Enhancing Readability in Construction Safety Reports Using a Two-Step Quantitative Analysis Approach

Abstract

This study addresses the limitations of South Korea’s Design for Safety (DfS) reports, which are a critical component of construction safety reports (CSRs) but rely heavily on text, limiting readability and visual comprehension. While previous studies have highlighted the readability challenges in construction safety documents, few have quantitatively combined layout and readability assessments using objective metrics. To enhance information delivery, this research proposes an improved CSR format and quantitatively evaluates its effectiveness compared to the conventional format. A two-step analysis was conducted using document layout analysis, pixel-based methods, and the Flesch Reading Ease Score (FRES) to assess layout and readability. The results showed that conventional CSRs consist of nearly 100% text, while the improved format integrates approximately 70% images and 30% text, enhancing visual clarity without altering content. The improved format achieved a higher average FRES score of 50.24 compared to 44.52 for the conventional format, indicating a 1.12-fold increase in readability. These findings suggest that the improved CSR format significantly enhances comprehension and information delivery. The proposed quantitative analysis method offers a practical approach for evaluating and improving document design in construction safety, and it can be applied to other fields to improve the effectiveness of written communication.

1. Introduction

1.1. Background of Research

In the dynamic and complex sector of construction, the success of projects across their entire lifecycle is a collaborative effort involving a diverse array of stakeholders [1,2,3,4]. This endeavor is underscored by the generation of a huge amount of project-related information [5]. Information generated throughout the project is recorded and stored in documents [6]. Construction-related documents serve not only as presentations of information but also play a crucial role as a medium for effective communication among project stakeholders [7,8]. Among the diverse documentation, reports stand out as the most frequently generated, offering a contrast to contracts and drawings [9,10]. These reports are systematically produced to track ongoing developments in design, construction, and oversight, ensuring a comprehensive view of the advancement of the project [11].

Although there are many types of reports, there has been growing interest in utilizing construction safety reports (CSRs) to prevent and mitigate accidents [12]. CSRs are a key tool for maintaining and managing safety on construction sites, conveying the risks of site tasks based on accident cases [13]. These reports facilitate communication among project participants, contributing to accident prevention and management [14].

For construction safety management, each country has representative CSRs, such as Design for Safety (DfS) in South Korea, Construction Design and Management (CDM) in the United Kingdom, and the Safe Work Method Statement (SWMS) in Australia. These CSRs can be used as preventative measures during the construction phase and for learning from accident cases to prevent future incidents [15]. In South Korea, construction designers create a DfS, which includes measures to reduce hazardous risk factors at the construction site [16]. In the United Kingdom, CDM mandates the inclusion of health and safety management in planning and design during the construction process [17]. The CDM 2015 regulations specify the creation of a Construction Phase Plan as a principal duty, outlining plans to safely complete work involving risk factors [18]. In Australia, the SWMS is prepared to perform high-risk construction work, detailing the process of risk identification and risk assessment to ensure safe work practices [19]. Thus, CSRs are utilized in various ways according to the regulations and purposes of each country.

While CSRs, with their numerous advantages for accident prevention, are expected to be actively utilized in each country, not all countries are utilizing them completely. In particular, their use in South Korea is not as prevalent, which can be attributed to several issues. CSRs in South Korea adopt a text-centric structure, facing limitations in the visual transmission of information and readability [20]. Akal [21] identified the text composition of bidding documents as a factor hindering readability in an analysis study of 34 subcontract bids. Additionally, Koc and Pelin Gurgun [22] found that “lack of visual representation” in construction contract documents could lead to conflicts. These issues can prevent the effective transmission of crucial information for accident prevention at construction sites, potentially leading to construction accidents [23].

Kang et al. [24] compared three types of American Association of Orthodontists (AAO) consent forms: the original AAO consent form, a readability-enhanced AAO consent form, and an AAO consent form enhanced with both readability and visual aids. They evaluated the improvement in patient and parent memory and comprehension by asking 90 pairs of patients and parents 18 questions. The study concluded that the consent form enhanced with visual aids significantly improved patient and parent memory and comprehension compared to the form with only improved readability. This result emphasizes the crucial role of high readability and visual information in enhancing memory and comprehension. While this example is drawn from the healthcare field, the underlying principle of improved comprehension through visual and readable materials is equally applicable to construction safety. Construction workers and site managers, like patients, need to understand complex safety information under time constraints and stress. Therefore, enhancing the readability and visual materials of construction safety reports could similarly yield better outcomes compared to text-centric reports, ensuring that critical safety messages are effectively conveyed. However, to verify if these improvements are genuinely effective, quantitative evaluation is required, and appropriate evaluation metrics need to be established.

There are various evaluation metrics for assessing documents. Waller [25] proposed the following criteria for evaluating what constitutes a good document: language criteria, design criteria, relationship criteria, and content criteria. Among these, readability is used as a sub-criterion for language criteria, and layout is used as a sub-criterion for design criteria.

Readability originates from the close relationship between the text and the reader, referring to the ease with which a reader can process and understand written text [26]. In other words, it indicates how easily a text can be understood considering the interaction between various text variables and the characteristics of the reader [27]. Hence, when measuring readability, it is important to consider not only the text itself but also the ability of the reader to understand the text [28]. However, assessing readability solely based on the linguistic characteristics of the text without considering the level of understanding or the background of the reader can be challenging. Therefore, assessing readability solely based on the linguistic characteristics of the text should be evaluated using formulas or indices [29]. Hence, various readability assessment indices can be utilized, with the Flesch readability test being a commonly used index [30]. Additionally, the layout of the report plays a crucial role in evaluating its readability [31]. Layout refers to the visual arrangement of a document, organizing and highlighting information to capture the attention of the reader [32]. A well-structured layout arranges information logically and consistently, reducing the cognitive load on readers and making it easier to locate important information [33]. This reduces the effort required for readers to read and understand the document, ultimately enhancing readability [34]. Therefore, readability and layout are closely related, and considering both factors together is essential for improving the overall quality and effectiveness of a document [31].

This study proposes an improved construction safety report by adding visual aids (images) to the existing text-centric reports. To verify the effectiveness of this improved report, the study will use two evaluation criteria: readability and layout.

1.2. Literature Review

There are various evaluation criteria for assessing documents, but this study focuses on readability and layout. Accordingly, we first analyzed previous studies that evaluated both the readability and layout of documents.

In the construction field, Sinyai et al. [35] identified ongoing issues with the readability and suitability of construction occupational safety and health materials for workers. This study aims to review examples of construction occupational safety and health materials to investigate how well these materials meet established readability and suitability standards. Using the standard online readability calculator (readability-score.com; accessed on 24 September 2024), the text’s difficulty was assessed, and the SMOG and Flesch–Kincaid formulas were used to generate grade level ratings required for understanding specific texts. Furthermore, the suitability was evaluated using the Suitability Assessment of Materials (SAM) and the Clear Communication Index (CCI), which comprehensively assess various factors including reading grade level and typography, in addition to layout factors. Through this, writers and publishers of construction occupational safety and health materials can identify opportunities to improve their materials by utilizing the readability and suitability testing tools presented in this study.

In fields other than construction, Liu et al. [36] evaluated the readability and layout of patient information leaflets (PILs) for cardiovascular disease and type 2 diabetes medications provided in the UK to determine if these leaflets are appropriate for use by older adults. The layout criteria for PILs were based on the guidelines from the European Union (EU) and the Medicines and Healthcare products Regulatory Agency (MHRA). The readability of the PILs was assessed using the Gunning Fog Index formula. This combined evaluation highlighted the need to improve both the layout and readability of the current leaflets.

Synthesizing the findings from previous studies, it is emphasized that evaluating both readability and layout through various evaluation tools highlights the need for document improvement. However, studies that evaluate both readability and layout together are limited. Moreover, the study by Liu et al. [36] did not use quantitative evaluation tools for layout but relied on qualitative assessments using EU and MHRA guidelines, which affected the study’s validity. For this reason, there is a need to develop new methodologies that use quantitative evaluation tools to assess both readability and layout together.

The quantitative evaluation tools used in previous studies to assess readability and layout were analyzed. First, studies that evaluated the readability of documents using quantitative evaluation tools were investigated.

Hall [37] evaluated the readability of original research articles published in surgical journals using the Flesch Reading Ease Index. The aim was to suggest a target Flesch Reading Ease Score for authors when writing manuscripts. The study calculated the Flesch Reading Ease Scores for 30 original articles published in Archives of Surgery, the British Journal of Surgery, and the ANZ Journal of Surgery in 2005. The results showed that the average scores for each journal were 12.4 (Archives of Surgery), 14.4 (British Journal of Surgery), and 18.6 (ANZ Journal of Surgery), concluding that original research articles published in surgical journals are difficult to read.

Emanuel and Boyle [38] collected and analyzed informed consent documents from four phase III clinical trials for COVID-19 vaccines to evaluate how difficult these documents were to read. The Flesch–Kincaid Grade Level (FKGL) was used to determine the grade level required to understand specific texts, and the Flesch Reading Ease Score (FRES) was applied to assess texts’ difficulty. The study found that all informed consent documents required a reading level of 9th grade or higher, with an average Flesch Reading Ease Score of 52.4, indicating a difficult reading level. The authors concluded that the informed consent documents for COVID-19 vaccine trials are difficult for the general public to understand.

Next, studies that evaluated the layout of documents using quantitative evaluation tools were investigated.

Reinert et al. [39] analyzed nine patient consent forms from a phase III clinical neuro-oncology study conducted at a German brain tumor center to evaluate their quality. The quality evaluation criteria included layout, readability, and ethical aspects, among others. The layout was evaluated based on three criteria: Well-Structured Information, Helpful Graphics, Symbols, or Figures, and Study Flow Chart. Each criterion was scored up to three points using specific evaluation criteria and scoring systems to quantitatively assess the layout of the documents.

Sinyai et al. [35] investigated how well 103 construction occupational safety and health materials met established readability and suitability standards. Suitability was evaluated using a checklist that included 22 items, some of which assessed layout factors. Each item was scored as 2 (excellent), 1 (adequate), or 0 (inadequate), using this scoring system to quantitatively determine how effectively the layout of the documents was structured.

The analysis of studies using quantitative evaluation tools to assess readability and layout revealed the following findings.

First, previous studies have applied various readability evaluation tools and layout scoring systems to different types of documents. However, unlike readability evaluation tools (e.g., Flesch Reading Ease, Flesch–Kincaid Grade Level), layout evaluation tools lack specific quantitative evaluation methods. Readability evaluation tools can assign scores based on predefined formulas, whereas layout evaluations can vary based on the evaluator’s subjective interpretation. For example, criteria, such as “Well-Structured Information” or “Helpful Graphics” in [39], can differ based on the evaluator’s understanding and interpretation. This variability can lead to inconsistent evaluation results for the same document, reducing the consistency of the evaluation. Therefore, it is necessary to select objective and consistent quantitative evaluation tools to jointly assess readability and layout and develop new methodologies for effectively evaluating documents.

Second, most existing studies analyzed readability and layout with a sample size of fewer than 30 documents. However, a small sample size makes it difficult to generalize the findings to the entire population, and the results based on such data cannot be considered reliable.

Therefore, the following research questions can be proposed. (i) How can methodologies for evaluating readability and layout be suggested? (ii) How can the results of readability be ensured with a small sample size?

Based on the existing literature, this study aims to confirm the improvement of layout and readability in the enhanced construction safety report using a two-step quantitative analysis method: document layout analysis and a pixel-based method and the Flesch Reading Ease Score. Additionally, as mentioned in previous studies, the reliability of validation results may not be sufficient due to a small sample size. To address this, we propose a framework that utilizes Monte Carlo simulation to generate 10,000 random samples, thereby producing more reliable results.

This paper is structured as follows. Section 1 introduces the background and rationale for the study. Section 2 details the materials and methods used, including data collection, document layout analysis, and readability assessment. Section 3 presents the results of the analysis. Section 4 discusses the findings in the context of existing research and practical implications. Finally, Section 5 concludes the paper with a summary of contributions and suggestions for future research.

2. Materials and Methods

This study proposes an improved construction safety report by adding visual materials (images) to the conventional text-centered reports and conducts a quantitative evaluation using document layout analysis and the Flesch Reading Ease formula to verify its effectiveness. The aim is to confirm the improvement in the layout and readability of the improved construction safety reports. The study is conducted in three phases, as follows.

First, this study collects 30 Design for Safety (DfS) reports. Then, it proposes 30 improved reports with added visual materials (images) to improve the conventional text-centered DfS. The effectiveness of these enhanced reports is then verified through the following steps. Second, the ratio of layout components (text and images) in the reports is calculated through document layout analysis and pixel analysis. This quantitatively verifies the improvement in the layout. Third, the Flesch Reading Ease formula, an indicator for evaluating the ease of reading text, is used to verify the improvement in readability. However, considering that the sample size of 30 reports may be insufficient to detect statistically significant differences, additional analyses are performed. In cases where the sample size is inadequate, Monte Carlo simulations are utilized to generate 10,000 random readability scores to derive more reliable results. This approach aims to estimate the characteristics of the population by generating a larger scale of random data based on the readability scores obtained from the limited dataset. This approach reduces statistical uncertainty due to limited data and contributes to a more reliable evaluation of the readability improvement of the improved CSR. The overall research framework is shown in Figure 1.

2.1. Collection of Data

The CSR utilized in South Korea’s construction industry primarily encompasses DfS and the Safety and Health Design Record [40]. Although operated under different legislation, it is assessed that there is no significant difference in the contents of these two reports [41]. Therefore, for improving CSR in South Korea, if one format can be selected for proposing improvements, it is expected that the same approach can be applied to the other format, as well. Hence, in this study, the authors chose to focus on DfS reports, which are extensively researched globally. DfS was introduced in 2016 to enhance construction safety in South Korea [42]. It involves identifying potential hazards in the early design stages of construction projects and mitigating or eliminating them throughout the project phases to ensure safety during construction [43]. For effective analysis and results, 30 DfS reports issued by public agencies were collected as the conventional CSR. Afterwards, we proposed 30 improved CSRs with added visual data (pictures) to improve the conventional text-based CSR. The formats of the conventional CSR and improved CSR are shown in Figure 2 and Figure 3, respectively. The comparison of the components between the conventional and improved CSR formats is presented in

Table 1. Comparison of DfS components.

No.	Component	Conventional DfS	Improved DfS
1	Title	-	Text
2	Hazard management in charge	Text	Text
3	Facility type	-	Text
4	Work type	Text	Text
5	Elements	-	Text
6	Solution stage	Text	Text
7	Reduce measures stage	Text	Text
8	Accident type (physical/human)	Text	Text
9	Hazard (accident causes)	Text	Text + Image
10	Risks	Text	Text
11	Hazard reduction measures (alternative)	Text	Text + Image
12	Reduction measures by third party	Text	-
13	Remaining hazard	Text	-
14	Evaluation aspect and purpose	Text	-
15	Evaluation table for alternative	Text	-
16	Note	Text	-
17	Final ver.	-	Text
18	Code	Text	Text

. In this study, the comparison of the components between the two CSR formats yielded several differences between the conventional CSR and the improved CSR. In the conventional CSR, certain components were omitted, whereas the improved CSR has supplemented these by providing richer information through text and images. First, items, such as titles, facility types, final versions, and elements, not listed in the conventional CSR were added in text form in the improved CSR. Additionally, in the improved CSR, sections like hazard (accident causes) and hazard reduction measures (alternatives) are visually presented using images. However, some items present in the conventional CSR were omitted in the improved CSR, including reduced measures by third parties, remaining hazards, evaluation aspect and purpose, the evaluation table for alternatives, and notes. This comparison reveals that the transition from conventional CSR to improved CSR did not alter the existing common information; rather, parts of the text were simply replaced with images.

2.2. DLA and Pixel-Based Report Evaluation

As a result of analyzing the above components, it was found that the text sections were simply replaced with images. To quantitatively confirm this, a DLA analysis was conducted. In this study, the authors aim to detect and visualize layout components using DLA, followed by the quantification of the pixel area size of each component based on the DLA results. Specifically, the authors analyzed how the ratio of layout components changed in the improved CSR compared to the conventional CSR, aiming to verify the improvements in the layout structure of the document. Hence, 30 samples of conventional and improved CSRs were analyzed and compared.

2.2.1. Extraction of Document Layout

DLA identifies and categorizes areas of interest within document images [44]. It involves finding the relationships and positions of document layout components, such as text and visual information (tables, figures, and charts), and other elements [45]. There are two fundamental approaches to DLA: bottom–up and top–down. The bottom–up approach analyzes the document from the level of small elements, like pixels, building up to the overall layout [46]. This method is useful for understanding the detailed characteristics of the document and accurately comprehending its structure [47]. Conversely, the top–down approach starts with large document areas, dividing the document into smaller areas, such as text, images, lines, etc. [48]. This method processes large document areas first, allowing for the sequential identification of detailed elements, which aids in understanding the overall structure [47]. This study employs a top–down approach to comprehend the overall structure of the document. The DLA analysis process using the top–down approach is as follows. First, to analyze the CSR layout stored in PDF format, the PDF page was converted to an image, and then image processing was performed using OpenCV. OpenCV is a powerful library for image and video processing, providing various functions, such as image loading, conversion, analysis, and object recognition [49]. Using the OpenCV library, PDF document pages undergo conversion from color to grayscale, thereby facilitating the computational identification of shapes and patterns within the document [50].

After image processing, the Canny algorithm was used to detect edges for line detection [51]. Edges represent parts of the image that delineate the contours of objects [52]. After obtaining information about the contours of objects in the image through edge detection, Hough Transform was used to detect lines [53]. Hough Transform is a technique used to find straight lines in an image, analyzing edge information found in the edge detection stage [54]. This process allows for the determination of the coordinates of the detected lines [55].

Subsequently, the process proceeds to the identification of text, images, numbers, other components, and graphical elements. The detected lines through image processing and edge detection are visualized by using bounding boxes [56]. These bounding boxes are created by drawing a square on the detected line coordinate to visually represent the component [57]. Bounding boxes created around text, images, numbers, and other elements are marked in specific colors. The characteristic colors for each component are defined as follows:

• Text: Text is marked in red and identified by the presence of alphabetic characters.
• Numbers: Numbers are marked in yellow and identified when they exclusively contain numerical characters within text elements.
• Images: Images are marked in blue and identified as pictures or photographs within the PDF document.
• Other elements: All other components are marked in orange and distinguished from text, numbers, images, and graphic lines.
• Graphic lines: Lines are marked in green and automatically detected through image processing and line detection processes. As graphic lines are not included in the analysis, they are not applied to subsequent pixel size calculation tasks.

2.2.2. Calculation of Pixel Size

A pixel represents the smallest unit of a digital image [58]. Based on the previously determined location and size information of each element, the size of each component is calculated in terms of pixel area. In this process, document components are classified into two main categories for calculation: text (text, numbers, and other elements) and images.

First, the pixel area of each component and the total area are calculated, as shown in Equation (1):

$${Pixel area}_i=\sum {Pixel area}_j=\prod {(Pixel width}_j, {Pixel height}_j)$$

(1)

where Pixel area_i means an area with a specific pixel range along with the ith color, Pixel area_j means an area with the jth pixel range, Pixel width_j means the width of the jth pixel, Pixel height_j means an area with the jth pixel range, and jth means the rectangle for the image and text in the DfS.

Subsequently, the relative size occupied by each component within the total area was calculated, as shown in Equation (2):

$$Pixel percentage={\displaystyle \frac{{Pixel area}_i}{\sum {Pixel area}_i}}\times 100(\%)$$

(2)

With these calculations, the authors quantitatively compared and analyzed how much text and images occupy the document layout, thereby verifying the improvement in the structure and layout of the document.

2.3. Verification of Improvement in Readability Based on the FRES

Based on the DLA analysis results from Section 3.2, the improvement in layout was verified through structural analysis. According to previous research, readability and layout are closely related, and a clearly organized layout can enhance readability [34]. Therefore, readability should be considered alongside layout. Consequently, the authors conducted a readability assessment for each report, along with a layout evaluation using DLA.

For readability assessment, various formulas can be employed [59]. Prominent among these are the FRES and FKGL, which are extensively used [60]. The FRES is one of the oldest readability formulas, and it is widely used by government agencies and organizations, recognized across various countries and academic research fields, and adopted by many researchers as a popular readability assessment tool [61,62]. Therefore, in this study, the authors employed the FRES to quantitatively verify the improvement in readability. The FRES is calculated based on the number of words, sentences, and syllables in the document [63], as shown in Equation (3):

$$\begin{array}{cc}\hfill Flesch Reading& Ease Score\hfill \\ & =206.835-1.015\times (Total Words ÷ Total Sentences)\\ & - 84.6\times (Total Syllables ÷ Total Words)\end{array}$$

(3)

The FRES is typically expressed within a range of 0–100, with higher scores indicating easier readability of the text [64]. Based on the evaluation results of each test, it is explained that scores above 50 require the comprehension level of a high school student [65]. The authors compared the readability of conventional and improved CSRs to verify how much the readability of the improved CSR had improved compared to the conventional CSR. The process for calculating the Flesch Reading Ease Score is as follows. First, it is necessary to extract text from PDF documents. For text extraction, OCR (Optical Character Recognition) technology was used. Specifically, in the Python programming environment (version 3.12.5), the pytesseract library was utilized to employ the Tesseract OCR engine. The Tesseract OCR engine, developed by Google, is a powerful open-source OCR tool that supports character recognition in various languages [66]. After the initial text extraction, it was found that the extracted text data contained errors and omissions. Therefore, a manual text correction process was applied to improve accuracy. This process involved checking the extracted text line by line against the original CSR documents and ensuring missing words were added and sentence structures were restored for clarity. To ensure consistency and objectivity during manual correction, the following guidelines were followed: (i) all corrections were performed by two independent reviewers with experience in construction safety documentation; (ii) a checklist of common errors (e.g., missing section titles, incomplete hazard descriptions) was used; and (iii) ambiguous corrections were cross-validated by comparing multiple report samples. The corrected text data were then input into an online text analysis tool to measure the number of words, sentences, and syllables per page. Based on these measured values, the Flesch Reading Ease Score was directly calculated by substituting the values into the calculation formula.

The readability scores of the 30 samples of conventional and improved CSRs represent a fraction of the population and are intended to analyze the statistical characteristics of the population. However, initial evaluations with a limited dataset present challenges in fully reflecting the characteristics of the entire population [67]. When the sample size is small, the results of statistical tests entail uncertainty, and Monte Carlo simulation has traditionally been employed to address this issue of uncertainty [68]. Accordingly, our analysis began by conducting an Anderson–Darling goodness-of-fit test (A-D test) on the readability scores of the 30 samples of conventional and improved CSRs to assess data distribution. Subsequently, we employed Monte Carlo simulation to generate random data, aiming to achieve a more detailed understanding of the population’s characteristics. The A-D test is a statistical method that fits the given data to various probability distributions to determine which distribution best explains the data [69]. The A-D value is a measure of goodness of fit obtained from the A-D test, indicating the difference between the given data and a specific distribution [70]. The smaller the A-D value calculated from the A-D test, the more closely the distribution matches the given data [71]. Following the A-D test applied to various probability distributions for the given data, the distribution of the readability scores of the 30 samples of conventional CSR was found to fit the beta distribution (A-D value: 0.1591). The beta distribution of the conventional dataset is shown in Equation (4):

$$f(x;\alpha ,\beta )={\displaystyle \frac{x^{\alpha -1}\cdot {(1-x)}^{\beta -1}}{B(\alpha ,\beta )}}$$

(4)

The readability scores of the 30 samples of the improved CSR yield the smallest A-D value of 0.1966 for the logistic distribution, suggesting that this distribution closely matches the data from the 30 improved CSR samples. The logistic distribution for the improved dataset is defined as shown in Equation (5):

$$f(x;\mu ,s)={\displaystyle \frac{e^{-(x-\mu )/s}}{{s^{\left({1+e}^{-(x-\mu )/s}\right)}}^2}}$$

(5)

After determining the data distribution using the A-D test, Monte Carlo simulation was used to generate random data from the defined probability distributions. Monte Carlo simulation refers to a probabilistic process specifically aimed at facilitating decision making under uncertain circumstances [72]. The objective of Monte Carlo simulation is to draw samples from the probability distributions assumed in the model to generate random variables [73]. The essence of this process is random sampling [74]. Because random samples effectively represent the population, their statistical values provide good estimates for the parameters [75]. Therefore, following the A-D test, random numbers were generated from each identified probability distribution, and these random numbers were used to extract data equivalent to the desired sample size [76].

In this study, Monte Carlo simulation was employed to generate 10,000 random readability scores. This approach aims to estimate the characteristics of the population by generating a larger scale of random data based on the readability scores obtained from the limited dataset.

3. Results

3.1. Overview of the Evaluation Process

This study employed a systematic evaluation process to assess the effectiveness of the improved construction safety reports (CSRs) in terms of layout and readability. The evaluation process was conducted in the following three key steps: (1) document layout analysis (DLA) and pixel-based methods were used to quantitatively assess the layout differences between the conventional and improved CSR formats; (2) the Flesch Reading Ease Score (FRES) was calculated for each report to evaluate the text’s readability using Optical Character Recognition (OCR) and subsequent text analysis; and (3) Monte Carlo simulation was conducted to address the limitations of small sample size, generating 10,000 random samples from the identified probability distributions to provide robust statistical evaluation. This systematic approach ensured that layout and readability improvements were assessed comprehensively and reliably.

3.2. Comparison Results of DLA and Pixel-Based Method

In the section titled “Extraction of document layout”, four components, such as text, image, numeric, and other elements, were detected with respect to their position for visualization using DLA (Figure 4). Figure 4 represents the results of DLA analysis using the 6th report out of 30 CSRs as an example. The sixth report was selected because it exhibited the most notable contrast in the text-to-image ratio when compared to both the conventional and improved CSRs.

Based on the results of the DLA analysis, pixel area sizes were calculated to compare and evaluate how the ratio of layout components changed between the improved CSR and the conventional CSR. First, the pixel ratio results of the four components of the improved report are shown in

Table 2. Four component pixel ratio of improved DfS.

No.	Text (Red)	Image (Blue)	Numeric (Yellow)	Other Elements (Orange)
1	33.40	66.20	0.20	0.20
2	32.20	67.40	0.20	0.20
3	34.40	65.20	0.20	0.20
4	31.00	68.60	0.20	0.20
5	34.20	65.40	0.20	0.20
6	33.40	66.20	0.20	0.20
7	31.90	67.70	0.20	0.20
8	29.97	69.67	0.21	0.15
9	33.34	66.25	0.23	0.17
10	30.90	68.70	0.20	0.20
11	33.30	66.30	0.20	0.20
12	31.10	68.50	0.20	0.20
13	35.73	63.83	0.25	0.18
14	29.00	70.60	0.20	0.20
15	34.25	65.34	0.24	0.18
16	30.46	69.17	0.21	0.16
17	31.70	67.90	0.20	0.20
18	32.50	67.10	0.20	0.20
19	31.20	68.40	0.20	0.20
20	30.80	68.80	0.20	0.20
21	30.40	69.20	0.20	0.20
22	34.50	65.10	0.20	0.20
23	36.00	63.60	0.20	0.20
24	34.30	65.30	0.20	0.20
25	35.20	64.40	0.20	0.20
26	31.20	68.40	0.20	0.20
27	34.50	65.10	0.20	0.20
28	34.74	64.84	0.24	0.18
29	33.30	66.30	0.20	0.20
30	32.90	66.70	0.20	0.20

. For the conventional report, the text component is 98.8%, image is 0%, numeric is 0.7%, and other elements are 0.5%. All 30 DfS report samples showed the same results, so these results were omitted in Table 2. As seen in Table 2, the pixel ratios of numeric and other elements in the improved report are significantly lower compared to the ratios of text and image. Therefore, when calculating pixels, the components within the document were classified into two main categories, text (text + numeric + other elements) and image, following the method described in Section 2.2.2 titled “Calculation of Pixel Size.” The results of calculating the improved report, classified into these two main categories, are shown in

Table 3. Two component pixel ratio of improved DfS.

No.	Text (T + N + O)	Image
1	33.80	66.20
2	32.60	67.40
3	34.80	65.20
4	31.40	68.60
5	34.60	65.40
6	33.80	66.20
7	32.30	67.70
8	30.33	69.67
9	33.74	66.25
10	31.30	68.70
11	33.70	66.30
12	31.50	68.50
13	36.16	63.83
14	29.04	70.60
15	34.67	65.34
16	30.83	69.17
17	32.10	67.90
18	32.90	67.10
19	31.60	68.40
20	31.20	68.80
21	30.60	69.20
22	34.90	65.10
23	36.40	63.60
24	34.70	65.30
25	35.60	64.40
26	31.60	68.40
27	34.90	65.10
28	35.16	64.84
29	33.70	66.30
30	33.30	66.70

. For the conventional report, text (T + N + O) is 100% and image is 0%. All 30 DfS report samples showed the same results, so the results were omitted from the table. The statistical analysis results of the text and image ratios between the improved and conventional reports are presented in

Table 4. Statistical analysis comparison of text and image ratios.

	Conventional CSR		Improved CSR
	Text	Image	Text	Image
Mean	100	0	33.13	66.87
Standard Deviation	0	0	1.86	1.86
Minimum	100	0	29.40	63.60
Maximum	100	0	36.40	70.60

. The analysis of the layout component ratios of 30 conventional CSRs revealed that the reports consisted entirely of text with no images. However, the analysis of the layout component ratios of 30 improved CSRs showed that text accounted for 33.13%, whereas images comprised about 66.87%. This confirms a significant decrease in text and a considerable increase in images. Thus, by quantifying the layout component ratio of improved CSR using DLA using pixels, the alteration in the document layout structure was verified.

3.3. Comparison Results of the FRES

To analyze the readability levels of 30 reports using the Flesch Reading Ease formula, text was first extracted from PDF documents using the Tesseract OCR engine. However, most of the extracted text was disorganized and jumbled, failing to maintain the original structure of the documents. Additionally, some text was not properly extracted, necessitating manual supplementation of the missing parts. To ensure the accuracy of the extracted text, sentences were reassembled to restore the original meaning and structure of the documents, and any missing text was verified. For example, in Improved CSR 6 (Figure 3), the pre- and post-correction text extraction states can be compared in

Table 5. Before text extraction modification.

No.	Text Extraction
1	Code No.
2	The braced wall wale bracket fell off and the braced wall collapsed
3	Hazard target (Constructor, User, Owner)
4	Application (-, -)
5	Elements (-)
6	Work (Earthwork)
7	Facility type (Apartment)
8	-
9	- (Non, -, Fall, Fire, -, Burst, ETC)
10	-
11	Type (Severity, Frequency)
12	Level H(8~25) M(4~7) L(1~3) ETC
13	Reduce measures stage (PPE, Administrative Controls, Engineering Controls, Substitution, Elimination)
14	- Ver.
15	division
16	Hazard (Accident causes)
17	Alternative
18	Picture/Drawing
19	Content
20	The braced wall wale bracket fell off and the braced wall collapsed
21	Reflect in the detailed drawing of the support angle for wale bracket
22	Prevent the wale from falling off by securing the bracket

and

Table 6. After text extraction modification.

No.	Text Extraction
1	Code No.
2	The braced wall wale bracket fell off and the braced wall collapsed
3	Hazard target (Constructor, User, Owner)
4	Application (Pre-construction phase, Construction phase)
5	Elements (Earthwork braced wall)
6	Work (Earthwork)
7	Facility type (Apartment)
8	Human (Fall, Fall down, Bump/Hit, Caught/Crushed, Amputation/Cut, Electrocuted, Burned, Chocked, ETC)
9	Physical (Non, Collapse, Fall, Fire, Explosion, Burst, ETC)
10	AS-IS, TO-BE
11	Type (Severity, Frequency)
12	Level H(8~25) M(4~7) L(1~3) ETC
13	Reduce measures stage (PPE, Administrative Controls, Engineering Controls, Substitution, Elimination)
14	Final Ver.
15	division
16	Hazard (Accident causes)
17	Alternative
18	Picture/Drawing
19	Content
20	The braced wall wale bracket fell off and the braced wall collapsed
21	Reflect in the detailed drawing of the support angle for wale bracket
22	Prevent the wale from falling off by securing the bracket

. In the pre-correction text data, items 4, 5, 8, 9, 10, and 14 did not extract properly. Missing words were manually added to complete the text data. Afterward, the completed text data was input into an online text analysis tool to measure the number of words, sentences, and syllables on each page. Based on these measured values, scores were derived by directly applying the values to the pre-designed Flesch Reading Ease Score formula.

The number of words, sentences, syllables, and Flesch Reading Ease Scores for each page of the 30 conventional and improved reports are shown in

Table 7. Flesch Reading Ease Score of conventional DfS.

No.	Words	Sentences	Syllables	Score	Summary
1	81	7	166	21.71	Very difficult to read
2	142	12	279	28.60	Very difficult to read
3	112	13	202	45.51	Difficult to read
4	201	18	339	52.82	Fairly difficult to read
5	145	11	227	61.01	Plain English
6	149	13	227	66.31	Plain English
7	204	17	326	59.46	Fairly difficult to read
8	171	15	324	34.97	Difficult to read
9	174	13	336	29.88	Very difficult to read
10	166	14	277	53.63	Fairly difficult to read
11	159	16	293	40.85	Difficult to read
12	140	14	247	47.43	Difficult to read
13	130	13	226	49.61	Difficult to read
14	149	14	272	41.59	Difficult to read
15	170	16	312	40.78	Difficult to read
16	208	16	369	43.56	Difficult to read
17	159	16	293	40.85	Difficult to read
18	156	16	275	47.80	Difficult to read
19	133	13	212	61.60	Plain English
20	171	15	295	49.32	Difficult to read
21	115	15	201	51.19	Fairly difficult to read
22	134	13	247	40.43	Difficult to read
23	105	13	214	26.21	Very difficult to read
24	163	11	307	32.46	Difficult to read
25	144	12	238	54.83	Fairly difficult to read
26	144	12	253	46.02	Difficult to read
27	120	12	241	26.78	Very difficult to read
28	168	14	310	38.55	Difficult to read
29	126	15	243	35.15	Difficult to read
30	124	12	195	63.31	Plain English

and

Table 8. Flesch Reading Ease Score of improved DfS.

No.	Words	Sentences	Syllables	Score	Summary
1	43	3	91	13.25	Very difficult to read
2	23	3	43	40.89	Difficult to read
3	36	4	61	54.35	Fairly difficult to read
4	44	5	77	49.85	Difficult to read
5	51	5	77	68.75	Plain English
6	49	5	66	82.94	Easy to read
7	55	5	81	71.08	Fairly easy to read
8	39	5	76	34.06	Difficult to read
9	55	4	99	40.60	Difficult to read
10	30	4	49	61.04	Plain English
11	34	5	73	18.29	Very difficult to read
12	47	5	76	60.49	Plain English
13	38	5	67	49.96	Difficult to read
14	53	4	95	41.74	Difficult to read
15	27	4	50	43.32	Difficult to read
16	61	5	107	46.06	Difficult to read
17	48	5	84	49.04	Difficult to read
18	49	5	81	57.04	Fairly difficult to read
19	48	5	75	64.90	Plain English
20	52	5	92	46.60	Difficult to read
21	27	4	43	65.25	Plain English
22	27	4	51	40.18	Difficult to read
23	36	5	68	39.73	Difficult to read
24	34	4	69	26.52	Very difficult to read
25	51	4	82	57.87	Fairly difficult to read
26	45	4	73	58.18	Fairly difficult to read
27	25	4	51	27.91	Very difficult to read
28	58	5	100	49.20	Difficult to read
29	35	4	62	48.09	Difficult to read
30	35	4	55	65.01	Plain English

. Among these metrics, the Flesch Reading Ease Scores are presented as a line graph in Figure 5 to visually compare the scores between the conventional and improved reports.

Overall, the authors observed that the readability scores of improved CSR are higher than those of conventional CSR, with the highest readability score of improved CSR being 82.94. The difference in scores between conventional and improved CSR is up to approximately 16.6, indicating a significant improvement in readability levels.

However, initial evaluations with a limited dataset present a challenge in fully reflecting the characteristics of the entire population. To address this, the characteristics of the population were more precisely analyzed using the A-D test and Monte Carlo simulation. After inferring the distribution through the A-D test, random readability score data for each report were generated through Monte Carlo simulation, producing 10,000 random readability scores for each, as shown in Figure 6. According to the analysis results in

Table 9. Statistical analysis of readability scores using Monte Carlo simulation.

	Conventional CSR	Improved CSR
Mean	44.52	50.24
Standard deviation	11.64	14.31
Minimum	16.99	1.28
Maximum	71.66	129.58
Q1	35.72	40.58
Q2	44.62	50.09
Q3	53.24	59.85

, the average readability scores of conventional CSR and improved CSR were 44.52 and 50.24, respectively, indicating that the readability of improved CSR was better than that of conventional CSR.

To determine if there is a statistically significant difference in the average readability scores between the two reports, a one-sided Student’s t-test was conducted. The following hypotheses were set before performing the test:

Null Hypothesis (H₀).

The average readability scores of conventional and improved CSR will be the same.

Alternative Hypothesis (H₁).

The average readability score of conventional CSR will be lower than that of improved CSR.

Before performing the one-sided Student’s t-test, an F-test was conducted to verify the difference in variances between the readability scores of the two groups, conventional and improved CSR. The results of the F-test are shown in

Table 10. F-test: two groups for variance.

	Conventional	Improved
Mean	44.52	50.24
Variance	135.46	204.69
Number	10,000	10,000
F ratio	0.66
P(T ≤ t) one-tail	0.00
F Critical one-tail	0.97

. The results of the F-test show a one-sided p-value of 0.00, which is less than the significance level of 0.05 (P(F ≤ f) < 0.05). Referring to the results of the F-test, their variances are suggested as unequal. Therefore, t-tests are based on the assumption that the variances between two sample data ranges are unequal. The one-sided unequal variances t-test results were revealed as follows (

Table 11. t-test: unequal variances.

	Conventional	Improved
Mean	44.52	50.24
Variance	135.46	204.69
Number	10,000	10,000
t Stat	−31.01
P(T ≤ t) one-tail	0.00
t Critical one-tail	1.64

). Because the p-value is lower than 0.05, the null hypothesis is to be rejected. In other words, hypothesis testing confirmed that the average score of improved CSR is higher than that of conventional CSR.

4. Discussion

According to the results confirmed through DLA and pixel-based layout evaluation, the conventional CSR format consisted of approximately 100% text, whereas the improved CSR format was restructured to include about 70% images and 30% text. In discussing these results, an analysis of prior research found that an increased ratio of images to text enhances document comprehension [77]. Furthermore, visually enhanced materials improve readability and aid in understanding the text [78]. Thus, an increase in the image ratio within the report directly contributes to the improvement of the overall readability of the document. Consequently, the conclusion can be drawn that the improved CSR, characterized by a higher image ratio, facilitates reader comprehension of the document, thereby implying that the improved CSR has contributed to an improvement in readability [79].

According to the Anderson–Darling distribution fitting and the generation of 10,000 readability scores based on the FRES through Monte Carlo simulation, the average readability score of the conventional CSR was 44.52, while the improved CSR’s score was 50.24, indicating an approximate 1.12-fold improvement in readability. An analysis of prior research considering the impact of readability on document comprehension confirmed that a higher level of readability positively influences document comprehension [80]. This indicates the significant role that readability plays in understanding documents. However, it is important to note that both the conventional and improved CSRs are still rated as fairly difficult to read according to the FRES thresholds. While the improved CSR shows statistically significant enhancements in readability, this does not guarantee that end users, such as construction workers or safety officers, will find the reports intuitively easier to understand. Therefore, these results should be interpreted as a preliminary indication of improved text design rather than definitive proof of better user comprehension. Additionally, the results of the one-sided F and t-tests confirmed that there is a statistically significant difference in readability scores between the two reports. This statistical difference indicates that the observed disparity implies a meaningful difference in readability between the two groups. In essence, this supports the conclusion that the readability of the improved CSR is statistically superior to that of the conventional CSR.

In line with recent studies on construction safety communication and documentation practices, our findings align with prior work emphasizing the crucial role of visual clarity and language accessibility in improving understanding and compliance. For instance, while previous research [35] has highlighted issues of readability and layout in safety documents, our study extends these findings by employing a two-step quantitative evaluation method, combining document layout analysis with pixel-based methods and the FRES, and validating results with Monte Carlo simulation. Unlike purely qualitative assessments or limited-scope studies, our approach offers a more objective and replicable framework for evaluating construction safety reports. Moreover, the study fills a gap in the literature by providing comprehensive, quantitative evidence that improved layout (integrating images and reduced text density) contributes to enhanced readability metrics. This contrasts with earlier works that focused primarily on readability scores without considering layout or that used subjective assessments of layout without quantitative validation. Therefore, our study not only reinforces but also expands upon the current understanding of document design for construction safety communication, offering practical and methodological advancements.

To enhance conventional text-based CSRs, an improved report incorporating visual materials (images) was proposed. To verify whether this enhanced report is superior to the conventional one, a quantitative evaluation was conducted using document layout analysis, pixel analysis, and the Flesch Reading Ease Score. This evaluation confirmed whether the layout and readability of the improved CSRs were better. Ultimately, it was quantitatively demonstrated that the improved CSRs are superior to the conventional CSRs. Additionally, to address the potential lack of reliability in the verification results due to the small sample size, a framework utilizing Monte Carlo simulation was presented, which contributed to enhancing the reliability of the findings in this study.

Beyond these findings, the improved CSR format holds several practical implications for the construction industry. By integrating clearer visual representations and improved text layout, these enhanced reports can serve as effective tools in safety communication and training, potentially reducing misunderstandings of safety guidelines. In practice, the improved format could be incorporated into standard safety documentation protocols, design guidelines, or digital safety management systems, ensuring consistent and comprehensible reporting across projects. Furthermore, the enhanced CSR format could be adopted in worker training programs to reinforce safe practices through more accessible content. These practical applications could lead to increased adoption of CSRs in South Korea, where reliance on text-heavy reports has limited their effectiveness.

5. Conclusions

This study successfully enhances text-based construction safety reports by incorporating visual materials (images) and evaluating their effectiveness through readability and layout assessments. The readability was assessed using the Flesch Reading Ease, while the layout was evaluated through document layout analysis (DLA) and pixel-based methods. The research validates the effectiveness of the improved reports using a two-step quantitative approach, demonstrating that the inclusion of visual materials significantly contributes to better readability and a more effective layout.

The contributions of this study are as follows. (1) A method for quantitatively verifying improvements in the report layout and readability using a two-step quantitative analysis method that employs DLA and pixel methods along with the FRES was presented. (2) The improved report format proposed for this study acts as a more effective means of safety reporting and information delivery, potentially enhancing safety on construction sites. (3) The quantitative analysis method proposed in this study can be utilized in fields beyond construction to improve document layout and readability, thereby increasing the effectiveness of information delivery. (4) The quantitative analysis method proposed for critically evaluating and improving the conventional CSR format contributes to the enhancement of technical document quality, which is expected to aid in raising standards in professional fields.

The limitations of this study are as follows. (1) The dataset used in this study represents only a fraction of CSRs, and the study did not cover a broader variety of report types and formats, presenting challenges in generalizing the two-step quantitative analysis method. (2) This study focused primarily on the visual aspects of reports, such as layout and readability, to assess document quality. However, evaluating document quality based solely on layout and readability has its limitations. (3) It is also emphasized that readability formulas overlook the prior knowledge, interests, and motivation of the reader. (4) The study lacks direct validation from actual users, as no feedback was collected from report readers, nor were comprehension tests conducted to verify whether the improved readability scores translate to better understanding in practice. The improvements demonstrated are limited to quantitative metrics rather than validated user experience outcomes.

Considering the above limitations, future research will proceed as follows. (1) The research will expand to cover various types and formats of reports beyond the examples limited to DfS reports, aiming to enhance the generalizability of the two-step quantitative analysis method. (2) By incorporating additional content quality evaluation metrics beyond layout and readability, the study will aim to achieve more comprehensive and integrative results through evaluations of various aspects, not limited to their visual components. (3) Future research will include user validation through feedback collection and comprehension testing to verify whether improved readability metrics correspond to actual improvements in user understanding and usability of the reports.