Factors of Influence on Evacuation Behaviour: Survey Results from the Riverine Floodplain Communities in Bangladesh

The data for this study were collected using a questionnaire survey administered in the three flood-affected districts (Nilphamari, Rangpur, and Gaibandha) of the northern region of Bangladesh. The questionnaire survey was conducted from April 2019 to May 2019. Focus group discussions (FGD) with participating local leaders and residents were also used to collect relevant information. These interviews (FGD) allowed in-depth exploration of a variety of information, including, but not limited to, the characteristics of the flood, vulnerability and capacity, flood early warnings and evacuation, sheltering, and relevant issues that riverine people experienced during the 2017 flood.

The household-level questionnaire survey was administered in three unions, selecting one union from each district (Purbachhatnai from Nilphamari, Gajaghanta from Rangpur, and Belka from Gaibandha) (Fig. 1). These three unions are located on the right bank of the Teesta River in Bangladesh and have flat topography (Table 2). The principal flooding mechanism is the overflow of the riverbank. This overflow of the riverbank is accompanied by the peaks of floods along with widespread transportation and deposition (Zwoliński 1992). Floods are recurrent in these three unions. The questionnaire used was pretested and it recorded households’ socio-economic information, flood characteristics, vulnerability and capacity of household, perception, household response, and other relevant information. An individual household was considered the primary sampling unit in this study. A complete list of all riverine households in the selected unions was not available. To select households from the riverine area, a systematic random sampling procedure was used. It was decided that the number of households to be interviewed from each union should be proportional to its size. The total number of households for these three unions was 18,972 (BBS 2013). A sample size of 377 was calculated using Cochran’s Formula (Cochran 1977). The calculated sample sizes for each union were: 68 for Purbachhatnai, 158 for Gajaghanta and 151 for Belka. In each selected household, the male or female head of the household or the elderly was interviewed.

Fig. 1

The descriptive analysis was performed using Pearson's χ² test. A multivariate logistic regression model was used to determine the factors affecting a household's decision to evacuate. Because evacuation decision is in binary form (yes or no), logistic regression is an appropriate statistical technique.

To identify factors of influence on evacuation decisions, a total of nine independent variables were selected based on the theoretical understanding, literature review and availability of data (Table 3). It is noted that all components of SCT are not entirely applicable for Bangladesh (Paul 2012). Therefore, the analysis of this study can be seen as a partial application of SCT.

The average age of the household head was 48.9 years at the time of the questionnaire survey. Table 4 shows that only 20 of the 377 households were headed by a female member. About 33.4% of the households reported a monthly income of <6000 BDT, whereas only about 23.9% earned >10,000 BDT. The data also show that 48% of all households were agricultural landless (Table 4). Among the agricultural landholders, around 66% (128 out of 195) households owned between 0.21 acres and 1.00 acres. Agriculture was the primary occupation for 63.1% of all households, followed by wage labour (22.8%). Concerning the level of education, around two-thirds of the households had studied beyond the primary level (Table 4). These socioeconomic characteristics of the households clearly indicate that riverine people are poor and have a lower level of education. These factors make riverine people more vulnerable to hydro-meteorological hazards, particularly floods, compared to people in other parts of the country.

The results of the questionnaire survey revealed that 82% of households received flood early warning in the 2017 flood. However, only 40% of households were evacuated from their residence in a safe place including flood shelter (4.8%), school building (4.8%), flood protection embankment (21.0%), and neighbour's house (18.0%) (Fig. 2). The highest percentage of households from Gajaghanta (χ² = 19.96, df = 2, p ≤ 0.010) were evacuated in a safe place. Around 44% of households received a warning message from a local leader or through a megaphone announcement. Mass media also plays an important role and only 26% received warning information from television/radio/newspaper. Nearly 60% of the households received a warning message from their social connection (e.g. neighbours, relatives, friends). Around 18% of households got warning information from the mosque. However, only three households received information from NGO officials.

Fig. 2

All the explanatory variables (Table 2) were included in the binary logistic regression model. Results of the binary logistic regression model of factors affecting evacuation behaviour to riverine flood disasters are presented in Table 5. Multicollinearity among the explanatory variables was checked using correlation analysis. The model is statistically significant, likelihood ratio χ² = 86.50, p ≤ 0.001. The Hosmer and Lemeshow χ² test was non-significant for the model.

The study results showed that the age of the household head was negative and significantly related to evacuation. This implies that older household head is less likely to evacuate. This could be explained by the mobility issues as the age of the household head for the majority of the households with only two family members were >45 years. In the context of a hurricane in the USA, Alawadi et al. (2020) and Adeola (2008) reported that older household heads were less likely to evacuate. This could mean that disaster managers should design the evacuation plan in such a way that will better accommodate the elderly.

The results also showed that the probability of evacuation decreases with the increased level of education. This implies that educated households are less likely to evacuate. A review of the field data reveals that the majority of educated households with an education level of >9 grade had an income of > 10,000 BDT/month, possess agricultural lands and mostly lives far away from the riverbank (>1 km) and modified their houses with strong materials to protect them from being flooded. This finding is different from other studies (Medina, Moraca 2016) that showed that educated households are more likely to evacuate. However, Lim et al. (2016) did not find any relation between education and evacuation.

Households with disabled/chronically ill members are more vulnerable to natural disasters than others (Islam et al. 2018). The findings of this study suggest that the likelihood of evacuation was higher among households with disabled/ill members. Studies reported that concerns about the well-being of vulnerable family members (e.g. chronically ill, disable) often lead to partial evacuation (Paul 2012). This finding necessitates the proper evacuation plan for the disabled and chronically ill members to evacuate in advance of a flood. These findings are consistent with Adeola (2008) that reported households with disabled people were more likely to evacuate than households without disabled people.

Table 5 shows that the evacuation rate was lower among the households that earned > 10,000 BDT/month. This result is in the line with previous findings that showed that wealthier households are less likely to evacuate (Medina, Moraca 2016). This might be because higher-level income groups have adapted more strategies at their disposal to mitigate risk against flood.

The result shows that the evacuation rate was lower among the households who owned livestock compared to households without having livestock, but this difference is not statistically significant. This may be because shelters are not equipped to keep livestock which demotivates households to evacuate. Therefore, instead of taking shelter in a safer place with their livestock, they prefer to stay at home with their livestock. Studies noted that cattle owners from the coastal region of Bangladesh are less likely to evacuate to cyclone shelters (Saha, James 2017).

Affiliation with NGOs did not appear as a significant predictor of evacuation behaviour. This may be because NGOs are mainly supporting livelihood enhancement overlooking the long-term resilience of riverine people. Besides, floods are recurrent in the study areas. Thus, the evacuation behaviour was not significantly different across affiliated NGO groups.

The height of floodwater inside the room was positively related to evacuation. This implies that riverine people still consider the ‘wait and see’ approach for evacuation. People refuse to evacuate as long as the floodwater does not rise within their homestead. Another reason could be related to their experience with the previous flood (Ourn, Suntornvongsagul 2015), which leads them to act upon seeing the characteristics of the flood. In the case of risk management, it is thus essential to raise the awareness of people to act upon getting the evacuation notices. Besides, the warning messages should be disseminated in an easily understandable way. This result is in the line with previous findings that showed that households who experienced deeper inundation were more likely to evacuate than those who experienced shallow flooding depth (Medina, Moraca 2016).

The location of the house from the riverbank did not influence evacuation behaviour. This is because, irrespective of their house location, a majority of the surveyed households faced inundation of their houses in 2017 and thus no significant differences were observed. Some scholarly evidence showed that proximity to water positively influenced hurricane evacuation (Wilmot, Bing 2004).

Surprisingly, the results revealed a negative relationship between evacuation and warning messages. This may be because more than half of the households received warning messages <6 h, which discouraged households to evacuate in a safe place. This finding is consistent with earlier research findings of Shah et al. (2012). Results of the FGD, (Purbachhatnai) revealed that the early warning system in the study area is very poor. In most cases, they received information about the increase or decrease of the water level of the Teesta River. But they do not know how much the water level might increase in the coming days. People usually use environmental cues (e.g. cloud colour, wind direction, exceptional behaviour of animals, observing celestial bodies, and river watercolour) to guess river flood (Mondal 2012). However, people become helpless when there is an onrush of water from the upstream due to the opening of barrage gates (Mondal et al. 2020).

On the other hand, the probability of evacuation was significantly higher in Gajaghanta than in the other two survey locations.