In the following, we present the results of our literature review, specifically (i) what application domains have been covered, (ii) the number of use cases focused on, (iii) whether the studies focused on ICT-induced patterns of production or consumption, and (iv) the methodological approaches applied.
3.1. Application Domains
Assessments of indirect environmental effects of ICT address how and to what extent ICT as an enabling technology changes patterns of production and consumption in domains other than ICT. We classified all of the assessments according to the application domains they covered and derived a set of common application domains. Finding a collection of application domains that are extensive and mutually exclusive is challenging. For example, the domain
dematerialization, as used by British Telecom (BT), refers to how ICT “replaces the need to manufacture, publish, print and ship newspapers, documents, books, CDs and DVDs for residential customers” [
52] (p. 20), Hilty et al. use the term
virtual goods to describe ICT’s capacity to enable “a shift from material goods to services” [
11] (p. 1262), whereas Coroama et al. [
76] use the term
electronic media to cover the transition from paper-based books to e-book readers and from physical travel to video conferencing. Producing and delivering a newspaper online instead of paper-based could be classified under all three mentioned domains; however, video conferencing would be a part of electronic media, as defined by Coroama et al. [
76], but not part of dematerialization as defined by BT [
52] or virtual goods as defined by Hilty et al. [
11].
Despite these issues, we identified seven common application domains. These are mainly based on two well-cited studies of the overarching indirect environmental effects of ICT [
6,
11] and allowed for us to classify almost all other studies identified in the literature review (see
Table 3). Most studies cover the application domains
virtual mobility,
smart transport and
virtual goods (see
Figure 2), followed by
smart buildings,
smart energy,
smart production, and
shared goods. Other application domains mentioned are
smart agriculture,
smart water, or
smart waste management; however, these are less frequently assessed.
Two studies could not be classified with respect to application domains. Laitner et al. [
26] conduct a regression analysis of historical macroeconomic time series data about the United Sates (U.S.) economy before and after the introduction of the semiconductor and thereby implicitly cover all potential application domains, without explicitly mentioning them. Røpke and Christensen [
75] assess how ICT—in general—changes everyday life, also without focusing on specific application domains.
3.2. Number of Use Cases
Most of the studies we identified assess specific ICT use cases (e.g., e-books, videoconferencing). Studies estimating the overall impact of ICT often select a number of the most common or prevalent use cases from various application domains and aggregate the environmental impacts across all use cases (e.g., [
6,
7,
55]). We have to consider that the studies use different abstraction levels and definitions for use cases, which is why it is difficult to match the use cases across studies. Therefore, the numbers provided in the third column of
Table 2 and in
Figure 2 and
Figure 3 are to be interpreted with caution. From a methodological perspective, it is essential to distinguish between studies that are focusing on one use case only and studies investigating several use cases because in the latter case, interactions between use cases can (and should) be studied. Therefore, we distinguish between “single-use-case studies” and “multi-use-case studies” in the following.
In total, we found 30 “single-use-case studies” and 21 “multi-use-case studies”. The latter usually apply relatively simple estimation methods to determine a specific environmental impact for each use case (e.g., GeSI applies the “ICT enablement method” (ICTem) in its SMARTer studies to estimate the ICT-induced GHG emission reduction potential for a collection of use cases [
6,
9,
10,
77]). There seems to be a trade-off between the depth of analyzing each use case vs. the scope of domains and use cases that are covered by the studies. Therefore, multi-use-case studies are often close to back-of-the-envelope calculations, also called “Fermi calculations”, which try to derive a rough estimate from a few simple assumptions [
78]. In contrast, the single-use-case studies usually apply methods allowing for a deeper analysis, including life cycle assessment or partial footprint (e.g., [
39,
40]). Mostly, the aim of these assessments is not just to estimate the environmental impact of the use case under study, but also to unveil the hidden mechanisms and impact patterns behind the use case in order to derive recommendations for policies or ICT application design. In search for deeper analysis, some studies also use simulation models. Xu et al. [
59] create an agent-based model to investigate the impact of increasing Internet penetration on consumers’ use of traditional and e-commerce book retailing schemes. Hilty et al. [
11] apply System Dynamics modeling to investigate the impact of ICT on the energy, transport, goods, services, and waste domains, and how these impacts affect total energy consumption and GHG emissions.
Three studies have no focus on specific use cases. Picha Edwardsson [
38] qualitatively explores the environmental impact of scenarios for future media use. As mentioned above, the studies by Laitner et al. [
26] and Røpke and Christensen [
75] could not be related to specific application domains.
3.3. Patterns of Production and Patterns of Consumption
ICT changes both the patterns of production (e.g., by changing manufacturing processes) and patterns of consumption (e.g., by changing individual media use). As can be expected, changes in production and consumption patterns are closely interrelated. For example, optimization of logistics has decreased the cost of logistic services (the service can be produced at a lower price and faster), such that e-commerce retailers can afford to offer free delivery and return to consumers, which dramatically changed consumer online shopping behavior (e.g., the online retailer Zalando had an order return rate of roughly 50% in 2013 [
79]).
12 of the assessments identified in our literature review focused on ICT’s impact on patterns of production. Moberg et al. [
39], for example, compares the environmental impact associated with production, use, and disposal of paper-based books vs. e-books. Such studies commonly use product-oriented assessment methods, such as LCA or partial footprint.
35 assessments focusing on ICT’s impact on patterns of production also consider changes in patterns of consumption. Many of these studies use ICTem. They first assess the impact of ICT on production processes and then the reaction of consumers to it. GeSI [
6], for example, calculate the GHG emissions that are associated with the provisioning of ICT-based learning, health, and transport services, and then estimate how many consumers will adopt these solutions in future.
Only three assessments focus on ICT’s impact on patterns of consumption exclusively. For example, Atkyns et al. [
57] use survey results to assess employee telecommuting behavior, as well as drivers and challenges of telecommuting adoption, without assessing the actual environmental impact of telecommuting compared to conventional commuting. These studies use consumer-centric assessment methods to identify changes in individual consumption, such as interviews or surveys.
3.4. Methodological Approach
Researchers use a variety of approaches for the assessment of indirect environmental effects of ICT. The assessments identified in our literature review used 15 approaches, namely agent-based modeling (ABM), system dynamics (SD), life cycle assessment (LCA), partial footprint, the “ICT enablement method” (ICTem), regression analysis, descriptive statistics, material input per service unit (MIPS), transport models, vehicle drivetrain models, scenario analysis, literature review, meta-analysis, interviews, and surveys. LCA, ICTem, and partial footprint are by far the most frequently used assessment approaches, whereas simulation methods and qualitative approaches are less often applied. In the following we describe the approaches and how they are applied in the field of indirect environmental effects of ICT. We exclude descriptive statistics, interviews, surveys, vehicle drivetrain models, literature review, and meta-analysis, as these are too generic. We further add the Software Sustainability Assessment method (SoSa), a recent approach proposed in the ICT4S community to assess the environmental impact of software systems [
80,
81].
Figure 3 subsumes meta-analysis, scenarios, transport models, vehicle drivetrain models, regression analysis, descriptive statistics, surveys, and MIPS under “others”. “Qualitative methods” include interviews and literature reviews.
Life cycle assessment (LCA) is used to estimate the environmental impact of a product system, evaluated with environmental indicators, by modeling all exchange of energy and matter between the product system and its environment [
82]. There are different types of LCA, which we do not distinguish in this study. Finnveden et al. [
83] provide an overview about recent developments in LCA. For indirect environmental effects of ICT, LCA typically compares the environmental impact of two product systems that differ with regard to ICT application. For example, Moberg et al. [
39] compare the environmental impact of reading paper-based books and reading books using an e-book reader. By applying LCA, they find that the production of an e-book reader causes approximately the same amount of GHG emissions as the production of 30 to 40 average books.
Many authors in the field of indirect environmental effects of ICT focus their analysis on selected life cycle stages only. For example, in their analysis of telecommuting, Kitou and Horvath [
56] evaluate the energy consumption of homes, offices, and ICT equipment, looking at their use phases only. A more comprehensive LCA would at least include the emissions that are caused by the production and disposal of the ICT equipment or other crucial assets. In line with ISO 14067, which specifies a “partial carbon footprint of a product” as the “sum of greenhouse gas emissions […] and removals […] of one or more selected process(es) […] of a product system […], expressed as CO
2 equivalents […] and based on the relevant stages or processes within the life cycle […]” [
84] (p. 2), we call such approaches
partial footprints, even if the environmental indicator is not GHG emissions. Such studies calculate the emissions or energy consumption for selected processes only, without applying a full life cycle approach.
Material input per service unit is a product-oriented assessment approach developed by Schmidt-Bleek [
85] to measure the resource productivity of services. It calculates the natural resources required throughout the life cycle of a product per unit of service delivered.
System dynamics (SD) is “a method that permits researchers to decompose a complex social or behavioral system into its constituent components and then integrate them into a whole that can be easily visualized and simulated” [
86] (p. 3). The interaction among system elements is modeled by connecting stocks with material flows, such as water running through pipes (flow) and increasing the water level in a bathtub (stock), and stocks and material flows with information flows [
86]. The key strengths of SD are that it helps decomposing complex systems into causally connected variables and that it can be executed by computer simulation to observe the behavior of the system over time. It is for these strengths that SD is often used in policy analysis. In the literature review, we found only one application of SD. Hilty et al. [
11] used SD to simulate the impact of ICT on environmental sustainability in the year 2020 (starting in the year 2000) in order to evaluate policy scenarios.
In
agent-based modeling (ABM), a system “is modeled as a collection of autonomous decision-making entities called agents. Each agent individually assesses its situation and makes decisions on the basis of a set of rules” [
87] (p. 1). In a simulation experiment, agents repeatedly interact with each other and with their environment. Their collective action determines the behavior of the system as a whole [
87]. ABM is especially useful to study emergent phenomena, e.g., macroeconomic phenomena emerging out of behavior at the micro level [
88]. Xu et al. [
59] use ABM to test different e-commerce book retailing schemes, the reaction of consumers to it, and how these affect the CO
2 emissions that are associated with book retailing.
Scenarios “denote both descriptions of possible future states and descriptions of developments” [
89] (p. 723).
Scenario analysis is a method in the area of future studies. Future studies are a collection of methods to “explore possible, probable and/or preferable futures” [
89] (p. 724). Comparing different scenarios that are based on different assumptions about future ICT development can provide insights on the environmental consequences of ICT application. Arushanyan et al. [
90] use scenario analysis in combination with LCA and develop a framework specifically for the environmental and social assessment of future ICT scenarios.
The
ICT enablement method (ICTem), as introduced by GeSI in 2010 [
77], can be used to quantify the carbon-reducing effect of ICT use cases. ICTem is useful to quickly provide a rough estimate of the environmental impact of an ICT solution. The approach is close to a Fermi problem or “back-of-the-envelope calculation”. In the SMART 2020, SMARTer 2020 and SMARTer 2030 reports [
6,
9,
10], GeSI uses ICTem by
identifying GHG abatement levers (e.g., reduction in transport demand),
estimating baseline emissions,
estimating the level of adoption of the use cases in the population,
estimating the impact on GHG emissions per unit of adoption, and
estimating the rebound effect (for an example see
Figure 4).
A feature that distinguishes ICTem from a partial footprint is that ICTem focuses on the mechanisms that cause the changes of environmental impact. Such studies almost exclusively present favorable indirect environmental effects of ICT, even though the method would also allow for estimating the size of unfavorable effects (e.g., by including induction effects or obsolescence effects [
1]).
Studies that are focusing on the transport domain usually develop a
transport model and assess how ICT changes transport. Transport models are usually combined with a partial footprint approach. Siikavirta et al. [
71], for example, model the impact of different e-commerce schemes on road truck delivery and estimate the avoided fuel consumption and resulting GHG emissions.
Using
linear regression analysis [
91], Laitner et al. [
26] estimate how the relationship between energy consumption (dependent variable) and economic growth and semiconductor investment (independent variables) in the U.S. changed after the introduction of semiconductor technologies. The application of regression analysis for indirect environmental effects of ICT can be manifold, for macroeconomic effects (see Laitner et al. [
26]) or for specific ICT applications (e.g., the effect of a traffic management system on the concentration of particulate matter in a city). However, it always treats the assumed causal mechanism as a black box and it does not reveal underlying system structures.
Even though we could not find application examples, we would like to mention the
software sustainability assessment (SoSa) method, a recent approach to assess the environmental impact of software systems. SoSa analyzes the immediate, enabling, and systemic impacts of software systems on “economic, social, environmental and technical” sustainability [
80] (p. 1). The result is similar to a causal loop diagram and helps to understand the relevant impacts of a software system to improve software design [
80,
81].