Hypotheses are testable explanations and predictions.

Many people think of a hypothesis as "an educated guess" (Moriarty, 1997). Is "an educated guess" a good definition of a hypothesis?

An "educated guess" is actually not a bad definition for a hypothesis -- simply an incomplete definition.

In practice, hypotheses are not really "guesses," but instead represent our best understanding of a system based on all the information that we currently have. 

But like guesses, hypotheses are tentative statements. We don't know if hypotheses are true or not. Therefore, one of the most important aspects of hypotheses is that we must be able to determine whether hypotheses are true or not true: hypotheses must be testable

For example, a testable hypothesis could be:

"If I write my paper a week before the deadline and discuss it with my instructor before revision, I will get an A on the paper."

The hypothesis is specific enough to make a testable prediction. However, the hypothesis is tentative: discussing a paper with the instructor is not a guarantee of an A (although still a good strategy to get a higher grade). 

An example of a statement that is NOT testable would be:

"I was a falcon in my past life." 

Because there is no way to measure anything about past lives, we cannot test a statement about past lives. Therefore, the statement cannot be a hypothesis.

A  more specific definition of a hypothesis could be:

DEFINITION:   "A hypothesis is a tentative, specific, explanation or prediction of a phenomenon or an observation that can be rejected by experimental data."

Expressing a definition using only one sentence is concise. However, this definition of a hypothesis has a problem: the sentence expresses two ideas (explanation and prediction), which can be confusing (Strode, 2015). Therefore, it is useful to analyze (break apart) the idea of a hypothesis by defining two separate terms:

DEFINITION:  A "General Hypothesis" is a tentative, specific explanation of a phenomenon that can be rejected by experimental data."

DEFINITION:  A "Measurable Hypothesis" is a tentative, specific prediction that can be rejected by experimental data."

For example, one framework for hypothesis testing is to write statements of the form "If __________, then __________, because __________." The "If, then, because" framework is useful because it includes (and separates) both General and Measurable hypotheses. 

The "If __________, then __________" statement corresponds to the Measurable hypothesis.

The "because __________" statement corresponds to the General hypothesis.

The hypothesis "If I write my paper a week before the deadline and discuss it with my instructor before revision, then I will get an A on the paper" could be an example of a Measurable Hypothesis that emerges as a prediction of a General Hypothesis. We might expect a high grade because of our General Hypothesis that "Instructor feedback contributes to effective revision and leads to higher-quality writing."

Given our definitions, it will be useful to explore General and Measurable hypotheses in more detail. 

In summary, we can distinguish between two different types of hypotheses: “General” hypotheses (scientific models) and “Measurable” hypotheses (testable predictions that emerge from the scientific models). Scientific models explain phenomena in ways that generalize to many different situations. General scientific models can then lead to many different predictions. Predictions are most useful if the predictions are measurable – i.e. expressed in terms of measurements that are possible to make. 

However, there are also other types of hypotheses. For example, statistical hypotheses  are usually necessary to compare the results of actual measurements to the predictions of Measurable Hypotheses, i.e. to “test” the Measurable Hypotheses. 

Specifically, statistical hypotheses test the ("Null") prediction that observed patterns in data happened by chance through random sampling.

If observed patterns are unlikely to have happened by chance, then the patterns may have happened for a scientifically meaningful reason.

However, the results from testing statistical hypotheses are only one piece of evidence that contributes to testing measurable hypotheses. Other evidence, and reasoning that is both valid and sound are also important for testing measurable scientific hypotheses.

A thorough discussion of statistical hypotheses is beyond the scope of our current discussion. However, we can note that the question of how best to test research hypotheses using statistics is an area of active scientific debate.