The Climate Effect of Digitalization in Production and Consumption in OECD Countries Thomas Kopp1 and Steffen Lange2 1 The University of Göttingen, Platz der Göttinger 7, 37073 Göttingen, Germany 2 Institute for Ecological Economy Research, Potsdamer Str. 105, 10785 Berlin, Germany Abstract—How does increasing digitalization affect the Department of Trade and Industry claims that digital- environment? A number of studies predict that digital- ization improves an economy’s ecological sustainabil- ization will ultimately reduce environmental degradation ity by increasing resource and energy use efficiencies but seem to overestimate the emission-reducing effects of digitalization through increases in resource efficiency, while (E NERGIE, 2015). According to the Association of Ger- underestimating substantial rebound effects and negative man Engineers, digitalization may result in increases environmental impact of the construction and maintenance in resource efficiency of up to 25% (R ESSOURCENEF - of complex digital infrastructures. Additionally, the envi- FIZIENZ, 2017). And the ’Global e-Sustainability Initia- ronmental benefits of decreasing consumption of one-time tive’, an international network of IT companies, argues usage goods may be outweighed by the environmental costs of the production of ICTs and the increasing use of digital that digitalization has the potential to decrease global technologies. carbon emissions by an impressive 20% (G E SI and ACCENTURE, 2015). This paper analyzes the relationship between degrees of countries’ level of digitalization and environmental indica- Looking closer at these studies reveals that such pre- tors by use of a panel data set of 37 economies. It is the dictions are founded upon weak empirical bases. Some first paper to differentiate between emissions associated publications simply postulate the likely potential of digi- with a country’s production and those connected to a country’s consumption, accounting for emissions related talization to decrease environmental pressures with little to exports and imports. The level of digitalization in subsequent quantitative analysis (F ORSCHUNG, 2014; production is approximated by companies’ investments B UNDESREGIERUNG, 2014; W ISSENSCHAFT, 2013). in digital technologies. The chosen indicator to measure Others, while based on more concrete empirical cal- consumers’ proclivity to digital technologies is online shop- culations, nevertheless overestimate positive effects and ping behaviour. We address the problem of changes in unobserved heterogeneity by using the recently developed underestimate negative ones (G E SI and ACCENTURE, Group Fixed Effects estimator. 2015), for a discussion see H ILTY and B IESER (2017). Results indicate that the beneficial environmental effects The scientific literature on the environmental effects of of digitalization on reducing climate gas emissions slightly ICT usually differentiates between effects on different outweigh the undesired environmental effects, both in pro- levels. Most taxonomies have in common that they duction and consumption. Ultimately, we find that increases in digitalization have a net positive effect on the natural include first higher order effects (RØPKE, 2012; H ORNER environment. et al., 2016a; P OHL et al., 2019). The definition of first order effects is similar throughout the literature. It entails Keywords: ecological footprint, digitalization, environ- the energy, resource use and emissions associated with mental throughput, industry 4.0. the production life cycle. This entails the production, use phase and disposal of ICT. I. I NTRODUCTION However, the entire environmental effects of ICT in- volve additional mechanisms (A RVESEN et al., 2011; There is a widely-held consensus among politicians and H AKANSSON and F INNVEDEN, 2015). Such higher order economists that increases in digitalization will have a effects are manifold. Which effects are incorporated net-positive effect on the environment. The German into the analysis and how they are systemized varies II Literature review throughout the literature. Some of these effects tend to macroeconomic shocks parallel to digitalization makes have positive and some to have negative environmental it even more complicated to measure each of the three consequences. A recent list of effects includes substi- mechanisms as along with the overall effect of digi- tution effects, optimization effects, beneficial effects, talization. Indeed, it is difficult to clearly separate the direct rebound effects, indirect rebound effects, induction environmental effects caused by digitalization from those effects, sustainable lifestyles and practices, transforma- effects caused by other key factors, like the continuing tional rebound effects, induction effects and systemic globalization of world trade, economic policies aimed at transformation and structural economic change (P OHL climate change and environmental threats, urbanization, et al., 2019). Several authors do not further categorize and population growth, among others. these higher order effects (B ÖRJESSON R IVERA et al., Due to these challenges in measuring the environmental 2014; P OHL et al., 2019). However, there exist also effects of digitalization directly, an alternative method several categorizations amongst the higher order effects. – complementary to the existing ones in the literature B ERKHOUT and H ERTIN (2004) differentiate between – is to compare economies within the same historic indirect and systemic effects and H ORNER et al. (2016a) setting through a differences in differences approach. between application and systemic effects. Many authors Do economies experiencing faster digitalization increase differentiate between two levels of higher order effects or decrease their environmental throughput compared to (i.e., second and third order). This categorization has economies with slower digitalization over the same time first been introduced by B ERKHOUT and H ERTIN (2001). period? H ILTY and A EBISCHER (2015) see the life cycle effects of production, use and disposal at the first order level, Nearly no studies exist on the macro level to allow for referring to them as direct effects. At the second order, capturing both negative and positive effects of digital- so-called enabling effects include process optimization, ization on biosphere use with an explicit differentiation media substitution, and externalization of control. Me- between production and consumption effects. One of few dia substitution means that with increased digitalization existing studies providing such evidence is S CHULTE information is distributed through new forms of media, et al. (2016), who assess the effects of increasing ICT use for example replacing books by tablets or Kindles, or re- on energy consumption in production through a cross- placing audio playback devices with streaming services. country panel data analysis. Our approach adds to the Externalization of control captures all processes that are existing body of knowledge by explicitly differentiating out of the hands of consumers or businesses using a between the environmental effects of increasing digital- specific technology, such as the need to regularly acquire ization caused by production and consumption. new hardware due to software update cycles. Third order effects are labelled systemic effects and encompass II. L ITERATURE REVIEW rebound effects, emerging risks and transition towards sustainable patterns of production and consumption. The life-cycle impacts of ICT use have been subject to a number of investigations. The production and application Accurately measuring the effect of digitalization on of ICTs themselves require a substantial input of raw the environment using these different classifications of materials and energy. However, existing studies on the effects is a challenging task (H EIJUNGS et al., 2009; natural resource costs associated with the production and F INKBEINER et al., 2014; M ILLER and K EOLEIAN, employment of ICTs are limited. 2015). It would be possible to investigate the life-cycle impacts, as well as the productivity increases, on a According to a recent study by M ALMODIN et al. (2018), microeconomic scale by estimating changes in environ- the use of ICTs accounts for 0,5% of global material mental productivity for the production of specific goods use. However, these numbers vary highly concerning the and services, or by measuring the energy and resources specific material being considered. Indeed, ICT accounts used to produce and use ICTs. However, even this is for 80-90% of depletion of certain materials, such as a difficult task (H ILTY, 2015). Estimating the effect of indium, gallium and germanium. Several studies have digitalization on labor productivity and economic growth investigated ICT’s effects on the demand for energy. is even more difficult, as various other factors come A NDRAE (2015) estimate that the use of ICTs accounted into play. Digitalization takes place within a certain for 8% of global energy use in 2010 and expect this share historic situation of the world economy. The fact that to rise strongly in the coming years. VAN H EDDEGHEM economies undergo a multitude of transformations and et al. (2014) estimate that the share of global energy 2 III Methodology consumption due to a certain subset of ICTs (communi- and consumers have more income to spend on other cation networks, personal computers, and data centres) goods and services. This implies that the energy and grew from 3.9% to 4.6% in 2012 alone. These numbers natural resources saved through increased productivity certainly show that such direct effects of ICTs must be can be used for other productive purposes, either to considered when estimating whether growing levels of produce more of the same goods and services or to digitalization serve to increase or decrease environmental produce additional other goods and services. Many au- throughput over time. thors have found evidence for rebound effects in different areas of the digital economy, including C OROAMA et al. Many examples of increases in environmental produc- (2012), M OKHTARIAN (2009), and A RNFALK et al. tivity through increased ICT use have been observed. (2016) for virtual meetings and video conferencing, and More efficient movement of robots can decrease their B ÖRJESSON R IVERA et al. (2014a) for production of energy use in manufacturing.1 In one of the first studies ICT hardware. Another example is in the increasing on the subject, L ENNARTSON and B ENGTSSON (2016) efficiency of processing units. The so called “Koomey’s find that improvements in robotic movement efficiency law” states that the energy efficiency of processing units are associated with a decrease in energy consumption of doubles every 1.5 years (KOOMEY et al., 2011). If the up to 40%. C OROAMA et al. (2012) find that replacing amount of processing units would stay the same, energy in-person business meetings with virtual conferences use by such units would decrease along a logarithmic could decrease the carbon footprint of such meetings by pattern with a half-life of 1.5 years, quickly approaching 37% - 50%. Electronic invoicing can decrease energy very low levels. But at the same time, the amount of use compared with traditional invoicing methods and computations of the processing units produced and used switching to online newspapers and magazines rather grows over time. The bottom line is that the growth in the than consuming conventional print publications can have number of processing units is higher than the rate of in- a substantial and far-reaching positive environmental crease in productivity. This impressive growth can at least impacts as well (M OBERG et al., 2010). Various studies partly be explained by technological developments. The have investigated the environmental effects of online newer, more efficient units allow for media substitution, vs. offline retailing at the micro level (H ORNER et al., resulting in more aggregate use. For example, with the 2016b; M ANGIARACINA et al., 2015; L OON et al., 2015). larger and more energy-intensive processing units in the Which one is more efficient depends on various factors 1990s, it was simply not feasible to invent a functional such as population density and the specific conditions smartphone. of delivery, implying that online shopping can actually be more environmentally harmful than traditional retail. In summary, the existing research suggests that digi- The same is true for media substitution regarding online talization has the potential to increase environmental video streaming compared to renting DVDs (S HEHABI productivity in many economic areas. But even if this et al., 2014). holds true, the production and maintenance of a digital infrastructure (first order effects) and potential rebound In addition, rebound effects can also be observed. In and other higher order effects could outweigh the benefits a nutshell, rebound effects refer to the phenomenon described by the first mechanism, possibly leading to where an increase in production efficiency leads to a an increase in total environmental throughput after all. lowering of consumer prices. This, in turn, leads to Only by an aggregated analysis that takes all mechanisms higher demand, resulting in an expansion of total pro- into account can we investigate whether digitalization duction. The corresponding increase in natural resource increases or decreases environmental throughput. use may overcompensate the ecological efficiency gains, leading to a net-increase of biosphere use (B ERKHOUT et al., 2000). An overview of the literature on rebound III. M ETHODOLOGY effects in regards to ICT usage is provided by G OSSART (2014). If digitalization helps to increase energy and A. Estimation method resource productivities, the costs for energy and other resources decrease. Economically speaking, this means This analysis relies on the recently introduced Group that the production function shifts downwards, a new Fixed Effects (GFE) estimator. It was developed by B ON - HOMME and M ANRESA (2015) and has been used by equilibrium of lower price and higher quantity is reached, few studies so far (G RUNEWALD et al., 2017; KOPP and 1 Note that this refers to robot steerage, not the introduction of robots. D ORN, 2018). Its development has been motivated by 3 III Methodology problems with conventional panel fixed-effect analysis, B. Identification strategy which implicitly assumes unobserved heterogeneity be- tween countries to stay constant over time. To tackle this The effects of digitalization can be decomposed into issue, in the first stage the GFE assembles all countries those effects related to the production of goods and into groups, according to the changes in the observables. those related to the consumption of goods. The former In the second stage the panel estimation is exercised, includes increased technical efficiency due to the use of supplemented by dummy variables for each of the groups ICT in production processes, while the latter refers to instead of individual country effects. The GFE also changing consumption patterns, such as the switch from solves the problem of low degrees of freedom in fixed- conventional analogous and offline practices to digital effect panel estimations, which require a big number of and potentially online ones. To allow for a differentiation dummy variables (one dummy per section, e.g. country). between the effects of these two areas we approach Since the GFE bundles all countries within a relatively the question from two sides, first from the production small number of groups (all literature reviewed that perspective and then from the consumption perspective. employs the GFE estimator relies on less than ten groups (B ONHOMME and M ANRESA, 2015; G RUNEWALD et al., To measure production-side effects, we measure all re- 2017; KOPP and D ORN, 2018)), the number of covariates sources that are used in one country’s industrial pro- decreases strongly. duction and investigate how deeply resource use in that country is affected by the country’s level of industrial Four control variables are included, following digitalization. This level of digitalization is captured by G RUNEWALD et al. (2017) and KOPP and D ORN the total yearly investments of all firms in information (2018): the share of the population living in urban areas, and communication technology. Environmental through- as well as the shares of the GDP being generated in put is proxied by each country’s CO2 emissions. the agriculture, the manufacturing, and service sectors, respectively. This leads to the following equation to The analysis on the consumption side considers all be estimated, for both production and consumption resources used during the production of the goods con- analyses: sumed in one certain country (even if produced abroad) and associates them with a measure of digitalization on ln CO2 = ln Digi + ln GDP the consumer side in one country. Defining an aggregate X + ln GDP Digi + X measure to account for all aspects of digitalization on the 4 (1) consumer side is complicated, as it encompasses several + GF E + c + "; i dimensions. This means, generally speaking, that new or i=1 additional products and services are consumed that were not previously imagined to complement or substitute where CO2 stands for climate gas emissions and Digi existing ones. This also includes the purchasing process for the level of digitalization.2 GDP denotes each itself, which is involved in every purchasing act, and country’s GDP. GDP Digi is a cross term capturing might therefore serve as an effective proxy for the interaction effects between GDP and the measure of consumers’ openness to new technology and willingness digitalization on the outcome variable.3 X is the vector to use them. This paper therefore proxies digitalization of control variables (x1 ; x2 ; x3 ; x4 ) and the vector of on the consumption side by the share of individuals who the respective coefficients (1 ; 2 ; 3 ; 4 ). GF Ei stands ordered consumer articles online during the last three for the coefficients of the GFE-groups, of which one is months.4 The environmental throughput caused by the omitted from the estimation due to collinearity. c is a consumption of goods in a country is proxied by the constant and " an error term. The dependent and the sub-index CO2 emissions of the ecological footprint. The key explaining variables are described in the following critical difference between the two measures of biosphere sections. use is that the former captures the CO2 emissions pro- duced within the countries while the latter also accounts 2 The identification of the digitalization effect is laid out in the next for emissions imported and exported through trade. section. 3 This allows for the possibility that the effect of one of the variables depends on the state of the other, i.e. that digitalization may affect 4 As this decision is controversial some critical reflections and ideas biosphere use in richer countries systematically different than it does on alternatives to this measure are provided in the “Outlook” section in less well-off countries. IV-C. 4 IV Results, discussion, and outlook C. Data and exported. Since the database provides the EF in the form of “global hectares”, it was converted back to CO2 On the production side the key explanatory variable, emissions, based on average sequestration capacity of the digitalization in a country’s production, is the sum forests, which is the measure used to construct the EF in of all investments made by all companies into ICT the first place. The control variables are the same as for infrastructure and software that is used for more than the production side. Descriptive statistics of all variables one year. This variable is provided by the OECD (2017). entering the consumption side regression are provided The dependent variable is CO2 emissions generated in Table (II). The values diverge slightly from the ones by all production processes carried out in one country, provided in table (I); because – due to different data proxying the environmentally detrimental output caused availabilities – the countries included in the analysis vary by production. This variable, as well as the controls, were slightly. taken from the World Development Indicators, provided by the World Bank. Descriptive statistics of all variables TABLE II: Summary statistics of all variables entering entering the production side regression are provided in the consumption side regression. Table (I). mean sd min max TABLE I: Summary statistics of all variables entering lnEFP 5.77 1.36 2.43 7.88 lnOnlineShopping -1.57 0.85 -3.91 -0.33 the production side regression. lnGDP 22.00 1.52 18.03 24.39 Urban 72.62 12.84 49.69 97.82 mean sd min max Manu 13.72 4.48 4.08 24.83 ln CO2 6.56 1.33 4.65 9.96 Agri 2.36 1.68 0.25 9.03 lnICTinvest 2.73 0.39 1.03 3.48 Serv 62.75 7.03 42.92 78.31 lnGDP 23.10 1.20 21.03 26.04 Urban 77.56 7.58 57.92 88.91 177 observations Manu 17.52 4.15 9.06 27.80 Agri 2.95 2.08 0.55 11.68 Serv 61.39 6.05 44.60 76.38 The consumption side is estimated as follows: 318 observations ln CO2C = C ln OnlineShopping + C ln GDP So on the production side the following model is esti- + C ln GDP OnlineShopping mated: X GF E + c + " ; + C XC (3) 4 ln CO2P = P ln ICTInvest + P ln GDP + i C C i=1 X GF E + c + " ; + P ln GDP ICTInvest + P XP 4 (2) where subscript C indicates the consumption side coef- + i P P ficients. i=1 The countries entering the analysis, their descriptives where subscript P indicates the production side coeffi- and group assignments are displayed in Table (V) in cients. the appendix. For the production side, the panel covers On the consumption side the key explanatory variable is the years 1990-2009 and for the consumption side 2008- the share of people who used the internet to purchase 2014. The number of observations per group is displayed goods or services during the last three months. The in Table (IV) in the appendix. data were provided by EuroStat, the statistics service of the European Commission (E URO S TAT, 2018). We IV. R ESULTS , DISCUSSION , AND OUTLOOK generated the dependant variable based on the carbon sub-index of the ecological footprint (EF), provided by A. Results the Ecological Footprint Network (L IN et al., 2016). Unlike other accounts of emissions the EF captures not Results of both regressions are displayed in table (III). only the ones produced in one country, but also accounts The inclusion of the interaction terms impedes a straight- for the ecological backpack carried by all goods imported forward interpretation by simply observing the estimated 5 IV Results, discussion, and outlook coefficients. To facilitate an intuitive interpretation, fig- Fig. 1: Effects of lnICT-Investments and lnGDP on CO2 emissions produced within the country. ures (1) and (2) visualize the effect of digitalization within the range of the GDP and digitalization levels in the data in the form of heatmaps. TABLE III: Regression results. (1) (2) VARIABLES ln CO2 ln EF P lnICTinvest -3.835*** (0) lnGDP ICTinvest 0.171*** (0) lnOnlineShopping -2.055*** (6.88e-05) lnGDP OnlineShopping 0.0826*** (0.000356) lnGDP 1.577*** -0.190 (0.000139) (0.706) lnGDP2 -0.0195** 0.0247** (0.0456) (0.0331) The shading indicates the size of the EF of the respective measure. Urban 0.0290*** 0.00841*** The dots represent the distribution of lnICT-investments and ln GDP (0) (0.00283) of all countries in our sample. Manu 0.00705* 0.0412*** (0.0888) (4.63e-06) Agri 0.0669*** 0.157*** Fig. 2: Effects of lnOnline-Shopping and lnGDP on CO2 emissions (0) (4.83e-08) (including imported emissions). Serv -0.0231*** 0.00433 (2.88e-09) (0.549) Assignment1 -0.438*** 0.742*** (0) (3.61e-09) Assignment2 0.448*** 0.879*** (0) (0) Assignment3 0.734*** 0.801*** (0) (0) Assignment4 0.253*** 0.424*** (0) (1.64e-05) " -21.12*** -4.855 (6.41e-06) (0.395) Observations 318 177 R2 0.989 0.957 pval in parentheses *** p<0.01, ** p<0.05, * p<0.1 B. Discussion The shading indicates the size of the EF of the respective measure. The dots represent the distribution of ln Online Shopping and Figure (1) shows that the effects of digitalization in ln GDP of all countries in our sample. production on emissions depend on the level of GDP in a given country. In the lower quartile of the GDP distribution, increasing levels of digitalization lead to a reduction in environmental throughput, while for the For both production and consumption, the effects of upper quartile the opposite is true. At the sample mean digitalization on environmental throughput appear to the effect is more ambiguous. be relatively small compared with the effects of GDP. Nevertheless, the effect may still be substantial. To fully On the consumption side (Figure 2) the effects appear understand the marginal effects, we differentiate the clearer: digitalization leads to a reduction in environmen- parametrized equations with respect to their respective tal throughput. However, the effect grows weaker as GDP measurements of digitalization, ICT investments, and increases. online shopping behaviour. 6 Bibliography Differentiating equation (1) with respect to Digi yields V. C ONCLUSION To the best of the authors’ knowledge, this paper is one of @ ln CO2 1 the first to analyze the impact of increased digitalization = Digi 1 + GDP on environmental throughput at the macroeconomic level. @Digi GDP Digi (4) + It is the first paper that differentiates between consump- = tion and production side effects. We make use of a unique Digi dataset linking national CO2 emissions to digitalization On the production side, equation (4) yields -0.22 at the in production and net CO2 levels after trading to data sample mean. At this point in the data, an increase in ICT on digitalization levels in consumption. To answer the investments by 10% (that is, by 1.65) would decrease research questions, we apply the newly developed Group ln CO2 by 0.3664. This amounts to 5.6% of all CO2 Effects estimator. emissions caused in production, ceteris paribus. The results of this study provide the first evidence of The computation is the same on the demand side. At its kind that the CO2 -decreasing benefit of digitalization the sample mean, an increase in Online Shopping by might outweigh its environmental costs. Indeed, there 10% would decrease ln CO2 caused through a country’s is evidence that the net effect of digitalization on CO2 consumption by 3.4%, ceteris paribus. emissions and environmental throughput are in fact pos- itive. However, supplemental research is required to test C. Outlook the robustness of these results against a wider range of digitalization measures on the consumption side. The global effects of digitalization are widespread and far-reaching in scope, encompassing multiple industries B IBLIOGRAPHY and sectors (H ILTY et al., 2014; W ILLIAMS, 2011; B ÖRJESSON R IVERA et al., 2014b). This work specif- A NDRAE, Anders S G (2015). “Method based on market ically focuses upon two aspects central to digitalisation. changes for improvement of comparative attributional On the production side, we focus on the life cycle life cycle assessments”. en. In: The International Jour- impacts reflected through the ever-increasing use of nal of Life Cycle Assessment 20.2, pp. 263–275. ISSN: digital hardware, proxied by firms’ ICT investments. On 0948-3349, 1614-7502. the consumption side, digitalization manifests itself in a A RNFALK, Peter, Ulf P ILEROT, Per S CHILLANDER, and general proclivity and openness of consumers towards Pontus G R ÖNVALL (2016). “Green IT in practice: vir- digitalization (second and third order effects), which tual meetings in Swedish public agencies”. In: Journal can be measured effectively through online shopping of Cleaner Production. Advancing Sustainable Solu- behaviour. tions: An Interdisciplinary and Collaborative Research However, one potential caveat of this study that must be Agenda 123, pp. 101–112. ISSN: 0959-6526. taken into account is the potentially questionable validity A RVESEN, Anders, Ryan M. B RIGHT, and Edgar G. of online shopping behaviour as a proxy for consumer H ERTWICH (2011). “Considering only first-order ef- openness towards digital services. To account for this fects? How simplifications lead to unrealistic technol- potential weakness, the next step is to expand upon the ogy optimism in climate change mitigation”. en. In: study by testing our results for robustness by adding more Energy Policy 39.11, pp. 7448–7454. ISSN: 03014215. dimensions on the consumer side, in order to provide a more complete picture of consumer-side digitalization. B ERKHOUT, Frans and Julia H ERTIN (2001). Impacts of This can be achieved by executing our existing analysis Information and Communication Technologies on En- with an index composed of data on more second-order vironmental Sustainability: speculations and evidence. effects, including media substitution, process optimiza- report to the OECD. Tech. rep. tion, and externalization of control. These can include – (2004). “De-materialising and re-materialising: digital data on internet penetration rates, average internet speed, technologies and the environment”. en. In: Futures share of Netflix subscriptions vs. DVD rentals, quantity 36.8, pp. 903–920. ISSN: 00163287. of decentralised energy systems/smart grids, sales of Amazon Kindles, inter alia. These data would then need to be normalized and aggregated to an index. 7 Bibliography B ERKHOUT, Peter H G, Jos C M USKENS, and Jan W René S CHEUMANN, Laura S CHNEIDER, and V ELTHUIJSEN (2000). “Defining the rebound effect”. Kirana W OLF (2014). “Challenges in Life Cycle In: Energy policy 28.6, pp. 425–432. Assessment: An Overview of Current Gaps and Research Needs”. In: Background and Future B ONHOMME, Stéphane and Elena M ANRESA (2015). Prospects in Life Cycle Assessment. Ed. by “Grouped Patterns of Heterogeneity in Panel Data”. In: Walter K L ÖPFFER. Dordrecht: Springer Netherlands, Econometrica 83.3, pp. 1147–1184. ISSN: 00129682. pp. 207–258. ISBN: 978-94-017-8696-6 978-94-017- B ÖRJESSON R IVERA, Miriam, Cecilia H ÅKANSSON, 8697-3. Åsa S VENFELT, and Göran F INNVEDEN (2014a). F ORSCHUNG, Bundesministerium für Bildung und “Including second order effects in environmental as- (2014). Die neue High-Tech Strategie der Bun- sessments of ICT”. In: Environmental Modelling & desregierung: Innovationen für Deutschland. Tech. Software. Thematic issue on Modelling and evaluating rep. Berlin. the sustainability of smart solutions 56, pp. 105–115. ISSN : 1364-8152. G E SI and ACCENTURE (2015). Smarter 2030. ICT Solu- tions for 21st Century Challenges. Tech. rep. Brüssel. – (2014b). “Including second order effects in environ- mental assessments of ICT”. In: Environmental Mod- G OSSART, Cédric (2014). “Rebound effects and ict: elling & Software. Thematic issue on Modelling and A review of the literature”. In: ICT Innovations for evaluating the sustainability of smart solutions 56, Sustainability. Ed. by Lorenz M. H ILTY and Bernard pp. 105–115. ISSN: 1364-8152. A EBISCHER. Berlin: Springer International Publish- ing. B ÖRJESSON R IVERA, Miriam, Cecilia H ÅKANSSON, Åsa S VENFELT, and Göran F INNVEDEN (2014). “In- G RUNEWALD, Nicole, Stephan K LASEN, Inmaculada cluding second order effects in environmental as- M ART ÍNEZ -Z ARZOSO, and Chris M URIS (2017). sessments of ICT”. In: Environmental Modelling & “The Trade-off Between Income Inequality and Car- Software. Thematic issue on Modelling and evaluating bon Dioxide Emissions”. In: Ecological Economics the sustainability of smart solutions 56, pp. 105–115. 142, pp. 249–256. ISSN: 09218009. ISSN : 1364-8152. H AKANSSON, Cecilia and Göran F INNVEDEN (2015). B UNDESREGIERUNG, Die (2014). “Digitale Agenda “Indirect Rebound and Reverse Rebound Effects in 2014–2017”. In: ed. by Bundesministerium für the ICT-sector and Emissions of CO2”. In: Joint Wirtschaft und E NERGIE, Bundesministerium des I N - Conference on 29th International Conference on Infor- NEREN , and Bundesministerium für Verkehr und dig- matics for Environmental Protection/3rd International itale I NFRASTRUKTUR. Conference on ICT for Sustainability (EnviroInfo and ICT4S), SEP 07-09, 2015, Univ Copenhagen, Copen- C OROAMA, Vlad C, Lorenz M H ILTY, and Martin B IR - hagen, DENMARK, pp. 66–73. TEL (2012). “Effects of Internet-based multiple-site conferences on greenhouse gas emissions”. In: Telem- H EIJUNGS, Reinout, Gjalt H UPPES, and Jeroen G UIN ÉE atics and Informatics. Green Information Communica- (2009). “A scientific framework for LCA”. In: Deliv- tion Technology 29.4, pp. 362–374. ISSN: 0736-5853. erable (D15) of work package 2. E NERGIE, Bundesministerium für Wirtschaft und (2015). H ILTY, Lorenz M (2015). “The role of ICT in labor Industrie 4.0 und Digitale Wirtschaft. Impulse für productivity and resource productivity–are we using Wachstum, Beschäftigung und Innovation. Berlin: technological innovation the wrong way?” In: Novat- BMWi. ica 234/ 2015.Special Issue for the 40th anniversary of the Journal, pp. 32–39. E URO S TAT (2018). Digital Economy and Society Dataset. H ILTY, Lorenz M. and Bernard A EBISCHER (2015). “ICT for Sustainability: An Emerging Research Field”. F INKBEINER, Matthias, Robert ACKERMANN, In: ICT Innovations for Sustainability. Ed. by Lorenz Vanessa BACH, Markus B ERGER, Gerhard M. H ILTY and Bernard A EBISCHER. Vol. 310. Cham: B RANKATSCHK, Ya-Ju C HANG, Marina G RINBERG, Springer International Publishing, pp. 3–36. ISBN: Annekatrin L EHMANN, Julia M ART ÍNEZ -B LANCO, 978-3-319-09227-0 978-3-319-09228-7. Nikolay M INKOV, Sabrina N EUGEBAUER, 8 Bibliography H ILTY, Lorenz M and Jan B IESER (2017). Opportunities M ANGIARACINA, Riccardo, Gino M ARCHET, Sara P ER - and Risks of Digitalization for Climate Protection in OTTI , and Angela T UMINO (2015). “A review of Switzerland. Tech. rep. Zürich. the environmental implications of B2C e-commerce: a logistics perspective”. In: International Journal of H ILTY, Lorenz M, Bernard A EBISCHER, and Andrea E Physical Distribution & Logistics Management 45.6, R IZZOLI (2014). “Modeling and evaluating the sus- pp. 565–591. ISSN: 0960-0035. tainability of smart solutions”. en. In: Environmental Modelling & Software 56, pp. 1–5. ISSN: 13648152. M ILLER, Shelie A. and Gregory A. K EOLEIAN (2015). “Framework for Analyzing Transformative Technolo- H ORNER, Nathaniel C., Arman S HEHABI, and Inês L. gies in Life Cycle Assessment”. en. In: Environmental A ZEVEDO (2016a). “Known unknowns: indirect en- Science & Technology 49.5, pp. 3067–3075. ISSN: ergy effects of information and communication tech- 0013-936X, 1520-5851. nology”. en. In: Environmental Research Letters 11.10, p. 103001. ISSN: 1748-9326. M OBERG, Asa, Clara B ORGGREN, Goran F INNVEDEN, and Sara T YSKENG (2010). “Environmental impacts H ORNER, Nathaniel C, Arman S HEHABI, and Inês L of electronic invoicing”. In: Progress in Industrial A ZEVEDO (2016b). “Known unknowns: indirect en- Ecology, an International Journal 7.2, pp. 93–113. ergy effects of information and communication tech- nology”. en. In: Environmental Research Letters 11.10, M OKHTARIAN, Patricia (2009). “If telecommunication is p. 103001. ISSN: 1748-9326. such a good substitute for travel, why does congestion continue to get worse?” In: Transportation Letters 1.1, KOOMEY, Jonathan, Stephen B ERARD, Marla pp. 1–17. ISSN: 1942-7867. S ANCHEZ, and Henry W ONG (2011). “Implications of historical trends in the electrical efficiency of OECD (2017). ICT investment Dataset. computing”. In: IEEE IEEE Annals of the History of P OHL, Johanna, Lorenz M. H ILTY, and Matthias Computing 33.3, pp. 46–54. F INKBEINER (2019). “How LCA contributes to the KOPP, Thomas and Franziska D ORN (2018). “Social environmental assessment of higher order effects of equity and ecological sustainability - can the two be ICT application: A review of different approaches”. en. achieved together?” In: cege Discussion Papers 357, In: Journal of Cleaner Production. ISSN: 09596526. pp. 1–32. R ESSOURCENEFFIZIENZ, Zentrum (2017). L ENNARTSON, B and K B ENGTSSON (2016). “Smooth Ressourceneffizienz durch Industrie 4.0. Tech. rep. robot movements reduce energy consumption by up to Berlin. 30 percent”. In: European Energy Innovation Spring, RØPKE, Inge (2012). “The unsustainable directionality p. 38. of innovation – The example of the broadband tran- L IN, David, Laurel H ANSCOM, Jon M ARTINDILL, sition”. en. In: Research Policy 41.9, pp. 1631–1642. Michael B ORUCKE, Lea C OHEN, Alessandro G ALLI, ISSN : 00487333. Elias L AZARUS, Golnar Z OKAI, Katsunori I HA, Derek S CHULTE, Patrick, Heinz W ELSCH, and Sascha E ATON, and Mathis WACKERNAGEL (2016). Working R EXH ÄUSER (2016). “ICT and the Demand for Guidebook to the National Footprint Accounts: 2016 Energy: Evidence from OECD Countries”. en. Edition. Tech. rep. In: Environmental and Resource Economics 63.1, L OON, Patricia van, Lieven D EKETELE, Joost D E - pp. 119–146. ISSN: 0924-6460, 1573-1502. WAELE, Alan M C K INNON , and Christine RUTHER - S HEHABI, Arman, Ben WALKER, and Eric M ASANET FORD (2015). “A comparative analysis of carbon emis- (2014). “The energy and greenhouse-gas implications sions from online retailing of fast moving consumer of internet video streaming in the United States”. goods”. en. In: Journal of Cleaner Production 106, en. In: Environmental Research Letters 9.5, p. 54007. pp. 478–486. ISSN: 09596526. ISSN : 1748-9326. M ALMODIN, Jens, Pernilla B ERGMARK, and Sepideh VAN H EDDEGHEM, Ward, Sofie L AMBERT, Bart L AN - M ATINFAR (2018). “A high-level estimate of the ma- NOO , Didier C OLLE, Mario P ICKAVET , and Piet D E - terial footprints of the ICT and the E&M sector”. In: MEESTER (2014). “Trends in worldwide ICT electric- pp. 148–168. 9 VI Appendix ity consumption from 2007 to 2012”. en. In: Computer Communications 50, pp. 64–76. ISSN: 01403664. W ILLIAMS, Eric (2011). “Environmental effects of infor- mation and communications technologies”. In: Nature 479.7373, pp. 354–358. ISSN: 0028-0836, 1476-4687. W ISSENSCHAFT, Promotorengruppe Kommunikation der Forschungsunion Wirtschaft – (2013). “Umset- zungsempfehlungen für das Zukunftsprojekt Industrie 4.0”. In: Abschlussbericht des Arbeitskreises Industrie 4.0. VI. A PPENDIX TABLE IV: Observations per group. N production N consumption 1 52 41 2 73 152 3 46 35 4 117 77 5 75 52 10 VI Appendix TABLE V: Mean values of key variables. CO2 ICTinvest assignment prod. EF OnlineShopping assignment cons. GDP Australia 1367 18.54 4 1838 1.3e+10 Austria 289 11.32 2 391 0.36 2 5.5e+09 Belgium 399 544 0.30 2 7.1e+09 Bulgaria 197 267 0.10 5 9.0e+08 Canada 2318 16.75 2 3126 2.5e+10 Croatia 91 120 0.13 4 1.0e+09 Czechrepublic 526 728 0.17 3 3.8e+09 Denmark 232 20.51 5 306 0.57 4 4.5e+09 Estonia 82 106 0.17 1 3.6e+08 Finland 282 11.09 4 268 0.41 1 3.6e+09 France 1513 15.49 5 1963 0.39 2 3.4e+10 Greece 312 423 0.13 4 4.0e+09 Hungary 210 273 0.20 3 2.6e+09 Ireland 187 10.15 3 247 0.34 4 3.3e+09 Italy 1829 12.48 2 2359 0.11 2 3.2e+10 Japan 4973 13.27 5 6403 8.1e+10 Latvia 31 42 0.14 5 4.0e+08 Lithuania 62 79 0.11 5 6.8e+08 Luxembourg 37 49 0.51 2 7.0e+08 Malta 10 13 0.32 5 1.3e+08 Montenegro 10 13 0.04 6.8e+07 Netherlands 694 880 0.52 2 1.2e+10 Newzealand 130 20.56 4 160 1.7e+09 Norway 279 292 0.55 1 8.4e+09 Poland 1478 1948 0.20 2 8.8e+09 Portugal 201 276 0.12 3 3.5e+09 Romania 461 626 0.04 4 3.6e+09 Serbia 243 324 0.03 7.6e+08 Slovakia 168 237 0.23 2 1.7e+09 Slovenia 71 101 0.19 2 8.4e+08 Southkorea 1996 11.17 3 2641 1.2e+10 Spain 1282 13.78 4 1680 0.20 3 2.0e+10 Sweden 241 22.65 1 322 0.52 5 6.6e+09 Switzerland 163 16.66 1 229 0.62 7.2e+09 Turkey 1554 2061 0.05 4 1.6e+10 Unitedkingdom 2076 23.58 4 2669 0.63 2 3.1e+10 Unitedstates 20518 29.54 2 26680 1.8e+11 Blank spots in the table refer to non-observed values in the respective dataset. 11