This study employed a cross-sectional design conducted from January to May 2023 to investigate safety climate, safety behavior, and work-related injuries in 22 randomly selected large-scale construction sites across five major cities in Ethiopia. According to the official report from the Federal Construction Regulatory Authority, more than 120 large-scale construction sites were operating in the country during the study period. The cities included in the study were Addis Ababa, Adama, Bishoftu, Bahir Dar, and Hawassa, the cities were chosen for their high number of construction activities.

The sample size was determined using a single population proportion formula, accounting for a design effect of 2 and a 10% non-response rate, resulting in a total sample of 1250 workers.

A two-stage sampling approach was employed. First, 22 construction sites were selected using proportional representation from the total 88 construction sites operated in the selected regions. Specifically, 12 sites were chosen from Addis Ababa, three from Adama, two from Bishoftu, three from Bahir Dar, and two from Hawassa, based on the distribution of construction sites across these cities. In the second stage, workers from each selected site were chosen using a proportional-to-the-population-size sampling method, ensuring representation from different departments, including managers/supervisors, skilled workers, and unskilled workers. A list of workers, including their respective departments, was provided by each company and used as a sampling frame. Workers employed at the company for at least one year were eligible to participate. This criterion was chosen to estimate the one-year prevalence of self-reported injury rates. Study participants were then selected using a proportional-to-the-size approach. In most cases, we used the list of workers as a sampling frame to select individual participants. However, on some sites, it was challenging to use the list, especially when certain workers on the list were reassigned to other sites owned by the same company during the data collection period. In those cases, we randomly selected participants without using the list.

Data were collected through face-to-face interviews using structured questionnaires. We also observe each construction site using an observational checklist. The Nordic Safety Climate Questionnaire (NOSACQ-50) was used to measure the safety climate conditions in the construction companies and translated into more than 35 languages including Amharic (Kines et al. 2011; Marin et al. 2019; Yousefi et al. 2016). This validated tool consists of seven safety climate dimensions, comprising 50 items. Of these, 22 items evaluate safety at the management level, and 28 items at the worker’s level. A four-point Likert Scale (1 = strongly disagree, 2 = disagree, 3 = agree, and 4 = strongly agree) was used for the answers in the questionnaire. To measure safety behaviour, a set of nine items was adopted from previous studies (Mohamed 2002; Nadhim et al. 2018; Neal and Griffin 2006). Five items were used to measure the respondents’ level of participation in safety activities. Likewise, four items were used to measure the level of safety compliance. A five-point Likert scale was adopted to measure the response to each item, ranging from 1 to 5, in terms of strongly disagree, disagree, neutral, agree, and strongly agree, respectively. The details of the tool used in this study are outlined in the forthcoming publication (Abegaz et al. 2025).

A questionnaire template was loaded onto each data collector's smartphone with an Open Data Kit (ODK) and 20 trained and experienced data collectors participated in the data collection process. Pre-testing was conducted in one large-scale real estate development project site located around CMC area in Addis Ababa, which is not part of the actual study sites. Based on the pre-test results, adjustments were made to the data collection manual to clarify certain questions. Each interview session took on average 15 to 20 min. The interviews were conducted in a private setting with only the worker and one interviewer present; no personal identification was recorded, and the information the workers provided was kept confidential. Two senior supervisors monitored the data collection process. To ensure reliability and standardization across multiple interviewers, we implemented several measures before and during data collection. We provide comprehensive training that includes instruction on the study protocol, use of the ODK platform, and standardized approaches to administering the questionnaire, which is supported by preparing a detailed data collection manual. The training also included role plays and a field pilot to reinforce consistency. Additionally, supervisors closely monitored data collection in the field, conducting regular spot checks.

Data were analyzed using Statistical Package for the Social Sciences (SPSS) version 23 software (IBM Corp., Armonk, NY, USA). Data were checked for missing values, outliers, and inconsistencies. Descriptive statistics (frequencies, percentages, means, and cross-tabulations) were performed to summarize the data and the results were presented using tables and figures. The mean, standard deviation, and reliability of the safety climate and safety behaviour factors were calculated. The reliability coefficient for the safety climate and safety behavior scales was calculated using Cronbach’s alpha, with all dimensions showing reliability above 70%, indicating acceptable internal consistency. The multicollinearity was assessed using Variance Inflation Factors (VIF), and all the values were below the recommended level. Self-reported injuries were considered the dependent variable, while socio-demographic characteristics, company-related characteristics, work environment characteristics, safety climate, safety participation, and safety compliance were the independent variables. Bivariate and multivariate logistic regression analyses were performed using a binary logistic regression model. Odds ratios (OR) and 95% confidence intervals (CI) were reported, with a significance level set at p < 0.05.

Based on the Nationale Forskningscenter for Arbejdsmiljø (NFA) interpretation guideline of the NOSACQ-50 safety climate questionnaire, a score above 3.30 indicates a good safety climate in the workplace. Scores between 3.00 and 3.30 are considered fairly good, though some improvement is needed. Scores ranging from 2.70 to 2.99 are viewed as fairly low, indicating the need for improvement, while scores below 2.70 are considered low and signal a significant need for improvement (NFA 2017). Therefore, to make it more interpretable, we classified it into low levels (scores below 2.70 and scores ranging from 2.70 to 2.99) and good levels (scores ranging from 3.00 to 3.30 and scores above 3.30). We focused on the total safety climate score in this study.

Regarding safety participation and safety compliance, we interpreted the Likert scale weighted mean using the interval method for classification. This method divides the scale into specific intervals to assess the level of safety participation and compliance, providing a structured way to evaluate the responses based on the weighted mean score (Pimentel 2010). Accordingly, it is interpreted as strongly disagree (1.0–1.80), disagree (1.81–2.61), neutral (2.62–3.42), agree (3.43–4.23), and strongly agree (4.24–5.0). To make it more interpretable we classified it into low levels which include strongly disagree, disagree and neutral (1.0–3.42), and good levels include agree and strongly agree (3.43 – 5.0).

Large-scale building construction project: Building construction project with an estimated capital budget of above 100,000,000 Ethiopian Birr (ETB) which is nearly equivalent to $ 2 million.

Self-reported injury: Any physical injury during the work on the construction site that led to more than three days of absenteeism, two to three days of absenteeism, or required first aid without absenteeism, reported by the worker during the past year.