Energy & Buildings 260 (2020) 109559

Contents lists available at ScienceDirect

Energy & Buildings

journal homepage: www.elsevier.com/locate/enbuild

Nudging the adaptive thermal comfort model

Thomas Parkinson a, Richard de Dear b,∗, Gail Brager a
a
University of California Berkeley, Center for the Built Environment (CBE), Berkeley, CA, USA
b
The University of Sydney, Indoor Environmental Quality Lab, School of Architecture, Design and Planning, Sydney, NSW, 2006, Australia

a r t i c l e i n f o a b s t r a c t

Article history: The recent release of the largest database of thermal comfort field studies (ASHRAE Global Thermal Com-
Received 1 June 2019 fort Database II) presents an opportunity to perform a quality assurance exercise on the first generation
Revised 25 September 2019
adaptive comfort standards (ASHRAE 55 and EN15251). The analytical procedure used to develop the
Accepted 27 October 2019
ASHRAE 55 adaptive standard was replicated on 60,321 comfort questionnaire records with accompa-
Available online 30 October 2019
nying measurement data. Results validated the standard’s current adaptive comfort model for naturally
Keywords: ventilated buildings, while suggesting several potential nudges relating to the adaptive comfort standards,
Adaptive thermal comfort adaptive comfort theory, and building operational strategies. Adaptive comfort effects were observed in
HVAC all regions represented in the new global database, but the neutral (comfort) temperatures in the Asian
Mixed-mode subset trended 1–2 °C higher than in Western countries. Moreover, sufficient data allowed the develop-
Natural ventilation ment of an adaptive model for mixed-mode buildings that closely aligned to the naturally ventilated
Energy
counterpart. We present evidence that adaptive comfort processes are relevant to the occupants of all
Standards
Climate
buildings, including those that are air conditioned, as the thermal environmental exposures driving adap-
tation occur indoors where we spend most of our time. This suggests significant opportunity to transition
air conditioning practice into the adaptive framework by programming synoptic- and seasonal-scale set-
point nudging into building automation systems.
© 2019 Elsevier B.V. All rights reserved.

1. Introduction Successful implementation of a mixed-mode strategy includes a re-

laxation of the conventionally tight deadband between heating and
The provision of thermal comfort for building occupants stands cooling setpoints. Fig. 1 shows reductions in annual HVAC-energy
out as one of the largest end-uses of energy in the built envi- consumption of roughly 7–15% for every degree Celsius expansion
ronment, bearing significant responsibility for greenhouse gas in either direction beyond a temperature control dead-band of
emissions and their destabilizing effects on our global climate about 2 K [29]. Utilizing natural ventilation is one mechanism for
system [8,36]. One of the more common architectural answers maintaining comfort within those wider temperature ranges. Sig-
to these challenges is climate-responsive or passive design of nificant energy savings can therefore be achieved through an op-
buildings, where natural ventilation is substituted for mechanical erational change as simple as nudging setpoint temperatures [24].
conditioning to deliver comfortable indoor environments while at Challenging conventional comfort theory and practice of
the same time zeroing energy demand for heating, ventilation, and the time, de Dear and Brager [14] and Nicol & Humphreys
air-conditioning (HVAC). Where external climatic conditions or the [48] proposed adaptive comfort models as the appropriate tool
building program are not amenable to exclusive reliance on natural for designing naturally ventilated spaces and quantifying their
ventilation, the hybrid approach known as mixed-mode (i.e., a operational comfort performance. In the two decades since then,
combination of operable windows and mechanical conditioning) practitioners have applied the adaptive comfort approach to the
represents an alternative low-energy design strategy. By fore- design and operation of many naturally ventilated buildings. And
stalling the onset of mechanical conditioning for as long as outdoor comfort researchers have tested the model with thousands of new
weather conditions permit, a mixed-mode design minimizes HVAC right-here-right-now comfort data points from buildings scattered
energy demand without compromising occupant thermal comfort. across diverse climate zones around the world [15]. But the needle
is not moving fast enough in the promotion of climate-responsive
designs with minimal reliance on air-conditioning to abate green-
∗
Corresponding author. house gas emissions from the built environment. In The Healthy
E-mail address: [email protected] (R. de Dear). Workplace Nudge, Miller et al. [42] borrow ideas from Thaler &

https://doi.org/10.1016/j.enbuild.2019.109559
0378-7788/© 2019 Elsevier B.V. All rights reserved.
2 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559

Fig. 1. The potential HVAC energy savings associated with widened heating and cooling setpoints for a standardized grade-A reference office building in three American
cities with diverse climates. Modified after Hoyt et al. [29].

Sunstein [61] and use behavioral economics to discuss how “nudge We conducted a bibliometric analysis using the Scopus database
thinking” allows small, positive unobtrusive changes to promote to understand the impact of the adaptive concept on the thermal
healthy decisions. With this in mind, this paper aims to nudge comfort research domain in recent decades. A query of journal pa-
the adaptive thermal comfort model to increase robustness and pers and conference proceedings with titles, abstracts, or keywords
incrementally expand its scope of applicability for use in building containing the words ‘adaptive’ AND ‘thermal’ AND ‘comfort’ re-
design and operation in the hope that this will lead to improved turned a total of 1200 documents in April 2019. Fig. 2 presents
energy and comfort performance. these research publication events as a timeseries demonstrating
the growth in outputs in the last 20 years. Whilst traditional
1.1. Changing landscape of adaptive thermal comfort centers of thermal comfort research – UK, USA, Italy, Germany
and Australia – appear on the list of productive countries, relative
Based on the pioneering framework of thermal comfort by Nicol newcomers, including China, India, and Hong Kong are becoming
& Humphreys [47], de Dear and Brager’s adaptive comfort model increasingly prominent. Our analysis showed China currently
[14] was first codified by ASHRAE in 2004 [4]. It has since been ranked the second most productive country behind the UK, and if
replicated in other jurisdictions, notably the European Union [18], the current trajectory is maintained, it is poised to become number
and more recently in China [34] and India [40]. The model’s name one in the near future. The important takeaway is that the center
references a view of building occupants as active agents in the of gravity of adaptive comfort thinking is shifting from places like
achievement of thermal comfort. This idea marked a sharp de- UK, USA, and Europe towards emergent research hubs in Asia.
parture from the orthodox thermal comfort view of occupants as Increased research activity in Asian countries has been ac-
passive recipients of indoor climate [20]. By debunking the con- companied by efforts to localize adaptive models in increasingly
ventional assumption that thermal comfort could only be achieved specific contexts. Whilst the general adaptive principle has been
within a narrow band of indoor temperatures, the adaptive comfort repeatedly demonstrated across diverse settings, region-specific
model and derivative standards conferred legitimacy on passive adaptive models are not universally applicable. ASHRAE’s Standard
and low energy design strategies focused on natural ventilation. 55 adaptive comfort model and the European Union counterpart
The 1998 and 2002 publications proposing adaptive comfort of EN15251 were transformative because of their generalizability,
standards sparked a flurry of new research activity on the topic. the empirical basis of which was vastly more comprehensive than

Fig. 2. The number of research outputs and citations by year since the first paper on adaptive comfort by Nicol & Humphreys in 1973. Citation count refers to the year in
which the cited paper was published.
 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559 3

anything preceding them. But in the two decades since their if not available the data from GHCN database was substituted
endorsement, there has been a large number of thermal comfort (19,995 records), and if neither of these options were available,
field studies in unique contexts. The recently-released ASHRAE we resorted to historical climatic averages (27,593 records). This
Global Thermal Comfort Database II [22], with over 10 0,0 0 0 rows included daily temperature measurements from ASHRAE Database
of “right-here-right-now” thermal comfort field data from around I, (the basis of the current ASHRAE adaptive comfort standard),
the world, is an order of magnitude larger than its predecessor which were also supplemented with monthly meteorological data
that was used to develop the ASHRAE Standard 55 adaptive com- for those records where available.
fort model [13]. It is beyond the scope of the present paper to Unlike its predecessor, Database II does not explicitly identify
summarize the database, but a detailed description can be found building level metadata. As a result, directly replicating the anal-
in Földváry Ličina et al. [23]. ysis in the original ASHRAE Standard 55 adaptive comfort model
was initially impossible with Database II because the estimation of
1.2. Research aims thermal neutralities using the linear regression method [13] was
based on the individual building as the unit of analysis. To address
In light of the changing landscape of adaptive thermal comfort this we used simple heuristics to infer building identification num-
research over the last two decades, and the release of the ASHRAE bers (referred to as building ID in this analysis) across Database II
Global Thermal Comfort Database II - referred to hereafter simply by determining unique cases based on publication, city, condition-
as Database II - into the public domain, a follow-up analysis of the ing strategy, and season (summer and winter were collapsed to
adaptive concept seems timely. The availability of a large volume include autumn and spring respectively, merely for this purpose).
of new data from diverse climatic and regional contexts provides Such backfilling of meteorological data and building ID codes were
an opportunity to revisit the original ASHRAE 55 adaptive comfort necessary prerequisites to replicating the analytical strategy used
standard. In the interests of nudging our current understanding of to define the ASHRAE adaptive comfort model.
adaptive theory, the existing adaptive comfort model, and the ap-
plication of adaptive principles to building operational strategies, 2.2. Data analysis
our principal aims for this paper are as follows:
The analysis by de Dear & Brager [14] underpinning the orig-
1. Replicate the analysis by de Dear & Brager [14] on a larger and inal adaptive comfort model was based on field measurements
more representative dataset to validate the original adaptive of indoor operative temperature. This was preferred at the time
comfort model, as it was deemed more representative of the actual conditions
2. Assess differences in adaptive comfort principles across broad experienced by building occupants through its consideration of
regions of the world, both radiative and convective heat transfers. However, an analysis
3. Propose revisions to extend the limits of applicability of the of Database II by Dawe et al. (forthcoming) determined the median
adaptive comfort model beyond naturally ventilated buildings absolute difference between indoor air and radiant temperature
as currently specified in ASHRAE Standard 55–2017, measurements as 0.4 °C, meaning an even smaller difference in
4. Discuss the potential for nudging HVAC practices to incorporate operative temperature. Our own exploratory analysis of adaptive
adaptive comfort theory as an energy-reduction strategy. comfort using Database II showed very similar results when
using either air or operative temperatures. We also observed that
2. Method Database II had 26,700 records missing an operative temperature
value. Therefore, we used air temperature as the independent
Our analysis of Database II was intentionally designed to variable in the following analyses to enable us to access the
replicate the development procedure of the previous ASHRAE statistical power of the complete database.
adaptive comfort model to ensure backwards compatibility with The analytical precedent of the ASHRAE Standard 55 adaptive
associated standards. We used “R” (R [50]) and the “RStudio IDE” comfort model was replicated here on a subset of the modified
[54] along with the following packages: tidyverse [66], data.table Database II containing all records from office buildings having
[17], bibliometrix [3], comf [57], ggpmisc [19], here [44], coun- concurrent observations of indoor air temperature, thermal sensa-
trycode [2], rworldmap [60], climateeng [51], and grateful [52]. tion vote, and outdoor mean monthly temperature. The resulting
Relevant data visualizations are grouped by conditioning strategy - subset contained 60,321 records from a total of 135 inferred
air conditioned (AC) in black, mixed-mode (MM) in mustard, and buildings, including 15,203 records from the original Database I.
naturally ventilated (NV) in blue. We calculated coefficients based on the sample size from each
building ID and used them to weight the regression analyses. Fifty
2.1. Modified ASHRAE database II six percent of the sample was from Summer (or Autumn) and the
remaining from Winter (or Spring). A map showing the countries
We made some modifications to the public domain version and sample size of the field studies comprising the subsetted
of Database II in order to perform the analyses required for this database is shown in Fig. 3.
paper. The timestamps of measurements were retrospectively Following the ASHRAE Standard 55 adaptive model’s precedent
added by referring back to the original publications stemming we performed a simple linear regression to predict thermal sen-
from contributed datasets. These included the month and year of sation vote (ASHRAE 7-point scale) based on binned indoor air
the study as a minimum, with 50,287 timestamps retrieved. This temperature measurements (0.5 °C intervals) with building ID as
allowed us to attach more temporally specific meteorological data the unit of analysis. Twenty eight regression models failed to reach
to those records than the climatological averages currently in the statistical significance (p ≥ 0.05), resulting in linear models for 107
online version of Database II. Specific monthly temperatures for of the 135 building IDs. The neutral temperature for those building
the closest meteorological station were extracted from the Global IDs could have been determined using the Griffiths method but
Historical Climatology Network-Monthly (GHCN) database, a public that method has recently been shown to vary significantly be-
resource compiled by the National Oceanic and Atmospheric tween different contexts [49]. The statistically insignificant models
Administration [63]. Our revisions to the meteorological data in only accounted for 4% of the dataset and were therefore dropped
Database II were based on the following priorities: original data from the analysis. We determined a neutral temperature for each
from database contributor were preferred (59,995 records), but building ID based on a backwards solution of its regression model
4 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559

Fig. 3. World map showing the sample size by country in the subsetted database in our analysis. The UK has the largest contribution, but there is broad representation from
countries throughout Asia in Database II.

for neutral thermal sensation votes (TSV = 0). Fifteen buildings 3.1.1. Naturally ventilated buildings
with mean outdoor monthly temperature below 10 °C or above Starting the analysis with naturally ventilated buildings (NV)
33.5 °C did not significantly change the regression models and is logical given they are the focus of the ASHRAE Standard 55
were ultimately dropped from the analysis as they fall beyond the adaptive comfort model. The slope of the regression for NV build-
limits of the original ASHRAE 55 adaptive model. ings in Database II is 0.28 °C−1 , comparable to that of the original
ASHRAE Standard 55 adaptive comfort model (0.31 °C−1 ). The
Y-intercept term of the Database II NV regression model at 19.7 °C
3. Results is 2.1 °C higher than its counterpart in the original adaptive model
for NV buildings (17.8 °C), and 1 °C warmer than the value found
The first part of this section reports on the analysis of the in the EN15251 adaptive model which was based on an exclusively
subsetted Database II following the methods of de Dear & Brager European database.
[14] to verify the ASHRAE Standard 55 adaptive comfort model, We questioned whether the higher offset of the Y-intercept
and explores several potential nudges. The second part is based in the Database II NV model might be the influence of broader
on neutral temperatures determined using the Standard Effective regional representation following the inclusion of measurements
Temperature (SET) index instead of air temperature, as SET ac- from countries new to the larger dataset. To investigate this, we
counts for the six basic parameters in the human heat balance (ta , repeated the same analysis but on separate Western (Europe,
tr , RH, v, clo, met). The SET analysis includes detailed descriptions North America, Australia) and Asian (Middle-East, Indian sub-
of the indoor physical environmental conditions prevailing at the continent, and South, Southeast, and East Asia) subsets. Building
time the comfort questionnaires were administered, and allows IDs from Africa (n = 7) and South America (n = 3) were excluded
us to explore the differences between adaptive comfort models from this specific analysis because of insufficient data to perform
obtained from buildings with different conditioning strategies, and regressions for those regions. Fig. 5 shows that both indoor and
further nudge our understanding of the underlying mechanisms of outdoor temperatures are generally higher in the Asian subset
adaptation. compared to Western, resulting in a higher concentration of data
points in the top-right quadrant of the graph. For comparable
outdoor climates in both regions, the neutral temperatures in both
3.1. Adaptive thermal comfort model
NV and AC buildings in the Asian subset trended slightly higher
by a degree or two compared to their counterparts in the Western
The results of the weighted least square regression in Fig. 4
subset. And since this is occurring in both NV and AC buildings,
shows the relationship between the neutral temperatures and
it cannot entirely be explained by adaptation that is isolated to
mean monthly outdoor temperature for each building classified
free-running NV buildings (this will be discussed later). The spread
according to its conditioning strategy. It is the same data vi-
of neutral temperatures in Fig. 5 suggest that the warmer indoor
sualization used in the ASHRAE Standard 55 adaptive comfort
and outdoor temperatures from field studies in Asian cities in
model analysis. The differences between conditioning strategies
Database II may have driven much of the warmer displacement
are comparable to those found by de Dear & Brager [14] on the
of the Y-intercept term for the model reported in Fig. 4 from the
smaller Database I (∼21,0 0 0 records), while also revealing new
original coefficients reported by de Dear and Brager [14].
patterns that are described in subsequent sections.
 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559 5

Fig. 4. Neutral temperatures of buildings (determined using the same neutrality regression method as [14]) and the mean monthly temperature prevailing during each
building’s comfort survey. Each point shows an individual inferred building ID, and the point size is proportional to the weighting coefficient (sample size) attached to
that building ID when fitting the regression model. The colors of the regression lines and model coefficients indicate the conditioning strategy of the building. The grey
shading marks the 95% confidence interval around the fitted models. Light grey points are those buildings falling outside the original ASHRAE model’s outdoor temperature
domain (10, 33.5) that have been excluded from the regression analysis after confirming they didn’t make a difference. The original ASHRAE Standard 55 adaptive model
plus its associated 80% and 90% acceptability limits for NV buildings are superimposed for reference. Models for AC (R2 = 0.31, F(1,30) = 12.61, p < 0.01), MM (R2 = 0.53,
F(1,21) = 23.94, p < 0.0 0 01), and NV (R2 = 0.44, F(1,35) = 27.47, p < 0.0 0 0 01) were all highly significant.

Fig. 5. Neutral temperatures of buildings and the prevailing mean monthly temperature for buildings in Western (top) and Asian (bottom) countries. The colors of the un-
weighted regression lines and model coefficients indicate the conditioning strategy of the building. The symbol shape indicates the subset (circle = Western, Triangle = Asian).
Grey points represent buildings from the other subsets to aid comparison. The original ASHRAE Standard 55 adaptive model for NV buildings is superimposed for reference.
Models statistics in the Western subset for AC (R2 = 0.52, F(1,14) = 19.54, p < 0.001), MM (R2 = 0.30, F(1,9) = 1.123, p = 0.32), and NV (R2 = 0.50, F(1,14) = 23.08, p < 0.001) and
the Asian subset for AC (R2 = 0.00, F(1,13) = 0.06, p = 0.81), MM (R2 = 0.60, F(1,8) = 3.25, p = 0.11) and NV (R2 = 0.33, F(1,12) = 1.664, p = 0.22) show only two of the regressions
are significant.

3.1.2. Mixed-mode buildings model shown in Fig. 4. As anticipated, the MM regression line
Neither ASHRAE’s adaptive comfort model nor the European falls between the NV and AC adaptive comfort models reported
EN15251 version had sufficient field study data from mixed mode in that figure, but is more closely aligned to NV than AC. This
buildings (MM) to sustain any meaningful adaptive model regres- finding supports the notion that well-designed mixed-mode build-
sion analyses. But the current subset from Database II contains ings should operate first as naturally ventilated buildings when
25 separate MM buildings scattered across sufficiently diverse cli- and where possible, and use air conditioning to temper weather
matic zones to produce a statistically significant adaptive comfort extremes only when necessary.
6 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559

tries. The slope of the regression is the same for AC, MM, and NV
types, the only difference being that it extends to both cooler and
warmer temperatures in NV buildings. It should be noted that the
same trends and relationships were observed on the smaller subset
of indoor operative temperatures in our exploratory analysis.

3.2. Standard effective temperature analysis

The preceding analysis has been conducted on thermal neu-

tralities (comfort temperatures) that we derived by regressing
thermal sensation votes on concurrent indoor air temperatures. It
remains unclear if the differences between regions and building
conditioning strategies we reported in Section 3.1 result from a
systematic shift in the underlying adaptive perceptual processes
between these categories, or the effects of other human body
heat-balance parameters left unaccounted in our regression mod-
els. For example, do occupants of naturally ventilated buildings in
Asia deem warmer indoor temperatures to feel neutral because of
the higher air speeds typically found in their indoor climates (a
physical heat-balance effect), or are their thermal perceptions and
preferences being nudged by sustained exposure to warmer indoor
environments? Why do adaptive comfort principles manifest so
clearly in naturally ventilated and mixed-mode buildings, but are
dormant or heavily attenuated in air conditioned buildings? Is it
Fig. 6. The neutral temperature for each building ID plotted against the mean in-
because of some adaptive displacement in comfort expectations
door air temperature for that building. The colors of the regression lines and model driven by greater adaptive opportunity and a history of exposure
coefficients indicate the conditioning strategy of the building, and show no clear to warmer indoor temperatures, or is it simply an artefact of
difference between AC, MM, and NV types. Symbol shape indicates the regional different clothing patterns in buildings with different conditioning
classification of the building (circle = Western, Triangle = Asian). Models for AC
strategies in place?
(R2 = 0.96, F(1,30) = 1696 p < 0.0 0 0 01), MM (R2 = 0.97, F(1,21) = 713.5, p < 0.0 0 0 01),
and NV (R2 = 0.98, F(1,35) = 1202, p < 0.0 0 0 01) were all highly significant. To explore these questions about the underlying causal mech-
anisms of adaptive comfort, we used the same analytical strategy
3.1.3. Air-conditioned buildings reported in Section 3.1 but substituted Standard Effective Temper-
The results reported in Fig. 4 show a muted relationship be- ature (SET) in place of air temperature. SET is a comprehensive
tween the neutral temperatures in buildings operating under AC comfort index based on the concept of an equivalent temperature
and concurrent monthly outdoor temperatures. This is in line with that incorporates the six physical parameters known to affect com-
the original adaptive comfort model, and our analysis by region fort, namely air temperature, humidity, mean radiant temperature,
reported in Fig. 5 indicates the same pattern across climates air velocity, clothing and metabolic activity. We subset the records
and cultures. It is for this reason that adaptive principles have in Database II that had the full complement of input parameters
historically been discussed only in relation to highly permeable, required to calculate SET. The resulting dataset contained 46,280
naturally ventilated or free-running buildings, where indoor tem- observations, with 97 different building IDs yielding significant
peratures drift in the direction of prevailing weather and seasons. weighted least square regression models of thermal sensation
Conversely, indoor temperatures in air conditioned buildings were votes on SET.
assumed to be relatively independent of outdoor climatic condi- The following subsections step through the physical character-
tions because conventional practice is for setpoint temperatures istics of the indoor environments of these buildings as a function
to remain static throughout the year irrespective of trends and of mean indoor air temperature, before synthesizing them into the
fluctuations outdoors. Yet many occupants typically spend much final SET adaptive comfort analysis. Investigation of differences in
of their daily lives inside office buildings, so it is conceivable that metabolic rate has been omitted here because the near-universal
the environments inside our buildings exert some influence over use of the standard lookup table leads to reduced reliability and
adaptive thermal comfort, as well as the outdoor conditions. variability of met estimates within a population.
This line of reasoning prompted us to question whether adap-
tation to the thermal environment occurred for occupants of AC 3.2.1. Humidity
buildings. To test this hypothesis, we calculated the mean indoor The most common critique of the ASHRAE adaptive comfort
air temperature using all available records for each building within model that we have heard since it was published two decades
the database. In many cases this comprised measurements made ago is that it fails to consider the effects of humidity on occupant
over a few days, typical of field study research designs. The mean comfort. This looms large in the general public’s understanding of
indoor air temperature substituted mean monthly outdoor temper- thermal comfort, particularly for those in hot and humid climate
ature as the independent variable (x-axis) in the weighted least zones. Fig. 7 displays the psychrometric combinations of dry-bulb
squares regression. The resulting models shown in Fig. 6 indicate a temperature and humidity ratio means for each building ID in
much stronger statistical relationship (R2 0.96 ∼ 0.98) between the the SET subset. The majority of mean dry-bulb temperatures and
neutral temperature of a building and its mean indoor air temper- humidity ratios fall within the range of 20–27 °C and 0.005–
ature than was found when using the outdoor temperature as the 0.015 kg/kg respectively. The 12 building IDs with an average
independent variable (R2 0.31 ∼ 0.53). The neutral temperature for temperature and humidity above these thresholds were located in
a group of building occupants is generally close to the mean indoor Thailand, India, or Singapore, and only two were categorized as
temperature measured inside their building. Furthermore, there air conditioned. Notably, there were only three buildings in the
were negligible differences in the relationship between condition- original ASHRAE database that were above these thresholds. This
ing strategies of buildings and between Western and Asian coun- could lead one to question whether the effects of high humidity
 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559 7

Fig. 7. Psychrometric chart showing the distribution of the mean indoor temperature and humidity for each building ID. The point color indicates the conditioning strategy
and the size is relative to the number of data points. The shape indicates the region; circle = Western, triangle = Asian, square = other (Tunisia).

Fig. 8. The mean indoor air speed for each building ID by conditioning strategy. A quadratic regression forced through the origin improved the goodness of fit for both the
mixed-mode and naturally ventilated buildings over a linear model. The shape indicates the region; circle = Angle, triangle = Asia, square = other (Tunisia). Light grey dots
show all data points for comparison purposes.

in hot and humid climate zones may have been underrepre- indoor air speeds in AC buildings generally fell below 0.2 m/s, a
sented by the original adaptive comfort model. However, all 12 of level widely regarded as the just perceptible draught threshold
those buildings were found to have neutral temperatures above within the comfort envelope ([5] appendix I3). The four cases
27 °C SET, which corroborates the earlier findings from Fig. 6 in exceeding the 0.2 m/s average threshold were from field studies in
Section 3.1 showing the strong relationship between mean indoor buildings in China (Beijing) and India (Chennai) during summer.
air and neutral temperatures of buildings, irrespective of high There is generally higher mean air speeds in both MM and NV
humidity. The effects of relative humidity on thermal neutrality building subsamples, particularly when mean air temperatures
will be explicated further in the SET analysis of Section 3.2.4. exceed 27 °C. The association between indoor air temperature and
air speed is stronger in the NV subsample (R2 = 0.91) compared to
the MM buildings (R2 = 0.79).
3.2.2. Air speed
It is assumed that one of the key indoor environmental dif-
ferentiators between naturally ventilated and air conditioned 3.2.3. Clothing
buildings is that the former have higher indoor air speeds on Clothing insulation level represents the other key heat-balance
average. This is borne out in the data shown in Fig. 8. Mean parameter readily manipulated by occupants when adapting to
8 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559

Fig. 9. Mean clothing insulation level for each building ID by conditioning strategy. The shape indicates the region; circle = Angle, triangle = Asia, square = other (Tunisia).
Light grey dots show all data points for comparison purposes.

Fig. 10. The same adaptive comfort model analysis as shown in Fig. 4 earlier but using a subset of data with SET to establish the weighted least squares regression equation
to solve for the neutral SET temperature instead of air temperature. Each point shows an individual building ID, and the point size is proportional to the weighting coefficient
attached to that building ID when fitting the regression model. The colors of the regression lines and model coefficients indicate the conditioning strategy of the building.
The grey shading marks the confidence interval around the fitted models. Light grey points are those buildings falling outside the original model limits (10, 33.5) that were
excluded from the regression analysis after confirming they didn’t make a difference. Note: the original ASHRAE Standard 55 adaptive comfort model is superimposed for
reference but is not entirely accurate due to the different y-axis. The model for AC (R2 = 0.0016, F(1,36) = 0.057, p = 0.81) was not statistically significant, but those for MM
(R2 = 0.35, F(1,16) = 8.44, p < 0.01), and NV (R2 = 0.39, F(1,31) = 20.08, p < 0.0 0 01) were.

their indoor thermal environment. The mean clothing insulation building. Fig. 10 shows the relationship between the neutral SET
level was similar across the three building conditioning strategies temperatures and mean monthly outdoor temperature for each
(0.67 clo for AC, 0.74 clo for NV, and 0.70 clo for MM), but it’s building, classified according to conditioning strategy. The focus
important to look beyond the mean to discover that the role was specifically on the three physical factors beyond temperature
played by clothing as an adaptive thermal comfort mechanism that are most often discussed in adaptive comfort research, namely
appears to be quite different. Fig. 9 shows that the rate of clothing humidity, air speed, and clothing insulation.
insulation decrease per degree of indoor air temperature increase The results of the SET analysis in Fig. 10 narrows the difference
was greatest for NV buildings, moderate for MM buildings, and in the gradient or slope of the regression models between air
slight for AC buildings. However, this observed statistical relation- conditioned and naturally ventilated buildings compared to the
ship may simply reflect the wider range of indoor temperatures same analysis based on indoor air temperature (Fig. 4). The slope
(independent variable) found in MM and NV buildings. coefficient of 0.18 in the SET regression equation for NV buildings
is one third less than that obtained from indoor air temperature
(0.28). This suggests, in effect, that approximately one third of
3.2.4. Standard effective temperature
the adaptive comfort effect can be accounted for by heat balance
Presenting the basic heat-balance comfort parameters is help-
parameters included as inputs to the SET comfort index, notably
ful to characterize and differentiate the indoor environments of
air speed, clothing insulation, and humidity. The other portion
buildings within the database. To extend that insight and examine
might be attributable to factors not considered in the heat balance
the combined comfort effects of those parameters, we repeated
equations, such as adaptive opportunities, physiological differ-
the adaptive comfort analysis using the SET index. That is, we
ences, and historical patterns of variability that affect expectations
regressed thermal sensation scale responses on SET rather than
and perceptions of our thermal environments. This reiterates a
indoor air temperature to determine the neutral SET for each
 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559 9

finding made in the original ASHRAE adaptive comfort model by in this paper (especially Fig. 6 in Section 3.1.3) is that thermal
de Dear and Brager [14]. A reduced regression gradient was seen adaptation occurs in all indoor climatic environments, regardless
for MM buildings from 0.24 to 0.16. A much higher y-intercept of conditioning strategy. The research challenge is to define the
term was found for AC buildings when using SET. nudge of the adaptive model to account for this. The SET analysis
in Section 3.2 showed that using a comprehensive heat balance
4. Discussion comfort index to normalize the indoor environments found across
conditioning strategies accounted for a significant share of the dif-
In the following section we discuss the findings of our analysis, ference observed in the neutral temperature regression (comparing
specifically addressing the limits of applicability of the adaptive Figs. 4 and 10) between AC and NV buildings. That is to say that
comfort model in [5]. With the expanded database, we are now environmental or personal parameters in the human heat-balance
able to argue that the adaptive comfort principles are equally equation explain about one third of the comfort temperature vari-
relevant to all buildings irrespective of conditioning strategy, on ability between different buildings in our analysis of Database II.
the basis that occupants’ adaptive processes and expectations The closer alignment of the adaptive model across all condi-
are shaped by both indoor and outdoor temperature exposures. tioning strategies when based on SET is interesting, but begs the
This has important implications for adaptive comfort theory, the question - what about the unexplained differences? Work by Yao
ASHRAE 55 adaptive comfort model, and the design and operation et al. [68] and Schweiker & Wagner [58] look beyond statistics
of buildings for thermal comfort. We will walk through these in to develop approaches that consider adaptive process along with
the following paragraphs and propose a series of nudges in light cultural and climatic considerations to better understand different
of the present findings. comfort expectations of building occupants. Based on this dataset
and our analytical procedure, two additional potential hypotheses
4.1. Nudging adaptive comfort theory present: first is the personal control construct. This notion of
personal agency in creation of one’s comfort inside a building
In naturally ventilated buildings, the relatively high perme- was a core principle of the original adaptive model, and has
ability of the façade from operable windows means that indoor been empirically validated across diverse research designs [7,10].
conditions are usually closely coupled to the broader outdoor cli- Unfortunately, Database II does not have the necessary metrics
matic milieu. The use of outdoor temperature in adaptive comfort of occupant behavior and building descriptions such as adaptive
models reinforces the idea that the outdoor climate is the causal opportunities, and so a direct empirical test of this hypothesis is
driver of human thermal adaptive responses. However, the weak not possible here. However, on the basis of precedent studies by
statistical association between the comfort temperatures in air others, the causal nexus between personal control and thermal
conditioned buildings and their outdoor temperatures reported in adaptability seems plausible in these data from Database II. The
Section 3.1 begs the question of why outdoor climate drives com- second possible explanation is the representativeness of data
fort temperatures most strongly in naturally ventilated buildings? used in the analysis. The psychrometric chart (Fig. 7) shows the
This question was the focus of work by Fanger & Toftum [21] to indoor climates of AC buildings in the database were controlled
extend the PMV model to natural ventilated buildings in warm to relatively tight conditions. However, if the regression line for
climates, suggesting that those occupants had lower expectations AC buildings in Fig. 6 were extrapolated, we expect their neutral
of their indoor environment. It seems more likely that the typically temperatures would follow the same relationship observed in
strong correlation of indoor and outdoor temperatures in highly MM and NV buildings. Clearly the occupants of AC buildings are
permeable building designs (NV and MM) means that outdoor capable of adapting to temperatures beyond the narrow range
temperature is a reasonable proxy for the fluctuations of indoor they experience, but those buildings didn’t require much envi-
temperature. As a result, conventional adaptive models based ronmental or behavior adjustments. It is likely, therefore, that
on outdoor temperature can have high levels of predictive skill conventional adaptive comfort models fail to establish a strong
for indoor comfort temperatures, but only in climate-responsive statistical association with indoor neutralities in AC buildings
buildings that track the natural outdoor cycles. simply because there were too few data points in the warmer or
The overwhelming majority of our time is spent indoors, so it cooler ranges of indoor temperature due to the effectiveness of
is conceivable that the temperatures we are exposed to inside the HVAC systems in tightly controlling the indoor environment across
built environment exert the more powerful effect on our comfort a diverse range of external climates. This begs the question - if
expectations. Luo et al. [39] showed that indoor exposure can indoor temperatures in AC buildings were to drift up and down in
largely shape occupants’ expectations of their thermal environ- sync with seasonal cycles prevailing outdoors, with the potential
ment. In their analysis of Database II, Cheung et al. [12] reported for significant energy savings, would the same adaptive comfort
better predictive capacity of thermal sensation using only indoor responses be observed for the occupants of these buildings too?
air temperature compared to the fully-elaborated heat-balance This will be explored later in the Discussion section of this paper.
model (PMV-PPD) requiring six input parameters. This is far from The strong dependence of thermal neutrality on mean indoor
new thinking; a very familiar graph in one of Humphreys’ earliest air temperature depicted in Fig. 6 suggests the driver of adaptive
contributions on adaptive comfort shows a compelling regression thermal comfort is simply recent thermal exposure, regardless of
and correlation analysis (R2 = 0.96) between mean indoor air the engineering or architectural strategies in place. While the
or globe temperature and thermal neutrality observed for the simplicity of a universal adaptive theory applicable across all
building’s occupants, prompting the following comment: conditioning strategies holds intellectual appeal, on the surface
it might appear to counter the original empirical evidence [9,14].
“It is interesting that such an accurate prediction can be made
Yet this is easy to reconcile when all buildings, NV, MM and even
simply from a knowledge of the mean temperature experienced by
AC, are viewed as potential arenas of thermal adaptation provided
the respondents during the observation period…the range of re-
their occupants can avail themselves of adaptive opportunities
cent experience is better regarded as one of the factors which will
[7,43]. The greater the adaptive opportunity on offer, the larger the
contribute to the acceptability of the environment to which the re-
adaptive comfort effect. This should of course be bounded within
spondent is exposed” [31].
some reasonable range of conditions (see [35]), but the evidence
We know that indoor conditions are relatively static in air con- presented in this paper suggests that adaptation isn’t confined to
ditioned buildings, but the logical inference from our arguments just occupants of naturally ventilated buildings. Simplistic binary
10 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559

classification of buildings as AC or NV obscures the reality that for the model to reflect regional differences beyond that described
they are just polar opposites on a continuum of opportunity for by adaptation to climate alone. We believe that developing frag-
adaptive thermal behavior by their occupants [64]. mented country-specific models (based on smaller datasets) would
dilute the usefulness of the adaptive comfort model. Instead, an
4.2. Nudging adaptive comfort standards optional tweaking of the intercept term is proposed to reflect the
reported regional differences. To address the higher neutral tem-
When the first generation adaptive comfort standard was perature, we propose a maximum +1 K offset of the y-intercept
published [4], there was insufficient empirical evidence available term for buildings in Western countries, and a maximum +2 K off-
to sustain rational and defensible comfort guidelines for several set for those across Asia. This simple nudge maintains backwards
different contexts. The applicability of the adaptive model in compatibility with the original model, and offers an extensible
ASHRAE 55 was therefore restricted to naturally ventilated build- and comprehensive solution to increased calls for specific adaptive
ings. In this section we elaborate some of the implications of the comfort models emanating from Asia [32,40,45,46,59,62].
present analysis and propose three nudges to the model’s limits
of applicability specified in Section 5.4 of [5]. It is our belief that 4.2.2. Mixed-mode buildings
these nudges, one for each of the conditioning strategies analyzed, The ability of mixed-mode buildings to switch between nat-
will extend the relevance of adaptive principles across the built ural ventilation and mechanical conditioning systems has led to
environment and serve the interests of decreasing HVAC energy conflicting advice on whether heat-balance models or adaptive
use and enhancing occupant comfort. comfort principles should inform the thermal comfort require-
ments of those occupants. There was insufficient evidence in the
4.2.1. Naturally ventilated buildings first ASHRAE database to decide how to address mixed-mode con-
The emphasis of adaptive comfort theory on naturally ven- ditioning strategies in an adaptive comfort standard. As a result,
tilated buildings is largely because it is the context in which Standard 55 precludes the use of the adaptive comfort model for
occupants are most connected to outdoor temperature variations, any building with a mechanical cooling system. But given the lim-
and usually where adaptive control opportunities are most readily ited climates and programs of contemporary buildings that could
available. The larger Database II allowed us to explore regional truly be conditioned solely by natural ventilation, this restriction
differences in adaptive comfort and recommend a nudge in the represents a significant challenge to practitioners looking to design
standards governing its use in naturally ventilated buildings. Repli- innovative, low-energy buildings (e.g. [67]). ASHRAE Database II
cating the analysis of de Dear and Brager [14] showed that the contains 14,811 records of data from 22 mixed-mode buildings,
relationship between comfort temperatures and outdoor conditions offering a reasonable empirical evidence-base to reconsider this
for NV buildings (Fig. 4) is very similar to what was reported over limitation and propose a nudge to the standards to permit MM
twenty years ago on a smaller database. [5] Section 5.4 presents buildings to use adaptive comfort models.
the adaptive comfort model specifically for the purpose of defining The slope of the regression for mixed-mode buildings in
the range of “acceptable thermal conditions in occupant-controlled Fig. 4 generally aligns more closely with the naturally ventilated
and naturally conditioned spaces”, which is expressed as: counterparts and indicates a climate-responsiveness of building
occupants in mixed-mode spaces, different to those in centrally
tcomf = 0.31tpma(out) + 17.8(◦ C ) (1)
conditioned buildings. Overlapping confidence intervals suggest
where tcomf is the neutral operative temperature for indoor that some of this difference may simply be from statistical uncer-
comfort (°C) and tpma(out) is the prevailing mean outdoor air tainty, and the SET analysis in Fig. 10 shows greater convergence
temperature (°C). after normalizing indoor thermal environments. This observa-
The gradient term conceptually represents an index of thermal tion reinforces the interpretation of mixed-mode buildings as
adaptability, and both ASHRAE’s adaptive comfort model (0.31) naturally ventilated designs that have air conditioning capabil-
as well as CEN EN15251 (0.33) suggest that indoor comfort tem- ities on stand-by to mitigate overheating during more extreme
peratures drift about a third of a degree for each full degree weather and seasonal conditions. Defining the conditions which
shift in prevailing outdoor temperature [11]. The corresponding justify switching mixed-mode buildings between cooling strategies
gradient of 0.28 reported in Fig. 4 is slightly reduced but broadly remains a fruitful area of research, and will depend on many con-
comparable with existing adaptive comfort standards. The most textual factors for each building considering this. But beyond the
notable difference lies in the Y-intercept term of 19.8 °C, which is particulars of an individual building’s characteristics, this finding
two degrees warmer than the ASHRAE model (Eq. (1)) and one suggests that the occupants of mixed-mode buildings broadly have
degree warmer than CEN EN15251. similar expectations of their indoor thermal environment to their
The evidence presented in Fig. 4 indicates that, for all intents naturally ventilated counterparts. They are also able to utilize
and purposes, the original ASHRAE adaptive comfort standard many of the adaptive opportunities that may not be available to
for naturally ventilated buildings closely approximates occupants’ occupants of air-conditioned buildings.
adaptability to outdoor conditions, as evidenced by the similar gra- The proximity of the regression lines for mixed-mode and nat-
dient in the regression slope between the original model (Eq (1)) urally ventilated strategies in Fig. 4 provides compelling evidence
and the reanalysis (Fig. 4). However, exploring regional differences to nudge the limits of applicability of the ASHRAE Standard 55
in adaptive comfort revealed a potential nudge of the intercept adaptive comfort model to permit mixed-mode buildings. Our
term of the model for naturally ventilated buildings. It stands to findings are comparable to other results from field studies of occu-
reason that, along with climatic drivers and physiological adapta- pant comfort in mixed-mode buildings in Australia [16], India [28],
tion, there are cultural influences shaping thermal perception of and China [38] that found that the adaptive model more accurately
building occupants around the world [6,39,58,68]. The steeper gra- describes thermal comfort of occupants in those buildings than
dient for the Asian subset in Fig. 5 suggests occupants of buildings heat-balance methods. Recent studies in Brazilian mixed-mode of-
located in Asia are more thermally adaptive than what we saw fice buildings [55] found occupants adapted to indoor temperature
in the original ASHRAE adaptive comfort standard. Mean building fluctuations during the use of natural ventilation, as predicted by
temperatures reported in the psychrometric chart (Fig. 7) indicates ASHRAE’s adaptive thermal comfort model. When operating under
that most of the buildings in the upper temperature ranges were air conditioning mode, the Rupp et al. [55] study found the same
from Asia. Warmer neutral temperatures suggest there is a need muted dependence of indoor neutral temperatures on prevailing
 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559 11

outdoor temperatures. When considered alongside our Database II 4.3.1.1. Adaptive/ramped approach. During periods (or zones) when
analyses, these findings support nudging the limits of applicability (where) the spaces are conditioned purely by natural ventilation,
of ASHRAE Standard 55 to include the use of the adaptive model the adaptive comfort model should apply. If windows are then
for mixed-mode buildings. sealed and mechanical cooling is operating (changeover mixed-
mode approach), a transition period would maintain conditions
in the upper reaches of the adaptive comfort temperature range.
4.2.3. Air conditioned buildings If mechanical cooling has been operating for an extended period,
Although there has been some discussion around the appli- conditions would then ramp down to the top of the conventional
cation of adaptive comfort theory to mixed-mode buildings, it is PMV-based comfort zone recommended for air conditioned spaces.
uncommon to hear of adaptive principles guiding the design or
operation of air-conditioned buildings. Yet our present analysis 4.3.1.2. Adaptive/conserving approach. The adaptive comfort criteria
showing evidence of adaptation to prevailing indoor tempera- are consistently maintained through all operational scenarios.
tures by occupants of all conditioning strategies suggests there Natural ventilation is used exclusively as long as conditions are
may be scope to do so. Perhaps the most significant nudge is maintained within the adaptive comfort limits, and mechanical
to encourage practitioners to explore the potential for adaptive cooling is used only as needed to ensure the building temperature
comfort principles to inform design criteria, set-point tempera- does not exceed the adaptive comfort maximum temperature.
tures, and operation schedules of HVAC systems. If considering This is most appropriate for spaces that operate primarily in
the fundamental principles of the adaptive theory – we adapt to naturally ventilated mode during significant periods of the year,
the conditions that we are exposed to – there is no reason why and where the occupants are well-educated about building per-
the control logic of HVAC systems can’t be programmed to reflect formance and will play an active role in managing their own
seasonal drifts in temperature within an acceptable temperature thermal environment (i.e., there is sufficient adaptive opportunity
range [35]. This is expanded on in the next section, but it is also so that expectations are relaxed as well). This is the most energy-
relevant here for suggesting a potential nudge to ASHRAE Standard conserving approach, and we believe that the present analysis
55. To encourage such a shift in practice, standards bodies such supports its implementation.
as ASHRAE should relax the tight guidelines on air conditioned
buildings, most of which are premised on the artificial precision 4.3.2. PMV and the adaptive model in air conditioned buildings
of the PMV/PPD model’s predictions [12,30]. Air-conditioned buildings are often operated at much cooler
temperatures than ASHRAE’s comfort prescriptions. In the US
and comparable countries such as Australia, the typical design
4.3. Nudging practice
temperatures are set at around 22 °C and remain fixed through
all seasons (e.g. [1,41,53]). Indeed, summertime AC setpoints are
Nudging the adaptive comfort model in light of this new
actually closer to ASHRAE’s winter prescription for heavier clothing
analysis amounts to naught unless accompanied by some nudging
and 50% relative humidity. The US General Services Administration
of building design and operational practices. The pathway between
(GSA) Public Building Service [25] investigated the potential for
research and practice in the built environment domain, especially
energy savings and performance gains and identified the optimal
in relation to indoor environmental quality, has traditionally been
temperature for office spaces in summer up to 25.6 °C. A signif-
via building codes and regulatory documents such as standards.
icant proportion of its buildings were operated at temperatures
But ultimately it’s the creative application of these standards in
well below that, leading to 61% of building users feeling too cold.
building design and operation that will ensure such impact. Small
Despite the adaptive comfort messaging from the American Society
nudges in practice can potentially have profound effects on both
of Heating, Refrigerating and Air-conditioning Engineers, comfort
energy and comfort performance of buildings.
practice in centrally conditioned buildings remains steadfastly
maladaptive all year-round.
4.3.1. Applying the adaptive model to mixed-mode buildings What is often overlooked in over-simplistic interpretations
At first glance, the assessment of thermal comfort in a mixed- of the PMV/PPD model is that the upper temperature limit in
mode building requires the evaluation of three different operating ASHRAE Standard 55’s 0.5 clo summer comfort zone stretches
regimes. all the way to 27 °C at 50% relative humidity with air speeds
below 0.2 m/s. Even the neutral mid-point of the summer comfort
a. Occupied hours when spaces are conditioned solely by natural zone is at about 25.5 °C. This disconnect between a standard
ventilation. written for the air conditioning industry [5] and common air
b. Occupied hours when spaces are conditioned by mechanical conditioning practice remains, to date, largely unexplained. One
conditioning only. potential hypothesis points to industry concerns about the effects
c. Occupied hours that fall within an hour or two of the transition of temperatures warmer than 22 °C on occupant cognitive perfor-
from one mode of space conditioning to the other. mance, based largely on Seppanen and Fisks’ [56] meta-analysis
study purporting to show a single temperature optimum for per-
There is no agreement in either the research or professional formance at precisely that temperature of 22 °C. Labor costs of an
communities about how best to define thermal comfort operating office building are widely acknowledged to be orders of magnitude
conditions for mixed-mode buildings in any of these three oper- higher than the energy savings accruing from implementation
ation regimes. There does seem to be agreement, however, that of either PMV/PPD or adaptive models for occupant comfort,
the chosen approach requires considered discussion between the therefore providing warmer and more comfortable temperatures
design team and the building owners (and/or managers for the is considered to be too costly in commercial buildings according
occupants, if different), as well as eventual occupant education to the adherents of the single temperature optimum “inverted U”
about the building operation and the occupants’ role in managing model [65]. The science underpinning the putative optimum per-
their own thermal environments. We recommend that design formance temperature of 22 °C has been thoroughly debunked (e.g.
teams consider two approaches for applying the comfort criteria [26,27,69]) and supplanted by the “extended-U” model. The latter
for mixed-mode buildings. These suggestions form our proposed fits a much larger body of experimental data than Seppanen and
nudge to practice when working with NV buildings. Fisks’ inverted U model [37] and it indicates optimum cognitive
12 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559

performance being maintained across a temperature range broadly peratures in air conditioned buildings. This could lead to improved
consistent with the adaptive thermal comfort zone. Nevertheless energy performance without sacrificing occupant comfort.
the buildings sector in western countries has strenuously ignored
advice to let indoor temperatures drift away from the 22 °C set- Declaration of Competing Interest
point, regardless of occupant discomfort, clothing insulation and
seasonal trends. We wish to confirm that there are no known conflicts of
The evidence presented in this paper indicates that cooling interest associated with this publication and there has been no
set-points in AC buildings are currently too low by any standard. significant financial support for this work that could have influ-
It is difficult to recommend suitable temperatures based on the enced its outcome. We confirm that the manuscript has been
results of AC buildings presented in Fig. 4 because the neutralities read and approved by all named authors and that there are no
were derived from building occupants adapted to overcooling. But other persons who satisfied the criteria for authorship but are not
given the strong dependence of neutrality on mean indoor tem- listed. We further confirm that the order of authors listed in the
perature presented in Fig. 6, it is clear that comfort temperatures manuscript has been approved by all of us. We confirm that we
in air conditioned buildings could potentially be higher if indoor have given due consideration to the protection of intellectual prop-
temperatures were to be nudged upwards at a slow enough rate erty associated with this work and that there are no impediments
to allow for adaptation. Regarding the most appropriate neutral to publication, including the timing of publication, with respect
temperature, we believe the summertime (0.5 clo) graphic comfort to intellectual property. In so doing we confirm that we have
zone on ASHRAE 55’s psychrometric chart remains the most ra- followed the regulations of our institutions concerning intellectual
tional choice. This would translate to set-points between 24–27 °C property. We further confirm that any aspect of the work covered
for relative humidity of 50% and typical air conditioned occupied in this manuscript that has involved either experimental animals
zone air speeds below 0.2 m/s. But how can we transition from or human subjects has been conducted with the ethical approval
current summertime practices (∼22 °C) to temperatures more of all relevant bodies and that such approvals are acknowledged
closely aligned with the PMV/PPD comfort model recommenda- within the manuscript.
tions of 24∼27 °C? The CoolBiz and Setsuden campaigns [33] are
precedents for abrupt changes to setpoint temperatures. But given Acknowledgements
the hypersensitivity of building occupants to changes in their work
environments, a sudden step-change in summertime setpoints The project was performed within the framework of the Inter-
by as much as 5 °C would very likely elicit complaints. Instead, national Energy Agency - Energy in Buildings and Communities
incremental steps spread out over a couple of weeks designed to Program (IEA-EBC) Annex 69 “Strategy and Practice of Adaptive
nudge indoor temperatures in the direction of outdoor climate Thermal Comfort in Low Energy Buildings.” Thomas Parkinson
could be feasible because it affords time for occupant’s adaptive is supported by the Republic of Singapore’s National Research
behavioral and perceptual mechanisms to adjust. Foundation Singapore through a grant to the Berkeley Education
Alliance for Research in Singapore (BEARS) for the Singapore-
5. Conclusion Berkeley Building Efficiency and Sustainability in the Tropics
(SinBerBEST) Program. The authors wish to acknowledge the
The recently released ASHRAE Global Thermal Comfort significant contributions made by Ariel Li on the modification of
Database II formed the basis of an investigation of the original the ASHRAE Database II.
ASHRAE adaptive comfort standard and exploration of additional
research questions. The aim of the analysis was to reappraise the Supplementary materials
scope of applicability of the adaptive comfort standard, assess
potential regional differences in adaptive comfort responses, and Supplementary material associated with this article can be
propose nudges to adaptive comfort theory, the adaptive com- found, in the online version, at doi:10.1016/j.enbuild.2019.109559.
fort model and standard, and building design and operational
conventions. We confirmed that the model remains valid in ap- References
proximating occupants’ comfort response in naturally ventilated
[1] S. Aghniaey, T.M. Lawrence, The impact of increased cooling setpoint temper-
buildings using prevailing outdoor conditions. The larger and more ature during demand response events on occupant thermal comfort in com-
geographically diverse database revealed that mean indoor tem- mercial buildings: a review, Energy Build. 173 (2018) 19–27 https://doi.org/10.
peratures in buildings in Asia were typically warmer than other 1016/j.enbuild.2018.04.068.
[2] V. Arel-Bundock, N. Enevoldsen, C.J. Yetman, “Countrycode: an r package to
regions, and that occupants in those buildings were better adapted convert country names and country codes, J. Open Source Softw. 3 (28) (2018)
to those conditions. There were sufficient mixed-mode buildings 848 https://doi.org/10.21105/joss.00848.
to determine that the relationship between neutral temperatures [3] M. Aria, C. Cuccurullo, bibliometrix: an R-tool for comprehensive science map-
ping analysis, in: J. Informetr., 11, 2017, pp. 959–975. Elsevier.
and outdoor conditions in such buildings was aligned more [4] ASHRAE, Standard 55 - Thermal Environmental Conditions for Human Occu-
closely with naturally ventilated buildings than those that were pancy, 2004, Refrigerating and Air-Conditioning Engineers, Atlanta, 2004.
air-conditioned. We also discovered clear evidence that people [5] ASHRAE, Standard 55 - Thermal Environmental Conditions for Human Occu-
pancy, 2017, Refrigerating and Air-Conditioning Engineers, Atlanta, 2017.
adapt most strongly to prevailing indoor temperatures irrespective [6] A. Auliciems, Towards a psycho-physiological model of thermal perception, Int.
of conditioning strategy (i.e., NV, MM and AC buildings). J. Biometeorol. 25 (2) (1981) 109–122 https://doi.org/10.1007/bf02184458.
These findings carry significant implications for both comfort [7] N. Baker, M. Standeven, Thermal comfort for free-running buildings, Energy
Build. 23 (3) (1996) 175–182 https://doi.org/10.1016/0378- 7788(95)00942- 6.
standards and building practice. As our suggested nudges, we rec-
[8] U. Berardi, A cross-country comparison of the building energy consumptions
ommend that the current ASHRAE 55 Adaptive Comfort Standard and their trends, Resour. Conserv. Recycl. 123 (2017) 230–241 https://doi.org/
remain in place, but propose an optional 1–2 K offset of the inter- 10.1016/j.resconrec.2016.03.014.
[9] G.S. Brager, R.J. de Dear, Thermal adaptation in the built environment: a
cept term for buildings located in Asia. We also believe the current
literature review, Energy Build. 27 (1) (1998) 83–96 https://doi.org/10.1016/
findings strongly support a modification of the limits of applicabil- S0378-7788(97)0 0 053-4.
ity, which should permit the use of the adaptive model for mixed- [10] G. Brager, G. Paliaga, R.J. de Dear, Operable windows, personal control and oc-
mode buildings. Finally, we suggest that building operation should cupant comfort, ASHRAE Trans. 19 (2004).
[11] S. Carlucci, L. Bai, R. de Dear, L. Yang, Review of adaptive thermal com-
better recognize occupants’ ability to adapt to indoor conditions by fort models in built environmental regulatory documents, Build. Environ. 137
implementing adaptive comfort algorithms to define setpoint tem- (2018) 73–89.
 T. Parkinson, R. de Dear and G. Brager / Energy & Buildings 260 (2020) 109559 13

[12] T. Cheung, S. Schiavon, T. Parkinson, P. Li, G. Brager, Analysis of the accuracy on [40] S. Manu, Y. Shukla, R. Rawal, L.E. Thomas, R. de Dear, Field studies of thermal
PMV – PPD model using the ASHRAE global thermal comfort database II, Build. comfort across multiple climate zones for the subcontinent: india model for
Environ. 153 (2019) 205–217 https://doi.org/10.1016/j.buildenv.2019.01.055. adaptive comfort (IMAC), Build. Environ. 98 (2016) 55–70 https://doi.org/10.
[13] R.J. de Dear, “A global database of thermal comfort field experiments, ASHRAE 1016/j.buildenv.2015.12.019.
Trans. 104 (1b) (1998) 1141–1152 V.. [41] J. Mendell, Mark, A.G. Mirer, Indoor thermal factors and symptoms in office
[14] R.J. de Dear, G.S. Brager, Developing an adaptive model of thermal comfort and workers: findings from the US EPA BASE study, Indoor Air 19 (4) (2009) 291–
preference, ASHRAE Trans. 104 (1998) 145–167. 302 https://doi.org/10.1111/j.1600-0668.2009.00592.x.
[15] R.J. de Dear, T. Akimoto, E.A. Arens, G.S. Brager, C. Cândido, K.W.D. Cheong, . . . [42] R. Miller, P. Williams, M. O’Neill, The Healthy Workplace Nudge: How Healthy
Y. Zhu, Progress in thermal comfort research over the last twenty years, Indoor. People, Culture, and Buildings Lead to High Performance, Wiley, 2018.
Air 23 (6) (2013) 442–461 https://doi.org/10.1111/ina.12046. [43] A.K. Mishra, M. Ramgopal, Field studies on human thermal comfort — an
[16] M.P. Deuble, R.J. de Dear, Mixed-mode buildings: a double standard in occu- overview, Build. Environ. 64 (2013) 94–106 https://doi.org/10.1016/j.buildenv.
pants’ comfort expectations, Build. Environ. 54 (2012) 53–60 https://doi.org/10. 2013.02.015.
1016/j.buildenv.2012.01.021. [44] K. Müller, (2017). Here: a simpler way to find your files. https://CRAN.
[17] M. Dowle, A. Srinivasan, (2019). Data.table: extension of ‘Data.frame’. https: R-project.org/package=here.
//CRAN.R-project.org/package=data.table. [45] A.T. Nguyen, M.K. Singh, S. Reiter, An adaptive thermal comfort model for hot
[18] EN 15251, Indoor Environmental Input Parameters for Design and Assessment humid south-east asia, Build. Environ. 56 (0) (2012) 291–300 https://doi.org/
of Energy Performance of Buildings—addressing Indoor Air Quality, Thermal 10.1016/j.buildenv.2012.03.021.
Environment, Lighting and Acoustics, Comité Européen de Normalisation, Brus- [46] F. Nicol, Adaptive thermal comfort standards in the hot–humid tropics, Energy
sels, 2007. Build. 36 (7) (2004) 628–637 https://doi.org/10.1016/j.enbuild.2004.01.016.
[19] P.J. Aphalo, Learn R ...as you learnt your mother tongue, Leanpub, Helsinki, [47] J.F. Nicol, M.A. Humphreys, Thermal comfort as part of a self-regulating
2016 https://leanpub.com/learnr. system, Build. Res. Practice 1 (3) (1973) 174–179 https://doi.org/10.1080/
[20] P.O. Fanger, Thermal comfort : Analysis and Applications in Environmental En- 09613217308550237.
gineering, Danish Technical Press, Copenhagen, 1970. [48] J.F. Nicol, M.A. Humphreys, Adaptive thermal comfort and sustainable thermal
[21] P.O. Fanger, J. Toftum, Extension of the pmv model to non-air-conditioned standards for buildings, Energy Build. 34 (6) (2002) 563–572 https://doi.org/
buildings in warm climates, Energy Build. 34 (6) (2002) 533–536. 10.1016/S0378-7788(02)0 0 0 06-3.
[22] V. Földváry Ličina, et al. (2018), ASHRAE global thermal comfort database II, [49] R. Rupp, J. Kim, E. Ghisi, R. de Dear, Thermal sensitivity of occupants in differ-
v4, UC Berkeley Dash, Dataset, https://doi.org/10.6078/D1F671 ent building typologies: the Griffiths Constant is a variable, Energy Build. 200
[23] V. Földváry Ličina, T. Cheung, H. Zhang, R. de Dear, T. Parkinson, E. Arens, . . . (2019) 11–20, doi:10.1016/j.enbuild.2019.07.048.
X. Zhou, Development of the ashrae global thermal comfort database ii, Build. [50] R. Core Team, R: a language and environment for statistical computing, R
Environ. 142 (2018) 502–512 https://doi.org/10.1016/j.buildenv.2018.06.022. Foundation for Statistical Computing, 2019. http://www.R-project.org/.
[24] A. Ghahramani, K. Zhang, K. Dutta, Z. Yang, B. Becerik-Gerber, Energy sav- [51] C. Rasmussen, (2016). Climateeng: tools to calculate climatic comfort related
ings from temperature setpoints and deadband: quantifying the influence of parameters. https://github.com/chrras/climateeng.
building and system properties on savings, Appl. Energy 165 (2016) 930–942 [52] F. Rodriguez-Sanchez, (2018). Grateful: facilitate citation of R packages. https:
https://doi.org/10.1016/j.apenergy.2015.12.115. //github.com/Pakillo/grateful.
[25] GSA Public Buildings Service (2009) Energy savings and performance gains in [53] A.C. Roussac, S. Bright, Improving environmental performance through innova-
gsa buildings – Seven Cost-Efffective strategies. (Accessed 26 May 2019 https: tive commercial leasing: an Australian case study, Int. J. Law Built Environ. 4
//www.gsa.gov/cdnstatic/GSA_SevenStrategies_090327screen.pdf). (1) (2012) 6–22 https://doi.org/10.1108/17561451211211714.
[26] P.A. Hancock, A dynamic model of stress and sustained attention, Hum. Factors [54] RStudio Team, RStudio: Integrated Development For R, RStudio, Inc., Boston,
31 (5) (1989) 519–537 https://doi.org/10.1177/001872088903100503. MA, 2018 URL http://www.rstudio.com/ .
[27] P.A. Hancock, H.C.N. Ganey, From the inverted-U to the extended-U: the evo- [55] R. Rupp, R. de Dear, E. Ghisi, Field study of mixed-mode office buildings in
lution of a law of psychology, J. Hum. Perform. Extrem. Environ. 7 (2003) 5–14 southern brazil using an adaptive thermal comfort framework, Energy Build.
http://dx.doi.org/10.7771/2327-2937.1023. V.158 (2018) 1475–1486.
[28] A. Honnekeri, G. Brager, S. Dhaka, J. Mathur, Comfort and adaptation in mixed– [56] O. Seppänen, W.J. Fisk, Some quantitative relations between indoor environ-
mode buildings in a hot-dry climate, in: Proceedings of 8th Windsor Confer- mental quality and work performance or health, HVAC R Res. 12 (4) (2006)
ence: Counting the Cost of Comfort in a Changing World Cumberland Lodge, 957–973.
Windsor, UK, 10-13 April 2014, London: Network for Comfort and Energy Use [57] M. Schweiker, S. Mueller, M. Kleber, B. Kingma, M. Shukuya, (2019). Comf:
in Buildings, 2014. functions for thermal comfort research. https://CRAN.R-project.org/package=
[29] T. Hoyt, E.A. Arens, H. Zhang, Extending air temperature setpoints: simulated comf.
energy savings and design considerations for new and retrofit buildings, Build. [58] M. Schweiker, A. Wagner, A framework for an adaptive thermal heat bal-
Environ. 88 (2015) 89–96 https://doi.org/10.1016/j.buildenv.2014.09.010. ance model (ATHB), Build. Environ. 94 (2015) 252–262 https://doi.org/10.1016/
[30] M.A. Humphreys, J. Fergus Nicol, The validity of ISO-PMV for predicting com- j.buildenv.2015.08.018.
fort votes in every-day thermal environments, Energy Build. 34 (6) (2002) [59] M.K. Singh, S. Mahapatra, S.K. Atreya, Adaptive thermal comfort model for
667–684 https://doi.org/10.1016/S0378-7788(02)0 0 018-X. different climatic zones of North-East India, Appl. Energy 88 (7) (2011)
[31] M.A. Humphreys, Field studies of thermal comfort compared and applied, 2420–2428.
Build. Serv. Eng. 44 (1976) 5–27. [60] A. South, Rworldmap: a new R package for mapping global data, R J. 3
[32] M. Indraganti, R. Ooka, H.B. Rijal, G.S. Brager, Adaptive model of thermal com- (1) (2011) 35–43 http://journal.r-project.org/archive/2011-1/RJournal_2011-1_
fort for offices in hot and humid climates of India, Build. Environ. 74 (2014) South.pdf.
39–53 https://doi.org/10.1016/j.buildenv.2014.01.002. [61] R.H. Thaler, C.R. Sunstein, Nudge: Improving Decisions About Health, Wealth,
[33] Y. Iwahashi, C. Tanabe, S. Tsushima, N. Nishihara, M. Hiraoka, H. Komoda, and Happiness, Yale University Press, New Haven, 2008.
S. Tabuchi, Effect of power saving measures on comfort, productivity and en- [62] D.H.C. Toe, T. Kubota, Development of an adaptive thermal comfort equation
ergy conservation, J. Env. Eng. AIJ 79 (704) (2014) 901–908. for naturally ventilated buildings in hot–humid climates using ASHRAE RP-884
[34] B. Li, R. Yao, Q. Wang, Y. Pan, An introduction to the chinese evaluation database, Front. Arch. Res. 2 (3) (2013) 278–291 https://doi.org/10.1016/j.foar.
standard for the indoor thermal environment, Energy Build. 82 (2014) 27–36 2013.06.003.
https://doi.org/10.1016/j.enbuild.2014.06.032. [63] V. Trouet, G.J. Van Oldenborgh, KNMI climate explorer: a web-based research
[35] P. Li, T. Parkinson, G. Brager, S. Schiavon, T.C.T. Cheung, T. Froese, A data-driven tool for high-resolution paleoclimatology, Tree-Ring Res. 69 (1) (2013) https:
approach to defining acceptable temperature ranges in buildings, Build. Envi- //doi.org/10.3959/1536-1098-69.1.3.
ron. 153 (2019) 302–312 https://doi.org/10.1016/j.buildenv.2019.02.020. [64] A.C. van der Linden, A.C. Boerstra, A.K. Raue, S.R. Kurvers, R.J. de Dear, Adap-
[36] O. Lucon, D. Ürge-Vorsatz, A. Zain Ahmed, H. Akbari, P. Bertoldi, L.F. Cabeza, tive temperature limits: a new guideline in the Netherlands: a new approach
N. Eyre, A. Gadgil, L.D.D. Harvey, Y. Jiang, E. Liphoto, S. Mirasgedis, S. Mu- for the assessment of building performance with respect to thermal indoor cli-
rakami, J. Parikh, C. Pyke, M.V. Vilariño, Buildings, in: O. Edenhofer, R. Pich- mate, Energy Build. 38 (1) (2006) 8–17 https://doi.org/10.1016/j.enbuild.2005.
s-Madruga, Y. Sokona, E. Farahani, S. Kadner, K. Seyboth (Eds.), Climate Change 02.008.
2014: Mitigation of Climate Change. Contribution of Working Group III to the [65] P. Wargocki, D.P. Wyon, Ten questions on thermal and indoor air quality effects
Fifth Assessment Report of the Intergovernmental Panel On Climate Change, on the performance of office work and schoolwork, Build. Environ. 112 (2017)
Cambridge University Press, Cambridge, United Kingdom and New York, NY, 359–366.
USA, 2014. [66] H. Wickham, (2017). Tidyverse: easily install and load the ‘Tidyverse’. https:
[37] O. Seppänen, W. Fisk, Some quantitative relations between indoor environ- //CRAN.R-project.org/package=tidyverse.
mental quality and work performance or health, HVAC and R Research 12 [67] L. Yang, H. Yan, J.C. Lam, Thermal comfort and building energy consumption
(2006) 957–973, doi:10.1080/10789669.2006.10391446. implications – A review, Appl. Energy 115 (2014) 164–173.
[38] M. Luo, B. Cao, J. Damiens, B. Lin, Y. Zhu, Evaluating thermal comfort in mixed- [68] R. Yao, B. Li, J. Liu, A theoretical adaptive model of thermal comfort—Adaptive
mode buildings: a field study in a subtropical climate, Build. Environ. 88 (2015) predicted mean vote (aPMV), Build. Environ. 44 (10) (2009) 2089–2096.
46–54 https://doi.org/10.1016/j.buildenv.2014.06.019. [69] F. Zhang, R. de Dear, P. Hancock, Effects of moderate thermal environments
[39] M. Luo, Z. Wang, G. Brager, B. Cao, Y. Zhu, Indoor climate experience, migra- on cognitive performance: a multidisciplinary review, Appl. Energy 236 (2019)
tion, and thermal comfort expectation in buildings, Build. Environ. 141 (2018) 760–777 https://doi.org/10.1016/j.apenergy.2018.12.005.
262–272 https://doi.org/10.1016/j.buildenv.2018.05.047.