Week 5 - biases and pitfalls

a summary describing the population (the truth)

estimate of the truth, subject to error. Samples give incomplete data

More replication is good so long as the replicates are true replicates Pseudo replication (artificially increasing sampling size maybe by treating sub-samples as replicates) gives standard error and p values that are too small Conclude significant difference between groups when there is no difference Type I error - reject the null hypothesis

treatment has no effect on response variable.

sometimes multiple sub-samples are averaged to make a single sample (multiple leaves one one tree)

one complete set of treatments and one control

Comparing two means - see if there is a difference between them Large t value is more likely to give a small p value Ideal to have equal sample sizes (maybe decide to have an unbalanced design and have one group large and one small) Unbalanced can mean larger overall sample which outweighs small but balanced samples