week 9 - cognitive models of decision making

Question 1

What is speeded decision making?
When do we use sdm?

Answer 1

making decisions under time pressures
everyday use

Question 2

What some sdm paradigms?
Give exam of task question for each?

Answer 2

Choice RT tasks (can this animal fly?)
Flanker Task (is this central letter a consonant or a vowel?)

Letter comparison (Are these the same sequence of letters or not)

Question 3

What can you measure from sdm tasks/paradigms?

Answer 3

reaction time RT = speed
proportion of correct responses = accuracy

Question 4

What is the speed-accuracy trade off?
Does this change with age

Answer 4

-different participants prioritise speed or accuracy in sdm tasks
-yes, older adults prioritise accuracy whereas younger, speed

Question 5

What is the task in drift-diffusion model DDM? (conceptual model)
What is measured in the DDM?

Answer 5

stimulus presented and then you have to press a button as a response
reaction time = time from stimulus presentation and response/press button

Question 6

What happens during the RT in the conceptual DDM?
Thus what is the RT equal to?
How can this equation be rephrased?

Answer 6

you encode the stimulus information Te, then you make a decision Td, then you make the motor response Tr

RT=Te+Td+Tr

RT= non-decision time (Ter) + decision time (Td)

Question 7

If you had a bias to a certain response, where would the response line start?

Answer 7

starting point would not be in the middle (bias)

Question 8

In DDM, when do you make your motor response?

Answer 8

when you reach the boundary of a certain response (have drifted up to it)

Question 9

What does the boundary separation represent? how does it change?

Answer 9

speed accuracy trade off:
get smaller: you prioritise speed -> faster response, smaller Td

get larger: you prioritise accuracy -> more accurate answer, Td increases

both at expense of each other

Question 10

Is the drift rate unique to an individual?
Why are cognitive researchers interested drift rate?

Answer 10

yes -usually stays the same
drift rate in individual is linked to other traits like fluid intelligence

Question 11

Who introduced the DDM?
When?

Answer 11

Ratcliffe 1970s

Question 12

What is the aim of Counterstike?

Answer 12

5 versus five game: 5v5 (number in each team)
Idea of game is that terrorists have to plant bomb and antiterrorists have to remove it (or just kill all terrorists)

Question 13

Hyde et al., who was recruited for the CS go study?
What did they have to ask participants?
What were the paradigms used?

Answer 13

casual, expert, pro
total hours played, weekly hours of play, current ranking and how do you rate your playing yourself

Choice RT: is it green or blue with a switch task -> is it spiky or round? randomly added

Question 14

What groups did Hyde et al. use?
What are the low medium and high difficulty tasks used?

Answer 14

casual
experiences
aspiring
semi/professional

low - only one paradigm in each block
medium - mixed block but same paradigm repeated
high - mixed block randomly switch between

Question 15

Hyde et al.
Any significant results in RT between participant groups in low medium and high tasks?

Answer 15

low and medium = professional were sig. lower than all other groups
high = no difference between groups

Question 16

Which study investigated why expert gamer are faster than amateur/aspiring?
Which model did they use?

Answer 16

Voss et al. used a DDM using fast-dm-30 with maximum-likelihood MLE

Question 17

Hyde et al.
What was the result of the drift rate (v)?

Answer 17

v was significantly higher in expert compared to other groups but error bars are large -> not very convincing

Question 18

Hyde et al.
What was the result of the boundary separation (a)?

Answer 18

no different in boundary separation (a)

Question 19

Hyde et al.
What was the most significant result?
How does the DDM further investigate this result?

Answer 19

largest change in faster non-decision times (Ter) in expert group compared compared to all other groups
-helped just go beyond speed and accuracy measurement and look at cognitive processes -> now look at if its the Te or Tr which gets faster

Question 20

What were the conclusions from Hyde et al.?
However what was Hyde et al. critical and contradicting arguement to these conclusions?

Answer 20

faster decision making in expert gamers was due to these cognitive processes:
-no effect on boundary separation -> suggests speed differences are not just due different strategic approach
-faster decision making was mostly due to faster non-decision time Ter

however people may choose to play this game if they already have a higher drift rate. thus, CSgo doesnt train people and make their decision making faster because their decision making is already fast due to higher drift rate

Question 21

How do you change the separation boundary?

Answer 21

by changing your response time or accuracy by changing your strategy + what your prioritise

Question 22

What are the main parameters investigated in DDMs?

Answer 22

drift rate (v), boundary separation (a), and non-decision time Ter

Question 23

What did the study by von Bastian and Oberauer 2013 investigate?
What measures did they investigate?
What was the resukts?

Answer 23

Effect of practice on decision making

1Visual matching training: say whether the two faces are the same (with different expression and sunglasses)
2say whether numbers are the same
3LSD: say whether they are the same image

-investigated speed and accuracy
-training caused RT to get faster but lower accuracy

Question 24

What did Reinhartz 2023 do with the data from von Bastian and Oberhauer 2013?
What model was used?

Answer 24

-used DDM to further investigate the cognitive processes underlying the results from vonBastion and Oberhauer 2013
-DDM using fast-dm-30 with Kolmogorov-Smirnov minimisation criterion

Question 25

What is Kolmogorov-Smirnov minimisation criterion similar to? How does it work to fit model?

Answer 25

- like RMSD - minimises the difference between observed and predicted CDF -> minimises maximum error

Question 26

What did Reinhartz et al. 2023 find out using DDM? for each parameter? what does this suggest?

Answer 26

-increased drift rates (v) for all tasks -> suggests training/practice improved evidence accumulation during decision making + increase was slow and steady with no plateau -> continual improvement -boundary separation decreases rapidly but levels out-> suggests participants favoured speed over accuracy OR just got bored -rapid decrease in non-decision time but levels out -> suggesting faster Te and Tr

Question 27

What was the change in boundary separation and non-decision time in Reinhartz et al. 2023? what do these results suggest?

Answer 27

rapid decrease in beginning but then levels out at the end -> suggests strategic changes and perceptual learning are the main drivers of rapid improvement in the first few sessions

Question 28

What is evidence accumulation in decision making?

Answer 28

gather more evidence/information to get better at answering paradigm questions

Question 29

Why did using the DDM by Reinhartz et al. 2023 help understand the results from von Bastian and Oberhauer 2013?

Answer 29

using DDM helped to understand what drive the observed practice-related changes in speed and accuracy (increase in speed but decrease in accuracy)

Question 30

What did Bastian et al 2022 investigate?

Answer 30

used DDM to model processing of speed training in different aged adults

Question 31

What was the issue with Bastian et al 2022? What is the solution to this?

Answer 31

-large sample size but was taken across 3 different cities with different languages -> need to account for within-sit and across-session dependencies -hierarchical bayesian estimations

Question 32

What were the four tasks in Bastian et al 2022?

Answer 32

simple: press button when stim appears choice: is it green or blue? switch: switch between two tasks dual: do both tasks at the same time !!

Question 33

How to do bayesian parmeter estimations in steps?

Answer 33

1) use prior distributions of possible parameter values 2) based on values in 1) you iteratively update these values to generate posterior distributions 3) get mean of the posterior distributions -> used as a point estimate of the parameter value OR mean gives the variance of the distribution as measure of certainty over this estimate =credibility interval

Question 34

What is bayesian estimation?

Answer 34

combining prior knowledge and observed data to derive a range of likely parameter values

Question 35

What is unique about transparency of doing bayesian parameter estimation?

Answer 35

have to state/publish what your prior is 1) what were you prior distributions of possible parameter values

Question 36

How do you calculate posterior distributions in bayseium parameter estimation? What algorithms does this method use?

Answer 36

using the Markov Chain Monte Carlo MCMC Metropolis-Hastings Algorithm Gibbs Sampling

Question 37

What is a conceptual overview of the Metropolis-Hastings algorithm?

Answer 37

1)choose plausible parameter value (target) which is then used to calculate the posterior distribution (current sample) 2) add random noise to current sample = proposal distribution 3) compare proposal distribution and target distribution 4) determine next sample: use accepted proposal otherwise reuse current sample (do step 2 onwards again) 5) go back to step 2) until enough samples have been obtained

Question 38

For MCMC Metropolis-Hastings method of Bayesian parameter estimation, if proposal > target which distribution do you accept? ditto if proposal

Answer 38

proposal > target = proposal! proposal < target = accept proposal BUT with only with a certain probability p=proposal/target ratio

Question 39

What is Bayesian useful for in modelling?

Answer 39

hierarchical models -> Hierarchical DDM

Question 40

What are the component of a Bayesian HDDM?

Answer 40

observed data latent variable with parent (hyper)prior estimated for each PARTICIPANT latent variable estimated for sample as a WHOLE

Question 41

wHat is Docker HDDM?

Answer 41

program which helps with bayesian HDDM -

Question 42

what should a trace of 25,000 MCMC iterations

Answer 42

flat hairy catepilar

Question 43

Bastian et al 2022 What were the results of training in terms of speed and accuracy?

Answer 43

every age group became faster at task but no changes in accuracy

Question 44

Bastian et al 2022 How did age group affect speed abd accuracy of task due to training?

Answer 44

age negatively correlated with speed + older you are, less you improved in speed no change in accuracy

Question 45

Bastian et al 2022 What were the effects of training on drift rate? Why was this contrary to expectation?

Answer 45

-no changes in drift rate due to training apart from decrease in drift rate in switch task (bad) -Bastian et al 2022 expected drift rates were better with increasing age but there was better drift rates in anyone 30+

Question 46

Bastian et al 2022 what were the effects of training on boundary separation?

Answer 46

In the beginning, older adults focus more on accuracy than speed (compliant with current literature) and vice versa but after many trials ALL groups abandon accuracy and prioritise speed

Question 47

Bastian et al 2022 what were the effects of training on non-decision time?

Answer 47

non-decision time decreased in ALL groups due to training but non-decision time increased with age

Question 48

Why was using the HDDM useful for von Bastian et al. 2022?

Answer 48

HDDM provided more insight into age differences in speed: -drift rates increased in older age -training reduced primarily boundary separation and non-decision time

Question 49

Overall why do we use DDMs for these paradigms?

Answer 49

-captures both speed and accuracy in same model -offers theoretical insight into the cognitive processes behind changes in speed and accuracy due to training and individual(age)