1. INTRODUCTION
Human–building interactions (HBIs) are shaped by a complex interplay of multi-domain environmental conditions, personal characteristics and contextual factors (Hong et al. 2017). However, research to date often adopts a reductive approach, overlooking the interactions between these elements (Mylonas et al. 2024). Occupants’ perceptions and behaviours are influenced by individual environmental domains (i.e. thermal, visual, acoustic and air quality) and their combined effects (Schweiker et al. 2020). Consequently, the single-domain approach dominating HBI research (e.g. Zhang & Barrett 2012) limits the predictive accuracy of occupant behaviour models, as actual behaviours emerge from synergies and conflicts across domains (Lyu 2023). Recognising this limitation, a growing number of studies now examine cross-modal interactive effects between environmental domains (Deng et al. 2024; Sun et al. 2024) and their combined effects on overall occupant satisfaction (Tang et al. 2022; Yang & Lian 2025). A holistic understanding of multi-domain indoor environmental quality (IEQ) and the drivers behind HBIs has been identified as a key research priority in the International Energy Agency’s Energy in Buildings and Communities Programme (IEA EBC) Annex 79 (O’Brien et al. 2020). Chinazzo et al. (2022), within Annex 79, proposed a set of quality criteria for multi-domain indoor environmental studies, but the application of these criteria reveals that comprehensive investigations across all four domains remain rare. Most existing studies address only two domains (Inkarojrit 2008; Li et al. 2015) and lack methodological frameworks that integrate diverse data collection methods across relevant spatial and temporal scales.
In parallel, significant individual differences in HBIs arise from personal characteristics such as preferences, personality traits, habits and physiology. While growing attention has been paid to differences in comfort perception (Arakawa Martins et al. 2022), these factors are rarely incorporated into occupant behaviour models. In open-plan workplace environments, personal motivations are frequently intertwined with contextual factors (e.g. building and system design and group dynamics) in shaping the actual behaviours (Hong et al. 2020; Stazi et al. 2017). Social–psychological constructs (e.g. intentions, norms and perceived control) can predict energy-related behaviours beyond environmental cues (Weerasinghe et al. 2023), while traits such as conscientiousness and extraversion have been linked to variations in window operation (Hong et al. 2020). Similarly, habitual responses and energy-saving attitudes can influence thermal comfort behaviour independent of measured conditions (Xu et al. 2023). Despite these insights, much of the research still relies on subjective surveys, while direct monitoring of real-time behaviours remains limited. Post-occupancy evaluation frameworks such as BOSSA (Candido et al. 2016), CBE IEQ (Zagreus et al. 2004) and HOPE (Bluyssen et al. 2011) capture global user satisfaction, but overlook the dynamic nature of behaviour. Their reliance on periodic self-reports means that real-time actions, motivations and environmental conditions at the moment of decision-making are often missed.
Although few formalised frameworks exist, several studies have laid important groundwork. Gunay et al. (2016) proposed integrating occupant behaviour into building performance simulation tools through rule-based and stochastic models, but without real-time motivational input. Schweiker et al. (2020) and O’Brien et al. (2020) highlighted the importance of multi-domain comfort assessment, yet existing implementations often combine decoupled subjective surveys with physical measurements. Recent empirical studies (Kim & Park 2023; Deng et al. 2024) have collected rich behavioural datasets, but tend to focus on single actions or domains and rarely record occupants’ self-reported motivations at the time of action. Advanced sensing at the desk level (e.g. Arakawa Martins et al. 2022) also remains disconnected from motivational and contextual drivers.
Moving beyond these limitations requires an empirically grounded methodology that: (1) achieves high spatial and temporal resolution to capture individual-level HBIs; (2) integrates real-time feedback on behavioural motivations alongside monitored actions; and (3) combines these with long-term surveys to account for personal traits and contextual factors. To address these gaps, this study introduces a holistic, multi-domain field study methodology, abbreviated as eCOMBINE, to investigate HBIs in office buildings. The eCOMBINE framework adopts an integrated approach to study the relationships between occupant behaviour, multi-domain environmental comfort, combined IEQ factors and building energy (i.e. heating, ventilation and air-conditioning (HVAC)-related energy consumption). The feasibility and acceptability of such an approach from the occupant’s perspective are assessed through the field campaign, with the research potential illustrated through exemplary analyses.
2. METHODS: PRESENTATION OF THE eCOMBINE FRAMEWORK
The eCOMBINE framework, which stands for ‘Interaction between energy, COMfort, Behaviour and the INdoor Environment’, is a step-forward multidimensional data-collection framework with a dedicated focus on open-plan offices. It employs a mixed experimental approach (Figure 1) that integrates objective data (e.g. occupant behaviour, building energy use, and multi-domain environmental conditions at both office and workstation levels) with subjective occupant feedback on (dis)comfort and the motivations behind their actions. The eCOMBINE framework builds on existing approaches and unites objective environmental monitoring with subjective occupant feedback. The eCOMBINE framework can support a wide range of research topics (Figure 1), including, but not limited to, the following:
linking global environmental stimuli (thermal, indoor air quality (IAQ), visual, acoustic) to user perceptions, behaviours and preferences
exploring real-time motivations behind user control actions (Barthelmes et al. 2026)
studying how combined environmental factors are associated with behaviours such as window opening (WO)
investigating group dynamics and social norms in shared environments
quantifying individual exposure to discomfort and pollutants to assess health impacts (Gonzalez Serrano 2023; Khovalyg et al. 2023)
creating calibrated models to evaluate occupant behaviour and optimise IEQ and energy performance (Barthelmes et al. 2023) and
identifying the minimum sensor set and resolution needed to infer behaviour cost-effectively (e.g. for IAQ-related measurements in Gonzalez Serrano 2023).
Figure 1
Overview of the eCOMBINE framework.
Note: A mixed experimental approach is used for collecting objective and subjective data. Its research capacity depends on the combination of various data sources.
This paper presents the measurement methods and questionnaires used within the eCOMBINE framework, aiming to support their replication in similar research contexts. While survey-based subjective data collection is widely applicable, some objective measurements, such as energy metering, are specific to the characteristics of individual buildings. To assess the feasibility and occupant acceptance of the eCOMBINE framework, it was implemented in two open-plan offices. The insights and lessons learned from these case studies are shared in this paper.
2.1 OBJECTIVE MEASUREMENTS
Objective measurements encompass multi-domain environmental data (both indoor and outdoor), occupant behaviour tracking and building energy use. The following subsections provide details on each measurement type.
2.1.1 Indoor environmental measurements
For each environmental domain, key parameters, typically defined by standards with set thresholds (Khovalyg et al. 2020), should be measured, along with supplementary indicators that provide a deeper insight into the specific environment. All environmental measurements considered in the eCOMBINE framework are summarised in Appendix A in the supplemental data online. The rationale for selecting each parameter is explained below:
To assess the thermal sensation of people and their degree of thermal dissatisfaction, measurements should reflect the occupant’s immediate microclimate and potential sources of local discomfort. Operative temperature, which combines convective and radiative heat exchange, is a key metric in thermal comfort and requires data on air temperature, air speed and mean radiant temperature (MRT), often derived from globe temperature per ISO 7726 (1998). In practice, dry-bulb air temperature is often used as the main parameter for moderate indoor environments, where certain assumptions about MRT and air speed can be reasonably made. Humidity should also be measured to estimate thermal sensation. Thermal discomfort indicators, such as draft rate, vertical air temperature differences and radiant temperature asymmetry, with acceptable levels defined in ISO 7730 (2005), can provide further insight into the thermal environment.
Regarding IAQ, CO2 concentration is a key indicator of ventilation performance, with levels above outdoor concentrations categorised by ISO 17772 (2018). Elevated carbon dioxide (CO2) levels in occupied spaces indicate insufficient ventilation, which may lead to the build-up of other indoor pollutants. Monitoring additional air pollutants—such as particulate matter with aerodynamic diameters < 2.5 µm (PM2.5) and < 10 µm (PM10), total volatile organic compounds (TVOCs), carbon monoxide (CO), ozone (O3) and sulfur dioxide (SO2)—is recommended by guidelines (e.g. ASHRAE 2010; WHO 2021; WELL 2025), as these pollutants impact health, comfort and cognitive performance. PM2.5, in particular, is closely linked to long-term health impacts and is often used in burden-of-disease assessments (e.g. Logue et al. 2012).
Regarding the visual environment, temporal and spatial variations in lighting conditions within office environments should be monitored using illuminance measurements. The minimum recommended illuminance level for offices is 500 lux, as specified by ISO/CE 8995-1 (2025). Additional measurements that can provide a more refined understanding of the visual environment are: (1) high dynamic range images from a representative viewpoint (e.g. occupant perspective) over one sunny day in each season to estimate glare risks through discomfort glare evaluations according to EN 17037 (CEN 2018); and (2) spectral power density (at 5 nm resolution) to specify precisely the quality and composition of light.
For the acoustic environment, the key parameter is sound pressure level measured with A-weighting integrated over time, whose threshold values are specified in the standard ISO 3382-3 (2022). The sound pressure level at different octave bands can provide further insight into the acoustic environment to which people are exposed.
Indoor environmental measurements should capture both temporal and spatial variations across thermal, air quality, visual and acoustic conditions, but optimal sensor placement and sampling strategies differ across domains. Air quality and thermal sensors are typically positioned near occupants with unobstructed airflow, lighting sensors must align with the field of view, and acoustic sensors often require more remote placement to capture background noise without interference—while also raising potential privacy concerns. Sampling frequency similarly varies: long-term averages may suffice for acoustics, whereas air quality, lighting and thermal comfort often require high-frequency, occupant-proximal measurements. Standards such as ISO 7726 (1998), ISO 7730 (2005) and EN 16798-1 (CEN 2019) provide guidance for sensor placement within occupied zones, typically defining measurement heights and positioning relative to the occupied or breathing zone rather than prescribing fixed distances in metres. However, these guidelines are primarily intended for compliance or defined-condition assessments and are not directly transferable to continuous, occupant-proximal field monitoring. For long-term field studies, spatial deployment strategies therefore require context-specific adaptation. Regularly occupied areas, variable-use spaces (e.g. meeting rooms) and environments for vulnerable populations are often prioritised. Because IEQ domains have distinct requirements for positioning, frequency, cross-interferences and privacy, comprehensive assessment generally necessitates deploying multiple sensor types. In this study, several instruments were therefore combined to implement the eCOMBINE framework (see Appendix A in the supplemental data online). The choice of instruments was guided by a trade-off between cost, reliability and accuracy, with a focus on deploying primarily low- and medium-cost devices.
2.1.2 Outdoor, behavioural and energy data collection
Outdoor environmental measurements should be conducted on-site to draw more accurate conclusions about how ambient conditions affect occupant perceptions, behaviours and the overall performance of the building. They also help create accurate, site-specific weather files for energy simulations. Key outdoor thermal parameters such as outdoor air temperature and solar irradiance should be supplemented with outdoor air quality data (e.g. PM2.5 levels). In the eCOMBINE framework, a rooftop weather station was installed, with sensor selection and set-up (detailed in Appendix A in the supplemental data online), guided by practical requirements, particularly the need for autonomous operation.
Behavioural data include occupancy status, whether or not a person is at a desk and the interactions with the building, such as opening or closing windows, adjusting blinds, and switching lights on/off. While various technologies are available for occupancy tracking, more individualised data can be obtained using sensors placed at each desk. The eCOMBINE framework uses reed switch sensors for window monitoring, slat angle and movement sensors for blinds, and smart plugs to monitor lamp energy use and lighting behaviour. Within the scope of this framework, energy monitoring focuses primarily on HVAC-related energy use (heating, cooling and ventilation), as plug loads other than monitored lighting were not included.
Monitoring energy use by HVAC systems is crucial for closing the gap between predicted and actual building performance. The monitoring approach and instrumentation should be guided by the specific HVAC set-up and the capabilities of the building management system (BMS). If metering is in place, zone-level data can often be accessed via the BMS; otherwise, temporary equipment should be installed based on the system set-up. For zone-specific analysis, thermal energy use can be estimated with heat meters. Ventilation energy use can be assessed by measuring airflow and the temperature difference between outdoor and supply air.
2.2 SUBJECTIVE MEASUREMENTS
For collecting subjective feedback from building occupants, the eCOMBINE framework proposes a series of questionnaires: two types of long-term (LT) questionnaires and three types of point-in-time (PIT) questionnaires. Figure 2 shows the occurrence and timing of each survey over an exemplary 2-week study period (for a comprehensive overview of the surveys, see Appendix B in the supplemental data online; and for their full text, see Appendices C–F online). The questionnaires were designed based on the research objectives, established standards such as ISO 10551 (2019), and insights from prior studies (e.g. Arsenault et al. 2012; Chinazzo et al. 2019). Occupants’ subjective feedback, collected through PIT and ‘action’ surveys, was temporally synchronised with objective measurements of the multi-domain indoor environment and tracked behaviours by matching survey time stamps with sensor data using a 5-min interval resolution and corresponding window sensor IDs.
Figure 2
Surveying schedule and frequency for long- and short-term surveys within the eCOMBINE framework.
Note: LT-A = background information of global comfort; LT-B = seasonal perception of global comfort; PIT-A = perception of global discomfort in the workplace; PIT-B = motivations behind window and blinds control behaviour; and PIT-C = motivations behind light-switching behaviour.
2.2.1 Long-term (LT) surveys
The objective of LT survey A (LT-A) (see Appendix C in the supplemental data online) was to collect information on employees’ general personal data and information not sensitive to seasonal variations (e.g. personal characteristics, work routines and global personal preferences). The objective of LT survey B (LT-B) (see Appendix D online) was to collect long-term information and the environmental perception of the occupants over the past 2 weeks when monitoring occurred. This included the occupants’ general comfort preferences, perceptions, satisfaction, knowledge of controls and usual group dynamics in the office.
2.2.2 Point-in-time (PIT) surveys
The PIT ‘comfort’ survey (PIT-A) had the objective to investigate occupants’ perception of the global environment and their global comfort (thermal, IAQ, visual and acoustic) at a given moment (see Appendix E in the supplemental data online). A general prerequisite for the design of all PIT surveys was the minimisation of the number of questions and inputs (or ‘clicks’) needed to reduce survey fatigue. PIT-A was divided into two short sections: the ‘contextual info’ (e.g. clothing and activity level) and the ‘global comfort panel’ (including all four environmental dimensions) to gather all necessary information. The objective of the ‘action’ surveys (PIT-B and PIT-C) was to investigate self-reported motivations behind window, window blinds and light control actions each time an employee interacted with these controls (see Appendix F online). This procedure allowed the capture of the ‘real’ motivational multi-domain and contextual drivers behind the HBIs (e.g. if the measurements indicate that the indoor environment is too warm, the real motivation behind a window-opening action might still be related to other motivations, such as the desire to let in fresh air, or others). To provide a user-friendly and easy-to-use interface for PIT-B and PIT-C surveys to the users, the authors developed a tailored mobile application called OBdrive (Barthelmes et al. 2025). The OBdrive app enables one to survey real-time, impulse-driven motivations, reflecting multi-domain and contextual factors behind HBIs. The type of actions and selectable motivations on the application are shown in Appendix F in the supplemental data online.
3. METHODS: FRAMEWORK IMPLEMENTATION
3.1 CASE STUDY OFFICES
Two case study buildings located in Switzerland, featuring open-plan office layouts, were selected to implement the eCOMBINE framework. The details of the buildings are provided by Khovalyg et al. (2023); a comparative overview of the offices is presented in Table 1. The floor plans of the chosen open-plan offices, labelled as A and B, are shown in Figure 3. The selection criteria for suitable offices and occupants to implement the eCOMBINE framework were the following:
Office requirements: (1) open-plan offices with various facade orientations and limited control automation (e.g. operable windows); (2) at least 10 occupants per office to study group dynamics; and (3) access to technical building data (e.g. envelope details, energy use and BMS access).
Occupant requirements: (1) the majority willing to join training and give feedback during seasonal studies; and (2) representative of typical office workers, without known behavioural biases.
Operational requirements: (1) management approval for monitoring and occupant interaction during work hours; and (2) research team access to spaces and systems for equipment installation.
Table 1
Overview of the characteristics of two open-plan offices.
| OFFICE ID | SURFACE AREA (m2) | ORIENTATION | WALL-TO-WINDOW RATIO | U-VALUE OF WALLS (W/m2K) | HEATING/COOLING SYSTEM | VENTILATION SYSTEM | AUTOMATIC ELEMENTS | OPERABLE ELEMENTS | PARTICIPATING OCCUPANTS |
|---|---|---|---|---|---|---|---|---|---|
| Office A | 259 | North, east | 0.42 | < 0.18 | Hydraulic radiant ceiling system | Mechanical | Desk lights, ceiling lights and thermostats | Windows (tilt–turn), blinds | 13 (65% from total) |
| Office B | 242 | South-west, south-east | 0.55 | < 0.18 | Wall radiators (only heating) | Mechanical | 31 (88% from total) |
Figure 3
Floor plan of the case studies and location of the measurement devices.
Note: An example layout of the monitoring set-up during winter is shown.
The eCOMBINE framework was applied in five monitoring campaigns (Table 2), each lasting two continuous weeks and repeated in at least two seasons to assess seasonal effects. Campaign timing was guided by typical seasonal conditions and sufficient office occupancy, avoiding holidays. However, in 2020, COVID-19 restrictions became the main constraint, limiting summer data collection to office B only. The participation in the study was voluntary; a total of 44 office employees took part in the study.
Table 2
Overview and timeline of the monitoring campaigns.
| CAMPAIGN ID | CASE STUDY | SEASON | PERIOD |
|---|---|---|---|
| A-F | Office A | Fall | 28 October–9 November 2019 |
| A-W | Office A | Winter | 27 January–7 February 2020 |
| B-F | Office B | Fall | 18–29 November 2019 |
| B-W | Office B | Winter | 17–28 February 2020 |
| B-S | Office B | Summer | 17–28 August 2020 |
Figure 4 shows the placement of all sensors. It was practically challenging to place instruments in a preferred position described in Section 2.1.1, and they were bundled and mounted on a stand placed on the desk, resulting in measurements taken closer to the occupant’s head level. To capture the vertical air temperature difference, measured between ankle level (0.1 m) and head level (1.2 m), pairs of sensors were installed whenever the desk configuration allowed one sensor to be mounted on a desk leg (Figure 4). Wireless RoomZ under-desk sensors were used to monitor occupancy, Reed KumoSensors for window activity, KumoSensor Tag Pro devices for tracking blinds, and myStrom wireless smart plugs to capture light switch behaviour. HVAC energy use in the case study offices was estimated indirectly, as dedicated meters were unavailable and zone-specific consumption could not be isolated. The thermal energy supplied by the conditioning system was measured using heat flux sensors (HFP01 from Hukseflux) placed on radiant ceiling panels (office A) and radiators (office B). For ventilation, the thermal energy used for preconditioning was estimated by measuring the volumetric airflow rate and the temperature difference between the outdoor air and the supply air entering the office space, using a TESTO ventilation hood. The mobile application OBdrive (action surveying) was installed on dedicated (not personal) mobile phones and placed close to the controls (windows, window blinds and lights) (Figure 4) using double-sided adhesive patches to facilitate easier interaction with the app during data collection. The remaining surveys were implemented on the LimeSurvey platform. The LT surveys were emailed to the employees before the start of the campaign for LT-A and after or on the last day of each 2-week monitoring period. A PIT-A survey was sent via an automated email script to participants twice during workdays (at 10.00 and 15.00 hours).
Figure 4
Overview of the multi-domain experimental set-up for the objective data collection.
Note: B1 = occupancy detection; B2 = window reed sensors; B3 = angle sensors; B4 = wireless smart plugs; red dot on the radiator = heat flux sensor for recording heating energy; 1–5 = sensors for indoor thermal environment; 8–9 = sensors for indoor air quality; 11–12 = sensors for visual environment; and 17 = a sensor for acoustic environment. For details about the equipment labelled 1–17, see Appendix A in the supplemental data online.
3.2 DATA ANALYSIS
To demonstrate the applicability and effectiveness of the eCOMBINE framework, a preliminary analysis of the case study data was conducted. This included: (1) descriptive statistics summarising participants’ survey experiences and response rates across seasons; (2) a temporal comparison of occupant control behaviours (e.g. WO) during pre-survey, survey and post-survey periods using tracked sensor data; and (3) logistic regression models assessing the association between subjective environmental preferences (e.g. feeling too warm or too cold) and concurrently measured environmental parameters (e.g. operative temperature, CO2 concentration, illuminance and noise levels). These analyses aimed to validate the consistency and behavioural sensitivity of the framework, and to highlight the added value of integrating subjective feedback with objective environmental data.
4. RESULTS
Key findings related to the data collection strategy are now described, particularly participant experience and the acceptability of the instrumentation. Objective data analyses on energy performance and the thermal environment and IAQ are reported by Barthelmes et al. (2023) and Khovalyg et al. (2023), respectively, with a multi-domain IEQ analysis to follow in future work. Participant demographics, work routines, control use and environmental preferences are detailed in Appendix G in the supplemental data online.
4.1 EFFICIENCY OF THE SUBJECTIVE DATA COLLECTION
The success of the mixed-method approach depends not only on effective objective data collection but also on active occupant participation and their willingness to provide feedback. Results on the response rates for the LT (seasonal) surveys LT-A and LT-B and the PIT survey PIT-A are summarised in Figure 5a.
Figure 5
Surveying of occupants: (a) response rates to surveys during different campaigns (adjusted response rate – response rate related to the seasonal tracked peak occupancy); (b) frequency of self-reporting interactions (answers to the question ‘Over the last two weeks, how often did you report your interactions on the mobile phones installed close to window and blinds?’); and (c) reasons for not taking action even when feeling uncomfortable (answers to the question ‘Over the last two weeks, it happened that I felt uncomfortable, but I did not interact with controls, because …’).
For the LT surveys, a decline in response rates and durations could be observed over the seasons. The more pronounced decrease in response rates (and the missing campaign in office A during the summer) is also explained by the context of the COVID-19 pandemic, as many workers were still working remotely. Indeed, in office B, the peak occupancy measured by desk occupancy sensors was 79% in fall, 81% in winter and 48% in summer, while in office A, peak occupancy was 77% and 92% in fall and winter, respectively. The seasonal response rates have therefore also been normalised according to peak occupancy rates. With regard to the LT surveys, a slight drop in adjusted response rates can be observed along the seasons; in office A, there was a drop by 23% from fall to winter, while in office B, there was a drop by 6% from fall to winter and by 9% from winter to summer.
With regard to the daily comfort surveys (PIT-A), there were 22% fewer PIT reports in office A and 9% fewer PIT reports from fall to summer in office B. Comfort survey (PIT-A) response rates declined over time, likely due to fixed survey times (10.00 and 15.00 hours) that conflicted with participants’ work schedules, particularly in office A during winter. Overall, response rate variations were smaller in office B than in office A, with 73% of employees working the full week (see Appendix G in the supplemental data online).
Considering all campaigns, 70% of participants who responded to the seasonal survey indicated that they had always or almost always reported their actions in the mobile application OBdrive (Figure 5b). Interestingly, over 25% of participants reported that they had not interacted with controls because they were too focused on their work (Figure 5c), and over 10% indicated that they did not interact with controls because they did not want to make other co-workers feel uncomfortable. Another 10% indicated that they lacked the time to act. The types of ‘inactions’ due to personal cognitive, social and practical reasons can only be investigated through subjective responses. Over 5% of respondents indicated that they prefer to restore the comfort with other actions (e.g. put more clothes on/off).
The effectiveness of the surveys was evaluated using the completion times (see Appendix H in the supplemental data online) and user feedback. The LT surveys typically took less than 10 min to complete, while most daily comfort surveys were completed in about 1 min. Some longer times (e.g. 2 min for PIT-A in office B during fall) were due to outliers where respondents left the survey open before submitting. Maintaining short survey durations (under 10 min for LT and around 1 min for daily surveys) was essential to achieving high response rates, as discussed above.
Figure 6 summarises feedback from the post-campaign survey. Over 80% of occupants were comfortable receiving comfort surveys twice daily, and more than 70% found it easy to report actions using the mobile phones placed near windows and blinds. Most participants felt that the response options on the phones accurately reflected their reasons for taking actions. However, three respondents noted missing options. Upon investigating the possible reasons through post-interviews, it was found that a ‘getting back to normal/getting back to a closed environment’ option for window-closing would have been an additional wished-for answer option. Occupants reported WO more consistently than window closings (WC) in all campaigns (A-F: 32 WO versus 25 WC; A-W: 12 WO versus 9 WC; B-F: 22 WO versus 19 WC; B-W: 28 WO versus 18 WC; and B-S: 15 WO versus 10 WC). The less frequent reporting of WC may be due to their association with time-related factors, such as participants’ departure after work, which could reduce the motivation or consistency of providing subjective feedback, or because the windows were closed by non-participants.
Figure 6
Post-campaign survey results regarding comfort (PIT-A) and action surveys (PIT-B).
Note: Numbers refer to the completed responses from both offices.
4.2 EFFECTIVENESS OF THE MONITORING IMPLEMENTATION
Ideally, employees should maintain their usual work habits during the monitoring period. However, the presence of researchers and equipment may influence behaviour, a phenomenon known as the Hawthorne effect (Adair 1984). To reduce this, sensors were installed outside working hours. It remained uncertain whether visible, interactive interfaces influence occupant behaviour, possibly reducing interactions due to self-reporting fatigue, or increasing them out of curiosity. In one campaign (in B-W), sensor data showed a slight increase in WO compared with before and after the study (Figure 7), suggesting minimal behavioural change, as no Hawthorne effect was reported by Barthelmes et al. (2021a, 2021b). At the end of the field study, 90% of respondents said their use of windows or blinds remained unchanged, 7% felt influenced by the study, and 3% were unsure or could not recall.
Figure 7
Logged window-opening actions from wireless window state loggers before (n = 22), during (n = 28) and after (n = 18) in office B’s winter measurement campaign (B-W).
To assess whether the sensors disturbed occupants, it was essential to understand how participants perceived the campaign. Figure 8 shows that 81% were not bothered by the installed equipment, while six participants, three of whom had sensor stands on their desks, reported some discomfort. Some expressed concerns about having sensors near them (e.g. fearing that sensors might emit signals that could negatively affect their health) or had privacy concerns about reporting personal or monitored data to their employers.
Figure 8
Post-campaign survey results regarding the perceived intrusiveness of the eCOMBINE experimental set-up.
Note: Responses from both offices A and B.
4.3 COMPLEMENTARITY OF SUBJECTIVE (SELF-REPORTED) AND OBJECTIVE (MEASURED) DATA
The comparison between tracked and self-reported occupant actions for WO/WC revealed discrepancies, indicating that subjective inputs alone are insufficient for office-level analysis. Participants more consistently reported WO (n = 109) than closings (n = 81), and non-participants did not report actions at all. By combining both sensors and self-reports, observations were validated using multiple methods.
Sensor reliability, tailored to each case study, was also critical. Figure 9 shows substantial variation in the proportion of self-reported window actions across days and sessions. While some sessions (e.g. B-F and B-W) show relatively consistent reporting, others (e.g. A-W) display sparse and inconsistent self-reports. This is supported by Table 3, where recall values, indicating how many tracked actions were reported, range from 0.25 to 0.61. Precision is generally moderate to high (0.56–0.72), suggesting that when participants did report, the reports were usually valid. Overall, the data highlight underreporting across sessions, with noticeable differences in reporting behaviour between sites.
Table 3
Confusion matrix and key performance metrics assessing the alignment between self-reported and tracked window opening actions.
| CAMPAIGN ID | FN | FP | TN | TP | PRECISION | RECALL | ACCURACY |
|---|---|---|---|---|---|---|---|
| A-F | 30 | 13 | 33,911 | 30 | 0.70 | 0.50 | 0.99 |
| A-W | 24 | 5 | 32,507 | 8 | 0.62 | 0.25 | 0.99 |
| B-F | 11 | 8 | 22,565 | 16 | 0.67 | 0.59 | 0.99 |
| B-W | 13 | 5 | 22,607 | 20 | 0.56 | 0.61 | 0.99 |
| B-S | 29 | 16 | 22,609 | 13 | 0.72 | 0.31 | 0.99 |
[i] Note: True positives (TP) indicate correctly reported tracked actions; false negatives (FN) are tracked actions with no corresponding report; false positives (FP) are reported actions without a matching tracked event; and true negatives (TN) represent instances where neither a tracked nor a reported action occurred. Precision reflects the accuracy of self-reports, recall measures the completeness of reporting and accuracy captures the overall correctness of classification.
Figure 9
Window-opening actions: percentage of reported (by employees) window-opening actions in the total tracked (with window reed sensors) actions during the different monitoring campaigns.
Note: For IDs, see Appendix A in the supplemental data online.
5. DISCUSSION
5.1 RESEARCH POTENTIAL OF THE ECOMBINE FRAMEWORK
The eCOMBINE framework was fairly received by occupants in the case-study buildings, as presented in the results, and facilitated exploration of multiple research avenues. First, single-domain analyses can be examined in the context of multi-domain environmental parameters, e.g. predicting thermal preferences from different environmental variables (see Figure S1 in the supplemental data online). Furthermore, it enabled correlating environmental stimuli with user perceptions and behaviours, providing insights into the combined effects of different environmental dimensions on individual and multi-domain responses (see Figure S2 online). Additionally, the OBdrive mobile application allowed for the investigation of real-time motivations behind user control actions (see Figure S3 online), revealing how different motivation types (e.g. thermal, air quality, visual, noise and context) interact with user perceptions (see Figure S4 online).
Compared with other established approaches such as BOSSA (Candido et al. 2016), CBE IEQ (Zagreus et al. 2004) and HOPE (Bluyssen et al. 2011), eCOMBINE extends these methods by linking occupant motivations directly to sensor-tracked actions and incorporating fine-grained, multi-domain environmental measurements at both individual and workspace levels. It serves as a replicable framework for advancing both explanatory and predictive models of occupant behaviour.
Few existing frameworks from IEA EBC Annexes 69 and 79 (O’Brien et al. 2020; Schweiker et al. 2020) and occupant-behaviour modelling tools (Stazi et al. 2017; Gunay et al. 2016) provide valuable insights, but typically overlook the real-time self-reported reasons for control actions. In contrast, eCOMBINE provides a structured way to collect multi-domain and motivational data, which may support more detailed analyses of occupant behaviour.
5.2 LIMITATIONS
Although comprehensive and continuous measurements across spaces and time, such as in the eCOMBINE framework, are often desirable, they may not always be economically viable or necessary. Initial installation costs represent only part of the picture, as long-term expenses related to maintenance, data management and system oversight can be substantial. Moreover, large volumes of data pose challenges regarding storage, processing and compliance with data protection regulations, particularly when linked to personal or behavioural information.
The benefits of more granular or frequent data collection should therefore be critically assessed relative to the effort and resources required. In many cases, a targeted and well-designed monitoring set-up that produces actionable insights for building operation and user experience can be more valuable than an abundance of underutilised data. Striking this balance is essential for scalable, sustainable and impactful IEQ assessments.
The eCOMBINE framework is observational. The analyses estimate associations rather than causal effects. Causal inference in HBI research requires exogenous variation and rigorous control of confounding (Pearl 2009). Future work could employ strategies such as fixed-effects models, event studies, experiments, structural models or mediation analyses.
6. CONCLUSIONS
eCOMBINE enables extensive research into human–building interactions and supports a multidimensional approach to post-occupancy evaluation. The framework was generally well received by participants over a 2-week period across different seasons. A few occupants raised concerns about privacy (e.g. employer access to personal data) and health (e.g. under-desk sensors), highlighting the importance of clear information sessions explaining data management and potential risks. Most participants found the surveys clear and the answer options satisfactory, although some requested more nuanced motivation choices (e.g. restoring the usual environment as a reason for window closings, rather than simply feeling cold).
Testing the eCOMBINE framework in two case study offices also revealed that relying solely on self-reported data proved insufficient for office-level analysis, as not all occupants participated, and window openings were more frequently reported than closings. Post-campaign evaluations showed no notable behavioural shifts, suggesting minimal Hawthorne effect, though this should be monitored in future studies.
While eCOMBINE was developed with open-plan offices in mind, its core principles and tools, particularly its integration of objective and subjective data, hold promise for adaptation to other indoor environments such as classrooms, private offices or activity-based workplaces. Exploring its applicability in different contexts will help demonstrate its versatility and broaden its impact. Although the framework involves intensive deployment, this can be one possible step toward developing streamlined and scalable monitoring strategies that simultaneously study human–building interactions, occupant behaviour, multi-domain comfort and energy use in buildings.
ACKNOWLEDGEMENTS
The authors thank all the participants of the field study for their time and engagement. They also gratefully acknowledge the office supervisors for facilitating access to the offices and supporting coordination with the employees during the monitoring campaigns.
AUTHOR CONTRIBUTIONS
V.M.B.: methodology, software, formal analysis, investigation, data curation, writing—original draft, visualisation. C.M.: formal analysis, investigation, writing—review and editing. V.G.S.: formal analysis, investigation, writing—review and editing. K.L.: formal analysis, writing—original draft. J.W.: methodology, writing—review and editing, supervision. M.A.: conceptualisation, methodology, resources, writing—review and editing, supervision, funding acquisition. D.L.: conceptualisation, methodology, resources, writing—review and editing, supervision, funding acquisition. D.K.: conceptualisation, methodology, methodology, resources, writing—original draft, writing—review and editing, supervision, project administration, funding acquisition.
COMPETING INTERESTS
The authors have no competing interests to declare. M.A. was a member of the journal’s editorial board in 2025, but had no role in any editorial decisions involving this manuscript.
DATA ACCESSIBILITY
The measurement instruments and survey materials are provided in the supplemental data online. The OBdrive app is freely available to the research community.
ETHICAL APPROVAL
The study was approved by the École polytechnique fédérale de Lausanne (EPFL) human research ethics committee (study reference number HREC 036-219).
SUPPLEMENTAL DATA
The supplemental data for this article can be accessed at: https://doi.org/10.5334/bc.648.s1