=Paper= {{Paper |id=Vol-3182/paper13 |storemode=property |title=Empirically Grounded Agent-Based Policy Evaluation of the Adoption of Sustainable Lighting under the European Ecodesign Directive |pdfUrl=https://ceur-ws.org/Vol-3182/paper13.pdf |volume=Vol-3182 |authors=Gido Schoenmacker,Wander Jager,Rineke Verbrugge |dblpUrl=https://dblp.org/rec/conf/jurix/SchoenmackerJV21 }} ==Empirically Grounded Agent-Based Policy Evaluation of the Adoption of Sustainable Lighting under the European Ecodesign Directive== https://ceur-ws.org/Vol-3182/paper13.pdf

Empirically grounded agent–based policy evaluation
of the adoption of sustainable lighting under the
European Ecodesign Directive
Gido H. Schoenmacker1 , Wander Jager2 and Rineke Verbrugge3
1
Independent researcher, Amsterdam, the Netherlands
2
University of Groningen, Groningen, the Netherlands
3
University of Groningen, Groningen, the Netherlands

Abstract
Twelve years ago, the European Union began with the gradual phase-out of energy-inefficient
incandescent light bulbs under the Ecodesign Directive. In this work, we implement an agent-
based simulation to model the consumer behaviour in the EU lighting market with the goal
to explain consumer behaviour and explore alternative policies. Agents are based on the
Consumat II model, have individual preferences based on empirical market research, gather
experience from past actions, and socially interact with each other in a dynamic environment.
Our findings suggest that the adoption of energy–friendly lighting alternatives was hindered
by a low level of consumer interest combined with high–enough levels of satisfaction about
incandescent bulbs and that information campaigns can partially address this. These findings
offer insight into both individual-level driving forces of behaviour and society–level outcomes
in a niche market. With this, our work demonstrates the strengths of agent–based models for
policy generation and evaluation.

Keywords
Agent–based modelling, Policy evaluation, Innovation diffusion

1. Introduction
Twelve years ago, the European Union (EU) began with the gradual phase–out of energy–
inefficient incandescent light bulbs under the Ecodesign Directive (2009/125/EC) [1]. In
2019 it was estimated that this directive had reduced energy expenditure of household
lamps by up to 60%, saving the average EU family e130 annually [2]. Since this outcome
benefits consumers, it might be expected that EU legislation was unnecessary for the
adoption of energy–friendly lighting. The observed reality, however, was that household
consumers had for years been hesitant to adopt more energy–friendly lighting options,
prompting EU legislation [3].
In this work, we implemented an agent–based simulation to model and explain the
consumer behaviour in the EU lighting market. Because individual preferences and
AMPM’21: First Workshop in Agent-based Modeling & Policy-Making, December 8, 2021, Vilnius,
Lithuania
Envelope-Open gido@schoenmacker.nl (G. H. Schoenmacker)
Orcid 0000-0003-3946-928X (G. H. Schoenmacker); 0000-0003-3829-0106 (R. Verbrugge)
© 2021 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution
4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073
CEUR Workshop Proceedings (CEUR-WS.org)
complex social interaction affect this behaviour, agent–based models are able to capture
dynamics that would be difficult for traditional statistical models [4]. Our agents are
based on the Consumat II model [5]: They have individual preferences based on consumer
market research [6], gather experience from past actions, and socially interact with one
another in a dynamic environment. We simulated multiple scenarios to examine consumer
behaviour under different policies. In doing this, we aimed to answer two main questions:
(i) Can we explain the reluctance of the consumer to switch to energy–friendly lighting?
And (ii) Compared to banning household incandescent lighting, are there alternative
policies that might have been equally successful?

2. Methods
2.1. Model
The Consumat II model – Consumat for short – is based on human psychological meta–
theory and formulates drivers of complex behaviour in terms of needs, satisfaction, and
uncertainty. It has been successfully applied to numerous, mostly environmentally–related,
issues [7, 8, 9, 10, 11, 12]. When an action is required, a Consumat agent can engage
in one of four cognitive strategies based on its position along two axes: its degree of
satisfaction and its degree of certainty. The four strategies, in no particular order, are (i)
repetition, (ii) imitation, (iii) deliberation, and (iv) social comparison.
Agents with a low degree of satisfaction are willing to expend effort to make changes,
whereas high satisfaction leads to lower–effort strategies. Certainty relates to the degree
of belief in expected outcomes when taking actions. A low level of certainty stimulates
agents towards social strategies [13, 14], whereas high certainty encourages individually
determined strategies [15, 16]. Agents with high satisfaction and certainty will engage
in repetition, being satisfied with earlier actions and confident about results. High
satisfaction combined with low certainty results in imitation, looking to peers for insurance
in an uncertain environment. Agents with low satisfaction and high certainty engage
in deliberation, trusting themselves to analyse the market and improve their situation.
Finally, low satisfaction and certainty result in social comparison, where behaviour of
peers is more closely examined and only copied if it is expected to increase satisfaction.
In our model, the behaviours are operationalised as follows. Repetition simply replaces
a broken bulb with the same type of bulb. If a candidate bulb is no longer available,
the agent will perform deliberation instead. Imitation selects a random but similar peer
and replaces the broken bulb with a random available bulb. Peer difference is defined by
kpA − pB k1 where pN is the preference vector for agent N (see below). If a randomly
selected peer is not similar enough, the similarity requirement is loosened and a new
peer is selected until a similar enough peer has been selected. Deliberation considers all
available lamps and selects the one that will result in the highest satisfaction. Social
comparison selects a peer in the same way as the imitation action and from its inventory
selects the lamp that will result in the highest satisfaction.
2.2. Model parameters
To determine satisfaction, agents had individual preferences for lighting market–specific
characteristics that were shown to significantly affect the decision process when purchasing
light bulbs in market research by Kattenwinkel [6]. In short, 97 Dutch individuals
were questioned about their lamp purchase habits and considerations. After removing
respondents with missing data, the resulting number of individuals was 87. Examples of
questions included the number of lamps respondents had in use and to which degree the
opinion of social contacts affected lamp selection. The full list of questions is available
in [17].
Agent preferences were initialised based on archetypes from Kattenwinkel. In total,
there are 87 archetypes, where each archetype is an 11–dimensional vector representing
the number of lamps of an agents needs, tolerances for functional and colour requirements,
focus on energy usage for financial and environmental reasons, social–mindedness and
–agreeability, and baseline levels of experience/satisfaction with 3 different lighting types
(incandescent, CFL, and light–emitting diode (LED)). In total 1000 agents are instantiated
by sampling uniformly from (0.95v, 1.05v) for every characteristic v from a single random
archetype vector.
The lamp models and their properties in the models can be found in Supplemental
Table S1. Lamp lifetime is drawn from N (`, 5` ) where ` is the mean lifetime in months
from Supplemental Table S1. Every lamp is assumed in use for 6 hours daily. Because no
clear enough price progression is available, we assumed that LED lights become available
in 2006 and then every year until 2020 decrease in price by 10%. This means that
their prices drop from e12.50–30.00 in 2006 to e3.20–7.60 in 2020. Since these may be
influential assumptions, the price progression is varied by a random multiplication factor
of 0.5–2 that governs LED pricing progress. Similarly, the energy efficiency of each LED
bulb is assumed to grow with 5% from 2007–2020 to a maximum of 99% efficiency and is
subject to a second, independent random factor 0.5–2 that governs LED innovation.
These changes to LED lamp properties occur in every scenario. Changes to the pricing
and availability of incandescent lamps are scenario–dependent and are described in the
“Scenarios” section below.

2.3. Agent satisfaction
Agent satisfaction describes how satisfied an agent is with a particular type of lamp.
The following five lamp properties were found to significantly affect consumer behaviour
by Kattenwinkel [6] and thus included in the satisfaction function: colour discrepancy,
energy efficiency, ramp–up time, and initial purchase price. These five properties were
weighted by global relative importance from the same study and agent–specific properties,
specifically tolerances for functional and colour requirements and focus on energy usage
for financial and environmental reasons. More details including specific weighting values
can be found in [17].
2.4. Scenarios
2.4.1. No regulation
In this scenario, there are no changes to incandescent lamp pricing or availability.

2.4.2. Soft ban: regulation with mild effect
Because the effects of the Ecodesign Directive regulation are not clear–cut, we considered
two possible scenarios. In this first one, the Ecodesign Directive results in an annual price
increase of 10% between 2013–2018 for incandescent bulbs, but does not affect availability.
Like with the LED pricing and innovation, the incandescent price progression is varied
by a third, independent random multiplication factor of 0.5–2. Because old batches and
new bulbs for “industrial use” may still be sold, incandescent lamps may not not become
wholly unavailable to the consumer.

2.4.3. Hard ban: regulation with strong effect
In this second regulation scenario, incandescent lights become wholly unavailable to the
consumer in 2015 after a 20% price increase (with random multiplication factor) in the
years 2012–2014.

2.4.4. Information campaign
This scenario is the same as the “no regulation” scenario with the change that in the
year 2012, two agent properties are altered. The focus on energy usage for both financial
and environmental reasons is increased by 50%, simulating the possible effects of an
information campaign that raises awareness of the benefits of lamp efficiency. As a result,
agents in the model become more dissatisfied by energy inefficiency.

2.4.5. Soft ban & information campaign
This scenario combines the “soft ban” scenario with the “information campaign“ scenario
to investigate to which extent their effects are additive.

Each scenario was run 50 times.

2.5. Validation
Due to the unavailability of numeric information about the progression of household
lighting in the EU region, exact calibration and verification periods have not been assigned
for our simulations. In the Conclusions we discuss plausibility of our results based on
existing literature and technical reports.
Figure 1: Main simulation results showing the percentage of non–incandescent lamps in households
over time. “Non–incandescent” is an umbrella term for halogen, CFL, LFL, HID, and LED lighting.
Every line represents the mean results from 50 runs of a scenario. Similar plots with the standard
deviation included are provided in the Supplement. The legend is sorted by efficacy, more effective
measures at the top.

3. Results
The main results are shown in Figure 1, which contains the mean results for each scenario.

3.1. No regulation
Firstly, we simulated the EU lighting market without any regulation. This resulted
in slow adoption of energy–friendly alternatives: the incandescent light bulb remained
dominant. Over 50% of all lamps in consumer households remained incandescent bulbs
by 2025. Supplemental Figure S6 shows the same plot as Figure 1 with the standard
deviation for this scenario included. When more energy–efficient LED lighting became
affordable, a segment of the population that was previously engaged in deliberation
and social comparison became satisfied and switched to repetition. This can be seen in
Supplemental Figures S7–S10, that contain the relative frequency of behaviours. However,
the majority of the population remained unaffected. The main drivers for this behaviour
appeared to be (i) the tendency of the Consumat to prioritise initial purchasing costs
over total cost of ownership and (ii) a low level of interest in household lighting, leading
to complacency with the functioning of incandescent lighting.
3.2. Soft ban
Secondly, we simulated two possible effects of the Ecodesign Directive policy. The “soft
ban” consisted of a gradual price increase. This significantly increased LED adoption over
the previous scenario: in 2025 around 75% of lamps were non–incandescent. Supplemental
Figure S4 shows the standard deviation for this scenario. Because incandescent bulbs
were still available, and generally still the least expensive option at time of purchase, a
segment of the Consumat population remained unaffected by market innovation. More
and more Consumats who previously were unsatisfied with available options and engaged
in high–effort strategies adopted the newer, more affordable LEDs and going forwards
only engaged in repetition behaviour (Supplemental Figures S7, S9, S10).

3.3. Hard ban
Thirdly, we simulated a second possible effect of the Ecodesign Directive policy. In
this scenario, incandescent lighting becomes unavailable to consumers. Trivially, the
percentage of non–incandescent lighting quickly climbed to 100% with low deviation
(Supplemental Figure S2). Initially there is a spike in deliberation behaviour forced
by the impossibility of repetition (Supplemental Figure S9). Next, many Consumats
switch to social strategies, because the unavailability of their top choice resulted in higher
uncertainty (Supplemental Figures S8 & S10). Unhappiness with the remaining options
means that repetition behaviour remains low (Supplemental Figure S7) even long after
the ban.

3.4. Information campaign
Fourthly, we simulated an information–only policy that was aimed at informing the
Consumat about energy–efficient lighting without restrictions on incandescent bulbs.
This raised the energy–efficient lighting adoption by over 10 percentage points. The
effects of the random variables governing LED pricing and innovation speed strongly
affect adoption speed. If LED lighting quickly becomes more affordable, adoption reaches
its peak around 2019. Slow price drops result in much lower peak adoption around 2022,
not too different from the “no regulations” scenario (Supplemental Figure S5). While
consumer opinion was changed in this scenario, the main driving force of behaviour still
appeared to be financial considerations.

3.5. Soft ban & information campaign
Lastly, we combined the “soft ban” and “information campaign” scenarios. While this
scenario resulted in an increased adoption of around five percentage point over the “soft
ban” only scenario, there was a large overlap between scenario outcomes as can be seen
in Supplemental Figures S3 & S4. Both the “soft ban” and “information campaign”
scenarios appeared to reach a similar consumer audience, so that combining them does
not result in a fully additive effect.
4. Conclusions
In this work, we implemented an agent–based simulation of the EU lighting market
under different policies. Our main goals were to explain consumer behaviour and explore
alternative policies. We found that our model was able to offer an explanation for the
reluctance of the consumer to switch to energy–friendly lighting and that it could be
used to investigate hypothetical policy scenarios. Our model suggested that the adoption
of energy–friendly lighting alternatives was hindered by a low level of consumer interest
combined with high–enough levels of satisfaction about incandescent bulbs.
The most invasive scenario of making energy–inefficient bulbs unavailable was highly
efficacious, because it forced the consumer to adapt its behaviour. A milder regulation
that inflated prices of energy–inefficient bulbs, increased lighting adoption from below
50% to around 75% in 2025. An even less invasive option of an information campaign
was less successful, increasing adoption to above 60%. Combining milder regulation with
an information campaign did not significantly increase adoption.
The authors are not aware of representative research measuring the effects of the
Ecodesign Directive in the EU, either on the lighting market or the household lamp
distribution. In [2], effects of the Ecodesign Directive are estimated based on different
hypothetical scenarios. Similarly, the Model for European Light Sources Analysis [18]
presents numbers up until 2013. Koretsky [19] notes that in 2020, incandescent bulbs
are still available to purchase online. The most complete source on the evolution of the
lighting market may be Zissis et al. [20] showing that in 2019, fluorescent and LED sales
each made up about half of >90% of global sales, suggesting a <10% market share for
incandescent lighting. This would mean that actual non–incandescent penetration lies
between our hard and soft ban scenarios. This makes sense insofar that the actual market
results of the Ecodesign Directive also appear to lie in between these two scenarios:
incandescent bulbs are still available for purchase online and in hardware stores, but not
readily available (e.g. in supermarkets). Our scenarios did not consider the effects of
limited availability.
From the four possible behaviours in our model, repetition was by far the most common
(Supplemental Figure S7). Making non–disruptive changes to the market or consumer
opinion did increase adoption rate, but it failed to reach consumers who were set in their
ways. Our findings agree with literature in that, while social interactions are known
to significantly affect customer repeat behaviour (e.g. [21]), consumer habits are hard
to break without direct disruption [22]. Financial savings, increased availability, more
natural colouring, and environmental concerns are mentioned as leading factors in LED
adoption [23, 24, 25, 26] and also implicated by our model.
Randomness occurs in our model through consumer preference instantiation, peer
selection in social behaviour, and three random factors of 0.5–2 governing LED pricing
decrease, incandescent pricing increase, and LED innovation progression (increased life
span and colour temperature of LEDs). Sensitivity analysis showed that the largest
variation as seen in Figures S2–S6 occurs through the interaction of the two factors
governing incandescent and LED pricing. Even though our agents consider other factors,
initial purchasing price remains a significant consideration for many. This means that a
tipping point is reached as soon as incandescent bulbs are no longer the least expensive
option in stores, the occurrence and timing of which in our model is determined by the
two random factors governing price progression.
With our model, we showed an application of agent–based models in explaining con-
sumer behaviour and testing policies to affect this behaviour. A body of related work
using agent–based methods specifically for climate–relevant behaviours exists [4, 7].
Closely related to our application, Hicks et al. [27] studied the innovation diffusion of
energy saving lighting focusing on information and perception, concluding that increased
usage may (partially) offset energy savings. Relatedly, Muelder and Filatova [28] investi-
gated different formalisations of social theories in energy consumption, specifically solar
investments. Buskens [29] summarised the sociological background of innovation diffusion
in social networks and concluded that close social circles are especially important to
establish trust in new products.
Our work is limited by a number of assumptions about the development of the lighting
market and the effects of information campaigns, as well as the abstraction of consumer
behaviour into four strategies. We measured the effects of our assumptions on agent
preferences, pricing progression, and technical innovation by including independent
random variables that halved to doubled our projections, exploring a wide range of
possible progressions. Our model environment and agent preferences are realistic by
virtue of being based on empirical market research. The Consumat model itself, while
necessarily being a simplified abstraction, is strongly grounded in psychological theory
and exhibits complex behaviour at macro-levels [5]. This combination allowed us to
generate specific testable hypotheses about consumer behaviour.
In conclusion, our agent–based model of the lighting market was able to explain
consumer behaviour and evaluate counterfactual policies. Our findings offer insight into
both individual–level driving forces of behaviour and society–level outcomes in a niche
market. With this, our work demonstrates the strengths of agent–based models for policy
generation and evaluation.

References
[1] European Commission, Directive 2009/125/EC of the European Parliament and of
the Council of 21 October 2009 establishing a framework for the setting of ecodesign
requirements for energy-related products, 2009. URL: http://data.europa.eu/eli/dir/
2009/125/oj.
[2] D. Quack, 10 years anniversary - Phase out of incandescent lightbulbs: How has the
Ecodesign lighting regulation helped consumers save money over time?, Technical
Report, Oeko-Institut e.V., 2019. URL: https://www.beuc.eu/publications/eu_
ecodesign_lighting_report_2019.pdf.
[3] B. F. Mills, J. Schleich, Why don’t households see the light?: Explaining the diffusion
of compact fluorescent lamps, Resource and Energy Economics 32 (2010) 363–378.
doi:10.1016/j.reseneeco.2009.10.002.
[4] J. Castro, S. Drews, F. Exadaktylos, J. Foramitti, F. Klein, T. Konc, I. Savin,
J. van den Bergh, A review of agent-based modeling of climate–energy policy, Wiley
Interdisciplinary Reviews: Climate Change 11 (2020) e647. doi:10.1002/wcc.647.
[5] W. Jager, M. Janssen, An updated conceptual framework for integrated modeling of
human decision making: The Consumat II, in: Proceedings of the Complexity in the
Real World Workshop, European Conference on Complex Systems, 2012, pp. 1–18.
[6] G. Kattenwinkel, And then there was light!, Master’s thesis, University of Groningen,
2012. URL: https://feb.studenttheses.ub.rug.nl/8979/.
[7] W. Jager, Using agent–based modelling to explore behavioural dynamics affecting
our climate, Current Opinion in Psychology 42 (2021) 133–139. doi:10.1016/j.
copsyc.2021.06.024, psychology of Climate Change (2021).
[8] S. Schaat, W. Jager, S. Dickert, Psychologically plausible models in agent-based
simulations of sustainable behavior, in: A. Alonso-Betanzos, N. Sánchez-Maroño,
O. Fontenla-Romero, J. G. Polhill, T. Craig, J. Bajo, J. M. Corchado (Eds.), Agent-
Based Modeling of Sustainable Behaviors, Springer International Publishing, Cham,
2017, pp. 1–25. doi:10.1007/978-3-319-46331-5_1.
[9] G. Bravo, E. Vallino, A. K. Cerutti, M. B. Pairotti, Alternative scenarios of green
consumption in Italy: An empirically grounded model, Environmental Modelling
and Software 47 (2013) 225–234. doi:10.1016/j.envsoft.2013.05.015.
[10] A. Kangur, W. Jager, R. Verbrugge, M. Bockarjova, An agent-based model for
diffusion of electric vehicles, Journal of Environmental Psychology 52 (2017) 166–182.
doi:10.1016/j.jenvp.2017.01.002.
[11] M. Moglia, A. Podkalicka, J. McGregor, An agent-based model of residential energy
efficiency adoption, Journal of Artificial Societies and Social Simulation 21 (2018) 3.
doi:10.18564/jasss.3729.
[12] P. Van Oel, D. Mulatu, V. Odongo, D. Willy, A. Van der Veen, Using data on
social influence and collective action for parameterizing a geographically-explicit
agent-based model for the diffusion of soil conservation efforts, Environmental
Modeling & Assessment 24 (2019) 1–19. doi:10.1007/s10666-018-9638-y.
[13] A. Bandura, Social learning through imitation, in: M. Jones (Ed.), Nebraska
Symposium on Motivation, University of Nebraska Press, Lincoln, NE, 1962, pp.
211–274.
[14] G. Ellison, D. Fudenberg, Word-of-mouth communication and social learning, The
Quarterly Journal of Economics 110 (1995) 93–125. doi:10.2307/2118512.
[15] P. Anand, Foundations of Rational Choice Under Risk, Oxford University Press,
1995.
[16] W. Wood, D. Rünger, Psychology of habit, Annual Review of Psychology 67 (2016)
289–314. doi:10.1146/annurev-psych-122414-033417, pMID: 26361052.
[17] G. H. Schoenmacker, Agent–based consumer modelling of the Dutch lighting market,
Master’s thesis, University of Groningen, 2014. URL: http://fse.studenttheses.ub.
rug.nl/id/eprint/11623.
[18] Van Holsteijn en Kemna BV, Lot 8/9/19 Ecodesign Preparatory Study on Light
Sources, 2015. URL: https://ecodesign-lightsources.eu/documents.html.
[19] Z. Koretsky, Phasing out an embedded technology: Insights from banning the
incandescent light bulb in europe, Energy Research and Social Science 82 (2021)
102310. doi:10.1016/j.erss.2021.102310.
[20] G. Zissis, P. Bertoldi, T. Serrenho, Update on the status of led-lighting
world market since 2018, Publications Office of the European Union: Luxem-
bourg (2021). URL: https://publications.jrc.ec.europa.eu/repository/bitstream/
JRC122760/status_of_led_lighting_world_market_2020_final_rev_2.pdf.
[21] K. Toker-Yildiz, M. Trivedi, J. Choi, S. R. Chang, Social interactions and monetary
incentives in driving consumer repeat behavior, Journal of Marketing Research 54
(2017) 364–380. doi:10.1509/jmr.13.0482.
[22] B. Verplanken, W. Wood, Interventions to break and create consumer habits, Journal
of Public Policy & Marketing 25 (2006) 90–103. doi:10.1509/jppm.25.1.90.
[23] O. Leelakulthanit, The factors affecting the adoption of led lamps, International
Business & Economics Research Journal 13 (2014) 757–768. doi:10.19030/iber.
v13i4.8684.
[24] K. R. Cowan, T. U. Daim, Understanding adoption of energy efficiency technologies:
applying behavioral theories of technology acceptance & use to understand the
case of led lighting for commercial, residential, and industrial end-users, in: 2011
Proceedings of PICMET’11: Technology Management in the Energy Smart World,
IEEE, 2011, pp. 1–9. URL: https://ieeexplore.ieee.org/abstract/document/6017904.
[25] R. Dangol, M. S. Islam, M. Hyvärinen, P. Bhushal, M. Puolakka, L. Halonen, User
acceptance studies for led office lighting: Preference, naturalness and colourfulness,
Lighting Research & Technology 47 (2015) 36–53. doi:10.1177/1477153513514424.
[26] K. Cowan, T. Daim, Adoption of Energy Efficiency Technologies: A Review of
Behavioral Theories for the Case of LED Lighting, Springer London, London, 2013,
pp. 229–248. doi:10.1007/978-1-4471-5097-8_10.
[27] A. L. Hicks, T. L. Theis, M. L. Zellner, Emergent effects of residential lighting
choices: Prospects for energy savings, Journal of Industrial Ecology 19 (2015)
285–295. doi:10.1111/jiec.12281.
[28] H. Muelder, T. Filatova, One theory - many formalizations: Testing different code
implementations of the theory of planned behaviour in energy agent-based models,
Journal of Artificial Societies and Social Simulation 21 (2018) 5. doi:10.18564/
jasss.3855.
[29] V. Buskens, Spreading information and developing trust in social networks to
accelerate diffusion of innovations, Trends in Food Science & Technology 106 (2020)
485–488. doi:10.1016/j.tifs.2020.10.040.
Supplement

Table S1
Initial settings for the lamps available to the model. The high initial prices for LED lighting are
based on 2006 prices, when LED was just becoming available to consumers. The “Available” column
represents availability at the beginning of the simulation and changes over time (see main text).

Type Price Efficiency Colour Ramp–up Lifetime Available
e % % Sec. Months Y/N
LED 30.00 63 10 1 125 N
LED 25.00 60 10 1 167 N
LED 20.00 60 10 1 208 N
LED 15.00 60 15 1 167 N
LED 12.50 60 10 2 208 N
CFL 9.30 80 30 80 83 Y
CFL 8.40 90 15 80 83 Y
CFL 7.80 90 15 40 100 Y
CFL 7.80 90 15 40 83 Y
CFL 7.00 90 15 40 83 Y
CFL 5.00 70 15 1 17 Y
CFL 3.20 70 15 1 8 Y
CFL 3.20 60 15 1 17 Y
CFL 2.50 60 15 1 17 Y
Incandescent 3.00 30 5 1 17 Y
Incandescent 2.70 50 5 1 8 Y
Incandescent 1.80 50 5 1 8 Y
Incandescent 1.80 40 5 1 8 Y
Incandescent 1.40 50 5 1 8 Y
Figure S2: Main simulation results showing the percentage of non–incandescent lamps in households
over time. “Non–incandescent” is an umbrella term for halogen, CFL, LFL, HID, and LED lighting.
Every line represents the mean results from 50 runs of a scenario. Standard deviation of the “hard
ban” scenario is included in red.
Figure S3: Simulation results showing the percentage of non–incandescent lamps in households over
time. “Non–incandescent” is an umbrella term for halogen, CFL, LFL, HID, and LED lighting. Every
line represents the mean results from 50 runs of a scenario. Standard deviation of the “soft ban &
information campaign” scenario is included in red.
Figure S4: Simulation results showing the percentage of non–incandescent lamps in households over
time. “Non–incandescent” is an umbrella term for halogen, CFL, LFL, HID, and LED lighting. Every
line represents the mean results from 50 runs of a scenario. Standard deviation of the “soft ban”
scenario is included in red.
Figure S5: Simulation results showing the percentage of non–incandescent lamps in households over
time. “Non–incandescent” is an umbrella term for halogen, CFL, LFL, HID, and LED lighting. Every
line represents the mean results from 50 runs of a scenario. Standard deviation of the “information
campaign” scenario is included in red.
Figure S6: Simulation results showing the percentage of non–incandescent lamps in households over
time. “Non–incandescent” is an umbrella term for halogen, CFL, LFL, HID, and LED lighting. Every
line represents the mean results from 50 runs of a scenario. Standard deviation of the “no regulation”
scenario is included in red.

Figure S7: Simulation results showing percentage of repetition behaviour over time.
Figure S8: Simulation results showing percentage of imitation behaviour over time.

Figure S9: Simulation results showing percentage of deliberation behaviour over time.
Figure S10: Simulation results showing percentage of social comparison behaviour over time.