TD3 Specs
Unlike TD2, TD3 uses a different Markov Mode. Changes are in red.
Overview
Subjects see objects and make
2-alternative choices. Sometimes they
are rewarded. We keep track of their
choices, and how long they take to choose.
Each object represents a state that the subject is in. A markov model
describes the states and their transitions.
Subjects are randomly assigned to one of two conditions determining how
long they have to make each choice (1 second or 4 seconds).
Markov Model
All subjects are navigating
the same Markov model. They always make 300 choices. The model has 6 states. Each state is associated with a different
picture of an object (e.g., fork, tree, etc.).
In each state, the subject makes one of 2 responses (pressing the left Z
key or the right . key). Subjects start in State 1. The state+choice
determines (1) which state the subject goes into next, and (2) a reward, as
follows:
State Response
1 2
1 2 6
2 6 3
3 6 4
4 5 6
5 6 6
6 1 6
Reward
State Response
1 2
1 -2 2
2 2 -2
3 2 -2
4 -2 2
5 +25 2
6 1 1
You will have six jpeg images
of objects. These should be randomly
assigned to the six states (separate randomization for each subject). The assignment of objects to states should be
saved. While the program is being tested, simply display the
state “State 1”, etc.
Instructions
[expand
to 3 screens]
n this experiment, you will play a sort of
"game", giving you the opportunity to win a small amount of real
money (you will be paid at the end). You
will always make one of two responses -- pressing the left key or pressing the
right key. These keys have little
stickers labeled "L" and "R". Before you respond, you will see a picture in
the middle of the screen. You can think
of each picture as telling you what ""location"" you are
in. There are six locations, each
corresponding to a different picture (mountain, cave, etc)
Make whatever choice you want. When you make a response, the screen will
turn a blue color, and you will get a positive or negative reward. The computer will tell you how much you won
(or lost). Then it will "move"
you into the next state.
[page
break.]
As mentioned above, you should
think of each picture as telling you
where you are in an imaginary world.
Where you are, and what response you make, completely determines two
things: (1) your reward (or loss) and (2) what state you go into next. Luck or chance plays no role in this
game. What happens next depends on where
you are and what response you make.
The computer always displays
your total winnings in the bottom right-hand corner of the blue screen. If at the end of the game you are below zero,
you won't have to pay us any money. If
you are above zero, we will pay you whatever you have won.
[page
break]
Speed matters here. You will have 1 second [4 seconds] to
respond. [if it
is 1 second display this sentence: 1 second is not a lot of time, so this means
you have to respond based on your immediate hunch.] If you don’t respond within the allotted
time, you will hear a loud buzzer. The
buzzer will remain on until you make a response, so please try and respond as
soon as possible. You will make 300 responses during the experiment. Furthermore, the time between responses is
fixed. Therefore, if you respond more
quickly than you need to, this will not give you a chance to earn more money,
nor will it mean that the experiment finishes up any sooner. Therefore, we suggest that you take
approximately the designated amount of time (1 second [4 seconds]) to respond.
Sequence
Subject sees picture (as
determined by state). Accept only one of
the two designated responses. When they
respond, the screen should slightly change color, becoming bluish (with the
picture still present). Play the appropriate wave file. You win two cents”, “You lose two cents”,
“You win twenty-five cents”, etc. The wave files should be less than 3 seconds
long. The Reward Screen should remain
present for 3 seconds. At the bottom of
the Reward Screen, you should show the subject their total purse. Label that
e.g., “Total winnings: $.47”
If the subject does not
respond within the allocated time, play a loud buzzer sound until subject makes
a response.
Before going to the next
state, insert a 500 second blank screen.
The new screen should of course not be bluish anymore.
When the subject *does*
respond in time, you should still impose an artificial delay on the time the
next trial begins. That is, if they are
supposed to respond in 4 seconds, but they respond in 2.2 seconds, then you
should add 1.8 seconds to the presentation time of the Reward Screen (making it
4.8 seconds long).
Storage of Data
Save:
1. one
row per trial (1-300), with the following in each row:
a. trial#
b. state # they were in
c. object displayed
d. choice
e. reward
f. total reward
g. reaction time for choice
h. how long they were supposed to take (1000 or 4000)