Skip to content

Experimental design - Category Home

Experimental design - Topics

Freezing a statistical analysis plan before the treatment codes are unblinded Allocation concealment is a different problem from blinding Testing an interaction term is not the same as testing the simple effects Dose-response designs with three doses and the shape of the curve Load cell drift across a six hour test and the calibration check nobody schedules Why does the target chamber need a vacuum good enough that the beam does not scatter before it arrives Multiple comparisons: how many subgroups can you slice before the results are meaningless? What decides how much laser you actually need is the smallest effect your detector can resolve Factorial designs: testing two variables at once without doubling the work The theory-ladenness of observation undermines the idea of a neutral experiment Masking the group labels from the analyst so the statistics are done blind Pragmatic versus explanatory trials: does the question ask whether it works or whether it works here? How much pilot data is needed before predicting an effect size is honest Sequential testing: why peeking at the data changes the error rate Cluster randomised designs when the clinic rather than the patient is the unit Realigning a regenerative amplifier is one of the few lab jobs that becomes a full afternoon What to do when you cannot randomize participants Writing competing hypotheses before collecting data instead of one favourite Randomization: the cheapest way to defeat confounding variables The neutrino that arrived faster than light in 2011 and was never a discovery Difference-in-differences needs a parallel trend before the treatment starts Clock synchronization across three recording devices and the milliseconds that vanish Cook's distance flags the single observation driving the regression line Deception studies and what a debriefing has to accomplish to make them acceptable Multi-arm designs: four treatments sharing a single control group Threats to validity you only notice after reviewers point them out Intent-to-treat vs per-protocol: the analysis population changes the answer From a vague idea to a testable procedure: designing your first real experiment Non-inferiority margin: choosing it from historical data instead of convenience Observation and experiment are not interchangeable: laboratory vs field studies How many replicates does this experiment actually need? Why every experiment needs a control group and what counts as one