3. Method
3.4 Experimental study
3.4.2 Participants
Participants were recruited via the participant database Prolific (www.prolific.co). A total number of 40 participants were recruited via the database. The number of participants was
32 effect of f = 0.25. While the effects of naturalness and complexity on fascination and perceived restoration in the literature are found to be medium-to-large (Berman et al., 2014; Ommeren, 2019;
Twedt et al., 2019), in the present study the manipulation is expected to be more subtle, with expected effects smaller than the f = 0.35 found in Van Ommeren (2019) which manipulated both the color and the design of the visual stimulus. Six additional participants were recruited for pilot testing and correcting for potential drop out.
To be eligible to participate in this study, participants were required to be fluent in English and have a normal or corrected-to-normal vision. Therefore, these criteria were used as inclusion criteria in Prolific. Participants were asked to conduct the experiment on a laptop or desktop, not on a mobile device, so the full image was visible on the screen. Each experimental session lasted around 15 minutes and participants were compensated £1.90 for their time.
Forty-one responses were submitted in Limesurvey. Responses with a response time below 5 minutes or above 45 minutes were rejected. One response was rejected by Proflic due to the long response time but still submitted in Limesurvey after 2.5 hours. Because of the time limit, this observation was not included in the dataset. The remaining 40 participants (24 male, 15 female) included in the dataset had an average age of 25.5 (SD = 6.90, Min = 18, Max = 45). Participants were recruited from all over the world. Figure 29 shows the country of residence and the country where participants spent most of their lifetime. There were small differences between the country of residence and the country where participants spend most of their lifetime.
Figure 29. Geographical data participants
3.4.3 Stimuli
The patterns manipulated on complexity and naturalness were used as a façade geometry in a virtual room. The simulation was created from real-world meeting room 2.422 in the Atlas building on the campus of Eindhoven University of Technology (Figure 30). The dimensions of the room were
33 4.10 x 3.64 x 2.60 meters, and it has a floor-to-ceiling window with dimensions 3.64 x 2.60 meters.
The three window panels were considered as one single window. This room was recreated in Rhinoceros 6 with the use of the building’s floor plans. The office space was simplified by creating a flat ceiling and no interior lighting was added. The furniture placed in the room was chosen to look similar to the original office space but differs slightly in look and dimensions. An illustration of the simulated office environment is shown in Figure 31. The experimental stimuli, as presented in Figure 27, were in the dimensions of 260 x 364 mm, and were magnified with a factor of 10 to fit the target widow size. The added façade geometry had a thickness of 1 mm.
Figure 30. Real-world example office space Figure 31. Simulated office space
For daylight simulations, DIVA-for-Rhino 4.0 was used. Most of the materials used for this simulation were the default DIVA-for-Rhino generic materials with no roughness or specularity. The material properties for the main materials are shown in Table 6. To have daylight patterns on the walls and furniture a clear sky with sun was chosen. The windows were facing east, the same orientation as the office space in the Atlas building. The first of March at 09:00 a.m. was chosen as the day and time for the daylight simulations. This particular day and time were chosen because the simulation resulted in attractive light patterns on the wall and furniture in addition to the pattern of façade design. Daylight conditions were kept constant for all variations of façade geometry. The viewpoint of the camera was placed at a height of 1.40 m on the place where the door is normally placed. The wall not visible in the image is a closed wall so no light was interfering with the daylight entering from the façade.
Table 6. Material properties
Surface Type R G B Reflectance Specularity Visual
transmissivity
Ceiling Plastic 0.8 0.8 0.8 80 % 0 .
Floor Plastic 0.2 0.2 0.2 20 % 0 .
Walls Plastic 0.7 0.7 0.7 70% 0 .
Table Plastic 0.5 0.5 0.5 50% 0 .
Chairs Plastic 0.21 0.34 0.49 32 % 0 .
Façades Plastic 0.01 0.01 0.01 1 % 0 .
Windows Glass 0.87 0.87 0.87 . . 87 %
34 renderings were: ps 8 pt .15 pj .6 dj 0 ds .5 dt .5 dc .25 dr 0 dp 128 st .85 ab 3 aa .25 ar 128 -ad 2048 -as 1024 -lr 6 -lw .01. The resolution of the renderings was set on 800 x 600 pixels since in Limesurvey the image would be presented in similar dimensions. The daylight renderings were exported in a png format. An example of the daylight simulation with the highly complex (C4) and highly natural (N4) façade design is shown in Figure 32. All daylight renderings with variations in façade design are shown in Figure 33.
Figure 32. Example of daylight simulation with a highly complex (C4) and highly natural (N4) façade design
35
N4N3N2N1 C1 C2 C3 C4
Figure 33. Daylight simulations
36 The dependent variable “fascination” was measured with the three items from the subscale fascination of the short version of the Perceived Restorativeness Scale (Pasini, Berto, Brondino, Hall,
& Ortner, 2014). This short version contains 11 items from the renewed PRS and is based on ART and therefore consists of four sub-scales for fascination, extent, being away and compatibility. The subscale for fascination consists of three items: “This place has fascinating qualities” and “My attention was drawn to many interesting things” and “This place is boring”. Participants gave their answers on an 11-point Likert scale (0 = not at all; 10 = very much).
The dependent variable “perceived restoration” was measured on a scale similar to Twedt et al. (2019). Participants were instructed to think back to a time when they felt overwhelmed or stressed while images of the daylit indoor office environment were shown. For each of the presented images, participants were asked to rate to what extent this place would be a good place to (1) take a break and make you less stressed. Two questions for perceived restoration from Pals, Steg, Dontje, Siero, & van der Zee (2014) were added and rephrased as the previous question: “to what extent is this environment a good place to (2) relax and (3) renew your energy?”. Answers were given on a scale from 0 to 10 (0 = not at all; 10 = very much).
Similar to the pilot studies perceived naturalness and perceived complexity were added to check the manipulation for naturalness and complexity in the façade geometry. Perceived naturalness was measured by adding the question “How much does this façade design remind you of nature?”. For perceived complexity the question “How complex is this façade design?” was included.
Both questions are answered on an 11-point scale. The scale for the experimental study was different than the scale used in the pilot studies because with an uneven number in the scale there is a clear middle point. Last, the pleasantness of the scene was measured with the question “How pleasant would you rate this scene?“ similar to Chamilothori et al. (2019). This concept was measured on an 11-point scale as well (0 = not at all; 10 = very much).
The demographic questions that were added in this study were age, gender, country of residence and country where participants had lived most of life. This particular measure was added since nature differs around the world. All questions with their answer options that were included in the survey are shown in Table 7.
Table 7. Measurements
Concept Question Answer
Fascination This place has fascinating qualities 0 – 10 Likert scale My attention is drawn to many interesting things 0 – 10 Likert scale
This place is boring 0 – 10 Likert scale
Perceived complexity How complex is this façade design? 0 – 10 Likert scale
37 Perceived naturalness How much does this façade design remind you of
nature?”.
0 – 10 Likert scale Pleasantness How pleasant would you rate this scene? 0 – 10 Likert scale
Demographics What is your gender? Choice: male, female,
don’t want to say
What is your age? Open question
What is your country of residence? Drop-down menu In which country did you spend most of your
lifetime?
Drop-down menu
3.4.5 Procedure
Participants were instructed to open the link on a desktop or laptop to ensure that the full image was visible. On the first page of the survey, a consent form was presented explaining what was expected from the participants. By proceeding to the next page participants gave their consent. A copy of the consent form can be found in Appendix 7.2. The daylight images with variations in façade geometry were presented one at a time in random order. In the first round, questions about perceived restoration and fascination were presented under the image. After the first set of sixteen images, a page was shown indicating the end of the first round. In the second round, questions about perceived naturalness, complexity and pleasantness were presented under the image. Within the rounds, the questions were randomized and backward navigation was not possible. After the participants saw the set of sixteen images twice the participants answered a few demographic questions. After the participants finished and submitted the questionnaire participants received a link for a compensation of £1.90 for their time. An illustration of the procedure with the different parts of the experimental study is presented in Figure 34. Each experimental session lasted around 15 minutes.
Figure 34. Schematic overview procedure
3.4.6 Statistical Analyses
Data processing and analyses were conducted in STATA 14. The data was checked for skewness and kurtosis and tested for normal distribution with the Shapiro-Wilk test. The dependent variables fascination and perceived restoration were normally distributed in all groups, as well as the data for pleasantness. For perceived naturalness and perceived complexity, some groups showed a non-normal distribution indicated by a rejected Shapiro-Wilk test. For each of these groups, the data approximated a normal distribution. Additionally, since sixteen tests were conducted for each variable, significance levels should be corrected for the number of tests.
Consent
38
= 2.36, p = .003) and perceived complexity (F(15,624) = 2.42, p = .001) the null hypothesis, the assumption of homogeneity of variance in the different groups, was rejected. For fascination (F(15,624) = 0.520, p = 0.933), perceived restoration (F(15,624) = 0.897, p = .567) and pleasantness (F(15,624) = 1.27, p = .215) the test was accepted, meaning the variance for these variables is equal for each group.
For the three items for perceived restoration and the three items for fascination, a Cronbach’s alpha was calculated. Both scales were identified as strong with a Cronbach’s alpha of respectively 0.95 and 0.82. Therefore all three items were included in one construct for fascination and one for perceived restoration.
Linear Mixed Model (LMM) analyses were performed to investigate the effect of manipulated naturalness and manipulated complexity on perceived naturalness, perceived complexity, pleasantness, fascination, and perceived restoration (separate LMM analyses were run for each dependent variable). In these analyses, participant was added as a random intercept to correct for the grouped data on participant level. The results from the tests of normality and homogeneity of variance for perceived naturalness and perceived complexity showed that using a Linear Mixed Model should be approached with caution. The significance level for these tests was corrected for the number of hypotheses tested: the effect of manipulated naturalness and manipulated complexity.
Therefore the significance level was set using a Bonferroni corrected alpha level of α = .05 / 2 = .025.
Before the LMM analyses were run the Variance Inflation Factor was calculated to test for multicollinearity. The results showed there was no mean VIF higher than 2.5 and no individual VIF was higher than 10, meaning no multicollinearity was found. In case of a significant effect, post-hoc pairwise comparisons were done with the use of Estimated Marginal Means (EMM), to explore differences between groups. A Bonferroni-corrected alpha level was also used to correct for the six comparisons made α =.05 / 6 = .008.
For secondary objectives in this thesis, LMM analyses were performed to explore the effect of perceived naturalness, perceived complexity, and pleasantness on fascination and perceived restoration, with separate LMM analyses for each dependent variable. The significance level was again corrected with the number of predictors α = .05 / 3 = .025. Lastly, LMM analyses were performed to investigate the effect of the manipulation of naturalness and complexity and perceived naturalness, perceived complexity, and pleasantness on fascination and perceived restoration.
Significance levels for these analyses were set on α = .05 / 5 = .01. For both analyses participant was added as random intercept to correct for the multiple measurements per participant.
39
4. Results
In this section, the results of the statistical analyses for different parts of the conceptual model are explained, starting with the effect of manipulated naturalness and manipulated complexity on perceived naturalness, perceived complexity, and pleasantness. In the sections that follow, fascination is used as the dependent variable, first the effect of manipulated naturalness and manipulated complexity on fascination is tested, followed by the effect of pleasantness, perceived naturalness and perceived complexity on fascination. Finally, the independent variables, as well as pleasantness, perceived naturalness, and perceived complexity were included in the analysis to explore the effect of fascination. In the last sections, similar analyses were reported but with perceived restoration as predictor instead of fascination.
4.1 Effects of manipulated naturalness and complexity on perceived naturalness, perceived complexity, and pleasantness
In this first section, the effect of manipulated naturalness and manipulated complexity is tested against perceived naturalness, perceived complexity, and pleasantness. Figure 35 shows the part of the conceptual model that was tested in the following subsections including the standardized beta coefficients and significance.
Figure 35. Part of the conceptual model with standardized beta coefficients and significance outcomes.
4.1.1 Perceived naturalness
To ensure that the manipulation was successful for naturalness, the effect of the manipulation was tested against perceived naturalness. A Linear Mixed Model with manipulated naturalness and manipulated complexity as fixed effects, showed an effect of both manipulated naturalness (βnat = .544, SE = .08, z = 6.62, p < .001) and manipulated complexity (βcom = 1.03, SE = .08, z = 12.55, p < .001) on perceived naturalness. The graphs in Figure 36 and 37 illustrate the effect of both manipulated naturalness and manipulated complexity on perceived naturalness.
40
Figure 36. Mean ratings for evaluations of fascination for each façade design ordered by levels of manipulated naturalness (left) or by levels of manipulated complexity (right).
Post-hoc tests were conducted to investigate the contrasts between individual levels of manipulated naturalness and manipulated complexity on perceived naturalness. A Bonferroni-corrected alpha was used of α = .008. Figure 37 shows the comparisons of Estimated Marginal Means (EMM) for both manipulated naturalness and manipulated complexity on perceived naturalness.
Contrasts of Estimated Marginal Means for both manipulated naturalness levels and manipulated complexity levels on perceived naturalness and their significance are shown in Table 8. The comparisons for the naturalness manipulation showed that the lowest level of manipulated naturalness, N1, was significantly different from all other levels. Significant differences in perceived naturalness were also found between the levels N2 and N4 and between N3 and N4. The middle levels of manipulated naturalness N2 and N3 were not significantly different.
For the complexity manipulation, the pairwise comparisons showed that perceived naturalness for the lowest manipulated complexity level, C1, was significantly different from the other levels of manipulated complexity. A significant difference in perceived naturalness was found when C2 was compared with C3, as well as C2 and C4. No significant difference in the perceived restoration scores was found between C3 and C4. The contrasts between the EMM can be found in Table 8. These results show that both naturalness and complexity had an effect on perceived naturalness. This main effect of complexity showed that for a façade to be perceived as natural, it should hold a certain level of complexity.
Figure 37. Pair-wise comparisons of perceived naturalness ratings across manipulated naturalness levels (left) and manipulated complexity levels (right). Significance levels are marked as follow: * p = 0.05, ** p = 0.01, *** p = 0.001.
41
Table 8. Estimated Marginal Means (EMM) and Standard Errors (SE) of perceived naturalness ratings per naturalness level and per complexity level, and contrasts of EMM between complexity levels and between naturalness levels.
EMM (SE) N1 N2 N3 N4
* the difference is significant at a corrected significance level of α = .008
4.1.2 Perceived complexity
A Linear Mixed Model with manipulated complexity and manipulated naturalness as fixed effects showed an effect of both manipulated complexity (βcom = 1.91, SE = .06, z = 31.11, p < .001) and manipulated naturalness (βnat = .315, SE = .06, z = 5.13, p < .001) on perceived restoration. Figure 38 illustrates the main effects of complexity and naturalness, showing a strong manipulation of complexity.
Figure 38. Mean ratings for evaluations of perceived complexity for each façade design ordered by levels of manipulated naturalness (left) or by levels of manipulated complexity (right).
Post-hoc pairwise comparisons were conducted comparing EMM. Bonferroni-corrected alpha level of α = .008 was used as the significance level. The results of these comparisons are shown in Table 9. All levels of the manipulated complexity resulted in significantly different ratings of perceived complexity, illustrated in Figure 39 (left). Results from pairwise comparisons of both manipulated complexity as well as manipulated naturalness are shown in Table 9. The pair-wise comparisons of the levels of manipulated naturalness on ratings of perceived complexity showed that N1 differed significantly in perceived complexity from N4, but there was no difference found between N1 and N2, and N1 and N3. The patterns high in naturalness, N4, were found to differ significantly in perceived complexity from N2, but not from N3. These results suggest that there is a difference in perceived complexity between the low (N1 and N2) and the highest level (N4) of manipulated naturalness, illustrated in Figure 38 (right).
42 manipulation of the façade geometry on the concepts of naturalness and complexity was not entirely independent. Perceived naturalness and perceived complexity both were affected by the manipulation done for naturalness and complexity.
Figure 39. Pair-wise comparisons of perceived complexity ratings across manipulated complexity levels (left) and manipulated naturalness levels (right). Significance levels are marked as follow: * p = 0.05, ** p = 0.01, *** p = 0.001
Table 9. Estimated Marginal Means (EMM) and Standard Errors (SE) of perceived complexity ratings per naturalness level and per complexity level, and contrasts of EMM between complexity levels and between naturalness levels.
EMM (SE) C1 C2 C3 C4
C1 2.51 (.25) .
C2 4.37 (.25) 1.86* .
C3 5.91 (.25) 3.40* 1.54* .
C4 7.68 (.25) 5.17* 3.31* 1.77* .
EMM (SE) N1 N2 N3 N4
N1 4.78 (.25) .
N2 4.87 (.25) .09 .
N3 5.23 (.25) .44 .36 .
N4 5.60 (.25) .82* .73* .38 .
* the difference is significant at a corrected significance level of α = .008
4.1.3 Pleasantness
The effect of manipulated naturalness and manipulated complexity on pleasantness was tested with a Linear Mixed Model with manipulated naturalness and manipulated complexity as fixed effects. The scores for pleasantness for each façade are shown in Figure 40. No significant effect was found for either manipulated naturalness (βnat = .057, SE = .08, z = 0.70, p = .484) or manipulated complexity (βcom = -.118, SE = .08, z = -1.46, p = .144) on ratings of pleasantness, which is illustrated in Figure 41 as well.
43
Figure 40. Mean ratings for evaluations of pleasantness for each façade design ordered by levels of manipulated naturalness (left) or by levels of manipulated complexity (right).
Figure 41. Mean ratings for evaluations of pleasantness across levels of manipulated naturalness (left) and manipulated complexity (right). Error bars represent standard errors.
4.2 Predicting fascination
4.2.1 Effect of manipulated naturalness and complexity on fascination
In this section, the effect of manipulated naturalness and manipulated complexity on fascination is explored, as illustrated in Figure 42. The fascination scores of manipulated naturalness and manipulated complexity are illustrated in Figure 43 and Figure 44. Figure 43 illustrates the sixteen different façade geometries grouped by the levels of naturalness and by the levels of complexity. The left graph suggests there is an effect of manipulated complexity across naturalness levels on perceived restoration. An LMM analysis with manipulated naturalness and manipulated complexity as fixed effects showed that manipulated complexity significantly affected fascination (βcom = .393, SE
= .05, z = 7.22, p < .001), as was suggested from Figure 43. The effect of manipulated naturalness on fascination was not significant (βnat = .108, SE = .05, z = 1.98, p = .048). This effect of manipulated complexity on fascination supports our hypothesis (H4). However, the absence of a significant effect of manipulated naturalness on fascination rejects our hypothesis (H3). The effect of manipulated complexity on fascination was positive, meaning that high manipulated complexity in façade designs resulted in higher fascination scores, as shown in Figure 44.
44
Figure 42. Part of the conceptual model with standardized beta coefficients and significance outcomes.
Figure 43. Mean ratings for evaluations of fascination for each façade design ordered by levels of manipulated naturalness (left) or by levels of manipulated complexity (right).
Figure 44. Mean ratings for evaluations of fascination across levels of manipulated naturalness (left) and manipulated
Figure 44. Mean ratings for evaluations of fascination across levels of manipulated naturalness (left) and manipulated