1 Introduction
The world faces unprecedented challenges due to the rapidly increasing energy demand and the intensifying effects of climate change (
Change, 2007;
Cozzi et al., 2020). In current society, buildings are standing as the primary energy consumers with a consistent upward trajectory (
Bavaresco et al., 2021;
Programme, 2022). According to the Global Status Report on Buildings and Construction in 2020, buildings contribute 36% of global energy consumption and 37% of energy-related CO
2 emissions (
Hamilton et al., 2020). In the UK, about 40% of total energy consumption is used to maintain a comfortable indoor building environment (
Liu et al., 2014). Given the urgency of addressing climate change to improve human well-being, reducing the carbon emissions from buildings is of paramount significance (
Ackerly and Brager, 2013), with the energy conservation in buildings showing a significant potential in promoting sustainability (
Conti et al., 2016;
Gu et al., 2021).
As stated by
Nicol and Humphreys (2002), in buildings
“if a change occurs such as to produce discomfort,
people react in ways which tend to restore their comfort.
” Such kind of behaviour, like opening/closing windows, opening/ closing blinds, turning on/off air conditioners, is called adaptive behaviour (
De Dear and Brager, 1998), defining the interplay between the building users and building energy systems (
Balvedi et al., 2018). In existing studies, it has been well justified that occupant adaptive behaviour is a prominent factor influencing both buildings' energy consumption (
Ding et al., 2023;
Jian et al., 2015; Yan et al., 2015,
2017;
Yan et al., 2015a,
b;
Zhou et al., 2018) and indoor environment (
Qi et al., 2020;
Tahmasebi and Mahdavi, 2018,
2019). For example, it has been observed by Bahai and James (
Bahaj and James, 2007) that buildings with identical structures and designs could exhibit up to a 300% variance in their actual energy consumption, primarily due to the differing usage patterns of household appliances by the occupants.
Du and Pan (2021) studied 12 university student dormitories and found a similar result: the difference in energy consumption between occupants' air conditioning behaviour can be as high as 88.6%.
In naturally ventilated/mixed-mode of buildings, occupants' window behaviour (opening/closing windows) has garnered increasing attention to maintain comfortable and healthy indoor environment (
Sansaniwal et al., 2021;
Yu et al., 2022;
Yun and Steemers, 2008) with lower energy consumption (
Brittle et al., 2016;
Liu et al., 2013;
Tong et al., 2017). This behaviour has a profound influence on the performance metrics of buildings (
Bruce-Konuah, 2014;
Gu et al., 2021;
Pan et al., 2019;
Yun et al., 2012;
Zhang et al., 2017;
Zhang and Barrett, 2012;
Zhou et al., 2021a). For example, a building performance research (
Wang and Greenberg, 2015) highlighted the influence of window operations on office energy usage, as they found that a combination of natural ventilation and mechanical ventilation in summer would result in 17%–47% energy saving in HVAC systems. A Netherlands study has indicated that occupants' window-opening behaviour significantly affects the indoor environment during the heating season. The study found that the air temperature can decrease at an average rate of 0.18 K per minute when windows are opened. In the case of CO
2 concentration, this reduction amounts to 37 ppm per minute (
Boerstra et al., 2017).
Existing studies on window behaviour can be primarily categorized into four main types:
1) better understanding window behaviour, such as its driving factors (e.g., indoor/outdoor temperature and relative humidity) and patterns (e.g., regular modes and habits) (
Cheng et al., 2023;
D'Oca et al., 2015;
D'Oca and Hong, 2014;
Fabi et al., 2014;
Mahdavi et al., 2006;
Sun et al., 2018;
Wei et al., 2013;
Zhou et al., 2018);
2) developing new window behaviour models or enhancing existing models to lower the performance gap between predicted and actual building performance (
Gu et al., 2021;
Zhou et al., 2021b) and testing various modelling approaches (e.g., logistic regression analysis, random forest algorithm and XGBoost algorithm) (
Mo et al., 2019;
Sun et al., 2018;
Zhou et al., 2021b);
3) promoting building performance (e.g., improve the design of natural ventilation strategy (
Belleri et al., 2014), optimize algorithms for automated building ventilation (
Korsavi et al., 2021)) based on a better understanding of occupant window behaviour.
4) introducing novel data collection methods (e.g., optical cameras and experimental study) or research framework (e.g., novel longitudinal protocol and interdisciplinary framework integrated the theories from social science (
D'Oca et al., 2018;
Langevin et al., 2015a).
In existing studies, two paradigms, namely, engineering research (
Wagner et al., 2018) and social science research (
Bavaresco et al., 2020a) offer complementary strengths. Engineering is defined as “the application of science and mathematics by which the properties of matter and sources of energy in nature are made useful to people,” emphasizing the application and utility of knowledge to humanity (
Hynes and Swenson, 2013). Generally, engineering work engages in the design and analysis of technical equipment, processes, and systems, to address complex problems (
Pleasants and Olson, 2019). Existing window behaviour studies using engineering research methods may involve understanding the patterns of window behaviour, collecting window behavioural data, developing window behaviour models etc. Social science is defined as “the study of people: as individuals, communities and societies; their behaviours and interactions with each other and with their built, technological, and natural environments,” according to the UK Academy of Social Science (
Cary L. and Cooper). It is primarily on HOW and WHY people attend, perceive, think, and decide under different scenarios (
Hynes and Swenson, 2013). Existing studies using social science research methods have targeted to the reasons and rationales behind window usage, such as the influence of occupants' psychological factors, social factors, and cultural background etc.
Due to the differences between engineering and social science research methods, it is important to realize how they can be used to support window behaviour studies. Furthermore, a broad consensus has emerged in recent years (e.g., IEA EBC Annex 66 (
Yan et al., 2017), 79 (
O'Brien et al., 2020) and 95 (
Annex95, 2024)) that occupant behaviour is complex and requires contributions from various disciplines, to gain comprehensive understanding (
Barthelmes et al., 2021;
Hong et al., 2017;
Schweiker, 2017). To guide future research, it has become crucial to gain a deep understanding of studies conducted using both engineering research and social science research methods, including their aims, methodologies and theoretical foundations. Therefore, this study conducts a systematic literature review of window behaviour studies in office buildings, aiming not only to evaluate the methodological and empirical contributions from both paradigms but also to identify theoretical tensions and synergies. Furthermore, this review proposes a philosophically informed interdisciplinary framework, which serves as an initial attempt to support more philosophically grounded and theoretically coherent interdisciplinary research in occupant behaviour studies.
2 Methodology
2.1 Data collection
This study has adopted the systematic literature review method, which is “
a way of synthesising scientific evidence to answer a particular research question in a way that is transparent and reproducible,
while seeking to include all published evidence on the topic and appraising the quality of this evidence” (
Lame, 2019). This study follows the PRISMA protocol, which is widely used in systematic literature review and primarily encompasses four main steps, namely, 1) identification, 2) screening, 3) eligibility, and 4) inclusion (
Carlucci et al., 2020). This review method provides transparent and concise protocols and enables researchers to efficiently search and assess pertinent studies within a designated research area (
Kitchenham and Charters, 2007;
Tian et al., 2018). Additionally, the well-defined review criteria required in this method can also enhance the extraction of knowledge related to specific research queries and identify gaps within the research domain (
Grant and Booth, 2009;
Kitchenham and Charters, 2007).
These four stages, as shown in Fig. 1, succinctly outline the literature-searching strategy adopted in this review, the screening process, the criteria for literature selection, the rules for exclusion and inclusion, and the quantity of literature at each phase. This study has employed two primary electronic databases, namely, Web of Science and Scopus. They cover wide range of topics, contain high-quality literature, and provide advanced searching tools and filtering options (
Falagas et al., 2008). These advantages make them widely used in searching scientific literature (
Harzing and Alakangas, 2016). As this review work focuses on studies concerning window-opening behaviours in offices, the search criteria for both databases were: “occupant behaviour” OR “adaptive behaviour” OR “window behaviour” OR “user behaviour” OR “window opening behaviour” OR “window operation”, AND “window” OR “ventilation”, AND “office”, searching keywords in all title, abstract and keywords of existing literature. The review's scope encompassed articles published in peer-reviewed academic journals or national/international conferences, written in English. To ensure comprehensive understanding of the field, no restriction regarding to publication date was applied.
After the initial screening, literature was filtered based on three exclusion criteria in the eligibility phase:
1) excluding studies that do not encompass research on window-opening behaviours;
2) excluding studies that did not involve data collection;
3) excluding studies that do not include offices.
After the above procedures, 106 relevant studies considering office window behaviour were left for in-depth analysis.
2.2 Classification of studies
This review work has categorized the 106 studies into two groups: studies used engineering research methods, and studies used social science research methods. The categorization was carried out mainly by the inherent differences between engineering and social science research methods, as described above, based on the following three criteria:
1) Research aim: according to the aforementioned fundamental differences between engineering and social science studies. For the former, the studies often aim to address practical problems, such as improving building performance by understanding the impact of environmental parameters on window-opening behaviour and developing predictive models. For the latter, the studies often aim to investigate the mechanisms or rationales behind occupants' window behaviour.
2) Self-declaration of research method selection, such as for social science studies the expression could be “a novel longitudinal protocol from a theoretical framework in the psychology literature” (
Langevin et al., 2015a), and for engineering studies the expression could be “a year-long longitudinal monitoring of occupants” window opening behaviour (
Pan et al., 2016).
3) The authors' academic background information, such as whether their title includes “engineering” or “psychology”. This criterion was only applied when the above two criteria could not clearly identify the categorization of the study.
Based on these criteria, ninety-one studies (86%) were categorized as engineering-based studies, fifteen studies (14%) were categorized as studies used social science research methods. No studies were identified as exclusively social science-based.
3 Results
3.1 General findings of the review
The majority of the reviewed literature consists of journal articles, with 91 articles (86%). Key conference papers account for 15 articles (14%). Regarding sources, Building and Environment (n = 31) and Energy and Buildings (n = 14) are the two journals contributing the most significant number of articles included in this review. Regarding conference proceedings, The Windsor Conference (n = 5) and ASHRAE Transactions (n = 3) provided the most significant contributions. Further details on the number and distribution of journal publications are presented in Table 1.
Over the past forty years, there has been an escalating interest among researchers regarding occupants' window behaviour in offices. In the 20th century (i.e., C20 in Fig. 2), only two relevant publications were found. In the new century, the volume of relevant studies started to steadily increase, culminating in a peak in 2018, and has subsequently continued to attract a considerable degree of attention in the following years as shown in Fig. 2. It was also observed that the growth trend of engineering-based study differs significantly from that of studies involving social sciences. The former emerged as early as the late 20th century and has experienced rapid growth since 2008, continuing to the present. In contrast, the latter only began to appear a decade ago, and its growth trend has remained relatively modest.
This study generated a keyword visualisation map to provide an overview of the current state of research on window behaviour among office occupants (see Fig. 3).
Keyword co-occurrence analysis is a method that explores the knowledge structure of a field based on the topics of the reviewed studies (
Sedighi, 2016). It is also a key element of systematic literature reviews (
Balali et al., 2023). The visualisation map was created using bibliographic data exported from the reference management tool EndNote. The VOSviewer software was used to perform the keyword co-occurrence analysis, applying a full counting method. In the map, each node (circle) represents a keyword, with larger nodes indicating a higher frequency of occurrence. The proximity between nodes reflects the strength of their connections. Additionally, the colour of each node is associated with the recency of the topic, with older topics appearing in purple and newer topics in yellow.
Based on Fig. 3, two key research topics in the study of window behaviour among office occupants are thermal comfort and building performance prediction. Researchers have often focused on the impact of physical environmental factors, particularly indoor and outdoor temperatures, on occupants' comfort. Secondly, the primary aim of investigating occupant behaviour, including window behaviour, is to inform building engineering systems (
Healey and Webster-Mannison, 2012). As a result, many studies integrate data-driven modelling, simulation, and building physics to improve window behaviour and building performance predictions, thereby reducing the gap between predicted energy consumption in the design stage and actual energy consumption in building operations. Logistic regression has been widely recognised as a modelling approach in this field, while some researchers are exploring using more innovative machine learning models. In contrast, there is limited research on social science aspects, generally centred around theories of perceived control over occupant behaviour. A few emerging studies have investigated the influence of socio-psychological factors on window behaviour, but these remain a relatively minor focus within the field.
3.2 Engineering-based studies
3.2.1 Research aim
For most studies, deciding research aim is a necessary and critical step, so the research methodology can be properly decided (
Doody and Bailey, 2016). For all engineering-based studies collected in this review work, their research aims have been listed in Table 2, ranked by their number of appearances in existing studies. According to the summary, the research aims of engineering-based studies can be categorized into four types, namely, 1) to understand occupant window behaviour patterns based on non-psychological influential factors (
Sun et al., 2018;
Zhou et al., 2018); 2) to develop/improve window behaviour models for building performance simulation (
Nguyen et al., 2022;
Zhou et al., 2021a); 3) to promote building performance control based on a better understanding of occupants' window behaviour (
Kim et al., 2019;
Korsavi et al., 2021); 4) to develop more effective monitor methods for window usage (
Luong et al., 2022).
Over half of existing studies (51 of 91) aimed to understand window behaviours in offices in three main aspects, namely, 1) window-behaviour patterns, 2) driving factors, and 3) its impact on building performance and occupant productivity. Window-behaviour patterns generally focused on regular modes (
Zhou et al., 2018) and habits (
Cheng et al., 2023;
D'Oca et al., 2015;
D'Oca and Hong, 2014) of interaction between occupants and windows, including timing (
Cheng et al., 2023;
Pan et al., 2016;
Sun et al., 2018;
Yun et al., 2012), frequency (
Cheng et al., 2023;
Liu et al., 2012), duration (
Cheng et al., 2023;
Pan et al., 2016;
Sun et al., 2019), and extent of window opening (i. e., the degree to which the window was opened) (
D'Oca and Hong, 2014;
Warren and Parkins, 1984). For example,
D'Oca and Hong (2014) identified four window behaviour patterns, namely motivational, duration, interactivity, and position patterns, using data mining method.
About the driving factors on window behaviour, existing studies explored environmental factors, including both indoor (i.e., temperature (
D'Oca and Hong, 2014;
Schakib-Ekbatan et al., 2015;
Su and Wang, 2020;
Yun et al., 2012), relative humidity (
Fabi et al., 2014;
Sun et al., 2018), CO
2 (
Bruce-Konuah, 2012;
Fabi et al., 2014), PM
2.5 (
Zhou et al., 2018) and illuminance (
Sansaniwal et al., 2021)) and outdoor parameters (i.e., temperature (
D'Oca and Hong, 2014;
Liu et al., 2012;
Schakib-Ekbatan et al., 2015;
Zhou et al., 2018), rainfall (
Fabi et al., 2014), wind speed (
Sansaniwal et al., 2021), solar radiation (
Sansaniwal et al., 2021) and PM
2.5 (
Wei et al., 2015)), and non-environmental factors, including time of day (
Pan et al., 2018;
Yun and Steemers, 2008), occupancy (
Zhai et al., 2019), season (
Al-Atrash et al., 2018), habits (
Belafi et al., 2017), age (
Marín-Restrepo et al., 2020) and gender (
Wei et al., 2013). For example,
Zhou et al. (2018) conducted a field study in an open-plan office, finding that outdoor temperature, the daily work schedule of occupants, and the state of air conditioners influence window opening behaviour.
Some studies tried to understand the impact of window behaviour, with one major stream about the overall building performance (
Ahadzie et al., 2021;
Pan et al., 2017;
Schakib-Ekbatan et al., 2015) and another about occupants' satisfaction of work environment (
Gnecco et al., 2024;
Mao et al., 2022;
Yun and Steemers, 2010). For example, existing studies have confirmed that window behaviour could significantly affect the thermal performance (
Mao et al., 2022;
Raja et al., 2001;
Yun and Steemers, 2010), indoor air quality (
J. Kim et al., 2019;
Li et al., 2024;
Li et al., 2023;
Shen et al., 2012). Additionally, they also confirmed significant impact on the energy consumption of buildings (
Pan et al., 2016;
Schakib-Ekbatan et al., 2015). To quantify occupants' satisfaction of their working environment, existing studies used Likert-Scale as their survey tool (
Indraganti et al., 2018;
Manu et al., 2016). Window-opening behaviour, as a part of indoor environmental control, is a crucial indicator of occupants' satisfaction with their office environment (
Boerstra et al., 2014).
Almost half of the studies aimed to model window behaviour using mathematical methods. This work is important to drive building performance. For the modelling work, statistical approaches, such as logistic regression and Markov chain models, have been adopted (
Grassi et al., 2022;
Liu et al., 2021;
Sun et al., 2018;
Yun et al., 2008;
Zhang et al., 2018;
Zhou et al., 2021a). In recently years, with the fast development of computer science, advanced machine learning methods, such as artificial neural networks, deep learning, and gauss distribution, have also been applied to model window behaviour (
Banihashemi et al., 2024;
Markovic et al., 2018,
2019;
Pan et al., 2019). For example,
Wei et al. (2019) conducted a field study to collect environmental data from an office building and used it to train an artificial neural network (ANN) model, which outperformed logistic regression and Markov process models in predicting window behaviour.
Markovic et al. (2018) used deep learning methods to model window-opening behaviour, finding that their model, trained on data from three offices and evaluated in 49 others, achieved high accuracy, outperforming traditional prediction methods.
Some researchers tried to promote building performance control based on a better understanding of occupants' window behaviour. The patterns of window use by office occupants can be utilized to improve the design of natural ventilation in the early stages of building design (
Belleri et al., 2014), and the status of windows can also be used for more accurate detection of occupancy (
Naylor et al., 2018). Occupants' window behaviour is crucial to the automated control systems of building ventilation. For example,
H. Kim et al. (2019) and
Korsavi et al. (2021) conducted field studies in offices to understand window operation patterns and influential factors, using this knowledge to develop and improve algorithms for automated building ventilation control, thereby enhancing indoor environmental quality and occupant satisfaction.
A few researchers have recognised the Hawthorne Effect (
Yan et al., 2015a,
b), wherein study participants alter their typical behaviour due to the awareness of being recorded, as a limitation when using on-site devices like temperature and humidity sensors, carbon dioxide concentration sensors, and magnetic sensors for window detection. To mitigate this impact, researchers have developed methods for detecting window operations based on cameras outside the building. This detection scheme can automatically identify windows on the facade using cameras positioned outside the building and recognise the status of each window from the images (
Bourikas et al., 2018;
Luong et al., 2022). This approach avoids the Hawthorne Effect and significantly enhances the efficiency of window detection (as traditional methods require installing a sensor on each window).
3.2.2 Data collection methods
To achieve the above research aims, researchers have adopted a variety of methods for data collection. In this review, a detailed analysis has been executed based on the type of data acquisition, the variables studied, and the data collection methods (see Table 3). The data used for window behaviour studies often requires information in terms of both window behaviour and relevant driving factors.
3.2.2.1 Data acquisition of window behaviour
Regarding occupant window behaviour, data collected from offices can be primarily used to decide the status of the windows (whether open or closed and the extent of opening) (
Bourikas et al., 2018;
Herkel et al., 2008;
Nguyen et al., 2022), propose window usage patterns (frequency and nature) (
Rijal et al., 2009;
Zhou et al., 2018), or to reflect occupant window actions (
Liu et al., 2012;
Marín-Restrepo et al., 2020). For these three directions, different methods have been employed for data collection. For window status, most researchers (59 of 91) used data loggers/sensors (i.e., magnetic induction devices (
Gu et al., 2023;
Herkel et al., 2005;
Jia et al., 2019;
Naspi et al., 2018b;
Pisello et al., 2016;
Ren et al., 2014), microswitches (
Haldi and Robinson, 2009)) or cameras (
Bourikas et al., 2018;
Gilani et al., 2017;
Luong et al., 2022;
Zhai et al., 2019;
Zhang and Barrett, 2012) for automatic and continuous monitoring, providing realtime data for analysis (
Pan et al., 2017). A smaller number of researchers asked participants to self-record the status of windows during the study period using a daily survey sheet, with intervals from 15 min interval (
Jia et al., 2019) to 8 times a day (
Zhang et al., 2024). Some researchers also adopted observation methods to record window status, and they visited the offices under study 2–3 times per day (
Ooka et al., 2014;
Wei et al., 2013,
2014).
Regarding window actions, researchers believe that the dynamic nature of occupants' decisions about the current state of windows is influenced by their previous state (
Haldi and Robinson, 2008;
Wei et al., 2015). To capture this dynamic behaviour, it is essential to measure the time when occupants change the state of windows or, alternatively, to record the state of windows at high frequency. For instance, electronic measuring devices, such as sensors, offer significant potential for analysing the dynamic nature of window actions due to their high measurement frequency, as previously mentioned. Besides, some researchers asked occupants to use a questionnaire to record any adaptive actions taken in the office at specific times of the day (like morning, noon, evening) or a specified number of times (2–3 times per day), during the last 20 min (
Liu et al., 2012) to 60 min (
Al-Atrash et al., 2018;
Damiati et al., 2016;
Rowe, 2004). Some researchers opt for face-to-face surveys, personally visiting the office three times a day to inquire about and record the window behaviour undertaken by the occupants during these periods (
Marín-Restrepo et al., 2020). Additionally, in their study,
Barthelmes et al. (2021) required participants to self-report each window operation through an app installed on a smartphone near the window, including the motivations behind these actions. In the study by
Rupp et al. (2021), a team of researchers conducted real-time observations of office occupants' behaviour and used a standard electronic spreadsheet to record various adaptive actions.
For studying window usage patterns, researchers typically employed a cross-sectional questionnaire to gather relevant information (
Zhou et al., 2018). These questionnaires were designed for collecting data about the frequency, timing, duration, nature, simple reasons, and ease of use associated with occupant window operation (
Almeida et al., 2022;
Bitomsky et al., 2020;
Manu et al., 2016). Besides patterns, such questionnaires can also concurrently gather basic information about the occupants and their satisfaction with their office environment (
Indraganti et al., 2018;
Zhou et al., 2018), and this information assisted researchers' analysis in typical window behaviour patterns (
Zaidan et al., 2021;
Zhou et al., 2018).
Collecting environmental and non-environmental influential factors.
In terms of influential factors, although there is currently no official classification methods (
Pereira and Ramos, 2019), many researchers have divided them into environmental (e.g., indoor and outdoor environmental conditions) and non-environmental (e.g., personal, social, and contextual) factors (
Fabi et al., 2012;
Inkarojrit, 2005). This review summarised influential factors that researchers collected in their study, as indicated in Table 4.
Environmental factors investigated most frequently include indoor and outdoor temperature, appearing across many studies (
Grassi et al., 2022;
Gu et al., 2023;
Nguyen et al., 2022;
Rupp et al., 2021). Other commonly reported parameters are relative humidity, CO
2 concentration, wind speed, solar radiation, and indoor air velocity. Less commonly studied factors include PM2.5, TVOCs, and illuminance. Measurements were typically obtained via in-situ sensors (e.g., HOBO for temperature, anemometers for air velocity) or weather stations positioned nearby or on rooftops (
Gu et al., 2023;
Sun et al., 2018).
Non-environmental factors (Table 4) include three main groups: occupancy, personal factors, and contextual conditions. Occupancy data, present in 35% of studies, is usually detected via PIR sensors, CO
2 levels, door status, or direct observation (
Liu et al., 2021;
Wei et al., 2013;
Zhou et al., 2021a). Personal factors cover attributes such as clothing insulation, gender, age, habits, and preference. Contextual factors refer to elements such as the time of day, season, office layout, distance from windows, and office size. Most of these influential factors, such as occupancy, time of day, clothing insulation level, are easily quantifiable and can be collected using corresponding sensors or conducting surveys (
Liu et al., 2013).
3.3 Studies combining social science and engineering method
3.3.1 Research aim
The research aims of studies on office window behaviour that incorporate social science research methods can be categorized into three types: 1) to better understand occupant how and why use the window, 2) to develop new research methodology, and 3) to change occupant window usage by interventions. Table 2 (Section 3.2.1) has summarised the research aims.
As indicated in Tables 2, 9 out of 15 studies (60%) using social science research methods aimed to better understand the drivers/rationales of occupant window behaviour, especially those are difficult to quantify (
Langevin et al., 2015a). For example, researchers have explored the influence of many non-physical factors, including psychological factors, design and construction, operation and maintenance, on occupants' adaptive behaviour (
Liu et al., 2014). Besides, they focus more on social factors, which mainly include occupants' personality traits (
Schweiker et al., 2016), the number of people in the office (
Schweiker and Wagner, 2016), sociocultural attitudes (
Indraganti et al., 2015), social-psychological factors (
Weerasinghe et al., 2023), intention to share control and perceived behavioural control (
Bavaresco et al., 2020b,
2021), and occupants' perceived user control satisfaction (
Weerasinghe et al., 2022).
To better understand the impact of social factors on occupant behaviour, including window usage, some researchers have integrated social science theories into their studies, developed novel methodological approaches for a more comprehensive and in-depth understanding of occupant behaviour. For instance,
Schweiker et al. (2012) designed an innovative experimental procedure to quantify occupants' behavioural, physiological, and psychological adaptation processes in indoor environment.
Langevin et al. (2015a) created a new longitudinal study protocol drawing on psychological theoretical frameworks to better understand how and why office occupants interact with their environment, including window usage.
D'Oca et al. (2018) developed an interdisciplinary framework synthesized from building physics and social-psychological theories, these approaches enabled the exchange of socio-technical knowledge and promote co-learning on the interactions between humans and buildings.
Intervening and guiding occupant window-opening behaviour to achieve more energy-efficient behavioural strategies is a practical research attempt in window behaviour studies (
Neves et al., 2020). For example, to encourage window operation,
Bitomsky et al. (2020) utilized real-time display of indoor environmental quality to demonstrate the effectiveness of digital nudging. The study emphasized the foundation of behavioural economics and digital nudging theories in understanding and influencing occupant behavioural decision making.
Zaidan et al. (2021) in their research have evaluated occupants' previous window-opening habits to raise occupants' environmental awareness and provide behavioural adjustment recommendations. The study aimed to educate (intervened) occupants to adopt energy-saving adaptive behaviours. Similarly, in a study conducted in Karlsruhe, Germany (
Meinke et al., 2017), researchers found that providing building occupants with information on the comfort and energy implications of different cooling strategies significantly influenced their choices, demonstrating that feedforward information can enhance rational interaction with the built environment by integrating social science insights into human behaviour and decision-making.
3.3.2 Data collection methods
In this section, studies incorporating social science methods were analysed, as summarised in Table 3 (Section 3.2.2). Unlike engineering research methods, social science research methods focused mainly on exploring non-environmental factors that are difficult to quantify, so questionnaires and interviews have been widely used.
3.3.2.1 Data acquisition of window behaviour data
In these studies, researchers confirmed multidimensional triggers/drivers of window opening behaviour using sophisticated questionnaires (
Bavaresco et al., 2021). These triggers/drivers include Indoor Environmental Quality (IEQ) satisfaction (
Weerasinghe et al., 2022), perceived behavioural control (
Bavaresco et al., 2020b;
Schweiker and Wagner, 2016), intention towards sharing the control (
Bavaresco et al., 2020b,
2021), and occupants' perceived user control satisfaction (
Weerasinghe et al., 2022). These questionnaires typically adopted Likert scales, with a seven-point scale from 1 (utterly dissatisfied) to 7 (completely satisfied) to express occupants' satisfaction level to their surrounding environment (
Weerasinghe et al., 2022), and a five-point scale from 1 (no control) to 5 (complete control) to assess occupants' perception of control over windows (
Bavaresco et al., 2020b).
3.3.2.2 Collecting non-environmental factors
Regarding the depth and breadth of data collection, studies combining social science research methods have devoted considerable effort to researching non-environmental factors that are difficult to quantify, as shown in Fig. 4. These non-environmental factors can be categorized into:
● Perception of window control, originating from the social psychological theory of planned behaviour, evaluates an individual's assessment of their capability to perform specific behaviours, related to anticipated difficulty and controllability (
Ajzen, 2020). Researchers generally use the seven-point scale (
Liu et al., 2014;
Schweiker and Wagner, 2016;
Weerasinghe et al., 2022) or the five-point (
Bavaresco et al., 2020b) for collection.
● Social-psychological factors include concepts like group norms and subjective norms (
D'Oca et al., 2018) and intention, ease, attitudes, and expectations to share control of windows (
Bavaresco et al., 2020b,
2021;
Zaidan et al., 2021). Subjects can express their responses using a Likert five-point scale, where 1 represents a very negative answer, and 5 represents a very positive one (
Bavaresco et al., 2021).
To delve deeper into these factors, researchers have integrated theoretical frameworks from related scientific fields when designing their research methodologies. For instance,
D'Oca et al. (2018) proposed an integrated research framework combining building physics with social sciences to study the interaction between occupants and buildings in office environments. This framework employed the Drivers–Needs–Actions–Systems (DNAS) framework (
Hong et al., 2015), predominantly based on physical sciences, to rationalize occupant behaviour related to building types and comfort; it is combined with purely social theories, such as the Social Cognitive Theory (SCT) from
Bandura (1986) to explain environmental, cognitive, and behavioural factors in human decision-making processes within a social context; and references the Theory of Planned Behaviour (TPB) (
Ajzen, 1991) to refine DNAS. Based on the proposed framework, they designed a questionnaire survey, encompassing the collection of factors from building physics and social psychology. Besides some persistent interests in occupant behavioural studies, such as environmental comfort, satisfaction and thermal comfort, some questions were specifically designed to explore the influences of some non-environmental factors that are difficult to quantify, such as behavioural intention, control beliefs, perceived control, normative beliefs, subjective norms, and ease of sharing control from social psychology. One measure of these factors was using Likert-scale, which was defined as either 7-point scale or 5-point scale, with 1 indicating strong disagreement and 7 or 5 indicating strong agreement, such as “How would the following conditions influence your current productivity at work?” (
D'Oca et al., 2018) and “To what extent are you satisfied or not satisfied with the following conditions in your workspace?” (
Bavaresco et al., 2020b). Another measure was questions used for occupants' behavioural intention in sharing the control systems, such as “Do you have control to open or close the windows in your workspace?” (
D'Oca et al., 2018) and “Please tell us why you normally open the window(s) at work during different seasons?” (
D'Oca et al., 2018). Subsequently, in Brazil,
Bavaresco et al. (2020b) used this interdisciplinary framework to assess subjective aspects (i. e., intention to share control, ease to share, attitudes, expectations of sharing, knowledge to control) on the choices of occupant adaptive behaviours, like window usage. In the following study, they utilized this method to figure out the impact of multi-domain triggers on occupant behaviour (
Bavaresco et al., 2021).
Langevin et al. (2015a) built on existing data collection and analysis methods in occupant behaviour research, combined the theory of comfort-driven behaviour from The Perceptual Control Theory (PCT) to propose a longitudinal study protocol. This method included semi-structured interviews, longitudinal surveys (thrice-daily questionnaires), the collection of Personal values (social restrictions on behaviour and other reasons that occupants may not take available adaptive actions), and continuous measurement of the indoor environment surrounding the subjects.
Several researchers have employed experimental studies to investigate the socio-psychological factors influencing the adaptive behaviours of occupants in office environments, including window-opening behaviours. For instance, in 2016,
Schweiker et al. (2016) conducted a laboratory study to explore the impact of occupants' personality traits (three of the Big Five personality traits describing human personality, namely neuroticism, extraversion, and openness to new experiences) on behaviours such as window opening. Participants were asked to conduct office activities in a wholly controlled environment and complete a questionnaire every 90 min (six times daily). The questionnaire included occupants' thermal sensations and tendencies (7-point categorical ASHRAE thermal sensation scale and 5-point categorical preference scale) and a personality questionnaire (5-point categorical scale). Similarly, they used the this method to test the impact of office occupancy on perceived control and behavioural patterns (
Schweiker and Wagner, 2016).
Some non-environmental factors can be collected through open-ended questions in questionnaires. For example,
Indraganti et al. (2015) utilized open-ended questions (e.g., “why are you not operating the window?” “Who operates the windows in this place?” and “Do you freely operate the window?”) to understand the barriers faced by office occupants in India when operating windows. These barriers include aspects related to design and construction, operation and maintenance, environmental issues, sociocultural factors, and attitudinal challenges.
3.4 Useful theories and frameworks for behavioural studies
Wagner et al. (2018) in their work
Exploring Occupant Behavior in Buildings, have emphasized that any research plan should integrate methods, theories (within a theoretical framework), and research design. This importance should not be overlooked studying occupant window behaviour. Therefore, this study reviewed theories and frameworks of all 106 studies and found significant differences between engineering-based studies and studies used social science research methods, as shown in Fig. 5.
3.4.1 Theory
In studies using engineering research methods, only a few studies (10%) have mentioned the relevant behavioural theories, with an even smaller fraction developing a research framework based on those theories. Therefore, it seems like that engineering-based research focused more on methods that can be used to solve real problems, not theory development. According to the review work, most engineering-based studies aimed to develop and optimize models of occupant window-opening behaviour to enhance the accuracy of building performance simulation (
Grassi et al., 2022;
Li et al., 2015;
Ren et al., 2014). Their primary focus was mainly on practice and application, such as the utilization of new data collection methods (
Bourikas et al., 2018;
Luong et al., 2022), the employment of new data analysis techniques (
D'Oca and Hong, 2014;
Sun et al., 2019), the adoption of new modelling methods (
Nguyen et al., 2022;
X. Zhou et al., 2021a), and the implementation of behavioural models for building design and operation (
Belleri et al., 2014;
Gilani et al., 2017). Regarding theory, engineering-based studies on window-opening behaviour predominantly explored occupant behaviour through Humphrey's adaptive thermal comfort theory (
Nicol et al., 2012), which posits, “if the environment changes and occupants feel discomfort, they will take actions to restore comfort.” Although most studies do not explicitly mention the adaptive behaviour theory, they assumed that occupants would react to regain comfort. These studies concentrated on the environmental determinants of discomfort and related behaviours, with only a few attempting to explore social factors. For instance,
Langevin (2015b) investigated the rules of occupant proxy behaviour through Perceptual Control Theory (PCT), incorporating personal preference as a factor influencing occupant behaviour models. In the study carried out by
Ahadzie et al. (2021), they designed their survey content based on the Theory of Planned Behaviour (TPB).
In contrast, most studies (67%) incorporating the social science research methods emphasized relevant theories of occupant behaviour and developed comprehensive research frameworks. For instance,
Langevin et al. (2015a) developed a novel protocol for longitudinal studies grounded in psychological theoretical frameworks, aiming to elucidate the mechanisms and reasons behind the interactions of office occupants with their surroundings, such as operating windows. In the investigation conducted by
D'Oca et al. (2018), a research framework bridging disciplinary boundaries was established, merging insights from building physics and social science. This framework utilized an integration of the Drivers–Needs–Actions–Systems (DNAS) model, the Social Cognitive Theory (SCT), and the TPB, offering a pioneering interdisciplinary approach introducing innovative insights from social science.
Bavaresco et al. (2020b,
2021) and
Weerasinghe et al. (2022) also adopted this novel interdisciplinary research framework in their studies, and their investigations demonstrated that studies integrating the social science component often tightly weaved research methods with theory, with researchers aiming to validate the appropriateness of methods based on theoretical grounds.
3.4.2 Framework
In terms of frameworks, the reviewed studies predominantly used the term “research framework” without referencing a “theoretical framework”. A research framework, also known as a conceptual framework, usually follows a typical procedure, including 1) a literature review to identify research gaps need further investigation, and 2) select appropriate data collection and data analysis methods to fill the gaps (
Varpio et al., 2020). This procedure aids researchers in organizing and implementing their studies, ensuring the systematic coherence and consistency among different studies (
Kivunja, 2018;
Wagner et al., 2018). A theoretical framework, on the other hand, serves as the theoretical foundation guiding the research derived from one or more existing theories (
Kivunja, 2018). It is the interpretation of the work undertaken by researchers using relevant theories within a specific study, not only a summary of the researchers' views on their studies, but also a synthesis of the seminal thoughts within the research area (
Lederman and Lederman, 2015;
Varpio et al., 2020).
The review results indicated that most engineering-based studies (98%) followed a typical process. This process involved understanding the current state of the problem through a literature review, identifying gaps in the understanding of the phenomenon or issue, determining the specific research questions or variables to be investigated, and selecting the necessary research methods (
Haldi et al., 2017;
Jia et al., 2019;
Pan et al., 2017). However, only a few studies (10%) mentioned the theories they adopted and the research frameworks developed based on these theories (2%). Most researchers implicitly followed the aforementioned typical research framework. In engineering-based studies, the framework was usually established before the research began, and once data collection started, it was rarely modified. In contrast, in the social sciences, there is an expectation to apply theoretical frameworks to validate the appropriateness of methods based on theory. For instance, Peshkin expressed his concerns that research not driven by theory could be seen as flawed or, worse, overlooked (
Peshkin, 1993); Norman and others have stated that theoretical frameworks are crucial for all our work, and all research papers should have a robust theoretical framework to demonstrate the importance and significance of the research (
Lederman and Lederman, 2015). Thus, we observe that research on window behaviour that integrates a social science component places greater emphasis on theory (67% studies mentioned theories they adopted) and develops comprehensive research frameworks (60% studies) based on theory compared to engineering-based studies.
It is important to note that the engineering-based study is typically deductive, focusing on empirical data and quantifiable outcomes (
Robinson, 1998). In contrast, a social science based study is inductive, aiming to develop a deeper understanding of underlying mechanisms, such as human motivations, which cannot be measured directly (
Kuczynski and Daly, 2003). Theories offer a conceptual abstraction, enabling generalization across various contexts. For example, the TPB posits that social norms are a crucial factor in shaping intentions, applicable across all building and occupant types (
Ajzen, 1991). This theoretical perspective allows for a more comprehensive understanding of behaviours beyond specific empirical observations. Wanger provides a detailed description of frameworks in the context of occupant behaviour research in his book
Exploring Occupant Behavior in Buildings, illustrating that methods (specifying what to measure), research design (determining if variables have a causal relationship), and theory (within a theoretical framework) need to be combined to design any research plan (
Wagner et al., 2018).
3.5 Comparing engineering and social science research methods
This review highlights the contributions of engineering and social science research methods in studying occupant window behaviour in office settings. Through this review, key differences between the two paradigms have been identified, along with areas of overlap. The strengths and limitations of the data collection methods employed in both paradigms have been critically examined and summarised. Additionally, a research framework has been proposed to illustrate how social science research methods can be applied to study office window behaviour.
3.5.1 Difference and overlaps
This study employed a Sankey diagram to visualise the research aims, intended data types, data collection methods, analysed data, and theoretical foundations of the two research paradigms in office window behaviour studies (see Fig. 6). The varying thickness of the lines represented the strength of the connections between these elements. For example, “develop/improve models” and “understand window behaviour” were the primary aims in studies adopting engineering approaches to investigate office window behaviour. The thicker lines in the Sankey diagram indicated that these connections were observed in more studies. In contrast, fewer studies in the review employed social science approaches, resulting in thinner corresponding connections. It is important to note that the thickness of the lines did not indicate the significance of the connections but rather their relative frequency among the reviewed studies. The data used to construct this diagram, which determined the thickness of each connection, was derived from the findings presented in Sections 3.2, 3.3, and 3.4 of this review. To intuitively highlight the differences and areas of overlap between the two paradigms, grey lines were used to represent engineering research, while pink lines represented social science research.
Figure 6 shows the differences between the two research methods regarding research aims, data collection methods, key research variables, and underlying theoretical foundations. Engineering studies were largely based on (or implicitly assumed) Humphreys' adaptive thermal comfort theory and primarily focused on improving the accuracy of window state predictions and understanding the relationship between occupants' thermal comfort, window behaviour, and environmental conditions. To achieve these goals, continuous, high-precision data monitoring methods were required, with research variables predominantly comprising physical environmental factors (
Liu et al., 2012;
Yun et al., 2012) or physiological indicators such as age and gender (
Pan et al., 2018;
Wei et al., 2013). In contrast, studies employing social science research methods emphasized understanding occupant behaviour, motivations, and decision-making processes. These studies aimed to provide a holistic understanding of how and why building occupants interact with their environment (
Langevin et al., 2015a), focusing mainly on social factors such as psychological aspects (
Weerasinghe et al., 2023), socio-cultural factors (
Indraganti et al., 2015), and group dynamics (
Bavaresco et al., 2020). Such studies were often grounded in social-psychological theories, including the Theory of Planned Behaviour and Perceptual Control Theory. However, there were areas of overlap between the two paradigms. For instance, regarding research aims, both approaches explored the influence of non-environmental factors. Additionally, there was considerable overlap in data collection methods, as both research methods utilized sensor-based monitoring to record window states and environmental parameters while employing survey-based approaches to understand occupants' adaptive behaviours or window usage patterns. These differences and overlaps provided valuable insights for researchers, enabling a better understanding of the distinct emphases of different research methods and highlighting the potential for interdisciplinary collaboration.
3.5.2 Data collection methods
This review summarised various data collection methods employed in the study of office window behaviour: 1) automatic measurement using electronic devices (e.g., window status sensors (
Haldi and Robinson, 2009;
Pisello et al., 2016), environmental measurement sensors (
Cheng et al., 2023;
Gu et al., 2023;
Naspi et al., 2018b;
Sun et al., 2018), occupancy sensors (
Wei et al., 2019)); 2) various types of questionnaires (e.g., cross-sectional surveys (
Almeida et al., 2022;
Zhou et al., 2018), self-report surveys (
Jia et al., 2019;
Sansaniwal et al., 2021), openended questionnaires (
Indraganti et al., 2015), ASHRAE Scale (
Rowe, 2004;
Sansaniwal et al., 2021;
Schweiker and Wagner, 2016), Likert Scale (
Bavaresco et al., 2020;
Bavaresco et al., 2021;
D'Oca et al., 2018)); 3) in-depth interviews (e.g., semi-structured interviews (
Langevin et al., 2015a)); and 4) on-site observation (
Ooka et al., 2014;
Wei et al., 2013), as shown in Fig. 6.
Engineering-based studies mainly used electronic devices (e.g., sensors, camera), cross-sectional questionnaire (one-time questionnaire) or longitudinal questionnaire (self-reported questionnaire), and on-site observation. Regarding social science study, they mainly adopted questionnaire and interview methods to collect data. Due to the rise of interdisciplinary research, modern science has started to blur the boundaries between disciplines and encourage shared method usage (
Okamura, 2019). According to the review work, the questionnaire method has been commonly used in both engineering-based studies and social science studies, although used for collecting different type of information. In engineering-based studies, questionnaire were mainly used to capture demographic information (e.g., gender, age and habit) (
Almeida et al., 2022;
Pisello et al., 2016;
Wei et al., 2013) and occupant comfort levels (e.g., thermal comfort and indoor air quality) (
Shen et al., 2012;
Sun et al., 2019), while in social science studies, they were used to gather richer insights into motivations and perceptions related to window behaviour (
D'Oca et al., 2018), to deepen the understanding of this research topic. Similarly, scaled questionnaires were utilized in both fields. In engineering studies, the ASHRAE scale was employed to quantify occupants' sensation and preference for the environmental conditions (e.g., room temperature, humidity) (
Rowe, 2004;
Sansaniwal et al., 2021;
Zhai et al., 2019). The Likert scale was also used to measure and quantify occupants' satisfaction with their office environment (
Weerasinghe et al., 2023). In social science research, the Likert scale was applied to quantify more abstract socio-psychological factors such as group norms, intentions, ease, attitudes, and expectations regarding the shared control of windows (
Bavaresco et al., 2020,
2021;
Zaidan et al., 2021). Each method has its own strengths and limitations, as shown in Table 5. Understanding these can help researchers design experiments based on their specific research needs.
3.5.3 Contributions of social science research methods
This section further refined and synthesized the methodological insights from the reviewed studies and proposed a diagram to illustrate the application of social science research methods in window behaviour studies, as shown in Fig. 7. It is essential to understand how social science has contributed to window behaviour studies in existing research, as summarised in the following sections, both before, during, and after engineering studies.
1)
Before: social science research methods used in guiding the development and execution of subsequent engineering research. For example,
Langevin et al. (2015a) carried out a semi-structured interview at the beginning of study to uncover unexpected insights into occupant window behaviour and provide substantial information for developing and interpreting subsequent longitudinal study, such as the preliminary determination of research samples and variable ranges.
2)
Alongside: social science research methods used along with engineering research approaches in data collection. In these studies, social science questions were combined in the questionnaires to collect information about occupants' intentions, ease, attitudes, and expectations in using office windows (
Bavaresco et al., 2021;
Liu et al., 2014). For example,
Bavaresco et al. (2021) employed an interdisciplinary framework (questionnaire) to collect subjective aspects of occupants' adaptive behaviour choices, such as the willingness to share control, the perceived ease or difficulty of sharing, attitudes and expectations regarding sharing of control.
3)
After: social science research methods used in better interpreting the potential drivers/constraints of the monitored window behaviour in engineering studies. For example, in the study carried out by
Indraganti et al. (2015), they used open-ended questionnaires to qualitatively describe the constraints, such as sociocultural aspects and attitudinal impediments, on the monitored window operation.
This section illustrated how engineering and social science research methods have been combined in existing window behaviour studies, highlighting the practical contributions of social science methods across different research stages. For example, semi-structured interviews at the early stages of a study can help identify key influencing factors. Based on these findings, structured surveys can be developed to capture subjective information, which can then feed into subsequent statistical analysis. Simultaneously, field measurements of physical conditions facilitate the integration of subjective and objective data, thereby enabling a more comprehensive understanding. Alternatively, if engineering-based analyses yield ambiguous or unexpected results, follow-up interviews or open-ended questionnaires can help uncover underlying social or cultural explanations. While these practices offer applicable methodological precedents, they are often implemented in a case-by-case or intuitive manner without a clear guiding framework. To better support interdisciplinary integration, it is necessary to reflect on the philosophical foundations of these approaches, particularly the ontological, epistemological, and methodological assumptions that underpin them. This reflection is taken up in the following section.
4 Discussion
4.1 Paradigmatic foundation and epistemological tension
4.1.1 Ontological assumptions: what is occupant behaviour?
In reviewing research on window-opening behaviour among office occupants, it is evident that the majority of existing studies (86%) fall within the field of engineering. These studies are typically grounded in a positivist ontological framework (
Alharahsheh and Pius, 2020), operating under the implicit assumption that the world is composed of observable, measurable phenomena governed by universal laws. Occupant behaviour, in this context, is conceptualised as a reactive response to environmental stimuli, such as temperature and humidity. This ontological perspective is particularly prevalent in data-driven modelling and predicting of occupant behaviour in buildings (
Haldi et al., 2017;
Nguyen et al., 2022).
Such approaches, however, often treat human behaviour as analogous to physical phenomena, presuming a deterministic causal structure (e.g., occupants adjust their environment upon experiencing discomfort)―without sufficiently interrogating the ontological assumptions embedded in these modelling approaches. As argued by
Rosenberg (2018), a fundamental distinction between the natural science and social science lies in their respective study objects: the former investigates entities that do not respond to the theorist's interpretations, whereas the latter concerns intentional agents whose actions are shaped by beliefs, norms, and social meanings. When behavioural studies adopt the naturalist ontology, it often overlooks the meaningfulness and normativity of human action. For instance, in existing studies of window use in shared offices, window opening has been reduced to a passive response to environmental conditions, rather than being understood as a socially situated practice-one shaped by hierarchical dynamics, spatial control, group negotiation, and anticipatory concerns about others' reactions (
Liu et al., 2021;
Naspi et al., 2018b).
Social science traditions, particularly those influenced by interpretive sociology and critical realism, underscore the meaning of people's behaviour (
Porpora, 2015). Within these perspectives, human behaviour is not merely a response to external stimuli, but is construed as a practice embedded in structures, norms, beliefs, and situated interactions. While a small number of studies using social science research methods have begun to acknowledge social variables and agency in their examination of occupant behaviour, these engagements remain largely superficial in ontological terms. In most cases, behaviour is still implicitly treated as a set of measurable attitudes or preferences (as discussed in Section 3.3.2), but not as intentional, situated actions shaped by social norms, institutional contexts, or power relations. As such, the underlying ontological assumption, that occupant behaviour can be meaningfully understood through its embeddedness in social structure and meaning, has yet to be substantively developed in the reviewed literature.
4.1.2 Epistemological commitments: how to understand behaviour?
Having explored the ontological assumptions embedded in different disciplinary perspectives, this section turns to their epistemological commitments, i.e., how each tradition conceptualises knowledge about occupant behaviour and what counts as a valid explanation. The distinction between these epistemological orientations has significant implications in studying window-opening behaviour.
In existing engineering-based studies (reviewed in this study), window-opening behaviour was commonly conceptualised as a passive response to physical environmental variables, such as indoor and outdoor temperatures, wind speeds, and humidity. When developing behavioural models for window usage, occupant behaviour was usually treated as the logical outcome of a set of initial conditions and empirical regularities. Therefore, their logic of explanation is profoundly informed by the deductive-nomological model inherited from the natural sciences (
Rosenberg and McIntyre, 2019). Within the positivist tradition, explanation and prediction are considered as the two sides of a coin, and only theories that yield successful predictions are deemed explanatory (
Rosenberg, 2018). Consequently, within engineering research, the more predictable a behaviour is, the more “valid” or “useful” the model is.
It is noteworthy that some studies, particularly those incorporating social science methods or theoretical frameworks, have begun to articulate an alternative logic of explanation. Rather than reducing behaviour to mechanistic responses to external stimuli, these studies seek to understand the actor's perspective and the social context in which decisions are made. For example,
Indraganti et al. (2015) explored the sociocultural and attitudinal constraints that may hinder adaptive behaviours. Similarly,
Bavaresco et al. (2021) have argued that future research must engage more seriously with subjective dimensions of occupant experience, including intention, attitudes, ease, negotiation, and expectations surrounding the shared control of the indoor environment. These studies resonate with the Verstehen tradition in social science, which prioritises understanding why actors behave in specific ways rather than merely when they will do so.
As argued by
Porpora (2015), explaining human behaviour necessitates a distinction between causes and reasons: whereas the former appeals to repeatable natural regularities, the latter foregrounds actors' beliefs, motives, and the social meanings they ascribe to their actions.
Rosenberg (2018) likewise emphasized that the goal of social science is not to predict outcomes but to interpret the structure of meaning that underpins behaviour.
4.1.3 What are the methodological approaches and research aims?
The divergent ontological and epistemological assumptions outlined above are reflected in the methodological strategies and research objectives pursued by engineering and social science studies. Methodologically, engineering approaches tend to search for stable relationships between occupant behaviour and relevant variables. Social science approaches, in contrast, are oriented towards understanding how actors express or suppress their agency within structural constraints. This divergence reflects more profound differences in ontological and epistemological commitments: while the former treats behaviour as observable and predictable physical responses, the latter views behaviour as socially situated practice, requiring interpretation within context.
Within the body of engineering literature reviewed in this study, occupant window behaviour is typically conceptualised as responses to environmental stimuli. The primary aim of these studies is to extract quantifiable behavioural patterns and construct predictive models. Most of these investigations follow a standardised process in which behaviour is modelled as an input-output mechanism triggered by predefined “drivers”, employing statistical and machine learning techniques, such as logistic regression or random forest, to enhance predictive performance. In this framework, agency is often simplified to a function of input variables, rather than recognised as the capacity of intentional, reflective actors to make judgments in context.
From a social scientific standpoint, the engineering-based approach overlooks the agentic and interpretive dimensions of behaviour. As argued by
Porpora (2015), “
we cannot explain or even identify individual action without appeal to actors'
intentions”. Therefore, the recognition and explanation of behaviour is not simply a matter of tracing causal sequences, but of interpreting the subjective meanings that actors attach to their actions. Traditions within social theory, such as phenomenology, interpretive sociology, and ethnomethodology, have long emphasized the agentic capacities of individuals as embedded within and shaped by social structures, including norms, hierarchies, and interactive contexts. Social science research tends to adopt methodological approaches aimed not at prediction, but at interpretation and contextual understanding. Rather than treating behaviour as a function of measurable inputs, this approach focuses on uncovering the meanings, intentions, and situational judgments that underlie human action. Typical methods include qualitative interviews, focus groups, and ethnographic observations, which allow researchers to explore the lived experience of occupants and the socio-cultural structures in which their behaviours are embedded. The research aims in such studies are generally aligned with understanding how behaviours are shaped by organisational dynamics, cultural norms, institutional constraints, or interpersonal negotiation, especially in multi-person settings. Although few reviewed studies fully adopt such approaches, they offer valuable yet underexplored pathways to better understand occupant behaviour (as discussed in Section 3.5.3).
4.1.4 Can these paradigms be integrated?
Studies on occupant behaviour in engineering and social science diverges not only in research aims and methods, but also in deeper ontological and epistemological commitments, in short, in their paradigmatic foundations. This divergence is not merely methodological; it reflects fundamentally different ways of construing the nature of human behaviour. As thus, uncritical attempts to combine these paradigms may generate theoretical inconsistencies. For occupant behaviour studies, failure to differentiate between the nature of causes and reasons may result in the misapplication of physicalist causal models to phenomena that require interpretive understanding. This form of paradigm misalignment may distort behavioural explanations and introduce biases into predictive models. It may also lead to misguided interventions, energy strategies that fail to align with the actual judgement logic and behavioural boundaries of occupants.
This does not imply that paradigmatic integration is impossible. On the contrary, given the complex, layered nature of human behaviour, the integration of divergent paradigms is both necessary and intellectually viable. As noted by
Porpora (2015), research began from some paradigmatic starting point. Indeed, certain paradigms may become dominant within fields. Yet researchers are not bound by these starting positions. Empirical evidence can either support a theory or expose its limitations. When theories are challenged by real-world findings, researchers must revise their conceptual frameworks. Through this ongoing process of empirical testing and refinement, meaningful dialogue between different paradigms can gradually emerge. Similarly,
Rosenberg (2018) also outlined a naturalist middle ground, in which explanation and prediction are not seen as inherently incompatible. From this perspective, researchers may both extract empirical regularities and acknowledge the purposive, meaningful nature of human behaviour.
Bhaskar (2010) has given a coherent framework for such integration in his interdisciplinary work,
Contexts of Interdisciplinarity:
Interdisciplinarity and Climate Change. It offers a systematic account of cross-paradigm collaboration, with three interrelated dimensions, namely, Meta-theoretical Unity, Theoretical Pluralism and Tolerance, and Methodological Specificity, which are collectively termed the “holy trinity” of interdisciplinarity. Based on this work, this study has developed a prototype framework for interdisciplinary research on occupant window behaviour, as shown in Fig. 8(a):
1) Meta-theoretical Unity: Clarifying the dual nature of behavioural phenomena
This review has shown that occupant behaviour is shaped by both physical variables and embeddedness in social interactions and normative structures. This duality demands a meta-theoretical position that recognises human behaviour involves both causal regularities and interpretive meaning structures.
2) Theoretical Pluralism and Tolerance: Accommodating engineering logics and social explanations
Different theoretical approaches can yield complementary insights at distinct levels or stages of inquiry. Engineering models offer a strong predictive capacity when capturing behavioural trends and frequencies, making them valid for forecasting, intervention, control, and system design. Social science frameworks, in turn, uncover the institutional and cultural logic shaping behaviours, for instance, how social norms inhibit adaptive action or how social hierarchies constrain access to environmental control. As such, we advocate a non-exclusive positioning of prediction and explanation as mutually enriching rather than competing objectives.
3) Methodological Specificity: A problem-driven mixed-method strategy
A core tenet of interdisciplinary research is that methodological choices should follow the nature of the research question, not disciplinary conventions. In occupant window behaviour studies, where both physical conditions and social dynamics are relevant, this principle calls for the careful coordination of engineering and social science methods.
Some reviewed studies have taken initial steps toward such integration. For instance,
Langevin et al. (2015a) used qualitative interviews prior to model development to uncover latent behavioural drivers, and
Bavaresco et al. (2020) embedded social variables, such as intention and perceived control, into statistical models. These practices demonstrate that mixed methods can be more than parallel tools; they can form a cohesive strategy to match complex research problems with appropriate layers of explanation.
To advance this integration, Fig. 8(b) proposes a structured explanation map that visualises how social science and engineering methods can be aligned throughout the research process. At the top left of the diagram, the process begins with problem–method matching, emphasizing that methodological choices should be driven by the nature of the research problem rather than disciplinary conventions. Specifically, the framework distinguishes between three types of research problems:
a) those involving observable variables, which are well-suited to engineering approaches (e.g., field monitoring and statistical modelling);
b) those involving latent variables, such as beliefs or intentions, which call for social science methods like indepth interviews or interpretive analysis;
c) those involving mixed variables requiring an interdisciplinary strategy.
Instead of combining methods in an ad hoc manner, the diagram encourages purposeful alignment between research stages (e.g., early-stage interviews, to variable selection, to statistical modelling, and finally to later-stage interpretation). For example, quantitative analysis might reveal behavioural regularities, while qualitative insights unpack their social motivations. The diagram also introduces “negotiation nodes”, where methodological or epistemological coordination is especially critical, such as incorporating latent social variables into SEM models or interpreting ambiguous findings through post-study interviews. This framework offers a more reflexive and systematic approach to mixed-methods research, ensuring that each method serves a clearly defined explanatory function and is selected in response to the specific demands of the research problem.
Moreover, social science theories are reflexive, that is, their interpretations may feed back into the social world and shape the very behaviours they aim to explain (
Rosenberg and McIntyre, 2019). In the context of occupant behaviour, for instance, making individuals aware of the social norms, hierarchies, or habitual routines underlying their actions may prompt shifts in how they engage with building systems. As such, incorporating these reflexive insights into engineering design or energy strategies can serve not only predictive functions but also educative and normative purposes, supporting behaviour change by altering occupants' self-understanding and perceived agency.
In sum, paradigm integration is not merely a philosophical possibility; it is a practical necessity for understanding complex behaviours and designing meaningful interventions. Future studies may advance this agenda by pursuing a research trajectory based on meta-theoretical negotiation, theoretical juxtaposition, and methodological integration, thereby producing more socially responsive and explanatory models of occupant behaviour within the built environment domain.
4.2 Empirical tensions and contradictions
Despite efforts to integrate subjective data across engineering and social science studies, empirical challenges persist. One common issue is recall bias, where participants misremember past actions (
Althubaiti, 2016;
Coughlin, 1990). For instance, in studies using self-reported window-use surveys, occupants often failed to accurately recall the frequency or timing of their behaviours (
Belafi et al., 2018). Similarly, social desirability bias (
Teh et al., 2023) may arise when participants respond in a way that aligns with socially accepted norms rather than their actual actions, such as overstating energy-saving habits in environmental surveys (
Bavaresco et al., 2020a). These biases highlight the limitations of relying solely on self-reported data and call for cross-validation with objective measurements.
In interview-based studies, the interviewer effect can also distort results (
O'Muircheartaigh and Campanelli, 1998;
West and Blom, 2017). Respondents may give more conforming or agreeable answers depending on the interviewer's perceived authority, tone, or even gender (
Davis et al., 2010). However, a casual or informal interview setting (e.g., a coffee room) may elicit more authentic responses than a formal office environment (
Swain and King, 2022).
Further complications arise in cross-cultural research, where semantic discrepancies can undermine scale-based measurements (
Froman and Owen, 2001). In Japan, terms like “cool” and “warm” were perceived positively, leading to misinterpretation of the ASHRAE thermal comfort scale, where those terms denote discomfort in English (
Damiati et al., 2016). Similar translation-related distortions were identified in Korea (
Kim et al., 2022). Methods such as the Double Translation Process (
McGorry, 2000) have been used to enhance semantic equivalence in such contexts (
Bavaresco et al., 2020b), but the underlying issue of cultural-linguistic interpretation remains a potential challenge for interdisciplinary studies relying on subjective data.
Together, these examples demonstrate that subjective information is shaped not only by memory or perception, but also by cultural context, linguistic framing, and social expectations. These tensions reinforce the need for research designs that are both methodologically reflexive and philosophically informed, particularly in studies seeking to integrate engineering precision with social insight.
5 Conclusion
This study systematically reviewed 106 studies investigating office window behaviour, with particular attention to the aims, methodologies and theoretical foundations employed in engineering and social science research. Given the urgency of reducing building-related carbon emissions, occupant-controlled natural ventilation is critical in enhancing energy efficiency, indoor air quality, and thermal comfort. A deeper understanding of how occupants interact with windows is therefore indispensable.
Guided by the PRISMA protocol, this comprehensive analysis revealed that, although the importance of interdisciplinary collaboration in window behaviour research has been widely recognised, current collaborations primarily occur at the methods level, with limited engagement in theoretical development and research frameworks. This review conducts an in-depth analysis of the aims, methods, theoretical foundations, and contributions of both engineering and social science research in the study of window behaviour, as well as their respective advantages, limitations, and potential challenges. The key findings are summarised as follows:
● Clarifying research aims and methods:
This review identified distinct research aims, and methods in engineering and social science research on window behaviour. Engineering studies typically focus on quantifying environmental responses and building predictive models, while social science research is more concerned with interpretive understanding of occupants' intentions, social dynamics, and behavioural reasoning. This study provides a structured comparison that highlights advantages and limitations within each method.
● Deepening analysis of paradigmatic and epistemological tensions:
Going beyond surface-level methodological comparisons, this study offers a detailed philosophical analysis of the ontological and epistemological foundations that underpin engineering and social science approaches. It shows that the two paradigms differ not only in research techniques but also in how they conceptualise occupant behaviour, what they consider valid knowledge, and how they construct explanations. By unpacking these tensions, the review clarifies why interdisciplinary collaboration often struggles and lays the groundwork for more coherent theoretical integration.
● Developing a philosophically grounded interdisciplinary framework:
To address the fragmentation observed in current research, the study proposes an interdisciplinary framework informed by meta-theoretical reasoning. This framework aligns causal modelling with social interpretation, offering guidance on how to combine methods according to the nature of the research problem. It introduces the concept of “problem–method matching” and maps out how engineering and social science methods can be coordinated across research stages. By doing so, it supports a more principled, reflexive, and epistemologically coherent approach to interdisciplinary research on occupant window behaviour.
This study highlighted the potential and importance of interdisciplinary research in the study of office occupants' window behaviour, particularly in multi-occupant office spaces and cross-cultural studies. Furthermore, it offers a novel theoretical contribution by proposing a cross-paradigm framework grounded. This framework responds directly to the fragmented nature of current interdisciplinary research, where methodological combinations often lack epistemological coherence. By integrating meta-theoretical foundations, diverse theoretical perspectives, and problem-driven methodological strategies, this review provides a structured approach for future studies to combine engineering precision with meaningful social interpretability. The proposed framework, may also serve as a transferable model for broader occupant behaviour research, promoting socially responsive, philosophically consistent, and practically actionable research in the built environment domain.
2095-2635/2025 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd.
PDF
(4558KB)
