Tafirenyika Sunde and Thulani Dlamini
Description
The deposited materials contain the de-identified survey dataset, anonymised qualitative accounts, coding documentation, supplementary material and R files required to reproduce the quantitative analysis. The analysis uses R version 4.3 or later and the open-source packages haven (version 2.5 or later) and jsonlite (version 1.8 or later). Download all files into one folder. Rename the downloaded survey .sav file as source_data.sav, or provide its original path when running the script. Install the required packages by running: install.packages(c("haven", "jsonlite")) From a command prompt or terminal opened in the download folder, run: Rscript reproduce_analysis.R source_data.sav analysis_outputs Alternatively, open R and run: source("reproduce_analysis.R") The script imports the survey data and preserves its variable labels. It audits the number of observations and variables, missing values, scale ranges, exact variable equivalences and repeated response profiles. The outcome-equivalent recruitment variable is excluded to prevent outcome leakage, while the duplicate strategic-direction variable is removed to prevent double weighting. The script then constructs six equal-weight capability domains: governance, financial controls, asset–liability management, risk management, stakeholder networks and leadership. It calculates the capability-system index as the equally weighted mean of these domains. It also calculates the weakest-domain index and within-profile dispersion. Dimensionality is assessed using the Kaiser–Meyer–Olkin statistic, principal component analysis and parallel analysis based on 2,000 independently permuted datasets. Associations with perceived DFI effectiveness are estimated using Spearman’s correlation. Confidence intervals are obtained from 10,000 response-profile-cluster bootstrap resamples. The script also performs unique-profile, leave-one-profile-out and profile-clustered ordinary least-squares sensitivity analyses. The principal verification values are 220 complete observations, 25 source variables and 69 unique response profiles. The expected KMO statistic is 0.951, and parallel analysis should retain one component explaining approximately 90.8% of item variance. The expected Spearman correlation is approximately 0.936 for the capability-system index and 0.924 for the weakest-domain index. Results are written to the analysis_outputs folder as CSV, JSON and text files. These include the integrity audit, descriptive statistics, demographic profile, reliability audit, PCA loadings, parallel-analysis results, bootstrap associations, sensitivity tests and an analysis digest. Minor differences in the final decimal place may arise across R or package versions. The qualitative files provide an audit trail for the reported thematic interpretation but are not treated as statistically reproducible evidence.
Files
Steps to reproduce
The deposited materials contain the de-identified survey dataset, anonymised qualitative accounts, coding documentation, supplementary material and R files required to reproduce the quantitative analysis. The analysis uses R version 4.3 or later and the open-source packages haven (version 2.5 or later) and jsonlite (version 1.8 or later). Download all files into one folder. Rename the downloaded survey .sav file as source_data.sav, or provide its original path when running the script. Install the required packages by running: install.packages(c("haven", "jsonlite")) From a command prompt or terminal opened in the download folder, run: Rscript reproduce_analysis.R source_data.sav analysis_outputs Alternatively, open R and run: source("reproduce_analysis.R") The script imports the survey data and preserves its variable labels. It audits the number of observations and variables, missing values, scale ranges, exact variable equivalences and repeated response profiles. The outcome-equivalent recruitment variable is excluded to prevent outcome leakage, while the duplicate strategic-direction variable is removed to prevent double weighting. The script then constructs six equal-weight capability domains: governance, financial controls, asset–liability management, risk management, stakeholder networks and leadership. It calculates the capability-system index as the equally weighted mean of these domains. It also calculates the weakest-domain index and within-profile dispersion. Dimensionality is assessed using the Kaiser–Meyer–Olkin statistic, principal component analysis and parallel analysis based on 2,000 independently permuted datasets. Associations with perceived DFI effectiveness are estimated using Spearman’s correlation. Confidence intervals are obtained from 10,000 response-profile-cluster bootstrap resamples. The script also performs unique-profile, leave-one-profile-out and profile-clustered ordinary least-squares sensitivity analyses. The principal verification values are 220 complete observations, 25 source variables and 69 unique response profiles. The expected KMO statistic is 0.951, and parallel analysis should retain one component explaining approximately 90.8% of item variance. The expected Spearman correlation is approximately 0.936 for the capability-system index and 0.924 for the weakest-domain index. Results are written to the analysis_outputs folder as CSV, JSON and text files. These include the integrity audit, descriptive statistics, demographic profile, reliability audit, PCA loadings, parallel-analysis results, bootstrap associations, sensitivity tests and an analysis digest. Minor differences in the final decimal place may arise across R or package versions. The qualitative files provide an audit trail for the reported thematic interpretation but are not treated as statistically reproducible evidence.