Chapter 3: Research Design, Methods and Study Site
3.3 Research Methods
3.3.2 Data Collection
The data used for this study were collected between October, 2010 and June, 2011 from Kibungo, Lanzi, Nyingwa and Dimilo villages in the case study area in three phases. A multi-method strategy as shown above was used. A range of methods and techniques were applied in the collection of data to increase their validity and reliability for broader understanding of the research question (Johnson and Onwuegbuzie, 2004; Downward and Mearman, 2007; Gray, 2009). The use of multiple methods has been highlighted by different researchers because it is felt to improve consistency across methods through the process of
"triangulation", by looking at the problem from diverse viewpoints (Olsen, 2003; Olsen, 2004; Downward and Mearman, 2007; Denzin, 2009; Bryman, 2012). Nichols (1991) argues that even when a survey is useful, it is often best used together with other complementary methodological tools. This strategy employs qualitative and quantitative methods in three phases that are characterised by a literature review, household surveys, focus group discussions, key informant interviews and observations as detailed in the next sections.
3.3.3 Research Phases
The first phase of the research involved the use of a qualitative research design to support the inductive direction of the thesis (Shuttleworth, 2008). This phase involved a review of literature, observation of farms, focus group discussion, semi-structured key informant interviews (Bernard, 2006; Babbie, 2008) with CARE-WWF Tanzania officers administering the program, and a focus group discussion with 8 participating and non-participating farmers between October and November 2010. Material collection in this phase sought to generate grounded knowledge on elements such as targeting, eligibility rules, payments, and land change management requirements of the EPWS programme, its institutional context of implementation and the farmers’ reasons for participating and not participating.
In the second phase, household surveys using a structured questionnaire were administered to EPWS program participants and non-participants. The questionnaire was tested with a small number (N=7) of households and in one
focus group meeting with village leaders in November, 2010. The main fieldwork was then conducted from March to May 2011 in four villages of Kibungo Juu ward namely Kibungo, Lanzi, Nyingwa and Dimilo. The number of households surveyed was 233 with 116 program participants and 117 non-participants. Household heads were selected from each village using stratified random sampling generated through the wealth ranking technique (see section 3.3.5), through which households were categorised into poor, middle and rich to ensure the representativeness of the sample (Chambers, 1994; White and Pettit, 2004). In the third phase, the quantitative findings arising from phase two were explored further with 32 semi-structured key informant interviews and 16 focus group discussions. The key informant interviews were conducted with CARE Tanzania EPWS program officers, village leaders, 8 representatives from EPWS groups in each program village and 8 EPWS participating and 8 non-participating households. Focus group discussions were used to capture divergent viewpoints such as the determinants of participation decisions (Hopkins, 2007). Following guidance by Hopkins (2007) and Creswell and Plano Clark (2007), participants with experience and knowledge of the phenomenon under investigation were selected. Representatives of local organizations and participating and non- participating households were selected for focus group discussions with separate focus group discussions conducted with EPWS participating and non- participating households in each program village. The size of focus group discussions was between 8-10 people. The key informant interviews and focus group discussions were conducted in ‘Swahili’, audio recorded and then transcribed into English.
3.3.4 Household Survey
Household surveys were conducted in the second phase of the research using a structured questionnaire administered to EPWS program participants and non- participants. This was informed by a quasi-experimental design (see section 3.4) to enable the estimation of the counterfactual effect of the program from the control and treatment group (Baker, 2000; Ferraro and Pattanayak, 2006; Dunning and Hyde, 2008). The treatment group was defined as households who had implemented SLM practices under EPWS program and whose farms had been measured for payment or had already received payments, while the control group was composed of households who had not registered their names with EPWS program. The treatment group comprised 116 household heads and the control group had 117 household heads. These households were selected from a stratified random sample generated through a participatory wealth ranking exercise (see section 3.3.5).
Household surveys were conducted during the second phase of the research process from March to May 2011 in four villages. A mixture of both closed semi- quantitative and open qualitative questions (Downward and Mearman, 2007; Gray, 2009) were used to investigate the determinants of landholders' decision to participate in the EPWS program and livelihood and environmental impacts of the program (types of data collected for each research objective are shown in subsequent chapters). A household was described as people who live and sleep in the same compound, including absentees (Randall et al., 2011). The heads of households were selected for survey using stratified random sampling technique
generated through wealth ranking technique to ensure representativeness (Chambers, 1994; White and Pettit, 2004).
3.3.5 Wealth Ranking
According to Van Campenhout (2010), participatory wealth ranking can be a best way of generating poverty indicators and profiles that incorporate local perspectives on poverty. To perform wealth ranking in this study, Chambers’ (1994) steps and advice for maintaining the strength of the wealth ranking exercise sensitive to local circumstances and expertise were followed. In the first step, key informants who had lived longest in the community and who knew most about the livelihoods of households in the area were selected following Mukherjee (1998) and Stocking and Murnaghan (2001). Village and sub-village leaders were selected first and, with their help, two other representatives were selected to form a wealth ranking group of 6 – 8 people.
Lists of names of the household heads from each village were collected from village records in the second step. The lists were verified with key informants to harmonise shared similar names, adding new households and deleting deceased individuals. Through a collaborative effort between the researcher and key informants, a simple fivefold wealth classification approach was adopted based on the five different types of livelihood capital assets (Table 3.1) identified by Carney(1998) and Ellis (2000).
Table 3.1. The five livelihood capital assets used to characterise wealth in the wealth ranking exercise
Type of capital asset Description
Natural capital Land, water and biological resources that are utilised by people e.g. forest resources
Physical capital Assets created by economic production processes e.g. buildings, irrigation canals and roads
Human capital The labour available to the household including education, skills, and health
Financial capital Stocks of money to which the household has access e.g. saving and access to credit in the form of loans Social capital Community and wider social ties on which households
can rely Source: Ellis (2000: 16)
Based on the key informants’ translation of these criteria of poverty or wealth (Mukherjee, 1993; Adams et al. 1997), a number of proxy indicators for each type of capital asset were identified by the key informants enabling the classification of households into three groups according to wealth level- rich, middle and poor.
Wealth ranking derived from participatory methods is a useful tool in deriving poverty indicators and profile poverty or wealth of one particular village (Southgate et al., 2010; Zhang and Pagiola, 2011). However, when it comes to comparing villages, participatory wealth ranking does not allow comparison across villages because criteria used in one village may vary in another village, resulting in non-comparable distributions (Place et al., 2007; Southgate et al., 2010). Due to this limitation, this study used additional wealth grouping method in addition to participatory wealth ranking used to classify households into three groups – rich, middle and poor to facilitate sampling of the household survey. The other method involved calculation of an asset level for each household, based on monetary values. This method provided the basis for the division of the entire
sample into asset quartiles. The total asset value for each household was based on the type and number of assets owned and valued at the average price across the full sample (see section 3.5.10.5).
3.3.6 Sampling
As the level of analysis in this study was the household, a household survey was carried out in the four villages of Kibungo, Nyingwa, Lanzi and Dimilo. However, defining the household as a unit of analysis is challenging due to the complexity and variability of the arrangements that people make to facilitate provision of food and/or other essentials for living. In the study villages and Morogoro Region in general, the household unit is defined in terms of rights to land, with every village household being allocated its own residential plot and farmland by the village government (Lyamuya et al., 1994). New households are formed as adult children move out of their parents house to marry and start their own families, with their own areas of farmland (Lyamuya et al., 1994).
Informed by Baker (2000), participatory wealth ranking was followed by the random sampling exercise to select a representative sample of 233 household heads for interviews from EPWS participating (treatment group) and non- participating (the control group) households in the study villages (see table 3.2). The treatment group was defined as those households who were enrolled in the EPWS program while the control group were those who were not enrolled in the EPWS program but possessing the same characteristics as the former. The sample sizes represented between 16 and 21% of the village households (Table
3.2). Stratified sampling was used to ensure representation of all wealth groups and all sub-villages as well as both gender groups. For each household in the sample, one respondent was interviewed; the person being either the head of the household or a spouse of the head of household.The gender distribution of the total sample is 65.1% male and 34.9% female respondents, which allowed for gender disaggregated analysis. Table 3.2 shows the village study sites and sample sizes, while wealth distribution of household survey sample is presented in Table 3.3.
Table 3.2. Village study sites and sample sizes
Village name Number of
households Households sampled Percent of population 1.Dimilo 227 48 21 2.Lanzi 275 55 20 3.Nyingwa 434 70 16 4.Kibungo 279 60 21 Total 1215 233 19
Table 3.3. Wealth distribution of household survey sample
Village Poor Middle Rich Total
Kibungo # of observation 19 35 6 60 % 31.7 58.3 10.0 100.0 Nyingwa # of observation 18 45 7 70 % 25.7 64.3 10.0 100.0 Dimilo # of observation 22 20 6 48 % 45.8 41.7 12.5 100.0 Lanzi # of observation 14 29 12 55 % 25.5 52.7 21.8 100.0 Sample # of observation 72 129 31 233 % 31.0 55.6 13.4 100.0
In addition to participatory wealth ranking, the classification based on the total value of assets of the household was performed to generate grouping of households that allowed inter-village comparison. As such, the measurement of
asset values helped to overcome limitations that could be experienced from participatory wealth ranking method (Zhang and Pagiola, 2011). Table 3.4 shows the comparison of the results of the participatory wealth ranking and the assert quartile wealth ranking. Both types of wealth grouping were highly significantly positively correlated (Spearman: r=0.250, p<0.00; Kendall’s tau_b: r=0.273, p<0.00).
Table 3.4. Comparison PRA wealth grouping and asset quartile groups
1 (poorest) 2 3 4 (richest) Total (n) Wealth group by PRA Poor 35 15 11 12 72 Middle 19 34 44 32 129 High 5 9 3 14 31 Total 59 58 58 58 233
Note: The total asset value (ranges) of the asset quartile groups in TSH are as follows: 1 < 651,000; 2 = 651,001 - 908,000; 3 = 908,001 – 1,412,000; 4 > 1,412,001
3.3.7 Interview Procedure
Using a semi-structured questionnaire, the interviews were administered directly by the research team, without interference from local officials and the use of face to face interview methods ensured a high level of completeness and accuracy of the data. Meetings with the interviewer every day prior and after household visits ensured consistency and the recording of additional notes and observations. The completed questionnaires were field checked in the evenings to minimize errors and missing data items. Most interviews were carried out outside of people’s houses, a setting at which respondents would feel familiar and not be inhibited in their responses. The setting furthermore allowed the enumerators to assess the characteristics of housing through personal observation without directly asking and to note down any other personal observations about the living conditions of the particular household.