1 Introduction

Over the past two decades, the proportion of retail sales attributed to e-commerce exhibited a consistent annual increase until the onset of the Covid-19 pandemic in early 2020. As uncertainties due to the pandemic began to spread in the first half of 2020, temporary closure of non-essential stores led to a decline in the sales in brick-and-mortar sales and a significant rise in e-commerce sales. Between the first and second quarter of 2020, adjusted total retail sales witness a reduction of approximately 4%, while e-commerce sales increased by more than 30% in the same timeframe (U.S. Census Bureau, 2021a). Non-essential stores like clothing and clothing accessories stores were especially impacted by these events, as the total retail sales (unadjusted) of clothing and clothing accessories dropped from $19.2 billion in February 2020 to $2.8 billion in April 2020 – an 85% decrease (U.S. Census Bureau, 2022). Meanwhile, unadjusted online sales of clothing and clothing accessories increased by about 40% between the first and second quarters of 2020 (U.S. Census Bureau, 2021b).

The sudden shift from shopping in traditional brick-and-mortar stores to online stores prompted market researchers and transportation planners to anticipate permanent changes to consumer shopping behavior in favor of online shopping (Sheth, 2020). Forecasts envisioned a significant reduction in shopping-related passenger travel demand, which would ultimately lead to a decrease in traffic congestion and energy consumption. However, as vaccination began in December 2020 and the lockdown restrictions were gradually lifted in the first quarter of 2021, shoppers became more comfortable slowly returning to their former shopping habits. By the second quarter of 2021, adjusted total retail sales had grown more than 28% from the same quarter a year ago, about twice higher than the 14.3% growth rate of e-commerce during that same period (U.S. Census Bureau, 2022).

Although these patterns seem to suggest that the shopping behavior seen during the lockdown phase of the pandemic was mostly temporary and shoppers may have returned to their usual shopping habits, a more plausible scenario involves the emergence of hybrid shopping habits, encompassing aspects from both the pandemic and pre-pandemic eras. While extensive research has explored the interaction between online and in-store shopping behaviors and its implications for e-commerce supply chains, retail strategies, urban planning, and travel demand modeling (Gevaers et al., 2014; Le et al., 2021; Mokhtarian, 2004; Perboli & Rosano, 2019; Rotem-Mindali & Weltevreden, 2013; Zennaro et al., 2022), the emergence of novel interactions resulting from the recent Covid-19 situation, particularly for non-essential products which were predominantly inaccessible in-store during the lockdowns, remains an area requiring investigation. Moreover, potential variations in the relationships between online and in-store shopping behavior for products within the same product class may differ have yet to be investigated in the existing literature (Zhen et al., 2018).

Thus, this study endeavors to investigate the interaction between online and in-store shopping behaviors for non-essential experience goods. This study specifically focuses on “experience” goods because, following the definition offered by Weathers and Makienko (2006) that experience goods are goods consumers “feel they need to directly experience” to adequately evaluate the goods’ quality, they tend to be purchased in-store and present more challenges for online shopping to substitute in-store shopping compared to “search” goods (Chocarro et al., 2013; Zhen et al., 2018). To fulfill the aim of this study, online shopping activity data for four distinct experience goods (CSWJ, BH, TKB, and HGT) were collected within the state of Florida during the first quarter of 2021 (February to April). Also, information related to consumers’ personal and household characteristics, coupled with their attitudes, were collected to provide insights into the factors affecting consumer shopping behavior.

This study contributes to the existing body of literature in several ways. First, it addresses a call for a comprehensive examination of shopping behaviors across different product types within the same product class (Zhen et al., 2018). The selected product types present varying purchase patterns both online and in-store, offering an opportunity to examine potential variations in the relationship between online and in-store shopping behavior for distinct experience goods. Furthermore, this study leverages the structural equation modeling framework to explore the mediating role of attitudes between exogenous variables and shopping behavior – an aspect scarcely addressed in the shopping and travel behavior literature, except few select instances within discrete choice modeling frameworks (Meister et al., 2023; Schmid et al., 2016; Wang et al., 2023). Lastly, acknowledging Covid-19 influences on shopping- and travel-related attitudes (Asgari et al., 2023; Jiang et al., 2023; Thomas et al., 2021), this study considers attitudes towards unattended delivery and mobility preferences, among other factors, on shopping behavior across the specified product types. The results of this study are expected to enrich our understanding of the evolving dynamics between online and in-store shopping behaviors in the wake of the pandemic-induced shifts.

2 Literature review and hypotheses

2.1 Travel effects of e-commerce

The exploration of the travel impacts of e-commerce can be traced back to the energy crisis of the 1970s when transportation researchers and engineers sought ways to improve energy savings (Mokhtarian, 1990). Information Communications Technology (ICT) was seen as a tool that could be used to substitute travel, and lead to a reduction in traffic congestion, energy, and environmental costs. It appears that ICT tended to substitute mandatory activities (e.g., work), but complement maintenance (e.g., shopping) and discretionary activities (e.g., going to restaurants) (Andreev et al., 2010). However, the research on the complementarity effects of e-commerce (or online shopping) on travel was far from conclusive, owing to the complexity of the topic (Le et al., 2021; Mokhtarian, 2004; Rotem-Mindali & Weltevreden, 2013).

Many studies have shown evidence of complementarity, that is, online shopping increased shopping travel (Cao, 2012; Cao et al., 2012; Ding & Lu, 2017; Etminani-Ghasrodashti & Hamidi, 2020; Farag et al., 2006; Hiselius et al., 2015; Ramirez, 2019; Xue et al., 2021; Zhen et al., 2016; Zhou & Wang, 2014). Contrastingly, some studies have found substitution effects (Irawan & Wirza, 2015; Shi et al., 2019; Zhen et al., 2018) and others have suggested neutrality or some form of modification effects on shopping duration, travel time, or trip distances (Hoogendoorn-Lanser et al., 2019; Kedia et al., 2019; Lachapelle & Jean-Germain, 2019). Mixed results have also been recorded by studies that examined the reciprocal effects between online and in-store shopping. For example, studies like Ding and Lu (2017); Etminani-Ghasrodashti and Hamidi (2020); and Farag et al. (2007) found that in-store shopping led to more online shopping, while other studies like Zhou & Wang (2014); and Xue et al. (2021) that analyzed the 2009 and 2017 National Household Travel Surveys (NHTS) datasets, respectively, indicated that in-store shopping frequency negatively affected online shopping frequency among U.S. residents. Among the many factors that could explain the mixed findings of past studies (e.g., the stage of the shopping process considered, specific shopping timeframe measured, the performance measures used, the research design and analysis methods, the geographical areas under study, sample characteristics, etc.), the characteristics of the shopped-for product(s) considered may be moderating the relationship between online and in-store shopping behavior.

Given this understanding, some shopping and travel behavior studies classified products into grocery versus non-grocery or daily versus non-daily products, and it appears that online shopping for non-grocery or non-daily products tended to yield complementarity effects on in-store shopping while online shopping for grocery or daily products tended to exhibit substitution or non-complementarity effects (Dias et al., 2020; Joewono et al., 2019; Suel et al., 2015, 2018). Another common product classification system that has been leveraged is the search/experience goods classification system. This classification system was first developed by Nelson (1970) based on the dominant attribute of products. Search goods were defined as goods whose essential qualities can be accurately known prior to purchase, while the qualities of experience goods can only be ascertained after purchase or during consumption. Others like Weathers and Makienko (2006) have not only argued that the search/experience goods distinction makes products channel-dependent, a modification of the definitions has also been proposed. “Experience goods”, for example, was re-defined as products “consumers feel they need to directly experience in order to evaluate the product’s quality”. That is, the greater the need to use the senses to evaluate a good, the more experience qualities the good possesses. Although products have varying degrees of search and experience qualities and may not fit perfectly into this binary classification system, products such as books, standard electronic appliances, and dried foods have been used to represent search goods, while clothing, cosmetics, groceries, and fresh foods have been used to represent experience goods (Cao, 2012; Chocarro et al., 2013; Liu et al., 2016; Schmid et al., 2016; Zhai et al., 2017; Zhen et al., 2018). Considering these products in analyzing shopping behavior within the search-experience goods dichotomy, the literature has found that online shopping intention was higher with search goods (Chocarro et al., 2013) while shopping for experience goods tended to yield more information search and product trials that lead to more shopping trips (Zhai et al., 2017). Also, the higher risk of purchasing experience goods without knowing their quality entirely predisposed shoppers to travel to the store, even when the travel time for the shopping trip was long (Farag et al., 2006; Lee et al., 2015; Zhen et al., 2018). And while some studies have examined online and in-store shopping interactions for different product types, to the knowledge of the authors, there has been no study that considered product types within the same product class (Zhen et al., 2018). Given these discussions, the first hypothesis is proposed:

  • H1. Online and in-store purchase frequencies (all four product types) have reciprocal influences.

2.2 Attitudinal Influences

In examining shopping behavior, shoppers’ attitudes, perceptions, and experiences may affect online and/or in-store shopping behavior. Expectedly, having a positive perception toward online shopping increased both online shopping intention and actual online purchasing behavior, while a positive perception toward in-store shopping increased the tendency to shop in a physical store (Farag et al., 2006; George, 2004; Irawan & Wirza, 2015). Some studies have shown that the use of and experience with the Internet or technology have a positive effect on online shopping (Cao et al., 2012; Farag et al., 2006), while others have found that Internet usage and technology experience increased the likelihood of both online and in-store purchases (Irawan & Wirza, 2015; Unnikrishnan & Figliozzi, 2020). Also, perceived risk or consequences with the use of the Internet or online websites has been negatively associated with online shopping attitude, which in turn affected both the intention and the decision to purchase online (Crocco et al., 2013; Hsu et al., 2014). Internet risk perception has also been linked with some dimensions of trust (i.e., trust in the website, the Internet, or the vendor) that likewise affected the attitude and perceived risk toward online shopping (George, 2004; Hsu et al., 2014). Valuing recreational shopping (i.e., shopping for fun or leisure) is associated with shopping behavior, as those who valued recreational shopping tend to prefer shopping in-store and are more likely to be loyal to a physical store than to an online store (Crocco et al., 2013; Lee et al., 2015; Scarpi et al., 2014). The perceived convenience of a channel, emotional and social fulfillment, lifestyle, pro-exercise, and pro-environment attitudes may affect shopping behavior (Etminani-Ghasrodashti & Hamidi, 2020; Lee et al., 2017; Mokhtarian & Tang, 2011; Titiloye et al., 2023). Moreover, Zhen et al. (2016) considered four different product types in examining the effects of novelty seeking, shopping enjoyment, cost consciousness, and time consciousness, and it was found that the attitudes had differing effects on the product types, suggesting the importance of differentiating and specifying product types in examining shopping-related attitudes.

It should be noted that these studies were conducted before the occurrence of the Covid-19 pandemic. Since shopping behavior and attitudes evolved across the Covid-19 timeline, some attitudinal effects on shopping behavior may have changed also. For example, the preference and convenience of a travel mode choice have been shown to influence shopping behavior, but Covid-19 fears negatively affected attitudes toward public transport in 2020 (Thomas et al., 2021). And although the attitude toward missing attended deliveries did not show significant effects on shopping behavior (Shi et al., 2019), the rise in the use of contactless delivery calls for its examination (Jiang et al., 2023). Therefore, the second hypothesis is proposed:

  • H2. Attitudes (related to online shopping, mobility preferences, data privacy and security risk, shopping enjoyment, technology savviness, and unattended delivery) influence online and in-store purchase frequencies.

2.3 Socioeconomic and demographic factors

Socioeconomic and demographic characteristics such as age, gender, income, education, employment, car ownership, and household characteristics are all known to influence online or in-store shopping behavior in some sort. Many studies have found a negative association between age and online shopping, with explanations ranging from the high internet usage and technology savviness of the younger generations to the leadership positions that many older individuals take in their households in purchasing a large number of items through in-store patronage (Cao et al., 2012; Crocco et al., 2013; Farag et al., 2006; Irawan & Wirza, 2015; Kedia et al., 2019; Lee et al., 2015; Unnikrishnan & Figliozzi, 2020). However, Shi et al. (2019) found a positive relationship between age and online shopping frequency. It should be noted that the findings of Shi et al. (2019) were based on a sample of relatively young individuals, and perhaps may be suggestive of a non-linear relationship between age and online shopping.

The influence of gender on online shopping behavior varied. Some have indicated males tended to make more online purchases as compared to females who tended to make more and longer shopping trips (Farag et al., 2006; Irawan & Wirza, 2015; Titiloye & Jin, 2023; Xue et al., 2021). Others like Lee et al. (2015) and Hoogendoorn-Lanser et al. (2019) have observed no gender effect on online shopping frequency. Contrastingly, Ramirez (2019) found that females tended to make not only more online purchases but more purchases in general (i.e., online and in-store) than men. Some researchers have found that distinctions by product type affected the shopping behavior of each gender. Chocarro et al. (2013) found that gender affected channel choice in purchasing search goods (e.g., books and plane tickets), but not in purchasing experience goods (e.g., T-shirts and personal computers). Lee et al. (2015) observed that females tended to shop for clothing and home supplies while males more frequently purchased electronics and sporting goods. Also, Zhen et al. (2016) demonstrated that women tended to make more frequent trips in shopping for clothing and daily goods than men, who tended to make more frequent trips in shopping for electronics.

Some studies identified positive associations between online shopping and income as well as educational levels (Cao et al., 2012; Crocco et al., 2013; Farag et al., 2006; Lee et al., 2015; Ramirez, 2019; Xue et al., 2021), but few others have suggested negative associations (Irawan & Wirza, 2015; Shi et al., 2019). Studies on the effect of employment status on shopping behavior seem scanty. However, Lee et al. (2015) and Xue et al. (2021) found that full-time workers were more likely to shop online than other employment groups. Vehicle ownership is positively linked with the frequency of online shopping (Farag et al., 2006; Kedia et al., 2019) and negatively linked with shopping trips (Irawan & Wirza, 2015; Shi et al., 2019). It may be that since those who have cars tend to be wealthier and have a higher value of time, frequent commuting to physical stores is less desirable and often replaced by online shopping. Although results seem to differ by country, studies conducted in the U.S. have indicated that car ownership is not linked with a higher likelihood of online shopping, and public transportation commuters may be more likely than others to make frequent online purchases (Farag et al., 2006; Ramirez, 2019; Xue et al., 2021).

Also, household size and other household characteristics have been found to influence shopping behavior. Some studies like Kedia et al. (2019) have demonstrated a negative association between the number of children and online shopping. However, others like Chocarro et al. (2013) have shown that household size had a positive association with shopping online, or a negative association with in-store shopping (Zhen et al., 2016). Other factors that have been found to influence shopping behavior are the number of productive family members (Irawan & Wirza, 2015), possession of a driver’s license (Xue et al., 2021; Zhen et al., 2016) or the number of household members with drivers’ license (Irawan & Wirza, 2015; Ramirez, 2019), the variety or quantity of products to be purchased (Mokhtarian & Tang, 2011), travel time or distance to store (Etminani-Ghasrodashti & Hamidi, 2020; Lee et al., 2015), spatial attributes (Motte-Baumvol et al., 2017; Zhen et al., 2018), and day or time of the week (Chocarro et al., 2013; Schmid et al., 2016; Zhou & Wang, 2014). Accordingly, the third hypothesis is proposed:

  • H3. Socio-demographics, household characteristics, and return patterns influence online and in-store purchase frequencies.

2.4 Attitudes as mediators

Although the effects of attitudes and other explanatory variables have been considered in past studies, few studies have related attitudes and other exogenous variables in their modeling framework. For example, Cao et al. (2012) considered only the direct effects of shopping attitudes, internet experience, demographics, household responsibility, and shopping accessibility on shopping behavior. Also, Etminani-Ghasrodashti and Hamidi (2020) considered various mediating variables but incorporated attitudes only as exogenous variables. Some studies that incorporated the relationship between attitude and socio-economic variables into their modeling framework did so mostly within a discrete choice modeling framework that was used to analyze stated preference channel choices, and preference heterogeneity was better explained within the models (Meister et al., 2023; Schmid et al., 2016; Wang et al., 2023). Thus, the fourth hypothesis is proposed:

  • H4. Attitudes mediate the influences of socio-demographics, household characteristics, and return patterns on online and in-store purchase frequencies for experience goods.

3 Research design

3.1 Questionnaire

The questionnaire used in this study was designed to provide insights into Florida residents’ online and in-store shopping behavior for different product types. Information in the questionnaire comprises three major sections:

  • Socio-economic, demographic, and household characteristics: Basic personal and household data collected include age, gender, income, education, ethnicity, race, marital status, employment status, housing type, household size, number of children, number of older adults, members with driver’s license, and vehicle ownership.

  • Shopping behavior: Respondents were asked to indicate the number of times they had purchased both online and in-store in the past month for four types of experience goods, respectively. The product types presented to respondents were a) clothing, shoes, watches, and jewelry (CSWJ), b) beauty and health products (BH), c) toys, kids, and baby supplies (TKB), d) home, garden, and tools (HGT). The purchase frequency had seven different levels from “never” to “more than 30 times” in the past month. In addition, respondents were asked if they had returned any purchased products (to a store, USPS, or via courier services) in the past month, by indicating Yes or No.

  • Attitudinal factors: To explore how respondents’ attitudes influence shopping behavior, respondents were asked to use a five-point Likert scale (i.e., from “strongly disagree” to “strongly agree”) to express how much they agree with each statement. These statements covered attitudes toward various aspects related to shopping, including shopping methods (online or in-store), local stores, delivery experience, social interactions, online data security, cost consciousness, time consciousness, and convenience. The respondents were also asked about their general lifestyle preferences, including technology use, mobility preferences, and environmental issues.

3.2 Data collection and sampling

The survey used in this study was created and implemented in Qualtrics, an online survey tool that has been widely used in academic research because of its advantages over other similar tools, such as a large array of options for presenting questions, translation of a survey into multiple languages, monitoring of respondents’ time spent on survey questions, among others. Due to Qualtrics’ large panel base of 200 million members from which randomly selected respondents within a target population can be reached (Qualtrics, 2023), services for survey distribution and recruitment were acquired from Qualtrics. The survey was fielded to our target audience, Florida residents who were at least 18 years old, in both English and Spanish. Respondents were randomly recruited following a quota sampling approach based on age group, gender, household income, education, race, and ethnicity to ensure that sample aggregates roughly follow the population distribution profile of the Census data for the state of Florida.

A total of approximately 4,000 responses were recorded, some of which were screened out due to ineligibility related to their ages, out-of-state status, being unsure about providing their best answers, survey abandonment, etc. Furthermore, low-quality responses relating to issues like speeding, straight-lining, inattentiveness, or duplicates were removed. Following data cleaning, the final sample size of 2,257 respondents was used for analysis.

Figure 1 illustrates the geographical distribution of respondents’ residential locations, demonstrating a well-distributed representation throughout Florida.

Fig. 1
figure 1

Geographical distribution of the sample

Full size image

Table 1 shows the comparison of the sample stratification between the study sample and U.S. Census data for the state of Florida (US Census Bureau, 2019). It should be noted that the US Census data for education is only available for people aged 25 and above, while the data for race, ethnicity and income level have no age specification. From Table 1, the sample distribution among the demographic groups generally follows the Census distributions, except for education. The discrepancy regarding those with very low education was primarily due to their low response rate and perhaps a disproportionately high number of low-quality responses that were filtered out during data cleaning. However, the large sample size can compensate for this discrepancy.

Table 1 Sampling characteristics (N = 2257)
Full size table

Figure 2 revealed that shopping for CSWJ had the highest online purchase frequency, as about 60% and 15% of respondents shopped online for CSWJ at least once and more than five times, respectively, in the past month. The product type that attracted the highest in-store purchase frequency is BH (i.e., less than a quarter of shoppers never shopped in a store for BH), followed by CSWJ. Comparing the online and in-store purchase patterns for each product type, CSWJ and TKB both showed similar online and in-store purchase patterns, while more shoppers purchased BH and HGT in a store rather than online, possibly owing to the relatively higher difficulty in adequately evaluating the quality or suitability of BH and HGT (Weathers & Makienko, 2006).

Fig. 2
figure 2

Online and in-store purchase frequencies by product type

Full size image

4 Structural equation model

4.1 Model formulation

The models that are commonly used to examine shopping behavior and its determinants are regression models (Farag et al., 2006; Kedia et al., 2019; Ramirez, 2019; Shi et al., 2019), probit models (Dias et al., 2020; Zhen et al., 2016, 2018), and structural equation models (SEMs) (Cao et al., 2012; Ding & Lu, 2017; Etminani-Ghasrodashti & Hamidi, 2020; Irawan & Wirza, 2015; Xue et al., 2021; Zhou & Wang, 2014). In this study, the relationship between online and in-store purchase frequencies for each product type was investigated using SEM—a modeling approach that has gained widespread usage because of its advantages in measuring not only direct effects but also indirect and total effects of exogenous variables (or independent variables) on endogenous variables, and within endogenous variables (or dependent variables). A variable is said to have a direct effect on another when no mediating variables are involved, while an indirect effect involves the mediation of one or more variables with another. The total effect is the sum of the direct and indirect effects. The formulation of the structural portion of an SEM can be expressed as:

$$y=By+\mathrm\Gamma x+\xi$$
(1)

where,

\(y=({M}_{y}*1)\) column vector of endogenous variables and \({M}_{y}\)= number of endogenous variables,

\(x=({M}_{x}*1)\) column vector of exogenous variables and \({M}_{x}\)= number of exogenous variables,

\(B=({M}_{y}*{M}_{y})\) matrix of coefficients representing the direct effects within endogenous variables,

\(\mathrm\Gamma=(M_y\ast M_x)\) matrix of coefficients representing the direct effects of exogenous on endogenous variables, and

\(\xi =({M}_{y}*1)\) column vector of errors.

In presence of latent variables (i.e., attitudinal factors), the SEM quantifies them based on a series of observed variables (i.e., attitudinal statements). This is usually referred to as the “measurement model” and is formulated as:

$$Q=\Pi Z+\zeta$$
(2)

where,

Q = v 1 vector of latent attitudinal factors, a subset of endogenous variables y,

Z = w 1 vector observed attitude such as the response in Likert scale,

Π = v w matrix of coefficients of the regression effect of Z on Q, and

ζ = v 1 vector of error terms.

The conceptual model for the SEM was designed to examine the relationships between online and in-store shopping frequencies, and to adequately model the complex interrelationships between the variables. It is expected that the endogenous variables (i.e., online purchase frequency and in-store purchase frequency) exhibit reciprocal effects within the one-month observational window. We specify the residual covariance between the endogenous variables since the existence of mutual causal effects may be indicative of some similar unmeasured causes for both variables. This type of model, shown in Fig. 3, is a non-recursive model with direct feedback loops.

Fig. 3
figure 3

Conceptual model design for the SEM

Full size image

As has been highlighted in the literature review section of this paper, socio-demographic, household attributes and attitudes have direct associations with shopping behavior. Thus, observed variables such as age, gender, income, education, ethnicity, race, marital status, employment status, housing type, household size, number of children, number of older adults, members with a driver’s license, vehicle ownership, and return pattern were examined as exogenous variables that affect shopping behavior. It is expected that these exogenous variables also affect people’s attitudes, and attitudes, in turn, affect shopping behavior. Hence, in addition to the frequencies of online and in-store shopping, our model considers latent attitudes as endogenous variables. That is, the latent attitudes can be predicted by exogenous variables after being measured in a measurement equation.

4.2 Model estimation

The Maximum Likelihood (ML) is the most prevalent estimation method used in most past studies that have explored online shopping behavior using SEM (Cao et al., 2012; Ding & Lu, 2017; Etminani-Ghasrodashti & Hamidi, 2020; Irawan & Wirza, 2015; Xue et al., 2021; Zhou & Wang, 2014). However, in the presence of ordinal endogenous variables, a modified approach known as the “Diagonally Weighted Least Squares” is suggested in the literature (Asgari et al., 2023; Mîndrilă, 2010; Muthén & Muthén, 2007; Rhemtulla et al., 2012; Sarker et al., 2023), and was used in this study. The DWLS approach is similar to other ordered response models in the sense that it separately converts the ordinal endogenous variables \(y\) into latent continuous variables \({y}^{*}\), and consequently calculates their associated thresholds and polychoric correlations (Bollen, 1989). The SEM model then minimizes the difference between the sample statistics and the model-implied estimates using the weighted least squares fit function \({F}_{WLS}\) (Muthén, 1984):

$${F}_{WLS}={[S-\sigma \left(\theta \right)]}^{\prime}{W}^{-1}[S-\sigma \left(\theta \right)]$$
(3)

where,

\(\theta =\) Model parameters

\(S=\) Vector of sample statistics (i.e., threshold and polychoric correlation estimates)

\(\sigma =\) Vector containing model-implied elements of Σ(θ)

\(W=\) Weight Matrix

The SEM was developed with the lavaan package in R (Rosseel, 2012). All the exogenous variables were re-coded as dummy variables to allow for the representation of their categorical nature (e.g., race, housing type, marital or employment status, etc.) and for better interpretability (Garavaglia & Sharma, 1998). The initial model included only paths between the attitudes and each of the endogenous variables. One variable at a time was added to link the attitudes and the endogenous variables so that the statistical significance of the variables and the effects of the additions on the overall model fit could be evaluated. Links that were not significant were removed, while those that were significant but materially reduced the model fit were excluded for later re-inclusion after all other links had been added.

The most recommended model fit indices for Maximum Likelihood estimation methods and their acceptable thresholds are Root Mean Square Error of Approximation (RMSEA) < 0.06, Comparative Fit Index (CFI), and Tucker-Lewis Index (TLI) > 0.950. In the case of ordinal models with DWLS estimation, the literature suggests lower (more relaxed) thresholds (Savalei, 2020; Xia & Yang, 2019).

5 Results and discussions

5.1 Interactions between online and in-store shopping

Table 2 shows the standardized direct, indirect, and total effects of the interactions between the endogenous variables for each product type, and the fit indices for each model indicate that the models performed reasonably well. Note that in a non-recursive model, an endogenous variable can have indirect effects on itself. For example, the indirect effect of online shopping on itself in the feedback loop comprises all the indirect paths from online to in-store, then to online, back to in-store, to online, and the cycling process continues until infinity. The indirect effect is thus determined by subtracting the direct effects from the total effects, which can be computed when the infinite sum of the powers of the coefficient matrices converges to zero (Bollen, 1989). A more detailed discussion on indirect and total effects within a non-recursive modeling framework can be found in Bollen (1989).

Table 2 Interactions between online and in-store purchase frequencies
Full size table

Our findings confirm the first hypothesis of this study. The results show that there were positive effects between online and in-store shopping frequency for CSWJ and TKB in both directions, indicating complementary effects. The effects of online shopping on in-store shopping frequency were larger than those in the reverse direction of influence. For BH and HGT, online shopping had positive effects on in-store shopping frequency, but no effects were found in the reverse direction. These findings are generally consistent with some studies that found online shopping frequency to exhibit a positive effect on in-store shopping frequency for non-daily or non-grocery products (Cao et al., 2012; Dias et al., 2020), and other studies that demonstrated that in-store shopping also increased online shopping in general (Ding & Lu, 2017; Etminani-Ghasrodashti & Hamidi, 2020; Farag et al., 2007).

The results indicate that online shopping did not replace in-store visits for experience goods. Individuals who often shopped online for these products were likely to also make more shopping trips. The literature suggested that these shoppers might use online shopping to not only make purchases but to also prepare for physical shopping purposes – e.g., to find out where and when to shop or experience their products before purchasing (Farag et al., 2007). Regarding the reverse effects of in-store on online shopping, a possible explanation might be that in-store shoppers may have purchased some items in the store, become attracted by some other items, and then gone back to use the Internet or store websites to find better alternatives, or similar products at a lower price or to purchase items online when stores are closed (Ding & Lu, 2017). It should be noted that no effect was found on online shopping frequency from in-store shopping for BH and HGT products, suggesting that shopping behavior does not always yield “reverse” complementary effects for all types of experience goods.

For this paper, this section mainly focuses on the CSWJ product type for a detailed discussion of the model results. The general findings were similar among the different product types, some distinctions were observed and will be discussed at the end of this section.

5.2 Influence of latent attitudes for CSWJ products

Table 3 shows the results of the measurement portion of the model (also known as confirmatory factor analysis). The model reveals six significant latent attitudinal factors. Accordingly, F1 (factor 1) measures respondents’ preferences for alternative travel modes. F2 captures concerns about the security and privacy of shoppers’ personal information at online stores. F3 measures the joy of shopping at mostly physical stores. F4 relates to attitudes toward the benefits of online shopping. F5 indicates the level of use and dependence on technology. F6 captures concern for delivery left in compounds unattended to. It is worth noting that prior to the confirmatory factor analysis (i.e., measurement model), an exploratory factor analysis was conducted to ensure that the internal consistency reliability of each construct ≥ 0.60, no cross-loadings ≥ 0.30; and Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy ≥ 0.7 (Watkins, 2018).

Table 3 Measurement model for CSWJ products
Full size table

Table 4 presents the results of the structural component of the model. In view of the impacts of the latent attitudes on shopping frequency, the model shows that those who preferred to use alternative travel modes (e.g., public transport) had a higher tendency to make their purchases online. Although mobility preferences tended to differ by study location or country, our finding agrees with the study conducted by Ramirez (2019) which showed that public transportation commuters in the U.S. were more likely than others to make frequent online purchases.

Table 4 Direct, indirect, and total effects on endogenous variables for CSWJ products
Full size table

As expected, a positive direct effect was observed from technology savviness to online shopping frequency, which complies with the overall findings in the literature (Cao et al., 2012; Farag et al., 2006). A similar positive direct effect was observed in “pro-online shopping” attitude, indicating that these amenities of online shopping (i.e., “shopping 24/7”, “finding items in high demand”, and “having a greater variety of choices”) contributed to the decision to shop online for those who found these features attractive. As expected, pro-online shopping attitude also reflects a direct negative impact on in-store shopping frequency, but it was canceled out by the positive indirect effects mediated through online shopping frequency.

The results show that those who enjoyed shopping tended to shop more frequently in-store (Cao et al., 2012; Crocco et al., 2013), which indicates that certain elements in physical shopping cannot be replaced by online shopping, especially for those who are not under any time pressure. Similarly, concerns about data privacy and security during online transactions tended to increase in-store shopping frequency, but with no direct (negative) effect on online shopping. This might indicate that privacy and security concern was not a determinant factor in discouraging online shopping, people still shop online for some of their needs even in the presence of such concerns. Past studies, however, have reported negative effects of perceived risk (and its antecedents) in using the Internet for online shopping (Crocco et al., 2013; George, 2004; Hsu et al., 2014). This may be a unique finding of this study suggesting that the COVID-19 experience may have neutralized the negative concerns on data privacy and online security. Similar to findings in the literature (Kedia et al., 2019), our model showed that those who had a negative attitude toward unattended deliveries were likely to make less frequent online purchases for CSWJ products.

5.3 Influence of exogenous variables for CSWJ

In this section, we only focus on the direct effects of the exogenous variables, which include a variety of personal and household attributes. Regarding the impact of generation (age group) on shopping behavior, Millennials tended to shop more frequently online for CSWJ products, while Gen-Z showed a direct positive effect on in-store shopping frequency. Also, older baby boomers were less likely to shop in-store, while the effects of younger baby boomers on shopping at both channels were insignificant. Although overwhelming evidence in the literature suggests that younger generations tended to shop more frequently online (Cao et al., 2012; Crocco et al., 2013; Farag et al., 2006; Irawan & Wirza, 2015; Lee et al., 2015; Unnikrishnan & Figliozzi, 2020), Gen Zers’ frequent in-store shopping for CSWJ may owe to the fact that they are less likely to be working full-time and would tend to have lower time-pressure than the Millennial generation. Concerns about the high transmissibility and fatality of Covid-19 among older adults may explain older baby boomers’ negative effect on in-store shopping frequency. We also found that blacks or African Americans tended to shop more frequently online than other racial groups, which is not at odds with past studies that suggested that, in shopping for food or groceries, blacks tended to be more oriented towards online shopping than other groups (Kim & Wang, 2021; Mercatus, 2021).

Our model also indicates a direct negative impact of one-person households on online shopping frequency, which consequently results in an indirect negative impact on in-store shopping. This might be reasonable in the sense that living alone reduces shopping needs in general. In addition, one can see that those who experienced at least one product return in the past month were more likely to shop online, probably due to the free-return policy implemented by many online shops. Since shoppers are not usually penalized for returning purchased items, customers can risk buying multiple items they like, test or experience them, keep the ones they like and easily return the rest. And in turn, they go online and purchase more items.

There are some variables in the model (shown in Table 4) that did not have significant direct effects on shopping frequency but had significant indirect effects mediated through attitudes, e.g., gender, income, education, employment, marital status, housing type, and household characteristics. Homemakers’ shopping behavior was mostly affected by their joy of shopping. Less concern for unattended delivery was associated with highly educated individuals, those from wealthy households, and those living in detached single houses, which increased their tendency for online shopping. Household characteristics contributed to varying attitudes, which in turn influenced their shopping frequencies. Figure 4 shows the path diagram for the CSWJ model, and the next section provides a detailed examination of these indirect effects mediated by attitudes.

Fig. 4
figure 4

Path diagram for CSWJ

Full size image

5.4 Indirect effects for CSWJ

Table 5 shows the effects of the exogenous variables on the latent attitudes, which in turn influenced individuals’ shopping behavior. It is observed that the effect of gender on shopping behavior was mediated by various attitudes. Females tended to enjoy shopping, value the benefits of online shopping (i.e., shopping 24/7, having more choices, and finding items in high demand), do not prefer alternative travel modes, and have less concern about unattended delivery.

Table 5 Effects on Mediators (CSWJ)
Full size table

The positive relationships between female and pro-online shopping and the joy of shopping suggest that females shopped more frequently both online and in-store than males, but due to different underlying reasons. Also, females’ negative attitude toward alternative travel modes mediated their online shopping frequency, and their lesser concern toward unattended delivery may be encouraging more online shopping. Thus, the results suggest that females had indirect positive effects on both online and in-store shopping for CSWJ products, which is supported by the literature (Ramirez, 2019).

Gen Zers and Millennials were alike in their preference for alternative travel modes and technology savviness, but Millennials tended to have less data privacy and security concern. Baby boomers’ lower technology savviness might have predisposed them to less online shopping. Our analysis also indicates that those living in households with income levels lower than $15 k tended to enjoy shopping, while those with higher income levels (between $100 k and $150 k) tended to be more tech-savvy and not as much concerned about unattended delivery compared to other income groups. These led to higher in-store shopping frequency for the low-income groups and higher online shopping frequency for the higher-income groups, albeit both groups showed positive indirect effects on both online and in-store shopping frequency. On the other hand, individuals with low-to-medium income ($35 k to $50 k) were less likely to be pro-online shopping, which resulted in negative indirect impacts on both online and in-store shopping frequencies.

Education did not show any direct effects on shopping frequency, but the indirect effect of education revealed that frequent online and in-store shopping were mediated through different attitudes. The tendency of those with lower levels of education (i.e., having an associate degree or less) to shop online less frequently is mediated by their low-tech savviness. Also, post-graduate degree holders’ lower in-store purchase frequency compared to their online purchase frequency was mediated by their less concern for unattended delivery. These findings are consistent with the literature (Cao et al., 2012; Lee et al., 2015; Ramirez, 2019; Xue et al., 2021). Regarding employment, nonworkers (i.e., homemakers, retired, and currently unemployed) were more likely to have attitudes that lead to more in-store shopping and less online shopping for CSWJ products. For example, homemakers tended to enjoy shopping; retired individuals tended to be less tech-savvy and did not prefer alternative travel modes, while unemployed individuals were more likely to be concerned about unattended delivery.

Households with three or four members tended to be more pro-online shopping and prefer alternative modes, respectively, which encouraged frequent online shopping for CSWJ products. This finding not only makes sense considering that our initial result indicated that living alone tended to negatively affect shopping online, but it is also consistent with previous studies that found that the number of children (or family size) increased the likelihood of online purchases or reduced the likelihood of in-store purchases (Chocarro et al., 2013; Zhen et al., 2016). Also, those without vehicles in the household tended to prefer alternative modes, which positively affected their online shopping frequency. A similar effect is observed with those in households where no member had a driver’s license. However, the positive impact of living with no driver’s licensed member on data privacy concerns resulted in a slightly higher in-store frequency compared to online shopping. Viewing two-vehicle and three-or-more-vehicle households, the results are mixed. In the former case, negative indirect effects are observed on both online and in-store shopping, mediated through the negative effect on preferences for alternative travel modes, with higher frequencies for in-store shopping. For the latter, our model shows a positive effect on pro-online shopping, resulting in higher frequencies of online shopping compared to in-store shopping activities.

Housing type is another variable that showed significant effects on attitudes. Accordingly, those who lived in two-to-four-bedroom apartment units were more likely to prefer alternative travel modes, enjoy shopping, and have higher data privacy/security concerns. This might stem from their land use and built environment attributes, or perhaps correlated with their lifestyle patterns. Additionally, those living in townhouses were likely to enjoy shopping, while those living in single-detached houses were less likely to be concerned about unattended deliveries. The insignificant direct effects found between housing type and shopping behavior may be suggesting that online shopping behavior is not affected by built environment characteristics when shopping for non-grocery experience products (Beckers et al., 2018; Lee et al., 2017).

5.5 Highlights for other product types

For the purpose of comparison, Table 6 shows the direct effects of the explanatory variables on the endogenous variables for each of the four product types. The effects of attitudes were very similar: pro-alternative travel modes and pro-online shopping were both associated with online shopping frequency, while the joy of shopping and data privacy/security concerns led to frequent in-store shopping for all four product types. However, technology savviness and concerns for unattended delivery influenced online shopping for CSWJ products only and did not show significant effects for other product types. Also, individuals 56 years or older tended to shop less frequently in-store (though not all effects were significant). Again, the higher fatality rate of Covid-19 among older individuals may have discouraged them from shopping in-store.

Table 6 Direct effects of explanatory variables for all product types
Full size table

There were some distinctions in the effects of the socio-demographic characteristics. For example, females showed a positive (significant) direct effect on online shopping frequency for only BH products. Also, working full-time showed a positive (significant) direct effect on online shopping frequency for only HGT. These distinctions suggest that, while the common use of “clothing” to represent experience goods in measuring shopping-travel effects is indeed applicable to other experience goods, it does not capture some important distinctions in the effects of the explanatory variables (especially socio-demographic characteristics) among different types of experience products.

6 Policy implications

The findings in this study carry significant implications for transportation and urban planning. The complementarity effects found suggest that as e-commerce becomes more attractive to consumers through potentially improved retail services, the continuous growth of e-commerce post-pandemic could potentially exacerbate traffic congestion and transportation challenges. To mitigate these effects, transport planners and policymakers should implement measures that encourage off-peak purchases and deliveries, such as congestion pricing and time slot allocation approaches. Urban designers and planners should focus on integrating shopping areas with diverse land use types, including residential neighborhoods and office spaces. Furthermore, promoting carpooling and incentivizing retail companies that offer subsidized prices for group shopping can enhance shopping efficiency and minimize the need for separate shopping trips. Given that the results concerning preferences toward the benefits of online shopping showed a positive effect on online shopping and a negative effect on in-store shopping frequency for CSWJ, retailers should emphasize specific advantages of online shopping like 24/7 shopping, finding items in high demand, and having a variety of choices. This might offer more transportation benefits rather than promoting e-commerce in general.

The study also highlights pertinent implications for the retail sector. In cases like BH and HGT, where online shoppers tend to frequently shop in-store but in-store shoppers rarely shift to online channels, retailers should enhance in-store shopping experience and customer engagement with investments in store layouts and interactive technology, along with exclusive in-store promotions that incentivize in-store visits. For CSWJ and TKB, adopting an omnichannel approach can foster cross-channel shopping. Also, since concerns about unattended delivery discouraged online shopping for CSWJ, retail strategies that adopt a hybrid system where both the vendors and the customers ensure the safe delivery of their ordered products may encourage more online shopping. The use of parcel boxes for delivery is suggested especially for boomers, unemployed individuals, and others who have unattended delivery concerns. Furthermore, the positive effects of privacy and security concerns on in-store shopping implies the need for retail companies to emphasize their privacy policies in a clear and easily understandable fashion to cultivate trust and loyalty among online customers. While there were distinctions in the effects of explanatory variables on shopping behavior across the four product types, it appears that older adults (above 55 years) tended to shop less frequently in a store, perhaps due to relatively higher Covid-19 health concerns. Thus, there is a potential market of e-commerce for older adults, as online retail companies could consider strategies that help older adults overcome obstacles associated with tech savviness, such as simplifying shopping websites, offering different navigation options, using larger font sizes, and outreach campaigns, etc. This could also help to alleviate equity concerns that may arise with boomers, low-educated individuals, and others who are less tech-savvy, as the increasing e-commerce usage and delivery demand gradually lead to a replacement of some of the traditional stores.

7 Limitations and recommendations

While this study provides valuable insights, there exist limitations that should guide future research endeavors. First, the results of this study focused on and are limited only to experience goods. Future research should explore other product classes (e.g., search goods) to comprehensively grasp online and in-store shopping dynamics both across and within diverse product classes. Also, the products considered may have varied degrees of search and experience qualities within them, thus oversimplifying the characteristics of the products. For example, some products within the BH product type like toothpaste and deodorants may have more search qualities, compared to cosmetics and fragrances. Future research should consider products with more distinct characteristics to better comprehend how product characteristics may impact shopping behavior. Moreover, regulations on shipping products containing liquids, alcohol, and flammable materials may further complicate shopping behavior and should be explored in future research. Lastly, the influences of spatial attributes and store accessibility on shopping behavior were not considered in our study. There is a need for future studies to account for shopping behavior variations in locations with different urbanization patterns and built environment characteristics.

8 Conclusion

The initial closure of non-essential brick-and-mortar stores at the outset of the COVID-19 pandemic prompted a shift in consumer behavior towards online shopping. Although the subsequent advent of vaccines in early 2021 restored some confidence in physical shopping, the newfound preference for online shopping persisted. This evolving shopping landscape necessitated an examination of the reciprocal interplay between online and in-store shopping, particularly for non-essential experience goods that were largely inaccessible during pandemic-induced lockdowns.

In our investigation of the interaction between online and in-store purchase frequency, we distinguished non-essential experience goods into four distinct types (CSWJ, BH, TKB, and HGT) to account for potential variations in outcomes. Upon controlling for various socio-demographic and attitudinal factors, our findings showed significant complementarity effects between online and in-store shopping, indicating the propensity of shoppers to exhibit cross-channel shopping when shopping for experience goods. However, for BH and HGT, the reciprocal relationship was asymmetrical, with in-store shopping showing no substantial effects on online shopping.

Additionally, personal attitudes were influential determinants of shopping behavior, acting as significant mediators for individual and household attributes. Consistent patterns regarding the influences of attitudes emerged across the product types, where a preference for alternative travel modes and pro-online shopping heightened online purchases, while the joy of shopping and data privacy/security concerns were the two main drivers of in-store purchases for all four product types. In contrast, tech savviness and unattended delivery concerns respectively encouraged and discouraged online shopping frequency for CSWJ.

It should be noted that the pro-online shopping attitude examined in this study encapsulates factors that motivate the use of online shopping, distinct from those that discourage in-store shopping like “avoid crowds” and “avoid going to stores” during the pandemic. Although both discouraging factors were assessed, neither were retained in the factor analysis, implying that the effect of the observed pro-online shopping attitude is not related to the pandemic and is therefore more likely to persist beyond pandemic circumstances. Moreover, the lingering concerns about online privacy and security do not appear to be decisive factors that discourage online shopping. This could be attributed to the situation during the pandemic, where social distancing restrictions forced people to shop online despite these concerns, and the online shopping experience during the pandemic may have neutralized the negative impacts of privacy and security concerns.

Ultimately, this study offers important insights into the complex dynamics between online shopping and in-store shopping across distinct non-essential experience goods. The observed complementarity effects suggest that transport planners should not expect e-commerce to significantly reduce shopping trips as far as non-essential experience goods are concerned. Furthermore, while there was consistency in the effect of online on in-store shopping across the product types, the variations in the “reverse” relationship of in-store shopping on online shopping suggest that distinct types of products, even within the same class (experience goods, in our case), contribute to heterogeneity in the relationship between online and in-store shopping. Lastly, the factors found to affect consumer demand (online, in-store, and both channels) should be leveraged by retailers and transport planners, informing the development of effective retail and marketing strategies and traffic-mitigating measures.