Vous êtes sur la page 1sur 10

Predicted mean response times for a two-stage

model
In an experiment participants were asked to respond to a visual target as
quickly as possible by making a saccade to this target. In a part of the trials
the visual stimulus was accompanied by an auditory stimulus, which could
be at the same location or a different one. Also the point in time at which
both stimuli were presented was varied. In all conditions where two stimuli
were presented, reaction times were faster than in the condition with just the
visual stimulus.
In Figure 1 the model to describe the data from the experiment is illus-
trated.

Visual
Processing
Central
Processing

Onset Auditory
Asynchrony Processing

Figure 1: Two-stage model in which peripheral processes run parallel, fol-


lowed by a central processing stage.

In the so-called V-version of the model, central processes can start as


soon as the visual processes have ended. If no interaction between the two
processes take place, the model can be simplified to the version shown in
Figure 2. If response times of the first stage are exponentially distributed
and central processing times are normally distributed, the total response
times are ex-Gaussian distributed.

Visual Central
Processing Processing

Figure 2: Two-stage model in which peripheral visual processing is followed


by a central processing stage.

If response times are measured from the onset of the visual stimulus, the
expected response time equals:

1
1
E(RT ) = µ +
λA
where µ is the mean of the Gaussian distribution, and λA the rate of the
exponential distribution. The variance equals:
1
V ar(RT ) = σ 2 +
λ2A
where σ is the standard deviation of the normal distribution. In order to get
a model that can explain facilitation in the case of two stimuli an assumption
is added: if the auditory process finishes before the visual process, central
processes are speeded up.
In order to compute the probability of the auditory process finishing ear-
lier than the visual process, the city-block representation of the race can be
used. In Figure 3 the block-diagram of the race process under consideration
is presented.
If the visual process starts first, the only path in the diagram in which
the auditory process finishes first, is: {T10 , T11 , T12 }. The probability that
this path is taken equals:
λA
exp(−λV · τ ) ·
λA + λV
where τ equals the stimulus onset asynchrony.
If the auditory process starts the race, the auditory process can finish
during the SOA with probability 1 − exp(λA · τ ), or the path {T01 , T11 , T12 }
can be taken, with probability:
λA
exp(−λA · τ )
λA + λV
Therefore the expected response time in the model with additional as-
sumption equals:
1
E(RT ) = +µ+π·δ
λv
where
exp(−λV · τ ) · λAλ+λ
(
A
if the visual process starts first
π= V
1 − exp(λA · τ ) · λAλ+λ
V
V
if the auditory process starts first

and where δ is the amount of facilitation.


In a second version of the model, the winner of the race determines the
duration of the parallel stage of the model. By this race-assumption the

2
T20 T21 T22

Visual T10 T11 T12


process

T00 T01 T02


SOA

Auditory
process
Figure 3: City-block representation of the race.

facilitation for two stimuli is in part accounted for my statistical facilitation.


The expected response time for this model can again be decomposed in the
duration of the race and the duration of central processes. The expected
response time for the central processes is not different in the race version of
the model. In order to determine the expected duration of the race, again
the city-block-diagram can be used.
In the case the visual stimulus starts the race, there is a probability that
the visual stimulus is processed during the SOA equal to

1 − exp(−λV · τ )

The expected response time in this case equals:


1
λV

3
If the visual stimulus is not entirely processed during the SOA, which has
probability exp(−λV · τ ), the expected response time equals
1
λV + λA
Similar reasoning for case in which the auditory starts first results in the
following predicted duration of the race:

1


 λV
− exp(−λV · τ ) λV (λλVA+λA ) if the visual stimulus starts first
E(RT race) =
1
− exp(−λA · τ ) λA (λλVV+λA )



λA
if the auditory stimulus starts first

Simulation Study
To get an impression of shape of the reaction time distribution predicted by
the model several datasets were generated using the model. This is done
according to the following scheme for the model where the duration of the
visual process determines the duration of the first stage:

x1=number generated by exponential distribution with mean V


x2=number generated by exponential distribution with mean A
x3=number generated by normal distribution with mean mu and std sigma

if x1<x2+SOA then
rt=x1+x3
else rt=x1+x3-delta

This algoritm was used for three SOAs and for 10000 datapoints per
1 1
SOA. The following parameters were fixed: λV = 200 , λA = 300 , µ = 200,
σ = 50, and δ = 50. The means and standard deviations that were found
are presented in Table 1. Histograms of predicted response times are shown
in Figures 4 to 6.

Table 1: Means found for a simulation study with SOAs equal to 0, 50, or
-100 ms together with predicted means.
SOA mean RT Std RT predicted RT
0 389 188 370
50 393 192 377
-100 367 187 391

4
SOA=0
2500

2000

1500

1000

500

0
0 200 400 600 800 1000 1200 1400 1600 1800

Figure 4: Simulated RT distribution for the V-model and SOA=0.

The shape of each of the reponse time distributions is quite similar to the
shape of a typical reponse time distribution. Because of the normal distri-
bution it is always possible to generate negative reponse times. Therefore it
might be better to use a truncated normal distribution instead.
Response times according to the race model version of the two-stage
model are generated using a different decision rule in the algorithm:
if x1<x2+SOA then
rt=x1+x3
else rt=x2+SOA+x3-delta
The means obtained in the simulation for the race version of the model
are shown in Table 2. Histograms are presented in Figures 7 to 9.

Matlab-code
The following matlab-code was used to generate data according to the V-
model.

5
SOA=50
3000

2500

2000

1500

1000

500

0
0 500 1000 1500 2000 2500

Figure 5: Simulated RT distribution for the V-model and SOA=50.

Table 2: Means found for a simulation study with SOAs equal to 0, 50, or
-100 ms together with predicted means.
SOA mean RT Std RT predicted RT
0 303 129 290
50 326 127 274
-100 243 137 260

x=[];
m_v=200;
m_a=300;
mu=200;
sigma=50;
delta=50;
soa=0;

for i=1:10000,
x1=-m_v*log(rand(1,1)); %duration visual process

6
SOA=−100
3000

2500

2000

1500

1000

500

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000

Figure 6: Simulated RT distribution for the V-model and SOA=-100.

x2=-m_a*log(rand(1,1)); %duration auditory process


if x1<x2+soa,
x2=mu+sigma*randn; %transforming z-values
x=[x x1+x2];
end;
if x1>=x2+soa,
x3=mu+sigma*randn-delta;
x=[x x1+x3];
end;
i
end;
hist(x,25);
gemiddelde=mean(x)
standdev=std(x)

Response times according to the Min-model were generated using the


following code:

7
SOA=0
2500

2000

1500

1000

500

0
0 200 400 600 800 1000 1200 1400 1600 1800

Figure 7: Simulated RT distribution for the Min-model and SOA=0.

x=[];
m_v=200;
m_a=300;
mu=200;
sigma=50;
delta=50;
soa=0;

for i=1:10000,
x1=-m_v*log(rand(1,1)); %duration visual process
x2=-m_a*log(rand(1,1)); %duration auditory process
if x1<x2+soa,
x2=mu+sigma*randn; %transforming z-values
x=[x x1+x2];
end;
if x1>=x2+soa,
x3=mu+sigma*randn-delta;

8
SOA=50
3000

2500

2000

1500

1000

500

0
0 500 1000 1500 2000 2500

Figure 8: Simulated RT distribution for the Min-model and SOA=50.

x=[x x2+x3+soa];
end;
i
end;
hist(x,25);
gemiddelde=mean(x)
standdev=std(x)

9
SOA=−100
3000

2500

2000

1500

1000

500

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000

Figure 9: Simulated RT distribution for the Min-model and SOA=-100.

10

Vous aimerez peut-être aussi