"Survey science" implies an experimental approach in which an experimental protocol is constructed that is followed repetitively on a number of occasions in order to accumulate a large corpus of data. Such a protocol is established prior to conducting experiments and collecting data, and is usually not altered during the course of the experiments in response to the data required. Collecting these data may involve dozens of iterations of a measurement or millions of such iterations, and the accumulated data may fill one laboratory notebook or many computer tapes.
Implicit in many such surveys and the protocols that guide them is the notion that a single, well-executed measurement will not suffice to produce unambiguous conclusions and that repetitive performance is required to achieve that end. This requirement for repetition may be necessitated by the heterogeneity or variability in the subjects of the study, unreliability in the measuring technique, lack of uniform competence amongst a large group of experimenters, and so forth.
The results of these measurements are commonly susceptible to statistical analysis, and more often than not, introduction into computerized data banks for storage, analysis, and retrieval. Examples of such "survey data" include clinical trials of drug regimens, DNA sequencing, epidemiological studies, other types of population studies, ecological surveys, and gathering of x-ray crystallographic data.
A quite distinct method of doing science is to follow an experimental course in which a succession of distinct, unique manipulations is performed to reach the end result. Here, the precise experimental course cannot be laid out in advance, since, importantly, the outcome of an initial manipulation will determine the precise nature of those that immediately follow it; moreover, each of the steps in such a succession may itself be invested with an elaborate logical structure that determines the nature and interpretation of its outcomes.
While the design of each of the steps in such an experimental succession may be dictated by generally accepted technical practices and logical models, the precise nature and outcome of any given step is not predictable, since it will be determined by those steps preceding it. In contrast to survey-type experiments, in these manipulative experiments the protocol is constantly altered in response to the most recently obtained results. When compared with survey-type research, the logical