Introduction

1For a number of years, the scientific literature has been looking at the relationship between digital technology and day-to-day household practices (work, shopping, leisure, etc.), particularly with respect to mobility (Aguiléra et al., 2012; Athanasiadou & Theriou, 2021; Ozbilen et al., 2021). The debate initiated by P. Mokhtarian (2002, 2004) regarding complementarity, substitution or neutrality impacting digital technologies on shopping practices is still underway, partially because of the complex processes involved (Zhou & Wang, 2014). The rapid growth of online shopping coupled with the increasing availability of products and delivery solutions, means that scientific research is constantly playing catch-up. In addition, the Covid-19 pandemic led to an unprecedented boom in e-commerce. As the volume of deliveries increased, retailers and their delivery operators introduced a range of delivery services for consumers, from home delivery, fast delivery including next-day and even same-day service to parcel lockers, in-store pick-ups and click-and-collect options for food or other goods such as clothes and electronics. Today, in many countries across the world, almost every category of product can be bought online and either collected at pick-up points of different kinds or delivered directly.

2This means that the relationship between online shopping and individual mobility practices can be looked at not only in terms of the frequency of online purchases but also in terms of the type of delivery method chosen (Aziz et al., 2022). As described by Mokhtarian (2002; 2004), delivery service choice indicates a household’s willingness to replace its shopping trips by delegating them to other people. The outsourcing is total in the case of home delivery, but only partial when out-of-home collection is the option chosen. For food pick-ups (by car or on foot) or in-store click-and-collect services, there is no delegation or substitution, since people still travel to a shop. Out-of-home collection points are often cheaper for households than home delivery, though delivery costs are falling (Dablanc et al., 2017). For online retailers and delivery operators, the aim of out-of-home collection points is to cut delivery costs by reducing the size of delivery rounds (several parcels left at a single point) and minimising the costs that arise when customers are not at home (several passes for the same parcel) or delivered goods are damaged, lost or stolen (World Economic Forum, 2018). On the whole, out-of-home collection points and their logistics solutions tend to develop in urban or high-density areas, where demand for e-commerce is likely to be higher (Buldeo Rai et al., 2022). Conversely, several studies have highlighted the potential contributions of e-commerce in mitigating accessibility issues in areas with few shops (Motte-Baumvol, et al., 2017). In particular, grocery e-commerce is seen as a way to tackle the problem of ‘food deserts’ (Newing et al., 2022; Shannon, 2016), meaning that e-commerce could also offer a solution for better supply to households in areas with few or no shops.

3These two contrasting hypotheses – on the one hand that online shopping benefits households in less densely populated areas, and on the other that online shopping is essentially an urban practice – refer directly to arguments about the socio-spatial diffusion of e-shopping. As formulated by Anderson et al. (2003), the first proposition, known as the diffusion of innovations hypothesis, argues that residents of highly urbanised areas are more likely to engage in online shopping than those living in less densely populated areas. This argument is based on the view that technological diffusion and internet adoption are primarily urban phenomena. People living in densely populated and high-amenity areas tend to have social characteristics, particularly in terms of education, which increase the likelihood of their buying online. The second proposition, known as the accessibility or efficiency hypothesis, argues that households in low-amenity areas are more inclined to shop online in order to compensate for their relative lack of access to physical stores. The diffusion and efficiency hypotheses did not take delivery services into account, as the only available option in low-density areas was considered to be home delivery. As internet access has become widespread, e-shopping has significantly increased and there is a wider range of delivery services. As a result, diffusion and efficiency claims need to be examined in conjunction with the availability of delivery services. For instance, the quality of delivery services in terms of price or speed significantly influences consumer choices (da Silva et al., 2019). Reliability is also a factor (Bhalerao & Gujar, 2019) and the failure rate of home deliveries is a major concern for consumers and impacts their preferences for out-of-home collection services such as click-and-collect (Oliveira et al., 2019; Buldeo Rai et al., 2020; Buldeo Rai et al., 2021).

4While the social roots of the diffusion hypothesis may be subject to question as e-commerce has become a mass phenomenon, another possible driver might be the fact that urban areas are the ‘easiest’ locations to deliver to. They are typically the areas of maximum population, offer numerous delivery services and are characterised by shorter delivery distances (Morganti et al., 2014). While home delivery might appear more challenging in lower density areas, it is facilitated by the kinds of households located there (Belton Chevallier et al., 2016). Moreover, different delivery services, such as click-and-collect for everyday consumer goods (ECG), have gained popularity in these areas where household car usage is prevalent (Pernot, 2021). A range of variables are employed to characterise the spatial dimension of online shopping practices, including indicators for access to shops and transport services, housing types, density levels, and public statistical zoning (Farag et al., 2007; Ren & Kwan, 2009; Cao et al., 2013; Zhou & Wang, 2014; Loo & Wang, 2018; Cheng et al., 2021; Shi et al., 2023). Although the accessibility of different delivery solutions is less commonly considered, existing research emphasises its significance. For instance, the quantity and diversity of delivery solutions in a neighbourhood positively impact the frequency of online shopping and consumer satisfaction with delivery services (Xiao, Wang, Liu, 2018). Furthermore, the presence of pick-up points along typical household routes increases the use of this delivery method (da Silva et al., 2019).

5While the determinants of online shopping have been extensively studied, the same cannot be said for the determinants of product delivery services. Although an increasing number of studies consider or acknowledge the existence of multiple delivery services, most focus primarily on home delivery, which continues to be seen as the dominant mode despite its declining market share relative to pick-up points and other out-of-home collection services, in France, Europe (FEVAD, 2022), and worldwide (Precedence Research, 2021). Ultimately, what are the socio-spatial determinants of the choice of online shopping delivery services? To what extent do they coincide with the determinants of online shopping, and with the efficiency or diffusion of innovations hypotheses outlined above?

6Based on a quantitative questionnaire-based survey conducted in France in 2016 with a representative sample of 633 households, this article discusses the factors that explain delivery solution choices (home or pick-up) for online purchases of physical goods, distinguishing between everyday consumer goods (ECG) and other products. Socio-spatial characteristics are assessed by the usual quantitative indicators (age, gender, number of children, income, urban area, etc.). They are also measured subjectively through respondents’ perceptions of their neighbourhood and their previous experiences of home delivery. This article consists of 5 parts. Following this introduction, we describe the data and variables and then the modelling approach employed. The last two sections present the results of different models, leading into the conclusion with a discussion of the possible consequences in terms of delivery services and the mobility practices associated with them.

Data and variables

The LivMob data

7In this article, we used the data from the survey entitled “Parcel deliveries and e-consumer mobility: characterization of practices and flows” (LivMob). The survey was conducted within the framework of the ELIPSS (Longitudinal Internet Study for the Social Sciences) panel run by the Centre for socio-political data (CDSP), Sciences Po/CNRS, a representative panel of the adult population resident in metropolitan France. The panel members are randomly selected and are provided with a tablet and a mobile Internet subscription. Every month, they respond to surveys drawn up by researchers. An annual survey of respondents is conducted to collect and update the panel’s demographic and socio-economic data.

8The LivMob survey, conducted in 2016 with 633 respondents, focused on online and in-store shopping practices. The first part of the questionnaire dealt with the purchase of everyday consumer or e-grocery goods, i.e. food, hygiene and cleaning products. The second part dealt with other purchases, which we will refer to here as online shopping (excluding e-grocery goods). The survey had some biases. While the individuals recruited were aged between 18 and 79, the way the survey is conducted, using a tablet, meant that respondents had to have some familiarity with ICT. Apart from this recruitment bias, 40 people had to be removed from the database because they did not provide any sociodemographic information, which meant that their responses could not be processed. The final sample consisted of 593 individuals. Based on the weighting provided by the Centre for Socio-Political Data (Sciences Po/ CNRS), this sample of 593 can be considered representative of the French population aged between 18 and 79, familiar with ICT and resident in mainland France, though the weighting was calculated on the basis of the initial sample of 633. Finally, 58 individuals did not answer the question about their income, so income values were allocated by Multiple Correspondence Analysis (MCA) based on the other available sociodemographic variables. The sample is presented in Table 1. The sampling unit is the individual. However, since everyday consumer goods are bought and consumed at household scale, the unit of analysis is the household.

9The LivMob data may appear relatively old and geographically limited. Nevertheless, as well as being representative, they have the advantage of including several types of goods that can be purchased online (ECGs or e-grocery and other online purchases), and of offering different delivery solutions for each of these types of goods. Although the types of goods that can be purchased online (in particular ready meals with different applications) and the types of delivery solutions (in particular lockers) have since become more diverse and prevalent, these services have been available and used in France for longer than elsewhere. In fact, since the 1980s and well before the US or the UK (for example), remote purchase pick-up points have existed in large numbers in France. Pick-up brands and networks then continued to develop with e-commerce in the 2000s (Morganti et al., 2014), again well before similar solutions were implemented in other developed countries. We therefore believe that analysis of the LivMob data is still relevant in shedding light on how online shopping practices vary according to the type of goods purchased and the delivery solutions chosen (home delivery vs. out-of-home collection).

Sociodemographic and spatial factors of online shopping and delivery services

10Online shopping frequency and delivery service choice may be governed by the objective characteristics of the respondents but they are also influenced by their subjective experience of e-shopping and delivery services as well as their perception of their residential location. The factors we used for our analyses are based on those used in previous studies on online shopping and delivery services. First of all, online shopping practices depend on the types of products or services considered, tangible or intangible, frequent or occasional consumption, light or bulky items, and so on (Jindal et al., 2021; Lee et al., 2017). That is why food products (e-grocery) and non-food products (online shopping) are separated in our model.

11Secondly, the sociodemographic characteristics of the households play a significant role as they are reflected in their lifestyles, albeit imperfectly. For instance, income is one of the explanatory factors for the propensity to make online purchases across all types of goods. Higher income inherently facilitates purchases in general, including online purchases (Spurlock et al., 2020). Higher income may also be associated with a greater likelihood of belonging to more educated population segments or professions more inclined to use digital technologies (López Soler et al., 2021). Gender may be an explanatory variable for specific types of purchases, such as food or family products (Bhalerao & Gujar, 2019; Handayani et al., 2020; Saphores & Xu, 2021). For example, households containing women, particularly working women with young children, tend to make more intensive use of e-grocery or online shopping. Age, which is still reflected in differences in the generational adoption of digital practices, tends to be associated with lower overall use of online shopping for all types of goods (Lee et al., 2015). However, for specific types of goods (such as clothing, digital products) or purchasing channels (like CtoC), younger buyers are more likely to make online purchases (Ladhari et al., 2019). Age also influences the choice of shopping method, i.e. computer, tablet, cell phone or in-store (Dorie & Loranger, 2020). However, the impact of sociodemographic factors in the choice of delivery method has scarcely been investigated in the literature, despite findings that suggest that individuals in China aged between 19 and 59 exhibit higher demand for pick-up services in China (Luo et al., 2022), and that delivery costs play a crucial role in the trade-off between pick-up and home delivery (da Silva et al., 2019). Most of these sociodemographic factors are included in our analysis as described in Table 1.

12Thirdly, as mentioned in the introduction, the spread of e-shopping may also be influenced by spatial factors, although – as pointed out by Song (2022) – research on the geography of e-shopping is sparse. Several studies such as Cao et al. (2013) or Clarke et al. (2015) show that internet users in urban areas with easy access to shops are more likely to shop online than others. They tend to buy more because they are better educated or are higher earners. Nevertheless, low access to shops in low-density areas also stimulates e-shopping, which is increasingly practised in rural areas, especially e-grocery. Other studies, such as Beckers et al. (2018), note that online shopping overall is linked to social and economic characteristics (male, 30 years old, well-educated and high-income) and that urbanisation level does not have a significant impact on e-shopping, at least in Belgium. So the geography of demand for online shopping may vary between urban and rural areas according to income. Other research such as Farag et al. (2006) also shows that e-grocery shopping in the Netherlands is more common in urban areas than in rural areas. For this reason, our analysis considered two spatial factors to assess their effects on e-shopping: housing type in the area of residence, and the size of the urban unit. More precise data such as the municipality of residence were not available. We will see from the subsequent description of latent variables that more subjective factors concerning access to shopping have also been considered.

Table 1: Description of the sociodemographic variables of LivMob households

	Variable	Modality	N	% in column
Women	Employment	No women or not in employment	95	16.1
		In employment	498	83.9
	Profession	No women, not in employment or other profession	330	55.7
		Employer and own-account workers	22	3.7
		Professionals and managers	99	16.7
		Intermediate employees	142	23.9
	Degree	No women	76	12.8
		Pre-Baccalauréat	186	31.4
		Baccalauréat ⴕ	108	18.2
		Two-year technical or university degree	96	16.2
		Bachelor’s degree or more	127	21.4
Men	Employment	No men or not in employment	139	23.5
		In employment	454	76.5
	Profession	No men, not in employment or other profession	330	55.8
		Employer and own-account workers	33	5.6
		Professionals and managers	127	21.4
		Intermediate employees	126	21.2
	Degree	No men	124	20.9
		Pre-Baccalauréat	180	30.4
		Baccalauréat	84	14.2
		Two-year technical or university degree	88	14.8
		Bachelor's degree or more	117	19.7
Household	Couple with child	Yes	220	37.1
	Household Head Age	24 or less	9	1.5
		25 to 29	21	3.5
		30 to 34	41	6.9
		35 to 39	58	9.8
		40 to 44	71	12
		45 to 49	72	12.1
		50 to 54	67	11.3
		55 to 59	78	13.2
		60 to 64	54	9.1
		65 to 69	80	13.5
		70 and above	42	7.1
Residential location	Urban Unit Size	Rural	149	25.1
		Urban area 2,000 to 4,999 residents	20	3.4
		Urban area 5,000 to 9,999 residents	36	6.1
		Urban area 10,000 to19,999 residents	34	5.7
		Urban area 20,000 to 49,999 residents	42	7.1
		Urban area 50,000 to 99,999 residents	49	8.3
		Urban area 100,000 to 199,999 residents	33	5.6
		Urban area 200,000 residents and more (except Paris)	143	24.1
		Paris urban area	87	14.7
	Housing type in the area	Scattered housing	134	22.6
		Unscattered housing	255	43
		Urban apartment buildings	84	14.2
		Mixed and other	120	20.2

No women (respectively no men) means that there is no adult woman (respectively no adult man) in the household
ⴕ Baccalauréat is the french secondary school diploma or high-school degree

Latent variables to further our understanding of online shopping delivery choices

13Because some of the phenomena we seek to explain in the model are poorly represented by a single observed variable, we developed 4 latent variables. Latent variables are used to identify a phenomenon that is not directly observed or observable, but can be measured by a combination of observable variables. An exploratory factor analysis (EFA) was conducted to identify observable variables that could be used to construct latent variables. The EFA explored the underlying correlations between the different observable variables and defined those that could be combined to construct one or more latent variables. The Kaiser-Meyer-Olkin (KMO) test and Bartlett’s test, which both produced values of more than 0.7, were used to identify variables that were suitable for factor analysis (Fabrigar & Wegener, 2011). The different latent variables were then identified respectively by EFA, with a value of rotating components higher than 0.5 as recommended to constitute a factor. All latent variables, i.e. factors, were recognised as reliable, with a Cronbach’s Alpha value greater than 0.65 (Hair et al., 2009).

14The first latent variable is the respondents’ perception of the lack of amenities in their neighbourhood (or NPLA). Three binary variables are used. The first highlights the lack of shops perceived by 22% of the sample, while the second covers the perception of the lack of meeting places and facilities in 10% of our sample. Finally, the third refers to the neighbourhood’s isolation perceived by 11.5% of the sample. Thus, the higher the value of the NPLA variable, the more negative is the respondents’ opinion of their neighbourhood of residence, revealing a perceived lack of amenities (in the broad sense). By considering the more subjective or perceived characteristics of the household’s territory of residence, this variable allows us to test whether the perceived lack of amenities in a neighbourhood generates online purchases or is associated with the choice of one delivery method rather than another. The NPLA variable is particularly important because it helps to measure individuals’ perceptions of their neighbourhoods or regions, as highlighted by the Uncertain Geographic Context Problem (UGCoP) (Chen & Kwan, 2015; Kwan, 2012). For instance, Arranz-López et al. (2022) demonstrated that the use of e-commerce for various types of goods (books, clothing, food) influences individuals’ perceptions of their neighbourhood’s accessibility. Conversely, the perception of the neighbourhood and its subjective characteristics can also impact individuals’ adoption of e-commerce and their choice of different delivery solutions, as online shopping practices and in-store shopping practices mutually influence each other (Dias et al., 2020).

15Two other latent variables specific to online shopping practices were constructed: intensities of online e-grocery purchases (as ECG) on the one hand and intensities of other types of online purchases (except e-grocery) on the other hand. They are constructed using ordered variables relating to frequency of purchase: the date of the last order, the history of online purchasing or the share of internet purchases in total purchases, and finally the probability of shopping on the internet (see Table 3 below). If the respondents declare that they often (every month or week) buy a certain type of product online, if they declare that they will buy more or as much in the future and if they have recently ordered goods, they are considered to be intensive users of online shopping.

16The fourth and final latent variable corresponds to the experience of incidents affecting home delivery of goods bought online, excluding groceries. Four binary variables correspond to incidents such as delivery no-show and leaving the package with a neighbour and/or at a pick-up point. Each of these incidents was reported by around a third of the respondents. In the end, the higher the value of this variable, the more delayed or cancelled deliveries respondents will already have experienced.

17As we will see later, these latent variables (whose construction is summarised in Table 3 below) can play an important, even central role in explaining the choice of delivery services for goods bought online (home delivery or out-of-home collection).

Table 3: Latent Variables (derived from the survey variables)

Latent Variable	Observed Variable	OV Modality	Type		Std.all Model home delivery		Std.all Model pick-up point
NPLA	Lack of shops	Yes	binary		0.758		0.758
	Lack of meeting places and facilities	Yes	binary		0.800	***	0.815	***
	Isolation	Yes	binary		0.907	***	0.892	***
Online Grocery Purchase Intensity	Online grocery frequency	“Once a week or more“ “Once a month or more“ “Less than once a month“ “Never“	ordered	0.689			0.796
	Last order date (e-grocery)	“During the last 15 days“ “During the last month“ “During the last 3 months“ “During the last 6 months“ “Over 6 months ago“	ordered	-0.903		***	-0.952	***
	Likelihood of e-grocery shopping in the next months	“Yes, as often as today“ “Yes, but more often“ “Yes, but less often“ “No“	ordered	-0.961		***	-0.913	***
Online Shopping Intensity	Online shopping (except e-grocery)	“Once a week or more” “Once a month or more” “Less than once a month” “Never”	ordered		0.886		0.888
	When did you start buying online?	“Less than 1 year ago“ “Between 1 and 3 years ago“ “Between 4 and 5 years ago “ “More than 5 years ago ago “	ordered		0.660	***	0.657	***
	Last order date (online shopping)	“During the last 15 days” “During the last month” “During the last 3 months” “During the last 6 months” “Over 6 months ago”	ordered		-0.848	***	-0.85	***
Home-Delivery Experience	The parcel was dropped off at a neighbour’s	Yes	binary		0.680		0.653
	The parcel was dropped off at a pick-up point	Yes	binary		0.574	***	0.637	***
	The parcel was dropped off at a post office	Yes	binary		0.791	***	0.773	***
	No one was home when the delivery arrived	Yes	binary		0.729	***	0.724	***

Data: LivMob 2016, Tools: R with the Lavaan package (Rosseel, 2012)
Signif. codes: 0 “***” 0.001 “**” 0.01 “*” 0.05 “.” 0.1
All values of the estimates are standardised and allow comparisons between variables.

Theoretical framework of model and model fit

18We used Structural Equation Modelling (SEM) with latent variables. SEM is a very general framework for modelling, in which equations are estimated as a system. The two models tested and examined propose the same five nested levels of effects (Figure 1 below). The advantage of a nested SEM model in avoiding multicollinearity lies in its structured organisation of variables into hierarchical levels, which reduces correlations between exogenous variables within each level. Hence, the first level aims to take better account of household income as a variable in the model in order to avoid problems of multicollinearity, by considering the effects of household composition, in particular for dual-income couples, and the effects of the spouse’s occupation type (if applicable). The second level aims to take better account of the perceived lack of neighbourhood amenities. The third level seeks to explain the intensity of online shopping. Because the relationship between “online grocery intensity” and “online shopping (other than grocery) intensity” is potentially bi-directional, we decided to allow the error terms in the equations for the two latent variables to be correlated. The fourth level aims to analyse the home delivery experience, i.e. problematic delivery experiences. Finally, the fifth and last level concerns the mode of collection of grocery and shopping deliveries, i.e. pick-up point (click-and-collect for groceries and pick-up collection for other purchases) or home delivery. For each of the intermediate levels (A to E), the variables explained are also used as endogenous explanatory variables for the following levels, either directly or indirectly by mediation effects.

19Two models were estimated with limited differences to the variables used in the 5th level (F and G). Indeed, despite the possibility of allowing the error terms in the equations to be correlated, the inclusion of the pick-up and home delivery modes in the same model generated significant multicollinearity problems. We therefore propose a model for deliveries to out-of-home collection points (pick-up points or food click-and-collect) and another model for home deliveries. Overall, in the Structural Equation Modeling (SEM) process, the model was carefully constructed with an emphasis on including only statistically significant variables. Variables that did not demonstrate statistical significance were excluded, ensuring that the model remains streamlined, theoretically sound, and interpretable. In SEM, the selection of variables is a critical step that directly influences the model’s fit, validity, and robustness. This careful selection helps to maintain model parsimony and to avoid the pitfalls of overfitting, which can obscure meaningful relationships and reduce the model’s generalisability.

20The coefficients of the two models presented in Section 5 were estimated using the lavaan R package (Rosseel, 2012). The estimation method used below is the Weighted Least Squares (WLS) method as it is best suited to dealing with binary and ordered variables and to estimating indirect effects between two variables via one or more other variables (Golob, 2003). Since WLS uses correlation matrices, the resulting coefficients are standardised, which facilitates comparison between coefficients in a given equation and across equations.

Table 2. Model goodness of fit indicators

	Home delivery model	Pick-up point model
cfi	0.997	0.990
chisq	295.947	370.432
npar	88.000	82.000
rmsea	0.008	0.016
srmr	0.060	0.065
Tli	0.999	0.996

21The estimation of the two models outlined in Figure 1 (see Table 2 above) produces a model with very good scores on the GOF indicators that are commonly used in SEM analysis (Golob, 2003). The root mean square error of approximation (RMSEA) and the standardised root mean square residual (SRMR) are both clearly below 0.05, which is judged satisfactory. Moreover, the comparative fit index (CFI), which compares the proposed model with an unrestricted base model, exhibits a value of 0.95 (a good model should have a value of more than 0.90).

22For clarity, the results are summarised using forest plots, showing standardised estimates for significant variables at each level of the models. Comprehensive details of the results are provided in Table 4 in the appendix.

Results

Income and neighbourhood perceived lack of amenities (NPLA): social and residential explanatory factors

23In the first regression, where the dependent variable is income (Figure 2-A), the results indicate significant positive effects for both men and women in managerial or intermediate professions. For men, being employed and having a higher level of education also has a positive impact on income. Similarly, age is positively associated with income. On the other hand, the presence of children in the household has a negative effect on income. Additionally, spatial variables play a role: residing in areas of unscattered housing is linked to higher income. The results of this income regression are as expected and do not reveal any anomalies.

24With regard to the perception of a lack of neighbourhood amenities (NPLA), which is summarised by a scarcity of shops, meeting places, and a relatively isolated location, the results also highlight this variable’s association with residential location (Figure 2-B). Living in larger urban areas (more than 200,000 inhabitants or Paris) tends not to be associated with a perceived lack of amenities. Conversely, households located in areas with dispersed detached housing are more likely to perceive a lack of amenities. Also, male managers are more likely to feel that their neighbourhood is poorly equipped. This regression indicates that NPLA primarily depends on residential location characteristics and lacks a significant social dimension except in the case of male managers.

Online Grocery and Online Shopping: intensities of uses with multiple and distinct drivers

25According to the regression results for intensity of online grocery shopping (Figure 3-C), the primary influencing factor is the woman’s profession. Women in management positions tend to have more intense e-grocery purchasing practices. The presence of children in the household also contributes to higher intensity of e-grocery purchase. Conversely, online grocery intensity significantly decreases with age. Finally, e-grocery is associated with an intermediate level of education. It is noteworthy that e-grocery practices are not associated with income or residential location variables. Therefore, e-grocery purchase is linked primarily to household sociodemographic characteristics.

26When considering the intensity of online shopping (excluding e-grocery), it becomes evident that the determinants display substantial divergence, as they are nearly all distinct (see Figure 3-D). Age, which is negatively correlated with intensity of online grocery purchase, is a common factor in most digital practices. However, in this case, income is positively associated with online shopping intensity, whereas living in an urban apartment building is negatively correlated. In particular, there appears to be no link between female characteristics and online shopping intensity, whereas male characteristics exhibit a strong connection: men with moderate or high levels of education and intermediate employment status are more likely to show higher online shopping intensity. In this case, therefore, income and residential location are linked with online shopping intensity, but there is no association with any specific female or household profile. Instead, intensity is associated with men with higher levels of education working in intermediate professions.

27When we look at the determinants of the delivery experience (as depicted in Figure 3-E), it appears that the primary driver is intensity of online shopping (excluding e-grocery). A higher level of engagement in online shopping practices is strongly associated with a greater likelihood of various home delivery experiences. There are few other factors that are significantly correlated with the home delivery experience. Once again, there is a negative correlation between increased age and digital usage. Being a male in employment also emerges as a positive factor. Generally, however, experience of home delivery is not significantly linked to residential location, income, or the social characteristics of either women or men.

Online Grocery and Online Shopping collection: lack of amenities is a determining factor for home delivery

28The regression results for grocery collection or delivery (see Figure 4-F) indicate that the primary determinant is the extent of online grocery shopping, whether for home delivery or out-of-home collection (click-and-collect). This finding is logical, as the incidence of a delivery method is inherently dependent on the intensity of online shopping practices.

29Moving forward, we observe distinct determinants for home delivery and out-of-home collection. Starting with home delivery, the perception of neighbourhood isolation emerges as the second significant factor. Households in which there is a perception of a lack of nearby amenities are more inclined to opt for home delivery of e-groceries. Another influential factor positively affecting the likelihood of home delivery is the age of the head of the household. In contrast, households containing a man employed in a managerial role are less likely to choose home delivery for their e-grocery orders.

30When examining the probability of using an out-of-home collection point (click-and-collect) to collect online groceries (Figure 4 F), only age affects the intensity of online grocery shopping and shows a negative correlation with this collection method. This implies that the choice of out-of-home collection for groceries is not explained by the survey variables alone, suggesting that other factors need to be explored. In contrast, in the case of home delivery, the correlated variables provide more insight into the underlying mechanisms.

31The regression results for online shopping collection or delivery services (as illustrated in Figure 4-G) once again emphasise that, logically, online shopping intensity is one of the primary drivers of use of delivery services. However, the magnitude of this effect is significantly greater for home delivery. This implies that as the intensity of online shopping increases, home delivery becomes the preferred choice.

32Another interesting difference between delivery services pertains to the Neighbourhood Perceived Lack of Amenities (NPLA) variable. NPLA is positively linked to home delivery but negatively associated with out-of-home collection (pick-up point). This suggests that home delivery is preferred over pick-up in areas perceived as having fewer amenities. Pick-up point delivery is not directly linked to any other determinants, further underscoring the conclusion that the choice of this delivery service is not adequately explained by the survey variables alone. In contrast, several other variables display significant correlations with home delivery. Income shows a negative correlation, while living in a neighbourhood characterised by unscattered houses demonstrates a positive link. This aligns with the findings of Morganti et al. (2014), which indicate that home delivery operates most effectively in such neighbourhoods. Even though parking and congestion issues are more prevalent in urban settings, distances between deliveries are shorter in areas of unscattered housing compared with scattered housing. Consequently, unscattered neighbourhoods offer the most effective delivery solutions both for the delivery provider and the end customer. Finally, it is worth noting that male employers or own-account workers are also more likely to opt for home delivery, perhaps on account of their professional activities.

Discussion and conclusion

33With the massification of online shopping and the spread of delivery services that has occurred in France over the past decade, the residential characteristics of households have little impact on the propensity to buy products online (food or otherwise). In fact, the prevalence of online shopping for different types of goods can be attributed primarily to socio-demographic determinants rather than residential factors. Nevertheless, the residential or spatial dimension of these practices remains influential in the choice of delivery service (home delivery vs out-of-home collection).

34For e-grocery shopping as for other types of goods, sociodemographic variables play a significant role in their use. For e-grocery, households with children or households in which there is a female manager are more likely to order online. For other types of goods, variables linked with the profession and level of education of the man in the household are more significant, along with income. For both types of goods, age is one the main significant factors. These results reflect the consequences of several phenomena. Organising food shopping, particularly in France, remains the responsibility of women in households, especially in households with children (Pernot, 2021). Men are far less likely to be involved in these activities, especially when they are employed. However, they are more likely to buy other goods, especially personal items (Kanwal et al., 2022), also partly because of the greater diversity of products in this category (Couclelis, 2004). More surprisingly, spatial factors do not come into play in explaining the practice of online shopping (food or non-food). Sociodemographic factors also remain important in explaining the choice of delivery services. While age is not an issue in the choice of delivery service for non-food products, it is for e-grocery shopping. Increasing age is negatively correlated with the probability of using out-of-home collection points for e-grocery shopping and positively with the probability of home delivery. For e-grocery shopping, households with men in professional management positions are less likely to opt for home delivery, while households with male employers or own-account workers seem to be more likely to use home delivery for other goods, possibly because they experience greater time pressure and have less time for shopping (Lesnard & Saint-Pol, 2009).

35If the more significant variables of delivery services are e-shopping intensity (the more people buy online, the more they use home delivery and out-of-home collection points), the objective and subjective (or perceived) spatial factors for their part explain the choice of one delivery service over another. Living in an area of scattered housing or in the suburbs tends to increase the choice of home delivery for non-food products. In addition, however, perceptions of the neighbourhood and its access to shops or amenities play a significant role in the choice of home delivery and out-of-home collection. More than the mere fact of living in an area of scattered housing, the perception of living in a poorly served neighbourhood tends to increase the use of home delivery (for e-grocery and non-food products) and to reduce the use of out-of-home collection points (for non-food products only). Home delivery remains more likely in less dense areas or where shops are considered less accessible. Out-of-home collections (for e-grocery via click-and-collect or for other products at pick-up points) are more likely to be used in areas of greater density or areas where amenities are perceived as more easily reachable. These results tend to show that the spatial attributes of online shopping (for food or other products) do not so much affect the intensity of the practice but rather the choice of delivery method. Moreover, the findings are enlightening in that they show that distinguishing between online purchases according to the type of goods purchased and the delivery service used to obtain these goods is an effective way to explore the spatial attributes of e-shopping (Zhen et al., 2018).

36Our results are based on a survey for which the year of completion (2016), residential scope (France) and number of observations (600 households) are undeniable limitations. Nevertheless, it is important to remember that French households were already frequently shopping online in 2016, although the pandemic subsequently made the practice more common. In addition, many delivery services, particularly out-of-home collection points, such as pick-up points or click-and-collect points, were already available in France at the time of the survey, whereas they developed later in many other countries. Derived from mail-order catalogue sales, pick-up point networks began to develop in France in the 1980s and expanded with the spread of e-commerce. Click-and-collect food collection points are a more recent development, enjoying great success from the 2010s onwards, and households have been quick to embrace them. In fact, click-and-collect grocery collection is a direct extension of normal, supermarket based consumer practices in France. Of course, e-commerce-related delivery services have further diversified since then (lockers, collection times, e-tailer subscriptions, etc.) and grew in scale with the global pandemic. Nevertheless, in 2016 France was already a prime location for observing the development of delivery solutions that were only introduced later in relatively developed countries, such as the United States or China (Song, 2022). Given the small number of observations, the LivMob survey was an exploratory study that was intended to take on further with a more recent survey involving a larger number of households and more diverse types of goods or delivery services.

37In conclusion, our article shows that the drivers of online shopping are mostly social with respect to intensity of use and more spatial with respect to the use of different delivery services. This means that the innovation and efficiency hypotheses proposed by Anderson et al. (2003) may not fully explain the socio-spatial patterns observed when multiple delivery services are available for different types of goods, and could be further refined. Beyond being a generalised or relatively uniform practice across the country, therefore, the socio-spatial dimensions of online shopping are not limited to the propensity to use e-shopping per se but relate to the choice of different delivery services. The latter may themselves be unevenly distributed geographically, as shown by Newing et al. (2022) in the case of food deliveries. If out-of-home collection points are less frequently used in areas of lower population density or in areas perceived as less well endowed with amenities, it is because they are less present or more difficult to access and because home delivery is chosen to avoid possibly long and complicated car journeys (Belton Chevallier et al., 2016). As a result, it may be concluded that shopping trips in less densely populated areas or areas with fewer amenities are more likely to be outsourced to home delivery drivers. Conversely, we can conclude that they would be more partially outsourced, to out-of-home collection points, in more densely populated or better served areas.

38With the increasing diversity of out-of-home collection points and the widespread use of e-commerce as a supply solution, it may be wondered to what extent this socio-spatial specialisation of delivery services and outsourcing of shopping trips may be transformed by the emergence of “unmanned” collection points, as suggested by Vakulenko et al. (2018). With the Covid-19 pandemic and the desire to offer cheaper and more efficient delivery, services with few or no personnel have the advantage of limiting the spread of a virus and reducing staff costs for different types of deliveries. For this reason, automated lockers networks are being developed on a massive scale in many countries such as China and Australia (Ma et al., 2022). France is no exception to this trend, and lockers operated by last-mile providers have proliferated across the country since 2022. At that time, Mondial Relay and Pick-up (La Poste) operated only a small number (respectively 300 and 600) of lockers. Today, in 2024, Mondial Relay offers 5000 lockers and Pick-up 2400 (information provided on their websites). Moreover, many retail chains – such as Decathlon or Carrefour – have installed them in their stores for the collection of goods that they also sell online. According to an OpinionWay study for Mondial Relay, lockers are mostly popular because of “their simplicity of use, proximity and their extensive access times” (FEVAD, 2024), which allow people to collect and retrieve their online purchases at almost any time of day when located in public or accessible places. Less costly for households and operators, lockers can be installed in a wide range of locations, particularly in sparsely populated areas. As a result, automatic lockers may encourage the emergence of out-of-home delivery in less densely populated areas, and lead to an increase in car trips that may previously have been reduced by the availability of home delivery.