Study area, sample selection, and data collection
This research includes Dhaka city as a study area (geographic location). A quantitative research approach using a structured questionnaire was used to obtain primary data. This study considers Dhaka city’s residents as a population for sample selection. However, the criteria for this study comprise Dhaka city voters who respond to the survey with their opinions. Voters’ living experiences allow them to make insightful observations about Dhaka’s social sustainability. A manageable sample size is required since getting data from all voters and the researcher’s financial and time restraints is challenging. The sample size for this study was calculated using G*Power 3.1.9.7; the explicated sample size was 287, and the real power was over 0.80, indicating a reasonable sample power (Chin, 2001). Thus, the minimum sample size for this study should be 287 (see Supplementary Fig. S1). Residents of Dhaka city provided 573 responses for this study.
In addition, a multistage sampling technique was used to choose the participants. In the first stage, Dhaka city voters were selected using purposive sampling. Using a systematic sampling technique, 23 wards from city corporations were chosen in the second stage. In the third stage, a systematic sampling technique was used to select the houses of target respondents, namely voters, by collecting information on voters from the ward commissioner’s office and the Bangladesh Election Commission. Finally, the researcher obtained 573 responses from Dhaka city residents.
Preliminary study
As part of the pre-testing procedure, this study examined the content validity of the survey questionnaire. To check the content validity of the individual item (I-CVI) and overall scale (S-CVI) scores, a structured questionnaire was sent to six highly experienced and top authorities, including Directors of Urban Planning and Development, City Planners, Consultants, and Program Analysts from national and international platforms with four scale degrees of relevance (consistency, representative of concepts, relevance to concepts, and clarity in terms) (see Supplementary Table S3). According to experts’ recommendations, 01 items must be merged with other existing items, and 02 items must be rearranged. Finally, 62 items were chosen under 11 social sustainability themes for the preliminary study based on expert comments and relevance ratings. For the preliminary research, the study collected 109 responses from the residents of Dhaka city. After conducting the pilot study, some questions were re-examined to understand the language better.
Research variables
This study selected 62 items under 11 social sustainability themes as an independent variable (Table 1). The study’s dependent variable consisted of 05 socially sustainable urban development items. The selected social sustainability and socially sustainable urban development items are adopted from different scholarly literature (see Supplementary Tables S1 and S2).
Data processing
To eliminate errors, this study examined outlier identification, missing data, normality assessment, multicollinearity assessment, and reliability assessment. The study used the Mahalanobis D2 measure to detect outliers, where 09 observations out of a total response of 573 were removed as outliers (see Supplementary Table S4). With no outliers, the total number of responses to this study was 564. In addition, the study used SPSS version 22 to check for missing data and found no missing values in survey items. This study evaluated the dataset’s normality by examining its skewness and kurtosis. All the skewness and kurtosis values for each item were within the threshold level ±2 (see Supplementary Table S5). It indicates that the study’s data were distributed normally.
Using SPSS, the study also examined multicollinearity via tolerance and variance inflation factor (VIF) (see Supplementary Table S11). The coefficient outcome presented that the ‘Tolerance’ value is more significant than 0.10, and the VIF value is less than 10. Hence, this study did not identify multicollinearity issues that could aid future statistical analysis. Cronbach’s alpha was used to analyze the reliability of 11 themes, as indicated in Supplementary Table S12. The result of the reliability analysis showed that the overall Cronbach’s Alpha value was 0.951 with 62 items. Also, Cronbach’s Alpha scores for all individual variables varied from 0.899 to 0.957, suggesting that all variables achieved greater than 0.70, a much higher reliability level. Thus, all the measuring variables meet the required threshold value of Cronbach’s Alpha, which is acceptable, valid, and reliable for this study.
Data analysis
To propose a social sustainability model, this study employed exploratory factor analysis (EFA) and structural equation modeling (SEM) to perform confirmatory factor analysis (CFA). The researcher used a different dataset for EFA (219 responses) and CFA (345 responses). According to Henson and Roberts (2006), using the same dataset for EFA and CFA can be potentially misleading and uninformative. Green et al. (2016) further noted that applying EFA and CFA on the same dataset demonstrates only two integrated modeling approaches. The authors suggested using EFA and CFA on a different dataset. This study used separate EFA and CFA datasets to obtain the scientific findings.
EFA is a statistical method for analyzing and interpreting interrelationships between multiple variables regarding their common underlying factors (Hair et al., 2022). Accordingly, this study used EFA to refine the data to find a set of interrelated constructs that reveal the actual structure of the constructs. For EFA, the study used 219 responses using the principal component analysis (PCA) under the SPSS. Moreover, CFA is a technique to test how well the measured variables represent a smaller number of latent constructs that can confirm or reject the measurement theory (Hair et al., 2022). It is also utilized to determine the structural model’s unidimensionality, validity, and reliability for this study. Examining the acquired data, SEM assesses and analyses the correlations between observable and latent variables (Zheng et al., 2019; Kawesittisankhun and Pongpeng, 2020). This study evaluated and examined the relationships between the observable and latent variables using the CFA by SEM method through AMOS 26.0. Using CFA, the study also assessed both components of SEM, namely the measurement and structural models. The SEM method looks at the measurement error that leads to good confirmatory results. Consequently, the SEM method was utilized in this study to analyze the association between the independent (social sustainability) and dependent variables (socially sustainable urban development). This study used 345 responses for CFA and considered five observations per item.
Findings
This study proposed a model of social sustainability for socially sustainable urban development in Dhaka based on 564 responses using 62 items under 11 variables. The following sections discuss the study’s findings.
Exploratory factor analysis
The Kaiser–Meyer–Olkin (KMO) value of this study’s exogenous and endogenous variables was 0.902, representing adequate data sampling. Bartlett’s test of sphericity is highly effective as the Sig value was 0.000, demonstrating that there is no multicollinearity across the constructs and that all components are appropriate for EFA. Table 2 illustrates KMO and Bartlett’s test in EFA.
This study used communalities analysis to examine the interrelationship of 67 (exogenous and endogenous) items. Based on the findings, only SC5 (practicing social and ethical values) had an extraction value of <0.40 (see Supplementary Table S6). Consequently, the value of 66 items was more significant than 0.40, and the SC5 item was removed from further consideration. Moreover, the scree plot revealed 12 extracted factors, including 67 items with eigenvalues greater than 1. The scree plot assumes the curve began to flatten between 11 and 13 factors, resulting in the retention of 12 factors. Figure 1 depicts the scree plot of all retrieved factors’ Eigenvalues.
Moreover, the 12 factors with eigenvalues greater than one are explained in Supplementary Table S13. The outcome showed that the extracted 12 factors explain 78.58% of the total variance. Besides, this study used PCA with varimax rotation to analyze the 67 items relevant to socially sustainable Dhaka. Supplementary Table S7 explains the rotated factor matrix. The results presented that the factor loading of 66 indicators, which were put in a twelve-factor matrix, was more significant than 0.50. To establish the internal consistency of this study, just one indicator, SC5, which had a factor loading of <0.50, was determined to be deleted. Hence, 66 indicators of the 12-factor matrix were significant for further analysis.
SEM-based confirmatory factor analysis
Before conducting CFA using SEM, the study tested outlier, missing data, normality, and multicollinearity concerns in the dataset (refer to data processing section). This study used a single and a full measurement model to validate the measurement models. This study used CFA in the measurement model to check the relationship between all constructs by establishing reliability, validity, and unidimensionality and evaluating the model’s initial overall fit.
Measurement model
In assessing a single measurement model, twelve measurement models were examined that the study found after conducting EFA. CFA was performed on all constructs used in this study, and the average variance extraction (AVE), composite reliability (CR), individual item reliability (R2), and goodness-of-fit (GOF) indices were used to evaluate the measurement model’s validity. A total of 11 items were eliminated from the 66 items in the single measurement model due to Modification Indices (MI) values (>15) and Squared Multiple Correlations (R2) (<0.30). Hence, 55 items were retained under 12 factors and selected to assess the full measurement model (see Supplementary Table S8).
To assess the full measurement model, this study considered several issues, i.e., factor loading, R2 value, MI value, standardized residual covariance (SRC), and GOF. The two-headed arrow linked the full measurement model, showing the constructs’ covariance. This study’s goodness-of-fit indices did not produce adequate results in the first or second iteration. The model was a rerun, and the final iteration had better goodness-of-fit indices than the second iteration; for example, chiSq/df = 1.583 with a cutoff point <5, RMSEA = 0.041 with a cut-off point < 0.08, CFI = 0.953 with a cut-off point >0.90, GFI = 0.855 with a cut-off point >0.90, IFI = 0.953 with a cut-off point >0.90, TLI = 0.947 with a cutoff point >0.90, PGFI = 0.727 with a cut-off point >0.50, and PNFI = 0.784 with a cut-off point >0.50 (see Supplementary Fig. S2). Even though the goodness-of-fit index (GFI) was.855, Hair et al., (2011) stated that GFI values of more than 0.80 are acceptable. In a complicated model, the GFI with a lower value was accepted (Byrne, 2010; Hair et al., (2011)). Thus, all the goodness-of-fit indices for the full measurement model were satisfied. Ten items out of 55 were excluded because of their MI value and SRC (higher than 2.58), as shown in Supplementary Table S9. Likewise, this full measurement model attained unidimensionality (factor loading >0.50), Construct Validity (achieved the fitness indices), Convergent Validity (see Supplementary Table S10), Discriminant Validity (refer to Table 3), and Construct Reliability (see Supplementary Table S10). Hence, 45 items were retained and selected for assessing the structural model (see Supplementary Table S9).
Structural model
After completing the full measurement model, this study evaluated the structural model using 41 items (without dependent items) under 11 variables. The study assessed the R-square values, standardized residual covariance, goodness-of-fit indices, and modification indices to validate the structural model. According to Henseler et al. (2009), the R-square value must be higher than 0.25. R-square values of 0.75, 0.50, and 0.25 are typically considered substantial, moderate, and weak, respectively (Henseler et al., 2009; Hair et al., 2011).
This study’s structural model was intended to examine the interrelations of the variables that relate to independent to dependent variables. The one-headed arrow explains the relationship between the variables in the structural model, which depicts independent and dependent variables. The structural model has a chance to boost the goodness-of-fit indices in the first iteration, especially for the GFI value. After re-specifying the model and the results of goodness-of-fit indices were improved and pretty good in the 1st iteration, e.g., chiSq/df = 1.594 with a cut-off point <5, RMSEA = 0.042 with a cut-off point < 0.08, CFI = 0.956 with a cut-off point >0.90, GFI = 0.866 with a cut-off point >0.90, IFI = 0.956 with a cut-off point >0.90, TLI = 0.950 with a cut-off point >0.90, PGFI = 0.722 with a cut-off point >0.50, and PNFI = 0.779 with a cut-off point >0.50 (refer to Fig. 2). Though the goodness-of-fit index (GFI) value was 0.866 with a cut-off point >0.90, it was more significant than 0.80, which is also acceptable, as suggested by Hair et al. (2022) and Byrne (2010). GFI values between 0 and 1 are also acceptable (Hair et al., 2022).
Furthermore, the R2 value for the endogenous variable was 0.75 (Fig. 2), representing the influence of constructs the health facilities, gender equality and women’s empowerment, urban poverty, and slums improvement, urban children, aged, the disabled, the scavengers, transportation availability, satisfied with space, open space, social capital, social justice, safety, and education facilities, which was 75%. Therefore, the R2 value of this study was substantial, as it was above the suggested threshold. The structural model of this study was appropriate as it adequately achieved all the model fit indices. The above findings assumed that social sustainability significantly influenced socially sustainable urban development, eventually revealed by the best model fit.
Table 4 shows the items removed and retained following the structural model assessment. Three out of 41 items were removed because of standardized residual covariance (greater than 2.58). Thirty-eight items remained, showing that the structural model met the required criteria. Hence, the structural model (final iteration) is considered more suitable and significant for comprehending the relationship between social sustainability and socially sustainable urban development. Finally, 38 indicators (items) under 11 themes (variables) of social sustainability are essential for Dhaka city’s socially sustainable urban development.
Hypothesis testing
In this study, EFA and CFA validated the proposed model, and now the hypothesis must be tested. This study assessed hypothesis testing based on the relationship between the factors in the structural model. Through the structural model, path analyses were performed in this study. Using twelve valid constructs for the full measurement and structural models, the structural model was employed to test the hypothesis. To determine whether a p value is significant, the beta coefficient value (β) and critical ratio (C.R.) were established for hypothesis testing. The findings of the structural model analysis are presented in Table 5, which also includes standardized estimates, standard errors, critical ratio/t values, and p values at the significant level.
The results of the structural model indicated that eleven hypotheses were statistically significant. It shows that all hypotheses positively correlate with the study’s outcome. The results of the hypothesis testing made it clear that social sustainability has a direct influence on socially sustainable urban development. The social sustainability model for socially sustainable urban development in Dhaka city is finally shown in Fig. 3.