Elements of Research Design
Good research design consists of an explicit, logical plan for connecting data and theory. In all types of research – participant observation, surveys, experiments – the major components of this plan are the same:
Formulating research questions (and hypotheses, if appropriate)
Selecting a research site where the research questions can be addressed
Developing a sampling strategy for selecting observations required to answer the research questions or test hypotheses
Choosing methods to collect data needed to answer research questions
Creating a plan for managing, documenting, and archiving data
Selecting methods for analyzing data to answer specific research questions and test hypotheses
This list makes clear how research questions ideally permeate all major components of research design – sampling, data collection, and data analysis. Well-designed studies (and grant proposals) make these links explicit.
Medical anthropologists and others have offered practical advice to enhance the links between different elements of research design. LeCompte and Schensul (2010, p. 188) recommend that researchers use a “data collection matrix” to fit questions to methods. Their matrix involves listing what researchers want to know, what types of data they will need to find out, where and from whom they can find such data, and how they will collect it.
Carey and Gelaude (2008, p. 237) emphasize the logic of design in the development of interview materials. They recommend creating an “intents list” to make explicit the rationale for each question in an interview. In a study of HIV risk, Carey’s team duplicated the semistructured interview guide and, below each question, added a one-line description of why the question was being asked. They also created a spreadsheet with three columns: (1) project objectives, (2) specific interview questions, and (3) citations related to the rationale or design of each question. These tools ensured that the design of interview materials was tightly integrated with the research questions and also proved useful in interviewer training and data monitoring. Powis (2017) described his use of this approach for ethnographic research on the experiences of expectant fathers in Dakar, Senegal (Powis 2020). Carey and Gelaude (2008, pp. 244–252) also recommend applying the logic of an intents list to data management and analysis plans before the project begins.
Having an explicit, logical plan to connect data and theory enhances the validity of all types of research. DeWalt and DeWalt (2011) discuss how to design research with participant observation. Spradley’s (1979, 1980) Developmental Research Sequence is a model for designing ethnographic research with progressively more focused methods of data collection and analysis. Bernard (2018) shows how the logic of experimental design helps researchers avoid common threats to validity – even if you never run an experiment. Johnson (1998) discusses research design in systematic anthropological research, and LeCompte and Schensul (2010) show how the logic of research design cuts across specific methods of data collection and analysis. Each of these sources provides many examples from research in medical anthropology.
Basic Research Designs
True (1996) noted that research design plays different roles in the training of medical anthropologists and epidemiologists. In epidemiology and allied fields, researchers recognize a set of basic research designs that have different strengths and weaknesses for answering particular types of questions. Medical anthropologists are generally not trained to think in these terms. However, familiarity with the range of research designs commonly used in other social and health sciences can enhance the validity of research in medical anthropology and increase our impact on neighboring disciplines and policymakers.
Figure 4.4 illustrates a typology of basic research designs. The first major distinction is between observational and experimental designs.
Figure 4.4 Basic research design options in medical anthropology and neighboring disciplines.
Experimental Designs Experimental designs are distinguished by two features: random allocation and manipulation of the key causal variables. In classic experiments, researchers randomly assign participants to either an intervention or control group and measure one or more dependent (outcome) variables in both groups. Participants in the intervention group are then exposed to a treatment designed to test the causal effect of an independent (explanatory) variable, and both groups are measured again on the dependent variable. Random allocation, when done well, makes groups comparable with respect to unmeasured variables, such that whatever differences emerge between groups after the intervention are likely to reflect the true causal effect of the intervention.
Experimental designs are not common in medical anthropology, but there are successful examples. Shain et al. (1999) combined ethnography and a randomized trial to test the effect of culture- and gender-specific interventions to prevent sexually transmitted infections in African-American and Mexican-American women in San Antonio, TX. They first collected ethnographic data (observations, 25 focus-group discussions, 102 in-depth interviews) on the cultural context of sexual behavior, perceptions of risk, and motivations for behavior change. They used this information to design culturally appropriate messages about recognizing risk, committing to change, and communicating about sex. Then, 424 Mexican-American and 193 African-American women were randomly assigned to receive either the culturally appropriate messages (intervention group) or standard counseling (control group). One year later, Shain et al. found that women in the intervention group were 49% less likely to have a sexually transmitted infection than were the controls.
Even if you never intend to run a randomized trial, being familiar with their strengths and weaknesses maximizes impact across disciplines. Smith-Morris and colleagues (2014) describe a collaboration in which ethnography was built into a randomized clinical trial. Their experience highlighted the complementarity of approaches and suggested a need for “more clinical and trial-based applications of medical anthropology” (p. 157).
Observational Designs Most research in medical anthropology, as in other health-related social sciences, is observational. Observational studies lack the defining features of experiments – random assignment to comparison groups and control over independent variables – and so are not well suited to demonstrate causal effects. But they are preferable to experiments for exploratory questions and have other advantages in confirmatory research, including higher external validity (generalizability), greater feasibility, and often fewer ethical objections.
We can distinguish three broad classes of observational designs: cross-sectional, longitudinal, and case-control studies (Figure 4.4). These three types of studies are recognized as the basic design options in epidemiology and biomedical sciences, but they are also used to varying degrees in medical anthropology. Cross-sectional studies, in which data are collected during one point in time, are the most common type. Although data collection often lasts for months or years, the study is still considered cross-sectional if the data are taken to represent one point in time. Thus, even long-term ethnographic research is usually cross-sectional in design (Gravlee et al. 2009).
In longitudinal designs, the data pertain to two or more points in time. In the last 50 years, longitudinal studies that incorporate panel data – repeated measures from the same units of observation at different points in time – have become increasingly important across the social, medical, and public-health sciences, but they remain relatively rare in anthropology. The lack of panel studies in anthropology represents a mismatch between theory and method, because panel data are particularly suited to the study of continuity and change – central areas of anthropological inquiry (Gravlee et al. 2009). For example, there is long-standing debate about the consequences of market integration and culture change for the health of Indigenous peoples. Most relevant studies, however, use cross-sectional designs,