How long is too long? Examining waiting times and stress in human-elevator interaction

Time perception

According to past studies, various factors influence time perceptions, such as variety and types of different stimuli (Buffardi, 1971; Mo, 1975), subjective emotions (Droit-Volet & Meck, 2007; Tipples, 2008), stimulus types, temporal paradigms (Cui et al., 2023), age (Hancock & Rausch, 2010), and gender (Hanson & Buckworth, 2016). These studies reported that negative emotions increased the perceived times more than positive emotions, and word stimuli showed relatively smaller distortions of time perception compared to visual and sound stimuli. From another perspective, a past study reported that even metabolic rates delay time-flow perceptions (Oshakbayev et al., 2024). In addition, several studies focused on the cognitive models of time estimation (Zakay & Block, 1996), and reported that prospective judgments are longer and less variable than are retrospective judgments (Block & Zakay, 1997).

These studies provided the essential basic knowledge to understand what factors affect time perception, establishing critical guidelines for experimental design for related works that focus on time perception. On the other hand, since the main purpose of this series of studies was to verify the impact of detailed factors on time perception under strictly controlled conditions, their research aims are different from ours. We focused on directly verifying the relationship between waiting times and stress when using elevators.

Relationships between waiting times and stresses

Various empirical studies have been conducted on how the unavoidable waiting times for various services affect satisfaction with service and stress. For example, a past study investigated the perceived waiting times reported by passengers at various types of transport stops and stations and compared it with the actual waiting times measured by recorded videos; the perceived waiting times exceeded the actual waiting times (Fan, Guthrie & Levinson, 2016). Another study investigated different kinds of transportation services (Ayodeji, Rjoub & Özgit, 2023) and reported that passenger satisfaction with waiting at airports is essential for customer loyalty in long-term sustainability. Waiting times strongly influence such different services as fast-food restaurants and health services (Abdul Hamid, 2021; Bielen & Demoulin, 2007; Pruyn & Smidts, 1998). From another perspective, a past study reported that for web users, the tolerable waiting times for information retrieval is about 2 s (Nah, 2004). Similar to such human-computer interaction, people perceived more stress when the response time of a conversational robot exceeds 2 s (Shiwa et al., 2009), although overly hasty response times are also perceived as strange by people in interactions with anthropomorphized robots (Shiomi et al., 2018; Shiwa et al., 2009).

Improving satisfaction with waiting environments is one effective way to decrease the perceived stress caused by waiting times (Bielen & Demoulin, 2007; Nah, 2004). One traditional and famous approach related to satisfaction while waiting for an elevator is to install a mirror (that diverts the attention of those who are waiting) in an elevator hallway (Sasser et al., 1978). Another study placed television sets in waiting environments and described the customer satisfaction with their waiting environments (Pruyn & Smidts, 1998). Such approaches are useful and effective in real service contexts.

However, although these past studies investigated the relationships between waiting times and stress related to satisfaction in various service settings, they neglected situations where passengers wait or take elevators. Nor has the relationship between their waiting times and perceived stress in such situations received adequate focus.

Human-elevator interaction

Some studies focused on understanding how people use elevators. For example, a past study analyzed the in-cabin behavior of passengers using multiple sensor devices and identified the preferred standing positions and movement patterns in elevators (Robal et al., 2022). Another study investigated preferred standing positions in elevators through a web-based survey to design acceptable behaviors of mobile robots that share elevators with humans (Gallo et al., 2022). From an engineering perspective, researchers developed voice-based control functions and investigated voice-based interactions with smart elevators to improve user experiences (González-Docasal et al., 2024; Meenatchi, Aishwarya & Shahina, 2016; Shiomi, Kakio & Miyashita, 2024).

These studies provided knowledge that contributed to achieving more natural and smooth interaction between passengers and elevators in a social context, although they focused less on understanding the gaps between perceived waiting times and actual waiting times, as well as the relationships between these times and perceived stress. Another study investigated the relationships among waiting time, transit time, and perceived stresses in elevator use cases; unfortunately, their experiment was an online survey and focused on specific lengths of waiting times (Bird et al., 2016). Therefore, the relationships between waiting times and perceived stresses in actual settings remain unknown.

Summary of our study

Our current study investigated the above two research questions through an empirical study where participants rode an actual elevator in a real building. First, we analyze the gap between perceived and actual waiting times in two situations: waiting for the elevator and riding it. Then we analyzed the relationships between the perceived stress and perceived/actual waiting times.

Environment and data collection settings

Figure 1 shows the experimental environment. We used an elevator in our laboratory. The elevator has elevator car operating panels on both sides of the door. There were no mirrors or posters inside the elevator. It is 2.3 m high, 1.35 m wide, and 1.6 m deep. We installed a camera inside it and on the ceiling near the elevator’s hallway to record video images during the experiment.

Participants

We recruited 30 participants (15 women and 15 men), all native Japanese speakers (mean age = 34.2 years, SD = 9.00), and conducted 48 trials with each participant (1,440 trials in total). The numbers of participants and trials were determined based on empirical heuristics and the constraints imposed by participants’ limited time availability; i.e., there was no a priori justification for the sample size. They were enrolled through a local commercial recruiting company and paid for their participation. All participants were physically healthy and had no mobility issues, resulting in similar riding times among individuals. For the data recording, they wore jackets with prominent numbers on them. The participants were strangers to each other and had no prior acquaintance before the experiment.

Measurement

The participants measured their perceived waiting times (the waiting time was the interval beginning when the participant pressed the call-button in the lobby, and ending when the elevator arrived and its doors fully opened, and the transit time was the interval beginning when the participant stood the elevator, and ending when the elevator stopped and its doors fully opened at the destination floor) and their perceived stress during these two waiting situations. Because we continuously measured these perceived waiting times, we employed a prospective paradigm similar to past studies (Lahera, 2022). Such an approach is more accurate than a retrospective paradigm (Brown, 1985) and appropriate for multiple measurements.

We also measured the actual waiting times in seconds, the number of people waiting together, the number of times the elevator was missed during waiting, the number of people riding together, and the number of times the elevator stopped during transit, using recorded videos. The perceived stress was evaluated in an 11-point response format where 0 denotes no stress, and 10 is extreme stress. We employed an 11-point response format because a past study argued that using 11 scales closely approximated interval data (Wu & Leung, 2017).

Procedure

All the procedures were approved by the Advanced Telecommunications Research Institute International Review Board Ethics Committee (525H). Written informed consent was obtained from all participants. Multiple participants simultaneously joined the experiment to reproduce various situations in elevators. We capped the number of participants (for any one situation) to five to prevent excessive congestion in the elevator even when all were riding at the same time. All participants first took the elevator on the third floor. They were provided with detailed instructions prior to the experiment; each participant received a printed sheet that illustrated the above timing rules, and a standardised oral briefing. Specifically, they were instructed to perform the following steps in sequence: (1) enter the elevator and select a predetermined floor number as instructed, (2) get off at the designated floor upon arrival, (3) refuse the elevator the next time it arrived (i.e., deliberately skip it), and (4) wait for the subsequent elevator and then enter it, selecting another predetermined floor different from the previous one. In addition. they were not permitted to use timing devices, such as smartphones or watches, nor were they allowed to speak with other participants during the experiment. The primary reason for prohibiting smartphones was to prevent participants from accurately measuring waiting/transit times.

After the briefing, the participant completed one practice ride to a different floor. During this practice, the experimenter verified that the participant recorded the perceived waiting time and the perceived transit time as instructed. Only after successful practice did the experimental trials begin. Each participant completed 48 trials within a single session lasting less than 2 h. We asked the participants to write down their perceived waiting times and stress levels experienced while waiting for and riding the elevator on pre-printed article immediately after each ride. Through the experiment, each participant visited every destination floor the same number of times, but the order of descents varied across participants; the next destination floors were described on the pre-printed article. Within each group, we counterbalanced these orders to minimise potential order effects. We conducted the experiments in six groups, each consisting of five participants, over a period of 3 days, with two groups per day.

Perceived and actual waiting/transit times

We analyzed the relationships between the perceived and actual waiting times for the waiting and transit times¹ . In this analysis, we focused on (1) whether the perceived waiting times were longer/shorter than actual waiting times, and (2) whether the repetition effects occurred in our experiment as reported in a past study (Matthews, 2011) that repeated trials tend to shorten perceived time durations, because the participants completed the elevator-riding task 48 times in this study. Therefore, we compared the average perceived/actual waiting and transit times between the first and last halves of the trials.

Firstly, we conducted repeated two-factor repeated-measures ANOVA (type factor: perceived and actual, and time factor: first and last) for their average waiting time (Fig. 2), and the results showed that a significant main effect for the type factor (F (1, 29) = 56.354, p < 0.001, partial η² = 0.660). There were no significant main effects for the time factor (F (1, 29) = 0.184, p = 0.671, partial η² = 0.006) and the interaction effect (F (1, 29) = 0.001, p = 0.990, partial η² < 0.001).

We also conducted repeated two-factor repeated-measures ANOVA (type factor: perceived and actual, and time factor: first and last) for their average transit time (Fig. 3), and the results showed that a significant main effect for the type factor (F (1, 29) = 51.425, p < 0.001, partial η² = 0.639). There were no significant main effects for the time factor (F (1, 29) = 1.718, p = 0.200, partial η² = 0.056) and the interaction effect (F (1, 29) = 0.556, p <= 0.462, partial η² = 0.019).

In addition, we analyzed the correlation between the times in each situation. We found a statistically significant positive correlation for waiting time (r = 0.510, p = 0.004), but no significant correlation for transit time (r = 0.332, p = 0.073).

These results showed that the perceived waiting times were significantly shorter than the actual waiting times under our experimental conditions, and no evidence of repetition effects on estimated times within our experimental setting.

Relationships between waiting/transit times and perceived stress

We analyzed the relationships between waiting times and perceived stress with SPSS and performed a linear mixed-effects analysis of the relationship between them² .

First, we analyzed the relationships between waiting times and perceived stress (Figs. 4 and 5). The model included fixed effects for the number of people waiting together, the number of times the elevator was missed during waiting, and the perceived waiting time or actual waiting time, with repeated measures for trials and random effects for participants. The results indicated no significant effects for the number of people waiting together or the number of times the elevator was missed during waiting. Therefore, we used a model that only included fixed effects for the perceived waiting time or actual waiting time, with repeated measures for trials and random effects for participants. The model was estimated using the Akaike Information Criterion (AIC). We compared the model fit indices between the perceived waiting time and the actual waiting time; the AIC showed that the model with the perceived waiting time showed a better fit (the perceived waiting time: AIC = 5,490.294, the actual waiting time: AIC = 5,767.842). A significant effect of the fixed effect for the perceived waiting time was also found (F (1, 27.689) = 164.372, p < 0.001), although there was no significant correlation between the perceived waiting time and the perceived stress (r = 0.180, p = 0.342).

Overall, the estimated model of the perceived stress based on the perceived waiting time can be described as

(1) $$PerceivedStressOfWaitingTime=1.87+0.06\ast PerceivedWaitingTime.$$

We also analyzed the relationships between the transit times and the perceived stress (Figs. 6 and 7) using the same analysis procedures with a linear mixed-effects analysis. The model included fixed effects for the number of people riding together, the number of times the elevator stopped during transit, and the perceived transit time or actual transit time, with repeated measures for trials and random effects for participants. The results showed that the number of times the elevator was missed during waiting was not significant for either perceived or actual transit times. After excluding this non-significant variable, we compared the model fit indices for perceived vs. actual transit times by using the AIC. The AIC showed that the model with the perceived transit time showed a better fit (the perceived transit time: AIC = 5,252.486, the actual transit time: AIC = 5,610.283). Significant effects were found for the perceived transit time (F (1, 23.947) = 62.579, p < 0.001) and the number of people riding together (F (1, 6,043.062) = 13.039, p < 0.001), although there was no significant correlation between the perceived transit time and the perceived stress (r = −0.080, p = 0.673). Overall, the estimated model of the perceived stress based on the perceived transit time and the number of people riding together can be described as

(2) $$\begin{array}{rl}PerceivedStressOfTransitTime=& 0.45+0.10\ast PerceivedTransitTime\\ & +0.22\ast NumOfPassenger\end{array}.$$

Perceived waiting times were shorter than actual waiting times

Unlike past studies (Fan, Guthrie & Levinson, 2016), the perceived waiting times were significantly shorter than the actual waiting times in each waiting situation (elevator’s arrival and riding it) under our specific experimental conditions. A possible reason might be that we asked the participants to record their perceived waiting times and perceived stress only in each waiting situation, i.e., situations with relatively small cognitive loads. Perhaps such simple tasks enabled them to focus on temporal measurements. Moreover, we employed a prospective paradigm to measure perceived waiting times; a past study reported that participants felt shorter durations under small cognitive loads, and the prospective paradigm provided shorter time lengths than the retrospective paradigm (Brown, 1985).

One possible explanation for the shortened perceived time is the participants’ attention directed toward the passage of time. Previous studies have indicated that reduced attention to the passage of time can lead to shorter or inaccurate estimations (Brown, 1985; Sawyer, Meyers & Huser, 1994). However, participants were not permitted to use timing devices, such as smartphones or watches, nor were they allowed to engage in conversations with other participants during the experiment. Indeed, our video analysis confirmed the absence of conversational interactions. Thus, we assume that participants’ attention was adequately focused on the experimental task itself.

Perceived stress estimation

Our linear mixed-effects analysis showed the effectiveness of using perceived waiting times to estimate perceived stresses compared to actual waiting times. This finding is consistent with previous studies that reported relationships between perceiving waiting time and stress (Abdul Hamid, 2021; Ayodeji, Rjoub & Özgit, 2023; Bielen & Demoulin, 2007; Pruyn & Smidts, 1998). Another study reported that passengers preferred a longer transit time than a longer waiting time (Bird et al., 2016); our equations also support this phenomenon.

Equation (1) for perceived stress while waiting indicated that the stress value approaches the middle value (five) after waiting for 1 min. This suggests that a majority of passengers feel stress when the waiting time exceeds 1 min. In a web survey of 400 people by a Japanese watch company (CITIZEN, 2023), 66.5% of the respondents felt stressed when they had to wait for more than a minute for an elevator, a result that resembles those shown by our model.

In addition, Eq. (2), which describes perceived stress during riding, indicates that stress levels were influenced not only by the transit time but also by the number of passengers. It is plausible that the limited space inside the elevator, compared to the lobby, increased stress due to the proximity of other passengers. Moreover, the coefficient indicating the impact of increased transit time on stress is greater than that for waiting time in Eq. (1), possibly due to the confined, enclosed nature of the elevator environment, which may cause an invasion of personal space (Hall, 1966) by close physical proximity to other passengers.

As an additional analysis, we examine whether repetition effects occurred in our experiment in the context of perceived stress, similar to the perceived/actual waiting and transit times analysis. For this purpose, we also compared the average perceived stress levels between the first and last halves of the trials using paired t-tests. The results revealed no significant differences in the perceived stress during waiting (t(29) = 0.364, p = 0.719, d = 0.066) and the perceived stress during riding (t(29) = 0.670, p = 0.508, d = 0.122). Thus, we found no evidence of repetition effects on perceived stress levels.

Reducing stress in elevator riding

From the perspective of reducing the stress of those taking elevators, an engineering approach, which improves the movement efficiency of elevators, i.e., shortening objective waiting times, can indeed be effective; however, a cognitive approach that reduces the perceived waiting times may be even more beneficial. For example, the installation of mirrors or televisions in elevator lobbies, as reported in past studies (Pruyn & Smidts, 1998; Sasser et al., 1978), are one cognitive strategy to reduce perceived waiting times. Furthermore, since waiting-time satisfaction has been shown to mitigate stress (Bielen & Demoulin, 2007; Nah, 2004), providing information to passengers waiting for an elevator by devices represents another useful method to enhance satisfaction during waiting periods. From this perspective, both equations are useful for managing the appropriate contents for passengers. For example, an elevator system can identify overly frustrated passengers by tracking them, measuring their waiting times, and changing their information contents depending on such passenger characteristics as age, gender, and demeanor. Such an approach, which resembles recent advertisement systems to attract them (Alhalabi et al., 2021; Sewmina, Navarathna & Senanayake, 2023), might decrease their perceived waiting times.

Limitations

This study has several limitations. First, the experiment has a simple setting in a building with just four stories. Hence our experiment settings cannot cover situations that require longer waiting times, such as in high-rise buildings. The elevator’s size is also fixed, and the experiment was only conducted in Japan. Additional investigations on different kinds of elevators and cultures are essential to generalize the knowledge from this experiment.

We did not confirm the participants’ abilities in time perception in advance; therefore, the experiment results might have been influenced by their characteristics. In addition, we did not explicitly prohibit counting time on their heads. Therefore, it is possible that silent counting influenced their time estimates. However, at the same time, participants had to note their perceived time and stress on article, and monitor the floor on which to disembark, which may have implicitly reduced the likelihood of sustained counting. Finally, no a priori power analysis was conducted to justify the sample size; we acknowledge this explicitly for transparency (Lakens, 2022; Sasaki & Yamada, 2023). We likewise refrain from including a post‑hoc power analysis because of its well‑documented limitations (Hoenig & Heisey, 2001). Future studies that incorporate pre‑registered power calculation (e.g., using G*power (Faul et al., 2007)) and larger samples will allow more rigorous statistical interpretation.