Experimental design

Question 1

Observational study

Answer 1

o In observational studies, the treatment is not under experimental control.
o Monitoring through time/ space

Question 2

Experimental study

Answer 2

o A designed experimental study has at least two treatments (e.g. a control and treatment) that are assigned to units/subjects
o Properly designed experiments allow causal effect of treatment on a response variable(s) to be evaluated (using statistics)

Question 3

Clinical trial

Answer 3

• A clinical trial is an experimental study in which two or more treatments are applied to human participants.

Question 4

Principles of experimental design

Answer 4

Reduce bias in estimating/testing treatment effects using:

Control groups
- control groups receive no treatment of interest but experience similar conditions to treated subjects

Randomization
- The random assignment of treatments to units
- Aim is to eliminate/reduce effects of confounding variables
- Methods: random number generator, block randmization, stratified randmization

Blinding
- Neither experimenter nor subjects know treatment assignments

Reduce sampling error using:

Replication
- apply each treatment to multiple, independent experimental units
- larger sample sizes  but do not remove error

Balance in numbers between treatments
- All treatments have equal sample sizes

Blocking
- Divide experimental units into groups/ blocks that share similiar features.
- Randomely apply treatments within blocks.
- Allows variation due to differences in blocks to be discarded.
- Also reduces bias

Question 5

Factor vs factorial design

Answer 5

Factor – single treatment variable whose effects are of interest to the researcher

Factorial design – investigates all treatment combinations of 2 or more variables + can measure interactions between treatment variables
- Interaction – effect of one explanatory variable depends on state of the other variabl
- Using extreme treatments  power of an experiment to detect a treatment effect

Question 6

Experimental design for observational studies

Answer 6

Observational studies should employ as many of the strategies of experimental studies as possible to minimize bias and limit the effect of sampling error.

Randomisation not possible so use …
* Stratified sampling – partitioning experimental units into homogeneous sub-groups
* Matching – matches exp units using like-for-like approach to minimise erroneous variation (e.g. treatment to similiar control)
* Adjustment – uses statistical methods to correct for differences between treatment and control groups eg. use different stats test if unequal sample sizes

Question 7

Sample size: plan for precision

Answer 7

Must:
- decide how much uncertainty can be tolerated (margin of error)
- Estimate the standard deviation

Equation: n=8(p/MOE)^2

MOE= 1.96 x SE

Wider margins of error = can afford smaller sample sizes
With higher within sample variance = need larger sample sizes for any given margin of error

Take result w/ pinch of salt as we are estimating pop’n sd – even if estimate were correct which is unlikely our sample sd (s) would likely be different to actual sd due to sampling error

Question 8

Sample size: plan for power

Answer 8

Must:
- Decide minimum difference between null and alternative (D)
- Estimate the standard deviation

Equaton:
n=16(sd/D)^2

The smaller the minimum difference then the larger the sample sizes