Does thermal control improve visual satisfaction? Interactions between occupants’ self- perceived control, visual, thermal, and overall satisfaction

Occupants’ satisfaction had been researched independently related to thermal and visual stimuli for many decades showing among others the influence of self- perceived control. Few studies revealed interactions between thermal and visual stimuli af-fecting occupant satisfaction. In addition, studies including interactions between thermal and visual stimuli are lacking different control scenarios. This study focused on the effects of thermal and visual factors, their interaction, seasonal influences, and the degree of self- perceived control on overall, thermal, and visual satisfaction. A repeated- measures laboratory study with 61 participants running over two years and a total of 986 participant sessions was conducted. Mixed model analyses with overall satisfaction as outcome variable revealed that thermal satisfaction and visual satisfaction are the most important predictors for overall satisfaction with the indoor environment. Self- perceived thermal control participant (10.3%- 26.6%). Condition explained only 0.003% 1.4% of variance across all models. Different values were found in the moderator models (M.ov.sat.mod1, M.ov.sat.mod2): Participant accounted for 5.8– 16.2% of variance (residual variance: 78.5% to 84.9%), and up to 5.3% of variance was accounted by condition.


| INTRODUC TI ON
Human beings are continuously exposed to multiple indoor environmental exposures from different domains at the same time. These domains include thermal and visual stimuli leading to a perception of thermal or visual comfort. Thereby, thermal comfort is defined as "that condition of mind which expresses satisfaction with the thermal environment". 1 Visual comfort means "a subjective condition of visual well-being induced by the visual environment". 2 The overall evaluation of perceptions from different domains leads to a level of overall satisfaction with indoor environmental conditions.

| From single to multidomain studies
Occupants' satisfaction had been researched independently related to thermal and visual stimuli for many decades. [3][4][5][6] In contrast, multidomain studies, for example, considering thermal and visual aspects and their interaction, are scarce. 7,8 Following the definition by Torresin et al, these multidomain approaches can be distinguished into crossed (main) effects and combined effects. Combined effects are defined as the effects of two or more distinct domains, for example, thermal and visual, on a third domain, for example, overall satisfaction. In contrast, crossed effects are analyzing a main effect from one domain, for example, thermal stimuli, on another domain, for example, visual perception.
A recent review on multidomain studies came to the conclusion that results from multidomain studies are not conclusive and partly contradictionary. Related to thermal and visual conditions, inconsistent results related to the existence or direction of an interaction exist and depend on lighting conditions (illuminance level, intensity, and spectrum) and outcome measure (thermal sensation vs. thermal comfort). 8 Relevant for the study presented here is the current state of knowledge related to combined effects on overall satisfaction, crossed effects on thermal and visual satisfaction, and the influences of perceived control and season.

| Crossed effects on thermal and visual satisfaction
With respect to crossed effects, Chinazzo et al 13 investigated the effect of daylight transmitted through three colored glazing types (blue, orange, and neutral) on thermal responses and overall comfort, at three temperature levels (19°C, 22°C, and 26°C). Statistical analysis revealed a color-induced thermal estimation, independent of temperature levels: With a blue glazing, people felt colder and less comfortable than with a neutral one. With an orange glazing, people felt warmer and more comfortable than with a blue one.
The same authors also investigated the effect of daylight illuminance levels (~130 lux, ~600 lux, ~1400 lux) on thermal responses under three temperature levels (19°C, 23°C, and 27°C) and found a cross-modal effect of daylight on thermal responses. 14 According to Huang et al, 11 occupants judged their environment as thermally acceptable even when they are dissatisfied with lighting levels. Nicol and Humphreys 15 found that lighting use decreases with increasing indoor and/or outdoor temperatures. Lighting exposure can alter thermal comfort or thermal sensation: Some studies found an effect of light intensity on thermal sensation but not on thermal comfort (eg, 16,17 ), while others found the opposite direction. 18 Despite crossed and main effects of individual domains, the level of perceived control was shown to influence satisfaction with temperature (eg, 19,20 ). According to Brager et al, 21 study participants with different degrees of control-even when they experienced the same thermal environment, activity, and clothing levels-had significantly different thermal responses, though their field study design does not permit a distinction between individual differences and differences in perceived level of control. Kwon  Newsham 24 found that manual control of blinds and electric lighting for example, thermal control options affects not only thermal but also overall satisfaction with indoor environmental conditions.
• To account for the intra-and interindividual differences in self-perceived thermal control, future buildings should put users into a position in which they have as much control as possible over thermal and visual conditions to increase satisfaction, for example, to install easy-to-use thermal controls to improve thermal satisfaction.
• Further studies and careful methodological considerations are necessary before improvements in the design of the indoor environment can be suggested, for example, dynamic lighting adjusted to thermal conditions or vice versa to contribute to energy savings.
can lead to better thermal comfort. In summary, experiencing and/ or having control over room conditions improves overall, thermal, or visual satisfactions and overall satisfaction. These interactions are found in some studies 19,[25][26][27] but not in others. [28][29][30] At the same time, recent reviews showed that studies analyzing interactions between thermal and visual influences on levels of satisfaction, which are dominantly performed within laboratory environments, rarely permit participants to control their thermal and/or visual stimuli. 7,8 Torresin et al identified solely the study by Pellerin and Candas,31 which permitted participants control over stimuli in experimental studies dealing with multimodal interaction.

| Seasonal influences
In addition, studies on the interaction between thermal and visual comfort mainly focused on a limited number of seasons (summer preferred), or the study participants were exposed to the experimental conditions for different time periods. 32 At the same time, the authors expect seasonal influences on the interaction between thermal and visual stimuli due to differences in their appraisal. For example, warm conditions appeared more preferable in winter compared with summer 33 and there is a preference for higher illuminance levels and higher glare acceptance in spring compared with autumn and winter. 34

| Research objectives and approach
In order to overcome these identified gaps in the literature, the overall objective of this study was an increased understanding of interactions between visual and thermal stimuli and their effect on thermal, visual, and overall satisfaction with room conditions under different control scenarios and different seasonal influences.
In detail, the specific objective of our study was testing the conceptual model shown in Figure 1 representing the hypothesized relationship between thermal and visual stimuli and thermal satisfaction, visual satisfaction, and overall satisfaction with season and self-perceived control as moderators. According to our knowledge, this is the first study looking at the effect of control on interactions between thermal and visual stimuli in semi-standardized laboratory experiments in all of the four seasons. On this basis, we conducted an experimental repeated-measures study systematically varying thermal and visual stimuli together with the level of control among four seasons.

| Experimental facility
This study was conducted in the LOBSTER facility (http://lobst erfbta.de, 35,36 ), a free-standing experimental building with two fully equipped offices and two workplaces each. The room air and surface temperatures can be controlled individually. conductance level) were also measured but not analyzed here.

| Experimental protocol
The basis of the experimental protocol was a repeated-measures design for which participants were surveyed up to four seasons and six different thermal and visual conditions each.
In the years 2016-2018, two distinct age-groups were invited: young (aged 18 to 32 years) and older (aged 50 to 70 years). Two distinct age-groups have been chosen in order to include not only young and healthy university students and thereby increase the generalizability of results. Including an older age-group is meaningful, because existing literature suggests that age affects thermal 4,5 and visual 37 perception.
The six conditions were characterized by either thermal or visual conditions being fixed and the other conditions being controllable by the participants as presented in Table 1. In the Tx conditions, the temperature range was fixed, while visual conditions controllable by participants. In the Vx conditions, participants were able to adjust the temperature set point according to their preferences, while illuminance levels were fixed. It should be noted that in none of the conditions, control over visual/thermal conditions was completely removed: Participants were still able to change their clothing, tilt the window, or change their head/working position in relation to the façade. Therefore, the perceived level of control was assessed and included in statistical analysis rather than the control condition (see below). In addition, in order to reduce repetitions for participants and due to previous results showing a large difference to conditions without any control opportunity, 38 no conditions with both fixed temperature and illuminance levels were introduced.
The repeated-measures design was chosen to reduce the variance due to interpersonal differences. Participants were invited F I G U R E 1 Conceptual model underlying the analysis of this study to the LOBSTER facility for four days in each of the four seasons ( Figure 2). Participants performed two sessions per day, leading to eight sessions per season. In each session, one of the six above described conditions was performed. The order of the six conditions was randomly assigned to each participant in every season. Therefore, a condition was either conducted in the morning or afternoon. The two additional sessions created by the difference from the six conditions and eight sessions were added in order to counterbalance daytime effects by repeating two of the six conditions at different times of the day. Therefore, each participant conducted per season one of the three T conditions and one of the three L conditions for a full day. A complete participation would lead to 16 days x 2 sessions a day = 32 measuring points in 24 different conditions (six experimental conditions multiplied by four seasons).
On the first day, the participants received instructions regarding the schedule, room characteristics, and control opportunities (light, thermostat, external blinds; depending on test condition), and written informed consent was obtained. Then, sensors for skin conductance level, skin temperature, and heart rate were applied. Note that analysis of physiological data is not included here. After entering one of the experimental office rooms, the participants had to fill in the first questionnaire. Questionnaires started automatically at different time points in the morning, before, and after lunch break, and in the afternoon (see 2.3 for further details). Between answering the questionnaires, the participants were allowed to work on their own projects, research on the Internet, or read a book. The participants

| Data collection
Date collection included physical parameters, questionnaire items, and physiological data as described in the following.

| Physical data
The list of sensors and their accuracy is presented in Table 2, their position in Figure 3. With respect to the visual parameters, the horizontal illuminance level was measured at every workplace 1.90 m TA B L E 1 Characteristics of the six distinct study conditions. Thermal and visual experimental stimuli were either predetermined and not modifiable by participants (eg, 20°C in condition T1) or subject to participants' preference (eg, thermal stimuli in condition L1)

| Questionnaires
The participants had to respond to four different questionnaires (start and end, background, and conclusion) on the computer.  In the corresponding literature regarding thermal and visual satisfaction, these ratings are usually assessed by different scales (eg, a 4-point scale in thermal, but a 5-point scale in visual comfort research). In order to harmonize ratings, thermal, visual, and overall satisfaction was assessed by visual analogue scales (VASs) ( Table 3).

TA B L E 2 List of sensors and their accuracy
The end questionnaire had to be filled right before lunch break and before leaving the LOBSTER in the late afternoon. This questionnaire consists of the same questions like the start questionnaire and additionally asked about activities in the past three hours in the office room.
The background questionnaire consists of items related to influencing factors on comfort sensation such as body height, weight, sex, age, clothing degree, quality of sleep, thermo-specific selfefficacy, and thermal and visual preferences.
The concluding questionnaire was collected from the participants at the end of last day of their participation in each season.
This questionnaire includes the same items as the end questionnaire together with three scales of the NEO-FFI 39 : extraversion, neuroticism, and openness.
During the day, the participants were asked to push one of several buttons on the computer when they drink something or when they change something on their clothes. The corresponding information was not included in the analysis.
Age-group was assessed beforehand, defining "young" as people aged 18 to 32 years and "older" aged 50 to 70 years. Sex was assessed when participants started the questionnaire.

| Participants
In total, N = 61 participants took part in the study of which 25 (   lattice 46 ) gave a review of the relationships between analysis variables and study conditions per participant. Based on this, different intercepts and slopes for participants and condition were assumed (an exemplary xy plot is presented in the supporting information).

| Analysis methods
Season was operationalized in the model as running mean outdoor temperature (T rm ). The running mean outdoor temperature was calculated using measured hourly outdoor temperatures of the seven days prior to the experimental day and applying the equation given in EN 15251. As such, seasonal influences were covered by variations in observed outdoor temperatures.
Three independent analyses were conducted with the dependent variables overall satisfaction (ov.sat), thermal satisfaction Fixed effects were all independent variables including season (operationalized via T rm ), age-group, and sex. The ranges of the dependent and independent variables are different scales, for example, 100 for satisfaction votes (−50 to +50) and several thousand for illuminance. These differences in ranges require normalization of all variables in order to use them in a single model and apply them properly. Therefore, all variables were normalized using function scale() in R, which means that each value was first divided by the standard deviation of the variable in question and then the variable was centered around 0 by subtracting the mean. Outliers were excluded based on Cook's distance with a cutoff value of d < 4 × mean(d). 53 Only models without outliers are reported.
Starting with the null model, the analysis took the following steps: 1. Models with only physical variables included, with and without control as covariate.
2. Models with only questionnaire items regarding thermal and visual satisfaction, with and without control as covariate.
3. Models combining both physical factors and questionnaire items referring to thermal and visual satisfaction, with and without control as covariate.
4. Moderator models, considering PMV and control as moderator.

| RE SULTS
Results are grouped according to the conditions during the experiments, the temperature and light preferences of participants, and the mixed model results for overall, thermal, and visual satisfaction.

| Conditions during the experiments
In total, 986 participant sessions were conducted.   70 Basically, a "+" sign denotes that two variables, but not their interaction, are considered; a "*" sign will include two or more variables and their interactions. The notation (X|Y) determines whether the random factor is modeled for intercepts only (1|random factor), slope only (0+fixed factor|random factor), or both (1+fixed factor|random factor).

| Temperature and light preferences of participants
The questionnaire ratings for overall, thermal, and visual satisfaction along with self-perceived thermal control are summarized in Table 7 and Figure 5. Disregarding individual conditions, as observable in Table 7, mean ratings for visual satisfaction were on average the highest followed by overall satisfaction. At the same time, overall satisfaction had the least variance followed Ratings for overall and thermal satisfaction were higher in L conditions and T2 compared with conditions T1 and T3.
Ratings for visual satisfaction were higher in T conditions than in L conditions, that is, visual satisfaction was higher when participants were able to control visual conditions.

| Crossed and combined effects on overall, thermal, and visual satisfaction
In the following, for each independent variable analyzed, that is, overall, thermal, and visual satisfaction, first, the analyzed models' performances are presented followed by a more detailed description of one of the tested model, selected based on goodness-of-fit parameters. Based on the moderator model (M.ov.sat.mod1), thermal satisfaction and visual satisfaction were the most important predictors for overall satisfaction (see Table 8). Thermal satisfaction had more influence (0.57) on overall satisfaction than visual satisfaction (0.34). An increase in control corresponds to a slight increase in overall satisfaction (0.09). T rm had a tendency to influence on overall satisfaction vote (0.04), but the value was lower than for perceived control. There was a significant moderation effect of control on the relationship between thermal and overall satisfaction but not for visual and overall satisfaction. The moderator models suggest that: -the effect of thermal satisfaction on overall satisfaction was different for different values of control (interaction term therm. sat: control = −0.12) indicating that by increasing control, the effect of thermal on overall satisfaction decreased or in other words control slightly reduced the effect of thermal on overall satisfaction (control as moderator as visualized in Figure 6).
-the influence of thermal on overall satisfaction was higher (0.54) than the effect of visual on overall satisfaction (0.37).
Other interactions between control, thermal, and visual satisfaction were nonsignificant, and the same applies to sex, age-group, and T rm . and overall satisfaction is high as well.

| Thermal satisfaction (therm.sat)
Analyzing thermal satisfaction as dependent variable revealed that in the model with only physical variables as predictors (M.therm. sat.phys) outdoor illuminance and the calculated PMV based on measured physical parameters were significant and including selfperceived thermal control explained variance increased up to 66%.
Details for the mixed models with thermal satisfaction as dependent variable are described in the supporting information.
In the following, the moderator model with the highest R 2 m value (M.therm.sat.mod2) is described in detail. Table 9 presents estimates for this moderator model including physical parameters, visual satisfaction, and self-perceived control. Significant variables are PMV, visual satisfaction, and self-perceived thermal control, but not the interaction between visual satisfaction and self-perceived control. The models' behavior is visualized in Figure 8, which shows clear differences with and without control on thermal satisfaction, highlighting the main effect of control (0.55).

| Visual satisfaction (vis.sat)
Variance explained by the fixed effects in the models with visual satisfaction as dependent variable is in general extremely low, that is, R 2 m values below 0.1 (see supporting information). Considering fixed and random effects, random intercept models with physical variables explained 36% of variance in visual satisfaction, revealing no significant fixed effect. As expected, self-perceived thermal control was not a significant predictor for visual satisfaction in any of the models. Details for the mixed models with visual The models' behavior is visualized in Figure 9, demonstrating the small effect even of these two main variables on visual satisfaction.

| DISCUSS ION
In this study, we tested the hypothetical model presented in Figure 1 through a carefully designed experimental study. This discussion will first focus on thematic considerations and later on the methodological aspects.

| Thematic considerations
This section discusses results obtained for the analysis of interactions on overall satisfaction, combined effects on thermal satisfaction, and visual satisfaction. In each section, results from the mixed models will be discussed alongside observations from the descriptive analysis. As mentioned in the introduction, authors are not aware of other studies analyzing interactions between thermal and visual aspects under varying control scenarios so that direct comparisons to existing studies need to be done carefully.
The major findings to be discussed are: 1. self-perceived control moderates the interaction between thermal and visual conditions on overall satisfaction and influences thermal satisfaction.
2. overall satisfaction and visual satisfaction are strongly affected by subjective ratings rather than physical parameters.

thermal and visual satisfaction are the most important predictors
for overall satisfaction with thermal satisfaction having a stronger influence, but their interaction being not significant. Crossed effects were observed for thermal satisfaction on visual satisfaction and for visual satisfaction on thermal satisfaction.
In the present study, perceived thermal control reduced the effect of thermal satisfaction on overall satisfaction suggesting control being an important moderator of the influence of thermal satisfaction on overall satisfaction. Once people have control over the thermal environment, dissatisfaction with the thermal environment less affects overall satisfaction. Also, according to mean values presented in Figure 5, perceived control was highest While results suggest evidence for control as moderator, several limitations have to be mentioned here. First, perceived thermal control was measured as a single-item self-perceived control question related to the thermal environment. There are evidences that one could obtain other results when, for example, using selfefficacy scales rather than one item measures of perceived control. For example, Hawighorst et al 58 found significant influences of self-efficacy on overall satisfaction/comfort. In addition, future studies should add self-perceived control of visual environment to F I G U R E 7 Predicted overall satisfaction according to observed thermal satisfaction, visual satisfaction, self-perceived control, and season. Note that variables were scaled, that is, a value of 1.0 in overall satisfaction corresponds to a value of +50 on the visual analogue scale. Low T rm refers to a scaled value of −1, that is, 5.4°C before scaling, and high T rm refers to +1, that is, 18°C. Low control refers to −1, that is, a value of −30 on the visual analogue scale, and high control refers to +1, that is, +31.5 enable the assessment of its effect on visual satisfaction or as moderator to the influence of visual satisfaction on overall satisfaction.
Second, previous studies have shown that perceived control is lower in a two-person office compared with a single-person office. 25 This study was conducted in two-person offices, but did not permit assessing the effect of the other person in the same room on the level of perceived control. Therefore, part of the variance in perceived thermal control may be due to differences in the way participants'    self-assessed satisfactions also known as a common methods bias. 62 A VAS was applied here, because of its advantages over categorical scales in terms of data type obtained (continuous, rather than ordinal). In addition, typically applied scales to assess thermal satisfaction, visual satisfaction, and perceived control are based on distinct numbers of response categories ranging between 5 and 7.
Future analysis will need to assess these potential uncertainties in the results introduced through the choice of a VAS. In addition, interindividual differences might have been increased by the type of work performed by the participants. While participants were free to choose the type of work, some worked on their computer, while others engaged in reading tasks. This natural setting added further uncontrolled variance to the data. The introduced random-effects term likely captured parts of interindividual differences with this respect, but is not able to capture intraindividual day-to-day differences in the type of work performed. Such aspects need to be further controlled or at least monitored in future studies in order to quantify the corresponding effect and its influence on the results presented here.
Within this context, it is also worth discussing are the large

| Methodological considerations
The experimental design applied aims at a combination of elements found in less-controlled field studies and highly controlled laboratory studies as did earlier studies by the authors (see, eg, 25

| Limitations
Additional limitations and related tasks for future studies are as follows. The results are based on data from a laboratory study and should be validated in field studies, for example, real-world office buildings in all seasons. The present data were analyzed using mixed models instead of multilevel modeling (eg, lavaan package in R 69 ).
Some of the models analyzed, but not presented here, did not converge because of the small sample size. Further research should replicate the results with bigger sample sizes and by analyzing data with, for example, multilevel modeling. The physical variables for thermal and visual environment controlled in this study were operative temperature and illuminance level, respectively. Other environmental variables related to thermal and visual environments such as humidity and correlated color temperature of the lighting could not be controlled, but were measured. However, they could not be included in the analysis at this stage due to the relationship between complexity of the model and available sample size. A larger sample size will be required to include these factors in future studies.

| CON CLUS IONS
To date, little information is available on the modeling of relationships between thermal and visual comforts with overall satisfaction when taking into account confounding factors such as season, age, sex, or self-perceived control. The present study is according to our -Physical environmental conditions did not have a significant effect on overall and visual satisfaction likely due to the ranges of conditions applied and the measurement instruments applied.
In conclusion, this study highlights interactions and cross-modal effects between overall, thermal, and visual satisfaction and the important role of self-perceived control. As such, we recommend to carefully assess thermal, visual, and control scenarios jointly and not independently in future research studies and for future building design and operation strategies. Based on this, improvements in the design of the indoor environment can be made, for example, dynamic lighting adjusted to thermal conditions or vice versa to contribute to energy savings.

ACK N OWLED G EM ENTS
The analysis benefitted from discussions within IEA EBC Annex 66 and 79. The present study "Validation and modeling of user interactions and their algorithmic implementation in building automation including IEA Collaboration EBC Annex 66 (ValMoNuI)" was funded by the Federal Ministry of Economics and Energy (BMWi, 03ET1289B).
The funding source had neither involvement in study design, in the collection, analysis, and interpretation of data, nor the writing of the report and the submission for publication. M.S. was supported during the preparation of the manuscript by a research grant (21055) from VILLUM FONDEN. We would like to express our thanks to all research assistants and participants for supporting data collection.

CO N FLI C T O F I NTE R E S T
The authors certify that they have no affiliations with or involvement in any organization or entity with any financial interest (such as honoraria; educational grants; participation in speakers' bureaus; membership, employment, consultancies, stock ownership, or other equity interest; and expert testimony or patent-licensing arrangements), or nonfinancial interest (such as personal or professional relationships, affiliations, knowledge, or beliefs) in the subject matter or materials discussed in this manuscript.

AUTH O R CO NTR I B UTI O N
All authors revised the manuscript and agreed with its submission.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ina.12851.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.