The sharing economy is not always greener: a review and consolidation of empirical evidence

Papers for this review were retrieved from Web of Science (WoS) and Scopus. The basic search string was constructed based on previous sharing economy reviews, and limited to peer-reviewed studies and conference proceedings in English, published between 2004 and 2022. The basic search string was refined in an iterative process to specifically target papers on environmental performance and digital sharing platforms using titles, keywords and abstracts (see table 1). The final string was composed of four sub-strings which covered sharing economy and sharing economy platforms across different consumption domains (search sub-strings A, B, C, and D), and one sub-string honing in on environmental impacts. The search string can be expressed with Boolean Operators as:

((A or B or C or D) and E).

Finding 1659 papers in WoS and 1858 in Scopus, we downloaded the results as reference files. Duplicates were removed using Python code, which left 2255 papers. We then screened papers first by their title and abstract, and then by reading the full text (N = 460). Removing any remaining duplicates, we compiled the final dataset which included 149 relevant papers as well as 6 additional papers that were highly cited by others, yet did not come up in the search (see figure 1).

Zoom In Zoom Out Reset image size

The PICO framework helps to define the inclusion criteria used in this review. The study population (P) includes platforms catering to consumers and private households. The intervention (I) includes any physical consumption mediated by the digital sharing economy. To clarify, activities such as knowledge transfer are excluded. Eligible studies must compare (C) the sharing economy to prevalent consumption modes and report outcomes (O) in terms of environmental impacts or behavioral changes from which environmental impacts can be easily derived. We include both qualitative and quantitative studies, as long as they include an empirical analysis. Table 2 provides a detailed explanation of inclusion and exclusion criteria.

Data perpetration for analysis included two main stages: assignment of environmental impact scores and classification by features related to the platforms' studied or papers' research design.

The papers in our final dataset included work from a wide range of disciplines, and thus reported results using different measures (e.g. number of people not buying a car, or kg of CO₂eq per km of bike ride). To consolidate the empirical body of knowledge and assess the environmental impacts of the sharing economy enable an analysis of all papers, the research team therefore examined the full text and manually assigned each paper an environmental impact score, ranging from negative, through mixed to positive. Note that we did not evaluate the scientific quality of the papers nor their methods. Thus, the environmental impact scores reflect the reported results as written up by each paper's original authors.

Next, papers were classified based on different features which could potentially affect environmental outcomes. These include platform related features such as the type of product shared (i.e. sharing domain), sharing model, and product ownership structure, as well as research design features such as the type of impact examined (environmental impacts or consumption changes with relevance for environmental outcomes), the scope of the analysis (i.e. product level or sector level analyses), and whether added platform operations or displacement were considered. Added operations refer to activities required to support platform operations. Within shared fashion this would be added transport and cleaning services, while in mobility operations may include Uber trips with no passengers (also known as 'dead-head' miles), or rebalancing bike stocks between different locations. Displacement refers to studies that examined the specific consumption pattern that sharing displaced, for example, determining the transport mode one would use if car-sharing were not available (also known as modal shift).

Table 3 presents a detailed description of all features considered in our analysis. A list of all papers included in our review, their classifications and their assigned environmental scores is available in the SI.

Common guidelines for synthesizing papers which report environmental impacts suggest statistically examining heterogeneity in reported results (Collaboration for Environmental Evidence Synthesis Assessment Tool, see Woodcock et al 2014). This includes statistically evaluating potential reasons for heterogeneity in the reported environmental outcomes. Similar to prior studies (e.g. Ivanova et al 2020 Galante et al 2023) we examined potential sources of heterogeneity such as platform or study design features (i.e. effect modifiers) which could potentially affect reported results of the sharing economy's environmental performance in comparison to prevailing consumption patterns.

In our analysis, we used an ordinal probit regression model to examine whether features significantly affected reported results as well as the direction of this effect. Reported results (the dependent variable) were coded as negative (1), mixed (2) or positive (3) according to the environmental impact score they were assigned. The features examined (the independent variables) were entered into the regression based on data availability and theory. Specifically, we gave preference to features which theory suggests would affect the environmental performance of sharing, including the sharing domain, region, the ownership model, the studies' system boundaries, and the inclusion (or lack thereof) of added operations and displacement. Other features, although of interest, were excluded due to small sample sizes (e.g. there were only 3 not-for-pay platforms in our dataset). All categorical predictors were entered as dummy variables into the regression. Across all regression models, we differentiated between shared transport types, while keeping accommodations and good sharing as separate groups given their smaller sample size. In Models 1–5, Goods served as the base category for comparison between subdomains, while in Model 6 which focused solely on shared transport, shared bikes served as the base category. To verify that our main findings were robust to the regression specifications, we repeated the analysis using alternative regression models (see SI).

Finally, in some cases the sample size limited our ability to test for interaction effects between different features (e.g. subdomain and region). Therefore, to dig in deeper and tease out underlying trends we also conducted a descriptive analysis of environmental impact scores. Of particular interest were potential differences between sharing domains which we thus further explored.

Results and specifications for the different regression models are presented in table 4. Descriptive statistics are available in tables SI-6 through SI-18. We find significant differences in environmental performance of sharing platforms across subdomains across all regression models. Specifically, studies on ride hailing and scooter sharing (Models 1–5, p < .001), along with papers on accommodations (Models 1–3, and 5) were more likely to report negative environmental impacts compared with studies on goods sharing (p < .001). Differences between subdomains are also evident in the descriptive analysis (see figure 3(b), table SI-8, and a more detailed overview in section 3.2 below).

Despite theoretical predictions, platform location and ownership model (P2P vs. B2C) did not significantly affect the likelihood to be assigned a specific environmental impact score (see regression Models 1 and 2 respectively). Descriptive analysis of these features did not present interesting results for platform region, but did reveal that papers on decentralized platforms where peers share with other peers are more likely to report negative outcomes (57% negative for P2P vs. 33% in B2C platforms, see figure 3(d) and table SI-9). Interestingly, 62% of papers on larger platforms reported negative results compared to only 21% of papers on smaller or medium platforms (figure 3(c) and table SI-11). Since, however, only 64 papers could be assessed in terms of platform size the sample size precluded a regression analysis.

To understand if and how research design choices affect reported results, we also examined different features related to study design. Contrary to our expectations, system boundaries, whether papers looked at the level of product, sector, or entire economy, did not significantly affect the likelihood that a paper would report positive or negative findings (see reg. Model 3). In contrast, accounting for platform operations emerged as statistically significant across all relevant models (Models 4–6, p < .001) and increased the likelihood that papers would report negative findings (53% negative when accounting for operations vs. 25% without, see figure 3(e) and SI table SI-16). Choice of data source displayed more negative results for public data on existing platforms (50%) compared to public data on theoretical platforms (13%, see figure 3(f) and table SI-12), though here too sample size was too small to support a regression given the 4 different source types.

When examining papers from all sharing domains, accounting for displacement did not significantly affect results. Yet we did find a main effect of displacement when including an interaction between displacement and the mobility sub domain (Model 6, p = 0.019). However, these results were sensitive to the choice of baseline category and should therefore not be seen as robust. A descriptive analysis of displacement suggests that there are differences between subdomains (see table SI-17). Negative results were more common in ride hailing papers that included displacement vs. those that did not, as many platform users substituted walking and public transportation for ride-hailing (Rayle et al 2016, Clewlow and Mishra 2017, Alemi et al 2018, Afroj et al 2019, Gehrke et al 2019, Lee et al 2019, Tirachini et al 2020, Schaller 2021). In contrast, within bike sharing displacement seemed to positively affect reported results, as some of the shared bike rides would displace more carbon intensive transport modes. Within car sharing, displacement had mixed results, depending on user profiles. For car dependent users, participating in car sharing lowered emissions but for car free users, car sharing increased emissions (Arbeláez Vélez and Plepys 2021, Martin and Shaheen 2008, Wang et al 2022).

Research on goods sharing mostly reports environmental benefits from increased use intensity of underutilized products (Martin et al 2019, Schneider et al 2019, Makov et al 2020, Wasserbaur et al 2020, Kerdlap et al 2021), though some of these may be eroded as more intensive use may shorten product lifespan (Retamal 2017, Zamani et al 2017, Schneider et al 2019). In addition, as evident from the regression results, some of these benefits can also be eroded as goods sharing often requires supportive logistic operations which may increase demand in other sectors, such as transportation and cleaning leading to problem shifting (Amasawa et al 2020, Behrend 2020, Makov et al 2020, Johnson and Plepys 2021, Kerdlap et al 2021).

While shared accommodation research is very limited (N = 8), two key findings emerge. First, leasing an entire apartment is more common than the more environmentally benign option of leasing a room in an otherwise occupied dwelling, with Airbnb consisting over 5% of housing stock in the cities reviewed (DiNatale et al 2018, Stergiou and Farmaki 2020, Muschter et al 2022). Second, Airbnb and shared accommodations might not displace hotel stays and instead may induce additional travel as lower prices allow users to take more frequent and longer vacations (Tussyadiah and Pesonen 2015, Farronato and Fradkin 2018, Sainaghi and Baggio 2020). Taken together, reports of reduced use intensity, and induced consumption suggest that shared accommodation increases rather than reduces environmental impacts.

Bike sharing papers (N = 48) report mostly positive results. Contrary to intuition, shared bikes are less efficient compared to private bikes, to a great degree because of an increase in bicycle stocks, but also because of platform operations (Ma et al 2018, Luo et al 2019, Chen et al 2020, Sun and Ertz 2021a, Wang and Sun 2022). In bikes, environmental benefits can also stem from bikes displacing private cars and results are predominantly positive in papers reporting bike displacement to cars (according to surveys around 20%: Lu et al 2017, Chen et al 2020, Lai et al 2021, Suchanek et al 2021, Cheng et al 2022), though these benefits are sensitive to the assumed displacement patterns (Kou et al 2020, Li et al 2020, Zhi et al 2022). However, when added operations (i.e. bike rebalancing) are accounted for, results shift becoming more negative (Sun and Ertz 2021a, Reck et al 2022, Wang and Sun 2022). Accounting for added operations, Li et al (2020) pinpoint the displacement tipping point at which bike sharing reduces emissions suggesting that at least 30% of shared bike rides should displace car rides. So, while bike sharing has potential to reduce environmental impacts, current papers which do not take platform added operations into account might overestimate the benefits.

In shared scooters, 13 out of the 17 papers report negative results mostly due to scooter's short lifespan and the high environmental costs associated with their production, as well as the emissions associated with their rebalancing and charging (Hollingsworth et al 2019, Moreau et al 2020, Severengiz et al 2020).

Most papers on car sharing (N = 40) report positive findings (60% positive vs. 13% negative), arising from modal shift (Lane 2005, Amatuni et al 2020) and reduced car ownership (Firnkorn and Müller 2011, Clewlow 2016, Nijland and van Meerkerk 2017, Becker et al 2018, Mishra et al 2019, Chapman et al 2020). However, in many papers the improvements are not directly related to sharing but rather related to newer or smaller cars (Cervero and Tsai 2004, Namazu and Dowlatabadi 2015), moving to electrical cars (Chicco and Diana 2021) or both (Fernando et al 2020, Migliore et al 2020).

Ride-hailing papers (N = 28) tend to report negative results, mainly due to users displacing low emissions modes of transport such as walking and public transportation (Rayle et al 2016, Tirachini et al 2020, Schaller 2021, Shi et al 2021), induced travel (Rayle et al 2016, Gehrke et al 2019) and added operations in the form of dead-head miles, where drivers drive around to pick up passengers potentially doubling the distance traveled (Oviedo et al 2020, Tirachini et al 2020, Schaller 2021). In contrast, ride-sharing papers (N = 15) predominantly report positive outcomes, mostly stemming from increased car occupancy (Ding et al 2019, Realini et al 2021, Sun and Ertz 2021a).

The expectation that sharing would relieve environmental burdens weighs heavily on the notion that sharing, like other access-based economy models (Tukker 2015), increases use intensity and enables the provision of a fixed amount of utility with smaller product stocks. Yet despite intuition, sharing does not necessarily reduce stocks (e.g. fleet size or the number of vehicles used overall), as illustrated by research on ride-hailing (Gong et al 2017, Zhang and Zhang 2018, Paundra et al 2020, Wadud and Namala 2022), and bike sharing (Ma et al 2018, Luo et al 2019, Chen et al 2020, Wang and Sun 2022). Furthermore, while sharing can potentially increase use intensity (Zhang and Mi 2018, Martin et al 2019, Schneider et al 2019, Amatuni et al 2020, Makov et al 2020, Wasserbaur et al 2020, Kerdlap et al 2021), it can also reduce use intensity, when for example, shared bikes and scooters are used less frequently than privately owned ones (Reck et al 2022) or when residential dwellings are converted into Airbnb vacation apartments (DiNatale et al 2018, Stergiou and Farmaki 2020).

Critically, measures of use intensity often reflect idle time which does not necessarily correlate with environmental impacts. Within mobility for example, 90% of GHG emissions are associated with the use phase rather than production (Allwood et al 2012). This means that GHG emissions are more a function of the passenger-km driven (Schäfer and Yeh 2020) than the number of cars on the road at any given time (i.e. fleet size). As such, reducing the number of cars should be coupled with increased passenger occupancy for it to effectively reduce environmental impacts (Amasawa et al 2020, Sun and Ertz 2021a).

Yet even when sharing increases use intensity and reduces stocks, it might not deliver the expected reduction in environmental burdens. Added platform operations can offset some or even most of the expected benefits of increased use intensity. Kerdlap et al (2021) suggests that the added transport and cleaning services required in stroller sharing offset a large portion of the environmental benefits. Similarly, managing shared scooter and bike stock locations and moving them from one place to the other to meet users' demand in free floating systems can have substantial environmental impacts (Hollingsworth et al 2019, Li et al 2020).

Platform operation might also be related to platform size and help explain why larger platforms are generally associated with more negative environmental impacts where smaller ones do not. As platforms expand so does their geographic reach. Uber drivers might have to drive more dead-head miles between passengers or trucks rebalancing bikes will drive further. Relatedly, as platforms scale the economic incentive to participate might increase, and as a result sharing ceases to be a marginal activity of owners monetizing unused stocks and becomes closer to conventional markets instead. Recent work by Cansoy and Schor (2023), reveals that as Airbnb became more popular, frequently rented listings (essentially vacation apartments) started making up a larger share of its total listing supply.

These examples demonstrate how crucial it is for researchers to take a full system's perspective and incorporate platform operations in environmental assessments. Failing to go beyond simplistic single product comparisons may often result in a biased result which does not capture the full extent of environmental impacts.

Changes in consumer behavior or in consumption patterns more broadly should also be taken into account as they can affect the environmental performance of sharing. First, sharing may not displace the products and services it is assumed to replace. For example, the literature on ride-hailing suggests that 21%–61% of trips did not displace car rides but more environmental modes of transport such as walking and public transportation (Rayle et al 2016, Clewlow and Mishra 2017, Alemi et al 2018, Afroj et al 2019, Gehrke et al 2019, Lee et al 2019, Tirachini et al 2020, Schaller 2021). Moreover, sharing may trigger added demand increasing overall environmental burdens rather than decreasing them. For example, Rayle et al (2016) reported that 8% of ride-hailing users surveyed would have otherwise stayed home, while Farronato and Fradkin (2018) reported that 42%–63% of nights booked on Airbnb would not have resulted in a hotel booking in the absence of Airbnb.

Second, sharing often saves users money which they then typically re-spend on additional products and services, a phenomenon referred to as re-spending or indirect rebound effect (Meshulam et al 2023). The rebound effect can occur when platform users save money (by collecting free food for example) and then use these savings to buy other things (Druckman et al 2011, Makov and Font Vivanco 2018, Sorrell et al 2020). Alternatively, rebound can emerge when owners in P2P (decentralized) platforms earn money by sharing their apartments (Cheng et al 2020) or when yacht owners spend only a portion of their money on maintaining yachts (Warmington-Lundström and Laurenti 2020). While research on sharing economy rebound effects remains scarce, the existing evidence from accommodations and goods sharing suggests that added consumption (or rebound effects) may negate a substantial part of the expected environmental benefits of sharing (Cheng et al 2020, Warmington-Lundström and Laurenti 2020, Meshulam et al 2023).

This review is limited to English papers only, and might be missing relevant non English papers. Moreover, given that our findings stem from extant literature, we can only draw conclusions on features widely researched. Consequently, our ability to draw insights on goods and accommodation sharing is limited due to the small number of papers. Similarly, while we did not find evidence that P2P platforms which operate as a two-sided market are necessarily better than platforms where resources are centrally owned (B2C platforms; Curtis and Lehner 2019, Curtis and Mont 2020), these results might be affected by research availability and confounded with other platform features such as domain and size, given the large number of papers on Uber and Airbnb. More work is needed to better understand the relationship between platform ownership and environmental impacts across a wider range of platforms and domains.

Future research should extend our findings and dive deeper into optimizing the sustainability benefits of the sharing economy. This is particularly important in light of growing interest in sharing among policy makers at the national and local level as well. Cities, in particular, have emerged as hubs for sharing. Past work has highlighted the major role that urban areas play in the growing popularity of the sharing economy. Davidson and Infranca (2016), for example, poise that 'platforms and networks that make up the sharing economy fundamentally rely for their value proposition on distinctly urban conditions'. Relatedly, Mont et al (2020) state that sharing in cities is 'particularly promising' from a sustainability standpoint. Indeed, urban features including high concentrations of economic activity, human population, consumption, and existing infrastructure, are thought to make urban areas exceptionally 'fertile breeding grounds' for sharing initiatives and innovations (Palm et al 2019, Mont et al 2020). Yet despite much interest in sharing as an urban phenomenon, as evident by the growing body of work focusing on urban governess of sharing activities (Davidson and Infranca 2016, Bernardi and Diamantini 2018, Akande et al 2020) the ways in which urban form affects sharing networks and the subsequent environmental impacts of sharing needs further exploration.

In a broader context, future research should also take into account rebound effects, growth imperatives, power imbalances, and the limitations imposed by capitalism, which hinder the potential for sharing, along with the commonly discussed concerns regarding greenwashing (Ivanova and Buchs 2023).

The sharing economy, along with the digital technologies it relies on, is a growing phenomenon, increasing its presence in everyday lives. Although our findings suggest that the sharing economy does not reduce environmental impacts by default, it is critical to note that it is not inherently negative either, and both platforms and their users could help make sharing more environmentally benign. Platforms can optimize their operations for minimal environmental impact (rather than costs alone). For example, platforms can optimize their stock rebalancing practices or place more emphasis on extending the lifespan of shared products through proper maintenance (Luo et al 2020, Severengiz et al 2020). In addition, platforms can strive to target specific consumer segments whose current consumption patterns are more environmentally intensive compared to sharing. Similarly, users can actively opt for more environmentally benign options such as waiting for shared carpooling (e.g. Uber pool) rather than single occupancy ride-hailing, forgoing dry cleaning of shared strollers, or deciding to collect free shared food only if it is within walking distance (Cai et al 2019, Makov et al 2020, Kerdlap et al 2021). Given growing interest in sharing as a sustainable consumption model, policy makers and sustainability advocates should push for more comprehensive analyses of sharing and how it can be optimized not only to reduce idle stock or scale up economically, but also on how sharing can be designed at the system level to minimize environmental impacts.