(c) Prediction

A useful hypothesis will enable predictions, by deductive reasoning, that can be experimentally assessed. If results contradict the predictions, then the hypothesis under examination is incorrect or incomplete and requires either revision or abandonment. If results confirm the predictions, then the hypothesis might be correct but is still subject to further testing. Predictions refer to experimental designs with a currently unknown outcome. A prediction (of an unknown) differs from a consequence (which can already be known).

(d) Experiment

Once a prediction is made, an experiment is designed to test it. The experiment may seek either confirmation or falsification of the hypothesis.

Scientists assume an attitude of openness and accountability on the part of those conducting an experiment. Detailed record keeping is essential, to aid in recording and reporting on the experimental results, and providing evidence of the effectiveness and integrity of the procedure. They will also assist in reproducing the experimental results.

The experiment's integrity should be ascertained by the introduction of a control. Two virtually identical experiments are run, in only one of which the factor being tested is varied. This serves to further isolate any causal phenomena. For example in testing a drug it is important to carefully test that the supposed effect of the drug is produced only by the drug. Doctors may do this with a double-blind study: two virtually identical groups of patients are compared, one of which receives the drug and one of which receives a placebo. Neither the patients nor the doctor know who is getting the real drug, isolating its effects. This type of experiment is often referred to as a true experiment because of its design. It is contrasted with alternative forms below.

Once an experiment is complete, a researcher determines whether the results (or data) gathered are what was predicted. If the experimental conclusions fail to match the predictions/hypotheses, then one returns to the failed hypothesis and re-iterates the process. If the experiment appears successful - i.e. fits the hypothesis – the experimenter often will attempt to publish the results so that others (in theory) may reproduce the same experimental results, verifying the findings in the process.

An experiment is not an absolute requirement. In observation based fields of science actual experiments must be designed differently than for the classical laboratory based sciences. Owing to ethical concerns and the sheer cost of manipulating large segments of society, sociologists often turn to other methods for testing hypotheses. In lieu of holding variables constant in laboratory settings, sociologists employ statistical techniques (e.g., regression) that allow them to control the variables in the analysis rather than in the data collection. For instance, in examining the effects of gender on promotions, sociologists may control for the effects of social class as this variable will likely influence the relationship. Unlike a true experiment where these variables are held constant in a laboratory setting, sociologists use statistical methods to hold constant social class (or, better stated, partial out the variance accounted for by social class) so they can see the relationship between gender and promotions without the interference of social class. Thus, while the true experiment is ideally suited for the performance of science, especially because it is the best method for deriving causal relationships, other methods of hypothesis testing are commonly employed in the social sciences.