Comparative analysis of emotional impression evaluations of rooms with different kinds of windows between scale-model and real-scale virtual conditions Kodai Ito a, Wataru Morishita b, Yuri Nakagawa a, Yae Aoyagi a, Yoko Watanabe b, Michiko Ohkura b a Graduate School of Science and Engineering, Shibaura Institute of Technology, Tokyo, JAPAN; b College of Engineering, Shibaura Institute of Technology, Tokyo, JAPAN The emotional impressions of a residential space s comfort and cleanness are critical for its residents. Although it is sometimes difficult to maintain good emotional conditions in the inner-city areas of huge city like Tokyo, appropriate windows might raise the impression of a residential space. To clarify the effect of the sizes and positions of living room windows, we designed and performed two experiments using scale-model and real-scale virtual environments. This paper describes these experiments and their analysis results using multivariate analysis and data mining techniques. Keywords: Emotional impression, Openness, Scale-model, Virtual environment, Multivariate analysis 1. Introduction For the last 15 years, narrow houses is one phenomenon that symbolizes Japanese urban housing. Narrow sites are often cheaper because of their obvious topographical disadvantages. In inner-city areas where the price of land is very high, narrow-site houses are attracting much attention for building houses at low cost. High-density residential environments have been created by rapid urban development that made many narrow sites and small houses. For planning narrow-site houses, one crucial goal is the "habitability" of such tight, cramped site conditions. Such emotional impressions as comfort and cleanness of a residential space are crucial for its residents. Although it is sometimes difficult to maintain good emotional conditions in the inner-city areas of huge cities like Tokyo, appropriate sized and situated windows might raise the space s impressions. In previous research, windows have the greatest impact on the openness of living spaces and that such openness is determined by both the size and view of a window (Inui et al., 1972a)(Inui et al., 1972b)(Ohkura et al., 2009). This study clarifies the impressions of living room windows in a narrow-site house. The relationship among view from window, size, and position influence the living space. Open views from windows give a comfortable feeling; windows sometimes cause negative emotions because through them others can look inside. The effects of windows vary, depending on their arrangements and peripheral factors. Therefore, to clarify the effect of living room windows, we designed and performed two experiments using scale-model and real-scale virtual environments. Scale models are often used to show the appearances of buildings. On the other hand, real-scale virtual environments are also used recently to help design yet to be built buildings and to reproduce buildings of the past (Ikeuchi et al., 2004), because of less cost and less time consuming compared with real-scale model. However, little has been done to record impressions of people in spaces of buildings created in a virtual environment. Yoshizawa et al. conducted a study in which brightness perception, space perception, and length perception of individual rooms were examined through the comparative experiments between virtual and real spaces (Yoshizawa et al., 2001). S. Seta et al. observed the emotional change induced by urban space using a system with video and computer graphic images of actual urban areas (Seta et al., 2004). Although, the examined impressions in the former study (Yoshizawa et al., 2001) were restricted only to physically sensory factors such as brightness and length, emotional factors were not examined, such as desirability of, and familiarity with the space. Concerning to the latter study (Seta et al., 2004), sensory and emotional impressions beyond sight were not recorded, because the experiment was narrowed to only video and computer graphics images displayed on a 29-inch flat display. Moreover, all comparisons in both studies were made only between virtual and real spaces, and there were no comparisons between spaces in virtual environment and between spaces in real environment. So, it remains unclear, within the context of a comparative experiment, how to switch from the real environment 1
Proceedings 19th Triennial Congress of the IEA, Melbourne 9-14 August 2015 to the virtual environment. Consequently, we designed an experimental system under the following conditions to perform the comparative experiments (Ohkura et al., 2005): - Use a large-sized display in virtual environment. - Build interior spaces identical in appearance in virtual and real environments, and record participants physical and emotional impressions of the respective spaces. We designed new experiments mainly based on this research. This paper describes our new experiments using scale-model and virtual conditions to compare the impressions of windows in a narrowsite house and their analysis results using multivariate analysis and data mining techniques. 2. Method We constructed our 1/6-scale model by cardboard for the walls and acrylic plates for the windows. The model s inside is covered with sheets of paper with a concrete texture because To-no-ie s walls are exposed concrete. The east-side wall is changeable to modify the openness of the living room s east side. The participants were imagined to be sitting on a living room sofa where their eye positions are 800 mm from the floor. The experimental scene for the scale-model is shown in Figure 1. The equipment for the real-scale virtual condition is shown in Figure 2 (a). A 3D projector and LCD shutter glasses were employed. The experimental scene for the virtual condition is shown in Figure 2 (b). Tou-no-ie (tower house in Japanese) is one of the most famous narrow-site houses in Japan. It has five stories and its bottom floor is only 20 square meters (Azuma et al., 1988). Based on this house, we built a 1/6-scale model of a living room with changeable walls with different windows to compare the effects of openness on such emotional impression factors as oppression and calm. The windows had different positions and sizes to produce different room impressions. At the same time, we prepared real-scale rooms in a virtual environment to compare the impressions using the same impression words as in the scale-model condition. Figure 3 shows five types of windows: 0: original (upper), 1: no window, 2: lower, 3: larger, and 4: half-transparent. Buildings and the sky can be seen through these windows. We investigated the emotional impression factors: wideness, oppression, calm, and openness. The impressions of the scale-model and real-scale virtual conditions were evaluated by the Magnitude Estimation Method (Stevens, 1957) where the scores of the original were set to 100. If a participant felt less wideness for the living room without windows, he gives a score below 100. The following is the experiment s procedure: 1. The experiment s outline was explained. 2. The original condition s room was displayed for 20 seconds. 3. A room with one of the other three window types was displayed for 20 seconds. 4. The participants scored the windows based on the four emotional impression factors. 5. Steps 3 and 4 were repeated three times. Wall Figure 1. Experimental scenes of 1/6-scale model condition with changeable wall. 2
Proceedings 19th Triennial Congress of the IEA, Melbourne 9-14 August 2015 Wall Glasses Projector (a) Equipment for real-scale virtual condition (b) Experimental scene of virtual condition. Figure 2. Equipment for real-scale virtual and experimental scene conditions. Openness Original (upper) 1 No window 2 Lower 3 Larger 4 Half- transparent Figure 3. 1/6 Scale- model condition Real- scale virtual condition Wall view Five types of windows in 1/6-scale model and real-scale virtual conditions from sofa viewpoint. 3
3. Results and discussion 3.1 Experimental data preparation We performed our experiments with 49 (35 males and 14 females in their 10 s to 40 s) participants for the scale-model condition and 168 (118 males, 48 females, and 4 no-answer in their 10 s to 50 s) for the virtual condition. The experimental data were used for further analysis including multivariate analysis and data mining. Part of the analysis results of the scale-model condition was described in another paper (Nakagawa et al., 2015). This paper describes additional analysis results that clarified the window effects on the impressions and the similarities and the differences of the impressions from the rooms in both conditions. Because the range of the scores varied very widely among individual participants, the scores were normalized using the four scores of each participant by the formula below: Normalized score = (Score averaged score) / Standard deviation of scores. 3.2 Comparison of impressions Figure 4 shows the normalized averaged scores of the four impression factors for the four window-types and the two conditions. Table 1 shows the results of the difference tests. Tables 2 and 3 show the correlation coefficients among the impression factors. We obtained the following results: - The impressions of wideness, oppression, and openness differed among window types. The differences of calm among window types were small. - The impressions of the half-transparent windows resemble the original window (upper window) - The impressions of the scale-model and real-scale virtual conditions were closely resembled. - Some statistically significant differences exist between the impression scores of the two conditions. The absolute values of the impression scores are higher in the scale-model condition than in the virtual conditions in most cases. - Oppression and openness are strongly negatively correlated in both conditions. - Calm is negatively correlated with wideness and openness in both conditions, which means openness does not elicit a calm state of mind for narrow-site houses in big cities. These results closely resemble our previous research results (Ohkura et al., 2005). Figure 4. Averaged scores of four impression factors for four window types and two conditions (HT: Half transparent). Table 1. Results of difference tests between two conditions. Wideness Oppression Calm Openness No window * Lower ** ** Larger ** ** Half transparent * : p<.05 ** : p<.01 4
Table 2. Correlation coefficients among impression factors in scale-model condition. Wideness Oppression Calm Openness Wideness Oppression -.471** Calm -.450** -.092 Openness.163* -.706** -.387** * : p<.05 ** : p<.01 Table 3. Correlation coefficients among impression factors in VR condition. Wideness Oppression Calm Openness Wideness Oppression -.366** Calm -.402** -.276* Openness.026 -.667** -.252** * : p<.05 ** : p<.01 3.3 Principal component analysis results After removing the insufficient data of of the participants in their 30 s and over, we analyzed the data by principal component analysis and show the results in Table 4. The first principal component is related to openness and wideness; the second principal component is related to calm. Figure 5 shows the plots of the scores of the principal components for four window types and two experimental conditions, where the horizontal and vertical axes are the first and second principal component scores. We obtained the following from Figure 5: - Differences among window types are shown for both experimental conditions from the first principal component related to openness and wideness. The scores for both experimental conditions are also similar. - Differences among window types are shown only for the scale-model condition from the second principal component related to calm. The scores are mostly similar except the no-window types for the virtual condition. From these results, the scale-model condition clarified the differences among the window types more strongly than the real-scale virtual condition, which resembles the results of the previous section. Table 4. Load of principal component analysis. First principal component Second principal component Wideness.608 -.488 Oppression -.836 -.506 Calm -.339.872 Openness.819.207 Eigen value 1.853 1.297 Contribution ratio 46.3 32.4 Accumulation contribution ratio 46.3 78.8 5
Figure 5. Plots of scores of principal components. 3.4 Results of analysis using decision tree To clarify the degrees of the importance of such various factors as window types, experimental conditions, gender, participant ages, and impression factors, we performed analysis using decision trees. Figure 6 shows the following results: Node Impression factor Node Node Window type Window type Node Node Node Node Figure 6. Decision tree results. 6
- The most important aspects are the impression factors. Node 1 is the wideness and openness group. Node 2 is the group of the other two impression factors. - The second important aspects are window types. Nodes 3 and 5 are the groups of the lower- and larger-window types. From these results, the difference between the scale model and the virtual condition is considered relatively smaller than the impression factors and the window types. Such differences among participants as gender and age are also considered relatively small. 4. Conclusions We designed and performed two experiments using scale-model and real-scale virtual environments to clarify the effect of a living room window s sizes and positions on the emotional aspects of residents. The experimental data were analyzed by multivariate analysis and data mining techniques. We obtained the following from the multivariate analysis results and the principal component analysis: - The impressions of wideness, oppression, and openness differed among window types. The differences of calm among window types were small. - The impression of the half-transparent windows resembles the original (upper window). - The impressions in the scale-model condition closely resemble in the real-scale virtual condition, in common. However, we identified some statistically significant differences between the impression scores of the two conditions. The absolute values of the impression scores are higher in the scalemodel condition than in the virtual conditions in most cases. - Calm is negatively correlated with wideness and openness in both conditions, which means that openness does not produce calm states in narrow-site house in big cities. These results are almost the same as our previous research results (Ohkura et al., 2005). The following are the results of the decision tree analysis: - The most important aspect is the impression factors. - The second important aspect is window types. - The difference between the scale-model and virtual conditions is considered relatively small. - The gender and age differences of the participants are also small. Our obtained results should be addressed for our future work using our virtual environment for emotion research on space impressions. Acknowledgements We thank all the participants of the experiments. References Azuma, T., S. Azuma, and R. Azuma. 1988. White paper of To-no-ie. Tokyo: Sumai Library Publishing Company. (in Japanese) Ikeuchi, K. 2004. Digitally archiving cultural heritage. In Proceedings of the Tenth International Conference on Virtual Systems and Multimedia (VSMM2004), 13-17. Inui, M., T. Miyata, K. Watanabe. 1972a. Evaluation of Spaciousness 1. Transactions of the Architectural Institution of Japan, 192: 49-55. (in Japanese). Inui, M., T. Miyata, K. Watanabe. 1972b. Evaluation of Spaciousness 2. Transactions of the Architectural Institution of Japan, 193: 51-57. (in Japanese). Nakagawa, Y., Y. Aoyagi, M. Ohkura, and Y. Ito. 2015. Experiment on the impression of Tou-no-ie by Takamitsu Azuma for the resolution of a narrow site house. In Proceedings of 19 th Triennial Congress of the International Ergonomics Association (IEA2015). (to appear) Ohkura, M., Y. Aoki, and T. Aoto. 2009. Evaluation of Comfortable Spaces for Women using Virtual Environment - Objective Evaluation by Biological Signals-. Kansei Engineering International 8(1): 67-72. Ohkura, M., Y.Komatsu, Y.Shimada, T.Shibata, S.Nakayama, and Y.Watanabe. 2005. Comparison of the impression of the space between virtual environment and real environment. In Proceedings of the 11 th International Conference on Human-Computer Interaction, CD-ROM, Las Vegas. Seta, S., Matsumoto, N., Takagi, K., and Miwa, N. 2004. The influence of the urban space elements upon the estimation of sensitivity analysis Notation of emotional meaning of urban outdoor spaces based on brain waves Part 2-. Journal of Architectural Planning 577: 65-72. (in Japanese) 7
Stevens, S. S. 1957. On the psychophysical law. Psychological Review 64(3):153 181. Yoshizawa, N., Inamoto, J., Hirate, K., Oyama, Y., and Ono, K. 2001. A study on the reproduction of the actual space on the virtual simulation system for the residential environment -The evaluation on brightness, space and length perception and a sense of reality-. Journal of Architectural Planning and Environmental Engineering 550: 87-93. (in Japanese) 8