Lecture 1- Computational Models Flashcards
why do we model data
data doesn’t speak for its self
verbal theorising not a substitute for quant analysis
always multiple models to select from
model selection based on quantitative and qual judgement
not enough to have just a description
what is meant by demystifying in computational models
models seem difficult with all their parameters
but the computer programmes do ALL of the maths
what is a model
At a basic level a model is an abstract framework capable of capturing the structure in the data.
eg mean or a regression slope
A good model for a set of numbers could be its mean
Models should also be a simpler version of what they are trying to explain.
No point in replacing one thing you don’t understand with another
what is data description according to Lewandowsky and Farrell (2011
data description= describes relationship between variables (eg linear line)
what is process characterisation according toLewandowsky and Farrell (2011)
- an attempt to peak into the ‘black box’ - mind
unlike descriptive models their explanatory power lies in hypothetical constructs of mind rather than the data
they remain neutral i respect to specific implementations of how they process and characterise
what is process explanation according to Lewandowsky and Farrell (2011
- provides an up close view of whats in the black box (mind)
- mus implement HOW process occours- cant remain neutral
give some reasons behind why we model
- makes data simpler
- verbalising theories alone cant substitute for quant analysis
- data never speaks for itself- requires a model to help understand it
- There are always several models to choose from
Model selection based on quantitative and intellectual judgment - Even seemingly intuitive verbal theories can turn out to be incoherent or ill-specified
- Only instantiation of a quantitative model ensures that all assumptions of a theory have been identified and tested
- All researchers use models whether they like to admit it or not.
- “Formal (i.e., mathematical or computational) theories have a number of advantages that psychologists often overlook. They force the theorist to be explicit, so that assumptions are publicly accessible and reliability of derivations can be confirmed…” (Hintzman, 1990)
- Verbally expressed statements are sometimes flawed by internal inconsistencies, logical contradictions, theoretical weaknesses and gaps. A running computational model, on the other hand, can be considered as a sufficiency proof of the internal coherence and completeness of the ideas it is based upon…” (Fum, Del Misser, Stocco, 2007, p 136; as cited in Lewandowsky & Farrell, 2011)
whats the difference between a model and a theory
models take the place of a theory
models can also be used to make predictions
how do process characterisation models differ from descriptive models
- attempt to peek into the black box
- explanatory power lies in hypothetical constaructs of the mind not just in data (don just explain things based on mean etc - about mind that relays this data)
- hoever these models remain neutral with respect to specific implementations of processes they characterise
explain process explanation
Explanatory models provide an up close view of what’s in the “black box”
Must implement HOW processes occur at the models level of specification
Can’t just say that I and R influence recall and then estimate these parameters from the data.
These constructs must instead be computed from the models architecture
what can we expect from models
“What we hope for primarily from models is that they will bring out relationships between experiments or sets of data that we would not otherwise have perceived.” Estes (1975 p. 271).
Multiple models for different subclasses of phenomena?
Light as a particle vs. wave
GCM exemplars vs. rules or prototypes
Emergence of Understanding
Since computational models are computer programs and computers only do what they are told, doesn’t it follow that models will never generate anything truly novel?
NO!
EZ-Reader and skipping costs
to be unlike humans
name a weakness of models
Scope and testability
We want our model to be falsifiable but not false!
“Regardless of their form or function, or area in which they are used, it is safe to say that these models all have one thing in common: They are all wrong”. (MacCallum, 2003 p. 114)
This is OK, take a deep breath. Incorrect models can still be very useful.
Kepler’s model of the solar system is based on Newtonian Physics and therefore…WRONG
explain Verisimilitude
Popper (1963)
Verisimilitude refers to a “partial truth value”
We know our models are false but can continue to use them so long as they have verisimilitude and…
Have “lots of money in the bank” (Meehl 1990).
Models and theories gain money by predicting facts.
