Lecture 1B - Sampling, logistics, choosing

an experimental design
Sampling, logistics
1. Experimental design - why bother?
biologists need a plan
dealing with variation, especially in heterogenous environments
sampling, logistics, efficiency
making the science robust

2. Sampling - why bother?

It is usually unrealistic and unnecessary to measure 'everything' within a
biological system

Biotic: Populations

e.g.

width of daisy flowers

length of mussels
enzyme activity of yeast
grazing effects in field

Abiotic

concentration of reactant
soil moisture on hillside
pH levels of stream
light levels within forest

... but we need to ask: how, where, when, how many?

Oversampling is poor practice because:

1. we don't gain more information after a certain point

2. it wastes time, effort, resources
3. it can result in being overwhelmed by data
4. it can damage vulnerable environments
5. for 'destructive' sampling it wastes/removes biological material
6. it could be unethical, even illegal
3. Sampling Strategies
Random sampling

each part of the area being investigated has an equal chance of being
included in the sample

main options are:

Haphazard/arbitrary:
not truly random
Randomly-chosem coordinates, located by:
approximation
grid
using excel to randomly select
Randomly-chosen by:
pre-labelling

Stratified random sampling

Using a grid - powerful exp, design, which can avoid effects of confounding
factors
equal 'sampling effort'

Systematic sampling
Sub-sample whole population

4. Replication, repeats, pseudo-replication

Repetition of a measurement or sample:

Replication spatially seperate

each repetition is a replicate (n)

Replicates

can be repeated: over time

 BUT biological systems change over time!
can be repeated: within class

BUT may introduce bias

Psuedoreplicates

Spatially or temporally connected

Reproducibility, replicability, or repeatability

being able to repeat an experiment and get similar/the same results

Replicates

Need to be:

independent
unbiased
'representative'
sufficient in number to capture variation within the group being
investigated
able to help assess the reliability of the experiment procedure

5. Sample size

6. Sample Bias
7. Logistics
Time taken to move between sampling points introduces another factor

Experimental Design
1. Types of variables
2. What type of experiment?

Confounding factors?
attributing effects on Y?
 variation in Y?

making science robust

4. Controls
A control is a 'reference point', without which a comparison might be
difficult or meaningless
A control is identical to the treatment(s) in all respects, except for the
factor being investigated

Negative control

not expected to cause a difference

no added conditions/manipulation
treatment: 'blank', placebo

Positive control

expected to cause a difference

is manipulated
 treatment: known, well-established effects

When controls are not needed

'Observation studies':
natural groups
natural gradients
within-subject comparisons ('before/after')
repeated measures ('longitudinal' studies)
factorial studies
some pilot studies

When the control becomes the treatment?

HA: water is essential for plant growth

5. Experimenter effects
= another confounding factor
 bias - researchers expectations influence the outcome e.g. confirmation
bias
intentional
unintentional
'demonic intrusion' - researcher's actions influence the outcome

Blind
subject does not know ID of treatment

Double blind
subject and experimenter do know ID of treatment

6. 'Classic' exp. design

Between-group comparisons

Losses (e.g. deaths) during experiment is a risk for all exps. with small
sample sizes

Independent groups

all EUs randomly allocated to one of the groups

Advantages

Fully randomised
Quick, easy especially if variation in EUs is uncertain/large
Disadvantages
 Chance of unequal
Distribution, especially with low replication

Matching Pairs

similar EUs paired then allocated one group

Advantages

Stratification, blocking
Powerful exp. design
Known variation of EUs equally distributed
Disadvantages
Time-consuming to set up

Repeated measures

all EUs randomly allocated to one group, then other

Advantages

Distributes variation of EUs equally over both treatments

Disadvantages
'Order' effect; the sequences of treatments may have an effect
lingering effects

Random block design

used to eliminate/'cancel out' a possible confounding factor

Multi-factorial designs