Mapping construction risk in Indonesia: Province-level evidence from CWA records

Abstract

Background

This study examines patterns in CWAs in Indonesia, utilizing 14,962 accident records from 2014 to 2020. The analysis relates accidents to workforce composition, building and project characteristics, and seasonality, by conceptualizing accidents as a systemic phenomenon influenced by factors such as workforce, organizational context, and temporal context.

Objective

This study identifies areas with accident rates disproportionate to workforce size, analyses temporal trends, assesses associations between workforce characteristics and construction accidents, assesses the gap between accident risk and safety monitoring capacity, and recommends targeted safety interventions for high-risk projects and building types.

Methods

We summarized accident frequencies and proportions by category; applied chi-square tests for goodness-of-fit and Fisher's Exact Test; mapped provincial CWAs and an Accident Index to identify high-risk areas; analyzing trends and seasonal patterns using a time series approach; compared CWAs across 13 building and project types via one-way ANOVA and Welch ANOVA; and applied Pearson correlation and negative binomial regression for over dispersed count data.

Results

The analysis reveals an uneven spatial and seasonal distribution of CWAs with a peak in October–December. New projects, industrial, residential, and transportation, are at the highest risk. Some pairs of categories are not significantly related. The negative binomial regression indicates that the CWN-SE is negatively and significantly related to accidents.

Conclusions

Temporal factors, project type, and safety capacity influence construction accidents. More CWN-SE means fewer accidents than more technical people do. Safety policies must be risk-based and systems-oriented and take account of the dynamics of time and project characteristics.

Keywords

construction worker accidents construction safety negative binomial regression time series analysis fisher's exact test and construction safety policy

Background

Infrastructure development, employment, and economic growth are all significantly influenced by the construction industry.^1–2 However, due to its complex work environments and exposure to high-risk situations, construction remains one of the most hazardous industries.³ Many studies have analyzed and categorized high-frequency incidents that present a particularly high risk of serious injury or death using accident reports from the construction industry.^4–8 In the occupational safety literature, construction accidents are widely perceived as the result of layered causal mechanisms (e.g., interactions among human, technical, organizational and environmental factors) rather than as random events or the result of individual actions alone. Although construction is inherently dangerous, the degree of risk varies by nation. Local socioeconomic factors can influence the safety climate and, in turn, workers’ behaviours on individual construction projects.⁹ As an example of an advanced economy, the Korean construction industry in 2022 recorded 402 fatalities, accounting for 46.0% of all worker accident deaths.¹⁰ Construction worker accidents (CWAs) in developed nations are typically linked to organizational and managerial flaws.¹¹ In contrast, workplace safety and accident prevention in developing nations are frequently hindered by financial limitations, such as restricted funding for safety initiatives.¹² The accident causality framework identifies those differences in risk across national and regional contexts are a consequence of variations in institutional capacity, workforce composition, project complexity, and levels of regulatory oversight and enforcement. The effect of having certified safety experts on workplace safety has been the subject of numerous studies. Among Chinese construction workers (CWN), Hu et al.¹³ discovered relationships between psychological needs, safety motivation, and personality traits. By enforcing adherence to safety procedures and encouraging a culture of safety awareness, some researchers argue that having safety experts on site is crucial to lowering accident rates. However, other research has shown that safety experts alone do not always lead to fewer accidents, especially when they have little authority or when site management does not strictly enforce safety rules for specific construction site hazards.¹⁴ These contrasting results suggest that the influence of safety professionals is highly context dependent and is contingent on the organizational context and safety management system, which warrants an analytical approach capable of disentangling causal effects from exposure and reporting effects.

In Indonesia, construction plays a significant role in the country's development, primarily through government-led public infrastructure initiatives to build high-rise buildings, roads, and bridges. However, the Indonesian construction industry continues to face persistent safety challenges. This study aims to identify areas with high accident rates relative to workforce size, assess the relationship between workforce traits and construction accidents, and identify safety interventions specific to projects and building types. Relaxed enforcement of safety regulations, insufficient worker training, and inadequate use of personal protective equipment (PPE) are common issues contributing to high accident and injury rates.⁹ Moreover, construction crews often comprise a mix of skilled and unskilled laborers, many of whom may lack formal safety training, resulting in increased workplace risks.¹⁴ From a causal standpoint, heterogeneous workforce composition and organizational safety capacity constraints can raise risk exposure while influencing the rate of recorded accident reporting. Zahoor et al.¹⁵ emphasize the need to allocate resources to effectively implement safety measures and comply with government regulations to enhance construction safety. The construction industry in Indonesia has the highest workplace accident rate among sectors, with the reported incident rate rising from seven in 2017 to fifteen in 2020.¹⁶ The Employee Social Security System of Indonesia (ESSSI) department reported in 2024 that there were 4233 insurance claims for work-related accidents in the construction industry. Moreover, Arman et al.¹⁷ note that the growth of Indonesian construction accidents increased 5.65% from 2020 to 2021. Such a trend highlights the need for analyses that describe not only accident frequencies but also the causal mechanisms underlying variations in accident rates across space and time. This trend has prompted several studies to identify CWAs to enhance the enforcement and effectiveness of safety measures. Bria et al.¹⁸ analyzed fatalities resulting from construction accidents, while Latupeirissa et al.¹⁹ examined the causes of work accidents and their impact on road and bridge construction projects. Additionally, Machfudiyanto et al.²⁰ identified a strong correlation between construction accidents in Indonesia and unsafe worker behaviors, such as improper tool placement, failure to secure safety lines, and improper handling of materials, which frequently lead to falls or injuries caused by being struck or cut. Nugroho and Sumbara²¹ focused on the types of work accidents in Indonesia and proposed safety improvement measures. Megasari²² analyzed accidents in a specific province, revealing that male workers aged 25–35 were most affected, particularly during morning hours, thereby highlighting the need for targeted safety interventions for high-risk worker demographics and periods. Suti et al.²³ analyzed construction accidents in a hospital project using field observations and risk assessment, finding that while the majority of incidents were classified as moderate in risk, two specific accident variables were identified as substantial hazards, indicating the need for targeted mitigation efforts.

Although research on construction safety is expanding, the focus remains on the project or national aggregate level, with limited integrated provincial perspectives. Thus, there is a significant literature gap on an integrated analytical framework that accounts for workforce composition, project and building types, seasonal factors, and regional variations. This lack of perspective presents a significant challenge, as budget planning, safety oversight, and regulatory implementation are typically carried out at the provincial level. Existing research has analyzed the types and causes of construction work accidents, leading to recommendations for safety practices in the construction industry. However, Indonesia's construction sector faces a high and inconsistent burden of CWAs, yet policymakers lack an integrated, province-level view that jointly considers workforce composition, building and project types, and seasonality. Without an integrated causal framework at the provincial level, it is difficult to distinguish between the exposure effects, reporting bias, and causal determinants of safety.

An absence of exposure-adjusted metrics, inconsistent reporting quality, and limited evidence on how staffing aligns with risk hampers targeted safety interventions and effective resource allocation. Without integrated provincial analysis, policymakers struggle to prioritize safety interventions amid budget constraints, resulting in reactive rather than preventative policies. Furthermore, fragmented accident data impedes the development of evidence-based regulations that are consistent with regional risk profiles. These gaps are increasingly consequential amid rapid infrastructure expansion, complex project portfolios, and rainy-season conditions that elevate risk, especially in densely populated and industrial provinces. In this context, an analytical approach that integrates exploratory correlation analysis with count-based regression models is important for clarifying causal relationships and supporting the development of evidence-based safety policies. Without timely, risk-based action, the mismatch between accident burden and available safety oversight will persist, slowing progress toward systematic reductions in CWAs. Align with these gaps, this study aims to: (1) pinpoint areas with disproportionately high accident rates relative to workforce size, (2) examine temporal trends in accident occurrences, (3) evaluate the link between workforce characteristics and construction workplace accidents, and (4) propose practical recommendations for targeting safety interventions in high-risk project types and building categories. These aims respond to the urgent need to address ongoing safety challenges, reduce occupational injuries, and inform evidence-based policy by analyzing the correlation between the number of CWAs and CWNCWN.

Methods

The research methodology framework is designed to provide robust statistical and spatiotemporal analysis of the CWAs across Indonesia. Figure 1 shows the research workflow, highlighting the methodologies used in this research. The framework includes stages from the definition of the research objective to data collection and pre-processing, analysis, and finally the discussion and conclusion. The framework supports a thorough analysis of CWAs, facilitates the identification of relationships between CWN and CWAs, and assists in designing possible measures to improve the construction industry's safety performance in Indonesia.

Figure 1.

Research design flowchart visualization.

Data collection and data preprocessing

Accident data were obtained from ESSSI, including accident frequency and distributions by province, year, project type (new, renovation, maintenance), and building type (residential, commercial, industrial, among others). These employee records kept by the ESSSI included all individuals who had work-related accidents with substantial losses, including medical costs, treatment, or other losses incurred. The ESSSI department has a nationwide coverage of all the employees in Indonesia and thus requires protecting the privacy of workers’ accident records. Therefore, the meaning of CWAs in this study is defined as work-related incidents that involve significant losses recorded in the country's insurance system. The number of CWN per province was obtained from the Directorate of Construction Development Indonesia (DCDI). All of the data collected underwent pre-processing steps for handling of missing values and data standardization.

Analytical approaches

Descriptive statistical analysis and chi-square test

The data were analysed using descriptive statistics, including the number of CWAs per year, the distribution of the CWN, project types, and building types. This step included reporting the results of data pre-processing, including the identification of the most descriptive variables of interest for the analysis. These significant descriptive variables were analyzed using the chi-square goodness-of-fit test and Fisher's exact test, which included the determination of the suitability of data distribution and the relationship between the different variables. The chi-square goodness of fit test was then utilized in the determination of the distribution of data among the different variables, which included the identification of the variability of the mean rankings.²⁴ Based on the study by Cao et al.,¹⁴ which utilized the Chi-squared automatic interaction detection test to determine the effect of safety experts on the Chinese construction workforce, the study utilized the test in the determination of the distribution of the variables. The chi-square goodness of fit test is defined by the equation in Equation 1 .^25–28 The use of Fisher's exact test with weighting, such as TF-IDF, can be justified as a significance test approach, with the weight value closely related to the negative logarithm of the p-value of the relationship between term occurrences in documents.²⁹ Fisher's Exact Test does not use asymptotic approximations and returns exact p-values (avoiding biased significance values), resulting in more valid statistical inferences. Significant decisions are made based on exact sig. (two-sided) with a significance level of α = 0.05. The data was analyzed using the equation in Equation 2 , which included the determination of the chi-square statistic, which was analyzed using the equation in Equation 1 .

X^{2} = \sum (\frac{{(O_{i j} - E_{i j})}^{2}}{E_{i j}} + \frac{{(O_{i j + 1} - E_{i j + 1})}^{2}}{E_{i j + 1}} + \dots + \frac{{(O_{i j + n} - E_{i j + n})}^{2}}{E_{i j + n}})

(1)

\sum E_{i j} = \frac{R_{i} * C_{j}}{N}

(2)

Where X² is the value of the chi-square test statistic. The term O_ij is the observed frequency in the cell in row i, column j, and while E_ij is the expected frequency if the variable were independent. In Equation 2, R_i is the total number of observations for row i, C_j is the total number of observations for column j, and N is the total number of all observations.

Spatial distribution of CWAs

Spatial distribution analysis is an applied quantitative approach to the spatial distribution of phenomena such as populations, resources, and incidents that are spatially arranged and/or dense. Tran et al.³⁰ employed spatial–temporal exposure analysis using the HISTEA model, which incorporates accident scenarios into a four-dimensional Building Information Modelling (4D-BIM) space to investigate overlaps of construction activities across space and time. Their findings suggest that predictive hazard identification based on spatial and temporal patterns is useful for minimizing recurrent accidents during the construction stage. Similar to Duan et al.,³¹ who employed complex network theory to analyse the spatial and temporal patterns of risk in trajectories of CWN and revealed risk clustering and transition patterns at the macro and micro scales. In this study, a spatial analysis was conducted to identify the risk conditions of CWAs across the whole province of Indonesia based on Accident Index values.

Figure 3 illustrates the distribution of the CWAs across the study area, encompassing all provinces of Indonesia. Equation 3 calculates the distribution value (Dist.Value) by dividing the CWA rate per 1000 CWN by the safety rate per 1000 workers. Dist.Value_i is the distribution value of the i-th province, Number_of_CWA_i is the total number of CWA in the i-th province, while Number_of_Total_Worker_i is the total number of CWN in the i-th province.

D i s t . V a l u e_{i} = (\frac{N u m b e r_o f_C W A_{i}}{N u m b e r_o f_T o t a l_W o r k e r_{i}}) * 1000

(3)

In addition, the AI is determined using Equation 4 . The values obtained from AI provide information on the position of the province with respect to the national average, where Accident Index > 1 represents a more accident-prone province (dangerous), and Accident Index < 1 represents provinces with a lower accident rate (safer working conditions).

A c c i d e n t I n d e x = (\frac{D i s t . V a l u e_{i}}{A v e r a g e \sum D i s t . V a l u e})

(4 )

Time-series analysis

Time series analysis, which is a statistical tool used to analyze trends of observations over time, is imperative in understanding the dynamics surrounding CWAs. Time series analysis enables the analysis and understanding of fluctuations and trends in accident rates, thereby guiding strategies for the prevention and management of accidents. In particular, the time-series method selected in this study is used not only for prediction but also to understand the temporal structure and seasonal patterns of traffic accidents. The main goal of this study is to identify long-term trends and recurrent seasonal variations that have direct implications for safety policy planning and resource allocation. In this respect, the Moving Average (MA) and the STL techniques were selected for their transparency, interpretability, and appropriateness to the research's analytical goals, especially for the explicit separation of trend, seasonal, and residual components. This approach not only produces numerical predictions but also provides researchers and policymakers with information on when and how accident rates rise or fall over time.

In this case, time series analysis was performed using the MA technique³³ and the Seasonal-Trend Decomposition in Loess (STL)³⁴ to identify overall trends across seasons. The outcome of this analysis directly addresses research objectives two and four. The MA techniques were applied to identify seasonal trends in the data. Additional Seasonal Indices (SI) were calculated to determine overall seasonal patterns for each year. The calculation of the MA is done using Equation 5 , and the SI's are calculated using Equation 6 .

M A_{i} = \frac{(O_{i - 1} + O_{i} + O_{i + 1})}{3}

(5)

S I_{i} = \frac{O_{i}}{M A_{i}}

(6)

Where MA_i is the MA at the point of time i, O_i is the observed value at point of time i, while SI_i is the seasonal index at point of time i. The two formulas smooth the series using a MA, while the seasonal index identifies the seasonal variation in the number of accidents.

Comparative statistical analysis

In comparative statistical analysis, two or more datasets, groups, or variables are compared using quantitative methods to identify trends, similarities, and differences. Guo et al.³⁵ conducted a comparative analysis of unsafe act patterns in building construction and urban rail construction, demonstrating the effectiveness of using comparative statistical techniques to compare groups by performing analysis of variance (ANOVA). Honda et al.³⁶ perform ANOVA to evaluate the release of glass fiber pegs with different protocols. Furthermore, Guo et al.³⁷ focus on carbon emissions in the construction industry, calculating the life cycle of the construction sector and analyzing trends and characteristics of emissions. The use of comparative statistical analysis in this study aims to compare accident rates across various projects and building type groups. The results of this analysis provide an empirical basis for focusing on construction safety at the highest risk. Because Levene's test indicated a violation of the homogeneity-of-variance assumption (p < 0.001), the inferential analysis was based on Welch's ANOVA. The ANOVA table is presented for descriptive purposes, while the significance test uses Welch's F value. Specifically, Welch's ANOVA is a variation of the one-way ANOVA used to compare the means of three or more groups when the assumption of homogeneity of variance (equality of variances) is violated. IBM SPSS Statistics 31.0.2.0 is used to determine variance across the experimental groups (project type and building type groups), with a significance level of 0.05. Lastly, the final analysis is a post hoc Games-Howell test to identify which pairs of groups have statistically significant mean differences.

Correlation analysis

This study uses the Pearson correlation test to examine the association between CWAs and CWNs and to identify significant relationships useful for improving construction safety management. Chan & Aghimien³⁸ investigated the relationship between performance indicators and project success in a construction portfolio, specifically focusing on construction site safety. Furthermore, Wang et al.³⁹ applied the Pearson correlation coefficient to measure the linear relationship among multi-domain characteristics for bearing degradation, hence aiding in clustering of highly correlated characteristics before applying kernel principal component analysis (KPCA).

The relationships between variables were initially investigated using Pearson correlations. However, because the dependent variable was count data with evidence of overdispersion, the analysis was conducted using negative binomial regression. Negative binomial regression was developed to overcome the limitations of bivariate correlation, control for confounding variables, and provide a more accurate estimate of nonlinear relationships in accident data. The number of CWAs was used as the dependent variable, while the number of CWN-CC, CWN-SW, and CWN-SE served as independent variables. The regression results are presented as incidence rate ratios with 95% confidence intervals.

Results

The research findings are presented in this results section through a number of connected analyses that highlight the traits and dynamics of CWAs. To capture variations in risk across regions and over time, the initial analysis expanded through spatial and temporal approaches, focusing on understanding the distribution and underlying relationships among variables.

Descriptive statistical analysis and chi-square test

Data collection Indonesia

The 14,962 CWAs recorded in the ESSSI from 2014 through 2020 are provided in Table 1. The data reveal that the highest number of accidents per year occurred in 2019, accounting for 18.63% of the total number of accidents in the review period. This was followed closely by 2018, which contributed 17.23% of the accidents in the review period. The fewest accidents occurred in 2015, accounting for 11.86% of total accidents in the review period. The decrease in CWAs in 2020 can be attributed to the restrictions imposed in the country during the COVID-19 pandemic.

Table 1.

CWAs from 2014 to 2020.

Year	Number of CWAs	Percentage(%)	Project type
Year	Number of CWAs	Percentage(%)	New	Renovation	Maintenance
2014	1982	13.25%	1706	231	45
2015	1774	11.86%	1567	168	39
2016	1936	12.94%	1734	168	34
2017	1865	12.46%	1622	209	34
2018	2578	17.23%	2253	278	47
2019	2788	18.63%	2461	274	53
2020	2039	13.63%	1794	199	46
Total	14,962	100.00%	13,137	1527	298

Examining how accident types were distributed for project types being new construction, renovation, and maintenance, some patterns clearly stand out. The vast number pertains to new construction (87.8%), totalling 13,137. Additionally, renovation, maintenance, and other types have significantly fewer instances, roughly 10.2% and 2.0% for renovation and maintenance, respectively. This is an obvious indicator that new construction projects are more prone to accidents, most likely due to increased hazard complexity, extended project duration, size, and constantly varying site conditions. This situation aligns with Chan et al.,²⁴ new workplaces are potentially more dangerous for workers due to unfamiliar environments, new features, and hazardous conditions.

In summary, the data indicate a definite need for enhanced safety measures, particularly in the construction of new public infrastructure. The data show that a discernible impact can be achieved by targeting safety risks in major new infrastructure projects to reduce overall CWA rates. The significant increase in accident rates in 2018 and 2019 would also indicate various industry or regulatory dynamics that influenced safety during this period.

Figure 2(a). illustrates the distribution of CWAs across building types, indicating distinct CWA intervals within construction categories. In this respect, the total number of accidents in the industry/factory stands at 2,577, closely followed by transportation facilities with 2564. Next are residential projects with 2515 CWAs, and commercial with 2479 CWAs. On the other end, worship places projects account for 82 CWAs, followed by cultural & entertainment, with 100 CWAs, and mechanical & electrical buildings with 273 CWAs. From the distribution, there are comparatively more dangerous project types with increased total numbers of CWAs, mainly in the industry/factory, transportation, and residential categories, as indicated by the notably higher prevalence of CWAs.

Figure 2.

(a) CWAs distribution by building type. (b) CWAs geographic distribution from 2014 to 2020.

Figure 2(b). shows the geographic distribution of CWAs across the Indonesian provinces. The distribution shows accident numbers are highest in industrialised areas, likely because these regions host more construction projects. For example, East Java and DKI Jakarta, both densely populated and highly urbanized, each reported more than 3000 accidents during 2014–2020, while less urbanized provinces such as North Maluku, Maluku, and West Papua reported far fewer CWAs.

Data collection from the directorate of construction development Indonesia (DCDI)

Table 2 summarizes the classification of CWN-SW (construction worker number-skilled worker), CWN-CC (construction worker number-certificate of civil expertise), and CWN-SE (construction worker number-safety expert) in Indonesia, which follows the structured framework established by PUPR Regulation No. 09/PRT/M/2013 and the Directorate of Construction Development Regulation No. 021/KPTS/LPJK-N/IX/2018. Under these regulations, skilled CWN are categorized into three levels (Levels 1–3) based on education and experience. Construction professionals and safety experts are likewise organized into three tiers: apprentice, expert, and master.

Table 2.

CWN classification in Indonesia.

CWN	Minimum requirements
CWN	Occupational classification	Education	Working experience(years)	Notes
CWN-SW³²	Level 1	A vocational degree	3	Pass the competency test
	Level 2	High school graduate	3
	Level 3	Elementary school graduate	3
CWN-CC³²	Master	Bachelor degree	5	Pass the competency test
	Expert	Diploma three graduates	5
	Apprentice	Diploma three graduates	3
CWN-SE³²	Master	Bachelor degree	8	Pass the competency test
	Expert	Diploma three graduates	6
	Apprentice	Diploma three graduates	3

The certification process requires meeting formal education criteria, demonstrating relevant construction experience, and passing written examinations and interviews. This framework is designed to ensure that personnel possess the competencies appropriate to their roles. In addition, firms bidding for construction projects must include certified team members, thereby mandating the presence of qualified personnel to uphold construction safety and quality standards. Overall, this regulatory system strengthens professionalism and competency in the construction sector and promotes adherence to industry best practices.

Figure 3 summarizes regional distributions of CWN-CC (CWN-Certificate of civil expertise), CWN-SW (CWN-Skilled workers), and CWN-SE (CWN-Safety expert). The distribution of accidents across provinces makes those variables significant. By knowing the ratio for each province across Indonesia, the geographic accident risk analysis in each region will be conducted using the same ratio. The Riau Islands record the highest CWN-CC and CWN-SW counts, with DKI Jakarta, West Java, and East Java similarly high. In contrast, CWN-SE is low in all regions, suggesting potential gaps in safety oversight that may relate to CWAs. These patterns argue for expanded targeted safety interventions in high-risk areas.

Figure 3.

CWN distribution by province.³²

Data preparation and Variable identification

Data on CWAs were obtained from the ESSSI for 2014–2020. Before analysis, the dataset was curated to identify the relevant variables, as summarized in Table 3. Data on construction worker expertise were also compiled to examine the relationship between CWAs and CWN.

Table 3.

Identified variables.

Variables	Abbreviation
Construction Worker Accidents	CWAs
CWA-Types of projects	CWA-TP
CWA-Types of building	CWA-TB
CWA-Location province	CWA-LP
CWA-Year of accident	CWA-YA
Construction worker number	CWN
CWN-Skilled workers	CWN-SW
CWN-Certificate of civil expertise (including CWN-SE)	CWN-CC
CWN-Safety expert	CWN-SE
Construction accident rate	CAR
Construction safety rate	CSR

The Construction Accident Rate (CAR) is defined as the percentage of CWAs relative to the CWN and is described using Equation 7 . The Construction Safety Rate (CSR) measures the proportion of CWN-SE (safety experts) responsible for workplace safety relative to CWAs and is computed using Equation 8 . Both CAR and CSR are calculated from 2020 CWA data to assess their distribution relative to the CWN in the same year. CWA and CWN-SE counts are categorized by province to enable province-level analysis.

C A R = \frac{N u m b e r_o f_C W A}{T o t a l_C o n s t r u c t i o n_W o r k e r} = \frac{C W A}{C W N - S W + C W N - C C}

(7)

C S R = \frac{C o n s t r u c t i o n_W o r k e r_N u m b e r_S a f e t y_E x p e r t}{N u m b e r_o f_C o n s t r u c t i o n_W o r k e r_A c c i d e n t} = \frac{C W N - S E}{C W A}

(8)

Chi-square test

Table 4 reports the results of the chi-square goodness-of-fit tests and tests of independence. The goodness-of-fit tests evaluated the distribution of CWA categories—CWA-TP, CWA-LP, CWA-TB, and CWA-YA—against a uniform expectation. All tests were significant (p < 0.001), indicating that the observed distributions deviated substantially from uniformity. The notes column provides additional context: new projects exhibited the highest share of CWA-TP; East Java, DKI Jakarta, and West Java reported the highest CWA-LP counts; industrial building projects had the most CWA-TB incidents; and CWA-YA cases peaked in 2019.

Table 4.

Chi-square and Fisher's exact test analysis results.

Analysis	Variable	Chi-Square x²	p-value
Chi-square goodness of fit test	CWA-TP	20,127	<0.001
	CWA-LP	46,082	<0.001
	CWA-TB	11,389	<0.001
	CWA-YA	420	<0.001

Variable Pairs	Fisher's Exact p-value	Note
CWA-TP & CWA-TB	< 0.001	Significant
CWA-TP & CWA-LP	< 0.001	Significant
CWA-TP & CWA-YA	0.123	Not Significant
CWA-TB & CWA-LP	< 0.001	Significant
CWA-TB & CWA-YA	< 0.001	Significant
CWA-LP & CWA-YA	< 0.001	Significant

The results of Fisher's Exact Test (see Table 4) indicate that the vast majority of pairs of CWA categories are related in a statistically significant way. Of the six pairs analyzed, five pairs showed a significant relationship (p < 0.05), namely CWA-TP × CWA-TB, CWA-TP × CWA-LP, CWA-TB × CWA-LP, CWA-TB × CWA-YA, and CWA-LP × CWA-YA. In comparison, the CWA-TP × CWA-YA pair did not show a significant relationship (p = 0.123). This result indicates that the relation between the CWA categories is dominant but not completely evenly distributed.

Overall, the findings indicate significant associations among most CWA variables, suggesting that types of construction accidents may be interrelated. Specific project types, locations, building types, and time periods exhibit statistically higher accident frequencies, offering policymakers and construction managers practical guidance for implementing targeted safety interventions.

Spatial distribution of CWAs

As shown in Figure 4, the highly populated province of East Java recorded the highest Accident Index during 2014–2020, indicating comparatively lower construction safety performance than other regions. This pattern may reflect the province's large volume of construction activity, as East Java is among the largest provinces, and it may also relate to the distribution of safety personnel. Although substantial construction activity occurs there, Figure 3 shows that East Java ranks ninth in the number of safety experts, with only 566 CWN-SE, suggesting a potential association between the availability of safety professionals and accident frequency.

Figure 4.

Accident Index across the province.

Figure 4 further summarizes construction activity conditions across provinces and flags those with an Accident Index>1, where an Accident Index of 1 is considered baseline, identifying 12 provinces with elevated risk. These findings call for heightened attention from construction stakeholders in higher-risk provinces to improve occupational safety.

Time-Series analysis

Figure 5(a) shows that the number of CWAs typically peaks in the fourth quarter (October–December) of each year. This pattern may reflect seasonal weather, with Indonesia's rainy season generally beginning in October and intensifying in November and December, and corresponding project scheduling pressures as many construction projects approach year-end completion. These observations underscore the need for stakeholders, contractors, site managers, and workers to anticipate seasonal and deadline-driven risks and take appropriate measures to protect workers.

Figure 5.

(a) The three-month Moving Average compared to the number of accidents; (b). Seasonal Indices from 2014 to 2020.

Figure 5(b) presents the seasonal indices by month and year. The fluctuations appear largely random, suggesting no substantial unmodeled temporal variation. Peak accident periods recur approximately every 2–3 months, implying that main safety initiatives would be most effective if scheduled at regular intervals to mitigate CWAs.

Comparative statistical analysis

To assess variation in CWAs across building types, we conducted a Welch's ANOVA with 13 categories, each comprising seven observations. The analysis examined whether the mean number of CWAs varied by building type, as shown in Table 5.

Table 5.

Levene's and Welch's ANOVA test results for building types.

Statistics test		Value	Notes
Levene's Test	Test (Based on Mean)	6759	Levene's Test indicates that the homogeneity of variances assumption is violated. Therefore, the use of conventional One-Way ANOVA is not recommended as the primary basis for inference due to the potential for increasing Type I errors. To address this, the analysis continued using Welch's ANOVA, which is more robust to violations of the homogeneity-of-variance assumption.
	df1	12
	df2	78
	p	<0.001
Welch's ANOVA	Welch F	55.056	There are significant differences in average building types.
	df1	12
	df2	29.474
	p	< 0.001

The Welch's ANOVA indicated a statistically significant difference in accident frequency by building type, F(12,29.474) = 55.056, p < 0.0001. Each group had the same number of observations (N = 7). The highest mean values were found in Industrial Factory (M = 368.14), Transportation Building Facility (M = 366.29), and Commercial Building (M = 354.14). Conversely, the lowest mean values were found in the categories of Cultural and Entertainment Facility Building (M = 14.29) and Worship Facility Building (M = 11.71). Overall, the total mean was 164.42 with a standard deviation of 148.78, indicating significant variation across building types (Figure 6).

Figure 6.

ANOVA result of: (a) The average number of CWAs by building type; (b) The variance in the number of CWAs by building type.

Figure 7(a) presents the average number of CWAs by building type. Transportation building facilities, industrial factories, residential buildings, and commercial buildings show the highest average accident counts (all above 350), indicating that these project types pose the greatest worker safety challenges. By contrast, worship facilities and cultural and entertainment facilities report the lowest averages (below 15), consistent with comparatively lower risk exposure or potential underreporting of minor incidents.

Figure 7.

Post hoc test result by Games–Howell for building type: (a) Group means with 95% confidence intervals; (b) distribution comparison; (c) distribution shapes; and (d) significance matrix.

Figure 7(b) further indicates that building type significantly influences the number of CWAs, underscoring the need for targeted investigation into specific risk factors and tailored mitigation strategies for higher-risk categories. For example, transportation building facilities exhibit both the highest variance (19,037.24) and the highest average accident count (366.29), whereas lower-risk categories, such as worship facilities, show substantially lower frequencies.

Further Games–Howell test results indicate that differences between building types are broad and consistent. Of the 78 paired comparisons, the majority of pairs showed statistically significant differences (p < 0.05), while only a small number of pairs were not significant (p ≥ 0.05). This finding indicates that mean differences between groups are not limited to a few extreme categories, but occur across building types. Specifically, Commercial Building (1) had a significantly higher mean score than Cultural and Entertainment Facility Building (2) (MD = 339.86, p < 0.001; 95% CI [261.92, 417.80]) and Worship Facility Building (13) (MD = 342.43, p < 0.001; 95% CI [291.06, 393.80]). Furthermore, the Industrial Factory (4) also showed a significant difference compared to Water Building (12) (MD = 295.29, p < 0.001; 95% CI [240.86, 349.73]) and Worship Facility Building (13) (MD = 356.43, p < 0.001; 95% CI [307.86, 405.00]). In addition, Transportation Building Facility (11) was significantly different from Cultural and Entertainment Facility Building (2) (MD = 351.86, p < 0.001; 95% CI [287.92, 415.80]) and the Water Building (12) (MD = 293.43, p < 0.001; 95% CI [237.06, 349.80]). Overall, this significant dominance of comparisons confirms that the building type factor has a strong and systematic influence on the variation of the analyzed values. Figure 7 presents a visualization of the result, where (a) presents the result of the mean separation with 95% CI between the top and low clusters; (b) and (c) display the shifts in location and distribution; while (d) presents the pattern of green cells representing statistically significant results, predominantly between clusters, and red cells representing non-significant results, predominantly within clusters. To sum up, a small fraction of pairs with similar means showed no significant difference, and the Games–Howell cluster analysis shows that the three building types (Commercial Building, Factory Industrial, and Transportation Building Facility) with the largest means consistently dominate the CWAs distribution. There are a number of intermediate subclusters, that are similar to each other but differ from the benchmark.

Table 6 shows that Levene's test for homogeneity of variance revealed that the assumption was not met (Levene = 15.474, df1 = 2, df2 = 18, p < 0.001). Because of heteroscedasticity, the analysis was conducted using Welch's ANOVA, a more robust test when variances across groups are unequal. Welch's ANOVA (Robust Tests of Equality of Means) revealed a significant difference in the mean Value variable between project types, Welch F(2, 8197) = 140.156, p < 0.001. As a result, we can conclude that the average value varies significantly between at least two groups: new, renovation, and maintenance.

Table 6.

Levene's and Welch's ANOVA test results for project types.

Statistics test		Value	Notes
Levene's Test	Test (Based on Mean)	15,474	Levene's test showed a p-value of < 0.001, indicating that the assumption of homogeneity of variances was not met (heteroscedasticity). As a result, the analysis was continued using Welch's ANOVA, which is more robust against violations of this assumption.
	df1	2
	df2	18
	p	<0.001
Welch's ANOVA	Welch F	140,156	The Welch ANOVA results indicate a significant difference in mean values between project types (p < 0.001). As a result, at least two of the three project types (New, Renovation, and Maintenance) have statistically significant differences in values.
	df1	2
	df2	8197
	p	< 0.001

Figure 8 shows pronounced disparities in CWAs across project types: new projects exhibit the highest mean accident counts and the greatest variance, renovation projects display intermediate levels, and maintenance projects show the lowest mean count and variability. These patterns suggest that new construction is associated with substantially higher accident risk than renovation or maintenance, underscoring the need to strengthen preventive controls and safety protocols in new-build settings.

Figure 8.

ANOVA result of the (a) average, and (b) variance of CWAs by project type.

The Games-Howell test results revealed that all pairs of project types differed significantly, with three out of three comparisons (100%) being significant and none being insignificant (0%). Project Type New (1) had significantly higher mean values than Renovation (2), with a mean difference of 1658.57 (p < 0.001) and a 95% confidence interval of [1262.86; 2054.28]. New (1) showed a significant difference from Maintenance (3), with a mean difference of 1834.14 (p < 0.001) and a 95% confidence interval of [1438.01; 2230.28]. Renovation (2) had a significantly higher mean value than Maintenance (3), with a mean difference of 175.57 (p < 0.001) and 95% confidence interval of [123.04; 228.11]. The confidence intervals for all three comparisons did not exceed zero, indicating that the differences between the groups were statistically significant and consistent. Thus, the Games-Howell post hoc test results confirmed that all project types differed significantly from one another, with the mean values in the following order: New > Renovation > Maintenance.

Figure 9 shows the results of the post hoc Games-Howell test. In (a), it is evident that there is no overlap for the new type of project relative to renovation and maintenance-type projects. In (b) and (c), show that there is a pronounced separation of distributions for the new type of project. Where (d) shows that there is a significant difference for ‘new-maintenance’ and ‘new-renovation’ pairs, and ‘maintenance-renovation’. This finding supports the conclusion that the differences in values across project types are statistically significant and are not affected by violations of the homogeneity-of-variance assumption, with new having the highest mean, followed by renovation and maintenance. The consistency of these findings supports the validity of the inference under heteroscedasticity conditions. The large effect sizes for the new type of project relative to renovation and maintenance-type projects are evident. In conclusion, the new type of project has a statistically and practically higher number of CWAs than the other two types, whereas maintenance and renovation types of projects do not differ significantly from each other.

Figure 9.

Post hoc Games-Howell test for type of project: (a) Group means with 95% confidence intervals; (b) Distribution comparison; (c) Distribution shapes; and d) Significance matrix.

Correlation analysis

Table 7 presents the results of the Pearson correlation analysis, descriptive statistics, and negative binomial regression. The PCV's sign denotes direction (positive/negative), and its magnitude is interpreted as low, moderate, or high according to predefined thresholds. The analysis indicates a positive but statistically non-significant correlation between CWAs and CWN-SE (PCV=0.015, p = 0.934). Similarly, the correlation between CWAs and CWN-SW was positive but not significant (PCV = 0.244, p = 0.170). The association between CAR and CSR was slightly negative (PCV = −0.034) and statistically non-significant (p = 0.846).

Table 7.

Pearson correlation and Negative Binomial Regression analysis results.

Analysis	Variable	PCV	Sig. (2-tailed)	Strength of correlation
Pearson Correlation Analysis	CWAs & CWN-SE	+0.015	0.934	Positive low degree
	CWAs & CWN-CC	+0.368*	0.035	Positive moderate degree
	CWAs & CWN-SW	+0.244	0.170	Positive low degree
	CAR & CSR	−0.034	0.846	Negative Low degree
*Sig. 0.05 level (2-tailed); Strength of correlation: Low (< ± 0.29); Moderate (±0.30 to ±0.49); High (±0.50 to ±0.99); or Perfect (±1)

Variable	Variance	Mean	Judgement	Notes
CWAs	7665.523	50.09	Variance greater than the mean	Overdispersion is proven → Poisson is unsuitable → Negative Binomial justified.
CWN-CC	131881706	7873.09
CWN-SW	759031156.09	20734.18
CWN-SE	1467359.608	710.29
CAR	313.805	15.79
CSR	71.758	20.00

Variable	B	Sth. Error	Wald χ²	p-value	IRR (Exp(B))	95% CI IRR
Intercept	2.331	0.296	61.821	< 0.001	10.287	[5.754 ; 18.392]
CWN-CC	0.000	0.000	18.878	< 0.001	1.000	[1.000 ; 1.001]
CWN-SW	0.000	0.000	3.542	0.060	1.000	[1.000 ; 1.001]
CWN-SE	−0.004	0.001	15.440	< 0.001	0.996	[0.994 ; 0.998]

Note: IRR = Incidence Rate Ratio; CI = Confidence Interval.

The Pearson Chi-Square (df = 1208) and Deviance (df = 1211) values are close to one, indicating that the negative binomial model fits the data well and effectively manages overdispersion in the construction worker accident data. Based on the descriptive statistics (see Table 7), the variance exceeds the mean, indicating overdispersion and suggesting that a negative binomial model is preferable to a Poisson model.

Table 7 shows the results of a negative binomial regression analysis of CWA. Statistically, the CWN-CC is strongly related to the number of accidents. However, the IRR value is very close to one, implying that the effect per additional unit is negligible in practice. After controlling for other variables, the CWN-SW does not have a significant effect on the number of accidents reported. This finding suggests that the relationship observed in the initial correlation analysis does not persist in the multivariate model, implying potential confounding or an indirect relationship. The CWN-SE variable has a significant and consistent negative impact on the number of accidents. The IRR of 0.996 indicates that each additional safety expert is associated with a 0.4% reduction in CWA, after controlling for CWN-CC and CWN-SW. Because the confidence interval does not exceed one, the effect is stable and statistically robust.

The CWN-SE has a negative and significant effect on the CWAs (IRR = 0.996; p < 0.001). This suggests that increased safety capacity contributes to reduced accident rates. After controlling for other variables, however, there is no significant effect on the CWN-SW.

Discussion

This study developed a province-scale model of CWAs that occurred in Indonesia between 2014 and 2020. The model encompasses 14,962 CWAs and examines the relationship between accident occurrences and factors including workforce, projects, and seasonal effects. The study found that the regional differences are significant, with East Java reporting the highest Accident Index and the lowest CWN-SE than the other provinces. The study also found that transportation facilities, industrial factories, residential, and commercial building projects have the highest average CWAs, and that new projects have higher accident occurrences and greater variation than renovation and maintenance projects. Regarding seasonal effects, the study found that CWAs are higher between October and December.

Based on the correlational analysis, the study supports a systems-based, rather than solely individual-based, approach to safety policy. Regulations mandating a minimum ratio of safety experts to the workforce, increasing the authority of safety experts in operational decision-making, and integrating safety functions across all stages of a construction project have the potential to significantly reduce workplace accidents. Furthermore, these findings emphasize the importance of using appropriate regression models for count data in occupational safety research, as this approach can uncover relationships not detected through simple correlation analysis. Thus, this study provides empirical evidence that investing in professional safety capacity is an effective and sustainable policy strategy for improving construction worker safety and offers a stronger analytical basis for formulating data-driven safety policies.

This Discussion section is organized to directly interpret and synthesize the empirical evidence presented in Section 3. Each of the six discussion themes is generated from different but related analytical findings. The empirical basis for the discussion on regional disparities and the deployment of safety professionals (Section 4.1) is the spatial distribution analysis of CWAs (Section 3.2) and the provincial variation in workforce characteristics. The descriptive and comparative analyses of accident distribution by building type and project characteristics (Sections 3.1 and 3.4) serve as inputs to the discussion of building type and project stage effects (Section 4.2).

The time-series and seasonal analyses (Section 3.3) provide the basis for the discussion on seasonality and operational planning (Section 4.3). The correlation and regression analyses (Section 3.5) provide evidence for the discussion of workforce composition, certification, and their differentiated relationships with accident outcomes (Section 4.4). These evidence-based insights are synthesized to derive implications for policy and practice (Section 4.5) and to formulate targeted policy recommendations (Section 4.6). The discussion themes are all explicitly linked to specific analytical results, so the interpretation is firmly grounded in the empirical findings.

Regional disparities and staffing of safety professionals

The concentration of CWAs in highly urbanized and industrialized provinces is consistent with the literature that identifies rapid infrastructure expansion and complex project portfolios as key drivers of elevated construction risk in Indonesia. Recent case studies of Indonesian construction accidents highlight road and heavy civil works as prominent contributors to the accident burden,¹⁸ aligning with the finding that transportation-related building types are high-risk for CWAs.

The results also suggest that provinces with intense construction activity may not have commensurate safety staffing (CWN-SE), raising concerns about on-site safety oversight capacity. This aligns with evidence that safety climate and management systems in Indonesia remain unaligned, with persistent gaps in enforcement, communication, and resource allocation that can undermine safety performance. Strengthening safety leadership, monitoring, and resource support, particularly in high-Accident Index provinces, has been recommended in prior Indonesian studies.⁴⁰

The observed mismatch between accident burden and safety-expert availability is particularly significant in the context of Indonesia's competency-based safety certification framework. The Indonesian Ministry of Public Works regulation that structures sub-qualifications for skilled workers and experts (including CWN-SE) provides a policy lever for scaling qualified personnel to match risk. Tailoring deployment of certified experts to high-risk provinces would be consistent with the regulation's intent to ensure competency at the point of work.

Building type and project stage effects

The ANOVA and descriptive results show statistically significant differences in CWAs across building types and project types, with transportation facilities and new projects standing out for both mean accidents and variance. These findings are supported by the observation that new-build settings typically involve heavy equipment, structural works, lifting operations, and dynamic interfaces among subcontractors, factors that international and Indonesian research repeatedly link to elevated incident rates.^18,41 In Indonesia specifically, roadway and heavy equipment interactions have been implicated in a disproportionate share of severe events.^18,42 More broadly, studies tie rework, complexity, and production pressure to higher injury rates, conditions more typical of large, new projects than of routine maintenance.²⁴

From a safety control standpoint, these differences argue for risk-tiered interventions by project type. For transportation facilities and other heavy works, stricter traffic separation, plant-person interface controls, lifting plans, and machinery authorization regimes are priority measures. For new-build projects, upstream design-for-safety, constructability reviews, and enhanced pre-task planning can reduce error-provoking conditions before they reach the site.⁴¹

Seasonality and operational planning

When interpreting these temporal patterns of accidents, it is important to consider potential reporting bias, especially for small-scale projects and certain types of facilities, such as religious sites. Safety management systems and accident reporting mechanisms are frequently under-institutionalized in these project categories, either due to a lack of resources, limited regulatory oversight, or the prevalence of informal work practices. As a result, some accidents, particularly minor ones, may go unreported, and the observed accident rate may reflect low reporting rather than true safety performance. This is especially important in seasonal analyses, because variations in recorded accidents are likely influenced not only by environmental and operational factors, but also by differences in reporting capacity across project types and construction scales.

The fourth quarter peaks (October–December) in CWAs coincide with Indonesia's wetter months and common year-end deadline pressures. While the data are not exposure-adjusted, this pattern is consistent with evidence that precipitation and adverse weather degrade site safety conditions, raising the likelihood of slips, falls, equipment instability, and operational deviations. As Manzoor et al.⁴³ noted, adverse weather conditions can increase accident risks, while Zermane et al.⁴⁴ noted that end-of-year project deadlines may intensify worker pressure, thereby increasing the likelihood of CWAs. Systematic reviews underscore precipitation and high winds among the most consequential weather factors for construction performance and safety, supporting seasonal hazard anticipation in planning.⁴⁵

Given the finding that accident peaks recur roughly every 2–3 months, it is pragmatic to align safety campaigns, planned training, and joint inspections on a similar cadence, front-loading activities in the early rainy season, and again before known workload spikes. External guidance for Asia also emphasizes climate-aware risk management and weather-contingent scheduling, which can be adapted to local contexts.⁴⁶

Workforce composition, certification, and the correlational results

Most pairwise correlations were weak and statistically non-significant, except for a positive association between CWAs and CWN-CC. This outcome likely reflects confounding by construction scale, in which provinces employing more certified civil experts may also host more complex projects, leading to more serious accident counts even if per-exposure risk is not elevated. The null associations between CWAs and CWN-SW or CWN-SE could similarly mask countervailing influences (e.g., better staffing where risk is higher), measurement noise, or insufficient variation relative to other determinants (project mix, supervision intensity, subcontracting structure). Prior Indonesian work documents gaps between formal policy and field implementation and highlights that safety outcomes hinge on management commitment, planning, and a robust safety climate, with all being factors not captured by headcounts alone.⁴⁷

The positive correlation between the CWN-CC and the CWAs does not necessarily imply that certified civil workers increase the risk of accidents; rather, it reflects the greater complexity and intensity of activities in large-scale projects. Projects that use a large number of CWN-CC are usually high-value, long-duration, and involve complex technical systems and are therefore inherently more exposed to the risk of occupational accidents. Furthermore, formal and strict mechanisms for reporting accidents are more common in large projects with CWN-CC. The regulatory requirements and better governance practices could lead to an increase in accident reporting, not because there are more accidents, but because incidents that were not previously recorded on smaller projects are now being systematically recorded. Hence, the positive correlation observed in the Pearson analysis is likely a reflection of the intensity of reporting and project exposure, rather than a direct causal relationship.

This interpretation is also consistent with the results of the Negative Binomial Regression, in which the effect of CWN-CC is no longer substantially practical after controlling for other variables. The IRR value very close to one indicates that adding certified civil engineers alone does not significantly decrease the accident rate. The role of CWN-SE, by contrast, remains relevant and consistent, confirming that not only the project's technical capacity but also a structured, professional safety function are key elements in controlling occupational accidents.

These results highlight the need to differentiate between indicators of project complexity and their causal determinants. The positive association between CWN-CC and accidents in the exploratory phase should be understood as an indicator of the large-scale project setting and more intense reporting practices. The count-based regression analysis, on the other hand, shows that investing in professional safety capacity is a more effective approach to reducing the risk of CWAs. These results caution against interpreting raw staffing numbers as protective without considering deployment, role authority, and integration into site safety decision-making. Evidence from broader settings suggests that training and competency matter, but effectiveness depends on design and reinforcement (e.g., active hazard-recognition training, simulation, feedback loops), as well as refresher frequency to mitigate knowledge decay.⁴⁸ This strengthens the case for targeted, high-quality training and empowered safety roles rather than relying on nominal counts.⁴⁹

Policy and practice implications

▪ Targeted deployment of safety expertise: Prioritize allocation of certified safety experts to provinces and project types with an elevated AI and variance, supported by auditing and coaching functions that can influence work planning, permits-to-work, and contractor oversight. This operationalizes the competency framework set by PUPR for impact at the site level. Chan et al.²⁴ observed that urbanized areas undergoing extensive construction and redevelopment experience complex interactions among project types and stakeholders, raising safety challenges. Similarly, Xu and Xu⁵⁰ emphasize region with strong economic development generally maintain better safety conditions and highlight the importance of safety standards. These results highlight the need for attention from construction stakeholders, especially local government, as policymakers in East Java, to improve construction safety.

▪ Risk-tiered controls for high-risk project types: Risk of transportation facilities construction⁵¹ and large new-build projects,²⁴ tighten plant/traffic risk management, lifting operations, and interface controls; incorporate design-for-safety reviews and constructability assessments to remove hazards upstream.

▪ Seasonal safety operations: Institute rainy-season playbooks: advance procurement of drainage and housekeeping supplies, weather-triggered stop-work criteria, and 2–3-monthly safety campaigns with targeted toolbox talks on slips, equipment stability, and electrical hazards. Align rosters and supervision levels with anticipated weather and deadline peaks.

▪ Training quality, retention, and safety enhancement: Shift from compliance-oriented training to methods shown to improve hazard recognition and retention (e.g., visual cueing, VR/AR simulations, adult-learning principles) and schedule refreshers to counter information decay. A key component of measuring success is tracking training transfer to the job via leading indicators.⁴⁸ Furthermore, reckless behavior is a leading cause of construction accidents, and effective mitigation strategies include job hazard analyses and pre-task safety briefings.^10,52 Most accidents result from multiple contributing factors rather than a single cause, necessitating comprehensive safety management practices and stakeholder coordination. Collective efforts are essential to reinforce safety regulations, improve enforcement, and cultivate a robust safety culture.⁵³ Moreover, adopting systematic safety management approaches can yield substantial improvements in safety outcomes within the construction sector.⁵⁴

Policy recommendations

Policy Recommendations for Reducing CWAs and Strengthening Safety Performance

Risk-Based Deployment of Certified Safety Experts

Certified safety experts should be deployed on a risk-based basis to provinces and project types with an elevated Accident Index. These experts must be granted clear on-site authority to approve permits and make stop-work decisions, ensuring that organisational safety considerations are integral to project execution. (b)

Implementation of Risk-Tiered Controls

In high-risk environments, particularly transportation facilities and new-build projects, it is essential to enforce stricter plant-person interface management, ensure competent lift planning, and conduct comprehensive design-for-safety reviews. These controls will help to proactively identify and manage hazards. (c)

Institutionalisation of Season-Aware Operations

Effective safety management should include adopting season-aware operational practices. This involves establishing rainy-season playbooks, setting weather-triggered stop-work criteria, and conducting refresher safety campaigns every 2–3 months to correspond with known weather accident peaks. (d)

Upgrading Training Methods

Training programmes should transition from compliance-focused briefings to evidence-based, retention-oriented approaches. This includes active hazard recognition exercises, utilisation of VR/AR simulations, and regular testing and feedback sessions. Scheduled refresher training should be implemented to reinforce knowledge and skills. From a policy perspective, the adoption of advanced training technologies such as VR/AR should be strategically phased and risk-targeted, given the trade-off between implementation costs and safety benefits. The initial investment for VR/AR-based training is higher than traditional training methods, but it may be given priority for high-risk project types, complex construction activities, and safety-critical tasks where the potential reduction in accidents provides the highest return on investment. Over time, reductions in accident-related costs (medical costs, project delays, insurance claims, productivity losses) can offset some of the initial costs, making technology-enhanced training a cost-effective safety intervention when applied selectively and strategically. (e)

Enhancing Data Integrity

To accurately assess risk and safety performance, it is vital to improve data collection and integrity. This can be achieved by tracking exposure data, such as worker-hours, which allows for the calculation of exposure-adjusted incidence and severity measures (e.g., CAR). Additionally, definitions should be standardised, and appropriate small-cell remedies applied. (f)

Intensifying Controls for Roadway and Heavy-Equipment Hazards

Stronger controls are required to manage roadway and heavy-equipment risks. Measures include establishing exclusion zones, ensuring only certified operators and banksmen are used, and implementing proximity alert systems to reduce the likelihood of accidents. (g)

Strengthening Construction Safety Management Systems

The implementation of construction safety management systems should be reinforced through audit-linked procurement incentives and continuous review and corrective action processes. This will drive consistent improvement and accountability in safety performance. (h)

Building Subcontractor Capacity

Safety capacity building for subcontractors is crucial. This can be achieved by providing training support, access to shared safety resources, and enhanced oversight from prime contractors. These steps ensure that safety standards are maintained across all tiers of project delivery. (i)

Publishing Provincial Dashboards

To sustain benchmarking and organisational learning, quarterly provincial dashboards should be developed and published. These dashboards must integrate both leading indicators (such as near-miss density and pre-task plan quality) and lagging indicators (including CWA counts, exposure-adjusted CAR, and severity).

The safety recommendations outlined above are illustrated in Figure 10.

Figure 10.

Key points and the implementation roadmap of recommendations.

Conclusion

This study provides the first province-level analysis of CWAs in Indonesia, utilizing a dataset comprising 14,962 records from 2014 to 2020. The findings reveal significant spatial disparities, with industrialized provinces such as East Java and Jakarta bearing disproportionate accident burdens. New construction projects, particularly in the transportation, industrial, residential, and commercial sectors, have been shown to pose a higher risk than renovation and maintenance projects. The occurrence of seasonal peaks in October–December, in conjunction with recurrent accident cycles at 2–3-month intervals, further emphasises the necessity of season-aware planning strategies. While the correlations between workforce indicators and accident rates were generally weak, the significant association between CWAs and CWN-CC highlights the importance of aligning workforce composition with project complexity. In practice, these results provide a clear empirical basis for policy makers and project managers to prioritize risk-based safety interventions, align the deployment of safety experts with project characteristics, and manage time for inspection and training more effectively and efficiently under resource constraints.

It is recommended that certified safety experts be deployed risk-based, that tiered controls be implemented for high-risk project types, that scheduling be season-aware, that training be evidence-based, that reporting be exposure-based, that the implementation of CSMS be strengthened, and that provincial dashboards be made transparent. These measures have the potential to address oversight gaps and enhance safety outcomes across a range of project contexts. Concurrently, limitations, including the lack of exposure-adjusted metrics, underreporting bias, enforcement variability, and masked project-level heterogeneity, impede extrapolation of findings to broader populations. By explicitly recognizing these limitations, this study highlights that the presented results should be considered a basis for informed policy decisions, as well as a foundation for the construction of more comprehensive safety data and evaluation systems in the future. It is recommended that subsequent research adopt exposure-based count models, incorporate project-level covariates, and utilise mixed-methods and quasi-experimental designs to evaluate targeted interventions.

The integration of provincial-level evidence with practical policy measures constitutes a significant contribution to developing a more systematic understanding of construction risk in Indonesia. Addressing the identified limitations and implementing the recommended strategies will be critical to reducing accidents, strengthening safety management systems, and safeguarding the workforce amid ongoing infrastructure expansion. This study is limited by the lack of systematic data on informal or uninsured CWNCWNCWN, who are probably at greatest risk of accidents, but are not represented in official records. Therefore, the accident patterns reported in this study are probably only a partial picture of the actual situation, especially in small-scale projects and the informal sector.

Footnotes

Acknowledgements

The authors would like to thank all individuals and institutions who supported this study.

ORCID iDs

Bangun Marpaung

Wei-Tong Chen

Ying-Hua Huang

Hew Cameron Merrett

Ethical considerations

Not applicable.

Informed consent

Informed consent was waived because the study used anonymized secondary data and involved no direct contact with participants.

Author contributions

Conceptualization: Wei-Tong Chen, Bangun Marpaung, and Ying-Hua Huang; Methodology: Bangun Marpaung and Wei-Tong Chen; Validation: Bangun Marpaung and Hew Cameron Merrett; Formal analysis: Bangun Marpaung; Investigation: Bangun Marpaung and Hew Cameron Merrett; Resources: Wei-Tong Chen and Ying-Hua Huang; Data curation: Bangun Marpaung Writing – original draft: Bangun Marpaung; Writing – review & editing: Wei-Tong Chen, Ying-Hua Huang, and Hew Cameron Merrett; Supervision: Wei-Tong Chen and Ying-Hua Huang; Project administration: Bangun Marpaung.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The datasets generated during and/or analyzed during the current study are available from the corresponding author on request.

References

Lopes

Banaitienė

. A model for construction sector development in middle-income sub-Saharan African countries. Technol Econ Dev Econ 2024; 30: 1229–1255.

Bilim

. Demographic analysis of occupational safety in the construction sector: strategies and insights for risk reduction. Buildings 2025; 15: 28.

Abukhashabah

Summan

Balkhyour

. Occupational accidents and injuries in construction industry in Jeddah city. Saudi J Biol Sci 2020; 27: 1993–1998.

Carrillo-Castrillo

Trillo-Cabello

Rubio-Romero

. Construction accidents: identification of the main associations between causes, mechanisms and stages of the construction process. Int J Occup Saf Ergon 2017; 23: 240–250.

Hamid

ARA

Noor Azmi

MRA

Aminudin

, et al. Causes of fatal construction accidents in Malaysia. In: IOP conference series: earth and environmental science. 220. IOP Publishing, 2019, pp.012044.

Lee

Kim

, et al. Trend analysis of construction industrial accidents in Korea from 2011 to 2015. Sustainability 2017; 9: 1297.

Kang

Ryu

. Predicting types of occupational accidents at construction sites in Korea using random forest model. Saf Sci 2019; 120: 226–236.

Khodabandeh

Kabir-Mokamelkhah

Kahani

. Factors associated with the severity of fatal accidents in construction workers. Med J Islam Repub Iran 2016; 30: 469.

Chan

Cristofaro

Nassereddine

, et al. Perceptions of safety climate in construction projects between workers and managers/supervisors in the developing country of Iran. Sustainability 2021; 13: 10398.

10.

Yoon

Ahn

Yum

, et al. Establishment of safety management measures for major construction workers through the association rule mining analysis of the data on construction accidents in Korea. Buildings 2024; 14: 98.

11.

Kaming

Olomolaiye

Holt

, et al. What motivates construction craftsmen in developing countries? A case study of Indonesia. Build Environ 1998; 33: 131–141.

12.

Kim

Son

Yum

, et al. Analyzing the risk of safety accidents: the relative risks of migrant workers in construction industry. Sustainability 2020; 12: 5430.

13.

Chan

. Characterizing the relationship between personality traits and safety motivation among construction workers. Heliyon 2023; 9: e20370.

14.

Cao

Chen

Cao

. Effect of occupational health and safety training for Chinese construction workers based on the CHAID decision tree. Front Public Health 2021; 9: 623441.

15.

Zahoor

Chan

Gao

, et al. The factors contributing to construction accidents in Pakistan: their prioritization using the Delphi technique. Eng Constr Archit Manage 2017; 24: 463–485.

16.

Manik

Putri

Azzahra

, et al. Sustainability in construction work: a perspective on accidents and building failure mitigation in Indonesia. In: IOP Conference series: earth and environmental Science. IOP Publishing, 2021, Vol. 794, No. , pp.012065.

17.

Arman

Suraji

Hidayat

. A conceptual framework for the process protocol of integration quality, safety, health and sustainable construction management systems for infrastructure projects in Indonesia. In: E3s web of conferences. EDP Sciences, 2025, Vol. 604, pp.16006.

18.

Bria

Chen

Muhammad

, et al. Analysis of fatal construction accidents in Indonesia—A case study. Buildings 2024; 14: 1010.

19.

Latupeirissa

Wong

ILK

Tiyow

. Causes of work accidents and its impact on the road and bridge construction projects. In: IOP Conference series: earth and environmental Science. IOP Publishing, 2021, November, Vol. 907, No., pp.012023.

20.

Machfudiyanto

Chen

Latief

, et al. Applying association rule mining to explore unsafe behaviors in the Indonesian construction industry. Sustainability 2023; 15: 5261.

21.

Nugroho

Sunbara

. Work accident analysis in the construction project of PT. XYZ. Int J New Technol Res 2021; 7: 44–56.

22.

Megasari

. Analysis of work accidents and work accident benefits in 2016 in east Java. Indones J Occup Saf Health 2022; 11: 248–255.

23.

Suti

Fisu

Dani

AAH

, et al. Work accident risk at hospital construction environment. In: Proceedings of the International Conference on Industrial Engineering and Operations Management, Harbin, China, 2021.

24.

Chan

Yang

Choi

, et al. Characteristics and causes of construction accidents in a large-scale development project. Sustainability 2022; 14: 4449.

25.

Peck

Olsen

Devore

. Introduction to statistics and data analysis. Third edition. United States, 2008.

26.

Rolke

Gongora

. A chi-square goodness-of-fit test for continuous distributions against a known alternative. Comput Stat 2021; 36: 1885–1900.

27.

Tallarida

Murray

. Manual of pharmacologic Calculations with Computer Program. Second ed. Temple University School of Medicine, 1981.

28.

Tan

Steinbach

Karpatne

, et al. Introduction to data mining. Second Edition. Pearson Education, 2019.

29.

Sheridan

Ahmed

Farooque

. A Fisher’s exact test justification of the TF–IDF term-weighting scheme. Am Stat 2026; 80: 146–156.

30.

Tran

SVT

Khan

Lee

, et al. A hazard identification approach of integrating 4D BIM and accident case analysis of spatial–temporal exposure. Sustainability 2021; 13: 2211.

31.

Duan

Zhou

Goh

. Spatial-temporal analysis of safety risks in trajectories of construction workers based on complex network theory. Adv Eng Inform 2023; 56: 101990.

32.

Direktorat Keberlanjutan Konstruksi Kementerian PUPR official website accessed September 12th, https://binakonstruksi.pu.go.id/dashboard-tenaga-ahli/ (2021).

33.

Cayo

TGF

Torres-Cruz

. Time Series Analysis Applied to Notifications of Work Accidents. arXiv preprint arXiv:2307.13748. 2023.

34.

Meng

Liu

Feng

. Prediction of construction and production safety accidents in China based on time series analysis combination model. Appl Sci 2022; 12: 11124.

35.

Guo

Tang

Liang

, et al. Comparative analysis of the patterns of unsafe behaviors in accidents between building construction and urban railway construction. J Constr Eng Manag 2021; 147: 04021027.

36.

Honda

Pelepenko

Monteiro

, et al. Fibre post removal using ultrasonic tips: a comparative in vitro study using different protocols. Aust Endod J 2025; 51: 47–54.

37.

Guo

Liu

, et al. Research on influencing factors of carbon emissions in the regional construction industry: a case study of jiangxi province. Sustainability 2026; 18: 69.

38.

Chan

Aghimien

. Safe working cycle: is it a panacea to combat construction site safety accidents in Hong Kong? Sustainability 2022; 14: 94.

39.

Wang

Zhao

Yang

, et al. Remaining useful life prediction of rolling bearings based on Pearson correlation-KPCA multi-feature fusion. Measurement ( Mahwah N J) 2022; 201: 111572.

40.

Lestari

Sunindijo

Loosemore

, et al. A safety climate framework for improving health and safety in the Indonesian construction industry. Int J Environ Res Public Health 2020; 17: 7462.

41.

Heredia Morante

del Río Merino

Ros Serrano

. The accident rate in the construction sector: a work proposal for its reduction through the standardization of safe work processes. Buildings 2024; 14: 2399.

42.

Rani

Isya

Iqbal

. Risk level analysis of work safety and health on national road routine maintenance project using self-management system. In: IOP Conference series: materials science and engineering. IOP Publishing, 2019, November, Vol.674, No., pp.012013.

43.

Manzoor

Othman

Waheed

. Accidental safety factors and prevention techniques for high-rise building projects–A review. Ain Shams Eng J 2022; 13: 101723.

44.

Zermane

Tohir

MZM

Zermane

, et al. Predicting fatal fall from heights accidents using random forest classification machine learning model. Saf Sci 2023; 159: 106023.

45.

Schuldt

Nicholson

Adams

, et al. Weather-related construction delays in a changing climate: a systematic state-of-the-art review. Sustainability 2021; 13: 2861.

46.

MARSH. Three key actions to mitigate construction risks from extreme weather, https://www.marsh.com/sg/en/industries/construction/insights/mitigate-construction-risk-extreme-weather.html (2025, Accessed 25 January 2026).

47.

Arifuddin

Tumpu

Reskiana

, et al. Study of measuring the application of construction safety management systems (CSMS) in Indonesia using the analytic hierarchy process. International Journal of Engineering Trends and Technology 2023; 71: 354–361.

48.

Albert

Routh

. Designing impactful construction safety training interventions. Safety 2021; 7: 42.

49.

Bęś

Strzałkowski

. Analysis of the effectiveness of safety training methods. Sustainability 2024; 16: 2732.

50.

. Analysis of the characteristics of fatal accidents in the construction industry in China based on statistical data. Int J Environ Res Public Health 2021; 18: 2162.

51.

Belanova

Ershova

Pyatkova

, et al. Assessment of the risks of construction of transport infrastructure facilities. Transportation Research Procedia 2022; 63: 1660–1667.

52.

Shahar

Abas

Basirun

, et al. Perspectives of contributing factors and prevention measures of struck-by accidents in from experienced safety personnel. Int J Sustainable Constr Eng Technol 2023; 14: 468–475.

53.

Krishnasamy

Rahman

Mohamed

. Assessment of occupational accidents in the Malaysian construction industry from 2015 to 2023: a study on ISO 45001 implementation, impact on workers, and safety recommendations. Int J Built Environ Sustainability 2025; 12: 217–235.

54.

Kineber

Antwi-Afari

Elghaish

, et al. Benefits of implementing occupational health and safety management systems for the sustainable construction industry: a systematic literature review. Sustainability 2023; 15: 12697.