Experimental Design

Bio300B Lecture 4

Richard J. Telford (Richard.Telford@uib.no)

Institutt for biovitenskap, UiB

15 September 2025

Starting with a hypothesis

  • Proposed explanation for observations
  • Testable

Null hypothesis H0 - no relationship
Alternative hypothesis H1 - relationship

Testing a hypothesis - types of evidence

  • Observational
  • Computer simulations
  • Experimental
  • Meta analysis and systematic reviews

Which is better? Why are all needed?

Dead Calluna vulgaris

Correlation != Causation

r = 0.95

Drought experiment

Controls

  • Turf transplants
  • Warmer/Wetter

Bad controls

After rotenone treatment have started, you want to test how the benthic fauna of a river is affected by the treatment.

Conditioned samples

H0 : The parasite Schistocephalus solidus does not affect food intake of three-spined sticklebacks.

Design: You collect infected and uninfected fish from the field and compare their food intake in the laboratory.

Systematic errors

Example: You want to test the yield of five different types of barley to determine what grain type is the best.

Design 1: Each grain type is planted in separate areas of the field.

Systematic errors

Design 2: Each grain type is randomly planted in several squares of the field to avoid systematic errors

Randomisation is the cure for systematic errors.

Block designs may also be useful.

Replication

One treatment and one control measurement not enough

  • Random variability
  • Need replication
  • Power analysis to estimate how many replicates needed

Randomisation

  • reduces bias
  • stratified random designs possible

Unbalanced design

Is an experiment with 100 treatment subjects and 10 control subjects valid?

What about 57 treatment subjects and 53 control subjects?

Why might you get an unbalanced design?

Pseudoreplication

Want to test the effect of different fish foods

Mass of each fish measured.

Can each fish be treated as an independent observation in your analysis?

Confounding factors

H0: age of a fish affects parasite resistance because host age affects the immune system.

But, size of a fish varies with age.
Size may also affect the immune system. Thus, size may be a confounding factor in the experiment.

Blocking

Blinding

  • Two videos: pigs bred for social breeding value, controls
  • Count positive and negative interactions

  • Same pigs. Video inverted.

Examples

Are there any problems with the following study designs?

Study 1

You want to test growth of salmon depending on three different food types. You have three aquaria with several fish and give one food type to each aquarium.

Study 2

You want to test if a virus enlarges the liver cells of infected salmon. You measure 60 cells, 30 from a fish you have infected and 30 from an uninfected fish, respectively.

Study 3

You have a D. pulex population in an aquarium and use these animals for an experiment. You place the daphnia individually in 70 ml jars, first the ones you will use as a treatment group, and then the control group. Then, you randomly place control and treatment jars in a climate chamber.

Conclusions

Design the experiment/observations to test your hypothesis

Consider how you will analyse the data when designing the experiment