4.4 Subjective analysis
4.4.5 Principal component analysis
To investigate the influential and distinct subjective characteristics of these sound- scapes, a principal component analysis (PCA) was performed on the responses to the semantic differential scales using a variance maximizing rotation of the original variable space (varimax). The regressed scores from each factor will then be used
in further analysis to help explain and summarise the complex objective interac- tions involved in soundscape perception. Table 4.11 shows the two components extracted with a criterion factor of eigenvalue>1, which account for 77.4% of vari- ance (Component 1 45.8%, Component 2 31.6%, Kaiser-Meyer-Olkin Measure of Sampling Adequacy = .710). Component Number 6 5 4 3 2 1 Eigenvalue 3.0 2.5 2.0 1.5 1.0 0.5 0.0
Figure 4.21.a: Scree plot
Component 1 1.0 0.5 0.0 -0.5 -1.0 Component 2 1.0 0.5 0.0 -0.5 -1.0 tranquil eventful exciting pleasant locquality quality
Figure 4.21.b: Rotated factor plot loadings
Figure 4.21: Scree plot and rotated factor plot for subjective rating variables
Figure 4.21 shows the scree and loading plot for each variable within the ex- tracted components. Component 1 consists of descriptions of “appreciation”, con- taining the variables: soundscape pleasantness, tranquillity, location quality and soundscape quality. Component 2 describes the “dynamics” of a soundscape, made up of ratings of excitement and eventfulness. Whilst all the other variables load very strongly along each component axis, tranquillity shows a strong positive loading on the relaxation component, as well as a comparable negative loading on the dynamics component.
Component 1 Component 2 How pleasant .881 Quality .851 Loc. quality .810 How tranquil .656 −.546 How eventful .892 How exciting .881
Table 4.11: Varimax rotated component matrix for subjective descriptors ( loadings<|0.2|removed)
exhibit distinct regions within the factor space. The exception is the variable tran- quillity, which has a relatively strong negative loading on the dynamics factor. This suggests that tranquil soundscapes are perceived as more relaxing and chaotic soundscapes are seen as being more dynamic in nature. Whilst this may seem obvious, it confirms participants correct interpretation of the studies semantic dif- ferential terms and the robustness of the data gathered. The distinctive loadings on the dynamics component of Exciting and Eventful suggest that these sound- scape components do not play a direct role in the subjective rating of quality.
Research Factor Expl. variance Description
Kang et al. [6] 1 26.0% Relaxation (comfort–discomfort, quiet–noisy, pleasant- unpleasant, natural-artificial, like-dislike, gentle-harsh)
2 12.0% Communication (social-unsocial, meaningful-meaningless, calming-agitating, smooth-rough)
Axelsson et al. [5] 1 49.0% Pleasantness (pleasant, appealing) 2 19.0% Eventfulness (eventful, lively)
Raimbault et al. [153] 1 67.0% Assessment (pleasant–unpleasant) linked to strength (quiet– loud)
2 15.0% Sound dynamics: temporal balance (steady–unsteady), spatial arrangement (organised–disorganised)
Viollon et al. [154] 1 46.6% Affective impressions, preferences (pleasant, comfortable, rural, friendly, silent)
2 18.0% Activity due to sound presence of human beings (bustling, marked by living creatures)
Kawai et al. [155] 1 25.0% Preference (irritating–relieving, unpleasant–pleasant, artificial– natural)
2 16.8% Activity (lively–deserted, joyful–empty, exciting–gloomy)
Table 4.12: First two factors emerging in PCA of soundscapes based on semantic differentials
The PCA outcome shows similarity with findings made by Kang, Axelsson and others detailed in Table 4.12, where the main principal components of re- laxation/pleasantness were extracted (Kang - Relaxation, Axelsson - Pleasant- ness) from semantic differential perceptions of soundscapes. Axelsson described his second component as Eventfulness, closely matching the outcome from the present study, whereas Kang describes his as Communication. It is worth men- tioning however that both of these studies focussed on urban soundscapes and utilised a large number of individual subjective variables. The limitations imposed by the use of mobile devices and unsupervised voluntary participation meant that requesting responses from subjects had to be kept to a minimum. This volun- tary choice available to participants also meant that there was no control over where people may capture these sound environments, meaning that urban and rural soundscapes are analysed as one. A supplementary question asking partic- ipants to label the soundscape as “rural” or “urban” would benefit future studies of this kind.
Another study with comparable results is that of Kawai et al [155]. They used a different methodology using a subjective card sort technique to evaluate the struc- ture that lies at the basis of peoples psychological evaluation of environmental sounds. The components extracted were: Preference and Activity, which corre- spond to the components extracted in the present study. Numerous other studies have also been found that have extracted comparable subjective principal compo- nents [156, 7].
To utilise these extracted components, 40 submissions were then manually selected to represent 4 subgroups of UK soundscapes to compare the subjective factor scores of these differing acoustic environments. This number was selected because of the time taken to manually ascertain if they could be grouped into the
following:
• Urban: inner city locations generally surrounding by large buildings and roads
• Rural: locations surrounded by countryside such as farmland
• Urban public space: inner city spaces, such as squares, plazas and mar- kets
• Urban park:inner city green space generally surrounded by large buildings and roads
These soundscapes factor scores are plotted in Figure 4.22, with the x-axis defined by the component of Appreciation and the y-axis representing Dynamics. The urban soundscapes form a cluster, predominately located on the left of the factor space. With low scores of Appreciation, it is clear that this soundscape group is considered to have an inferior sound environment. The Dynamics scores of the urban type show spread across the x-axis, signifying a lot of variation in the perceived dynamics of these urban soundscapes. Rural soundscapes sit in the opposite half of the factor space to urban group, showing slightly lower dynam- ics scores, but a marked increase in appreciation. The urban public space group have high scores on dynamics and have an even spread of appreciation scores, suggesting that these places are met with mixed emotions, but are generally con- sidered to be exciting and eventful spaces. Surprisingly, urban parks exhibit gen- erally lower perceived scores for dynamics than rural soundscapes. The energy of an urban environment should serve to increase perception of eventfulness and excitement, but this finding is contrary to that assumption. The generally large dy- namic range of rural soundscapes may be perceived as more more exciting and
Component 1 factor score 2 1 0 -1 -2 Componen t 2 factor score 2 1 0 -1 -2 Urban park Urban public space Rural Urban Urban/Rural Appreciation D yn a mi cs
Figure 4.22: Subjective component scores showing: urban, urban public space and rural soundscapes
eventful as the individual sound sources could be more defined against the low background levels.
The manually extracted soundscapes of type: urban, rural, urban public space and urban park, revealed varying and distinct subjective opinions of each. Urban soundscapes were considered to have an inferior sound environment when com- pared to the rural group, with lower appreciation scores in general. The perceived dynamics of urban soundscapes showed a wide range of dynamics ratings, with rural soundscapes being perceived as less dynamic. The urban public spaces scores suggest that these places are met with mixed emotions, but are generally considered to be exciting and eventful spaces. The low scores for urban park dy- namics when compared to rural soundscapes may be due to subjects rating urban parks as relative to the high dynamic nature of an urban soundscape, which you can assume they have just passed through to enter the urban park.
4.4.6
Small group study
The results from the small group study were compared with those from the main study, with Figure 4.23 showing the difference in ratings between the two locations and Table 4.13 detailing these mean values and significance of these differences.
Mean Quality Loc. quality Pleasant Exciting Eventful Tranquil
Urban roadside 3.43 4.14 2.93 4.64 6.50 2.21 Urban park 6.21 7.29 7.00 3.79 3.79 7.29 M-W U test U 20.5 6.5 2.0 69.0 17.5 2.0 Z -3.606 -4.258 -4.468 -1.369 -3.778 -4.480 Asymp. Sig. <.000 <.000 <.000 .171 <.000 <.000
Table 4.13: Mean ratings by location with Mann-Whitney U test
Location Urban park Urban roadside 95% CI 8 6 4 2 Tranquil Eventful Exciting Pleasant Loc. quality Quality
Figure 4.23: Subjective mean scores with 95% CI between locations
ratings and Mann-Whitney U tests are shown in Table 4.14. Mean ratings are provided to give a clearer indication of any differences present between groups.
Mean Quality Loc. quality Pleasant Exciting Eventful Tranquil
Small group 3.43 4.14 2.93 4.64 6.50 2.21 Random sample 3.60 3.50 3.60 4.50 5.90 3.70 M-W U test U 65.0 51.5 61.5 60.5 67.5 35.5 Z -.301 -1.102 -.515 -.562 -.152 -2.091 Exact Sig. .796 .285 .625 .585 .886 .042
Table 4.14: Urban roadside soundscape comparison
With the null hypothesis being that the two groups median values are equal, the clear non-significance of all tests but the rating of tranquillity show that there is significant evidence to accept the null hypothesis. The difference in tranquillity rating could be due to a number of factors, but the low number of subjects involved in the test may have produced this close to non-significant test result. This trend
is also seen in the urban park comparison in Table 4.15, where there is significant evidence to assume equal median values for the two groups. In the case of the urban park location, all subjective variables show this significant average equality.
Mean Quality Loc. quality Pleasant Exciting Eventful Tranquil
Small group 6.21 7.29 7.00 3.79 3.79 7.29 Random sample 6.60 7.60 6.20 3.50 4.00 6.10 M-W U test U 57.5 53.0 59.5 54.5 64.0 52.0 Z -.743 -1.040 -.631 -.931 -.363 -1.082 Exact Sig. .472 .341 .546 .371 .752 .312
Table 4.15: Urban park soundscape comparison
The small group study served to further validate the studies large scale method- ology. The supervised small group has shown significant agreement with the find- ings made from the random sample taken from the project dataset. Whilst the ob- served similarities does not fully validate the projects novel methodology, it does support it in terms of its potential to gather meaningful soundscape research data in light of its different methodology. A more robust validation would involve much larger numbers of subjects in the small group to improve the statistical robustness of the comparison.