Week 4 Flashcards
What is an experiment?
an experiment there is manipulation of one variable, usually while keeping everything else constant
What do experiments provide evidence for?
Experiments provide some evidence of cause and effect
What is an observational study?
Illuminate patterns (correlation or associations between variables), but are unable to fully disentangle the effects of measured explanatory variables and unmeasured confounding variables
What is a confounding variable?
a variable that masks or distort the
causal relationship between measured variables in a study
What is the limitation of observational studies?
An observational study might not be able to pull apart confounding variables because they will co-occur
Is correlation symmetrical?
yes both variables will be correlated with each other
Is Causation symmetrical?
no
– Smoking causes heart attacks, but heart attacks do not cause smoking
What is the goal when designing a experiment?
to eliminate bias and reduce sample error.
What is bias?
a systematic discrepancy between the
estimate you would obtain, IF you could
sample a population again and again, and the true population characteristic
What is sampling error?
The difference between an estimate and the population parameter being estimated caused by chance
What does reducing sampling error achieve?
increases precision
What do control groups do?
Eliminates bias
What does Randomization do?
Eliminates bias
What does Blinding do?
Eliminates bias
what does Replication do?
Reduces sample error
what does Balance do?
Reduces sample error
what does Blocking do?
Reduces sample error
what do Sub-replicates achieve?
Reduces sample error
Paired designs purpose?
Reduces sample error
•Consistency of conditions purpose?
Reduces sample error
What is a control group?
A simultaneous control group is a set of subjects that are treated in the same way as the treatment group in all ways except that the treatment is not applied
What will happen with randomization?
Randomization will roughly equalize confounding factors
The confounding factors will still exist, but the effect will not be skewed
What does Pseudo-randomization do?
almost always generates bias
definition of blinding
Blinding is when either (or both) an experimenter and subject are unaware of treatment and control
definition of replication
Replication is the term we use for running multiple treatments and controls at the same time
What is the importance of having a greater number of replicates?
Studies with a greater number of replicates will have a larger n and a smaller standard error as a result
What are replicates?
Replicates are not the number of plants or animals used, but the number of independent units
What is the definition of balance?
A study design is balanced if all treatments and controls are equal in number
What is the definition of Blocking?
Blocking involves assigning treatments randomly to replicates within ‘blocks’
A block of replicates share similar properties.
Differences among treatments are evaluated within blocks.
Paired design definition?
In a paired design the treatments are applied to replicates so that each replicate contains one spatially associated
treatment and control.
what does paired design do?
By pairing treatments and controls, a lot of environmental noise is reduced.
What are Sub-replicates
Sub-replicates is the term for multiple samples taken from the same replicate.
How are the results calculated when using sub replicates?
Usually the results of sub-replicates
are averaged to provide a final replicate result (data).
Examples of sub replicates?
- Multiple leaves from a single plant
- Multiple animals sampled at a single forest site
- Multiple plugs from a single bacterial plate
- Multiple aphids sampled from a single crop field
What are the reasons you may want to use extreme treatment?
• Treatment effects easiest to detect when the effects are large
• Small differences can be difficult to detect and require
larger samples sizes
What is the limitation of extreme treatments?
the effects of large dose may be different from the effects of smaller, more realistic
doses
What should extreme treatment be used for?
Use as first step in detecting an effect of one variable on another
What is a factor?
single treatment variable
Why might you want to do an experiment with more than one factor?
Factors night interact to affect the response variable in a way that you would not expect/see if you were to test each factor on its own
What does an experiment with Factorial design?
examines all treatments combinations of two or more variables
– It also can measure interactions between the treatment variables
– Interaction between two variables mean that the effect of one variable depends upon the other variable
What is Pseudoreplication?
Using sub-replicates instead of replicates in your analysis.
What is the consequence of Pseudoreplication?
This inflates the n above what it should be and (sometimes vastly) increases the chances of significance.
What is Co-correlation?
Two or more explanatory variables are correlating or ‘cocorrelating’ in a sample. This means the variables are (probably) measuring the same thing. i.e. ‘lighter carrying’ and ‘cigarette carrying’ are actually both measuring the behaviour of smoking.
What is the problem with too many explanatory variable?
If you have n samples and n explanatory
variables then your explanatory variables will explain all the variation in your data because there is one variable per data point. This isn’t informative.
