Here is a relevant fragment of my paper which will be published in the proceedings of the Society of Philosophy of Science (SPS) conference in Montreal (June 2012):
"3. Levels of Confidence in Scientific Claims: Data and Explanations.
A discovered experimental fact is usually presented to the scientific community, to be independently confirmed or refuted. Experimental results are accepted--at a high level of confidence--when they become reproducible on demand. Absence of such reproducibility justifies suspicion of possible errors or fraud. Methods of validation of theories (explanations of facts) are slightly different. A new scientific theory is also presented to a community of experts, to be independently evaluated. Their level of confidence in a theory depends on the validity of underlying assumptions and on the rigor of quantitative analysis. But even a most reliable scientific theory, called a law, is said to be falsifiable, in principle, when conflict with reproducible-on-demand data becomes undeniable (15). Such unusual conflict could trigger a scientific revolution (16).
To explain something usually means to identify causes and to construct a logically satisfying model of reality. An attempt to explain a fact, or to resolve an apparent logical conflict, usually leads to discoveries of other facts. A classical example was the discovery of planet Neptune, in 1846. A more recent and less widely known example was the discovery of a subatomic particle named neutrino. Experimental data collected in the 1920's showed that beta rays (electrons emitted in radioactive decay) had lower mean energies than expected on the basis of the theoretical E=mc2 formula. Austrian theoretical physicist W. Pauli solved this "logical inconsistency" by suggesting that tiny neutral particles, later named neutrinos, were responsible for the missing energy. His hypothesis was formulated in 1933. Experiments confirming the reality of neutrinos were performed, 23 years later. "