What Drives Choice of Coping Strategies Against Ex-Post Weather and Climate Shocks Among Smallholder Fish Farmers in Malawi?

The study was conducted in Dowa and Mchinji districts of central Malawi to understand how fish farmers make choice of village savings and loans, income generating activities and insurance to deal with weather and climate shocks. A multivariate probit model was used to analyse the drivers of farmers’ choice so that we know if the strategies are complements or substitutes. Data from 353 fish farmers was collected. The study has shown that education level, age of the farmer, farming experience and the hypothetical insurance bid are significant drivers of fish farmers’ choice. We recommend promotion of adult literacy, civic education and enhancement of extension services on the coping strategies as a proxy to formal education which has proved so crucial in the adoption of coping strategies.

advocate diversification away from a sole reliance on customary activities need to recognize the level and opportunity for market-based activities. In addition, these interventions must consider different sensitivities to different hazards as well as community homogeneity in order to effectively support rural communities to cope with climate variability. Uddin et al. (2014) in their study on the drivers of households coping strategies towards environmental degradation and climate change used logit model to identify the socio-economic factors affecting the farmers' adoption of adaptive strategies. The results showed that almost all households perceived that climate is changing and the key drivers to their choice of adaptive strategy towards climate change was the social economic status of the households under consideration. Further the adaptation strategies that were hinted among others include crop diversification, integrated farming system, crop insurance and use of irrigation farming.
Against this backdrop of hastily varying weather conditions and the severity of the impact on poor fish farmers, it is urgent that the various options available to smallholder fish farmers to cope are documented. A thorough understanding of farmers' choice of coping strategies is critical for the success of any efforts to mitigate the negative effects of weather and climate shocks on fish farmers' well-being. As argued by Berman et.al, (2014), better understanding of farmers coping process will enable more targeted interventions and policy formulations. This being the case, we investigated what drives fish farmers' choice of coping strategies against weather and climate shocks. We examined the factors affecting choice of coping strategies to deal with weather and climate shocks.
The findings of this study are crucial in informing the design and implementation of strategies to curb the negative repercussions of climate change on fish farmers' well-being. Further, the analysis of fish farmers' choice of coping strategies contributes on information as to how differences in socio economic characteristics affect their choices.

Approaches to modelling coping and adapting to climate change and variability
Some studies have used univariate analysis on adaptation studies in Malawi and elsewhere. For example , Zulu et.al, (2017) used univariate probit to analyze the determinants of individual households' coping strategies to weather and climate shocks. Farid et.al (2015) used multiple regression approach to examine the extent of adoption of improved farm practices by the farmers of Northern Bangladesh and determined the influencing factors for adopting these practices. Further, Montle and Teweldemedhin (2014) applied a multinormial logit to determine the factors affecting farmers' levels of strategy diversification in the face of climate shocks in Namibia. A departure in terms of the approach is Mulwa et.al (2017) who used a multivariate probit approach to model how farmers in Malawi respond to climate risks. The latter approach is more realistic considering that farmers use a multiple options to deal with shocks. Using the same econometric method the study contributes to the literature by using both the pond level (farm level) and household level data to analyse drivers of farmers' choice of coping strategies to deal with weather and climate shocks in the context of smallholder fish farmers in Malawi. To the best of our knowledge, no study has been done in Malawi that specifically target fish farmers' coping to weather and climate shocks. Many studies on climate change adaptation have focused on crop production other than specifically fish production.

Theoretical framework
The study established a theoretical framework following a random utility theoretical structure. Random Utility Model (RUM) defines a choice resolution of which an individual has a set of substitute coping strategies from which to select (McFadden, 1978). It is presumed that each coping alternative has its specific qualities which influence individual's choice. RUM aids us address fish farmers' choices over alternative coping strategies (us Saqib, 2004). This model is grounded on that an individual originates utility by selecting a number of alternatives. The utilities are latent variables, and the noticeable preference indicators manifest the essential utilities (us Saqib, 2004). Following Uddin et.al (2017) the utility that an individual derives from selecting strategy J, from a choice set (C) of alternatives can be described as follows: = , + (2.1) In equation (2.1), q is a vector of characteristics of each coping strategy = 1,2, … , as selected by an individual . Z is a vector of household specific characteristics. is the error term. An individual jointly selects coping strategy j from a set of alternative strategies. A strategy is assumed to be selected from an overall set of strategies only if and if only the expected utility is more than the actual utility of all other bundle of coping strategies (C) (Uddin et.al, 2017).

Methodology
Using face to face interviews, household level data was collected on 353 fish farming household heads in Dowa and Mchinji Districts of central Malawi. The data included household socioeconomic characteristics, pond characteristics, household's perception on climate and weather shocks, and households' coping strategies against weather and climate shocks. A multi-stage sampling method was applied to select the 353 households. Firstly, a purposive or judgment sampling technique was used to select the Traditional Authorities (TAs) with high probability of getting fish farmers in Mchinji and Dowa. Then, sections from each selected TA were also selected purposively in order to make sure that the sections with more fish farmers are captured. Finally, probability proportional sampling method was used to select villages to sample our farmers.

Model specification and estimation strategy
We followed a Multivariate Probit Model (MVP) approach as opposed to the univariate modelling. The multivariate probit approach concurrently models the effect of a set of explanatory variables on each different coping strategies and at the same time allowing for the possible relationship between unobserved disturbances, as well as the association among the adoptions of diverse strategies (Mulwa et.al, 2017). The equation in a system of multiple coping choices can be shown as = + = 1,2,3, … , (4.1) And the observable dichotomous choice variables are; Where is the utility that the fish farmer gets by choosing strategy from the given set of options. In the MVP model where the adoption of several coping strategies is possible, the error terms jointly follow a multivariate normal distribution (MVN) with zero conditional mean and variance normalized to unity (Kpadonou, et al., 2016). MVN (0, Ω) and the symmetric covariance matrix Ω is given by (4.3) In the equation (4.3) above, % (#ℎ ) is the pairwise correlation coefficient between the error terms. Kpadonou, et al., (2016), state that, the sign and the significance of the correlation coefficient, %, provide evidence on the nature of the relationships between adoption equations i.e. a complementarity for a positive sign and substitutability for a negative sign. The multivariate probit estimator allows the covariance between the error terms to be correlated across strategies but not across observations within a given practice as shown in equation below (Marenya & Barrett, 2007).

Dependent Variables
Being a multivariate approach, the study had three coping strategies namely village savings and loans (VSLs), rainfall indexed insurance (RII) and other income generating activities (OIGAs) of which fish farmers choose from. These choices were also the dependent variables. The strategies were chosen in line with the literature on adaption to ex-post weather and climate shocks, augmented by the focus group discussions with the selected fish farmers per district.

Independent Variables
The study incorporated explanatory variables based on the review of the available literature on farmers' adaptation to weather and climate related shocks and the availability of the variables in the data set. The variables were mainly on socio-economic characteristics of the fish farmers (age of household head, household size, marital status, farmer education, and years of farming). Since the study is on aquaculture, the study also pays special attention to pond size, fingering expenses, feed expenses, fish sales and operational status 1 . The data shows that fish farming in Dowa and Mchinji is dominated by male headed households (78%) and the average age, years of education and years of farming of the household heads were 43.78 years, 6.21 years and 5 years respectively. Further, the average household size in the sampled households was 6 with over 49% of the households having 6 to 9 members. The mean annual income for the sample was MK221035.2 (US$291.71). There was however, high variation in income across the sampled farmers as evidenced by high standard deviation than the overall mean and median incomes. Furthermore, income is positively skewed 2 (skewed to the right) meaning that the majority of the households have their incomes below the mean income (Nicholson and Snyder, 2008). The data also reveals that during the 2017 stocking season 1 , 27.48% of the farmers in the entire sample had pond sizes in the range of 201-500 M 2 with the average being 1108 M 2 . In terms of expenditure on fingerings, 71.67% of the sampled households spent in the range of between MK0 to MK50000 and on average, the farmers spent MK8060.057 (US$10.75). Overall, an average of MK9085.27 (US$12.11) was spent on feeds. The average net fish sales for the farmers was MK35019.55 (US$46.69). Over 61% of the farmers were full time farmers.

Results and Discussions 5.1 Drivers of Fish Farmers' Choice of Coping Strategies against Weather and Climate Shocks 5.2.1 Diagnostic tests
A correlation matrix was developed which showed weak correlations between independent variables. This shows that multicollinearity was not a serious problem in the model. Furthermore, the Multivariate test for normality (mvtest with normality sub-option), was used in stata to test for normality and the results show that the errors of the fitted variables were normally distributed at the 1% level. Endogeneity was tested as proposed by Hausman (1978). Endogeneity makes identification of causal effect difficult because of biased estimates (Katengeza et.al, 2018). The possible sources of endogeneity in this study could be reverse causality. To control for endogeneity, the study used the control function approach as hinted by Katengeza et.al (2018). In the control function, each of the endogenous variables (bid price, income and operating status) was used to generate a new variable called a proxy variable (predicted variable) and it is these predicted variables that were used in the final model instead of the original endogenous variables. The idea behind the control function correction is to derive a proxy variable that influences the dependent variable and is dependent on the error term (Petrin & Train, 2009). The proxy variable makes sure that the variation in the endogenous variable be independent of the error and standard estimation approaches will again be consistent. The approach is used in the absence of proper instrumental variables of the endogenous variables 5.2.2 Farmers' choice of coping strategies Tables 5.1, 5.2 and 5.3 summarizes the drivers of farmers' choice of coping strategies. Being a multivariate probit model, the results were interpreted based on the fact that each choice had its own drivers in the overall and district disaggregated models.  Table 5.1 shows that years of schooling positively and significantly influenced the choice of insurance at the 5% level. A one year increase in education level increases the probability of adopting insurance by 0.077(7.7%). This is probably because education enables farmers to have access to new information and ideas, and therefore capacitating them to develop a more understanding on the importance of insurance against weather and climate shocks. These results concur with Berman et.al. (2014) and Asgary etal. (2004) who also found a positive relationship between education level and choice of coping strategies against shocks in western Uganda and rural Iran, respectively. The disaggregated model shows that education has a positive and significant influence at the 10% level on adoption of insurance in Mchinji District (Table 5.3), but could not significantly explain the adoption of insurance in Dowa District (Table 5.2). Table 5.1 shows that farmers who said yes to a given bid of the premium 1 had a probability of adopting insurance by 125.1% for the first bid and 106.4% for the second bid. The findings concur with Trang (2013) who found a positive and significant relationship between the bid value and adoption of insurance in a study on the determinants of rice farmers' willingness to pay for area indexed insurance in Vietnam. Across districts, Tables 5.2 and 5.3 reveal that bid price had an influence on choice of insurance in both Dowa and Mchinji. Table 5.2 shows that acceptance of the bid prices by farmers in Dowa raised the probability of adopting insurance by 100.3% for the first bid and 183.8% for the second bid. In Mchinji, acceptance of the bid prices raised the probability of choosing insurance by 145.6% for the first bid and 98.2% for the third bid (Table 5.3). Table 5.1 shows that number of years of farming had a significant influence on the adoption of insurance in the aggregate model at the 10% level of significance. A one year increase in the years of farming raised the probability of adopting by 7.7%. However, the disaggregated model in Table 5.3 reveals that the number of years of farming significantly raised the probability of adopting insurance in Mchinji District at the 5% level. A one year increase in the number of years of farming increased the probability of choosing insurance by 19.1%. This could be the case because being in the fish farming business for more years as was the case with Mchinji ,meant that the farmers had more years of exposure to weather and climate shocks and their impacts. Such farmers are likely to develop a skill(s) to deal with such shocks, in form of participating in an insurance scheme or otherwise. The findings are similar to those of Iheke and Agodike (2016) who noted that years of farming may give an indication of the practical knowledge a farmer has acquired on how to overcome certain inherent farm production problems, some of which are effects of climate change.

Effects of age of the farmer on choice of RII
Generally, age of the farmer significantly influenced farmers' choice of coping insurance against weather and climate shocks in the aggregate model at the 10% level. A one year increase in the age of the farmer reduces the probability of adopting insurance by 2.1%. This may be because most farmers in the study area are old and are probably risk averse 2 to the operations of insurance. They have lived in the area for many years and accumulated some experience pertaining to operation of insurance. Mulwa et.al. 2017 argued that older farmers with experience easily understand coping strategies and this may also explain why age of the farmer negatively influences adoption of insurance in fish farming.

Effects of operational status on choice of RII
Operational status of the farmer, which is the state of either being a full-time fish farmer or a part-time fish farmer, affected choice of insurance at the 1% level (Table 5.1). In the overall model, Table 5.1 further shows that being a full-time farmer reduced the probability of choosing insurance by 63.4% than being a part-time farmer. These results concur with Iheke & Agodike (2016) who noted that being a full-time farmer gives one a chance to understand inherent problems on one's farm more easily than when one is a part time farmer. This explains why full-time farmers in Dowa and Mchinji had a low chance of joining insurance than part-time farmers (Tables 5.2 and 5.3). In Dowa, being a full-time farmer reduced the probability of adopting insurance by 105% more than being a part-time farmer (Table 5.2). Operation status could not significantly affect the adoption of RII in Mchinji District (Table 5.3).

Effects of shock experience on choice of RII
In the overall past shock experience significantly influenced the choice of insurance at the 1% level (  Ghazanfar et.al. (2015).

Effects of marital status of choice of RII
Marital status of the household significantly influenced the choice of insurance at the 10% (Table 5.1). Being married reduces the probability of adopting insurance by 20.4%. However, the disaggregated models reveal that marital status significantly influenced the choice of insurance in Dowa District at the 5% level. Being married reduced the probability of choosing insurance as a coping strategy to weather and climate shocks by 42.1%. The observed results conform to the fact that married people may have more responsibilities and due to such, most may be resource constrained. Thus strategies that require initial/ upfront capital are likely to side line more married than unmarried people ( : Namwata et al., 2010. Table 5.1 shows that the bid values had no significant effect on village savings and loans (VSLs).The disaggregated models shows that acceptance of the second bid price reduced the probability of adopting VSL by 18.6% in Dowa (Table 5.2). This could be the case because the insurance would act as a substitute to the use of VSLs in coping to weather and climate shocks as would both require funds to finance them.

Effects of fish sales on choice of VSLs
Fish sales of the farmers influenced the choice of VSL as a coping mechanism to weather and climate shocks. In the overall model, fish sales affected choice of VSL at the 10% level (Table 5.1). A one percent increase in fish sales reduced the probability of choosing VSL by 12.6%. The observed results are in line with the fact that luck of enough income is a key driver for joining village savings and loans (McIntosh et al., 2013). Furthermore, the increased fish revenue capitates the farmers to concentrate much on the enterprise than other sources of income. The disaggregated models did not show any significant influence of income on the choice of village savings and loans.

Effects of the household size on choice of VSLs
In the overall model, household size significantly influenced the choice of VSL at the 5% level of significance (Table 5.1). A one member increase in the household affected choice of VSL by 8.1%. The observed results are due to the fact that increased members would require extra income for survival which VSL could provide. The disaggregated models show that household size had a significant influence at the 1% level in Dowa District. In Dowa, a one member increase in the household size increased the probability of choosing VSL by 29.9%.

Effects of value of worst damage on choice of VSL
Overall, Table 5.1 shows that value of the worst damage the farmer ever experienced significantly affected choice of VSL at the 1% level of significance. A percentage increase in the value of previous damages on fish production reduced the probability of adopting VSL by 17.5%. In Dowa, Table 5.2 shows that a one percentage increase in value of worst damages reduced the likelihood of joining VSL by 19.4% and in Mchinji, a one percent increase in the value of the worst damage reduced the likelihood of joining VSL by 20.4%. The findings are in line with Trang (2013);Mulwa et.al (2017);Ghazanfar et.al (2015) who noted that the value of damage a farmer ever experienced may determine the range of future choices of coping strategies in the future.

Other Income Generating Activities (OIGAs) 4.5.2.3.1 Effects of the household size on choice of OIGAs
Household size influenced the choice of OIGAs at the 1% level of significance. Overall, a one member increase in the household size raised the probability of adopting OIGAs by 10%. This observation may be due to that household size is crucial when it comes to household activities (Hassen et.al. 2012). The disaggregated models shows a significant effect of household size on the choice of OIGAs in Dowa District at the 5% level. A member increase in the household size increases the probability of adopting OIGAs by 13.5%.

Effects of operational status on choice of OIGAs
In the overall model, operational status significantly influenced the choice of OIGAs at the 10% level. Being a fulltime fish farmer reduced the probability of choosing OIGA by 29.9% (Table 5.1). In Dowa, operation status negatively influenced the choice of OIGAs at the 5% level (Table 5.2). Being a fulltime farmer reduced the probability of choosing OIGAs by 54.7% than being a part-time farmer. These results can be attributed to the fact that a full time fish farmer devotes a lot of time in the management of fish and has little time available for other activities (Iheke & Agodike, 2016).

Effects of gender of the household head on choice of OIGAs
Gender significantly affected the choice of OIGA as a coping strategy to weather and climate shocks. In the overall model, gender significantly influenced the choice of OIGA at the 10% level (Table 5.1). Being male raised the probability of choosing OIGA by 38.5%. The observed results conform to the fact that women are resource constrained. Thus strategies that require initial/ upfront capital are likely to side line more women than men (Abdullah et al., 2014: Namwata et al., 2010. Across districts, gender had a significant positive influence on the choice of OIGA Mchinji District at the 1% level (Tables 5.3). Being male raised the probability of choosing OIGA by 126.1% (Table 5.3).

Effects of shock experience on choice of OIGA
In the overall model, Table 5.1 shows that shock experience by the fish farmers influenced the choice of OIGA at the 10% level of significance. Past shock experience increased the probability of choosing OIGA by 1%. This observation can be attributed to the fact farmers who experienced some shocks before are capacitated to prepare for eventualities and may diversify their income sources through say, OIGA (Ghazanfar et.al, 2015). Across districts, the disaggregated models in Tables 5.2 and 5.3 show that past shock experience only significantly explained the choice of OIGA in Mchinji District at the 10 % level of significance. Shock experience increased the probability of choosing OIGA as a coping strategy among fish farmers by 82.1%. 4.5.2.3.5 Effects of farm coping strategies on choice of OIGA Generally, the availability of other farm coping strategies than RII, VSL and OIGA significantly affected the choice of OIGA at the 1% level (Table 5.1). Availability of other coping strategies on the farm increases the probability of choosing OIGA by 3.9%. The observed result may be attributed to the fact that other strategies may need some start-up capital of which OIGAs like small-scale businesses can provide. The disaggregated models shows significant effect of farm coping strategies in both Dowa and Mchinji at the 1% level of significance. In Dowa, availability of other coping strategies raised the probability of choosing OIGA by 2.7%. In Mchinji, availability of other coping strategies raised the probability of choosing OIGA by 6.1%.

Effects of feeds expenses on choice of OIGA
Generally, feed expenses influenced the choice of OIGA at the 5% level of significance in both Dowa and Mchinji Districts (Table 5.1). A percentage increase in the expenses on feeds reduces the probability of choosing OIGA by 19.3%. This may be due to increased pressure on the money resources the farmer may be endowed with (McIntosh et.al, 2010). In Mchinji, feed expenses influenced the choice of OIGA at the 5% level of significance. A percentage increase in the cost of feeds reduced the probability of choosing OIGA by 24.3%. No significant effect was observed in Dowa. Table 4.17 shows a significant and negative relationship at the 1% level among the three dependent variables which are also the coping strategies namely: village savings and loans (VSLs), other income generating activities (OIGAs) and insurance (RII). This implies that these coping strategies can be used as substitutes in order for fish farmers to deal with the weather and climate shocks they face. Table 5.4 shows a significant and negative relationship at the 1% level among the three dependent variables which are also the coping strategies namely: village savings and loans (VSLs), other income generating activities (OIGAs) and insurance (RII). This implies that these coping strategies can be used as substitutes in order for fish farmers to deal with the weather and climate shocks they face.

Conclusions and policy recommendations
The multivariate probit results indicated that education level, operating status, years of farming experience, age of the farmer, shock experience, marital status and bid price are significant drivers of the adoption of RII. Bid price, fish sales, household size and value of worst damage were very crucial in the adoption of VSLs in both districts. Finally, household size, operating status, gender of the household head, shock experience, availability of other farm coping strategies and feeds expenses were key in the adoption of OIGAs. Our study therefore concludes that years of schooling, bid price, operational status, gender, cost of feeds, farm coping strategies, fish sales, marital status, shock experience, age of the farmer, years of farming experience and value of worst damages are important drivers of choice of coping strategies against weather and climate shocks. We recommend promotion of adult literacy and civic education on weather and climate related issues which may also be part of enhancing extension services. The policy and program intervention should consider training lead farmers who may in turn train fellow farmers on how to deal with weather and climate related issues. These trainings should be extended to new fish farmers to enhance their preparedness in handling weather and climate shocks.