Chamberlin (1897, yes there is no typo in the century indicated) gives what should be obvious advice. If you have only one hypothesis, especially if it has been named after you or a friend, you will be loath to reject it. Lakatos (1970) convinces me that theories are never rejected until a better one is available. Work with multiple, alternative hypotheses and determine what data will be required to discriminate among them. Make them explicit ahead of time or else you will devolve easily into post hoc explanations rather than genuine, a priori hypotheses.
William of Occam, a British theologian of the middle ages, also made a useful suggestion that is and should be followed by most scientists most of the time. Note that it was not developed for science or by a scientist or philosopher of science and is not necessarily a good criterion to apply to science. The gist of Occam's razor is to avoid a more complicated model or explanation when the data can be fitted equally well or better by a less complicated model or explanation. In statistical regression, for example, one can exactly fit an nth order polynomial through n + 1 points, so why would one ever draw a polynomial of higher order through n + 1 points? A scientist might do so usefully if she or he had a mechanism in mind that yielded a higher-order polynomial. Occam's razor is very useful for preventing the complexity of theory from growing at a rate greater than is justified by the data, but it is not any absolute guide to scientific truth. A more complicated theory may be better and can be suggested, but there is no reason to adopt it unless it makes testable predictions that the simpler one did not and at least some of those predictions are fulfilled. Hypothesis generation requires suspension of Occam's razor, since a good hypothesis will fit not only existing observations but will anticipate uncollected ones. Using Occam's razor slavishly would lead to hypotheses that explained existing data and nothing more.
Mayr (1982, especially pp. 21-82 and 829-858) notes frequent traps. Biologists often seem to polarize issues. For example, take the ill-posed as well as polarizing question of whether competition or predation is the more important in structuring communities. Or, take the example of "bottom-up vs. top-down" control of community structure. Clearly both act, so this case is not one of mutually exclusive, alternative hypotheses. Nor are all possibilities included. These particular examples are two of many in ecology of (Mayr 1982, p. 844) "failure of opponents to see that the two opposing viewpoints did not exhaust the number of possible explanatory choices." More generally, it is important to recognize that multiple hypotheses can be overlapping or mutually exclusive and that the list in hand may be either exhaustive (cover all possibilities) or not. Work toward mutually exclusive, equiprobable (a priori), exhaustive, multiple alternatives, but recognize when and where you fall short of this ideal. Avoid the transparent trick of creating "straw" hypotheses so unlikely that they are of no interest or already have been beaten to death (e.g., the null hypothesis that fluid flow does not influence larval settlement). Keep in mind the etymology of "hypothesis"; work from or toward a thesis rather than with an ad hoc jumble of little hypotheses. How can you have hypotheses without a thesis? Beware of tautologies, cases where the "then" part of your statement follows as a logical necessity from the "if" part (Peters 1976).
It is easy to slip into teleology, explaining the "purpose" of a biological structure or behavior. Heritable genetic variation in structure and function allows natural selection for traits. Traits do not evolve for or with a purpose.
Some biological oceanographic studies suffer a peculiar kind of myopia or chauvinism. Specialists in a given taxon or on a particular trophic or size group tend to do whole-water experiments or field observations and interpret the results as due principally or entirely to their pet group. For example, many studies of "exudation" by phytoplankton do not exclude the likelihood that the contents of phytoplankton are spilled in herbivory or diffused from herbivore fecal remains. To carry the analogy too far (reductio ad absurdum), a shark victim "exudes" bodily fluids during attack and upon defecation (by the shark; no implicit reference to défaillance corporelle).
Oceanographers of many sorts make a particularly lame excuse for avoiding predictive theory and making up "just-so" stories about haphazard collections of data. They argue that oceanography is an observationally limited field, making [undirected] observation of paramount value. The argument also extends to the value of technological development independent of interesting questions: "If only we could make better observations..." Likewise, some oceanographers argue that because manipulations of whole systems and realistic laboratory simulations of whole systems are difficult or impossible, we might as well forget about hypothesis testing and just go ahead and collect observations. Those assertions may even sound plausible, but think about astronomy, wherein manipulation and observation are at least as difficult. Astronomy clearly benefits from systematic iteration between hypothesis generation and hypothesis testing. So should and can oceanography. Contrary to popular opinion, it is impossible simply to collect scientific observations without explicit or implicit theory. What you choose to measure or observe depends on your beliefs. It is important to make those beliefs explicit before collecting the observations, and to entertain alternative beliefs at the same time, as well as to question the quality of the data. To do less shortchanges the science.