Introduction to Computational Modelling

• Data never speak for themselve
• Need a framework to interpret data
• Verbal theorizing is not a substitute for quantitative analysis 
• Always multiple models to select from 
• Model selection based on quantitative and qualitative judgement 
• Psychology is moving towards this specification 
• Number of publications using a computational model has increased hugely 
• A quantitative model ensures that all assumptions of a theory have been identified and tested 

Ptolemaic model 

• Earth is at the centre of the universe
• Only plots a few of the planets 
• Sun orbits around the earth

Copernican model 

• Projection from earth to mars will end up tracing out the retrograde motion 
• At different points in time because of the relative speed
• Earth appears to overtake Mars 
• Mars will end up moving ahead again  
• Replaced the Ptolemaic model because it was simpler and involved the fewest assumptions 

Kepler’s model

• Assumed that the planets had elliptical orbits rather than circular 
• Could predict the location of planets accurately
• Additional power to this model can be attributed to quantitative fit 

Data description 

• Describes relationship between variables 
• E.g. mathematical function relating study time to course marks 

Process characterisation 

• Attempt to peek inside the black box (the mind or what processes occur in the mind)
• Explanatory power lies in hypothetical constructs of the mind rather than with the data 
• Remain neutral in how the implementations of processes in the black box are happening 

Process explanation

• Provide an up close view of what’s inside the black box 
• Attempts to implement how processes occur so cannot remain neutral 
• Can’t just say that I and R influence recall and then estimate the parameters from the data 
• The constructs must be computed from the model’s architecture 

• Process characterisation model of recall 
• Propose two different mental processes 
• I – the probability that a memory trace is intact
• R – the probability that the memory trace can be recreated from a less than intact structure 
• The probability of correctly recalling an item in a memory = I/1 – I x R 
• Error choice = 1 – I

Hulme et al (1997)

• Applied the MPT model to the recall of words with varying frequency 
• Assumed that R is greater for high frequency words and I is greater for words with earlier serial positions in the list 
• Easier to fill in the blanks with high frequency words 
• Predicted that high frequency words recalled more than low frequency which becomes more pronounced over serial position 
• Word frequency influenced recall but mostly for later serial positions 

Strengths

• Bring out relationships between experiments or sets of data that we would not have otherwise perceived 
• Multiple models for different subclasses of phenomena?
• Emergence of understanding
• Allow us to explore various implications of human behaviour and cognition 
• Like humans, they can learn
• Make precise predictions

Weaknesses

• Computational models are computer programmes so may only do what they are told 
• Will models never generate anything truly novel?
• Variable data
• Verisimilitude – enough truth about the model (partial truth value) 
• Can continue to use false models as long as they have verisimilitude