Lecture 3 - experimental design Flashcards
Two broad types of sampling
-mensuration “natural” experiment -manipulation experiments: = ‘real experiment’
Mensuration
Involves making measurements on units (e.g., plots) where treatments are not applied.
manipulation
Involves making measurements on ecological units to which treatments have been applied. This is what most people think of as an experiment
Experimental unit
The smallest division of the experimental material that can receive any of the treatments.
Replication
-repetition of the basic experimental unit (EU) -each treatment is applied to more than one EU -Critical for statistical analysis
Replicates
-Independent experimental units (note: subsamples are not replicates) - subsamples cannot be elevated to EUs because artificially inflates replication (used to happen.. would inflate until null hypothesis was rejected..) -How many replicates are enough? n great than or equal to 2.. -Statisticians recommend -10 replicates!!! - very demanding.. Usually need to trade the number of replicates against total number of experimental units you can use. If can’t get as many EUs do you keep replicates? treatments? - classic dilemma.
Necessity of replication
1) to estimate experimental error = variability (e.g., S2)
2) insurance against the intrusion of chance events in experiments (= ‘noise”)
ex of noise… Insect outbreak in one of the experimental units (e.g., pots) set up in a green house
A good experimental design….
a) know what your experimental unit is - to avoid pseudo replication (which would lead to inflation of EU)
b) randomize - removes bias
c) Get as many replicates as possible - for good statistical analysis - determines variability & adds precision
d) Make sure there is a control
Experimental designs
1) Completely randomized design 2) Randomized block design 3) Latin-square design 4) Systematic design
Completely randomized design
-experimental units are randomly placed -not recommended when replicates are few -not knowing which cage is control etc..
Randomized block design
-EU grouped in blocks -Blocks are groupings of homogeneity (why most popular design) -Each block should contain as many experimental units as treatments. Treatments are randomly assigned within the block.. Ex of treatments.. no burn, low intensity, high intensity -this design corrects the possible affect of a single environmental gradient -Perpendicular to gradient
Randomized block design (recommended because..)
a) completely randomized designs can lead to all the units with a particular treatment grouped together - this can be disastrous! b) whole block can be lost without compromising the experiment
Systematic design
recommended but with caution - bc of possibility of natural pattern..
Latin Square
- two gradients are known to exist (moisture and pH) - latin square corrects effects of the two gradients
Rules for latin square
*All treatments must be present in each block. *Treatments must not be duplicated in a block. *Same # across as down
Pseudoreplicates
-Not independent -Needs to be avoided -Fake replicates… -Consequence of assuming that subsamples are independent of each other and come from independent EU’s
Schematic representation of acceptable (A) and unacceptable (B) experimental designs
- Unacceptable -> some form of pseudoreplication
A-2 & A-3 good if lots of treatments
B-1 very bad because everything on one side gets some treatment.. Not random - so are dependents.. Could have contamination due to proximity…
B-3 even worse.. two chambers.. Would be desirable if using volatiles, but concern is if contamination within samples..
B-4 possible they are dependent
B-5 - means single event that is occurring and each individual is a replicate

Examples of Pseudoreplication
1) Burn Experiment: A 1000 ha plot of forest is burned and another plot is not burned. You could then sample 15 10-m2 quadrats on each plot - very flawed.. puts you at B-2 bc never had a chance of being burned/not burned.. More replicates the better.
2) Exclosure Experiment: A 1 ha plot of grassland is fenced and an adjacent 1 ha plot is not fenced. You could then sample 50 1-m2 quadrats on each plot.
**Their are no replicates in either of the above manipulation experiments!!! There are only two EUs in each example and each quadrat is a subsample!
*n=1
Exclosure studies in BC
-Look over slides with notes!!

Solution to pseudoreplication
-Solution to pseudoreplication is to establish more experimental units (= burn plots and exclosures in our examples).
Note: establishment of more experimental plots adds considerable to the effort and cost of the experiment