Steph Steinhardt "Big Data, Open Infrastructure and Care: Following the Rise ...
Greg taylor
1. Formal Modelling in the Social Sciences
Greg Taylor
— Oxford Internet Institute —
Summer Doctoral Programme 2014
G. Taylor (OII) Modelling 1 / 48
2. What is a model?
model, n. and adj. A simplified or idealized description or conception of a
particular system, situation, or process, often in mathematical terms, that is
put forward as a basis for theoretical or empirical understanding, or for
calculations, predictions, etc.; a conceptual or mental representation of
something.
Models are abstractions and simplifications of reality.
G. Taylor (OII) Modelling 2 / 48
3. What is a model?
model, n. and adj. A simplified or idealized description or conception of a
particular system, situation, or process, often in mathematical terms, that is
put forward as a basis for theoretical or empirical understanding, or for
calculations, predictions, etc.; a conceptual or mental representation of
something.
Models are abstractions and simplifications of reality.
World is very, very complex.
Objective: distil world to a core of interesting elements to admit a
tractable analysis.
G. Taylor (OII) Modelling 2 / 48
4. What is a model?
model, n. and adj. A simplified or idealized description or conception of a
particular system, situation, or process, often in mathematical terms, that is
put forward as a basis for theoretical or empirical understanding, or for
calculations, predictions, etc.; a conceptual or mental representation of
something.
Models are abstractions and simplifications of reality.
World is very, very complex.
Objective: distil world to a core of interesting elements to admit a
tractable analysis.
The trade-off:
1. retain the features that make model recognisable.
2. be simple enough to permit clear understanding.
G. Taylor (OII) Modelling 2 / 48
5. What is a model?
model, n. and adj. A simplified or idealized description or conception of a
particular system, situation, or process, often in mathematical terms, that is
put forward as a basis for theoretical or empirical understanding, or for
calculations, predictions, etc.; a conceptual or mental representation of
something.
Models are abstractions and simplifications of reality.
World is very, very complex.
Objective: distil world to a core of interesting elements to admit a
tractable analysis.
The trade-off:
1. retain the features that make model recognisable.
2. be simple enough to permit clear understanding.
For our purposes, a model is equipped with an apparatus that
allows analysis (is not purely descriptive).
G. Taylor (OII) Modelling 2 / 48
6. Where are models used?
In the social sciences modelling is most heavily concentrated in
economics, where it has been the dominant means of theory building
for a century and a half.
G. Taylor (OII) Modelling 3 / 48
7. Where are models used?
In the social sciences modelling is most heavily concentrated in
economics, where it has been the dominant means of theory building
for a century and a half.
These methods have spread to other ‘nearby’ disciplines, e.g. political
science, marketing, and computer science.
G. Taylor (OII) Modelling 3 / 48
8. Outline
Models as tools for logic
Models as laboratories
Models for concrete communication
Models as unified paradigms for knowledge
Other advantages
G. Taylor (OII) Modelling 4 / 48
9. Logical consistency
Consider the following statements:
1. If prices are high then manufacturers will produce more since it is
more profitable for them to do so.
High prices correspond to high production.
G. Taylor (OII) Modelling 5 / 48
10. Logical consistency
Consider the following statements:
1. If prices are high then manufacturers will produce more since it is
more profitable for them to do so.
High prices correspond to high production.
2. If prices are high then people will buy less; if people buy less then
manufacturers will produce less because they tend to produce
only what they can sell.
High prices correspond to low production.
G. Taylor (OII) Modelling 5 / 48
11. Logical consistency
Consider the following statements:
1. If prices are high then manufacturers will produce more since it is
more profitable for them to do so.
High prices correspond to high production.
2. If prices are high then people will buy less; if people buy less then
manufacturers will produce less because they tend to produce
only what they can sell.
High prices correspond to low production.
It is often hard to assess the logical consistency of even moderately
sophisticated rhetorical arguments about empirical relationships.
But also hard to diagnose what goes wrong: premise vs. conclusion.
G. Taylor (OII) Modelling 5 / 48
12. Smith’s ‘Wealth of Nations’ (1776) Book 1, Chap. VII
When the price of any commodity is neither more nor less than what is sufficient to pay the rent of the land, the wages of the
labour, and the profits of the stock employed in raising, preparing, and bringing it to market, according to their natural rates, the
commodity is then sold for what may be called its natural price.
The commodity is then sold precisely for what it is worth, or for what it really costs the person who brings it to market; for
though in common language what is called the prime cost of any commodity does not comprehend the profit of the person who
is to sell it again, yet if he sells it at a price which does not allow him the ordinary rate of profit in his neighbourhood, he is
evidently a loser by the trade; since by employing his stock in some other way he might have made that profit. His profit,
besides, is his revenue, the proper fund of his subsistence. As, while he is preparing and bringing the goods to market, he
advances to his workmen their wages, or their subsistence; so he advances to himself, in the same manner, his own subsistence,
which is generally suitable to the profit which he may reasonably expect from the sale of his goods. Unless they yield him this
profit, therefore, they do not repay him what they may very properly be said to have really cost him.
Though the price, therefore, which leaves him this profit, is not always the lowest at which a dealer may sometimes sell his
goods, it is the lowest at which he is likely to sell them for any considerable time; at least where there is perfect liberty, or where
he may change his trade as often as he pleases.
The actual price at which any commodity is commonly sold is called its market price. It may either be above, or below, or
exactly the same with its natural price.
The market price of every particular commodity is regulated by the proportion between the quantity which is actually
brought to market, and the demand of those who are willing to pay the natural price of the commodity, or the whole value of the
rent, labour, and profit, which must be paid in order to bring it thither. Such people may be called the effectual demanders, and
their demand the effectual demand; since it may be sufficient to effectuate the bringing of the commodity to market. It is different
from the absolute demand. A very poor man may be said in some sense to have a demand for a coach and six; he might like to
have it; but his demand is not an effectual demand, as the commodity can never be brought to market in order to satisfy it.
When the quantity of any commodity which is brought to market falls short of the effectual demand, all those who are
willing to pay the whole value of the rent, wages, and profit, which must be paid in order to bring it thither, cannot be supplied
with the quantity which they want. Rather than want it altogether, some of them will be willing to give more. A competition will
immediately begin among them, and the market price will rise more or less above the natural price, according as either the
greatness of the deficiency, or the wealth and wanton luxury of the competitors, happen to animate more or less the eagerness of
the competition. Among competitors of equal wealth and luxury the same deficiency will generally occasion a more or less eager
competition, according as the acquisition of the commodity happens to be of more or less importance to them. Hence the
exorbitant price of the necessaries of life during the blockade of a town or in a famine.
When the quantity brought to market exceeds the effectual demand, it cannot be all sold to those who are willing to pay the
whole value of the rent, wages and profit, which must be paid in order to bring it thither. Some part must be sold to...
G. Taylor (OII) Modelling 6 / 48
13. Modelling paradigm
Assumptions Calculation Results
correct
Who is involved?
What are their objectives?
How are they constrained?
New environment→new assumptions.
G. Taylor (OII) Modelling 7 / 48
14. Modelling paradigm
Assumptions Calculation Results
correct
Who is involved?
What are their objectives?
How are they constrained?
Who is involved?
What are their objectives?
How are they constrained?
New environment→new assumptions.
G. Taylor (OII) Modelling 7 / 48
15. Modelling paradigm
Assumptions Calculation Results
correct
Who is involved?
What are their objectives?
How are they constrained?
Who is involved?
What are their objectives?
How are they constrained?
correct
New environment→new assumptions.
G. Taylor (OII) Modelling 7 / 48
16. Modelling paradigm
Assumptions Calculation Results
correct
Who is involved?
What are their objectives?
How are they constrained?
Who is involved?
What are their objectives?
How are they constrained?
correct
New environment/phenomenon→new assumptions.
G. Taylor (OII) Modelling 7 / 48
28. Paradox resolved: high price =⇒ high quantity
0 q 1
0
p
1
D
S
Quantity
Price
q
p
D
G. Taylor (OII) Modelling 16 / 48
29. Paradox resolved: high price =⇒ low quantity
0 q 1
0
p
1
D
S
Quantity
Price
G. Taylor (OII) Modelling 17 / 48
30. Paradox resolved: high price =⇒ low quantity
0 q 1
0
p
1
D
S
Quantity
Price
q
p
S
G. Taylor (OII) Modelling 17 / 48
31. How general is it?
Are these results an artefact of my chosen demand and supply curves?
†See Appendix for rigorous proof of this. 18 / 48
32. How general is it?
Are these results an artefact of my chosen demand and supply curves?
D(p; d0) satisfies D1(p; d0) < 0, D2(p; d0) > 0, and D(0; d0) > 0.
S(p) satisfies S (p) > 0 and S(0) < D(0; d0).
D(·; ·) and S(·) are continuous.
†See Appendix for rigorous proof of this. 18 / 48
33. How general is it?
Are these results an artefact of my chosen demand and supply curves?
D(p; d0) satisfies D1(p; d0) < 0, D2(p; d0) > 0, and D(0; d0) > 0.
S(p) satisfies S (p) > 0 and S(0) < D(0; d0).
D(·; ·) and S(·) are continuous.
Equilibrium condition: D(p; d0) = S(p) =⇒ p = D−1(S(p); d0).
S (p) > 0, D1(p; d0) < 0 =⇒
∂
∂p
D−1
(S(p); d0) < 0.
Suppose we increase d0 (shift demand curve outward):
D1(p; d0) < 0, D2(p; d0) > 0 =⇒
∂
∂d0
D−1
(S(p); d0) > 0†
=⇒
∂p
∂d0
> 0.
So price increases. What about quantity? Using the chain rule:
∂S(p∗)
∂d0
=
∂p∗
∂d0
S (p∗
) > 0.
†See Appendix for rigorous proof of this. 18 / 48
34. Putting the model to work
Note: although model is mathematical, it is essentially qualitative.
The model resolves the apparent paradox, and makes the underlying
mechanism much clearer.
It is the insistence on understanding such mechanisms that
differentiates social science from ‘market research’.
G. Taylor (OII) Modelling 19 / 48
35. Putting the model to work
Note: although model is mathematical, it is essentially qualitative.
The model resolves the apparent paradox, and makes the underlying
mechanism much clearer.
It is the insistence on understanding such mechanisms that
differentiates social science from ‘market research’.
Model instantly tells us about effects of things that shift supply or
demand.
Example: flooding in Thailand
G. Taylor (OII) Modelling 19 / 48
36. Putting the model to work
Note: although model is mathematical, it is essentially qualitative.
The model resolves the apparent paradox, and makes the underlying
mechanism much clearer.
It is the insistence on understanding such mechanisms that
differentiates social science from ‘market research’.
Model instantly tells us about effects of things that shift supply or
demand.
Example: flooding in Thailand =⇒ reduced supply of hard disk
drives
G. Taylor (OII) Modelling 19 / 48
37. Putting the model to work
Note: although model is mathematical, it is essentially qualitative.
The model resolves the apparent paradox, and makes the underlying
mechanism much clearer.
It is the insistence on understanding such mechanisms that
differentiates social science from ‘market research’.
Model instantly tells us about effects of things that shift supply or
demand.
Example: flooding in Thailand =⇒ reduced supply of hard disk
drives =⇒ higher hdd prices.
G. Taylor (OII) Modelling 19 / 48
39. Putting the model to work
Example: EVE developers nerf cruise missile launcher
G. Taylor (OII) Modelling 21 / 48
40. Putting the model to work
Example: EVE developers nerf cruise missile launcher =⇒ reduced
demand for cruise missiles
G. Taylor (OII) Modelling 21 / 48
41. Putting the model to work
Example: EVE developers nerf cruise missile launcher =⇒ reduced
demand for cruise missiles =⇒ lower cruise missile prices.
G. Taylor (OII) Modelling 21 / 48
43. Putting the model to work
Example: Steve Balmer quits Microsoft
G. Taylor (OII) Modelling 23 / 48
44. Putting the model to work
Example: Steve Balmer quits Microsoft =⇒ increased future
profitability
G. Taylor (OII) Modelling 23 / 48
45. Putting the model to work
Example: Steve Balmer quits Microsoft =⇒ increased future
profitability =⇒ increased demand for Microsoft shares =⇒ increase
in share price.
G. Taylor (OII) Modelling 23 / 48
47. Outline
Models as tools for logic
Models as laboratories
Models for concrete communication
Models as unified paradigms for knowledge
Other advantages
G. Taylor (OII) Modelling 25 / 48
48. Putting the model to work
As well as explaining price changes ex post, the model can be used for
ex ante experimentation.
Let’s look at a hypothetical policy scenario to see how understanding
the model can help us.
G. Taylor (OII) Modelling 26 / 48
49. Putting the model to work
As well as explaining price changes ex post, the model can be used for
ex ante experimentation.
Let’s look at a hypothetical policy scenario to see how understanding
the model can help us.
Suppose there is a policy objective to decrease some illegal or socially
undesirable activity that is highly addictive/compulsive (e.g. online
gambling).
G. Taylor (OII) Modelling 26 / 48
50. Putting the model to work
As well as explaining price changes ex post, the model can be used for
ex ante experimentation.
Let’s look at a hypothetical policy scenario to see how understanding
the model can help us.
Suppose there is a policy objective to decrease some illegal or socially
undesirable activity that is highly addictive/compulsive (e.g. online
gambling).
Could (a) target sellers/suppliers with more severe penalties, (b)
target buyers/users with penalties, education programmes, or
remedial measures.
G. Taylor (OII) Modelling 26 / 48
52. Supply-side policy
0 q 1
0
p
1
D
S
Quantity
Price
q
p
S
↓ Supply
G. Taylor (OII) Modelling 27 / 48
53. Supply-side policy
0 q 1
0
p
1
D
S
Quantity
Price
q
p
S
↓ Supply
G. Taylor (OII) Modelling 27 / 48
54. Supply-side policy
0 q 1
0
p
1
D
S
Quantity
Price
q
p
D
↓ Demand
G. Taylor (OII) Modelling 28 / 48
55. Supply-side policy
0 q 1
0
p
1
D
S
Quantity
Price
q
p
D
↓ Demand
G. Taylor (OII) Modelling 28 / 48
56. Ineffective supply-side crackdowns
“Despite attempts to prohibit Internet gambling – most recently
through the Unlawful Internet Gambling Enforcement Act of 2006 –
American consumers nevertheless are wagering online more than $100
billion annually with operators based outside the U.S. [...] As a result,
rather than protecting consumers, the UIGEA has simply left the
millions of Americans who continue to find a way to gamble online
unprotected and vulnerable to unscrupulous gambling operators.”
— Jared Polis, U.S. Congress
G. Taylor (OII) Modelling 29 / 48
57. Outline
Models as tools for logic
Models as laboratories
Models for concrete communication
Models as unified paradigms for knowledge
Other advantages
G. Taylor (OII) Modelling 30 / 48
58. Slippery concepts
Many concepts in the social and related sciences are inherently hard to
pin down (information, knowledge, bias, etc.).
G. Taylor (OII) Modelling 31 / 48
59. Slippery concepts
Many concepts in the social and related sciences are inherently hard to
pin down (information, knowledge, bias, etc.).
Often, the biggest danger is not having the wrong definition, but
rather having no definition at all!
G. Taylor (OII) Modelling 31 / 48
60. Slippery concepts
Many concepts in the social and related sciences are inherently hard to
pin down (information, knowledge, bias, etc.).
Often, the biggest danger is not having the wrong definition, but
rather having no definition at all!
Usually easier to define things that can be measured (e.g. intelligent
vs. knowledgeable).
G. Taylor (OII) Modelling 31 / 48
61. Slippery concepts
Many concepts in the social and related sciences are inherently hard to
pin down (information, knowledge, bias, etc.).
Often, the biggest danger is not having the wrong definition, but
rather having no definition at all!
Usually easier to define things that can be measured (e.g. intelligent
vs. knowledgeable). Models can help us by making more things
measurable in a standardised fashion.
G. Taylor (OII) Modelling 31 / 48
67. Total surplus
0 q∗ 1
0
p∗
1
D
S
Quantity
Price
G. Taylor (OII) Modelling 36 / 48
68. Total surplus
0 q∗ 1
0
p∗
1
D
S
efficient!
Quantity
Price
G. Taylor (OII) Modelling 36 / 48
69. Welfare
We have a definition of welfare (there are others). Also definitions of
consumer surplus, producer surplus, and efficiency.
G. Taylor (OII) Modelling 37 / 48
70. Welfare
We have a definition of welfare (there are others). Also definitions of
consumer surplus, producer surplus, and efficiency.
The definitions are made concrete by the model.
The newly defined concept immediately becomes a useful analytical
tool:
Empirically: Most estimates put the consumer surplus from home
Internet use at about 2–4% of GDP.
Theoretically...
G. Taylor (OII) Modelling 37 / 48
71. Outline
Models as tools for logic
Models as laboratories
Models for concrete communication
Models as unified paradigms for knowledge
Other advantages
G. Taylor (OII) Modelling 38 / 48
72. Monopolised markets
We have been looking at a competitive market (recall that it was
competitive pressure that forced the price to equilibrium).
G. Taylor (OII) Modelling 39 / 48
73. Monopolised markets
We have been looking at a competitive market (recall that it was
competitive pressure that forced the price to equilibrium). We say
firms are price takers.
G. Taylor (OII) Modelling 39 / 48
74. Monopolised markets
We have been looking at a competitive market (recall that it was
competitive pressure that forced the price to equilibrium). We say
firms are price takers.
Some industries have one dominant firm—a monopoly. Such a firm
gets to dictate the price.
G. Taylor (OII) Modelling 39 / 48
75. Monopolised markets
We have been looking at a competitive market (recall that it was
competitive pressure that forced the price to equilibrium). We say
firms are price takers.
Some industries have one dominant firm—a monopoly. Such a firm
gets to dictate the price.
For this we need a new model.
G. Taylor (OII) Modelling 39 / 48
76. Monopolist 1: producer surplus
0 q∗ 1
0
p∗
1
D
S
Quantity
Price
G. Taylor (OII) Modelling 40 / 48
77. Monopolist 2: a better price?
0 q q∗ 1
0
p
p∗
1
D
S
Quantity
Price
G. Taylor (OII) Modelling 41 / 48
78. Monopolist 2
0 q q∗ 1
0
p
p∗
1
+
-
D
S
Quantity
Price
G. Taylor (OII) Modelling 42 / 48
79. Monopolist 3: welfare
0 q q∗ 1
0
p
p∗
1
D
S
Quantity
Price
G. Taylor (OII) Modelling 43 / 48
80. Monopolist 3: dead weight loss
0 q q∗ 1
0
p
p∗
1
D
S
Quantity
Price
G. Taylor (OII) Modelling 44 / 48
82. Monopolised markets
New market structure demanded a new model. But by working within
the same analytical paradigm,
(i) we were able to reuse existing tools and results;
G. Taylor (OII) Modelling 45 / 48
83. Monopolised markets
New market structure demanded a new model. But by working within
the same analytical paradigm,
(i) we were able to reuse existing tools and results;
(ii) we were able to compare the results of the new model to those of
the old in a very explicit fashion.
G. Taylor (OII) Modelling 45 / 48
84. Monopolised markets
New market structure demanded a new model. But by working within
the same analytical paradigm,
(i) we were able to reuse existing tools and results;
(ii) we were able to compare the results of the new model to those of
the old in a very explicit fashion.
Over time, this helps to build a body of knowledge that is unified in
the sense that it is mutually compatible and expressed in a common
analytical language.
G. Taylor (OII) Modelling 45 / 48
85. Monopolised markets
New market structure demanded a new model. But by working within
the same analytical paradigm,
(i) we were able to reuse existing tools and results;
(ii) we were able to compare the results of the new model to those of
the old in a very explicit fashion.
Over time, this helps to build a body of knowledge that is unified in
the sense that it is mutually compatible and expressed in a common
analytical language.
Important in topical studies such as ours since it prevents reinvention
of the wheel.
G. Taylor (OII) Modelling 45 / 48
86. Outline
Models as tools for logic
Models as laboratories
Models for concrete communication
Models as unified paradigms for knowledge
Other advantages
G. Taylor (OII) Modelling 46 / 48
87. Other advantages of formal models
We have seen that models are tools for logic, experimentation,
communication, and unification.
G. Taylor (OII) Modelling 47 / 48
88. Other advantages of formal models
We have seen that models are tools for logic, experimentation,
communication, and unification.
Models can help us in a number of other ways:
Encourage scientific transparency by making underlying
assumptions explicit and concrete.
G. Taylor (OII) Modelling 47 / 48
89. Other advantages of formal models
We have seen that models are tools for logic, experimentation,
communication, and unification.
Models can help us in a number of other ways:
Encourage scientific transparency by making underlying
assumptions explicit and concrete.
Facilitate ‘theoretical reproducibility’.
G. Taylor (OII) Modelling 47 / 48
90. Other advantages of formal models
We have seen that models are tools for logic, experimentation,
communication, and unification.
Models can help us in a number of other ways:
Encourage scientific transparency by making underlying
assumptions explicit and concrete.
Facilitate ‘theoretical reproducibility’.
Abstraction often helps separate the superficial from the
fundamental and identify the latter.
G. Taylor (OII) Modelling 47 / 48
91. Other advantages of formal models
We have seen that models are tools for logic, experimentation,
communication, and unification.
Models can help us in a number of other ways:
Encourage scientific transparency by making underlying
assumptions explicit and concrete.
Facilitate ‘theoretical reproducibility’.
Abstraction often helps separate the superficial from the
fundamental and identify the latter.
Generate theoretically-grounded and theoretically-meaningful
hypotheses for empirical work.
G. Taylor (OII) Modelling 47 / 48
92. Other advantages of formal models
We have seen that models are tools for logic, experimentation,
communication, and unification.
Models can help us in a number of other ways:
Encourage scientific transparency by making underlying
assumptions explicit and concrete.
Facilitate ‘theoretical reproducibility’.
Abstraction often helps separate the superficial from the
fundamental and identify the latter.
Generate theoretically-grounded and theoretically-meaningful
hypotheses for empirical work.
Provide structure for empirical work.
G. Taylor (OII) Modelling 47 / 48
93. Appendix: Proof
It was asserted that, for differentiable D(·, ·),
D1(p; d0) < 0, D2(p; d0) > 0 =⇒
∂
∂d0
D−1
(S(p); d0) > 0.
Suppose, on the contrary, D−1(S(p); d) > D−1(S(p); d ) for d < d. We
want to show that this leads to a logical contradiction. This and
D2(p, d) > 0 imply
D[D−1
(S(p); d); d] > D[D−1
(S(p); d ); d] > D[D−1
(S(p); d ); d ],
where the first inequality comes from applying D(·, d) to both sides
and the second follows from D2(p, d) > 0. Noting that
D(D−1(x, d), d) = x by definition, we have x > D[D−1(S(p); d ); d] > x,
which can never be true.
G. Taylor (OII) Modelling 48 / 48