CHAPTER FIVE: METHODS
5.4. Data Processing and Analysis
5.4.3. Data Processing for the Semi-Ethnographic Interview Results
Unlike the interviews based on the Theory of Planned Behaviour (TPB) and the Personal Construct Theory (PCT), the semi-ethnographic interviews did not have minimum requirements that should be met by the respondents in answering the questions. The semi-ethnographic interviews allowed the respondents to use their own terms in explaining their decisions regarding the improved paddy-prawn system (“pandu”). A list of leading questions, however, was still used in order to help extract the respondents’ main decision criteria (see Section 5 of Appendix B). Hence, all responses from the semi-ethnographic interviews were eligible for the analysis.
LeCompte (1999, pp. 82-83) summarized several approaches that might be applicable for analyzing the results from an ethnographic interview. The consensus appears to include the following steps:
a) code and classify the responses into several key terms;
b) outline the association among a group of key terms in the form of patterns or paths;
c) justify the links between the key terms and the paths through some assumptions relevant to the case studied; and
d) explain the outcomes and the implications to the case studied and general application.
With some modifications, the above steps were applied in this study. The analysis began with summarizing and coding the responses to form a database of key terms that may represent the respondents’ decision criteria concerning “pandu”. Some of the responses were also quantified; and the results were used in the TPB analysis (see Section 5.4.1 of this chapter).
Since each respondent appeared to elicit many terms, indicating a complex set of decision criteria, it was necessary to re-classify the key terms to produce the most relevant decision criteria for each respondent (Gladwin, 1989a, pp. 28-33). This was done by identifying the general themes of a group of decision criteria, and by
choosing some key terms based on their specific reference in the respondents’
explanations, indicating a particular decision option and/or a new direction of decision making path. For example, one participating farmer in Sugihwaras mentioned:
“I asked the farmer group, and re-thought again before deciding to apply pandu”
Another participating farmer in the same village also explained:
“I was interested in Pandu because the development of prawn and paddy was better and tidy, and because of the financial incentives…”
The first farmer suggested a decision criterion related to his effort in seeking more information (and assurance) despite being involved in the dissemination process. The second farmer indicated a financial motive underlying his interest on “pandu”. Both terms might apply for the decision option of “try, or not try, pandu”.
Other examples are:
Farmer 1: “I only allocated pandu for the smaller ricefield-pond because I did not have enough money.”
Farmer 2: “I tried (pandu) in a smaller plot because I was worried the trial would be unsuccessful…”
These farmers participated in the dissemination process, and did not directly mention why they tried “pandu”. Instead, they indicated that they directly applied “pandu”, but faced constraints, i.e. lack of capital and observable result. These constraints could be classified as decision criteria which persisted despite the provision of financial incentives and demonstration plots.
The above examples are only partial and may indicate only one decision option. A more complete explanation usually indicates several key but different phases of decision making. This is shown by the explanation from one farmer in Rejosari:
“I was interested in applying pandu after hearing from the explanation (dissemination) that the results would be doubled. I could easily learn about, and apply, pandu because technical assistance was provided.”
“I did not use jajar legowo (the recommended paddy planting distance) because the farm labourers did not want to… although using planting distance, pests and diseases could be controlled easier; paddy clumps and yield increase….” “I do not apply pandu again because water supply is limited, and I did not know that water is plenty this year.”
From this particular farmer, at least four decision factors for three decision options could be identified. The yield from “pandu” and the technical assistance were the first two factors determining the farmer’s intention to try “pandu” (first decision option). However, this farmer appeared to be a partial adopter, as the unfamiliarity of “jajar legowo” among local farm labourers had limited applying all the “pandu”
recommendations (second decision option). The farmer also added that the limited water availability was the reason for quitting “pandu” (third decision option).
All decision criteria and the relevant decision options were then classified and summarized. This process resulted in nine groups of decision criteria and three possible decision options. The nine groups of decision criteria represented:
a) the respondents’ belief in the potential of “pandu”;
b) the respondents’ efforts to seek information, either actively or passively; c) the respondents’ level of experience and reliance on their own practices;
d) the financial aspects related to “pandu”, which included capital, land possession, costs, incentives, yield (including prawn survival rate), prices/marketing
aspects, and “pandu’s” contribution to the fulfilment of household needs; e) the technical aspects related to “pandu”; this included the level of difficulty,
workload, time availability to work on the farm, the provision of technical assistance, knowledge and skills, age/strength, and the local farm labour system; f) suitability of “pandu” with the local agro-ecosystem; this included the suitability
of paddy being intercropped with prawn, water quality and availability, soil conditions, level of pest and disease infestation, the effects of different cropping patterns among farmers, the feasibility of ricefields for “pandu”, and commodity preference;
g) the level of “pandu” application: partial or full adoption;
h) the involvement and contribution of others in the respondents’ decision making process; and
i) the respondents’ general evaluation on “pandu”.
These nine groups were then used for developing several decision trees. The
construction of the decision trees followed the procedures set out by Gladwin (1989a, pp. 21-45), with some modifications. They are summarized as follows:
a) define the decision, and set the relevant assumptions;
b) apply the decision criteria (key terms) for each domain branch leading to different decision choices, followed by the next (underneath) branch(es). Each decision criterion serves as a decision constraint. Figure 5.12 provides an example of a decision tree development, although the structure may be different for different decisions and groups of farmers;
c) check the branches and the decision criteria involved for a logical decision making flow, and whether more branches (more constraints) are required before the final decision; and
d) test the aggregate decision tree to confirm the generalizability by involving a different group of respondents.
Figure 5.12 A direct approach to building an aggregate decision tree (modified from Figure 2.9 & 2.10, pp. 43-44, in Gladwin, 1989a)
In this study, no individual decision tree was developed. Instead, a group decision tree model was formulated directly using the decision criteria mentioned by each respondent. This was deemed to be time saving considering the large sample size and differences in the respondents’ adoption process.
In each village, the respondents were divided into two groups: participating farmers and non-participating farmers. The grouping was important as these groups might have a different level of understanding about “pandu”. The participating farmers were involved in the “pandu” dissemination process; while the non-participating farmers might only rely on indirect observation and information passed from their colleagues. The participating farmers also received financial incentives and technical assistance.
yes
Decision Choices: Try “pandu” vs Do not try “pandu” <reason 1 true?> no <reason 3 true?> no <reason 2 true?> no
Do not try “pandu” yes
Try “pandu” no
<constraint 2 true?> no
Do not try “pandu”
Try “pandu”
<constraint 1 true?>
A few of the non-participating farmers who tried “pandu” during the introduction period might also receive technical assistance, but not the financial incentives.
The two groups might also experience a different adoption process. The participating farmers might have to make three consecutive decision options: (i) whether to try “pandu”, (ii) whether to apply all recommendations, and (iii) whether to continue adoption. The second decision might indicate that although the participating farmers had committed to apply “pandu”, they might face constraints causing them to have different degrees of adoption. These three decision options did not necessarily apply only for the introduction year, as they might also apply in the subsequent years depending upon the farmers’ situations.
For the non-participating farmers, their first decision choice included try, or not try, “pandu”. If they decided to try “pandu”, they had to decide whether to continue. Since most of the non-participating farmers did not receive complete information about “pandu”, they were assumed to rely on observations of others’ application of “pandu”. This increased the possibility of partial/modified application among these farmers if they had decided to try “pandu”. Hence, the “full/partial” decision option was excluded.
The recognition of consecutive decision options enables the analysis to look deeper into the respondents’ adoption process, which covers not only the adoption decision, but also the modification and continuation choices. This is believed to portray the respondents’ real decision making process better. In addition, the decision grouping has simplified the procedures for developing the aggregate decision tree models, which Gladwin (1989a) and Murray-Prior (1998) referred to as time consuming and complicated.
Using the two farmer groups, nine groups of decision criteria and some decision choices, 12 decision trees were developed. Each decision tree contained decision criteria that served as constraints in the decision path. Most decision criteria used
were the ones leading to a decision. Hence, each respondent had to go through every decision path and pass each criterion (constraint). Some respondents had to stop when facing a particular constraint. This indicated the factor that led the respondents to their final decisions. Other respondents continued passing through several
constraints and then stopped. This showed that more decision criteria were involved in the decision making process. Some decision criteria might also be applicable for more than one decision tree, but these criteria were carefully selected in order to assure their relevancy.
After all decisions were identified, they were compared to the respondents’ answers to the open-ended part of the TPB questionnaires (see Section 7, Appendix B), which contained questions directly asking the respondents about their “pandu” application. All matched responses were counted giving a percentage showing the level of correct representation of the respondents’ actual behaviour in the decision tree model. These procedures represented the internal validation for the decision tree.
Some decision trees were also validated using a different group of samples to determine the generalizability of the decision tree models. The test involved 30 farmers from Sidobinangun village, the neighbouring village of both Sugihwaras and Rejosari. The test did not use the decision trees; instead, it employed an interview with 183 questions based on the key terms mentioned by the respondents from Sugihwaras and Rejosari (see Appendix H). From the 183 questions, only responses similar to the nine groups of decision criteria (see page 152) were used. The test procedures required the relevant responses from the test group to go through each path in the decision tree. Some respondents stopped when facing a particular constraint, while others continued until they passed through all decision criteria and came to the end-point of a decision branch. The number of respondents in every decision end- point was calculated, and this was checked against their answer related to their interest in applying “pandu”. Every matched decision increased the decision tree’s predictability power. The rule of thumb for a good decision tree model is if it can predict around 80-90 percent of individual choices (from various applications, see Fairweather, 1996; Gladwin, 1989a; Jangu, 1993; Murray-Prior, 1998). The validated
decision tree models then can be used for explaining or predicting similar behaviour in different case studies.
The test, however, only involved decision trees that were related to the decision option of “try or not try pandu” because farmers from Sidobinangun had not been introduced to “pandu”. The other decision options (“partial or modify pandu”, and “continue or discontinue pandu”), thus, were not tested. Nevertheless, these decision trees could still enrich the description of the farmers’ decision making processes. All decision trees for the two farmer groups in Sugihwaras and Rejosari are presented in Chapter Eight.