BIO3011 Lecture Notes - Lecture 2: Test Statistic, Multiple Comparisons Problem, Inverse Trigonometric Functions
31 May 2018
School
Department
Course
Professor
3. Experiments
• Controlled experiment (spatial control):
o Uses treatments and controls
o Only one variable is altered
o Eg. examine lizard activity in an enclosure with a heat lamp vs control
• Perturbation experiment (temporal control):
o No control is used
o All variables are kept the same except one that is varied or manipulated over time
o Eg. what happens to lizard body temp if ambient temp is changed
• Natural experiment (natural control):
o Attepts to idetify treds usig treatets ad otrols deteried y nature
o Eg. do lizards living in rocky habitats have different activity to forest?
• By reducing the effect of confounding variables, the experimenter can isolate and identify the
effects of the treatment variable
• Confounding variable:
o Confounding variables co-associate with or might have complex relationships with a
variable of interest
o Difficult to identify without an experimental approach
• Ways to reduce sampling error and bias:
• Generally more replication is always better but with enough replicates all differences will
eventually become significant because the p value depends on the degrees of freedom
• Balance = all treatments and controls are equal in number
• Blokig = ioles assigig treatets radoly to repliates ithi loks
- a block of replicates share similar properties
• Paired designs = treatments are applied to replicates so that each replicate contains one
spatially associated treatment and control
- often more powerful because less noise will interfere with the treatment response
• Sub-replicates = multiple samples taken from the same replicate
-usually averaged to provide a final replicate result
-eg. multiple leaves from a single plant
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
A block of replicates share similar properties: paired designs = treatments are applied to replicates so that each replicate contains one spatially associated treatment and control. Often more powerful because less noise will interfere with the treatment response: sub-replicates = multiple samples taken from the same replicate. Usually averaged to provide a final replicate result. Experimental design pitfalls: nested designs, difficult to analyse, where treatments are nested inside other treatments, eg. examining light then having soil a and b in each light, pseudoreplication: using sub-replicates instead of replicates in analysis. Be aware of independence: pseudo-randomisation: using haphazard allocation of units to groups. Use a randomisation technique: multiple comparisons: because a significance test has a 1 in 20 error rate built into it, the more comparisons you do the greater the risk of getting an error by chance. If you run 20 comparisons you have a 64% chance of one or more errors.
