starting hypothesis

Terms from Statistics for HCI: Making Sense of Quantitative Data

The original hypothesis that an experiment is designed to test. The gold standard statistical process is usually deemed to be that only hypotheses determined before the study should be admissible.
In practice researchers may notice new patterns in their data and then test a new hypothesis based on this, a process termed HARKing or Hypothesis After the Result is Known. This can be dangerous: it is a form of cherry picking and may lead to spurious, purely random results. One of the reasons for pre-registration is to discourage this practice. The ideal is that if you do notice such patterns you should perform a fresh study to investigate them, that is a fresh study where the observed pattern from the previous experiment becomes the original hypothesis of the new one.
However, sticking purely to the original hypothesis is a counsel of perfection that is not always appropriate or achievable, especially when dealing with data such as epidemiological data or historical data that it is impossible to recreate, or where the original data collection is very expensive or difficult to access. In such circumstances the crucial thing is to be aware that you are creating a new hypothesis and then, where possible, perform a suitable multiplicity adjustment such as a Bonferroni correction, and definitely be very clear in any published work that this was not the original hypothesis.

Defined on page 82

Used on page 82