Analytic induction refers to a systematic examination of similarities between various social phenomena in order to develop concepts or ideas. Social scientists doing social research use analytic induction to search for those similarities in broad categories and then develop subcategories. For example, social scientist may examine the category of 'marijuana users' and then develop subcategories for 'uses marijuana for pleasure' and 'uses marijuana for health reasons'. If no relevant similarities can be identified, then either the data needs to be reevaluated and the definition of similarities changed, or the category is too wide and heterogeneous and should be narrowed down. 
In the earlier sociological papers (from 1940s and 1950s) this term could also be used to mean the search for "universals" in social life, where "universal" meant an invariant, complete, positivistic propriety (i.e. "all black males between 35 and 40 vote for Democrats"). 
This principle was formulated in 1934 by Florian Znaniecki in order to identify universal propositions and causal laws. It is contrasted with enumerative research, which can only identify correlations and cannot explain outliers in statistical relationships. The principle was further refined by Alfred Lindesmith (1947), Donald Cressey (1950), W.S. Robinson (1951), and Howard S. Becker (1963). Taylor & Bogdan (1998) Eventually it became one of the classic research methods in ethnography.
According to Znaniecki analytic induction involves "inducing laws from a deep analysis of experimentally isolated instances", and is more advanced than simply defining and using terms in advance of research. It involves inductive (rather than deductive reasoning) and allows modification of social concepts throughout research. This allows an accurate reflection of the studied phenomena.
Analytic induction is based on ontological and epistemological realism which relies on Cartesian dualism and assumes an equation of truth (albeit, partial truths) with the faithful reproduction of the object as factual knowledge.
Researchers using analytic induction pursue three fundamental questions: Under what contexts do patterns arise? Under what contexts do the exceptions to the pattern arise? What significance do these patterns and exceptions hold? Like all research, analytic induction is a subjective process. However, researchers build their analyses carefully and provide enough information for readers to evaluate the credibility of their argument. 
The strength of analytic induction is its capacity to generate complex, theoretically rich understandings of social life. To some extent, analytic induction does enable researchers to theorize and anticipate contexts in which a phenomenon might arise, yet this is quite different from the generalizability of quantitative research. However, Analytic induction is never intended to study how the machinery of marginalization produced viable subjects for exploitation. Nor is it intended as a method for examining routine relations of power and privilege. The epistemological foundations on which it is moored render it ineffective for such analysis today. 
- Robinson, W. S. (1951). The logical structure of analytic induction. American Sociological Review, Vol 16, no 6, pgs 812-818
- Znaniecki, F. (1934). The method of sociology.
- Steven J. Taylor, Robert Bogdan, Introduction to Qualitative Research Methods, John Wiley & Sons, 1998, ISBN 0-471-16868-8