How the past influences the future: flood risk perception in informal settlements

ABSTRACT This study presents fresh evidence from an informal settlement in Accra, Ghana, examining how knowledge, understanding, experiences, and feelings about flood risk influence the flood risk perceptions of residents. The study adopted a mixed-methods approach, involving the collection and analysis of qualitative and quantitative data. We collected the data through seventeen interviews and 392 household surveys in Glefe, Accra, Ghana. We then conducted a thematic analysis of the qualitative data to understand participants' perceptions and the factors influencing their flood risk perceptions. The factors were used to produce hypotheses about flood risk perception. We subsequently performed regression analyses using the quantitative data to test the hypothesised relationships. The findings revealed that fear, flood experience, and coping experience were the major factors influencing residents' flood risk perceptions. Taken together, these factors had varying levels of influence on risk perceptions, with fear being the most statistically significant. However, it seems that experience held sway over residents' opinions, views, and perceptions. The perceived likelihood of future flooding events was therefore determined by residents' experience with flooding and coping. The study recommends incorporating the flooding and coping experiences of residents into adaptation mechanisms because these influence their perceptions of the flooding risks.


Introduction
Many urban areas in the developing world have seen a rapid increase in the frequency and severity of flooding hazards (Tasantab et al., 2020a(Tasantab et al., , 2020b. Such increased risk seems to be driven by climate change and urbanisation (Rana et al., 2020;Tasantab et al., 2021). While climate change may increase the frequency and severity of flooding because of intense precipitation (Christensen et al., 2013;Collins et al., 2013;IPCC, 2012;Kirtman et al., 2013), increased urbanisation puts more people in flood-prone locations (Amoako & Inkoom, 2018;Cobbinah et al., 2019). Many have argued that the effects of flooding hazards are felt more in informal settlements because of their vulnerability, poverty, and informality (Amoako & Inkoom, 2018;Tasantab et al., 2020a). It is further believed that developing countries are twelve times more likely to be affected by natural hazards than their developed counterparts (Brown et al., 2018;Tasantab et al., 2020a) because of the exacerbation of the depth and breadth of their vulnerability to disasters (UNDRR, 2019).
How these potentially vulnerable people perceive their risk of flooding matters in the adoption of adaptive response measures (Hudson et al., 2020;Rana et al., 2020). This research responds to priority one of the Sendai Framework that calls for an understanding of risk in all its forms, including vulnerability, capacity, exposure of persons and assets, hazard characteristics, and the environment (UN, 2015). As some of the most vulnerable people on earth, how those in informal settlements assess their risk can, therefore, complement scientific understanding and knowledge when context-specific policies, plans, and strategies are implemented to reduce risk (UN, 2015;UNDRR, 2019).
Few research publications on flooding in Ghana (Aboagye et al., 2013;Ahadzie et al., 2016;Tasantab et al., 2020b) have attempted to understand, in an in-depth way, the factors influencing the views, worries, and anxieties of people in informal settings about the probability of future severe flooding events. This area of research is thus still nascent in Ghana. This research, therefore, aims to investigate the underlying factors influencing flood risk perception/appraisal in a flood-prone informal settlement in Accra, Ghana. The findings will aid in the development of policies and programmes to enable the vulnerable populations of informal settlements to respond adaptively to flood risks (Sraku-Lartey et al., 2020). Beyond informal settlements in Ghana, this research will play a crucial role in efforts to adapt urban settlements to the changing climate. As governments worldwide invest in adaptation and resilience, understanding the factors that influence risk perception and the decisions and choices that result from these perceptions (Siegrist & Árvai, 2020;Tasantab et al., 2022) can help them deploy evidencedbased, community-centered risk communication measures to spur adaptation at the community and household level.

Risk perception and protection motivation
People's risk perception is framed by their opinions, culture, beliefs, worldviews, and biases (Guardiola-Albert et al., 2020;Samamdipour et al., 2019;Tasantab et al., 2020b). The way people perceive risk, process information about risk and make decisions regarding risk has been explained by many theoretical models, including the psychometric paradigm, the mental noise model, the risk perception model, the trust determination model, the negative dominance model, and the social amplification of risk framework (Paek & Hove, 2017). The psychometric paradigm provides a good theoretical foundation for this study because it explains how different people (non-experts and experts, those in informal settlements, and those in well-planned settlements) perceive risks differently (Siegrist et al., 2005). The paradigm recognises that non-expert people may treat the emotional and qualitative characteristics of hazards as serious concerns (Siegrist & Árvai, 2020).
Another dimension of this research is formed by underlying behaviour-change objectives. Risk perception in informal settlements needs to be understood, as priority one of the Sendai framework (UN, 2015) directs, so that appropriate measures can be developed and implemented to spur individual disaster risk reduction actions in line with sustainable development goal 11. The protection motivation theory (PMT) postulated by Rogers (1975) therefore provided the framework to support this research. This theory can be used to explain the decision-making processes and factors that influence risk perception (Tasantab et al., 2022). The proponents of the theory have postulated risk perception as a multifaceted construct that comprises perceived severity (including perceived probability) and vulnerability (Bamberg et al., 2017;Maddux & Rogers, 1983;Rogers, 1975Rogers, , 1983. While perceived severity (and probability) measures the likelihood of harmful consequences from hazards, perceived vulnerability assesses the individual's expectation of being exposed to the hazard. Rogers (1975) believed that the perception of a harmful event's seriousness and likely occurrence directly influences an individual's intention to protect themselves from the threat. These perceptions were believed to mediate the persuasive effects of risk communication by influencing protection motivation. Though fear may cause or result from risk perception (Raaijmakers et al., 2008;Terpstra, 2011;van der Linden, 2014), Rogers suggested that the essence of risk perception is not to emphasise fear as an emotion but as a protective motivation. It was understood that individuals will only be motivated to undertake flood risk protection measures when they perceive that there is a high likelihood of being inundated, and an equally high chance of harmful impacts from the hazard (Botzen et al., 2019;Bubeck et al., 2018;Grothmann & Patt, 2005).

Empirical relationship between experience and risk perception
Studies have observed that a relationship exists between experience and risk perception (Ogunbode et al., 2019;Osberghaus, 2017;Siegrist & Árvai, 2020;van der Linden, 2014). For example, van der Linden (2014) cited research indicating that personal disaster experience predicts climate risk perception. They also stressed that personal experience influenced affective responses to risks. In the African context, Adelekan and Asiyanbi (2016) and Mashi et al. (2020) reported that in some Nigerian cities people's awareness and personal experience shaped their perception of risk. In the Ghanaian context, studies have also elaborated on the flooding experiences of residents (Amoako, 2018;Mensah & Ahadzie, 2020;Tasantab et al., 2020b). However, the factors that influence the risk perceptions of informal settlement residents have not been sufficiently studied globally. For example, Siegrist and Árvai (2020) have argued that the relationship between experience and risk perception is mediated by factors such as socioeconomic variables. They also noted that risk perception is influenced by where people live and cultural differences. Thus, someone who lives in an informal settlement in Ghana may appraise their risk differently from one who lives in a formal settlement. Unfortunately, while most of the impacts of climate change-induced disasters are projected to occur in the informal settlements of developing countries, most of the discussions and theorising about disaster experiences are rooted in developed country contexts. That is why investigating how flood experience influences risk perception in informal settlements in a developing country context is important (Okaka & Odhiambo, 2019;Tasantab et al., 2022). As Siegrist and Árvai (2020) emphasised, it is important to improve understanding of risks so that risk communication capabilities can be improved.
Another empirical relationship that has not been sufficiently discussed is coping experience and risk perception. While Rogers (1983) revised the PMT to consider the influence of environmental factors and experience on threat appraisal (Rogers, 1983), it has still not been empirically established (at least in informal settlements) whether there is a statistically significant relationship between coping experience and risk perception. Many of the available studies more commonly investigated the relationship between experience and protective actions (Ogunbode et al., 2019;Osberghaus, 2017) and noted that having experienced negative emotions could affect people's perceptions of threat severity and individual vulnerability (Ogunbode et al., 2019). The PMT and its modified versions, therefore, did not consider the influence of coping actions on risk perception. Though Ogunbode et al. (2019) mention that coping could predict negative reactions to disaster risks, coping, in their context, referred to the emotional or psychological response to extreme events. It is relevant, therefore, to empirically establish the interaction between a coping experience (experience gained from undertaking coping measures, whether structural or non-structural) and risk perception. This is crucial in the context of informal settlements, since most measures taken to adjust to flooding are reactive and short-term coping measures (Mensah & Ahadzie, 2020;Tasantab et al., 2020b).

Risk profile of study area
Ghana has experienced repeated flooding events annually for several years. In the past 50 years, nearly 5 million people have been affected by annual flooding (Tasantab et al., , 2020b. Conservative estimates show that the last major flood disaster killed over 150 people (Tasantab, 2019;Tasantab et al., 2018Tasantab et al., , 2020b. The economic damage from flooding also surpassed USD780 million as of 2014 (Amoako & Inkoom, 2018;Tasantab et al., 2018;Tasantab et al., 2020b). Urban Ghana, especially Accra, seems to be the most affected. Evidence shows that most people had not recovered from the impacts of the 2015 floods two or more years after the events (Erman et al., 2018).
The settlement is inundated annually due to flooding caused by multiple factors, notably heavy rainfall and spillage of the Weija dam (Frick-Trzebitzky & Bruns, 2019; Owusu-Ansah et al., 2019). Though not many lives have been lost in Glefe, the economic damage caused by flooding is impoverishing (Erman et al., 2018;UNDRR, 2017). At present, many of the inhabitants respond to flooding risks through reactive (coping) measures (Amoako, 2018;Tasantab et al., 2020b). Coping refers to the short-term and reactive measures adopted during or immediately after flooding events to reduce the immediate impacts or to survive the hazard (Lavell et al., 2012;Tasantab et al., 2020a).

Mixed methods design
A mixed research methodology was used in this study. It comprised the collection and analysis of qualitative and quantitative data. The mixed methodology was adopted to provide a comprehensive understanding of the research problem. It was necessary to understand the perceptions and opinions of the residents on flood risk, subjective factors that required a nuanced view. It was also necessary to establish the statistical associations between the hypothesised relationships. The mixed methodology helped to achieve both objectives within the study. It also allowed for triangulation of the data. The quantitative and qualitative data were collected concurrently (Creswell & Plano Clark, 2011, pp. 58-88). However, the qualitative analysis was conducted first, and the factors found were used to formulate hypotheses. The quantitative data was then analyzed using regression to evaluate these hypotheses. This sequential approach to the analysis was necessary because time and resource constraints did not allow us to operationalise a full sequential design. The mixing of the data, therefore, occurred in a way that allowed the qualitative data analysis to build up to the quantitative data analysis. The qualitative data were collected through key informant interviews in Glefe. The interviews involved seventeen persons. These included the chief (1), the traditional custodian of the land (1), one (1) local government representative, local government unit committee members (2), and selected household heads based on the recommendation of the local government representative (12). The respondents comprised nine males and eight females. They were purposively selected from the community because their status as long-term residents meant they had more knowledge of the history of flooding events in the settlement. After the respondents consented to participate in the interviews, a semi-structured interview guide was used to facilitate face-to-face interviews. The questions were categorised into demographic and social characteristics (age, gender, household size, number of years lived in the community), flood experience and coping experience, and flood risk appraisal (flood risk perception). See supplementary materials for full interview questions. The views of the respondents were recorded, transcribed, and imported to NVivo for coding and thematic analysis. All respondent names were replaced with codes during analysis (e.g. HO1 = Household respondent 1 and LGR1 = Local government representative).
The quantitative data was collected through a Likert-scale-based questionnaire, using the same variables as the qualitative research. This helped to statistically assess the same issues covered in the qualitative data. The mode of data collection was face-to-face. The approach was adopted due to the informal nature of the community and the non-availability of data on the population. A list of indicators was obtained through literature and used in the questionnaire design. The indicators were measured from 1 = strongly disagree to 5 = strongly agree. The participants comprised 392 household heads (or their representatives) randomly selected from a total of 2368 households. Using a random route sampling approach and a map, the researchers visited each of the pre-selected houses for the data collection. The approach taken was that random numbers ranging from 1 to 1074 (number of houses in Glefe) were generated (using random.com). All the buildings on a Google map of the settlement were also labelled with the generated numbers. The settlement was then partitioned into sections and each research assistant was allocated one section. With the help of the numbered map, the researchers visited each house to recruit the household heads for participation in the study, starting from the house closest to the random starting point. The random starting point was the local taxi/minibus station. One household was drawn from each selected house and the household head was interviewed from that household. The random route (or random walk) approach was used because of the lack of a discernible settlement pattern and layout (Bauer, 2016;Tasantab et al., 2022). According to Flynn et al. (2013), a random route approach is a form of cluster sampling. If there were no eligible household members, or the household members refused consent, the next house in the direction of travel was selected to replace it. This was where the random route approach was more prominently used. It was hard to always follow the numbered buildings on the map, as some turned out to be dilapidated houses without occupants or kiosks used for other purposes. With the random route, replacement houses were found by conveniently choosing the next house. Figure 2 shows the map of Glefe with all buildings shown.
KoboCollect was used to collect the data. The data was then cleaned, processed, and analyzed using SPSS, WarpPLS, and Excel. The descriptive statistics, such as mean values and standard deviations, were obtained using SPSS. The rest of the data was processed with WarpPLS to generate factor scores for each of the latent variables. These standardised factor scores were imported to Excel to conduct the linear regression analysis. The research received human ethics approval (H-2018-0415) from the authors' institution of affiliation.

Limitations
A research endeavour is bound to have limitations and this study is no exception. One limitation was the fact that once the researcher selected a house to survey, they were expected to randomly select one household from that house, because there were instances of more than one household in a house. However, that was not always followed, as the houses were visited only once and only the household head at home during that visit was recruited. Also, the quantitative data was collected during a weekday when most men might have gone into the city for work. As a result, most respondents of the quantitative research were female. Notwithstanding the above, we believe the data was representative of the sample population, as the sampling was adequate and female household members had flood experiences, like the men.
Also, some respondents of the qualitative research were recommended by the Assembly member. It is possible that these recommendations were skewed to friends or family. To prevent bias, we recruited and interviewed additional residents to ensure that the data collected was representative of the settlement. We also ensured that general community opinions were captured by interviewing their democratically elected representatives, traditional leaders, and opinion leaders.

Introduction
The results of the qualitative analysis revealed that Glefe residents have experienced flooding events during the past 30 years. The respondents were mostly between the ages of 28 and 68 (eleven were above 50 years). Most of the residents were born and raised in the settlement. The number of years lived in the settlement ranged from 10 to 50 years, highlighting our focus on long-term residents. Most of the residents have had several experiences of flooding, giving them rich memories, feelings, and opinions about flooding experienced in the settlement. Each of the respondents had households ranging from 5 to 18 people, with an average household size of seven people. The residents were asked questions about their experience of living through flooding (flooding experience), what they did to cope with flooding (coping experience), difficulties after floods, and their appraisal of future flooding risks.

Flooding experience
The residents revealed that the community has flooded annually since the year 2000, with the worst flooding occurring in 2015. They attributed flooding risk in the community to urbanisation and population growth, resulting in high demand for land and the resultant encroachment of housing developments on floodable areas They further contended that the yearly spilling of the Weija Dam and the overflow of the two lagoons adjoining the community was further contributing to flooding risk in the community. They (15 respondents) therefore reasoned that flooding was part of the life of the settlement.
The analysis revealed that these issues influenced how they understood flood risk, experienced flooding, and perceived flood risk. One respondent noted that 'Flooding has been re-occurring for a long time. It is always massive, especially in the flood-prone They also pointed out the causes of the flooding. HO12 opined 'the lagoons bordering the community [are] filled with refuse and human excreta. Weeds have [also] taken over the lagoon … causing the flooding.' HO10 also observed that the flooding was caused by 'choked drains, resulting from poor drainage systems and improper disposal of refuse, because the laws that govern building construction and sanitation are not strictly enforced.' This 'has allowed people to build anywhere they find, causing flooding.'

Fear and emotional response to flooding
The experience of inundation and flooding-induced destruction appears to trigger fear of such inundation and destruction reoccurring. Residents, therefore, expressed fear of flooding and its impacts on them and their families.

Influence of flooding experience and fear
The flooding experience of the respondents was revealed as the major factor impacting their perception of flooding risk. Those who had experienced severe flooding or lost property and valuables during previous flooding events believed that future flooding and its impacts could be worse. They, therefore, expressed fear or dread because of that experience, with nine respondents believing flooding could worsen. For instance, HO10 remembered that the rainy season in Glefe is frightening. All you worry about is destruction either to persons or properties. I was terrified last time. I am still afraid for my life, which is mostly under threat in the rainy season. My family and property could suffer harm from floods in the future if nothing is done about the flooding. HO1 concurred: 'I was scared for my family during the recent flood. I am (still) terrified for our safety.' HO12 also opined 'the flooding could get worse because residents are still indiscriminately throwing refuse in drains. That makes it difficult for rainwater to flow freely, causing flooding.' HO15 also said, 'Severe rains could come in the future' and HO4 noted 'There is a great possibility for flooding to occur. The flooding could get worse.'

The outcome of coping actions impacts flood risk perception
The results further revealed that respondents who believed that their coping measures could prevent flooding of their properties felt safe and had the perception that future flooding was less likely to affect them. For instance, HO13 opined that 'I do not think it (flooding) can happen again because we have started implementing measures to prevent it.' While HO2 added that 'As community members, we have also used our little knowledge to put things in place to protect ourselves from future flooding. I do not think we are likely to experience a severe flood situation again.' HO14 also said, 'I do not think my family will suffer harm in the future because of the adaptation interventions I will put [in] place to avert such a situation.' Respondents HO3, HO13, and HO15 shared the view that 'Glefe is currently not under threat of flooding due to sea defense works.' However, LGR1 was of the view that the sea defense wall only protected a part of the settlement from tidal erosion.
Conversely, a small number of respondents perceived that coping actions were not sufficient to protect them and their valuables from adverse flooding. This group of people thus held the perception that future flooding could be worse. HO11 said 'I do fear for my safety because, besides coping with the situation over the years, nothing long-lasting has been done.' HO9 also said, 'There is nothing I can do to reduce the harm from floods.' The above analysis of the qualitative data revealed that flood experience, fear, and coping experience triggered perceptions of future flooding probability and severity. While it is evident that flooding experience and fear led to perceptions that flooding was highly probable and could be severe, the general view was that coping actions were sufficient to prevent the harm, leading to perceptions that the settlement was less vulnerable.

Hypotheses
Based on the findings of the qualitative analysis, the following hypotheses were formulated to statistically assess the relationship of flood experience (FE), fear (FW), and coping experience (CE) with flood risk perception/appraisal (FRA).
. H1: Fear has a positive relationship with the perception of flood risk in informal settlements.
H1 suggests that a heightened sense of fear of flood risk impacts leads to a strong conviction that future flooding would be severe and that people are vulnerable to flooding impacts.
. H2: Flood experience (FE) has a positive relationship with flood risk perception/appraisal in informal settlements.
H2 suggests that if informal settlement residents experience a flood event that hurts them, they will perceive that current and future flooding would be severe and highly probable.
. H3: Experience gained from coping with flooding has a negative relationship with flood risk perception/appraisal in informal settlements.
H3 postulates that if coping mechanisms were successful in reducing past flooding damage and severity, informal settlement residents would perceive that floods are manageable and thus less severe or probable, and vice versa. They may also perceive their properties as less exposed and themselves as less vulnerable to flooding.
Based on these hypotheses, the quantitative data was used to assess if flood risk perception/appraisal in informal settlements could be statistically predicted based on people's experience of flooding, coping, and fears.
6. Quantitative analysis of factors predicting flood risk perception

Descriptive statistics of indicators
The descriptive statistics of the socio-economic data show that most respondents were females (62.8%, n = 246), while the rest were male (37.2%, n = 146). The respondents 1 were mostly between the ages of 18 and 69 years (98.5%, n = 386). Their educational attainment ranged from primary school to postgraduate level, with 41.8% and 21.7% of respondents in the survey, respectively, attaining Junior Secondary and Senior Secondary education levels. In terms of tenancy, 46.7% of the respondents were homeowners while 42.1% were renters. Apart from the socio-economic data, the descriptive statistics of the indicators of flood experience (FE), coping experience (CE), fear (FW), perceived severity (PS), and perceived vulnerability (PV) were analyzed. Perceived severity and perceived vulnerability make up flood risk perception/appraisal. For the descriptive statistics of the indicators of the variables see Table 1.
As Table 1 portrays, the majority of residents responded neutral (3) or agree (4) (Likert scale ranged from 1 = strongly disagree to 5 = strongly agree) to questions about their flood experience. That explains their view that the impacts of flooding were severe, as indicators FE1 and FE2 show. In terms of coping experience, the majority of respondents agreed or strongly agreed that they rebuilt damaged walls (CE2), channelled water away (CE10), and repaired damaged roofs (CE11), showing that most coping measures were structural. They were also worried about their present and future safety as FW5, FW6, and FW7 show. Flood risk perception/appraisal was measured using perceived severity and perceived vulnerability. Responses to the indicators showed that the majority of residents believed flooding could become increasingly severe, their location was flood-prone, their housing was susceptible, and they were vulnerable. That was revealed by responses to PS and PV indicators. It was also clear that the residents required further information about the risks, as shown by responses to PS5 and PS6, underscoring the vital importance of risk communication. The results of the quantitative analysis, therefore, mirrored the qualitative results.

Simple linear regression of predictors of the residents' flood risk perception
Simple linear regression was conducted to test the hypothesis previously highlighted. These analyses use the variables fear/worry (FW), flood experience (FE), coping experience (CE), and flood risk perception/appraisal (FRA) found in the qualitative analysis. The variables FW, FE, and CE represent hypotheses H1, H2, and H3, respectively. Flood risk perception/appraisal (FRA) is a multifaceted variable comprising perceived severity (PS) and perceived vulnerability (PV).
6.2.1. Relationship between fear/worries about flooding and flood risk perception The analysis showed that the relationship between fear and flood risk perception/appraisal (FRA) 2 was statistically significant. The results further showed that R = 0.737, R 2 = 0.543, Adjusted R 2 = 0.542, Standard error = 0.677. The result thus indicates that 54% of the variance of FRA is attributable to fear. Fear was also positively correlated with FRA, indicating that fear triggered the perception of the high severity of flooding in the informal settlement, as hypothesis H1 postulated. Table 2 shows the regression coefficients. As shown in Table 2, the p-value is less than the 0.05 significance level.

Correlation between flooding experience and flood risk perception
The analysis further revealed a statistically significant relationship between flooding experience and flooding risk perception/appraisal (FRA) of residents. The figures were R = 0.5363, R 2 = 0.2876, Adjusted R 2 = 0.2858, Standard Error = 0.8451. The R-square figures show that flooding experience explained 28.76% of the variance of FRA. Flooding experience (FE) also showed a significant positive correlation with FRA, indicating that an increase in the severity of flood experience could result in a corresponding increase in FRA. This finding thus confirms hypothesis H2 of the study. Table 3 shows the regression coefficients.
6.2.3. Correlation between coping experience and flood risk perception Hypothesis H3 postulated that there exists a significant and negative correlation between coping experience and flood risk perception (FRA) in informal settlements. That means if coping mechanisms were effective in the past to reduce flooding impacts, the residents would perceive that they are less vulnerable, and flooding was less severe due to confidence that their coping mechanisms would remain effective. The opposite effect is assumed if coping mechanisms failed. The results showed R = 0.3236, R 2 = 0.1047, and Adjusted R 2 = 0.1024. This means that coping experience has a statistically significant but negative relationship with FRA. It also revealed that 10.47% of the variance of FRA is explained by coping experience. Thus, residents perceived that they were less vulnerable to flooding when there was a higher success rate of coping mechanisms. Table 4 shows the regression coefficients of the relationship.

Multiple regression of factors influencing flood risk appraisal
After the results of the simple linear regression, a multiple regression analysis was conducted to assess the combined influence of the three explanatory variables on flood risk perception/appraisal. This was necessary because, in real-life, multiple factors may contribute to or predict flood risk perception/appraisal, since risk perception is a multifaceted construct (Sjöberg, 2000). The results revealed R = 0.7469 and R 2 = 0.5579. The model was statistically significant. The three variables thus explained 55.79% of the variance of FRA.
The results revealed that flooding experience (FE) and fear (FW) were significant and positively correlated with FRA, as hypotheses H1 and H2 postulated. However, coping experience (CE) was not statistically significant and therefore different from the simple linear regression result. Further analyses were conducted to exclude multicollinearity problems in the model. Multicollinearity occurs when two predictors measure the same attribute of a construct in multiple regression models (Kock, 2015;Kock & Lynn, 2012). Multicollinearity checks revealed acceptable variance inflation factors (VIFs) of FE = 1.98, FW = 1.6, and PC = 1.54, indicating that multicollinearity may not be the issue. However, when CE was removed from the model, further analysis showed a reduction in the regression coefficient of FE. Thus, the inclusion of coping experience (CE) in the multiple regression increases the effect of flooding experience (FE) on FRA. Further analysis also revealed that there was a significant linear relationship between FE and CE, with a coefficient of −0.58, and R = 0.58, showing that the two variables are highly correlated and measure the same underlying construct. It may thus be advisable to combine the two variables in a multiple regression model. Furthermore, the exclusion of the CE from the multiple regression model did not reduce the explained variance of FRA, showing that CE contributes little to the prediction of flooding risk perception in the multiple regression model. Table 5 shows the results of the multiple regression analysis.

Discussion
This study addressed a gap in the literature that has received little attention in academic discussions. How do informal settlements perceive current and future risks? What influences these perceptions? Answers to these questions are needed due to the increasing population of informal settlements (Satterthwaite et al., 2020;Tasantab et al., 2022), their increasing exposure to disaster risks driven by climate change and the need for context-specific data that captures their culture, opinions, worldviews and perceptions (Tasantab et al., 2020b). As Cook and Overpeck (2019) discuss at length, the required impact on disaster risk in informal settlements can only be achieved when policies are driven by the voice and experience of residents. Existing Western-focused research on risk perceptions, as Hudson et al. (2020) argues, is insufficient to provide the nuanced understanding required to formulate policies needed to address risk and hazard issues affecting informal settlements. Research that does focus on informal settlements dwell mostly on their physical and socio-economic characteristics, and coping strategies (Ahadzie et al., 2016;Amoako, 2015;Owusu-Ansah et al., 2019;Satterthwaite et al., 2020).
In the context of Ghanaian informal settlements, this study is one of only a few (Tasantab et al., 2020b(Tasantab et al., , 2022 that have investigated the factors influencing risk and adaptation intentions. It is well known that informal settlements have often been neglected by governments and policymakers (Satterthwaite et al., 2020;Tasantab et al., 2022). It is thus important for researchers to let their voices and opinions be heard, and this study contributes to that.
The results reveal that when the residents of the Ghanaian informal settlement experience adverse impacts from a flooding event, they are likely to extrapolate from that experience that any future flooding, or its impacts, would be severe. Residents in this study believed that future flooding could be worse (more severe and more frequent) than past floods due to their flooding experiences. Hudson et al. (2020) and others (Guardiola-Albert et al., 2020) support this finding by showing that experience is an important predictor of risk perception. The results have underscored that in the Ghanaian informal settlement, negative experiences during past flooding events can and do influence residents to consider their vulnerability and to become informed about the severity of future flooding. These negative experiences serve as a mental shortcut to assess the probability and severity of future flooding. The majority of residents were concerned about flooding risks and are eager for credible information on flooding severity and potential impacts. This confirms the view that risk can be predicted based on experience (van der Linden, 2014). What is new, however, is the finding that the success of coping measures can be an indicator of whether risk is taken seriously or not. The results thus confirm the following hypotheses: . H1: Fear has a positive relationship with the perception of flood risk in informal settlements (supported). . H2: Flood experience (FE) has a positive relationship with flood risk perception/appraisal in informal settlements (supported). . H3: Experience gained from coping with flooding has a negative relationship with flood risk perception/appraisal in informal settlements (not supported 3 by the multiple regression).
The linear regression analysis and qualitative findings reveal that when the success rate of coping measures is high, complacency sets in, and people begin to belittle their exposure and vulnerability to disaster risks. The result of hypothesis 3, therefore, amplifies the view that response efficacy and self-efficacy, represented in this research by coping experience, not only predict adaptation intentions (Hudson et al., 2020;Tasantab et al., 2022) but also risk perception. Sometimes, coping strategies fail to be effective (Hudson et al., 2020;Parvin et al., 2016;Paul & Routray, 2010). What makes one coping measure fail while an alternative succeeds depends on several factors, including the severity of the flooding, the level of resources, and the extent of exposure (Paul & Routray, 2010). The findings reveal that when a coping measure fails in the informal settlement, it produces strong perceptions of community and individual vulnerability to the increased severity of flooding (Hudson et al., 2020;Parvin et al., 2016). However, when coping experience is positive (that is when coping measures work), the perception of vulnerability or exposure weakens. As shown by the qualitative analysis, some people believed that flooding may not be severe or affect the community because of the coping measures they had taken or government structural programmes, such as sea defense walls. However, the experiences the community shared in this study belie this belief. Thus, informal settlement residents' coping experience can evoke different responses to disaster risk.
Furthermore, fear, as discussed in the literature, could be a product of risk perception or may influence risk perception (Raaijmakers et al., 2008;Terpstra, 2011;van der Linden, 2014). The results of this study show that heightened fear about flooding can provoke strong perceptions about flooding severity and vulnerability. Also, unpleasant emotions attached to being inundated or suffering damage or losses during a flood event can evoke a strong sense of fear and the resultant belief that flooding risk could materialise and cause further damage. This explains why the regression analysis produced a strong statistically significant relationship between fear and flood risk perception/appraisal. Fear thus plays an important role by amplifying the sense of urgency about flooding risk. In this sense, fear may play a significant role in informal settlement flood resilience.

Conclusion
Informal settlement residents want to be heard. They want their knowledge, views and perceptions about disaster risks to influence policies and programmes that affect them. This study is one of the first in the context of Ghana to attempt an investigation of the factors influencing risk perception in informal settlements. Moreover, this study is also novel in analyzing the relationship between coping experience and risk perception. The findings strengthen knowledge of the adjustments of informal settlements to flooding risks, not just in Ghana but also in other developing countries with similar contexts. The findings provide context-specific information on local flood risk knowledge and practices that is crucial to the success of any flood risk reduction measures in the informal settlements.
The findings thus provide an opportunity for policymakers and disaster risk managers to leverage flooding experience, coping experience, and fear of future flooding to undertake initiatives encouraging informal settlements to adopt protective measures to build resilience to flooding. That may mean providing credible information on disaster risk predictions and people's vulnerability through evidence-based risk communication and encouraging informal settlements to consider their community and individual circumstances about flooding risk to prevent a false sense of security. For the informal settlement, the past is ever-present and is the lens through which they view flooding risk. As many informal settlements around the world continue to suffer the impacts of disaster risks, studies such as the present one are needed to understand their perceptions and acceptance of risk to inform risk communication and adaptive capacity-building measures.

Future research directions
This study explored risk perception in the context of Ghanaian informal settlements. One area of future research would be to provide a wholistic view of risk perception and the nuanced differences between formal, rich, educated sections of urban communities and informal settlements to offer further insights into how different segments of the urban society perceive disaster risks. It would also be useful to explore how community leaders and policymakers integrate scientific data and community risk perceptions in policies and programmes on disaster risk reduction. Notes 1. Based on 2010 population and housing census, Glefe was inhabited by 8,738 people (approximately 2,368 households and 1074 houses). The sample is representative of this population. Households in 36% of the houses participated in the survey. 2. The variable (FRA) was obtained by using WarpPLS to generate factor scores for each of the latent variables. To do this, we first processed the raw data using WarpPLS, generating standardized indicators. We then conducted structural equation modelling analysis of the relationship between PS and PV, using their indicators. Factor scores of PS and PV were generated through the analysis and added to the dataset as new standardized indicators, converting PS and PV from latent variables to indicators. Based on literature, PS and PV were then used as indicators of FRA. For more information on this process, see Kock (2020). 3. A negative and statistically significant correlation was evident in the simple linear correlation.
However, the multiple linear correlation was not statistically significant.

Disclosure statement
No potential conflict of interest was reported by the author(s).