A researcher may begin a research project with a broad question, such as "Why is air pollution in large cities growing worse?" However, a single experiment cannot address this question. The researcher must narrow the question -- for example, "Does traffic affect air pollution in Houston?" is a question that may be reasonably put to the test. Then the researcher formulates a testable hypothesis. In this case, it might be that increasing traffic negatively impacts air quality in Houston.
A thorough review of previous literature on the topic provides essential insights for the researcher. The research to be conducted must not be identical to research that has already been done. At the same time, new research should build upon the previous research. Results from earlier studies may show a need for revision or refinement of the hypothesis. Previous studies also provide ideas for methods to use in conducting the experiment.
Experimental design begins at the end, by deciding what data needs to be collected from the experiment. The researcher decides what qualities to measure, what measurements to use and how to analyze the data. This stage of the design often requires defining vague terminology -- for example, the factors that are included in air pollution. The researcher then designs an experimental procedure that will allow all the required data to be colleted. To provide meaningful comparison, the experimenter collects data from an affected area or group and from a "control," which is as unaffected by the factors being tested as possible.
Scientific research must avoid both the personal bias of researchers and bias in the design of experiments. Researchers should remain detached from the results of experiments and avoid using biased language in reports. To avoid bias in experiments, researchers should clearly state their assumptions, thoroughly document the procedures and systematically perform them. Ethical considerations are important for experiments that use human or animal subjects. Research ethics also includes full and honest reporting of the experiment, including any changes to the methods or unexpected results.