Background and Literature Review
Bangladesh is a South Asian country identified by the IPCC as one of the most vulnerable least developed countries (Intergovernmental Panel on Climate Change, 2014). It is the eighth-most populous country in the world and its administrative structure consists in eight divisions and sixty-four districts, each subdivided into upazila or thana, that are the second lowest level of regional administration in Bangladesh. The smallest rural administrative and local government units in Bangladesh are the unions. It has been ranked 7th on Global Climate Risk Index 2019 of the countries most aﬀected by climate change since 1998 (Eckstein, Hutfils, & Winges, 2018). Due to its geographical position and its environmental features – being a flat, low-lying deltaic country – Bangladesh is exposed to disasters of diﬀerent types, ranging from extreme temperatures, increasing number of severe floods, coastal erosions, higher frequency of cyclones and erratic rainfall. The climate of Bangladesh is considerably heterogeneous over the year and has rather marked seasonal variations (Thomas et al., 2013). More than 80% of the annual precipitation occurs during the monsoon season between June and September (see Figure A5) and recently the weather pattern has become more erratic with a shorter cool and dry season between November and February (see Figures A5 and A6) and a positive trend in the average yearly temperature between 1981 and 2012 (see Figure A10).
Bangladesh has a natural resources-based economy, predominant in agriculture, with more than 35% of the population living below the poverty line. Agriculture heavily depends on weather and there is evidence of the negative eﬀect of increases in temperatures and changes in rainfall patterns (Aragón et al., forthcoming; Burke & Emerick, 2016; Schlenker & Roberts, 2009) on crops. The impact of climate change can thus be amplified in countries as Bangladesh that heavily rely on agriculture. Rain fed agriculture hinges upon seasonal rainfall and South Asian monsoon is the most important climatic phenomenon directly linked to the intensity and frequency of rainfall and drought in the country (Dastagir, 2015). Natural disasters such as floods, droughts and storms, that are expected to increase with climate change, can have both short- and long-term impacts on households, who mostly depend on agriculture, aﬀecting savings, creditworthiness, loss of livelihoods and agricultural damages (Intergovernmental Panel on Climate Change, 2014; Maccini & Yang, 2009).
Bangladesh has a growing season extending throughout 12 months, that can be divided into three over-lapping cropping seasons. The three seasons are articulated following the production of three diﬀerent types of rice, which is the staple and the most produced crop of Bangladesh (Johnson, 1982). The three main varieties of rice are Aus, that is sown from March to June, during the pre-monsoon season, the Aman, that is the dominant monsoon season rice (from July to October), and the Boro, that is the dry season irrigated rice, sown from November to February. The crop rice calendar of Bangladesh is extended to the other crops and determines three diﬀerent growing seasons, characterized by diﬀerent weather characteristics: Kharif 1 is the pre-monsoon season that goes from March to June, rainfall is variable and temperatures are high. Besides the Aus rice, summer vegetables and pulses are grown during Kharif 1. Kharif 2 is the second part of the Kharif season. It is the monsoon season occurring between July and October and it is characterized by heavy rain and floods. Aman is the major crop grown, while other fruits and summer vegetables can be grown on high lands. Rabi is the winter dry season going from November to February, with low or minimal rainfall and low temperatures. Boro rice, wheat, potato, tomato, cabbage, spinach, Chinese cabbage, cauliflower and oil seeds are examples of winter vegetables (Paul & Rashid, 2016). Therefore, three diﬀerent growing seasons are defined, Kharif 1, Kharif 2 and Rabi, extending the approach in Guiteras (2009) and Carleton (2017), who, in their studies set in India, define only the Kharif (monsoon crop) growing season and Rabi (winter crop). Time series of the average temperature and total precipitation by growing season over time between 1981 and 2012 in Bangladesh are reported in Figures A11 and A12, where it is possible to observe an increasing trend in average temperatures in each growing season and a decreasing trend of rainfall only in the Rabi season, that has become drier over time.
Beliefs and Subjective Perceptions
The philosophical and phenomenological literature has extensively investigated the relationship between belief and perception (Smith, 2001). The main open question still unanswered is whether the perception involves the belief in the object of perception. Contemporary philosophers have conjectured that perceiving something leads to having a particular perceptual belief about that object (Armstrong, 1969).
In the framework of climate change, studies on the perceptions of this phenomenon have been conducted both in developing (Ishaya & Abaje, 2008; Mertz et al., 2009; Vedwan & Rhoades, 2001) and developed (Akter & Bennett, 2011; Leiserowitz, 2006; Semenza et al., 2008) countries. In both settings, results confirm that the majority of the population has already perceived climate change. This, however, is not immediately linked to a spread belief in climate change, as there is still a lively debate on the narrative behind the phenomenon of climate skepticism, that is present both in emerging and developed economies (Lejano, 2019). Regarding the definition of perception of climate change, Whitmarsh and Capstick (2018) provide an adequate interpretation of this concept that denotes a range of psychological constructs, that include knowledge, attitudes, concern, beliefs and perceived risk. This term captures the cognitive, aﬀective and evaluative dimensions of the internal representations of the individuals of the notion. These representations, however, are not exempt from influences of social processes and cultural context, that shape their formation (Whitmarsh, Seyfang, & O’Neill, 2011).
The articles that study the subjective perceptions of climate change have usually focused on the impact of diﬀerent socio-demographic factors. Maddison (2007) and Ishaya and Abaje (2008) have concluded that farming experience has an important role in the perception of climate change, whereas Semenza et al. (2008) showed that higher income is associated with a stronger perception of the changes in climate. Other drivers of the perception of climate change such as gender, ethnic background and newspaper reading (Leiserowitz, 2006), education, access to extension services and soil types (Gbetibouo, 2009) have also been studied. There is a small strand of the literature in developing countries that has compared farm surveys with data records from meteorological stations, but this has always been limited to descriptive comparison (Hageback et al., 2005; Thomas et al., 2013; Vedwan & Rhoades, 2001).
A particular feature that characterizes the perception process is the distinction between sudden-onset weather events and gradual long-term climatic changes (Intergovernmental Panel on Climate Change, 2014). In fact, climate change is usually interpreted as a long-term shift in climate patterns and usually refers to slow-onset changes such as increases in temperature and changes in rainfall patterns. Nevertheless, climate change can also be perceived through the increase in the frequency and intensity of extreme events such as floods, droughts, tornadoes, cyclones and hailstorms. In this regard, there is a strand in the psychological literature that investigates the diﬀerent impact that gradual changes and extreme weather events can have on subjective perceptions. The theoretical background behind this hypothesis associates more power to experiential processing than to analytical processing in driving decision-making and behavior. Therefore, close experiences of climate change could influence climate opinions and perceptions more than longer-term, gradual or distant climatic change. Studies (Druckman & Shafranek, 2017; Fownes & Allred, 2019; Hamilton
& Stampone, 2013; Joireman, Truelove, & Duell, 2010; Risen & Critcher, 2011) have concluded with results that confirm this hypothesis.
Another cognitive bias that can be related to the personal experience of climate change and the formation of subjective beliefs is the confirmation bias (Kahneman & Tversky, 1973; Nickerson, 1998). An individual suﬀers from confirmation bias if she tends to misinterpret ambiguous evidence as confirming her current beliefs (Rabin & Schrag, 1999). Coupled with the confirmation bias is the availability bias (Kahneman and Tversky, 1982), that arises when « […] people assess the frequency of a class or the probability of an event by the ease with which instances or occurrences can be brought to mind” (Kahneman and Tversky, 1982, 11) and that has already been studied in the context of how individuals form beliefs over the likelihood of natural disasters (Gallagher, 2014). The introduction of such biases in the beliefs formation process of the agents relaxes the assumption of Bayesian rationality, through which agents start with subjective beliefs over diﬀerent possible states of the world, they learn about an unobserved environmental change by observing the weather and then use Bayes’ rule to update the prior beliefs (Kelly, Kolstad, & Mitchell, 2005).
Literature on climate change has also introduced in the theoretical agents’ learning models the « recency bias » (Kala, 2017). This corresponds to the tendency of an individual to most easily remember something that has happened recently, compared to remembering something that may have occurred long time before.
Even in robust learning models, this bias persists since agents use old information in more forecasts and the agent would best respond by treating the earlier signals as less informative.
The analysis of subjective perception of climate change and its drivers is relevant in order to understand whether, besides technological, economic and ecological limitations, there can be « social barriers » that hamper adaptation to climate change (Raymond & Spoehr, 2013). Social barriers are defined as normative, cognitive and institutional obstacles to adaptation, where the first refer to the interactions between the individual and the social environment and the second refer to individual psychological and thought processes (Weber, 2016). Directly observing subjective perceptions can allow to empirically test the hypothesis that lack of adaptation is driven by a diﬃculty in recognizing changes in climate and eﬀectively adapt to them.
The IPCC has defined the adaptive capacity as the ability of systems, institutions and humans to adjust to climate change, including climate variability and extremes, in order to alleviate potential damages and cope with the consequences (Intergovernmental Panel on Climate Change, 2014). Thus, implementing an adaptation strategy should increase the capacity of a system to survive external shocks or change. The International Food Policy Research Institute (IFPRI) has defined adaptation as the process of improving the ability to cope with changes in climatic conditions both in the short term, e.g. seasonal or annual, and in the long term, e.g. in decades (Nhemachena & Hassan, 2007).
Agriculture has been identified as the primary channel through which the impacts of climate change are transmitted to poor and rural households. The literature (Di Falco et al., 2011) has identified a number of potential climate change adaptation strategies that rural households have undertaken in order to tackle the negative eﬀects of climate change on agriculture. Adaptation strategies can be classified into three main categories: i) reallocation of economic activities through the diversification between on-farm and oﬀ-farm activities (Barrett, Reardon, & Webb, 2001; Colmer, 2018; Molua, 2011); ii) changes in consumption and savings patterns (Fafchamps, Udry, & Czukas, 1998; Hisali, Birungi, & Buyinza, 2011; Kazianga & Udry, 2006); iii) on-farm production decisions (Binswanger & Rosenzweig, 1993; Hassan & Nhemachena, 2008).
In the first type of adaptation strategies, farmers could decide to leave agriculture because of considerable losses in productivity due to climate change and choose to seek employment in other sectors of the economy such as industry or services. Bryan, Chowdhury, and Mobarak (2014) found that Bangladeshi rural house-holds respond to incentives that relax their liquidity constraint when they make seasonal migration decisions during the lean period. This decision might include geographical relocation, with households choosing to migrate to more productive regions across states. Gray and Mueller (2012) have investigated population mobility in Bangladesh driven by natural disasters and found that, despite natural disasters having relevant eﬀects on long-term population mobility in rural Bangladesh, exposure to them does not have significant positive eﬀects on overall mobility and it might reduce mobility by increasing labor needs at the origin or by removing the resources necessary to migrate.
Rural households and farmers might also change their consumption and savings behavior in response to natural disasters and climatic changes exposure. Forward-looking agents would usually save to smooth their consumption during disasters or in the aftermath of them. Eskander, Fankhauser, and Jha (2016) showed that Bangladeshi rural households tend to shift from farm to non-farm employment in order to tackle negative shocks in the household income from exposure to floods and storms but maintain similar levels of savings.
A final set of strategies includes on-farm adaptation strategies. These strategies are usually implemented by households who cannot bear the cost of migrating or seeking an oﬀ-farm employment, and who heavily rely on the agricultural income. Therefore, they find more eﬃcient to adjust their farming practices, bearing a cost of eﬀort in terms of adaptation and learning new techniques. For this reason, these adaptation strategies are mostly related to gradual changes in climate such as changes in temperatures or precipitation, whereas migrating or seeking oﬀ-farm employment would be the most eﬀective strategy in response to exposure to extreme natural disasters such as cyclones or tornadoes. Examples of on-farm adaptation strategies are changes of cropping practices – in terms of timing and variety of plantations – diversification between crop and livestock activities, adjustment of quantities of inputs applied or the update of technological methods (Aragón et al., forthcoming). This set of strategies extends also to soil conservation techniques, shading and planting trees, change in the use of irrigation/groundwater and other watering strategies (Deressa et al., 2011). Maddison (2007) suggests that stratifying the adaptation strategies by diﬀerent changes in climate and perceptions, such as increases in precipitations, increases in temperatures, increases in cold waves or changes in the timing of rainfall, would provide greater insights on the adaptation decision-making process. For example, if temperatures are perceived to change, farmers might change variety of plantations or increase the use of water conservation techniques and the use of shading and sheltering techniques (Lobell & Burke, 2009). Furthermore, for a perceived change in precipitation and its timing, farmers would tend to respond by varying the planting date (Kala, 2017).
It might be argued that if farmers behave as forward-looking agents and they anticipate disaster shocks, they could adapt to them. However, such shocks cannot be perfectly anticipated because of their increasing high frequency, especially in south-east Asia (Mirza, 2011). Even though farmers usually show experience of coping strategies by considering seasonal risks and uncertainties in agriculture, the magnitude and frequency of stresses and shocks deriving from climate change are changing (Davies, Guenther, Leavy, Mitchell, & Tanner, 2009), making it more diﬃcult for them to adapt on time. For this reason, studying the subjective perceptions of climate change and their drivers can shed light on the adaptation decision-making process.
Household-level data for Bangladesh are taken from the Bangladesh Climate Change Adaptation Survey (BCCAS), that consists in a two-round survey. It is designed by the International Food Policy
Institute (2014a). Baseline data are collected as part of a study undertaken with 800 agricultural households in 40 randomly selected unions (administrative units) in Bangladesh. The survey is funded by the United States Agency for International Development (USAID) and it was designed and supervised by the IFPRI together with the Center for Development Studies (ZEF) and the Data Analysis and Technical Asssistance Limited (DATA). It was administered by the Bangladesh Centre for Advanced Studies. The first round of the survey was conducted from December 2010 to February 2011, and it covers data from the previous production year, between December 2009 and December 2010. The dataset provides information at the national level on demographic characteristics, land tenure, crop management, incidence and perception of climatic shocks and adaptation options for Bangladeshi rural households. A follow-up second round of the survey (International Food Policy Research Institute, 2014b) was conducted in the following year, in September 2012, and it covers data from the previous production year between September 2011 and August 2012. A timeline of the survey rounds with respect to the three agricultural growing seasons identified earlier for Bangladesh is reported in Figure A2.
The respondent is the head of the household. A household is defined as a group of people who live together and take food from the same pot. It counts as household member anyone who has lived in the household at least six months, and at least half of the week in each week in those months. People who do not share blood relations with the head of the household (e.g. servants, lodgers or agricultural laborers) are considered members of the household if they « have stayed in the household at least 3 months of the past 6 months and take food from the same pot » (International Food Policy Research Institute, 2014a). An agricultural household is defined as such if it complies with at least one of the following requirements: i) it was operating cultivate land (either owned, leased, shared or mortgaged); ii) it owned 5 or more livestock; iii) it raised 50 or more poultry. Bangladesh is divided into seven broad agro-ecological zones (AEZs), as grouped by the Bangladesh Centre for Advanced Studies (2014), based on the 30 AEZs categorization operated by the Soil Resource Development Institute (SRDI). The BCCAS covered 40 unions selected to represent proportionally the 7 AEZs, which are Barind Tract, Beel and Haor Basins, Floodplain, Himalayan Piedmont Plain, Modhupur Tract, Northern and Eastern Hills and Tidal Floodplains, that present very heterogeneous climatic character-istics. The heterogeneity in monthly precipitations across AEZs is documented in Figure A4. The number of unions randomly selected from each AEZ is reported in Table 1. For each sample union, twenty agricultural households were randomly drawn from a single village in each union, for a total sample of 800 households. Figure 1 shows the location of the unions in the survey and the seven diﬀerent AEZs in Bangladesh. Based on such agroecozone stratification, the sample was built to be national representative. However, the absence of a recent agricultural census made it impossible to assign weights. Therefore, observations are unweighted. More than 97%, or 766 out of 800 households, from the first round have been reinterviewed in the second round. The remaining 34 households could not be interviewed because they migrated or were not at home at the time of the survey. In particular, only 15 migrated. Given that the interest of the research is on subjective perceptions, the final sample includes only those households who have been surveyed in both waves, did not move between the two rounds and whose respondent was the same in both waves. For this reason, the resulting final sample is a balanced panel dataset with two periods and 714 observations each. The geographical
distribution of the households included in the final estimation sample is reported in Table A1. There are no substantial diﬀerences in the geographical distribution of the full and final estimation sample, which provides support to the absence of bias when focusing only on the 714 remaining households.
Households Table 2 presents key summary statistics of the households and their agricultural charac-teristics by survey wave and defines the setting for the empirical analysis. Most of the households have a male head (94%). The average number of years of education for the head of the household is 3.5, that would imply an incomplete primary education, confirmed by the fact that the literacy rate among the heads of the household is around 50%. There are on average 5 members in the household. The average farm does not seem to rely on hired labor force for farming, whereas above 70% of the households has at least one member of the household self-employed in farming activities. Child labor force occurs only in less than one fifth of the households. Most of the rural households (around 90%) own livestock (cattle, buﬀaloes, goats, sheep, pigs, chicken or ducks), whereas only less than a third of the sample owns fishery assets. This provides support to the fact that most of the households are mainly focused on farming activities and fishery is only a supplementary activities that can be either performed on the south coast or on tidal ponds. Even though the ownership of a tractor or power tiller is not very spread (around 2% of the sample), households usually own other agricultural equipment such as plough or threshers. In the first round, the average area of land holdings
is around 0.68 hectares and the average cultivable land is 0.56 hectares1. There is considerable heterogeneity in the soil type of the cultivable/arable land held by an household, that predominantly is clay-loam. The total cultivable land reported by the households in the survey is subject to diﬀerent operational status: around half of the cultivable land is owned by the household. Around 30% of the cultivable land of an household was rented in the first round, with the average share dropping to around 20% in the second round. These are the two most common statuses of cultivable land in the sample that make up to 75% of the total cultivable land of an household across waves. Remaining options such as mortgage, rent out or fallow have small non substantial shares.
Table of contents :
2 Background and Literature Review
2.2 Beliefs and Subjective Perceptions
2.3 Adaptation Strategies
3.1 House hold Data
3.2 Historical Weather Data
3.3 Extreme Events and Natural Disasters Data
4 Theoretical Framework
4.1 Subjective Perceptions
4.2 Behavioral Response : Adaptation Strategies
4.3 Expected Utility
5 Empirical Approach
5.1 First-Step:Subjective Perceptions
5.1.1 Confirmation Bias : An Instrumental Variable Approach
5.2 Second-Step:Behavioral Response
5.2.1 Do belief saffect behavior?
6.1 Heterogeneous impact of weather event son subjective perceptions
6.2 Beliefs,Accuracy and Behavioral Response
6.3 Robustness Checks