Statistics Theory L4 = Quasi-experiment Designs Flashcards
Quasi-experiments?
= studies of semi/uncontrolled events.
Studies of semi/uncontrolled events attributes? (2)
- Real world adds challenges that require careful planning.
- Conclusions about broader inferences are limited.
Why are conclusions about broader inferences limited in such studies? (5)
- In place of a control, a similar “reference” area might be used.
- Random sampling is less common.
- Randomised experiments are very rare.
- Replication might not be possible.
- No control over the perturbation we want to study.
Egs of studies of semi/uncontrolled events? (2)
- Wind energy impacts on flying vertebrates.
- Diesel spill impacts on stream ecology.
Explain the eg of Wind energy impacts on flying vertebrates? (2)
- Non-random placement of turbines, and infrastructure.
- No formal treatments or controls.
Explain the eg of Diesel spill impacts on stream ecology? (2)
- No before-after.
- No comparison to unaffected sites.
Therefore, Quasi-experimental designs attributes? (2)
- Increased uncertainty about conclusions regarding cause and effect (due to no randomisation).
- Reduced ability to infer to a wider population.
Thing to note about the types of Quasi-experimental (QE) designs?
Progression of designs begins with more-rigorous approaches with more-certain conclusions, to less-rigorous approaches with less-certain conclusions.
Types of QE designs? (4)
- Before-after/control-impact design.
- Impact-reference design.
- Response gradient.
- Before-after design.
Before-after/control-impact (BACI) design?
= experimental units are randomly allocated to treatment & reference areas before a treatment is applied.
BACI attributes? (2)
- No true control area, rather a “reference” area.
- Presumes that we have some control over the treatment/impact.
Beware of 2 things with the BACI design?
- Response variables vary naturally in nature, and changes in the response variable could be unrelated to the treatment.
- No matter how carefully the comparison areas are chosen, they always differ (differences could be unrelated to the treatment).
Diagram/”Procedure” of BACI design? (3)
- Before treatment is applied, measure variables.
- Apply treatment.
- After treatment is applied, measure variables of interest.
Impact-reference design?
= compares measurements on a treatment area (>1 area).
Impact-reference design attributes? (2)
- We try to make comparison areas as similar as possible.
- After only, no before data.
Beware in Impact-reference design?
Differences could be unrelated to treatment, but rather related to the natural variation in the system.
Are differences between R (Reference) & T (Treatment) caused by treatment or natural variation?
We can’t say that the treatment caused the difference.
“Procedure” of Impact-reference design? (3)
- Treatment/Impact.
- Apply treatment.
- Measure variables of interest (covariates) after treatment is applied.-
Response gradient design?
= where we select areas that allow for measuring the level of an impact & the level of the response.
Response gradient design attributes? (2)
- Could involve measuring the distance from impact.
- Could involve measuring the time since impact.
Distance from impact?
= measuring along a spatial gradient, with distance from the impact.
Time since impact?
= measuring X and Y variables repeatedly over time since an impact.
Eg of how Distance from impact may look like?
Along a gradient, you have 0 1 2 3 4 5 6….km.
Eg of how Time since impact may look like?
Along a gradient, you have 0 1 2 3 4 5 6….weeks.
Beware in Response gradient design?
Response could be confounded with naturally-varying factors in the study area.
Before-after design attributes? (3)
- One experimental area is used/focused on.
- Before is the reference to which the impact/treatment is compared.
- The least certain.
Beware in Before-after design? (2)
- Change could unrelated to the treatment.
- Measuring covariates is very important to describe other potential changes that coincide with the impact.
“Procedure” of Before-after design? (3)
- Reference area before treatment is applied.
- Apply treatment.
- Measure covariates after treatment is applied.