1.2 - Aims and hypotheses

Aims and research questions

An aim states the purpose of the investigation without assuming its eventual result.

An aim could be “to investigate whether background speech affects word recall”. A research question expresses that purpose as a question: “Does background speech affect the number of words recalled?” Both focus the investigation without claiming that speech necessarily reduces recall. “To prove speech ruins memory” assumes the answer and is too broad.

A research question should identify the relevant relationship or difference and be possible to investigate. “Why do people forget?” is a useful broad concern, but a single small study needs a narrower question. One study can compare recall after learning the same kind of word list in silence or while recorded speech plays.

For a correlation, the question concerns an association: “Is the number of revision sessions in the past week related to score on a ten-question quiz?” The wording does not imply that the researcher manipulates revision or establishes its effect. The method should fit the question rather than quietly changing an association into a causal claim.

Alternative and directional hypotheses

An alternative hypothesis predicts a difference or relationship that the research can test.

It goes beyond an aim by stating the expected pattern. An experimental hypothesis is an alternative hypothesis for an experiment. A directional, or one-tailed, hypothesis predicts the direction of the difference or association, such as higher/lower or positive/negative. “Music affects memory” lacks the direction and clear measures required for a precise directional prediction.

Suppose students learn 20 words for one minute, either in silence or while recorded speech at a fixed volume plays, then write as many words as they can remember after a two-minute delay. A directional alternative could state: “Students who learn the words in silence will recall more correct words out of 20 after the two-minute delay than students who learn them with recorded speech.” The conditions and outcome are operationalised: their exact manipulation and measurement are specified.

For a correlation, a directional alternative could state: “There will be a positive correlation between the number of revision sessions of at least 20 minutes in the past week and the number correct on a ten-question quiz.” It predicts association, not that sessions cause scores.

A direction should be chosen before inspecting the new results and justified by relevant theory or previous evidence. Choosing “higher” only because the collected scores happen to be higher turns a prediction into a description after the event. One-tailed and two-tailed refer to how the prediction relates to later statistical testing; this lesson teaches their meaning, while the testing calculations come later.

Non-directional hypotheses

A non-directional alternative predicts a difference or association without predicting its direction.

This is also called a two-tailed hypothesis. For the word-recall study, it could state: “There will be a difference in the number of correct words recalled out of 20 after a two-minute delay between students who learn in silence and students who learn with recorded speech.” Either group could recall more; the prediction remains a difference, not no difference.

For correlational research, it could state: “There will be a correlation between revision sessions of at least 20 minutes in the past week and number correct on a ten-question quiz.” Both a positive and a negative association would fit its direction-free prediction.

Non-directional does not mean vague. Both variables and relevant conditions still need clear definitions. It is suitable when there is not enough consistent prior evidence to justify a particular direction, or when either direction is of interest. A directional prediction is more specific but risks overlooking an effect in the other direction if the testing plan is inappropriate. The researcher decides before collecting the main data, rather than switching wording to suit the outcome.

Null hypotheses and testable wording

A null hypothesis states no population difference or association of the kind the study tests.

For the word study, a suitable null is: “There will be no difference in number of correct words recalled out of 20 after a two-minute delay between the silence and recorded-speech learning conditions.” Statistical testing asks whether the sample result is sufficiently incompatible with this null model to reject it. Sample scores can differ by chance even when there is no underlying population effect; the null does not demand identical scores for every participant.

For a correlation, use no association: “There will be no correlation between the number of revision sessions of at least 20 minutes in the past week and the number correct on a ten-question quiz.” Do not write that the two co-variables are “the same”, since sessions and quiz scores are different measures. Adding “any observed association is due to chance” is a common way to express the role of chance variation. Wording such as “no significant difference/correlation” is also commonly accepted when the variables are clear; it must not be read as proof that a later non-significant result establishes exact equality.

To formulate a testable pair, first decide whether the study compares conditions or measures an association. Specify the conditions or co-variables, define the outcome, then state an appropriate alternative and corresponding null. Check that the two statements concern the same variables and population. “Participants will improve” lacks a comparison or time point, while “people will feel better” lacks a clear measure.

For example, measuring each student's hours of sleep last night and response time on a fixed attention task calls for an association hypothesis. A directional alternative predicts a positive or negative association only if justified in advance. Its null predicts no association between those same measures. The hypothesis cannot by itself supply causal control that the study lacks.