Experimental design Flashcards
What types of hypothesis can you have have in fMRI?
- Haeomdynamic hypothesis. E.g., Condition A will result in a greater BOLD signal from region X than condition B. As the BOLD signal is what is directly measured, this type of hypothesis is most direct and can be causal.
- Neuronal activity hypothesis. E.g., Condition A will result in an increase of activity in the amgydala as compared to condition B. This is a correlative hypothesis, as BOLD signal is only a proxy for neural activity.
- Pyschological hypothesis. This is the most complex, the most inferential and the most difficult to prove. It would typically have to be corroborated with other types of data - and is only as strong its underpinning psychological theory.
What are the benefits and costs of within-subject design?
Less subjects need to be used to reach statistical significance, as there tends to be less variation within subjects than between subjects.
The downsides are that for experiments that require multiple runs, or have different levels of testing, the same subjects will have to come back and be scanned many times. There are many problems with this - from habituation, to detection of deception within paradigms, to learning effects, to drop out etc.
What are the benefits and costs of between-subject design?
Different groups can be used for different conditions, therefore there are less issues associated with repeated scanning. You may need to compare healthy to patient groups as part of your hypothesis.
More people are needed due to intersubject variation. It has to be determined whether the results are truly due to the independent variable, or due to some group difference (unless the independent variable is the group difference e.g. ADHD v.s. control)
Groups must be randomised and counterbalanced. Groups should be made as similar as possible for different confounding variables such as gender, age etc.
Interstimulus interval
The amount of time between stimuli
Stimulus Onset Asynchrony (SOA)
The amount of time between the start of one stimulus and the start of the next
Control condition issues
Your control condition should ideally be identical to the task condtion except for the independent variable that you are manipulating. This can be difficult to actually achieve.
For example, when determining activation in response to faces, a cross hair would not be an appropriate control. It is different in many ways - it shows activation of DMN, it doesn’t have similar visual features to a face etc.
Instead a scrambled image that is matched to faces in terms of frequencies, luminous variance etc. This is a good control in some respects - but it also cannot be named as recognisable object, which may be a problem depending on your hypothesis.
Instead, inanimate objects could be shown.
Types of stimulus presentation in block design.
Conditions could alternate with one another, showing task A then task B etc. This is useful for indentifying differently activated regions, but it will not show commonly activated regions because BOLD fMRI is a relative measure.
Instead they can alternate with resting task and well as two task conditions. This allows the individual response to both A and B to be identified and compared with one another.
Considerations of interval length in block design
The idea of block design is to maximise the difference between conditions, therefore haemodynamic response should be able to to return to baseline in between stimulus presentation.
However, it also needs to be timing efficient so that more runs can be done.
Block length can be increased to increase the SNR.
10s is a good balance
Limitations of inferences of BOLD increase/decrease
They are relative to one another so we cannot truly know if we see an increase or a decrease in decrease
Event related design: advantages and disadvantages
Some advantages
• Estimation of the shape and timing of the haemodynamic response • Same events can be analysed in different ways • Sorting events by long-term consequences
Some issues
• Detection power worse than blocked designs
Inefficient designs:
- Co-occurrence of processes • Processes follow too closely in time
- Poor at evoking response of interest; participant engagement • Poorly matched to experimental hypothesis
Experimental efficiency
The power of an experimental design to test a hypothesis. An efficient design can reject the null hypothesis even with only a small effect.
Definitions of a contrast
- Differences in intensity between quantities measured by imaging
- The physical quantity being measured e.g. T1 weighted imaging.
- A statistical comparison of the activation activated by two or more conditions
Subtraction
This where two conditions differ only in the independent variable. Thus the effect of the IV can be made by subtracting the dependent variable measurements from one another.
If the conditions differ in more than one way, then the experimental effect cannot be said to be necessarily due to the IV.
Any factor that varies with the IV is called a confounding factor. An important confounding factor to be considered are hidden causal factors, such as reduced performance.
Randomisation
This minimises the impact of confounding factors by ensuring that they vary randomly and thus do not correlate with the independent variable
Counterbalancing
This balances confounding factors by ensuring they have equal effect in all conditions, such as fatigue
