Choosing an analysis
The right analysis depends on two things: what kind of outcome you measured, and how the experiment was laid out. aStat suggests one from your columns, but you know how the trial was run.
First: what is the experimental unit?
The experimental unit is the thing a treatment was applied to independently. If a diet was fed to a whole pen, the pen is the unit, even if you weighed every pig. Analyzing each pig as if it got the diet on its own overstates the evidence (pseudoreplication). Use pen averages, or a mixed model that includes the pen.
Measurements (weights, gains, yields)
| Your question | Layout | Use |
|---|---|---|
| What do the numbers look like? | Any | Summary statistics |
| Is the mean different from a target? | One group | One-sample t-test |
| Do two groups differ? | Different animals in each group | Two-sample t-test |
| Did the same animals change? | Before and after columns | Paired t-test |
| Do three or more groups differ? | Completely randomized | One-way ANOVA |
| Do treatments differ, allowing for barns, litters, or fields? | Every treatment in every block | RCBD |
| Same, but blocks too small for every treatment | Incomplete blocks | BIBD |
| Same, blocking in two directions (row and column) | Each treatment once per row and column | Latin square |
| Do two factors act together? | Every combination tested | Two-factor ANOVA |
| Do groups differ after allowing for starting weight? | Group, plus a measurement taken before treatment | ANCOVA |
| How does one measurement change with another? | Two or more measurements | Regression or correlation |
| Do several measurements differ together? | Several measurements per animal | MANOVA |
The same animal or plot measured more than once
| Your question | Layout | Use |
|---|---|---|
| One factor on whole units, another within them | Ration per steer, aging time per cut | Split-plot |
| Do treatments differ over time? | Each animal on one treatment, weighed repeatedly | Repeated measures |
| Each animal gets every treatment in turn | Periods with a washout | Crossover |
| How repeatable is a trait? | Several records per animal, no treatment | Repeatability |
| Average growth curve, and how much animals differ in slope | Each animal measured at several times | Random coefficients |
| Heritability and the best sires | Progeny records by sire (and dam) | Sire evaluation (BLUP) |
Outcomes that are not measurements
| Outcome | Example | Use |
|---|---|---|
| Yes or no, overall rate | Dystocia rate in a herd | One proportion |
| Yes or no, by group, no model | Dystocia by sire breed | Two-way table |
| Yes or no, with treatments, blocks, or a dose | Pregnancy by protocol and herd | Logistic or probit model |
| Counts | Fecal egg counts, lesions, piglets born | Count model |
| Ordered scores | Locomotion score 1 to 5 | Ordinal scores |
Before the study
To find how many animals you need, use Power and sample size (the Plan a study button).