Study areas and sample size The required data for this study was derived from a cross-sectional dataset, collected through farm household survey based on their agricultural and different nonfarm activities intensity. Multi-stage sampling procedure was used to identify the 153 sample households from four villages. The number of households is not the same for each village. Sample households were categorized into farm households (income source is only agricultural activities), and part-time farming households (income source is both agriculture and non-farm activities) consisting of 59 and 94 households respectively. Data were collected through household survey, key informant interviews (KII) and focus group discussion (FGD) with farm households during July to November 2014. Analytical techniques The term welfare indicates a broader area and it is explained in different ways in different studies. In general, poverty alleviation is mostly used as economic wellbeing indicator (Reardon et al., 1992; Block & Webb, 2001; Ravallion & Datt, 2002; Holden et al., 2004). The term welfare indicates a broader area and it is explained in different ways in different studies. In general, poverty alleviation is mostly used as welfare indicator (Ravallion & Datt, 2002; Holden et al., 2004). Some studies use food consumption or calorie intake as a welfare indicator (Musyoka et al., 2014; Seng, 2015). However, only food consumption cannot fully indicate the whole standard of living of a household. Access of the rural people to basic services such as electricity, water, sewage facilities along with fundamental needs (food intake, consumption of cloth, housing, medicine and education services) are viewed as a reflection of the household’s welfare standing (Jesko & Lanjouw, 2006). The expenditure incurred on these various needs is vital to enhance the welfare status of households (Ismail & Bakar, 2012; Scharf & Rahut, 2014). Household’s insecurity, uncertainty and discrimination to get these facilities and services reducing their income and consumption level, as well as their welfare (Brück, 2004). Though savings is another part of the income provides household security, expected to expense in future for improving living standard. Therefore, it is generally assumed that more expenditure on the daily necessaries indicates more welfare situation. Household’s per capita total expenditure comprising of expense on food, clothing, education, health, transport, fuel and festival is used as household-level indicator of welfare in this study. Household’s total consumption expenditure is considered here instead of household income, as many empirical works in different countries stated that it can be measured with more accuracy than using income. It is generally assumed that, poor people expense less on consumption of food and other non-food goods and services comparing to rich people. Livelihood strategy is likely to be endogenous in rural household welfare estimation, as household adopts different strategy mainly to improve their livelihood situation, ultimately welfare. Participations in different income generating sources have effect on rural household welfare, and vice versa. For addressing this simultaneous causality between different livelihood strategies and welfare indicators, the Two Stage Least Square (2SLS) methods with instrumental variable is applied to estimate impact of the strategies on household welfare. Following model is used as a first stage to estimate this impact Description of the variables used in the model As the concern of this study is to analyze the impact of different livelihood strategy on welfare, welfare indicator namely household per capita expenditure is treated as the outcome variable. Natural logarithm is taken of the dependent variable to meet the linearity assumption. Descriptions of selected explanatory variables are briefly presented in Table 2. Selection of instrumental variables Following Imbens & Angrist (1994), Abadie (2003), Awotide et al. (2012), farmer’s access to improved rice varieties seed and fertilizer, and local market distance have been selected as instruments for households those involved in only agricultural activities. The first instrumental variable denotes a binary variable equal to 1 if a household has easy access to improved seed and fertilizer and 0 otherwise. Another instrument, distance from local market expected to influence agricultural income but not the outcome variable. But it does not have any direct impact on agricultural household’s welfare indicators. For analyzing the impact of income from a combination of agriculture and wage or self-employment on household welfare, following two instruments are used: (1) Distance from district level urban centre and (2) Share of non-farm employment at district level. Scharf & Rahut (2014) used proportion of the working population engaged in non-farm sector at village level. Due to unavailability of data at village level, this study followed Kilic et al. (2009), where he used this data at district level. Bangladesh agricultural census of 2008 is used to find out the share of non-farm employment in district level. In addition, for estimating welfare impact of the income, generated through a combination of agriculture and migration based nonfarm activities, two instruments are also used, namely family migration network and district level migration network.