Hypothesis testing is the one thing that separates science from all other human endeavors, so what is it?
Here are some characteristics of quality scientific hypotheses as gathered over various sources and based on experience. Note that in many cases, scientists must do a lot of intellectual work to go from fuzzy notions to crisp, powerful hypotheses.
Clarity - An hypothesis should be understandable to you and to others; dependent and independent variables should be identifiable; simplicity is prized.
The scope or domain over which an hypothesis applies should be implicitly clear or be specified.
Falsifiability - there should be a bona fide way that an hypothesis could be found to be wrong. Logical truisms, for instance, are not hypotheses.
Testability - An hypothesis should predict something observable so that there is a way to challenge it with new data or observations.
a. Classically, scientists often work by establishing null hypotheses of no effect or no difference. They then see how much their data disagree with this null hypothesis, a somewhat convoluted approach to knowing.
b. Other forms of reasoning, such as Bayesian Analysis, evaluate the likelihood of competing hypotheses given new data.
Importance and generality - Hypotheses that apply far and wide are more useful than ones that apply only to narrow circumstances.
Here is a presentation that does a good job covering some of the basics of hypothesis creation.
An interesting and rigorous approach is to "pre-register" not just your hypothesis but also your study design, thus avoiding the powerful temptation to invent an hypothesis after data are collected and analyzed.