161.0	2507.22 ± 528.692	116.3	187.4	16.3
SGHMIB	2022.11 ± 319.015	3612.29 ± 560.741	178.6	3501.78 ± 624.393	96.9	173.2	−3.1
MGS	3387.56 ± 180.084	4408.00 ± 410.900	130.1	4827.00 ± 173.738	109.5	142.5	9.5

.

The results from Table 7 (columns 3, 5, 6, and 7) show the comparative evolution of these sub-types of e-commerce during the pandemic compared with pre-pandemic (in percentages), post-pandemic compared with during pandemic, post-pandemic compared with pre-pandemic, and, in the last column, the net evolution of post-pandemic, where increasing trends are obvious compared to during pandemic, when practically the increase in e-commerce begins. Hence:

• During the pandemic, compared with pre-pandemic and post-pandemic times, the highest increase was registered by FB (334%), followed by BMGES (229.7%), GM (215.8%), SGHMIB (178.6%), FHF (177.1%), HPS (161%), with the minimum for MGS with 130.1%.
• In the post-pandemic period, compared to during the pandemic, FHS, EA, and SGHMIB recorded declines.

The last column of Table 7 underscores that FHF (−9.4%), EA (−4.8%), and SGHMIB (−3.1%) recorded a decline. However, sub-types of e-commerce, such as HPS (+16.3%), GM (+14.5%), and BMGES, registered a rising trend after the pandemic.

In Figure 14a–i, the PACF charts for these sub-types of e-commerce are presented to sustain (or not) the previous results. Additionally, the results of the independent samples Kruskal–Wallis test and Student t-test comparing the means of these sub-types of e-commerce across the three periods confirm statistically significant differences between the periods, particularly between the post-pandemic and pre-pandemic times. There are some similarities with the results for TEC, but there are also some particularities, as follows:

• For FHF, BMGES, food FB, and MGS, the partial auto-correlation coefficients are statistically significant for lag 1, with all the rest of the PACF coefficients falling between the horizontal lines. Within a 95% confidence interval, these aspects also indicate the existence of non-stationarity in this time series and suggest that first-order differencing is the remedy. The first lag is above the critical limit and the autocorrelation. The meaning of these lag values for partial auto-correlations implies that for these sub-types of e-commerce, at time “t” is correlated with its value at time “t−1”;
• For EA, all the PACF values are inside the critical limits with a 95% confidence interval;
• For CCA, the partial auto-correlation coefficients are statistically significant for lags 4 and 5, with all the rest of the PACF coefficients falling between the horizontal lines. Within a 95% confidence interval, these aspects also indicate the existence of non-stationarity in this time series, suggesting that the fourth-order and fifth-order differencing are the remedies. The fourth-lag (positive) and fifth-lag (negative) are above the critical limit and autocorrelation. The meaning of these lag values for partial auto-correlations outlines that for CCA at time “t” is correlated with its value at time “t−4” and “t−5”;
• For GM and SGHMIB, the partial auto-correlation coefficients are statistically significant for lags 1, 4, and 5; all the rest of the PACF coefficients fall between the horizontal lines with a 95% confidence interval. These aspects also indicate the presence of non-stationarity in this time series and suggest that fourth-order and fifth-order differencing are the remedies. The first-lag, fourth-lag (positive), and fifth-lag (negative) are above the critical limit and autocorrelation. The meaning of these lag values for partial auto-correlations implies that for GM and SGHMIB, at the time “t” is correlated with its value at the time “t−1”, “t−4”, and “t−5”;
• For HPC, the partial auto-correlation coefficients are statistically significant for lags 2 and 4, with all the rest of the PACF coefficients falling between the horizontal lines within a 95% confidence interval. These aspects indicate the existence of non-stationarity in this time series and suggest that the 4th-order and 5th-order differencing are the remedies. The second-lag and 4th-lag (positive) are above the critical limit and autocorrelation. The meaning of these lag values for partial auto-correlations implies that HPC at the time “t” is correlated with its value at the time “t−2” and “t−4”.

4.3. Testing the Seasonality of Consumer Buying Behavior Between Classical and E-Commerce Stores

The heat map graphical representation highlights the trend and some seasonality in consumer buying behavior in both traditional (store) and e-commerce settings. At the same time, to detail and analyze these, for both TSNSS and TEC, we will present in the following the results (also based on graphical representation, the most indicated for time series as our research data) of sequence charts due to their advantage to opt for possible transformation of data from time series by natural log transformation or difference.

Starting from the results from Figure 12, Figure 13 and Figure 14 of PACF, we carried out the research by applying quantitative transformations as follows: the difference (1) for MVP (store and e-commerce) and for non-store (classic trade) and the difference (1) and seasonally difference (1) for all the rest of the total store and non-store trade and total e-commerce types of trade.

Figure 15a–f and Figure 16a–f depict TSNSS and TEC, respectively:

• The time series without transformations (Figure 15a,c,e,g,I,k) and
• With transformations (Figure 15b,d,f,h,j,l));
• For difference (1) and seasonal difference (1, period 4) or (in the case of total retail trade) with transformations for difference (5) and seasonal difference (1, period 4) according to the PACF results from previous figures.

According to Figure 16a,c,e,g,i,k for TSNSS without transformations of the data, it is obvious there is an increasing trend after the COVID-19 pandemic time for all types of commerce or maintaining the trends for furniture and clothing, for example. There is a clear upward trend over time in retail sales, although it fluctuates. After the vertical dashed lines, the increase in sales appears more pronounced, suggesting an event-driven or seasonal effect. This line shows the different and seasonally adjusted TRT over time.

In Figure 16b,d,f,h,j,l, the transformations applied demonstrate that differencing removes trends and focuses on changes between consecutive periods to stabilize the mean. The seasonal adjustment removes seasonal effects (e.g., holidays, recurring patterns), and the differenced series shows strong spikes in both directions, indicating periods with significant changes in commercial sales. A post-spike stabilization could be observed, the fluctuations reduce, and the data stabilizes around the horizontal axis (zero).

For TEC, similar trends were observed after the COVID-19 pandemic, but the stabilizations (Figure 16b,d,f,h,j,l) are more pronounced.

Therefore, according to the sequences’ charts in Figure 15 and Figure 16, all the increasing trends in sales above the analyzed period can be attributed to the increasing trends during the pandemic time. All registered values in sales from the post-pandemic period are dependent on or directly correlated with values registered during the pandemic period for both types of commerce, classical or electronic. These conclusions are sustained by the ACF charts from Figure 17a–l.

The ACF practically confirms that the applied transformations for our quarterly data were a good decision and supports the extrapolation of buying behavior during the pandemic to the post-pandemic period.

5. Discussion

The COVID-19 pandemic has changed consumer behavior globally. This change had direct effects on e-commerce. In this context, the question of a comparative analysis of behavioral evolution in different regions, such as the US, Europe, and Asia, was raised. Studies have observed a common trend in the adoption of online shopping in these regions. A number of differences were confirmed, while others were questioned regarding traditional consumer behavior models during crisis periods [119,120,121]. In the US, the pandemic accelerated the transition to online payments, especially in the food sector. Studies [119,120,121] have demonstrated that American consumers have adapted e-commerce platforms for purchasing essential products. This change largely influenced consumer behavior, as they were forced to socially distance themselves due to fears of exposure in physical commercial spaces. This evolution highlighted a global trend, observed in Europe, particularly among European Union (EU) member states, as well as in Asia. In this space, the use of mobile applications for shopping and their frequency increased in the food sector during the quarantine period [122,123,124,125].

Tyrväinen and H. Karjaluoto [126] conducted a meta-analysis on online food shopping, based on dynamic data collected during the pandemic. The results demonstrated this transformation through the sudden increase in the perceived usefulness of online shopping in emergency contexts. Beyond these similarities, the way these changes have manifested differs subtly between regions. Thus, a rise in impulsive buying behavior was observed in the US. Consumers repeatedly purchased fitness products, home equipment, food, or everyday items, driven by the general anxiety related to the pandemic context and uncertainty [127,128]. However, in Europe, although impulsiveness was present, it was accompanied by an increased emphasis on sustainable purchasing practices. Wong et al. [129] state that European consumers have shown an increasing trend toward responsible consumption. This behavior highlighted increased awareness of the environmental impact of consumer choices. These changes are analyzed in detail in several recent studies. Some of them focus on consumers’ ecological orientations and their sustainable behaviors in the context of the fear generated by the pandemic. At the same time, the research by Dangelico et al. [130] examines the evolution of Italian consumers. This evolution marks an increase in sustainable shopping, driven by heightened environmental awareness during the pandemic. The study is directly influenced by socio-demographic factors such as age, education level, standard of living, income, and other direct factors. Agnani et al. [131] show the accelerated transition toward e-commerce in Asian countries, where small and medium-sized enterprises are developing demand forecasting models to adapt to the new realities of general crisis consumption.

Similarly to the other categories, there is a stark contrast between pre-pandemic and pandemic sales levels for non-store retail in the US. However, unlike some other categories, the transition for non-store retailers is immediate and sustained from the early pandemic onward, without a gradual build-up. This indicates a sudden shift as consumers quickly adapted to online shopping.

All results indicate the pandemic as a defining turning point. All the graphical representations from our research serve as a clear dividing line, where sales levels distinctly increase and never return to pre-pandemic levels. This observation highlights the pandemic’s pivotal role in both categories, but particularly for AO, where the increase appears more robust and immediate.

These particularities help highlight how consumer behavior adapted not only in response to immediate needs during the pandemic but also led to longer-term shifts in spending patterns, with each category reacting uniquely based on product type and perceived necessity.

Our research results, obtained through statistical and econometric analysis, underscore the pandemic’s role in accelerating e-commerce trends. Before the pandemic, non-store retailer sales were lower. Once the pandemic struck, online retail sales surged rapidly, marking a significant structural shift in the retail landscape that has continued to shape shopping behaviors. Practically, the heat maps together with the previous graphics and results highlight the pandemic as a structural shift for e-commerce, with the main features:

• Pandemic boost. The pandemic dramatically boosted non-store retailer sales, reinforcing the importance of e-commerce;
• Lasting impact. Sales remain elevated post-pandemic, suggesting a lasting increase in online shopping demand;
• Fast adoption: The rapid shift to darker colors during the pandemic indicates that consumers quickly pivoted to online shopping.

This paper highlights the pandemic’s transformative impact on retail, particularly by accelerating the growth and normalization of e-commerce as a primary shopping channel.

The COVID-19 pandemic marked a watershed moment for e-commerce in the US, accelerating its adoption across sectors and leaving a lasting impact on consumer behavior. This study presents robust evidence, both statistical and graphical, on how the pandemic has redefined retail trade, with significant implications for businesses and consumers alike.

The findings indicate an acceleration of e-commerce growth based on the fact that TEC sales grew significantly during the pandemic, with categories like AO experiencing a remarkable 195.3% increase compared to pre-pandemic levels. In addition, during the post-pandemic period, e-commerce sales stabilized at higher levels, with a cumulative increase of 194.6% for TRT compared to pre-pandemic periods. Concerning the sectoral insights, the results highlight that TEC of FB recorded the highest surge during the pandemic, with sales increasing by 334%, reflecting the critical need for home delivery of essential goods. Furthermore, the BMGES also registered sustained growth, with sales increasing by 261% post-pandemic compared to pre-pandemic levels. CGM grew by 187.5% during the pandemic, and its post-pandemic growth settled at a more modest 10% compared to the pandemic period.

The time series instruments and methods applied in this paper (PACF and ACF) highlight the sustained trends in the post-pandemic period. Thus, we demonstrate that the entire increasing trend post-pandemic is based (partially auto-correlated and auto-correlated) on the previously registered values of sales, especially for e-commerce, which is dependent on the value for t−1 and t−5 increasing registered during the pandemic time. The statistical analysis revealed that even as physical stores reopened, e-commerce sales did not regress to pre-pandemic levels.

Heat maps and scatter plots highlighted the persistent elevation in sales, particularly for NSR, which grew by 161.3% during the pandemic and maintained a 120.8% increase post-pandemic compared to pandemic levels.

Also, our research shows some structural shifts. The PACF and Kruskal–Wallis test results confirmed statistically significant differences in sales across pre-, during-, and post-pandemic periods. For example, the PACF analysis revealed clear seasonality and lagged correlations, indicating that pandemic-induced buying behaviors persistently influence current sales patterns.

These findings demonstrate that the pandemic was more than just a temporary disruption, as it permanently altered the retail landscape. Retailers that quickly adapted to online models gained a competitive edge, while consumers became accustomed to the convenience and flexibility of e-commerce.

6. Conclusions

In conclusion, this research highlights the pandemic’s role as a structural shift in e-commerce, with both statistical and practical evidence indicating its transformative and enduring impact. The challenge now lies in understanding how businesses can continue to innovate and sustain this growth in a rapidly evolving retail ecosystem.

The limitations of the study are related to:

• Data due to aggregated retail trade data, which can mask variations across different sectors, regions, or store types (e.g., physical stores vs. e-commerce), and the seasonally adjusted and external factors (e.g., economic shocks, pandemics) may still distort the results;
• Time framework limitations. The period analyzed may not be long enough to capture long-term trends or structural changes, which limits the robustness of the conclusions. Short-term fluctuations might dominate the results, making it hard to infer persistent patterns;
• Unobserved factors, such as market competition, consumer behavior, or global economic conditions, may also influence the results.

The findings may be context-specific (in the US) and not generalizable to other specific markets or time periods.

To resume our research results:

• The results of this study demonstrate that the COVID-19 pandemic had a profound and lasting impact on US consumer shopping behavior. Time series analyses revealed that e-commerce sales experienced a significant and rapid increase during the pandemic, with growth rates exceeding 150–200% in most categories compared to pre-pandemic levels. Importantly, post-pandemic sales did not return to pre-pandemic baselines but remained at elevated levels, confirming a structural shift in consumer preferences.
• PACF and ARIMA analyses highlighted clear patterns of seasonality and lagged correlations across different categories, indicating that consumer behaviors developed during the pandemic continue to shape current market dynamics. The ARIMA models provided robust evidence of persistence in e-commerce adoption, reinforcing the predictive value of these econometric tools for future retail forecasting.
• Inferential statistics (Kruskal–Wallis tests) confirmed statistically significant differences between pre-, during, and post-pandemic phases across all major retail categories and subcategories. Categories such as food and beverage, general merchandise, and non-store retailers recorded the strongest and most sustained growth, while furniture, electronics, and clothing displayed heterogeneous recovery patterns.
• Overall, the findings emphasize that the pandemic acted not merely as a temporary disruption but as a structural accelerator of digital commerce, fundamentally reshaping US retail trade and embedding e-commerce as a permanent consumer habit.

The analysis of how the COVID-19 pandemic transformed consumer behavior in the US reveals several important implications for both consumers and trade policymakers. First, the long-lasting and structural shift toward e-commerce is confirmed, representing a fundamental change in the dynamics of demand and supply. For consumers, the convenience, flexibility, and accessibility of digital channels have moved beyond mere situational advantages to become a behavioral norm. This shift has empowered consumers by giving them greater access to price comparisons, more diverse options, and increased informational transparency. Additionally, the data shows the emergence of hybrid consumption models: essential goods are mainly purchased online, while purchases of branded clothing, electronics, furniture, luxury items, and similar products continue to favor the traditional retail experience.

From policymakers’ perspective, these findings highlight the need to rethink the regulatory framework and trade support tools. The ongoing online sales following the pandemic suggest that public policies must adapt to a digital-dominated market, where data protection, transaction security, and fair competition are becoming increasingly vital. Meanwhile, declining traffic in physical shopping centers raises concerns for local taxes and urban planning, calling for transitional policies that promote the coexistence of physical and digital retail. The growth of e-commerce logistics also poses environmental challenges, underscoring the need for sustainable solutions and infrastructure that are integrated into urban mobility and smart commerce strategies. This analysis gains additional relevance when compared across different market structures. In Western Europe, where retail markets are mature, trends indicate a consolidation of e-commerce, influenced by stringent regulations focused on consumer protection and fair competition. In Central and Eastern Europe, where digitalization is less advanced, US market results can serve as benchmarks for accelerating the digital shift and bridging gaps with more developed economies. In Asia, the situation is more complex: in China, where digital infrastructure was already well-established, the analysis confirms the permanent shift toward online habits. In Southeast Asia, the rise in mobile payments and the integration of e-commerce into social platforms like WeChat or Grab represent similar structural shifts, but with unique cultural and institutional differences.

The methodological value of the study derives from the application of robust econometric tools, such as PACF and ARIMA, which highlight the relevance and seasonality of new consumption patterns. The practical utility of these models lies in their ability to provide a predictive planning framework for various economic actors and public institutions, thereby contributing to the optimization of inventory management, the adjustment of logistics chains, and the calibration of pricing strategies. This approach emphasizes the relevance of the research not only as a retrospective analysis of a major disruption but also as a tool to substantiate policies and strategies in an economic environment that is constantly reconfiguring. In conclusion, the research highlights that the pandemic has served as a catalyst for digitalization in retail, bringing about irreversible changes in consumer behavior and shifting the balance between traditional and digital channels. The implications for consumers are evident in the consolidation of access and bargaining power, while for policymakers, the need for proactive adaptation of regulations and economic infrastructure is crucial. From a comparative perspective, the US experience constitutes a laboratory for structural change, from which lessons can be extrapolated and adapted, with the necessary methodological precautions, to the European and Asian contexts. In this sense, the analysis becomes not just an empirical description of a crisis, but a basis for reflection on the directions of innovation and adaptation in a global retail system in transition.

Future research may consider the following:

• To extend the analyzed time frame, namely for a long-term period by taking into consideration the economic recovery periods, changes in consumer habits, or technological adoption;
• To focus on the sectoral analysis to identify where changes are most significant;
• To apply other predictive models based on ML and use granular data to account for seasonality and market dynamics;
• To explore the role of e-commerce and digital transformation in retail trade growth and stability, especially in times of disruption

In conclusion, we used the PACF and ARIMA approach for time series not with a prospective scope (to identify predictive values for each category of retail) but with a retrospective one, analyzing the main relationships between data from pre, during, and post pandemic e-commerce in US to determine if the COVID-19 pandemic changed the consumer behavior.