Lecture 5 - Introduction to Experimental Design

Experiment

A procedure where a researcher intentionally changes some variable to observe the effect of its action

Experimental Unit

The smallest portion of experimental material which is independently perturbed, the item under study for which some variable is changed (the subject)

Observational Unit

The smallest unit of which a response is measured. They are what the measurements are actually from within the experimental unit.

If there are multiple measurements/samples from one experimental unit, the samples are all observational units (this also shows pseudo & technical replication)

Observational Data

Not experimentation because we aren’t intentionally changing some sort of factor

Treatment

Also known as the independent variable or factor, is an experimental condition independently applied to an experimental unit. It is controlled by the research and there can be different values or levels of the treatment

Dependent Variable

Also known as the response, is the output that is measured after an experiment. This is what the researcher will measure and assess if changing the treatment/s induce any change.

An effect is the change in response variable caused by the independent variable. The magnitude of this effect is significant based on the researcher’s level of confidence, significance values, and so on.

Biological Replication

Each treatment is independently applied to each of several experimental units (humans, plants, or animals). This allows for generalisation to the population.

Technical Replication

Two or more samples from the same biological experimental unit source which are independently processed. Advantageous if proessing steps introdue a of variables. It also increases the precision with which comparisons of relative abundance treatments are made.

Pseudo-replication

One sample from the biological source is divided into two or more aliquots which are independently measured. Advantageous for noisy measuring instruments. It also increases the precision with which comparisons of relative abundance treatments are made.

Randomisation

A way to protect against biases by planning and desigining the experiment in a way that the variation caused by extraneous factors can all be combined under the general heading of ‘chance’.

It ensures each treatment has the same probability of getting either good/bad units, thus avoiding systemic bias.

It also cancels our population bias, ensuring any possible causes of experimetnal results are split equally between groups.

Blocking

Another way to control variability by making treatment groups more alike by dividing them into blocks so that units within the same block are more similar to each other than units from other blocks.

It deals with nuisance factors that may have some effect on the response, but is no interest (age, race, diet, etc)

Completely Randomised Design

When the experimental units are divided randomly into treatment groups where each group is also randomly assigned one treatment level.

Randomised Complete Block Design

It is similar to the previous, but treatment gorups are split into blocks before assigned a treatment. There will be b amount of blocks for how many ever t treatment levels there are. In the end, there should be b x t amount of experimental units.

We’d want the experimental units within each other to be as homogenous/similar as possible to avoid any unwanted variation.

Factorial Design

When there are two or more sets of treatments. Rather than studying each factor separately, all combinations of treatment factors are considered.

They only infer if any interaction effects, when the effect of one variable depends on the value of another variable, exists. If one does exist, the effect of one factor on the response will change depending on the level of the other factor.