Latent Psychological Pathways in Thermal Comfort Perception: The Mediating Role of Cognitive Uncertainty on Depression and Vigour

Abstract

Thermal comfort is the condition of mind that expresses satisfaction with the thermal environment, and it is assessed through subjective evaluation, according to the American Society of Heating, Refrigerating, and Air-Conditioning Engineers. While research has traditionally emphasised physical factors, growing evidence highlights the role of the state of mind in shaping thermal perception. In a prior Monte Carlo sensitivity analysis, six mood subscales—Anger, Confusion, Vigour, Tension, Depression, and Fatigue—were examined for how they affect the absolute difference between actual and predicted thermal sensation. Depression and vigour were found to be the most influential, while confusion appeared least impactful. However, to accurately assess the role of confusion, it is necessary to consider its potential interactions with other mood subscales. To this end, a mediation analysis was conducted using Hayes’ PROCESS tool. The mediation analyses revealed that confusion partially mediated depression’s effect in males and vigour’s effect in females. These results suggest that, despite a weak direct impact, confusion critically influences thermal perception by altering the effects of key mood states. Accounting for the indirect effects of mood states may lead to more accurate predictions of human sensory experiences and improve the design of occupant-centred environments.

1. Introduction

Thermal comfort is defined as “the condition of mind which expresses satisfaction with the surrounding thermal environment and is assessed by subjective evaluation,” as stated by the American Society of Heating, Refrigerating, and Air-Conditioning Engineers (ASHRAE) [1]. One of the most widely used models to assess thermal comfort is the Predicted Mean Vote (PMV)/Predicted Percentage of Dissatisfied (PPD) method, developed by Fanger [2]. This model is based on heat balance principles and considers four environmental parameters—air temperature (T_a), mean radiant temperature (T_r), air velocity (v_a), and relative humidity (RH)—as well as two personal parameters: basic clothing insulation (I_cl) and metabolic rate. The PMV model uses a seven-point thermal sensation scale, ranging from −3 (cold) to +3 (hot), with 0 representing a neutral thermal sensation, to predict the average thermal response of a large group of people [2]. While the PMV/PPD method provides a standardised and quantitative way to evaluate thermal conditions, it assumes steady-state environments and does not fully consider individual behavioral or psychological adaptations. In contrast, the Adaptive Thermal Comfort (ATC) model includes human behavioral, physiological, and psychological responses to the thermal environment, acknowledging that people actively interact with and adjust to their surroundings [3]. This adaptive approach is also relevant in psychological research, as it takes into account how people perceive and respond to their thermal environment, and its broader considerations may lead to different comfort assessments compared to the PMV/PPD model [4].

The phrase “condition of mind” in the ASHRAE definition highlights the psychological nature of thermal comfort [1]. This term indicates that thermal comfort is not only a result of physical measurements but is also influenced by individual perception, mood, and experience [5,6,7]. People may feel thermally comfortable or uncomfortable even under the same environmental conditions, depending on their mental and emotional state [4,8]. This shows that thermal sensation is not purely physical, but partly psychological. Understanding the role of the mind is essential for a full evaluation of thermal comfort, especially in contexts where personal well-being and subjective experience are important [9,10].

One key psychological factor that plays a role in thermal perception is mood state, which refers to a temporary emotional condition that can influence how individuals experience their environment. To evaluate the mood state, the Positive and Negative Affect Schedule (PANAS) [11] and the Profile of Mood States (POMS) [12] questionnaires are utilised. The PANAS was introduced in 1988 [11] and assesses two main dimensions of mood: Positive Affect (PA) and Negative Affect (NA). This tool includes 20 items rated on a five-point Likert scale and is used to measure both momentary emotional states and more enduring mood traits [11].

On the other hand, the POMS, developed by McNair et al. in 1971 [12], measures current mood through six subscales—Anger (ANG), Confusion (CON), Depression (DEP), Fatigue (FAT), Tension (TEN), and Vigour (VIG)—with 65 statements. In addition to the standard form, a short form of the POMS (POMS-SF) is available, which includes 37 items while retaining the original six subscales [13]. The POMS-SF [13] is frequently used in applied settings and large-scale research due to its reduced length and time efficiency, without compromising the reliability of the mood assessment. Participants are asked to reflect on the prompt “How are you feeling right now?” and rate each statement on a five-point Likert scale, ranging from 0 (not at all) to 4 (extremely), indicating how well each item describes their current emotional state. After completing the POMS questionnaire, Total Mood Disturbance (TMD) is calculated by summing the scores of the five negative subscales (the ANG, CON, DEP, FAT, and TEN) and subtracting the score of the one positive subscale (the VIG) [14,15]. These raw scores are then converted to T-scores to classify the participant’s overall mood state by Equation (1).

$$\mathrm{T-}\mathrm{S}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}=50+{\displaystyle \frac{10\times ({\mathrm{T}\mathrm{M}\mathrm{D}}_{\mathrm{p}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{i}\mathrm{c}\mathrm{i}\mathrm{p}\mathrm{a}\mathrm{n}\mathrm{t}}-{\mathrm{T}\mathrm{M}\mathrm{D}}_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}})}{\mathrm{S}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{d}\mathrm{a}\mathrm{r}\mathrm{d} \mathrm{D}\mathrm{e}\mathrm{v}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n} \mathrm{o}\mathrm{f} \mathrm{T}\mathrm{M}\mathrm{D}}}$$

(1)

A graphical representation of the mood classification is presented in Figure 1. In addition, the POMS questionnaire can be seen in Appendix A, and the subscales and their representations are depicted in Appendix B.

There have been some thermal comfort studies conducted by using the PANAS and POMS questionnaires. For instance, Lin and Zhang [16] conducted a study to examine the effects of psychological aspects on the thermal sensation and used the PANAS questionnaire to categorise the emotional states. When participants were in a stressed state, many of them reported feeling warm or slightly warm. In contrast, during times when they were not stressed or depressed, their perception of temperature was mostly neutral or slightly uncomfortable, with some feeling either too cold or too hot. Regarding humidity, participants who were stressed often described the environment as humid, while those in a more positive and relaxed state generally perceived the humidity as neutral [16].

The POMS offers a more comprehensive and sensitive assessment of mood changes than the PANAS [17]. The POMS includes 65 items covering six mood dimensions, which allows it to capture a wider range of emotional states. In contrast, the PANAS has only 20 items and measures just two dimensions, positive and negative effects. This makes the POMS more suitable for multidimensional and accurate analysis of mood, especially in experimental contexts.

Using the POMS questionnaire in thermal comfort studies is a new approach in the literature. Therefore, only a few thermal comfort studies that utilised the POMS questionnaire were available, and some information about these studies is given in

Table 1. Studies that utilised the POMS questionnaire to explore the effects of mood state on thermal sensation/comfort.

Study	Research Question	Results
Turhan et al. [18]	Do mood states influence individuals’ thermal sensation?	A significant difference was observed between the control and experimental groups, indicating that mood states affect thermal sensation.
Özbey et al. [4]	How does the current mood state, as measured by the POMS subscales, influence the absolute difference between perceived and predicted thermal sensation, and which mood subscales are the most sensitive to this difference?	The most sensitive parameters were found in VIG and DEP, while the least sensitive subscale was CON.
Çeter et al. [19]	How does emotional intensity relate to the absolute difference between perceived and predicted thermal sensation across genders?	Statistically, males were found to be more sensitive to changes in emotional intensity compared to females.
Özbey and Turhan [20]	Can a comfort temperature model be developed that considers both outdoor temperature and psychological mood states?	Comfort temperatures varied significantly among individuals in non-neutral mood states, with lower comfort temperatures observed in those experiencing higher or lower levels of concern.
Turhan et al. [21]	Can a fuzzy logic model predict thermal sensation accurately by including mood states as input variables?	The developed fuzzy logic model predicted thermal sensation with an accuracy of 92%.
Özbey et al. [22]	How is the Tension subscale of the POMS questionnaire correlated with thermal sensation?	For males, the items “shaky,” “uneasy,” and “anxious” were significant, while for females, “nervous” and “anxious” were significant predictors.
Turhan et al. [23]	Is it possible to model thermal sensation using a black-box approach that incorporates mood states?	Individuals in neither pessimistic nor optimistic mood groups tended to perceive warmer temperatures.
Yerlikaya et al. [24]	Can the previously developed black-box model [23] be validated using a data mining method like Multivariate Adaptive Regression Splines (MARS)?	The correlation coefficients between the MARS and the black-box model were 0.9426 (training) and 0.9420 (testing), confirming the high accuracy of modelling mood states with the MARS.
Özbey and Turhan [25]	How do individual POMS subscales influence thermal sensation, and what are the positive and negative effects of these mood states on perceived thermal comfort?	VIG was the most influential parameter, while CON and FAT had negative effects (feeling cooler) on thermal sensation.

.

In addition to psychological parameters, studies have consistently revealed a remarkable gender difference in terms of thermal comfort. According to Karjalainen’s meta-analysis, females tend to express more thermal dissatisfaction and need more individual temperature control along with adaptive actions [26]. The thermal sensitivity of females was found to be greater than that of males, and women feel significantly colder than men in cold environments [27,28,29,30]. Furthermore, females report more drowsiness and boredom than males at temperature extremes, suggesting that psychological variables such as mood and mental state differently influence males and females [19,22,31].

There is a well-established relationship between the mental and emotional states of human beings. Since CON is the only cognitive dimension of the POMS questionnaire [32], this dimension requires special attention due to its relationship with other mood states. All seven items of this dimension (Confused, Bewildered, Efficient (reverse-coded), Unable to Concentrate, Muddled, Uncertain, Forgetful) refer to lack of mental clarity, which is a decline in the cognitive functions of individuals, including perception, learning, memory, and judgement. When individuals are confused, their perception is clouded; they have difficulties with mental focus, i.e., concentration, presence in the moment or environment, and consequently, their mood states are affected. This relationship is an inevitable outcome of the complex yet well-established interaction between cognition and emotion [33,34,35]. For example, in terms of psychopathology, confusion—difficulty in concentration and thinking—is both a symptom and a diagnostic criterion for a broad array of depressive, trauma, stress-related, and anxiety disorders [35]. Similarly, studies on the POMS to date have shown that, across different samples, confusion is highly correlated with anxiety, depression [36], anger, and tension [37]. These findings suggest that a lack of mental clarity could be a powerful but possibly latent mental factor associated with the other mood states in the POMS questionnaire.

In a previous study, Özbey et al. [4] used a Monte Carlo sensitivity analysis to investigate how the six main subscales of the POMS questionnaire influence the absolute difference between predicted and perceived thermal sensation. That analysis was based on the idea that mood may contribute to discrepancies between actual and predicted thermal perception. The results showed that changes in mood states could cause differences of up to 1.32 points on the thermal sensation scale. Among the subscales, VIG and DEP had the strongest effects, causing changes of 0.31 and 0.30, respectively, while CON had the smallest direct effect, with a change of only 0.11. However, even though CON seemed to have a limited direct impact, it may still play an important indirect role by influencing the effects of other mood states. Therefore, the aim of this study is to examine whether CON acts as a mediator in the relationship among the other mood subscales (ANG, DEP, FAT, TEN, and VIG) and the absolute difference between TSV and PMV (ABS (PMV-TSV)) values. To the best of the authors’ knowledge, this is the first study to apply mediation analysis within a thermal comfort context using mood states as psychological predictors. By focusing on confusion as a mediating variable, this research provides a novel psychological perspective that extends beyond conventional direct-effect models and offers a more nuanced understanding of how mental states shape thermal perception.

The remainder of this paper is organised as follows: Section 2 describes the methodology, including data collection, variables, and the mediation analysis procedure. Section 3 presents the results of the mediation analysis. Section 4 discusses the findings considering previous research available in the literature. Finally, Section 5 concludes with key insights and suggestions for future work.

2. Materials and Methods

2.1. Participants and General Procedure

The overall research workflow is summarised in Figure 2. This study followed five main steps. First, ethical approval was received from the university’s Ethics Committee before beginning the experiments. Then, the experimental setup was prepared, including the testing area and the necessary equipment. In the next step, both subjective data (through questionnaires) and objective data (from environmental sensors) were collected. After data collection, the raw data were scored and organised to be ready for analysis. Finally, statistical tests such as t-tests and mediation models were applied to examine the relationship between mood states and thermal comfort.

A total of 1159 participants (440 female and 719 male), aged between 18 and 68 (Mean (m) = 23.78, Standard Deviation (SD) = 4.41), were included in the experiments. All participants met the inclusion and exclusion criteria that were set for this study and gave written consent for their volunteer participation.

2.2. Case Building

A university study hall in Ankara, Türkiye, was selected as the case building. Seasonal average air temperatures in Ankara are 1.6 °C in winter to 22.3 °C in summer, with spring and autumn averaging approximately 11.1 °C and 13.2 °C, respectively [38]. The space, totalling 365 m² (29.5 m length × 12.4 m width), is primarily occupied by researchers and students and relies solely on ten radiators fixed at 22 °C for heating; no active cooling is available. All measurements were taken within a central 10 m² zone to minimise door-induced airflow. Further architectural and material details are available in previous studies of [4,19,20,22,23,24]. In addition, the representative photo from the experimental area can be seen in Figure 3. Moreover, the schematic diagram showing the position of the sensors can be seen in Figure 4.

2.3. Instruments and Data Collection

To clarify the experimental setup, the main criteria are summarised below:

• Study conducted between August 2021 and August 2022, during office hours.
• Indoor conditions (air temperature, humidity, globe temperature, and air velocity) logged every minute with calibrated sensors at 1.1 m height.
• Outdoor temperature and humidity monitored using a separate data logger.
• Participants submitted their personal data (gender, age, clothing, mood, and thermal sensation) via a developed mobile application.
• A 15 min rest period before starting the experiments ensured a stable metabolic rate (1.1 met).
• Basic clothing insulation values were calculated based on ASHRAE-55 [1].
• The POMS questionnaire was used for psychological assessment, and the average scores of each subscale were calculated before conducting the data analysis.

Between 16 August 2021 and 1 August 2022 (09:30–16:30), subjective and objective measurements were recorded. Indoor environmental variables (air temperature, air velocity, globe temperature, and relative humidity) were logged every minute using a Delta OHM TCA 32.3 [39] thermal comfort data logger, while outdoor conditions (outdoor air temperature and outdoor relative humidity) were monitored with a DHT-22–based environmental data logger [40]. All indoor environmental sensors were mounted at a height of 1.1 m, which represents head level for a seated person. The specifications of the sensors used for objective data are given in

Table 2. The specifications of the sensors used to collect objective data.

Measurement Device	Parameter	Range	Accuracy	Resolution
DELTA OHM HD32.3TCA [39]	Globe Temperature	−10–100 °C	±0.1 °C	0.1 °C
	Air Velocity	0.02–5 m/s	±(0.05 + 5% of the measurement) m/s	0.01 m/s
	Indoor Air Temperature	40–100 °C	±0.1 °C	0.1 °C
	Relative Humidity	0–100% RH	±1.5%	0.1%
Environmental Data Logger DHT-22 [40]	Outdoor Temperature	−40–80 °C	<±0.5 °C	0.1 °C
	Outdoor Relative Humidity	0–100%	±2%	0.1%

.

Participants used a mobile application, which was developed by the authors, to submit general personal information (age, gender, height, and weight), clothing details, their POMS questionnaire responses, and thermal sensation votes on a 13-point scale [41] (Figure 5). Each participant took part only once; subjective entries required approximately 10–15 min, objective readings were recorded at 1 min intervals, and later they were averaged over the same interval. Pre-experiment rest of 15 min ensured a uniform metabolic rate (1.1 met—seated in a typing position [1]). Moreover, basic clothing insulation values were calculated according to the ASHRAE-55 [1].

Since those on long-term antidepressants or reporting recent metabolic or substance-use issues were excluded to protect the POMS validity, participants who had not been using antidepressant medication for the past 3 months were included in the study [42,43]. Moreover, participants who had used caffeine, alcohol, or smoked any kind of inhalant 12 h prior to the experiments were not included in the experiments, since these substances are also known for their effects on the central nervous system, with potential for altering the mood and physiological states and confounding the results obtained from the POMS questionnaire [44,45,46]. The last exclusion criterion was intense physical activity. In order to eliminate any changes in body temperature, heart rate, and overall metabolic rate, participants who had not engaged in any intense physical activity for the previous 12 h were included in the experiments. To avoid any influence from the ‘Observer Effect’ or ‘Social Presence Effect’ on the validity of the POMS questionnaire during the experiments, participants were not allowed to remain in close proximity to one another [47]. Other psychological traits, such as personality traits and individual characteristics, fell outside this study’s scope. The representative sequence of experimental tasks, from participant arrival through data recording, is illustrated in Figure 6.

For the psychological assessments, the 65-item POMS questionnaire [12] was utilised. Participants rated each item on a four-point Likert-type scale ranging between 0 (“not at all”) and 4 (“extremely”). Scores for each subscale—ANG, CON, DEP, FAT, TEN, and VIG—were calculated by averaging the item scores. Two items, “efficient” and “relaxed”, were reverse-coded. The items related to friendliness were not considered because they did not clearly differ from those in the VIG subscale [48].

2.4. Statistical Analysis

All statistical analyses were conducted using SPSS v.28 [49]. Before performing the statistical analysis, the normality of the data was checked, and outliers were determined by using the SPSS check data function. Cases that were detected as “abnormal” were removed from the dataset. Afterwards, to examine gender differences in the variables of interest, independent samples t-tests were performed. The null hypothesis for this analysis (H₁) is given below.

H1: 

There is no significant difference between male and female participants in the ANG, DEP, FAT, TEN, and VIG variables as measured by the POMS subscales.

In addition, the effect size was calculated in case there was a significant difference between the genders. Effect size is the standardization of the difference between means. In other words, it is a standard value that shows the extent of the difference between group means, as well as the strength of the independent variable’s effect on the dependent variable. In this study, the effect size was calculated using Cohen’s d [50].

A series of mediation analyses was conducted to examine whether “Confusion” acted as a mediating variable in predicting the absolute difference between predicted and perceived thermal sensation votes affected by five mood states—ANG, DEP, FAT, TEN, and VIG—and for the total population and each gender. The conceptual diagram for the mediation model, represented in the form of a statistical diagram, can be found in Figure 7.

A conceptual mediation model includes two outcome variables (M and Y) and two input variables (X and M). In this model, X affects both M and Y, while M also affects Y. The mediation model shows how a variable X influences an outcome Y through another variable M. There are two main paths from X to Y: a direct effect, where X directly affects Y, and an indirect effect, where X affects M, which then affects Y. The indirect path shows how X influences Y through M. In such models, M is called a mediator, but it may also be called an intermediary or intermediate variable in some fields [51]. The M and Y can be found by using the following equations [51]:

$$\mathrm{M}={\mathrm{i}}_{\mathrm{m}}+\mathrm{a}\mathrm{X}+{\mathsf{\gamma }}_{\mathrm{m}}$$

(2)

$$\mathrm{Y}={\mathrm{i}}_{\mathrm{y}}+{\mathrm{c}}^{\prime }\mathrm{X}+{\mathrm{b}}_{\mathrm{M}}+{\mathsf{\gamma }}_{\mathrm{y}}$$

(3)

Here, $i_m$ and $i_y$ represent the regression intercepts, while ${\gamma }_m$ and ${\gamma }_y$ are the error terms in predicting the mediator M and the outcome Y, respectively. The coefficients a, b, and $c^{\prime }$ are the regression coefficients among the related variables. Specifically, in Equation (3), $c^{\prime }$ reflects the direct effect of X on Y. This means that when two observations differ by one unit in X but have the same value for M, their predicted Y values will differ by $c^{\prime }$ units (as shown in Equation (4)) [51].

$${\mathrm{c}}^{\prime }=\left[\widehat{\mathrm{Y}}|\left(\mathrm{X}=\mathrm{x},\mathrm{M}=\mathrm{m}\right)\right]-\left[\widehat{\mathrm{Y}}|\left(\mathrm{X}=\mathrm{x}-1,\mathrm{M}=\mathrm{m}\right)\right]$$

(4)

Here, m stands for a specific value of the mediator variable (M), the vertical bar “|” means “given” or “conditioned on”, and the symbol $\widehat{Y}$ indicates the expected or predicted value of Y based on the model. Therefore, if two cases have the same value of M but differ by one unit in X, the difference in their estimated Y values is given by $c^{\prime }$ representing the direct effect of X on Y [51].

The indirect effect of X on Y through M is calculated by multiplying the coefficients a and b. For example, if a = 0.5 and b = 0.4, then the indirect effect is 0.2. This means that for every one-unit change in X, the expected change in Y—caused indirectly through M—is 0.2 units. In other words, X influences M, which then influences Y. Finally, the total effect of X on Y, represented by $c$, is the sum of the direct and indirect effects (as shown in Equation (5)) [51].

$$\mathrm{c}=c^{\prime }+(\mathrm{a}\times \mathrm{b})$$

(5)

The scheme of the mediation analysis for this study is given in Figure 8.

The mediation analyses were performed using the PROCESS v4.2 macro, a widely used statistical method to facilitate complex mediation and moderation analysis [51]. It employs ordinary least-squares regression and bootstrapping techniques to estimate indirect effects [51]. It was implemented in SPSS [49], and analysis was conducted with bootstrapping with 5000 resamples. Indirect effects were considered significant if the 95% confidence intervals did not include zero. A PROCESS Model 4 diagram illustrating this mediation structure is provided in Figure 8. The corresponding null hypothesis (H₂) for this analysis is as follows:

H2: 

Confusion does not mediate the relationship among the POMS subscales (ANG, DEP, FAT, TEN, and VIG) or the absolute difference between TSV and PMV.

3. Results

The mean and standard deviations for indoor and outdoor objective data measurements of parameters are given in

Table 3. Descriptive statistics for objective data measurements for indoor and outdoor environments.

Indoor Parameters	m	SD	Outdoor Parameters	m	SD
Ta (°C)	21.82	1.38	Toutdoor (°C)	11.61	8.73
Tr (°C)	21.94	1.36
va (°C)	<0.01	0.02	RHoutdoor (%)	57.34	21.11
RHindoor (%)	33.51	9.34

.

Indoor air temperature (T_a) and mean radiant temperature (T_r) were relatively stable, since the set temperature was 22 °C and the radiators were operating during the winter season. The experiments were conducted under the still air conditions, and since the experimental area was in the middle of the experimental room to avoid air drafts, the v_a remained very low. Detailed information about the collected objective data can be found in [20], in which the same dataset was used.

As described in the methodology, the normality was checked, and 47 cases (8 female, 39 male) that were detected as “abnormal” were removed from the dataset. Furthermore, independent sample t-tests were conducted to determine whether there were significant differences between male and female participants in terms of their ABS (PMV-TSV) scores and their average scores for the ANG, DEP, FAT, TEN, and VIG subscales. Descriptive statistics for the study variables are shown in

Table 4. Descriptive statistics for average scores of the POMS subscales and ABS (PMV-TSV).

	Female				Male				All
	m	SD	sk	K	m	SD	sk	k	m	SD	sk	k
ANG	1.13	0.65	0.45	−0.12	1.12	0.57	0.58	0.67	1.12	0.60	0.53	0.31
CON	1.35	0.52	0.72	0.78	1.34	0.50	0.43	0.79	1.35	0.51	0.55	0.78
DEP	1.02	0.64	0.70	0.56	1.02	0.53	0.63	0.99	1.02	0.58	0.67	0.86
FAT	1.62	0.69	0.31	0.07	1.66	0.66	0.10	0.51	1.65	0.67	0.18	0.30
TEN	1.52	0.62	0.46	0.17	1.48	0.57	0.49	0.73	1.50	0.59	0.49	0.49
VIG	1.86	0.58	−0.33	0.41	2.13	0.53	−0.75	0.78	2.02	0.57	−0.59	0.45
ABS (PMV-TSV)	0.70	0.55	0.98	0.42	0.75	0.55	0.99	0.50	0.73	0.55	0.99	0.46

, and the results of the t-tests are depicted in

Table 5. Comparison of male and female participants’ mean scores on ABS (PMV-TSV), ANG, DEP, FAT, TEN, and VIG.

	Levene’s Test for Equality of Variances	df	t	p	95%CID Lower	95%CID Upper
ANG	Equal variances not assumed (p < 0.05)	820.886	−0.164	0.870	−0.81	0.07
CON	Equal variances assumed	1110	−0.091	0.927	−0.06	0.06
DEP	Equal variances not assumed (p < 0.05)	798.881	0.112	0.911	−0.07	0.08
FAT	Equal variances assumed	1110	0.933	0.351	−0.04	0.12
TEN	Equal variances not assumed (p < 0.05)	863.203	−1.054	0.292	−0.11	0.04
VIG	Equal variances assumed	1110	7.869	<0.001 *	0.20	0.33
ABS (PMV-TSV)	Equal variances assumed	1110	1.196	0.232	−0.03	0.11

* p < 0.05 indicates statistical significance.

.

As shown in Table 5, there was a statistically significant difference in the VIG scores between males and females (t (1110) = 7.869, p < 0.05). Male participants (2.13) reported significantly higher VIG scores compared to female participants (1.86). However, no significant differences were observed between males and females in their mean scores for ANG, DEP, FAT, and TEN, or ABS (PMV-TSV). For these variables, the mean scores were similar across genders. It is important to note that, since there is a statistically significant difference in the VIG scores between males and females, the effect size (Cohen’s d) was calculated and found to be 0.491. This indicates that the difference in VIG scores between males and females has a modest effect size [50].

In order to determine whether confusion has a mediating effect in the relationships among the independent variables—ANG, DEP, FAT, VIG, TEN—and ABS (PMV-TSV), Hayes’ PROCESS analysis (Model 4) was conducted separately for males, for females, and for the entire sample. Sequential mediation analyses were conducted separately for each POMS subscale. Among the total of 15 analyses performed, only those that fulfilled all mediation criteria and revealed a mediation relationship (mediator) were reported in this study. A mediation effect was observed in only two models. Confusion acted as a partial mediator in the relationship between DEP and ABS (PMV-TSV) among males, and between VIG and ABS (PMV-TSV) among females. No significant mediating effect of CON was found when taking into account all the data. The model is given in Figure 9, and the results are depicted in

Table 6. Mediation effect of CON in the relationship between dependent variables—VIG_female and DEP_male—and dependent variables—ABS (PMV-TSV)_female and ABS (PMV-TSV)_male.

Model Pathways		β	Standard Error	t	p	LL95%CI	UL95%CI
VIGfemale	a	−0.096	0.043	−2.009	0.045 *	−0.171	−0.002
	b	0.104	0.051	2.169	0.031 *	0.010	0.210
	c	−0.118	0.046	−2.467	0.014 *	−0.202	−0.023
	c’	−0.108	0.046	−2.257	0.024 *	−0.192	−0.013
DEPmale	a	0.716	0.025	26.700	<0.001 *	0.622	0.721
	b	0.195	0.060	3.582	<0.001 *	0.096	0.330
	c	−0.083	0.039	−2.170	<0.001 *	−0.163	−0.008
	c’	−0.222	0.056	−4.092	<0.001 *	−0.338	−0.119

* p < 0.05 indicates statistical significance.

.

A mediation analysis was conducted using Hayes’ PROCESS macro (Model 4) to examine whether the relationship between the independent variables (VIG_female and DEP_male) and the dependent variables (ABS (PMV-TSV)_female and ABS (PMV-TSV)_male) is mediated by CON_female.

The results revealed a significant negative effect of VIG_female on the mediator (path a: β = −0.096, p = 0.045), and a significant positive effect of the mediator on VIG_female (path b: β = 0.104, p = 0.031). Furthermore, it was found that 1.39% (β²) of the variance in ABS (PMV-TSV)_female could be explained by VIG_female. The total effect of VIG_female on the outcome was also significant (path c: β = −0.118, p = 0.014), and the direct effect remained significant after controlling for the mediator (path c’: β = −0.108, p = 0.024). These findings indicate a partial mediation, suggesting that CON partially explains the relationship between VIG and ABS (PMV-TSV) for females. In addition, the ratio of the indirect effect to total effect is 8.92%.

On the other hand, for male participants, CON has a mediator effect on the relationship between DEP and ABS (PMV-TSV). The results revealed a significant positive effect of DEP_male on the mediator (path a: β = 0.716, p < 0.001) and a significant positive mediator effect on DEP_male (path b: 0.195, p < 0.001) Furthermore, it was found that 0.68% (β²) of the variance in ABS (PMV-TSV)_male could be explained by DEP_male. The total effect of DEP_male on the outcome was also significant (path c: β = −0.083, p < 0.001), and the direct effect remained significant after controlling for the mediator (path c’: β = −0.222, p < 0.001). These findings indicate a partial mediation, suggesting that CON partially explains the relationship between DEP and ABS (PMV-TSV) for males. In addition, the ratio of the indirect effect to total effect is 38.44%.

4. Discussion

In this study, the POMS subscales and their relationship to thermal comfort discrepancies (absolute difference between PMV and TSV) were examined for male and female participants, both separately and together. Overall, males and females showed very similar scores for the POMS subscales, with the only significant gender difference emerging for VIG in favor of males, leading to a rejection of our first null hypothesis. Our mediation analyses revealed that confusion emerged as a significant mediator of the relationship between VIG and thermal comfort discrepancies in females and as a stronger mediator of the relationship between DEP and thermal comfort discrepancies in males. This result led to the rejection of our second null hypothesis.

Within the scope of gender-specific analysis, the findings reveal some variations. Existing studies indicate that males often report higher levels of VIG [52,53,54], and our results are consistent with the extant literature.

Additionally, the absolute difference between the predicted thermal sensation vote (PMV), based on the ASHRAE-55 standard [1], and the actual reported thermal sensation vote (TSV) was utilised. This difference, calculated as ABS (TSV-PMV), provides insight into the potential mismatch between standard comfort models and subjective thermal experiences. Our findings indicate that the average absolute difference was 0.73 (SD = 0.55) across all participants, with a slightly higher deviation in males (m = 0.75) compared to females (m = 0.70). These results are consistent with prior studies reporting the difference between actual and predicted thermal sensation votes [55,56]. Moreover, previous studies have reported that mood states can lead to a notable deviation from predicted thermal comfort conditions, especially when individuals deviate from the psychological neutral state, such as when experiencing optimistic or pessimistic moods [20,23]. The observed discrepancies emphasise that while PMV is a foundational tool in thermal comfort prediction, it may not fully capture individual thermal perceptions under varying psychological states.

To place the current study within the broader context of existing research and to highlight its scientific contribution, a comparative review of relevant studies was conducted, as seen in Table 1. Previous thermal comfort studies that involve the usage of the POMS questionnaire have demonstrated that mood states significantly influence thermal sensation, as seen in [18], where experimental groups exhibited measurable differences based on mood. Other studies have focused on specific subscales of the POMS questionnaire, such as [3], which identified VIG and DEP as the most influential, and CON as the least. Additional studies have developed thermal comfort models that integrate mood states using various methodologies, including fuzzy logic, black-box modeling, and data-mining techniques like MARS [23,24]. However, none of these studies have explicitly investigated the mediating role of Confusion in modulating the impact of the other mood subscales on the difference between perceived and predicted thermal sensations in a gender-based approach. This study fills that gap by introducing mediation analysis to reveal hidden psychological pathways, thereby extending current knowledge and offering a novel approach to enhancing thermal comfort prediction. To understand the hidden effects of CON on the association among the other subscales of the POMS questionnaire and the absolute difference between perceived and predicted thermal sensation, the mediation analysis was conducted via Hayes’ PROCESS macro. The use of Hayes’ PROCESS macro enables robust estimation of mediation effects using ordinary least-squares regression and bootstrapping, ensuring statistical reliability.

From the previous study [4], CON has been found to have the least effect on the absolute difference between TSV and PMV (ABS (PMV-TSV)); in other words, the absolute difference between predicted and perceived thermal sensation. According to the literature, cognitive uncertainty may affect the responses on DEP and VIG, as indicated in [57]. Since confusion, by its nature, reflects a lack of mental clarity, its effect on two salient mood states can be explained by the effect of cognitive factors in perceiving and experiencing emotions. According to the cognitive appraisal theories of emotions, distinct emotions are formed and experienced by mental activities, i.e., appraisal of the inner and outer stimuli [58,59]. Since both the DEP and VIG subscales contain specific positive and negative emotions—such as unhappy, sad, and guilty in the DEP subscale, and cheerful, lively, and full of pep in the VIG subscale—mental confusion might have affected the cognitive appraisals of the participants, leading to a mediating role for confusion on the thermal comfort levels of participants. In other words, our findings can be interpreted as support for cognitive theories of emotions.

The results of this study also support the existing literature in terms of gender difference. There are possible reasons for this difference. The physiological and hormonal differences between the genders could be one of these reasons [60]. Another explanation can be the difference in sensitivity levels of males and females. It is known that females are more thermally sensitive than males [27,28,29,30]. A similar sensitivity difference in terms of emotions may lead to such variations. Unfortunately, this topic falls beyond the scope of this study, and future studies need to examine the possible sources of this difference.

The prediction of TSV plays a central role in optimizing thermal comfort and energy efficiency in built environments. Traditional models often rely on environmental parameters (air temperature, mean radiant temperature, air velocity, and relative humidity) and personal parameters (metabolic rate and basic clothing insulation). However, recent research has demonstrated that psychological factors, particularly mood states, significantly influence perceived thermal comfort. By incorporating these psychological inputs—such as those measured through the POMS questionnaire—into predictive models, the accuracy of TSV estimation can be substantially improved. This improvement in prediction precision would allow heating, ventilating, and air conditioning (HVAC) systems to respond more effectively to actual human comfort needs, thereby reducing unnecessary energy use. To implement such a system in real-world settings, as indicated in one of the previous studies [21], a technological framework could be established involving a mobile application for user input, environmental data loggers for real-time measurement of indoor and outdoor conditions, a central server to run the predictive model, and wireless communication modules such as Wi-Fi for data exchange and control. The TSV predicted by the model could be used to automatically adjust the setpoint temperature, allowing HVAC systems to respond dynamically not only to physical but also psychological comfort indicators. By including both psychological and physiological components in the decision-making process, such systems could provide a personalised and responsive approach to indoor climate control, which would not only enhance occupant well-being but also contribute to sustainable energy management in buildings. Hence, this personalised approach to thermal regulation has the potential to improve user satisfaction while simultaneously promoting energy efficiency in real-world applications.

5. Conclusions

This study explored the mediating role of the CON subscale in the relationship among other POMS subscales and the absolute difference between predicted and perceived thermal sensation (ABS (PMV-TSV)). While CON has been previously identified as the least influential subscale [4], the current findings reveal its indirect influence on thermal comfort perception, particularly through mediation effects. Key findings of the study include the following:

• Confusion significantly mediated the relationship between VIG and ABS (PMV-TSV) in female participants, with an 8.94% indirect-to-total effect ratio.
• Confusion also mediated the relationship between DEP and ABS (PMV-TSV) in male participants, with a 38.44% indirect-to-total effect ratio.
• As a conjecture, if we consider previously reported direct effects of VIG (0.31) and DEP (0.30) on ABS (PMV-TSV) from studies without gender differentiation, CON may contribute approximately 0.03 (8.94% of 0.31) in females and approximately 0.12 (38.44% of 0.30) in males. It is vital to emphasise that these values are approximations, based on prior literature that did not account for gender-specific analysis.
• These results indicate that the CON subscale plays a more critical role in thermal comfort perception than previously assumed.

Several limitations to this pilot study need to be acknowledged. This study was limited to the Turkish population, educational buildings, and the warm Mediterranean climate zone. Additionally, even though the age range of this study was 18–68, the majority of participants were undergraduate and graduate students, with 96.6% of them aged between 18 and 35 years. Therefore, changes in the sample population, building type, climate zone, or age range could influence the findings, as other studies have indicated [61,62,63].

Further research is needed to better understand the psychological dimensions of thermal comfort, particularly across different cultures and climates, and such research should integrate physiological measures, such as skin or core body temperature, to corroborate self-reported data and enhance model accuracy. Moreover, the inclusion of real-time environmental monitoring (e.g., high-resolution thermal imaging) to capture micro-climatic variations in occupied spaces might be valuable for understanding the effects of cognitive uncertainty. These findings may inform more adaptive and occupant-centered HVAC control systems by integrating psychological variables—such as mood state tracking—into thermal comfort models, potentially improving perceived comfort and energy efficiency.