Modeling in psychology: What

1
Modeling in psychologyWhats the point?

• Robert M. French
• L.E.A.D. CNRS
• U. de Bourgogne

2
What do we mean by a model The reduction of
a phenomenon to its essential elements as a means
of explaining it.
3
A good model must

• Reproduce, at least qualitatively, the existing
  empirical data.

• Explain this data.

• Make predictions that go beyond the existing data.

• Clearly indicate its context of application
  (i.e., its level of explanation).

• Be able to be probed in order to how its
  mechanisms produce the phenomena being modeled
  and to understand the limits of the model.

• Be falsifiable.

4
A probabilistic model/theory of the world

• Theory of the World
• Assume that past experience shows that an event
  A is as frequent as an event B. But suppose that
  for a period of time, we have only seen event A.
  We conclude that event B should become more
  likely until event B catches up to event A.

For example, suppose we get the following
sequence of heads and tails
?
T
T
T
T
H
H
H
H
T
H
H
H
T
T
T
T
T
T
H
Despite our naive theory p(H) ½, p(T) ½
5
A little more complicated
car
goat
goat
6
The host of a TV game show puts a goat behind two
of the doors and the new car behind the other.
He picks a volunteer from the audience and
explains that there are goats behind two of the
doors and a new car behind the other door. He
invites the volunteer to pick a door.
7
The volunteer picks Door No. 3.
3
1
2
8
3
1
2
The television host then opens Door No. 2, behind
which is a goat.
9
?
3
1
2
He asks the volunteer if he wants to change his
choice to Door 1 or keep Door 3. He may keep
whatever is behind the door of his final choice.
10
?
3
1
2
He asks the volunteer if he wants to change his
choice to Door 1 or keep Door 3. He may keep
whatever is behind the door of his final choice.
11
Our experience (probabilistic model of the world)
tells us that when there are two options and we
dont know the outcome of either, we choose at
random.
In this case, this theory leads us to the wrong
conclusion, because...
....WE MUST CHANGE DOORS !
12
A probabilistic model of basketball an
explanation of the  hot hand .
Thomas Gilovich, Robert Vallone, and Amos Tversky
recorded every basket shot by players for the
Philadelphia 76ers during the 1980-81 season.
Gilovich et al. (1985). The Hot Hand in
Basketball. On the Misinterpretation of Random
Sequences. Cognitive Psychology, 17, 295-314.
13
Their model of success at scoring in
basketball Statistically, if each shot is
modeled as an INDEPENDENT EVENT, the performance
of a player can be modeled with a coined biased
so that it reflects the seasonal shooting average
of the player. Heads basket made tails
basket not made. According to this model, there
is no such thing as a  hot hand .
14
1 basket made 0 basket missed
Not great, not bad
Warm up
0 1 1 0 1 1 1 0 1 1 0 1 0 0 0 0 0 1 1 0 0 0 1 1 1
1 0 0 1 1 1 1 0 0 1 0 0 0 1 0 0 1 0 0 1 1 0 0 0
0 1 0 0 0 1 1 1 1 1 1 0 1 1 1 1 1 1 1 0 0 1 0 0 0
1 0 0 0 1 0 1 1 0 0 1 1 1 1 1 0 0 0
Pass me the ball, Im HOT!
Percentages of shots made 55
Simulated by a 55/45 biased coin and the voice of
a basketball player.
15
Some reactions

• There are so many variables involved in making a
  shot, that this study makes no sense. -- Bobby
  Knight, Indiana U.
• Who is this guy? He does a study, so what? I
  dont give a damn what he found. -- owner of the
  Boston Celtics
• We once did a lot of sport. You tell ME whats
  wrong here. Its for you to explain, not for me
  to believe. -- my brother.

16
Consider our criteria

• Reproduces, at least qualitatively, the empircal
  data 

• Explains the data.

• Makes predictions

• Clearly indicates its context of application.

• Can be probed to discover its limits and to
  understand how its underlying mechanisms produce
  the effects being modeled.

NON

• Is falsifiable.

Conclusion Its a pretty good model of scoring
in basketball.
17
Models and Prediction
The Delphi Oracle predicts
An earthquake will level Athens in 8 days.
Migratory birds will fly to Africa a month early
this year.
Sun spots will begin at the end of 2007 this year.
18
Models and Prediction
The Black Box predicts
An earthquake will level Athens in 8 days.
Migratory birds will fly to Africa a month early
this year.
Sun spots will begin at the end of 2007 this year.
The Black Box has no explanatory power and is,
therefore, not a model
19
A model of how the cock crows

• Reproduces, at least qualitatively, the empircal
  data 

• Explains the data.

• Makes predictions

• Clearly indicates its context of application.

• Can be probed to discover its limits and to
  understand how its underlying mechanisms produce
  the effects being modeled.

• Is falsifiable.

20
Lets examine a little more closely the following
three criteria

• Clearly indicates its context of application.

• Can be probed to discover its limits and to
  understand how its underlying mechanisms produce
  the effects being modeled.

• Is falsifiable.

21

• Clearly indicates its context of application.

Without this, every model becomes false, simply
by saying,  Well, it doesnt explain this lower
level.  Neuroscientists do this all the time
 Where are the GABA receptors in your model? 
etc. For this reason, you must specify the
level at which your model is designed to explain
things.
22

• Can be probed to discover its limits and to
  understand how its underlying mechanisms produce
  the effects being modeled.

Without this, we cannot have the permanent and
necessary interaction between the model and
empirical data.
23

• Is falsifiable

Astrology
Freudian psychoanalysis
Greek/Hindu/Western mythology.
Evolutionary psychology??
All unfalsifiable!
24
Unfalsifiability
An interaction between a model and empirical data
is necessary BUT It can lead and too often
does lead to an unfalsifiable model.
25
Boxology The disease of cognitive modelers
26
Original model
Phonological module
New Data to explain
27
New data to be explained
New data explained
Original Model
Pre-linguistic Module
Phonological module
28
Visual-gusatory module
Original Model
Pre-linguistic Module
Phonological module
29

• But this can, and frequently does,
• lead to rampant boxological cancer.

30
(No Transcript)
31

• .and, once again, the system loses its
  explicative power and, along the way, also
  becomes unfalsifiable.

32
Consider our criteria

• Reproduces, at least qualitatively, the empircal
  data 

• Explains the data.

MAYBE

• Makes predictions

NO

• Clearly indicates its context of application.

NO

• Can be probed to discover its limits and to
  understand how its underlying mechanisms produce
  the effects being modeled.

???

• Is falsifiable.

NO
33
Evolutionary Psychology
Falsifiable or not??
In its currently practiced form, the answer is
(mostly) NO.
34
Connectionist models
35
What do cows drink?Connectionism provides a
bottom-up answer MILK
COW
MILK
DRINK
These neurons are active even the network has
never heard the word MILK
36


What do cows drink?



Symbolic AI gives a top-down answer


ISA(cow, mammal)


ISA(mammal, animal)


Rule 1

IF animal(X) AND thirst(X) THEN lacks_water(X)

IF lacks_water(X) THEN drink_water(X)

Rule 2


Conclusion
Cows drink WATER.


37
A good model of human cognition must be able to
give both answers, according to the context in
which it is asked the question.
38
We will be looking closely at a connectionist
model of categorization in young infants.
39
Connectionist model of Bilingual language
learning(French, 1998 French Jacquet, 2004)

• Input to the SRN
• - Two micro languages, Alpha Beta, 12 words
  each
• An SVO grammar for each language
• - Unpredictable language switching

Attempted Prediction
BOY LIFTS TOY MAN SEES PEN GIRL PUSHES BALL BOY
PUSHES BOOK FEMME SOULEVE STYLO FILLE PREND STYLO
GARÇON TOUCHE LIVRE FEMME POUSSE BALLON FILLE
SOULEVE JOUET WOMAN PUSHES TOY.... (Note absence
of markers between sentences and between
languages.)
The network tries each time to predict the next
element.
We do a cluster analysis of its internal
(hidden-unit) representations after having seen
20,000 sentences.
40
Summary
A model is the reduction of a phenomenon (or set
of phenomena) to its essential elements as a
means of explaining it. It must

• Reproduce, at least qualitatively, the existing
  empirical data.

• Explain this data.

• Make predictions that go beyond the existing data.

• Clearly indicate its context of application
  (i.e., its level of explanation).

• Be able to be probed in order to how its
  mechanisms produce the phenomena being modeled
  and to understand the limits of the model.

• Be falsifiable.