Testing an Ethnographic Decision Tree Model on a National Sample: Recycling Beverage Cans
Ryan, Gery W., Bernard, H. Russell, Human Organization
We report here on an ethnographic model of a decision that Americans make regularly: to recycle beverage cans or not. The model was derived from 21 ethnographic interviews and 70 structured interviews in Florida and North Dakota. Ethnographic decision models are not new, but we show here that these models can be tested for both internal and external validity. We test internal validity by comparing the model's predictions systematically to what people say about their own behavior. We test external validity by comparing the predictions of the ethnographic model to those of a representative sample of 386 people across the United States. The original model accounts for about 90% of the reported behaviors, while the national model predicts about 85% of the reported behaviors.
Key words: decision modeling, recycling behavior, ethnographic methods
We report here on methods to test the internal and external validity of ethnographic decision models (EDMs). EDMs are qualitative, causal analyses that predict real, episodic behaviors, rather than-as does so much social research-the intent to behave in a certain way. EDMs can be displayed as decision trees (e.g., C. Gladwin 1989), as decision tables (Mathews and Hill 1990; Young and Garro 1994), or as sets of rules in the form of IP-THEN statements. For example, Ryan and Martinez ( 1996) modeled the decision of mothers in rural Mexico to take their children to a doctor in response to an episode of childhood diarrhea. One of the rules in the model was: IF there is blood in the stool, OR IF the episode lasts more than eight days, THEN take the child to the doctor.
Typically, EDMs predict at least 80% of the behavior under study. Such effective models are easiest to build for questions about behaviors that can be answered yes or no, like "Did you buy a new computer in the last 30 days?" or "Did you go to Lagos any time during the past year?" However, researchers have used EDMs to understand more complex behavioral outcomes, like the price that people place on products (H. Gladwin 1970; Plattner 1984; Quinn 1978); the choice by farmers to plant this or that crop on their land (C. Gladwin 1976, 1989; Barlett 1977); where fishermen choose to hunt for fish (Gatewood 1983); and the allocation of tasks in households (Mukhopadhyay 1984).
Medical social scientists have long used EDMs to understand the choice, by lay people, of treatments for various illnesses (Hill 1998; Mathews and Hill 1990: Montbriand 1994; Ryan and Martinez 1996; Weller et al. 1997; Young and Cairo 1994). Breslin et al. (2000) applied the EDM method to referrals for outpatient treatment by clinicians of drug-abuse patients; Bauer and Wright ( 1996) modeled the decision by Navajo mothers to breast feed or use formula; Johnson and Williams ( 1993) modeled decisions by IV drug users in Houston to take the risk of sharing needles; and Beck (2000) used the method to model the decision by psychologists in British Columbia to report suspected cases of child abuse to the authorities.
EDMs are built from interviews with a relatively small number of people (20-60) and are usually tested on a similarly small and local sample. C. Gladwin et al. (2001), however, tested a model for the decision to evacuate in a hurricane (Andrew in 1992, Erin in 1995) on 954 respondents in South Florida, and H. Gladwin and Murtaugh ( 1984) tested a model for car buying on 114 cases selected from the National Transportation Survey of 1978.
In what follows, we explain ethnographic decision modeling in detail; derive and test a model for recycling beverage cans; and test the results of the ethnographic model for internal and external validity. It is well known that the single best predictor of beverage can recycling is the presence of a recycling bin at the moment the decision has to be made (see for example, Austin et al. 1993; Larson et al. 1995; Ludwig et al. 1998). We chose deliberately to model a decision with a well-known predictor in order to test the efficacy of our model, both at the local and at the national level. …