Sampling Design

Population - the collection of all possible sampling units

Sample population defined in terms of spatial, temporal and ecological attributes

Variation is everywhere

Good experimental deisgn minimises effects of random variation = increased precision

Quantify uncertainty as often measure a sample of population of interest

Estimate is of no value w/out some statement of uncertainty of estimate

Replicate samples - they may show variation

Scope of inference - population to which inference (conclusions) can be reasonably drawn, based on study

Scope of inference depends on sampling unit for replicate observations - only 1 sample site, only infer one region (otherwise not representative)

Increasing independent replicates increases randomisation which increases representation of a population

Simple random sampling - choosing sample sites where every possible sample that could be selected has predetermined probability of being selected

Stratified sampling - divides population to sep. groups, factor in treatments and variability in environment

Cluster (paired) sampling - link in treatment and non-treatment samples together. Good for paired t-tests

Systematic sampling - along transects (shows gradients)

Even if plots randomly located, some might be close to each other = 'auto-correlated'

Affected by same factors if in similar areas = not independent

Samples must be spatially independent

Regular patterns of individual type location are used to:

• Maximise possible distance
• Minimise spatial dependence

Danger of sampling patterning = samples not independent, affected by similar things

Latin square design = good but not if there are gradients in the site

Block design = great to get around gradients in site - each block has patches with all different treatments in

Pseudoreplication - use of statistics to test for treatment effects w/ data from experiments where treatments not replicationed OR replicates not statistically independent

Taking multuple samples from same area/experiment is not replication

Non-repeated before-after-control impact experiments are based on pseudorep. and should not be used in experimental setups - should run control and test experiments AT THE SAME TIME

Precision - measure of scatter, dispersion or replicability of measurements

• Measurements show variation - precision reduces this
• Low precision = noise

Accuracy - extent to which measurements are reliable estimate of true value

• Links sample to population
• Random errors and systematic bias reduce accuracy

Reliability - (subjective) refers to interpretations but sometimes to measurements

• Affected by precision and accuracy
• Depends on vailidity of assumptions made in measurements/calculations

Replication reduces reliability and increases precision

Used to quantify uncertainty - how likely event occurs

Probability distributions - function describing likelihood of occurence of possible experiment outcomes

Small number of random (chance) events = likely different outcome then expected - low accuracy 

Larger number of events = outcome of sample more likely to be closer to true (population) value = law of large numbers

When calculating probability consider if sample is with or without replacement