Contents - Previous - Next
This is the old United Nations University website. Visit the new site at http://unu.edu
Experimental and naturalistic field research
Prediction, cause, and causality
Selection of research communities
Collaboration with community people
Sample size and population definition
The household as a primary research unit
The structure of variables: categories versus variations
Quantitative and qualitative data and the EMIC/ETIC issue
PERTTI J. PELTO
Department of Nutrition Sciences. University of Connecticut, Storrs, Connecticut, USA
Although field research is the primary mode of data gathering in nutritional anthropology, much of the literature on scientific methodology is based on the model of experimental research in laboratory settings. Therefore, it is important to consider some of the specific conditions in field research that call for modifications of the classic experimental paradigm. Much confusion has resulted from the tendency to consider field research as "basically the same" as experimental investigation, except, perhaps, for the added logistic complexities. In this paper I will examine some of the specific features of field research that may call for logical structuring quite different from that usually seen in experimental research.
The pursuit of field-based research in human populations involves a complicated trade-off. On the one hand, careful control and manipulation of key variables is nearly always impossible to attain, and, even if possible, would pose serious ethical problems. On the other hand, well-constructed field research produces data that approximate the actual living situations of real populations. One aspect of this realism is the ability to include observations on a wide range of relevant variables, including features not initially provided for in the research design. Such realism in the research process is especially important for applied research, which characterizes much of nutritional anthropology.
Researchers must be very aware, however, that the realism - the apparent validity of fieldbased data is purchased at considerable methodological cost. Moreover, the problems of designing logical and statistical controls are formidable. Because extraneous variables cannot be controlled through experimental manipulation, the field researcher must adopt a quite different strategy for maximizing the possibilities of inferring predictive relationships and causal pathways. These special features of field research also lead to a somewhat different logic concerning interpretations and generalizations from the data analysis. Before analysing in detail experimental, quasi-experimental, and naturalistic research strategies in field settings, I will discuss site selection briefly.
Communities as Field Sites
Field-work in nutritional anthropology usually focuses, at least to some degree, on the unique constellation of features in a particular community, region, or population. Typically a field research site is selected because one wishes to develop generalizations concerning a particular type of community - e.g. a rural agricultural community, an urban migrant settlement, or perhaps a particular ethnic population. Thus, the selection of the research site is intended to contribute to a particular category of case-study.
A distinction can be made between community-based research and other field investigations. Most anthropological research, whether dealing with food and nutrition or other topical areas, is community-based in the sense that researchers attempt to identify a more or less clearly delineated community of interacting, interdependent households for which certain community-level characteristics are seen as important for defining and interpreting the data. In such a study, it is generally assumed that key aspects of the local ecological and economic system need to be described because these affect food use and nutrition patterns.
In contrast to the use of a community of households in a research design, we occasionally find field studies in which samples of households or individual respondents are drawn from a population - e.g. a suburban area - with no attention to or interest in specific community features. An example would be a field study of health-food users in a region, in which the research focused entirely on the individuals and their households, perhaps cutting across several communities haphazardly (Kandell and Pelto, 1980).
We can generally distinguish quite sharply between two basic types of research designs in field settings. Some researchers try to develop field experiments, in which the attempt is made to approximate as nearly as possible certain essential features of experimental method. The primary defining feature is the manipulation of one or more independent variables through selection of (and sometimes intervention in) appropriately varying research populations. Since anthropologists seldom engage in direct intervention themselves, the most usual type of field experiment is one in which:
The comparison of the scores of the experimental population and the control sample on a dependent variable, such as an anthropometric measurement, is then used to establish the probable effect or outcome of the programme. The study of "Plan Chontalpa" by K. Dewey (1981) is an example of this type of research. She noted that ". . . the conclusions drawn from [the study] depend on a comparison of families which have been direct "beneficiaries" of the Plan and families which presumably have been less directly affected' (p. 24).
Some researchers have operationalized a quasi-experimental research design by selecting two or more communities that demonstrably vary with regard to some independent variable of special interest to the researcher (Campbell and Stanley. 1963). Selection of communities as "more traditional" and "more acculturated" is a common format for operationalizing research on effects of "modernization," for example. R. Baer (1984) compared anthropometric data (the dependent variable) in three communities with different degrees of "commercialization of agriculture" and "access to resources" (the independent variables). In a similar vein, selection of communities with two or three different ethnic groups has been a common practice for "manipulating" ethnicity as an independent variable.
Field research projects vary considerably in the extent to which there may be a conscious focus on one or more independent variables underlying the initial selection of field sites. Of course, it is possible to manipulate an independent variable among different subgroups such as classes, subneighbourhoods, or households in a single community. However, for the kinds of research of interest in nutritional anthropology this is not a usual design.
Kerlinger (1973) refers to non-experimental research with the non-commital label of field studies, which "are ex post facto inquiries aimed at discovering the relations and interactions among sociological, psychological. . . (and other) variables in real social structures" (p. 405). Ordinarily there is no manipulation of independent variables. Although some methodologists have taken pains to point out the basic similarities of underlying logic between the two approaches (Kerlinger, 1973), I feel it is more useful to emphasize the differences between experimental and naturalistic methods. The natural history approach in nutritional anthropology resembles certain kinds of epidemiological studies and is fundamentally concerned with a focus on a particular dependent variable or set of such variables.
The differences between experimental and naturalistic designs in field research begin to emerge clearly if we examine two series of typical research questions appropriate for each category, as shown in table 1.
There are important, interesting questions for research on both sides of the list. What has been ignored, however, are some fundamental differences in the logic-in-use (Kaplan, 1964; Pelto and Pelto, 1978) that come into play depending on whether the primary focus is on the independent or the dependent side of the research "equation." In the next section, some of the major limitations of the experimental research design in field settings will be examined.
Limitations of Experimental Designs in Field Settings
In its classic form, the experimental method involves two conditions: the experiment and the control (non-experiment). This structure provides a dichotomous independent variable presence or absence of the experiment. Outcomes or effects, such as improved nutrition, are postulated for the experiment. In one of the more common research designs, improved diet and nutritional status are expected to occur as a result of a particular food programme (e.g. supplemental feeding), so the basic hypothesis-testing has the usual classic form:
X (programme) (r) Y (improved diet/nutrition)
Table 1. Types of research questions in field research
|Experimental research design, focused on independent or "experimental" variable ·||Naturalistic design, usually focused on dependent variable(s)|
|Does a nutrition intervention
have the predicted effects?
|What are the main factors that
relate to (or
predict) the range of variation in nutritional/
health status in a given community or region?
|What are the nutritional effects of an economic development programme (e.g. agricultural programme)?||How do economic factors and ideational/ cultural features interact in affecting peoples participation m a food programme?|
|What are the nutritional and health effects of a settlement programme?||What are the main factors that affect peoples' choices among indigenous and commercially available foods, health care, etc.?|
|What are the nutritional and health effects of urban migration?|
|To what extent are the differences between two ethnic groups (e.g. in nutritional status) due to their cultural belief systems?||When two or more ethnic groups are intermingled in the same community, can we identify the differences in nutritional status and health, and the main factors that account for the differences?|
|What are the effects of the introduction of a new cash crop in a region, where some people adopt the new crop and others do not?||When a new cash crop is introduced in a region, what are the main differences between the acceptor and non-acceptors of the innovation ?|
The requirement of ceteribus paribus (other things being equal) is crucial in this experimental paradigm because, in the multifaceted world of human food use and nutrition, there are always a variety of competing causal explanations for a particular outcome. Those other variables must be controlled, either through careful site selection or, more usually, with statistical "partialling out" of their effects. This requirement of "other things being equal" is extremely difficult to achieve in real-world field studies, however. Thus, the experimental design in field-work is always beset with two major challenges:
Since there is currently a great deal of research interest in the effects of urban migration and other features of the process of modernization on food patterns and nutritional status, it is useful to note a number of rather serious weaknesses in the basic form of the experimental research design when used in such studies.
First, "urban migration" cannot logically "cause" people to eat better and improve their nutritional status. Instead, the very general condition of migration is believed to affect certain results through quite complicated "intervening variables." Indeed, there appear to be many cases in the world in which urbanward migration has had the opposite effect. Hence, the general proposition must be restated as: "Urban migration produces dietary improvement provided certain conditions (a, b, c, and d) are not present, and provided one or more factors (f, g, h, or i) are present in the environment of the urban migrants.
Second, in practice it is extremely difficult to identify a suitable control group for the urban migrants, even though a control group is generally considered essential for making the claim that it was actually X (migration) that caused Y (improved diet). The most likely "control," of course, is the non-migrant population in the home (rural) community. However, migrants are always self-selected non-randomly from the original population, so there are always other systematic differences between the migrants and any particular group of non-migrants. Furthermore, in the period since migration, both the migrants and the non-migrants may have experienced economic changes, new cultural messages, political perturbations, and a variety of other "contaminating" effects. The observed differences between the migrants and the "controls" may well be due to such extraneous factors.
Some researchers have measured the dependent variable by means of a pre-test/ post-test design, getting dietary data on the migrants before they left the rural communities, and then repeating the observations in the urban setting. This is somewhat more convincing, but the apparent changes from pre-migration to post-migration may be due to any of a series of extraneous factors, not the least of which is the possible effect of the different context of measurement. That is, the situation in which detailed dietary observations are made in the city will differ substantially from the rural context, and generally the interviewers will be different people, perhaps eliciting a different degree of accuracy in responses.
The problem of using a control group as a means of operationalizing the independent variable is severe in practically all field-work. The issues are even more acute when two groups of different cultural background are selected for comparison. This research paradigm, too, is quite common, for example, in a comparison of Indians and Ladinos, or two neighbouring villages, or an agricultural village and a pastoralist village. Of course, it is always of some interest to find that two groups differ in terms of a given variable such as health beliefs, resort to store-bought foods, or degree of meat-eating, but when it comes to assigning causal significance to these differences, there are serious pitfalls. The selection of two groups for comparison on the basis of ethnicity has the effect of elevating their cultural difference to the forefront as the prime independent variable, presumably the causal variable. However, the differences in food behaviours, nutritional status, or other dependent variables may be the result of a wide range of other kinds of variables in terms of which the two populations differ. The populations may be different in age structure, occupational pursuits, sex ratios, socio-economic resources, relations with commercial resources, political ties, and a variety of other factors. Sorting out the prime causal variables can be a formidable task (Cook and Campbell, 1979).
Comparisons of Ethnic Groups Using Experimental Paradigms: Confounding Effects of Socio-economic Status.
In those many field studies in which differences in an independent variable in two or more ethnic groups are compared, the single most common and damaging omission is the neglect of socio-economic variation as a prime underlying variable. Many rural populations have the appearance of general poverty to such a degree that researchers feel that "the villages are practically all the same." However, even the most poverty-stricken of populations generally varies surprisingly from household to household, and village to village, in certain kinds of local resources that may significantly affect food behaviours and nutritional status (DeWalt and Pelto, 1977; Munoz de Chavez, 1974).
The key problem in manipulating the independent variable by means of two groups or two communities can be pictured from another perspective. The experimental paradigm often seeks to structure the independent variable to be dichotomous (two cases), with the assumption of homogeneity within each condition (or case) and major variation across the two or more cases. Even if the researchers seek out, and control for, all manner of confounding variables, the usefulness of the independent variable is often severely diminished because of ubiquitous intra-community or intra-group variability.
To some extent the experimental paradigm of scientific research falls into serious difficulties in field research simply because of the complexities we find in real human communities. However, the experimental paradigm also has an association with fairly narrowly defined, single-cause theoretical formulations. It works best when one particular independent variable is, in fact, a major actor in the system.
Focusing on the Dependent Variables: The Natural Approach
From table 1, we can see that some research questions considered important in both nutritional science and in anthropology - particularly practical, applied questions focus attention on prediction of a dependent variable. In its most general form, the research question is: "What are the causes or predictors of a particular condition or phenomenon, Y?" Of course, there are still many researchers who look for single, primary "causes," but more sophisticated researchers propose models of interacting variables in which certain factors may be of central importance, with various potential "intervening" and "contributing" factors. Consider the question: "What are the factors that contribute to the prevalence of obesity in a region?"
The strategy for research begins with an inventory of all possible contributing elements that are theoretically plausible, including high caloric intakes, lack of motor activity, metabolic disturbances, and so on. The plausible proximate causes can in turn be linked to a series of contributing factors, or "risk factors," including marital status, types of occupations, cultural definitions of appropriate body build, traditional meal patterns, and many others. Figure la illustrates the logical structure of this kind of field research.
Fig. 1. Conceptual differences between experimental and
naturalistic designs in field research
A. Experimental design
X--------------------- (intervening -------------------------Y
(experimental condition) variables) (dependent variable)
non-X (control group)-------------------------------------------- > not Y
Contrast between the experimental population and control population operationalizes the key independent variable as a dichotomy. In the absence of a control group, the pre-test/post-test contrast is sometimes used as the measure of the variable X (v. non-X).
B. Naturalistic design
X3----------------------(independent variables)----------------------> Y
There is no control group in the naturalistic design, as the assignment of values to the independent variables is determined through the naturally occurring range of variation on these characteristics among the households in the research community or region.
Several important features of the model of interacting variables illustrated in figure 1b are of special interest:
The statistical model for this kind of research is that of multiple regression analysis, with variant forms, as discussed by Robbins and Robbins in this volume (also see Asher, 1976). If we find that a given population can be usefully sorted into two groups - obese and non-obese - the appropriate statistical treatment might be a discriminant analysis (Klecka, 1980). This analysis, like the multiple regression, is intended to identify the constellation of variables and their various "weights" that best predict differences in the dependent variable, for example, obesity versus normal weight.
The logic of the natural history research model often leads to the selection of a particular population - in part of a city, a region, or rural community - within which data concerning both the dependent and independent variables are collected for a sample of individuals/households. Thus, neither independent nor dependent variables are manipulated in the original selection of research sites. Since none of the key variables have been established a priori through selection of particular research communities or groups, the key variables in the design can be developed and refined in the course of the research, and all relevant variables may be operationalized as ordinal measures, instead of being fixed in advance into dichotomous form.
Most methodologists note that departures from strict application of the experimental research paradigm reduce the credibility of causal inferences. The clear manipulation of the independent variable in an experimental research design in which there are both experimental and control groups is the strongest basis for assigning causality. On the other hand, the idea of cause itself raises some problems when applied to the complex interrelationships in human behaviour.
Researchers in human behaviour, as well as in other disciplines, are divided on the use of the concept "cause" and "causal modelling." For some anthropologists, avoidance of the term "cause" is seen as a shirking of scientific responsibility. Statements without causal linkages are "mere description," and, as Marvin Harris (1968) put it, "A nomothetic statement which lacks a causal arrow is a contradiction in terms; its absence can be tolerated only as a token of work in progress" (p. 620).
The opposing view has been forcefully expressed by Kerlinger (1973) in his Foundations of Behavioral Research: ". . . the study of cause and causation is an endless maze. One of the difficulties is that the word "cause" has surplus meaning and metaphysical overtones. Perhaps more important, it is not really needed. . ." (p. 393). Kerlinger strengthened his argument by quoting from Bertrand Russell (1953):
. . . the word "cause" is so inextricably bound up with misleading associations as to make its complete extrusion from the philosophical vocabulary desirable . . . the reason physics has ceased to look for causes is that, in fact, there are not such things. The law of causality . . . is a relic of a bygone age, surviving, like the monarchy, only because it is erroneously supposed to do no harm (p. 387)
Why is the term "cause" such a problem? In the first place, much of the logic of causality and the best force of the experimental method is bound up with the idea of single, unique, "necessary and sufficient" cause of a phenomenon. In the epidemiology of an older era, for example, specific micro-organisms were seen as the discrete causes of specific diseases. In its day, that orientation had some usefulness. The search for the specific agents responsible for typhoid, cholera, smallpox, tuberculosis, and other diseases brought about some very effective scientific advances in public health. In recent times the centre of interest in epidemiology, too, seems to have shifted to problems in which well-defined pathogens are absent, and complex, multiple pathways and agents combine to produce cardiovascular diseases, various neoplasms, and various states of nutrition and malnutrition (Susser, 1973). The factors and pathways affecting human food-use patterns are even more complex. Over-adherence to ideas of causality, therefore, faces a number of problems:
Abandonment of the search for "causes" should not be interpreted as a terribly radical shift in scientific objectives. Researchers are still likely to be well-focused on those variables considered to be most important because they have proved to be central in many previous studies. More important, there need be no abandonment of the concept of "independent variables" nor the concepts of intervening, confounding, and other variables in applications of well-known statistical models.
Perhaps the central message is that in terms of research design, nutritional anthropologists have better chances of success in identifying "the factors bringing about Y. e.g. adequate nutritional status" than in seeking clear answers concerning the effects of a particular independent variable.
The most common situation in nutritional anthropology seems to be that the researcher focuses on a particular population - a region, a specific ecological zone, or an individual community - in order to examine certain key parameters of food use, nutrition, and other variables. The first issue, then, is which community to select among a range of alternatives. In some instances the assigned task may be to characterize some features of a particular locale - hence selection must be made to represent that geographic region. But if the main objective is to examine some features of food use and nutrition in a particular type of population, then selection of the right region or situation to represent that larger category becomes difficult. To select a suitable site to represent some aspects of rural communities, urban poor neighbourhoods, or transitional areas, a first requirement is adequate preliminary reconnaissance of possible sites.
Some other principles of research site selection include the following:
These comments only touch the edges of the complex issues encountered in the choice of field research sites. Perhaps the best advice in the matter is to proceed carefully and to seek counsel constantly from other researchers as well as from people knowledgeable about the proposed research regions.
In many research projects, independent variables such as "modernization," "commercialization," or perhaps "degree of urbanism" are considered of key importance. Because of their global, multifaceted nature, it is essential that they be carefully defined with reference to specific features or "indicators" in the research region. Sometimes the variable can be operationalized through selection of three or more communities that represent different points along the selected dimension. Selection should be based on discussions with key informants and certain important indicators, such as numbers of stores and communication services.
Although variables such as degree of commercialization of agriculture may in part be operationalized by careful selection of communities, it is generally advisable to measure the same variable separately for each household or other research unit. For example, one cannot assume that all the households in a village or neighbourhood X are above average in commercial farming just because the community as a whole has a lot more commercial farming than does community Z. Curiously, the error of assigning community-level qualities to the household level has persisted in much anthropological research even though the pitfalls of this ecological fallacy have been well described in the social sciences literature (Langbein and Lichtman, 1978).
Also, it is important to make sure that other qualitative and quantitiative differences between the selected communities are carefully controlled. In the case of a study of the degree of commercialization of farming, these would include size, distance, ethnicity, cropping, and wage-labour differences.. Testing of hypotheses concerning the effects of the independent variable of commercialization can be initially formatted through comparisons of mean values across three or more communities; however, the same hypotheses should be tested using the total sample of households as units, eliminating community from the list of variables.
The logic of the multicommunity research design requires that variables such as "commercialization" and "socio-economic status" must be operationalized by means of criteria or indicators that are comparable across the several communities. Identification of appropriate indicators calls for careful ethnographic pilot research to ensure that a suitable cross-community range of criteria is identified.
Multiple community sites may also be selected simply to reflect the general range of variation in a region. In such instances, it is again important to carry out ethnographic surveying in the region, in addition to using published and unpublished descriptive sources to select communities that are logistically appropriate as well as representative of the range of variation in the region. The main steps in the process would be to:
A more novel and potentially rewarding mode of community selection that deserves attention is that in which a string of communities is identified along a waterway, roadway, or other line of communication. This sampling system may be particularly useful in situations where a geographic distance factor (e.g. distance from the market town) is considered significant. Also, sampling along a line that intersects several ecological zones can be effective.. For example, in some Andean areas researchers have referred to the feature of "verticality," as individual communities have important social links with zones higher up and farther down the steep, mountainous terrain, in order to maximize contact with different ecological zones (Brush, 1978). In this situation the selection of a single community or of two or three communities in the same ecological zone may fail to represent the variations in a region.
Random sampling of communities in a region would be desirable if a truly statistically representative picture were intended. Random sampling is seldom done, however, as usually no great power is added to the research design by claiming that the entire region is truly represented. None the less, it is important to be able to describe in some detail the ways in which a sample of communities represents, and fails to represent, the types of situations found in a valley, district, province, or other regional population.
Sampling in the City
Research in urban settings is seldom intended to be generalized to an entire city population so that random sampling across widely dispersed neighbourhoods in an urban area is not usually necessary. The most useful strategy often begins with selection of a subcommunity or sector within a city that maximizes some of the desired research variables without introducing too many extraneous complexities. Researchers should seek as much information as possible concerning the distributions of socio-economic differentials, ethnic variations, and other features characterizing an entire urban area. For example, if one intends to do dietary research on the Puerto Rican population in a North American city, one should try to obtain census data to identify the range of areas of the city where Puerto Rican populations represent different concentrations. As in our rural examples, it is often useful to select two or three or more different subcommunities, with different concentrations of the desired population, perhaps representing different socio-economic levels, public and private housing situations, and other variations. If locations of health services, stores, and other facilities are considered important, the subcommunities may be selected directly using these criteria.
Previous ethnographic and other research in a particular community or neighbourhood may be an important criterion for selecting locations. Of course, in some instances one avoids repeated research in the same neighbourhoods, but quite often the past research in an area provides valuable ethnographic, descriptive data for identifying key variables and strategies for selecting samples.
In some instances an urban project is developed in which ethnic variation is to be a prime independent variable. Maximum control of the ethnic variable is made possible if a series of different ethnic groups are all intermingled in the same area of the city. Under these conditions, ethnic variation is not confounded by geographic separation into distinct enclaves with different stores, different health facilities, and other features. Ethnic comparisons may run into serious problems with confounding variables when they are based on one group in the "north end," another in "western suburbs," and yet another group in the "downtown inner city."
If it becomes necessary to select two or more different subcommunities in the city to represent different ethnic groups, then ethnographic field-work is essential for a thorough exploration of the various economic and ecological differences between the subareas. Two or more neighbourhoods in a large city may appear similar in their economic and social qualities, yet what seem to be ethnic differences may turn out to reflect micro-ecological differences in the subcommunities, including different landlords, different access to services, and submerged differences in socio-economic status.
Field research generally involves intrusion of an academic or government-sponsored research group into a local region or community. Even the lone field-worker brings an element of intrusion, though the researcher also often brings in a certain amount of money, special resources such as transportation, and other economic advantages that may provide some benefits to the local people. The field-worker or field team assumes some sort of interactional position in the local and regional social organization, and relies on influential local people to give their endorsement or at least to acquiesce in the research activities. Very few field activities can proceed without overt support and approval from local people.
Since nutritional anthropological topics are often of considerable interest to local people and may have direct practical effects, it is useful to involve them quite actively in the research process. Local people may be the most sensitive and appropriate for some of the more intrusive kinds of observations in households. They are most familiar with features of local idiom, recipes, and patterns of household maintenance. Involvement of local people in early stages of the research process can accomplish the following:
The participation of local people is important not only for routine data-collection tasks but also for instrument construction. They can also play major roles in defining locally and regionally relevant sampling units and populations. Of course, there are wide variations in intellectual interests and styles within local cultural groups, so research directors and planners can expect to have difficulties in communicating with and motivating some local assistants, just as one generally experiences some difficulties with research assistants and graduate students.
In some areas, researchers have found it quite feasible to hire field assistants from the nearest university or other academic centre. However, there is often a considerable gulf between the university students (and secondary-school students) and the local people. Capability in the local language does not ensure that urban, university-based assistants will be sensitive and thoughtful concerning local cultural features and feelings.
Contents - Previous - Next