22.214.171.124 Observation as a means of deriving hypotheses, building models or refining theory: analytic induction and conventional ethnography
126.96.36.199.1 Analytic induction Observation is sometimes used as a means to inductively derive hypotheses from research where there is a lack of effective theory (Smelser, 2001). The usual intention is that such hypotheses will be subject to more 'rigorous' testing through formal surveys or other data collection techniques. In similar vein, where data analysis inductively suggests relationships between groups of variables, such as in factor analysis (see Section 8), then observation may be used to clarify the relationship and frame appropriate conceptual variables that encapsulate the factors derived from the statistical analysis.Brewer (2000) suggests that qualitative data can illuminate causal inferences from survey data and it can be used to add a longitudinal dimension to survey data.
However, observation can be used as a means in itself to derive theory through a process of analytic induction(Znaniecki, 1934). Znaniecki originally proposed analytic induction as an alternative to statistical approaches, which he referred to as 'enumerative research'. Znaniecki argued that sampling and statistical analysis produced correlations but could not account for exceptions in statistical relationships. Analytic induction, he suggested, was a more scientific approach to causal explanation. The approach was to initially test a hypothesis on one or a small number of cases first and to continue to do this until a disconfirming case is found, which leads to a revision of the hypothesis and further testing, or until the point is reached when further study will reveal nothing new (Znaniecki, 1934, p. 249).
The approach became established as a core method in sociology, but more often linked with ethnographic research than with quantitative approaches, which continued to prefer correlational analysis.
Analytic induction, historically, is the conventional approach to observational research/ethnography in sociology. AlfredLindesmith (1947) adopted it in his study of opiate addiction, which inspired HowardBecker's (1953) study on marijuana smoking: 'I wanted to experiment with the method of analytic induction, which I had found in a study too little appreciated, Alfred Lindesmith's book on opiate addiction' (Becker, 2003).
188.8.131.52.2 Analytic induction approach The analytic induction approach is illustrated inWhyte's (1949) study of a fast-food restaurant (Case Study: Whyte (1949)). The procedure is as follows.
First, to describe the situation that is the subject of the study (what people do, what the purpose of the organisation is and how people relate to one another, and so on), as Whyte does when describing the fast-food restaurant.
Second, make some initial statements about the social processes observed (often based on crude counts). Whyte notes the relationships that occur along the chain of communication between customer and cook. He highlights the tensions that occur when low-status workers initiate actions for higher-status workers.
Third, suggest some initial hypotheses or generalisations based on the observation; that is, proceed inductively, proposing hypotheses based on the data. Whyte hypothesises that 'minimizing direct contact between lower and higher status workers minimizes tension'.
Fourth, test out these initial propositions by looking for negative cases. If negative cases arise then modify or discard the propositions. Whyte examined situations in different restaurants to see whether his hypothesis was confirmed.
Fifth, gradually build up a model of the social organisation or the group interactive processes based on the propositions that survive the scrutiny of empirical data. Whyte did not suggest that he had tested a theory but that he was building up a model that described lines of communication in fast-food restaurants and helped explain the nature of friction within a service industry.
Thus, this analytic inductive approach works, essentially, by suggesting a model and then refining it by systematically seeking out instances that falsify aspects of the model. If it cannot be shown to be incorrect then it is taken as the best version of the model available. Gradually, by empirically falsifying the incorrect suppositions, a 'truer' model of what is happening should emerge (Becker, 1958; Becker and Geer, 1957,1960).
Whyte's approach was also subsequently codified as a set of steps inCressey's (1953) study of fraudsters. In his account, Cressey places more emphasis, in the final stages, on developing a theory rather than a model of interaction. Cressey's steps, which closely mirror Whyte's stages, are:
1. Identify the phenomenon to be explained. 2. Formulate a working definition of the phenomenon.
3. Formulate a working hypothesis to explain the phenomenon.
4. Study one case.
5. Assess whether the case data fit the working hypothesis.
6. If so study the next case. If not, either redefine the phenomenon to exclude the case or reformulate the working hypothesis.
7. Contine step 6 until the emerging theory has accounted for all cases: negative cases require redefinition or reformulation.
In theory, of course, this last stage could go on endlessly and the researcher has to make the decision that continued observation of cases (such as restaurants in Whyte's (1949) study) keeps showing the same patters of activity and thus there is no need for further observation. In practice, resource and time constraints are much more likely to determine how many cases are observed.
Logically, though there is a problem with this inductivist approach, namely that one has to logically infer from the known cases about the nature of unknown cases.
184.108.40.206.3 Analytic induction: a scientific approach? Znaniecki had claimed that analytic induction would lead to causal laws that would enable predictions. However, in practice, this element was dropped by subsequent users of the method, especially in the wake of a critique byTurner (1953), which, although endorsing analytic induction, denied that it could lead to causal laws. 
Despite the systematic approach of analytic induction, the conventional approach to ethnography does acknowledge that participant observation is as 'objective' or 'reliable' as the social survey in establishing proof or providing explanations. The conventional approach to ethnography. tjough,sees participant observation as fraught with dangers as a 'scientific' method but one that can be extremely revealing if approached systematically.
Interestingly, the debate in the 1940s has been recycled half a century later under the guise of 'postmodern ethnography' (see below).
Although originally dating from the 1930s, it should not be assumed that the analytic inductive approach is no longer used. Indeed, the conventional approach has been very resilient and continues to be adopted.
There is a view that Becker and colleagues (often referred to as late Chicago School, or the Chicago School of deviance) were far more radical in their approach to participant observation than they actually were. They certainly did not adopt any thing like a radical phenomenological position (Section 2.3.1) despite ostensibly advocating a symbolic interactionist perspective, which might have allowed a more radical casting of naturalistic data. On the contrary, they were conservative and, while taking issue with the overblown pseudo-scientific claims of the social surveys of the time, they sought to establish a degree of validity and reliability of their observational data, conceding to the positivistic norms. It was no surprise that later, their disciples, such as Denzin, engaged in a soul-searching self-critique of observational data.
220.127.116.11.4 Conventional observation research The conventional approach to observational studies, which draws on analytic induction, does not set out to discover and interpret meanings as phenomenologists might (see Section 3.3.2). Rather, it tests hypotheses about the nature of social interaction or organisational structures. Meanings are not of concern in themselves, but only as supplementary material for explaining why, for example, some social interaction is accomplished smoothly while other interaction is fraught. Whyte's (1949) CASE STUDY is a good example of this.
The conventional approach uses existing theoretical concepts and propositions to guide the analysis through a systematic collection, classification and reporting of 'facts'. Conventional ethnography is thus characterised by detailed empirical description to reveal social processes rather than broad causal generalisations. Whyte, was reluctant to make any broad causal generalisations although clearly pointing to factors that he thought caused friction within the restaurant's flow of work.
If participant observation is used, then the conventional approach suggests that it is crucial for the participant observer to maintain a balanced perspective. It emphasises detachment and highlights the importance of the researchers' reflexive accounts of their field role, which are neutral in conventional approaches to observational research (or ethnography more generally).
Subsequent developments of the conventional approach have taken up the issue of reflexivity as central, and argue that the researcher can never actually be neutral. Indeed, the reverse is true: all researchers are participants in some form or another and, therefore, the reflexive activity is crucial. Researchers should take account of their impact on the research situation and ensure that they communicate this clearly in any report (Hammersley 1990, Hammersley and Atkinson, 1983)