Week 10: Decision Making Flashcards
what is decision making?
- process of making choices between alternatives
- Process of transforming stimulus representations into specific behavioural patterns
what is perceptual decision making?
= judgement of the identity (or any other discrete or continuous sensory feature) of a stimulus
What are discrete vs continuous sensory features of a stimulus?
Discrete: identity, sex, orientation (upside/ inverted),…
Continuous: Color, orientation (Gabor 1-180degree), tempo of music
What is the signal detection theory?
- tried to understand how we make decisions under uncertainty (e.g. when stimuli can be easily confused)
- Not a theory of decision making, but a model
- We can break down influence of stimulus signal vs noise by using the theory
- Provides a precise language and graphic notation for analysing decision making in the presence of uncertainty
What are examples for (perceptual) decisions under uncertainty?
- disease or no disease (medical)
- Green light or not in fog (real life)
- Detect plane or flock of birds (radar operator)
What are Signal and Noise in the context of SDT?
- Signal („Target“) = what you are trying to perceive, the variable of interest
- Noise („distractor“, „lure“, „foil“) = anything else that could look like a signal
What are Hit, Miss, False Positives, Correct Rejection in the context of SDT?
Hit: there was a signal and a response
Miss: there was a signal and no response
False alarm: there was no signal and a response
Correct rejection: there was no signal and no response
- What is Sensitivity in the context of SDT?
= discriminability → how well can you tell apart signal from noise
- more discriminability → more hits and correct rejections - Less discriminability → more misses and false alarms
How can we increase discriminability?
- manipulate signal
- Manipulate noise
How can d‘ be determined?
d‘ = Z(hit rate) - Z (false alarm rate)
What is Decision criterion in the context of SDT?
Criterion = bias → determines what kind of errors you make
- high criterion (conservative) → less hits (more misses) and less false alarms
- Low criterion (liberal) → more hits (less misses) and more false alarms
- Has nothing to do with sensitivity → different decisions with same information
What does the decision maiking process look like according to SDT?
Step 1
Specify your decision in terms of internal response to signal and noise
Step 2
specify your threshold for making the decline
Decide what type of errors you want to make
→ criterion can be shifted higher to minimise false alarms or lower to minimise misses
What does the number and what does the type of misclassifications in SDT depend on?
Number: overlap of distributions of internal response to signal and noise
Type: strategy (criterion)
What is the ROC curve?
- ROC = Receiver Operating Charatcerstic
- A visuallization of d‘ (discriminability)
- ROC curve is sensitivity as a function of false positive rates
- All points on curve correspond to one d‘ with different criterion values
- Criterion determines point on the curve
- Shape depends on discriminability
What is the AUC
- AUC = Area under the curve
- A measure of accuracy
- Accuracy = proportion of Hits + correct rejections out of all trials
- Related to sensitivity (more sensitivity → more accuracy), but not the same
- Used to compare different model
- Non-parametric: useful in more complex cases
What are evidence accumulation models?
- evidence in the noisy environment is sampled over time (in SDT evidence stays the same)
- Decision has two parts:
- Which hypothesis is more in line with evidence?
- When to stop the sampling process and commit?
What is the random dot Kinematogram?
- Dots moving coherently in one direction against a background of dots moving in random directions
- Coherent dots = signal
- Incoherent dots = noise
- Motion coherence ( proportion of signal to noise)= strength of evidence
- information is spread out across time → study of accumulation of evidence across time
- Threshold = coherence level at which the correct direction can be identified on a given proportion if trials
What does accumulation to threshold look like?
- subject accumulates evidence for one or other alternative at each step
- Evidence is integrated until a decision threshold is reached
- Higher coherence → faster reach of decision threshold (steeper drift rate) and faster RT
- Strategies manipulate („respond fast“ vs „respond accurately“ )decision thresholds
- Includes: sequential sampling models, evidence accumulation models, race models
What effect does stimulation of input or output areas have on accumulation models?
- stimulation of input areas (sensory areas/MT) → increase in drift rate
- Stimulation of output areas (LIP/IPS) → upward short of the curve (biased starting point)
What is the RDM task?
- study by Shadlen and Kiani
- Macaques see a RDK and saccade to a target post presentation
- Receptive field of cell in MT covers area of stimulus location
- Receptive field of cell in LIP (parietal area that prepares saccades) covers target location
- The more evidence collected in MT that movement is to one direction, the higher the activity in corresponding LIP cell
what is neurometric vs psychometric information?
Neurometric : neural
Psychometric: behaviours Neurometrci curve is more sensitive
Evidence accumulation models
- MT codes momentary evidence
- LIP codes evidence integrated across time
- Signals in MT are noisy → integration in LIP is stochastic
- Accumulation (drift) rate is proportional to strength of sensory evidence → higher coherence the faster the accumulation
- Decision is made when accumulated evidence reaches a decisions boundary
- The DDM also provides an explanation for behavioural reaction time distributions
