All scientific research is based on the construction of models. A model is a symbolic and conceptual structure derived from real world observation. It must always have the status of an idea, an abstraction rather than a concrete reality. An excellent example of this is given by the Minkowski and Riemannian spaces of Special and General Relativity. Yet even in sciences which are not predominantly mathematical, the employment of abstract models is essential. Thus, once the data gathered from observations by many researchers around the world culminated in the construction of the double-helix model of DNA, the model became the primary focus of attention and discourse.
This is the point: one is obliged at a certain point to dispense with observations of the "real" world and just work with the model. From investigation of the properties of the model one then makes predictions that are then tested against reality, thereby strengthening or weakening one's faith in the universal validity of the model itself.
However, one's science can only be as good as one's models, and all models are (of necessity) gross oversimplifications of the reality one seeks to elucidate. To take one example: Hydrodynamics is replete with "paradoxes". These seemingly contradictory phenomena do not exist in nature, but are a consequence of the fact that all hydrodynamic models simplify certain quantities while completely discounting others (density, viscosity, temperature, compressibility, etc.), That's because one is desperately in need of finding equations one can solve . However the number of differential equations solvable in terms of elementary functions , or the number of "quadratures" (integrals with solutions in closed form), is quite limited.
The truth of the matter is that construction of a model is best described as a "slash and burn" operation. Literally one "hacks away" at the structure of the equation to find some mutilated reduction that one hopes comes close to capturing its complexity, ( which complexity is already an over-simplification of the natural phenomena from which it is derived!)
From the kinds of models, skewed through approximation or even the eliminating of fundamental variables, that one encounters with such frequency in the sciences, it is indeed possible to conclude almost anything. A notorious example is in the proliferation of bad science churned out by the big polluting industries and the government to "prove" that Global Warming is a myth, or a phantasm, a fabrication or irrelevant. This is easily done through emphasizing or belittling this or that combination of input parameters that structure the impossibly complicated models for global climate, from which, through reasoning on the model, one derives prediction of climate change.
Models are very welcoming to the natural human tendency to extrapolate. This is particularly true in the reconstructive sciences which try to recreate how things were in the past. Much of the scientific wisdom in such disciplines as archeology, paleontology and even history, reveals itself under close inspection, to be little more than wild speculation from a single bone fragment, or piece of pottery, or an imaginative filling in of missing links between two bits of data widely separated in time. To this one must add a healthy dose of science fiction. Indeed one is sometimes able to identify which science fiction novels the historian or archaeologist has been reading of late !
A significant amount of bad science is also derived from the correct application of "scientific method" to skewed, narrowly conceived, or flawed models. As has been pointed out by every philosopher from Hume to Popper to Kuhn, one cannot derive models from data and observations alone , that is to say, from "scientific method". Imagination, intuition, fad and fashion and personal prejudice all have their share in the construction of them.
Predictions based on a model, if one is not careful, may merely predict how the model itself behaves over time. Yet it often happens that the catalogue of quantifiable "observables" that go into the construction of the model is insufficient to the underlying reality. Biologists, for example, have a fatal tendency to confuse "pain" with "injury". They are far from being the same, however "injury" is easily quantifiable, while "pleasure" and "pain" can at most be vaguely correlated on some numerical scale. As consciousness can't be effectively quantified, the biological sciences either ignore it, or turn it over to the worst of all pseudo-sciences, Psychology.
Likewise, Economics is largely concerned with models based on the "hog theory"; that is to say, human needs are insatiable. As a result, major decisions by governments that affect all of us turn out to be arrived at through a process of rational decision-making based on "hogonomics". Although it may be true that human beings can never get enough of what they need, it is also the case that those needs that can be satisfied by monetary transactionsare surely limited. I may want to be a great violinist , however no amount of money or economic theory can gratify this desire. Nor did Nero become a great lyric poet through killing off his rivals.
Finally, take the example of Medicine. Restoring a person or a society to health depends on a model of health. Yet the contemporary perspective of what it is to be human is of a being so shabby and mediocre, that it comes as no surprise that much of modern medicine is based on high technology, invasive operations and the administration of worthless and poisonous pharmaceuticals.