A range of variables, identified based on expert knowledge and literature review, are considered important for estimating intangible flood damage. The variables are grouped by flood characteristics, building factors, and socio-economic characteristics. Figure 2 shows the eleven independent variables: flood depth, flood duration, building type, proximity to water bodies, business type, household size, years of living (residential) or years of operation duration (business), ownership, income, the presence of elderly individuals, and the presence of children. Despite the number of possible exogenous variables that could have influenced the psychological effects of flooding, this study focuses on eleven subset variables to take advantage of the contact time with respondents and the quality of responses. All of the explanatory variables were used to assess their correlation with the intangible damage. Whilst some of the respondents shared their recovery experiences of post-flooding during the interview, the recovery rates were not being considered explicitly as one of the variables in the multivariate analysis.

The survey questionnaire was designed to allow for separate analysis of residential and business. Residential buildings were classified as village-type, terraced, or bungalow, while businesses were categorized as micro, small-to-medium, or large businesses, similar to that proposed in KTA Tanaga Sdn Bhd (2003). These classifications represent the primary building structure types prevalent in Malaysia. The categorization of business premises type was based on the total number of workers, as specified by SME Corporation Malaysia (2022). Businesses were categorized into micro and small businesses based on the number of full-time permanent employees. Micro businesses were defined as having fewer than five employees, while small to medium businesses included those with five to thirty employees (SME Corporation Malaysia, 2022).

As for the socio-economic characteristics, three classifications were used for the demarcation of the residential sector income levels following the national standard and the parent study (i.e., Kabirzad et al., 2024) (Department of Statistics Malaysia, 2020). The classification is B40 (bottom 40 %), M40 (middle 40 %), and T20 (top 20 %). The B40 group included those earning less than USD 1130 per month, the M40 group covered incomes between USD 1130 and 2553.45 per month, and the T20 group consisted of households earning more than USD 2553.45 per month. For business, micro business income is between USD 73 367–3 668 380, and small-medium income is USD 3 668 380–7 333 170 (SME Corporation Malaysia, 2022).

The intangible suffering due to the psychological effects of flooding was valued using the WTP approach. The monetary values that the respondents are willing to pay to alleviate flood damages were taken as a proxy, where the absolute value of WTP originally in Malaysian Ringgit (MYR) was converted to US Dollars (USD). The study aimed to assess damage using multiple variables and understand the multiple variables' contributions. Figure 3 shows the process in which the survey was made in person and where the data screening for missing values and duplication is applied. Further explanation of the process is discussed in the next sub-sections. Ultimately, a multivariate regression analysis was undertaken to identify the contributing factors of intangible damage to residential buildings and business premises.

Intangible damages of flooding for households primarily involved diminished well-being, stress, anxiety, and sleep disturbances linked to the difficulty in managing possessions and recovering. For businesses, intangible flood damages might stem from concerns about employees' well-being and subsequent disruptions on operations and recovery (Lekuthai and Vongvisessomjai, 2001). To quantify the non-market values for alleviating psychological health resulting from flooding, the present study adopts similar work by Foudi and Osés-Eraso (2022) to estimate the WTP of people who have flood experiences and could experience flooding again,.

During the engagement sessions, conversations with household heads, business managers, or owners in flood-exposed areas were conducted in Malay, as many residents felt more comfortable speaking in their native language. The conversations were informal at first, but shifted to a more formal tone once participants were willing to engage further. The engagement was conducted cautiously to ensure interviewees received sufficient information about the study's purpose before delving deeper into the WTP questions. A sequential information-sharing process was applied before respondents were gradually brought into the topic and questions regarding WTP. Respondents can be emotionally affected by their past devastating flood experiences (Joseph et al., 2015), and this can be addressed by validating respondents' psychological experiences prior to questions being asked.

After listening to and validating their stories, they were briefed on the effects of stress and anxiety. Additionally, they were informed about the challenges of monetizing intangible flood damage, which requires the adoption of the contingent valuation method (Markantonis et al., 2012; Semrau et al., 2016; Joseph et al., 2015). To avoid confusion between intangible and tangible losses, and to focus WTP responses solely on psychological and mental health effects, the distinction between the two was explained to the interviewees. They were advised not to include financial or asset losses in their WTP values (e.g., Foudi and Osés-Eraso, 2022). Respondents were then presented with a dichotomous choice question on their willingness to pay for flood mitigation that potentially reduces the intangible damages of flooding that they experienced. Respondents who answered “yes” were then asked an open-ended question on how much they are willing to pay. Those who stated “no” were asked for their reasons. Interviews for open-ended questions were only continued for respondents who answered “yes”. From the face-to-face conversations, those who refused to pay were influenced by their strong views about the payment vehicle that should be part of the government's responsibility. Such a protest bid is common in similar social studies, and exclusions of the protest bids are necessary to reduce bias in the analysis.

Meanwhile, a number of questions were adopted to guide responses on WTP values associated with alleviating psychological burden. For instance, some of the interviewees have built concrete barriers at the opening of their house to protect from flood intrusion. Such a real case was used as a surrogate to the WTP value, but with emphasis on alleviating their psychological burden. Example questions are: (1) how much they are willing to allocate to reduce their stress and anxiety if the same flood event were to occur?, (2) if the same event as the worst that they had experience is to happen in the future and they were not at home, will the property-level protection barriers that they have constructed able to ease their psychological burden of not be able to save possessions while they are not there? If yes, then the costs for the barriers were asked, or (3) if the respondents do not yet have property-level barriers, they will be asked if they are willing to spend on one, and by how much to alleviate the same psychological burden as they have experienced during the worst flood experienced before. The price of WTP is adjusted for the inflation rate using the Malaysian Consumer Price Index calculator to maintain consistency and comparability across different periods (Department of Statistics Malaysia, 2021). Furthermore, the results were presented as an absolute value in US dollars (USD) for a flood event. Information on the selected variables for the multivariate analysis, such as building types, income level, etc., of the respondents was then collected. All information was stored using an online standard form, including the coordinates of the building's location where the interview took place.

The collected information of WTP represents the anxiety and stress at their personal level related to the businesses that they are managing, and the stress and anxiety stemming from impacts on productivity and disruption of sales, etc., manifested from the condition of flood events that they are in. Whilst the description is not exhaustive and pathways of intangible damages on individuals and businesses that they own or work in may overlap, the present study does not express a distinction. It is perceived that the intangible damages can be of any disturbance to the running of the businesses that cannot be monetised, whether it stems from a personal level of the business owner/worker related to the businesses, or from more specific losses to the business, such as loss of opportunity that they are stressed and anxious about. Moreover, the WTP does not consider economic losses that can be monetised indirectly, for example, due to business downtime. The exclusion was made clear during the interviews (e.g., Darnkachatarn and Kajitani, 2025).

The data that has been collected was filtered to identify duplicates and missing information for all variables. The data pre-processing was conducted for both sectors in preparation for the correlation and regression analysis. Missing information was imputed with values having similar characteristics to the ones recorded (e.g., Rodríguez Castro et al., 2025). All datasets of the same type of variable from different areas were combined into a single database. Building type, business size, the presence of elderly or children, and ownership status were treated as binary variables, whilst others were treated as continuous variables. Outliers in all continuous variables were identified and treated using winsorizing, the z-test, or three standard deviation cut-offs, and the Mahalanobis distance method. After completing this process, the normality, linearity, and homoscedasticity were assessed. Subsequently, a multicollinearity assessment between the variables to identify highly correlated independent variables was undertaken. The regression analysis on intangible damage for residential types of buildings indicated multicollinearity between two independent factors, which are flood duration and distance from the river. Therefore, the distance from the river was excluded to improve the accuracy and reliability of the regression, which leaves only nine independent variables.

For the normality test, some datasets were first transformed to accommodate the variables in normality using the log or log-log transformation (Svenningsen et al., 2020). For residential buildings, the datasets were log-log transformed except for independent variables, such as building type, presence of elderly, presence of children, and ownership status. Meanwhile, for commercial buildings, the datasets remained untransformed except for income data, which was log-transformed. Linearity was checked to ensure that the dependent variable is linearly related to each independent variable. Violations can reduce model accuracy, though slight deviations may be acceptable depending on context. In this case, residual plots confirmed a generally linear relationship, but some continuous variables did not meet the assumption. Another assumption is that the residuals exhibit constant variance, ensuring the absence of heteroscedasticity (unequal variance), Heteroscedasticity can bias regression estimates and reduce predictive accuracy. Therefore, statistical transformation, such as logarithmic adjustment or other methods, is applied.

After the pre-processing of datasets was completed, multivariate regressions were performed to explore the relationship between intangible flood impacts in terms of victims' psychological burdens and the considered factors. These models enabled the identification of key factors contributing to damage severity related to the three groups of factors: flood characteristics, building types, and socio-economic conditions (Lee, 2020). The regression analyses for this study considered three significance thresholds, such as 10 %, 5 %, and 1 % (e.g., Lamond et al., 2015). Table 1 presents the detailed descriptions of explanatory variables considered. The goal of the analysis is to identify predictors that statistically influenced the psychological burden of respondents due to flood events. Meanwhile, standardized coefficients β were utilized to determine the relative significance of each predicting variable (x), allowing for comparison of their effects on intangible damages (Y). The final models include error terms that address unexplained variations. These models were expressed as a general Eq. (1), encompassing all essential variables to ensure reproducibility for future research investigating flood impacts in comparable settings (Svenningsen et al., 2020). 1Damage(Y)=β0+β1x1+β2x2+…+βkxk+ϵ…. The model's performance was evaluated using the coefficient of determination (R2), which indicates the proportion of variance in the damage explained by the independent variables (e.g., Poussin et al., 2015). However, R2 is not always recommended as a sole indicator of model accuracy, as it can be artificially inflated by adding more independent variables (Jarantow et al., 2023). Furthermore, a low R2 does not necessarily indicate a weak relationship, as statistically it is heavily influenced by the variation in the independent variables (Hamilton et al., 2015). Therefore, its interpretation must always be contextualized within the specific research scope (Hair et al., 2018).

A total of 380 respondents were approached, of whom 217 provided valid responses. Eliciting monetary valuations for psychological impacts proved challenging, particularly when respondents were asked about their willingness to pay (WTP) to reduce flood-related psychological distress. Despite the use of a sequential and tailored interview approach to build respondents' trust, some participants expressed disagreement when asked about their willingness to make monetary contributions to safeguard themselves from the psychological effects of flooding.

Of the valid respondents, only 141 responses (107 residential and 34 businesses) expressed willingness to pay for disaster risk reduction measures, while 76 respondents (35 %) stated zero WTP or refused to contribute. Common reasons included limited income, the belief that flood mitigation is the government's responsibility, lack of trust, and the perceptions that flood impacts were not severe or were primarily emotional. Such “protest zero” responses are common in contingent valuation studies for flood reduction measures (e.g., Hanley et al., 2009; Brouwer et al., 2009; Jones et al., 2015), and are typically excluded to reduce bias (Foudi and Osés-Eraso, 2022). In this study, zero WTP responses were excluded to focus on identifying the drivers for the non-zero range of WTP. Furthermore, WTP was used as a proxy for the intangible damage, which is unlikely to be zero in the context of damaging flood events.

Most respondents were from residential buildings, reflecting limited commercial activity in the study areas. Local information indicated that several flood-affected businesses had permanently relocated to safer locations. Time constraints and demanding schedules also limited participation among business owners, resulting in excessively long waiting times and limited interviews. Consequently, the dataset is dominated by residential respondents. Given the difficulty of collecting socio-economic and psychological data, data validity relied partly on expert judgment.

Table 2 summarises the number of residential and business respondents. Forty-four percent (44 %) of residential respondents and sixty-seven percent (67 %) of business respondents were only from the Kota Bharu (Kelantan) study area, where terrace buildings accounted for nearly 40 % of all buildings, followed by low-cost housing. Respondents from the Segambut district of Kuala Lumpur were few, and primarily resided in terrae and low-cost houses. In Kajang and Dengkil areas of Selangor, most respondents also lived in terrace houses, while Kota Bharu district respondents were more evenly distributed across building types.

Among business respondents, the predominant type of business is micro-sized enterprises (76 %), followed by small to medium businesses. The small-to-medium businesses (SMEs) represented 24 % of the study sample. Kota Bharu recorded the highest flood depths for both residential and business premises, largely attributed to the major flood event in 2014. Although the case study sites experienced flooding in different years, all locations have been severely affected by flooding over the past decade.

Variations in intangible damage across residential households and businesses were analysed in terms of income groups, business size, and the distance from the river using both total and average WTP values. Total values refer to the cumulative WTP within each category. Meanwhile, the average values represent the mean WTP. Comparisons across income groups are presented in Figs. 4 and 5.

Bar charts were used to examine patterns of flood exposure and intangible damage across income groups and business sizes. The analysis focused on whether lower-income households and micro businesses experienced greater exposure and demonstrated higher WTP to reduce psychological impact. Intangible damage was assessed using both total and average WTP across income groups, business size, and buildings' distance from the river. categories.

Figure 4 presents that most residential respondents belonged to the bottom 40 % (B40) income group, followed by the middle 40 % (M40) income group, while the top 20 % (T20) group had the fewest respondents. Given that the interviewed people are randomly approached based on the criteria of their building's proximity to rivers, this distribution highlights the disproportionate exposure of lower-income households to flooding. Only 13 and 5 respondents belonged to the M40 and T20 groups, respectively, compared with 86 respondents from the B40 group. Average WTP value ranged from USD 46.6 to 186.4 across all income groups. While the B40 group recorded the highest total losses due to its larger sample size, median WTP values per household were comparable across all income groups.

Figure 5 presents results for the business premises. Micro businesses were more exposed to flooding than small to medium businesses, reflected in a larger sample size (24 versus 7). Micro businesses also showed a higher total WTP value, largely due to their greater representation. However, the median WTP value was similar across business sizes, consistent with the findings for residential households. This suggests that lower income or smaller business size does not necessarily imply significantly lower WTP.

Residential households reported higher intangible losses than businesses, reflecting their greater perceived need for interventions to reduce psychological and mental effects associated with flooding. However, comparable median WTP values across income and business size categories indicate that income alone cannot be identified as a decisive factor based on single-variable analysis. These findings may change with a larger sample size, highlighting the challenges of collecting data from flood-affected populations. While descriptive analysis provides useful insights, multivariate analysis is required to identify key determinants of intangible flood damage.

This section presents results of multivariate regression for residential and business sectors, examining building characteristics, socio-economic factors, and flood attributes influencing WTP. The objective is to identify the relative strength and significance of intangible damage drivers.

Despite the insights obtained, the analysis suggests that the current sample size may be insufficient to fully model intangible impacts across both sectors. A primary limitation is the lack of statistically significant associations in the business sector, which is largely attributable to small sample sizes. In the residential sector, while flood duration and family size were significant predictors, the distribution of the sample may result in bias and the statistical insignificance of the other variables. Specifically, smaller sample sizes for the middle income (M40) and high income (T20) than the B40 income group may limit the accuracy of the residential findings related to income level. Therefore, increased sample sizes from diverse flood-prone regions are necessary to improve model prediction and account for the heterogeneous nature of these sectors. Nonetheless, this requires laborious efforts and resources to target both the quality and quantity of sample sizes.