Description
Break an experimental task into steps, reason about how each responds to the manipulated factor, and time the steps so a predictive formula replaces a single 'optimum'.
A study that finds one best setting rarely transfers. Dix revisits an early mobile study that declared one font size best for searching paginated menus. Decomposed, the task is visual search of a screen, paging to the next screen and selecting the item. Larger fonts help search and selection but increase paging, so the optimum is a trade-off set by menu length and screen resolution. With step times taken separately from logs, an exact page count and an empirical page-down time, one could write a formula that gives the best font size for any menu length and predicts other tasks with similar steps. Thinking in mechanisms also separates interpolation from extrapolation and makes replication meaningful.
Attributes
- Discipline
- Research Design & Rigor
- Loss type
- Latent
- Stage
- Design, Evaluation
- Data
- Quantitative
- Effort
- Weeks
Steps
- Write the experimental task as a sequence of user steps.
- For each step, predict how the manipulated factor will affect it and in which direction.
- Identify trade-offs between steps that make any single optimum specific to the task and device.
- Instrument logs to time steps separately, and compute deterministic quantities (such as page counts) exactly.
- Combine the step models into a formula or algorithm that predicts performance for new task parameters.
- Use the model to make predictions before the study and to analyse and bound generalisation afterwards.
Sources
- Statistics for HCI (Dix, 2020)
Also known as
- Mechanism decomposition for generalising experimental results
- Unpacking the mechanism
Categories
- Research Operations
- Research Design & Rigor