Experimental unit

For each case, identify:

• what receives the treatment
• the experimental unit
• the number of experimental units

Calves

Pig pens

Fish tanks

Treatment structure

Research question

Does increasing dietary protein improve growth rate in growing pigs?

A useful research question identifies:

• the biological system of interest
• the factor that may explain variation
• the response used to evaluate the explanation

Design elements

Treatment factor

The treatment factor is the explanatory variable deliberately manipulated in the experiment.

In the pig example:

• factor: dietary protein concentration
• standard level: the current diet
• experimental level: the increased-protein diet
• treatment: the condition applied to an experimental unit

Response variable

The response variable records the outcome used to evaluate the treatment effect.

For growth rate, the study must specify:

• the measurement used to calculate growth
• the period over which growth is assessed
• the units of measurement
• whether measurements occur at animal or pen level

No comparison

What does this tell us?

• it describes growth under the new feed
• it does not show what growth would have been under another condition
• it cannot estimate the effect of changing feed

Control group

Add a contemporaneous group of pigs receiving the current diet.

The control group provides:

• the relevant alternative condition
• an estimate of outcomes over the same period
• a basis for estimating the treatment effect

The appropriate control depends on the research question.

Rival explanations

The design changes the diet, barn environment and management at the same time.

Design check

Before interpreting an experiment, identify:

• the research question and biological hypothesis
• the treatment factor and its levels
• the response variable and measurement schedule
• the experimental units receiving each condition
• the comparison required to estimate the treatment effect

Which n?

Before accepting that value, ask:

• how the pigs are housed
• where the feed is delivered
• which units receive diets independently

Two kinds of unit

The units may be the same, but they do not have to be.

Subsamples

A subsample is an observation taken from within an experimental unit.

Subsamples can:

• describe variation within the experimental unit
• improve estimation of an experimental-unit response
• reveal individual responses within a treated group

Subsamples do not add independent treatment allocations.

Measurements or units?

Three quick cases

1. Six pens receive each diet. One mean growth value is calculated for every pen.
2. One pen receives each diet. Twelve pigs are weighed in every pen.
3. One cow is measured weekly for ten weeks.

Pseudoreplication

Common sources include:

• several animals sharing one treated enclosure
• repeated observations from the same animal
• several samples or technical measurements from one unit

Independent units

When evaluating replication:

• identify the experimental unit from treatment application
• count units that could receive treatments separately
• keep subsamples and repeated observations linked to their unit
• report the treatment sample size using independent experimental units

Music and hens

Your diagnosis

Work in groups. Complete the diagnosis before suggesting a repair.

Treatment structure

Two problems

Minimum repair

If music must be applied to a whole room:

• include several independent rooms for each sound condition
• randomly allocate rooms to classical music or no music
• measure hens within every room and retain their room membership in the analysis

If treatments can be isolated within smaller enclosures, those enclosures could become the experimental units, provided several receive each condition independently.

Random allocation

It:

• starts from the eligible experimental units
• prevents investigator preference from determining treatment groups
• gives each unit a known opportunity to receive each condition
• supports an unbiased comparison of treatment effects

Sampling or allocation?

Researcher choice

The groups may have differed before treatment because of:

• baseline anxiety severity
• owner expectations and motivation
• home routines or other care

Confounding

Confounding can arise when treatment levels occur:

• in different rooms, barns or tanks
• during different periods
• in groups formed by systematic choice

More measurements within the same experimental units do not separate the effects.

Diet or barn?

Design repairs

The allocation must create independent treatment comparisons at the experimental-unit level.

Surgery questions

For every proposed experiment, ask:

• What is the experimental unit?
• Could each unit have received any treatment?
• How was the random allocation implemented?
• Does another variable change systematically with treatment?
• What alternative explanation remains inseparable?
• Does the comparison support the proposed causal conclusion?

Sources

Ruxton, Graeme D., and Nick Colegrave. 2016. Experimental Design for the Life Sciences. 4th ed. Oxford: Oxford University Press. ISBN 978-0-19-871735-5. Publisher page.