Assessing Open Space in Scotland: Reliability and Construct Validity of the Open Space Scale

Abstract

The design of open spaces is the subject of interest when searching for solutions to promote well-being and a better quality of life for dwellers, especially those who live in urban areas. A user-friendly open space that meets the needs of an area has become a major concern in sustainable city design, environmental health, and psychological health. Thus, a universal scale that can be applied in different places is needed to study the different needs of different areas. This study systematically adapted the Open Space Scale previously developed in Hong Kong and assessed the reliability and constructed validity of the adapted version in Scotland; 535 samples from Edinburgh and Glasgow completed the revised version of the scale. The Confirmatory Factor Analysis indicated a good model fit and factor loadings in the revised scale. Overall, the Open Space Scale-Revised demonstrated satisfactory measurement properties. In the future, this scale can be used with other scales for further analysis and more complex structural equation models. This scale can also be used in conjunction with other scales for various types of policy analysis to provide policymakers and urban planners with substantial data. For example, one can measure the influence of open spaces on physical and psychological health in an area, such as well-being and sense of belonging, to decide if it is required to improve or expand the proportion of open spaces in that area.

1. Introduction

The design of open spaces is often considered a vital part of town planning as open spaces provide chances for recreation, socialisation, and interaction [1]. Over the last few years, town planners have argued that satisfactory open spaces can enhance the sense of attachment of people in their communities [2]. Open spaces have been revealed to have a positive influence on people’s mental and physical health through numerous studies. For example, Topcu claimed that open spaces such as parks and gardens are essential to physical and social well-being networks [3]. A study by Scotland’s Nature Agency published in 2020 found that nearly 70 per cent of people believe that spending time outdoors in open spaces helps them reduce stress; 56 per cent of people agreed that open spaces have improved their physical health [4]. A study at the University of Edinburgh revealed that green spaces were related to lower stress levels among university students [5]. Another study in Sheffield found that the primary users of parks are children—many parents bring their children to parks to play, socialise, exercise, and connect with nature [6]. For older adults, previous studies have found that open spaces are associated with older adults’ mental health and thus their chances of achieving a successful retirement [7,8,9,10,11]. The research has also found that open spaces are crucial in promoting active ageing [12,13,14,15].

There are some points to consider when planning open spaces. The World Health Organisation suggested that eleven elements should be considered when planning open space, including services, environment, security, green space, outdoor seating, sidewalks, paths, road traffic, cycle lanes, buildings, and lavatories [16]. Moreover, some researchers have proposed the consideration of environment, facilities, location, and accessibility in planning open spaces [2]. These elements are vital because the environment, security, and amenities in an open space affect the health and social interactions of users of open spaces [17]; while location and accessibility are associated with the use of open spaces [18].

Many countries and localities define open space and specify the criteria that should be considered when planning open spaces. For example, in Hong Kong, an open space is a ‘statutory land use area that provides open space and recreational facilities for the public’, according to the Hong Kong Planning Standards and Guidelines [19]. Based on this definition, open spaces can be parks, gardens, rest areas, coastal walks, recreational areas, children’s play areas, and jogging trails in Hong Kong [19]. The Planning Standards and Guidelines were first published in 1990 when Hong Kong was still a British colony and they are still in use today.

However, there is no uniform definition of open space in the United Kingdom regarding size, quality, or description. Nor is there a general or statutory understanding of what is meant by open spaces. Traditionally, many of today’s open spaces were enclosed land in the past, often surrounding palaces or mansions for deer protection and landscaping for the entertainment and recreation of monarchs or landowners. Iron railings in most open spaces and parks were not removed until the Second World War [20]. The first legal explanation of open space is listed in Section 20 of the Open Spaces Act 1906, which states that open space is any land that is used as a garden, recreational area, or uninhabited [20]. For park, it was only defined in Section 15 of the Local Government Finance Act 1988 for council tax rating purposes [20]. Currently, the primary definition of open spaces is based on a passage in the Town and Country Planning Act 1990, which has a similar definition to Section 20 of the Open Spaces Act 1906 [21].

The British government basically has no clear requirements for the standards that should be considered when planning open spaces. In the National Planning Policy Framework, only Articles 98 and 100 refer to planning standards for open spaces [22]. Articles 98 and 100 only focus on the needs of facilities and public rights of access [22]. Apart from that, there is no clear standard requirement for open spaces. The provisions also simply require planners to conduct an assessment based on local situations and decide how to plan open spaces based on the information obtained from the assessment. Therefore, planning open spaces in the United Kingdom is often the responsibility of local councils. Since local councils in the United Kingdom have a certain degree of autonomy and political wrestling, each local council has different planning urban standards, which leads to some counties having enough open spaces, while some counties cannot meet the psychological and physical needs of residents [23]. For example, an ethnographic study in London in 2022 found that policy and planning interventions aimed at increasing green space use often led to unplanned consequences, including a decrease in actual use in some cases [23].

The characteristics of open spaces are the subject of interest when searching for solutions to foster well-being and a better quality of life for dwellers, especially those who live in urban areas. A user-friendly open space that caters to an area’s needs has become a major concern in sustainable urban planning, environmental health, and psychological health. Understanding the different needs and expectations of different communities regarding the use of open spaces is essential, especially in the post-COVID-19 era, when many studies found that open spaces are important for people to recover from COVID-19. In a recent study, 90 per cent of British respondents said that open spaces had enhanced their quality of life throughout the pandemic [24]. A report in the United States of America showed that woodland in green areas improved mental health during the pandemic [25]. One study also found that residents of Perth in Australia and Moscow in Russia appreciate and rely on open-air green areas to support their well-being, particularly during the pandemic [26].

However, in understanding people’s needs for open space, we have lacked effective measurement tools or scales to understand the needs of people in different regions. Urban planners often can only rely on their own experience or refer to the experience of other regions to design open spaces (especially in the case of the United Kingdom), which frequently leads to the inability to meet the needs of residents. Thus, a universal scale that can be applied in different places is needed to study the different needs of different areas. This study adapted the scale developed by Yu, who previously used the scale in Hong Kong and assessed the construct validity of the scale for Scottish adults [27].

2. Materials and Methods

2.1. Design and Participants

This study is quantitative and cross-sectional. The sample was collected in Edinburgh and Glasgow, two major cities in Scotland. A face-to-face survey was conducted in Glasgow Green and the Meadows in the summer of 2021. The selected locations are, respectively, the primary public open space in Glasgow city centre and Edinburgh city centre. The participants were asked to rate different open space attributes. A convenience sample of 535 was obtained. The recruited participants were aged 18 years and over and were in a good physical and mental state to complete the questionnaire.

2.2. Instruments

In this study, open spaces were measured by the Quality of Open Space Scale developed by Yu [27]. This is a 4-point Likert scale with 12 questions related to different open space attributes (

Table 1. Quality of open space scale developed by Yu [27].

No.	Item	Factor
V1	Are located at a safe location	Location
V2	Are near the residential area	Location
V3	Have enough walking pathways	Entrance
V4	Have adequate lighting	Environment
V5	Have easily identifiable entrances	Entrance
V6	Have adequate green areas	Environment
V7	Have satisfactory maintenance and management	Environment
V8	Have adequate barrier-free designs	Social and Recreational Facility
V9	Have different types of recreational facilities	Social and Recreational Facility
V10	Have adequate seating	Social and Recreational Facility
V11	Are easily accessible to all people	Entrance
V12	Have enough space for social activities	Social and Recreational Facility

Note. Questions start with “I think open spaces in my town/city”.

). The original scale had acceptable reliability (Cronbach’s α = 0.805) and four factors were suggested in the study: which were (1) Social and Recreational Facility, (2) Environment, (3) Location, and (4) Entrance [27].

2.3. Data Processing and Analysis

The questionnaire data were analysed with JASP version 0.16.13 [28]. Since the scale has been used for exploratory factor analysis and published before, this study directly conducted Confirmatory Factor Analysis to examine the fit of the data to different models (

Table 2. Comparison of five models.

Illustration	Diagram
No common factor model (null hypothesis)
Single-factor model
Uncorrelated model (no correlation between factors)
Correlated model (correlation between factors)
Second-order model/hierarchical model

). The analysis started with a baseline model (null model), reflecting the null theory that the latent construct of open space is constituted of numerous independent factors. Then, the single-factor model was examined, which evaluated the measured variables as a single latent factor. This model assumed that the twelve measurement variables of the Open Space Scale measure only one common factor rather than some individual factors. If this pattern is supported, the scale presents only a single dimension of the construct. The third model assumed that the Open Space Scale has some different factors that are independent of each other. If this pattern is supported, the scale can be divided into different factors, but the factors belong to different constructs that are uncorrelated. On the contrary, the fourth model allowed the correlation of latent constructs. This model assumed that the Open Space Scale has some different factors that are related to each other. This model, if supported, indicates that the scale can distinguish some distinct factors and implies the possibility of a hierarchical model. Finally, a single second-order model was tested. This model assumes that the Open Space Scale has some different factors and that a higher-order factor can cover these distinct factors. If this pattern holds, it means that the scores for these distinct factors can be summed up into a single factor score that is meaningful and interpretable.

Estimating methods such as the Maximum Likelihood method and the Generalised Least Square method are prominently affected by the distribution of variables. If the value of skewness of variables is more than 3 or the value of kurtosis is more than 10, it is deemed problematic [29]. Furthermore, the Maximum Likelihood is not appropriate for ordinal variables as it assumes that the observed variables adhere to a normal distribution. Since the Likert scale is technically an ordinal measurement, this study used the Diagonally Weighted Least Squares method instead of the Maximum Likelihood method, as the former is suitable for the ordinal data [29]. Although the Diagonally Weighted Least Squares method has no assumptions regarding the distribution of the observed variables, it assumes a normal latent distribution for every categorical variable [29].

This study would eliminate items with low loadings (λ < 0.4) to enhance the proportion of variance explained in constituent variables of the factors according to the recommendations by Nunnally and Bernstein [30]. Apart from factor loadings, the following indicators for evaluating the model fit were employed. First, the minimum fit function (χ²) was employed. A chi-square and degree of freedom ratio between 2 and 5 was considered satisfactory [31]. Nevertheless, the chi-square test is sensitive to the sample size and will easily generate a poor fit result with a large sample size (>200). Therefore, Hair et al. recommended that additional goodness-of-fit statistics should be used [32]:

• For absolute fit, a Standardised Root Mean Square Residual (SRMR) value of no more than 0.08 was deemed as a good fit [33]; a Root Mean Square Error of Approximation (RMSEA) between 0.05 and 0.08 was deemed as a good fit and below 0.05 was deemed as the best fit. A Goodness of Fit Index (GFI) of no less than 0.90 was deemed a good fit model [33].
• For the relative fit, the Comparative Fit Index (CFI) and Tucker–Lewis index (TLI) values of no less than 0.90 were deemed a good fit [33].
• Other indicators such as Normed Fit Index (NFI), Non-normed Fit Index (NNFI), Relative Fit Index (RFI), Incremental Fit Index (IFI), and Relative Non-centrality Index (RNI) were also used to determine the relative fit, although they were not always necessary. Only these indices of no less than 0.90 are deemed a model of good fit.
• The scale’s reliability was assessed through its internal consistency with Cronbach’s α. A Cronbach’s α of no less than 0.70 indicates a good internal consistency.
• Convergent validity, discriminant validity, and factorial validity were then tested, respectively, using composition reliability (CR), average variation extraction (AVE), and Kaiser-Mayer-Olkin (KMO) and Bartlett’s tests.

3. Results

3.1. Characteristics of the Respondents

A total of 535 samples were collected for the study; 272 of them were collected in Edinburgh and 263 were collected in Glasgow. The male respondents (n = 275; 51.4%) slightly outnumbered the female respondent (n = 260; 49.6%). Fifteen per cent of respondents (n = 81) were retired. Over two-thirds of the sample had university degrees (n = 363; 67.85%). The mean age was 36.94 (SD = 6.33; range = 18–72) (

Table 3. Sociodemographic of respondents (N = 535).

	n	%
Sex
Male	275	51.4
Female	260	48.6
City
Edinburgh	272	50.8
Glasgow	263	49.2
Employment status
Retired	81	15.1
Education level
Degree or above	363	67.9
Age
Mean	36.94
Standard deviation	6.33
Range	18–72

).

3.2. Model Fit Indices

Table 4. Model fit.

	Cut-Off	Baseline	One Factor	Yu (2021)	Three Factors Uncorrelated	Three Factors Correlated	Second Order Model
Absolute fit
χ2		509.336	135.395	82.538	202.800	98.571	98.571
df		66.000	54.000	48.000	54.000	51.000	51.000
p	≥0.05		0.001	0.001	0.001	0.001	0.001
GFI	≥0.90		0.912	0.945	0.871	0.976	0.976
SRMR	≤0.08		0.066	0.052	0.165	0.054	0.054
RMSEA	≤0.08		0.081	0.056	0.109	0.058	0.058
Relative fit
NFI	≥0.90		0.734	0.838	0.602	0.942	0.942
NNFI/TLI	≥0.90		0.776	0.893	0.590	0.962	0.962
RFI	≥0.90		0.675	0.777	0.513	0.924	0.924
IFI	≥0.90		0.821	0.925	0.673	0.971	0.971
RNI	≥0.90		0.816	0.922	0.664	0.971	0.971
CFI	≥0.90		0.816	0.922	0.664	0.971	0.971
Parsimony fit
PNFI	≥0.50		0.601	0.609	0.492	0.728	0.728
χ2/df	≤3	7.717	2.507	1.720	3.756	1.933	1.933

Notes. df = Degrees of freedom; GFI = Goodness of fit index; SRMR = Standardised root mean square residual; RMSEA = Root mean square error of approximation; NFI = Normed Fit Index; NNFI = Non-normed Fit Index; TLI = Tucker–Lewis Index; RFI = Relative Fit Index; IFI = Incremental Fit Index; RNI = Relative Noncentrality Index; CFI = Comparative Fit Index; PNFI = Parsimony Normed Fit Index; Underlined figure = unfit.

presents the results of comparative analyses of fit. The test started with the baseline model with χ² valued at 509.34 (df = 66). The one-factor model was first tested. The result indicated that this model was a poor fit, with χ² (135.40)/54 = 2.51, p < 0.001, CFI = 0.816, GIF = 0.912, NNFI/TLI = 0.776, SRMR = 0.066, RMSEA = 0.081.

The following model was based on the four factors model suggested by Yu (2021). The result reported the model was unfit in this sample according to the cut-off adopted in the study, with χ² (82.54)/48 = 1.72, p < 0.001, CFI = 0.922, GIF = 0.945, NNFI/TLI = 0.893, SRMR = 0.052, RMSEA = 0.056. The value of NNFI/TLI was slightly lower than the 0.9 cut-offs. While this study adopted the cut-off of 0.9, as suggested by Hu and Bentler [33], Hopper et al. have suggested that the cut-off score of NFI and NNFI could be as low as 0.8 due to the complexity of the formula [34]. Although this model could be legitimately deemed as a marginal fit, this study considered this model unfit as stricter criteria were used [33]. It was believed that a better model could be formed. Furthermore, the RFI of this model was also unfit.

In order to obtain a better model, this study then revised the scale based on the literature and combined Location and Entry as Accessibility. This was the result of careful consideration as combining Location and Entrance as Accessibility would cause the model to be more straightforward and concise. This study also moved V3 (Walking Pathway) to Environment. The Social and Recreational Facility was also renamed as Facility to better reflect the items on the scale. Model 4 was the modified three factors model and assumed that there was no correlation between each factor. This model reported a poor fit, with no indicator passing the cut-off (χ² (202.80)/54 = 3.76, p < 0.001, CFI = 0.664, GIF = 0.871, NNFI/TLI = 0.590, SRMR = 0.165, RMSEA = 0.109). The result suggested that the factors did not belong to different constructs that were independent of each other.

Model 5 was the modified three factors model and assumed a correlation between each factor. This model reported an excellent fit, with χ² (98.57)/51 = 1.93, p < 0.001, CFI = 0.971, GIF = 0.976, NNFI/TLI = 0.962, SRMR = 0.054, RMSEA = 0.058. The only problem with this model was the χ², the value of which should not be significant for a good fit. However, this test can be ignored when the sample size is more than 200, as the low p-value was possible due to the large sample size. Hence, this model was deemed a good fit, implying the suggested three factors were correlated.

Model 6 was the second-order model (hierarchical model), assuming three distinct open space factors could be summed up into a single factor. This model reported an excellent fit, with all indicators the same as Model 5 (χ² (98.57)/51 = 1.93, p < 0.001, CFI = 0.971, GIF = 0.976, NNFI/TLI = 0.962, SRMR = 0.054, RMSEA = 0.058). Therefore, whether one should use the first-order correlated model or second-order model subjects to the research objectives. The first-order correlated model allows one to study the relationship between three open space factors (Accessibility, Environment, and Facility). The hierarchical model can be used to assess the open space factors as an independent or dependent variable.

3.3. Factor Loadings

This study used Model 5 to check the factor loadings, covariances, and validity. Model 5 reported a good fit and the factor loading of each variable also passed the 0.4 threshold with p < 0.001 (

Table 5. Factor loadings.

					95% CI
Factor	Indicator	Estimate (λ)	SE	p	Lower	Upper	R2
Environment	V4	0.695	0.047	<0.001	0.603	0.787	0.483
	V6	0.845	0.043	<0.001	0.76	0.929	0.713
	V7	0.701	0.045	<0.001	0.613	0.79	0.492
	V3	0.732	0.044	<0.001	0.646	0.817	0.535
Facility	V8	0.493	0.051	<0.001	0.394	0.593	0.243
	V9	0.673	0.046	<0.001	0.584	0.763	0.453
	V10	0.825	0.05	<0.001	0.727	0.923	0.681
	V12	0.869	0.055	<0.001	0.762	0.976	0.756
Accessibility	V1	0.677	0.061	<0.001	0.557	0.798	0.459
	V2	0.83	0.061	<0.001	0.71	0.949	0.688
	V5	0.568	0.053	<0.001	0.464	0.673	0.323
	V11	0.723	0.062	<0.001	0.601	0.846	0.523

Notes. SE = standard error; CI = confidence interval; estimates were standardised.

). Some scholars such as Bogozzi and Yi suggested that factor loading should be at least 0.5 and not be larger than 1 [35]. It was noted that the factor loading of V8 did not pass this stricter threshold (λ = 0.493). However, this study decided not to delete this item as the factor loading was very close to 0.5 and this study adopted the 0.4 thresholds suggested by Nunnally et al. and Phinney and Ong [30,36].

Table 6. Factor covariances.

						95% CI
			Estimate (ψ)	SE	p	Lower	Upper
Environment	↔	Facility	0.713	0.069	<0.001	0.578	0.849
Environment	↔	Accessibility	0.566	0.085	<0.001	0.400	0.732
Accessibility	↔	Facility	0.647	0.080	<0.001	0.489	0.804

Notes. SE = standard error; CI = confidence interval; estimates were standardised.

shows the factor covariances of the model. Three factors had a strong effect, implying they were highly correlated with each other (Environment and Facility: ψ = 0.713, p < 0.001; Environment and Accessibility: ψ = 0.566, p < 0.001; Accessibility and Facility: ψ = 0.647, p < 0.0.001).

3.4. Convergent Validity (Construct Validity)

Convergent validity can be determined from composition reliability (CR) and average variation extraction (AVE). The composition reliability (CR) can be considered as the internal consistency of the constructs. Fornell and Larcker recommended that the CR of the latent variables should be higher than 0.6 [37]. Basically, CR is similar to Cronbach’s α, but due to the minor difference in the formula, the value of CR will be slightly higher than Cronbach’s α. Thus, this study adopted Cronbach’s α to check the internal consistency of the full scale and subscales. The Cronbach’s α of the full scale was 0.93, which was higher than the initial study in Hong Kong. The three subscales also showed an assuring internal consistency, with Environment = 0.833, Facility = 0.814, and Accessibility = 0.796.

The Average Variation Extraction (AVE) is a measure of the amount of variance that can explain the latent variable. Fornell and Larcker advised that the AVE should be higher than 0.5 [37]. However, if the value of AVE is higher than 0.5, the factor load must be higher than 0.7. Since factor loadings may not always be above 0.7 in actual situations, it is therefore acceptable that the value of AVE can be low as 0.36 [37]. The subscales reported satisfying AVE values, with Environment = 0.746, Facility = 0.730, and Accessibility = 0.706.

3.5. Discriminant Validity

Fornell and Larcker and also Gaski and Nevi proposed three criteria for discriminant validity [37,38]. First, the correlation coefficient between constructs should be smaller than one. Second, the correlation coefficient between constructs should be less than Cronbach’s α. Third, the correlation coefficient between constructs should not be more than the square root of the AVE. The model met all criteria, implying that the scale has discriminant validity (

Table 7. Matrix of the correlation coefficient, CR, and AVE between constructs.

	1.  Environment	2.  Facility	3.  Accessibility
1	CR/AVE = 0.833/0.746
2	0.713	CR/AVE = 0.814/0.730
3	0.566	0.647	CR/AVE = 0.796/0.706

).

3.6. Factorial Validity

The Kaiser-Mayer-Olkin (KMO) test and Bartlett’s test were performed to indicate the factorial validity and to ensure the factor distributions of the revised Open Space Scale were similar to the original scale. A KMO of no less than 0.70 implies a factorial and valid scale. The KMO value of this study was 0.752, which was comparable to the original version (0.781). Thus, it affirmed the factorial validity of the revised Open Space Scale. The finalised model of the Open Space Scale-Revised is shown in Figure 1.

4. Discussion

This study examines the reliability and construct validity of a scale for measuring open spaces in Scotland. The four-factor Open Space Scale was simplified to a robust and concise three-factor scale following a Confirmatory Factor Analysis on the Scottish sample. Factor 1 is Environment, Factor 2 is Facility, and Factor 3 is Accessibility. The scale was named Open Space Scale-Revised. The three factors represent the essential elements related to open spaces. The internal consistency score for the Open Space Scale-Revised showed that the retained factors are correlated, meaning that the three subscales are unique and different facets of the same underlying construct.

In the original study by Yu, the Walking Pathway was categorised in Entrance as the result of statistical analysis and the enclosed layout of open spaces in Asia [27]. The reserved factors may be more reflective of the actual situation and design of open spaces in most countries as the merging of Location and Entrance resulted after considering the actual situation and design of open spaces in the United Kingdom. In metropolitan areas in Asia, it is typical for open spaces to be surrounded by iron railings or walls. These open spaces, as a result, have a clear entrance, such as gates, in conjunction with walking pathways. However, in the United Kingdom (as well as many non-Asian cities), it is not always the case that open spaces have an entrance or gate as people can entry the open space from anywhere, as long as they can step into the green. Therefore, this study moved V3 to Environment as walking pathways are less related to entrances outside Asia.

This three-factor scale also reflects that environment, facilities, and accessibility are important elements of a good open space. This is also consistent with the analysis of Yung et al. [2]. However, previous studies in Britain have focused less on facilities, especially those that benefit the elderly, children, and socialisation [6]. Much of recent research in Britain has mainly focused on how green spaces affect health, stress relief, and even eye protection, especially during COVID-19 [4,5,24]. British legislation and planning guidelines also focus on greening and location and less emphasis on facilities [20,21]. This is understandable because most European and American countries have more urban green spaces than Asian countries and their development density is relatively low [39]. However, in Asia, there are more concrete buildings in cities such as Hong Kong, Tokyo, and Singapore [27,40]. The structure generated in this study reflects that the open space facilities are equally important. The respondents of this study also attach great importance to the facilities in open spaces as facilities in open spaces are highly related to the environment and accessibility. In fact, facilities can affect the interaction and health of open space users [17]. Future urban studies in Britain should pay more attention to the facilities of open spaces.

5. Conclusions

This study examined the reliability and construct validity of the new scale to evaluate open spaces in Scotland. The newly validated revised version of the Open Space Scale (Open Space Scale-Revised) is a more straightforward instrument with a balanced quantity of variables shared across the three factors. The Confirmatory Factor Analysis indicated a satisfactory model fit and factor loadings in the revised scale. According to the results, the revised subscales (facility, environment, and accessibility) were correlated. Overall, the revised scale demonstrated satisfactory measurement properties. The factor loadings and reliability scores also show the robustness of the instrument. In the future, this scale can be used with other urban planning related scales for further analysis and more complex structural equation models, such as the Neighbourhood Environment Walkability Scale [41]. This revised version of the Open Space Scale can also be used in conjunction with other scales for various types of policy analysis to provide policymakers and urban planners with substantial data. For example, one can measure the influence of open spaces on physical and psychological health in an area, such as well-being, attachment, and sense of belonging, by using the Sense of Community Scale to decide if it is required to improve or expand the proportion of open spaces in that area [27].