Models Class Session 13

Session 13: Path Dependence

13.1 Path Dependence
Path Dependence: Urn Models:
 What happened along the path determines  A class of models to help describe path
what happens now. History matters! dependence.
 In contrast to Markov Processes where path  Flesh out the logic what is path dependence,
history does not affect the equilibrium what causes path dependence, distinguishing
(statistical) state. History does not matter. different types of path dependence.
 Equilibrium, what happens in the long run.
 Distribution over all possible outcomes could be
path dependent
Path Dependence Example – Typewriter
Keyboard:
 Today the keyboard is QWERTY layout (next
to top row of keys – letters)
 Lots of initial layouts proposed but QWERTY
emerged in a ‘process of increasing returns’.
It became a standard as more using caused
more to want to use.

Path Dependence: Examples: 1) technology - e.g., AC vs DC, gas vs.

electric cars, 2) common law – e.g., precedence
 Outcome probabilities depend upon the where past law influences present law, 3)
sequence of past outcomes. institutional choices – single-payer vs multi-payer
healthcare system, pensions (defined benefits or
 History matters (in a probabilistic sense). defined contributions), 4) economic success.

Economic Success Example:

 Ann Arbor: Largest Public University
 Jackson: Largest four-walled prison
 Both in Michigan, 50 miles apart, with past
choices influencing present success.
 Note the influence of crime in the 1920’s
which grew Jackson, then the war which
reduced both, and then post WWII GI bill
grew Ann Arbor.
Increasing Returns: Increasing Returns Path Dependence:
 Ann Arbor hosts a University. Then  Logically completely separate concepts
hospitals, law schools, other educational Chaos Path Dependence: ESTIC – Extreme
things emerge to reinforce growth. Sensitivity to Initial Conditions. Chaos deals with
 Virtuous cycle – good building on good. initial points. PD on the path.
Markov Processes:
Recall: 1) Starting point doesn’t matter, 2) Path doesn’t matter, 3) History doesn’t matter
 What is violated is the ‘fixed state transition probabilities’.
 In the URN models, transition probabilities do change and that is why history matters.

1
Models Class Session 13

13.2 – Urn Models

Path Dependence (PD):
 Use Urn models to describe path-dependent versus phat-dependent (order doesn’t matter but
the set of things matter – [recall “n choose k” from discrete mathematics]). (≠PATH) = PHAT
 Also help us distinguish process outcomes in a given period that are path dependent and
processes that have equilibriums – a long run distribution over a set of outcomes, that are path
dependent.
 Will distinguish between path-dependent outcomes and path-dependent equilibrium.
Fixed # of
balls,
describes
roulette,
blackjack

Used to
describe
PD.
Simplest:
Polya &
Balancing

selected ball & new ball both back selected ball & new ball both back
to urn, P(R)n changes & is PD to urn, P(R)n changes & is NOT PD
Basic Urn Model: Urn contains Result 1: Any probability of red Result: The balancing process
balls of various colors. The balls is in equilibrium and is converges to equal percentages
outcome is the color of the ball equally likely. We could get 4%, of the two colors of balls.
selected. 23%, or 99% red balls. Shirt That is long run is 50% red and
choice example illustrates Polya 50% blue so is NOT PATH
process and history dependent. DEPENDENT
Result 2: Any history of B blue Application: Keeping political
and R red balls is equally likely. constituencies happy. Where to
P(RRRB)=P(BRRR)  frequency host the Olympic games.
of occurrence says nothing
about order. Observation: Equilibrium
Observation: percentage of red balls in the
percentage Period Outcomes: color of ball long run.
Uniform versus 50% long run in a given period
Path Dependent Perspective:
 Path Dependent Outcomes: color of ball in a given period depends on the path, or
 Path Dependent Equilibrium: percentage of red balls in the long run depends on the path.
Path Dependent Equilibria Path Dependent Outcomes
Polya Yes YES
Balancing ½ YES
History can matter at each moment but not have any impact in the long run. Depends on the
underlying process that is in effect. Examples: Manifest Destiny: (growth of US). Railroads: tracks
may follow one another but may also have been defined by economic routes. In short, if there is an
equilibria in the long-run, then the process is path independent. ???

2
Models Class Session 13

PHAT Dependent: Outcome probabilities depend upon past outcomes but not their order.
Polya is PHAT – Outcome probabilities do not depend upon the order in which outcomes occurred.
They only depend on the set of outcomes that occurred. RRBR  BRRR
Why the concern: exponential growth in path possibilities vs. linear growth in outcomes.
e.g., thirty rounds of 2 choices  230 possible outcomes vs. just 31 sets (Scott’s logic, any order of ball
selection that has the same ratio of red/blue balls). Permutations [2n] vs combinations [n choose k].
Two Questions:
1. Could something actually be path-dependent, depend on all this going different paths?
2. If there is one, can I construct it using my simple urn model?
Ans: Yes, there is a simple urn model called ‘the sway’
The Sway: U = {1 Blue, 1 Red}, Select and Return
In period t, add a ball of the same color as the selected ball and add 2t-s – 2t-s-1 of color chosen in each
period s<t.
Period 1 – pick a blue ball add a blue ball.
Period 2 – pick a red ball  add a red ball, and add
a blue ball for the blue ball picked in Period 1.
Period 3 – assume picked a blue ball add a red
ball again for the red ball picked in Period 2. Now
add two blue balls, (multiplying times two the blue
ball picked in Period 1.
Period 4 – assume pick a red ball, add a red ball.
Add a blue ball for Period 3. Add two red balls for Period 2. Add four (2x2), blue balls for Period 1.
Thus for each ball picked, I'm adding 1, then 2, then 4, then 8, then 16, then 32, as I move through
time. In short as move back in time decisions take on more and more weight.
If the past takes on more weight over time then you can get full path dependence.
e.g., first-movers, law, etc.
Summary:

Urn models can model some concepts of path vs. set dependence and the influence of ecosystem
development (my words) where the outcomes of the past can compound current and future
outcomes.

We learned how to construct a very simple path-dependent process, called the Polya process; we
constructed a balancing process, i.e., can have path-dependent outcomes in each period, but not
have path-dependent equilibrium. You can go to a particular thing each time. And constructed the
sway process – full path dependence and that showed us one way you can get full path dependence
and not just PHAT dependence is by having the weight of history increase over time, which is
something intuitively I think that a lot of historians and a lot of scholars of technology have felt was
true.

3
Models Class Session 13

13. 3 The Proofs:

Will prove both results: Will start with Result 2 and use to prove Result 1.
Result 2: Any history of B blue and R red balls is equally likely.
Example 1: Example 2:

Note: the denominator increases by 1 on each draw-replace-add same cycle. The numerator follows the number
of red (on R draw) or blue(on B draw) that are in the urn at that time, starting with one R and one B on first
draw. Observe p = ½ on every first draw.
Note: Example 2 Numerator – the order doesn’t matter as these have the same numbers one, two, one, two
multiplies in different orders. This follows from the number of the color when the color is drawn. Check with
Example 1 as well.
Quiz: If we want to find the probability of choosing a red ball, another red, then two blues, or in other words
P(R,R,B,B), what would be the numerators (in order) of the probability fraction?
[1,2,2,1]; [1,2,1,2]; [2,1,2,1]; [1,1,2,2]
Ans: [1,2,1,2]
Explanation: Any answer beginning with a 2 in the numerator must be incorrect, because there are only two
balls to start with. So that probability of choosing one - regardless of what color - must equal 1/2. Since we
want the second ball to be the same color as the first one chosen, the probability will be 2/3. In the third stage,
we want the a ball the color that has not been chosen yet; there is only one, so the probability is 1/4. The
probability of another blue is now 2/5.
Result 1: Any probability of red balls is an equilibrium and is equally likely. We could get 4%, 23%, or
99% red balls.
Argument is that p(4 B) = 1/5 but that
p(1R & 3B) in any order is 1/20 times 4
ways to get (1R & 3B) = 1/5. Similarly
p(4R) = 1/5 and p(1B&3R) = 1/5 and then
p(2R&2B), the only remaining
combination is 1 – (1/5 + 1/5 + 1/5 + 1/5)
= 1/5.
Note: the denominator increases by 1 on each draw-replace-add same cycle. The numerator follows the number
of red (on R draw) or blue(on B draw) that are in the urn at that time, starting with one R and one B on first
draw. Observe p = ½ on every first draw.
Shows logic of how any sequence has
the same probability. Note the use of
combination counting in the p(47B 3R)
case to account for all of the ways one
can draw 3 balls, (3x2x1) and there are
50 possible locations for the first R ball,
49 locations then for the second R ball,
and 48 places for the third R ball. But
the three balls are indistinguishable,
hence / (3x2x1).

4
Models Class Session 13

Note: started with the idea that maybe the number of purchases of a red shirt is driven by the
number that have previously been purchased. Any equilibium is equally likely. Any history of R and B
shirt purchases are equally likely.
Summary:
Given a simple model that the proportion of, the probability of somebody buying something is in
proportion to the number of previous people that had bought it, like the blue and red shirts. This
formal model yields two really interesting results: 1) any equilibrium is equally likely [ratio of red to
blue shirts] and 2) any history, with the same number of Blue shirt purchases and Red shirt purchases,
is also equally likely [order doesn’t matter, (n chose k)].

13.4 Path Dependence & Chaos

Path Dependent: Outcome probabilities depend upon the sequence of past outcomes.
Markov Process. A1: Finite States, A2: Fixed transition probabilities, A3: Can eventually get from one
state to any other, A4: Not a simple cycle. Markov Convergence Theorem – Given A1 to A4, a Markov
process converges to an equilibrium distribution which is UNIQUE.
Note: the Markov model is not path dependent because of A2 – fixed transition probabilities.
Thus history matters if the transition probabilities change.
Relating to Chaos – Recursive Function: Outcome at time t, x(t) & Outcome function F: X X
Example: F(X) = X+2 generates 1, 2, 5, 7, 9, 11, . . . or F(X1, X2, X3) = X4 or
Choas: Extreme Sensitivity to Initial Conditions (ESTIC): If initial points x and x’ differ by even a tiny
amount after many iterations of the outcome function, they differ by arbitrary amounts.
Tent Map (function):
X in (0, 1) with
{
The tent map is
chaotic.

ESTIC but not

Path Dependent

Quiz: Two men go for a walk. They are next-door neighbors named Roland and Micah, and they both set out
from their own doorstep. They both decide that they will walk for 20 houses and then take a break. If they stop
in front of a house with an even numbered address, they will stop walking; if, on the other hand, they stop in
front of a house with an odd numbered address, they will walk another 20 houses, with a maximum total
distance of one mile. Assume addresses are numbered consecutively from 0 on to 100. When both men have
stopped walking, Roland is a mile from home, and Micah is just down the street from his front door. This result
is an example of: (a) ESTIC (extreme sensitivity to initial conditions), (b) Path Dependence
Analysis: This is recursive with { and house are numbered in sequence. So if start at
even number will continue to the end at one mile. If odd, go 20 houses reach another odd and stop.
Ans: (a) ESTIC
Explanation: ESTIC means that the process - in this case, walking 20 houses at a time - is determined solely by
the starting point of the process. In this example, houses are numbered consecutively, and Roland and Micah
walk 20 houses at a time. This means that whoever lives in an odd-numbered house will stop each time at an
odd-numbered house, and thus have to continue walking until he reaches the limit of one mile. We can infer,
since Roland walks a mile, that Roland lives in the odd-numbered house. Micah, who lives in the even-numbered

5
Models Class Session 13

house stops after his first set of 20 houses.

The point is that the initial condition - whether each man lived in an odd- or even-numbered house, determines
the entire process. There are no factors along the way that alter either man's path (other than that initial
condition), meaning this isn't a path dependent process.

Independent (probabilistic): outcome doesn’t depend on Initial Conditions: Outcome depends on the
the starting point or what happens along the way. starting state, e.g., chaotic
Path Dependent: Outcome probabilities depend upon the PHAT dependent: Outcome probabilites depend
sequence of past outcomes. upon past outcomes but not their order.
Historians: it’s really about PATH or PHAT. Why Path over PHAT? A set of events over time in different orders
don’t make sense as an explanation of today. And a history determined by a few intial choices in an
underterministic process doesn’t make any sense either.

Path versut Fact: PATH says that sequence matters, Fact says the set
Matters. Why do they think it’s the path and not the set? Consider this
hard to imagine set of events.

When we think about these ideas. Path dependence. Fact dependence.

Independence. Sensitivity to initial conditions or chaos. What we see is
these simple models help us organize our thinking about the world might
look like. And we understand why a lot of historians focus so much on path dependence. Because it seems the
most reasonable. We also see why people who study gambling in casinos consider independence. And we see
why a lot of physicists are interested in things like chaos, because there are actual physical recursive that
produce this extreme sensitivity to initial conditions. the URN model, in particular, helps us draw bright lines
between path dependence, PHAT dependence and Independence.

6
Models Class Session 13

13.5 Path Dependence & Increasing Returns

Illustrate that path dependence and increasing returns are related but are not the same thing. Both
Sway process and Polya process generate increasing returns [positive feedback]. Balancing process
does not produce path dependence, not increasing returns, in fact, decreasing returns [negative
feedback].
Increasing Returns: Increaseing Returns≠PD: Question: Is increasing Returns
equivalent to path dependent
equilibrium?
Early gas car versus an electric car
evolution. With the rule shown at the
left, it is evident that the increasing
returns of the B (gas) balls will
overwhelm the R (electric) balls. But
increasing returns themselves do not
imply path dependence. Increasing
returns but not path dependent. P(B)
> P(R)  not path dependent where
P(nR)=P(kR) from Polya model.
Polya model is an example of Question: Could you get path dependence
increasing returns. More red without increasing returns? YES!
balls I pick, the more likely I (See Symbiots to Polya): This symbiosis
will pick more red balls in the model. Consider as a R-G set, & a B-Y set
future. with add R if R-G member is drawn and
Symbiots to Polya: Create two add B if B-Y member is drawn  a R-B
groups from four as shown Polya experiment, path dependent but
PD but no increasing returns not increasing returns. (not enough gain,
so to speak.) Path dependent but not
increasing returns.
Gas-Electric Model: IR but no PD
Symbiots Model: PD but no IR
Increasing returns is logically distinct of PD. Completely separate concepts. But can occur together.
Externalities: Another driver of path dependence. My choices affect you, positive or negative.
Public projects are BIG. They likely BUMP into each other. They CREATE externalities. Examples
include: Roads, Universities, Sewer Systems, etc.
Model: Projects {A, B, C, D, E}, Each has a value of 10, each creates externalities.
Externality Matrix: Diagonal Start: A = 10, would do
is the value of the project. AB = 10 + 10 – 20 = 0  not
Below diagonal is the AC = 10 + 10 + 5 = 25
externality between (i, j) ACD = 10 + 10 + 5 + 10 – 10  D?
projects. Consider project Start: B = 10
and order: AB = 0  not
Bottom line: Starting with A BC = 10 + 10 – 10  C not
gives different path BD = 10 + 10 + 30 = 50
dependencies of projects
than starting with B, e.g., AC A big cause of path dependence is More negative externalities create
vs BD. externalities. Dissertation: More more adverse dependencies that hinder
positive externalities  less PD. project adoption.

7
Models Class Session 13

13. 6 Path Dependent or Tipping Point?

Path Dependent: Outcome probabilities depend upon the sequence of past outcomes.
Path Dependent Outcomes: e.g., color of ball in a given period depends on the path.
Path Dependent Equilibrium: e.g., percentage of red balls in the long run depends on the path.
Tipping Points: Direct (Active) Tip and Contextual Tips (forest fire)
Note that an active tip has an association with path dependence since
and unstable node will go one way or the other, once started it will
continue to the next stable position.

Recall in tips there is an abrupt change in the outcome.

Measuring Tips: (measure of uncertainty) (a) diversity index (number of stable positions available), (b)
entropy (measure of information/uncertainty.) Recall (∑ ) (∑ )
Polya Process: Does this represent a tip system? Check diversity index for drawing four balls from the
urn. From previous calculations (shown below) could get 0, 1, 2, 3, 4 Red balls. So our diversity
calculation is 1/{(Pi)2 which equals 5 in this case. What happens after we pick the first ball which,
say, is Red? The similar calculations shows p(4R) = 0.4, p(3R)=0.3, p(2R)=0.2, p(1R)=0.1 and
( )
Recall Polya Result 1:

Any number of Red balls is an

equilibrium and is equally likely.
Therefore find probability of each
equilibrium.

Five possible outcomes {0,1,2,3,4}

Suppose first ball chosen is red ball.  Start with 2 Red & 1 Blue balls.
Possible different outcomes are? 1) RRR red balls in the next three
periods  total four reds. . 2) RRB or RBR
or BRR in next three periods  ( ) ( )( )( )( )
3) RBB or BRB or BBR ( ) ( )( )( )( )
4) BBB  ( )( )( ) . Note not equally likely!
Initial diversity index was 5 from Result 1. Based on
drawing four balls from the urn now computes to
3.33. Suggests something happened as we moved
along the path. But it is slow relative to the rapid
change seen in tipping phenomena.

So the Polya model does not represent a tipping

process.
The difference between path dependence and different [tipping] points is one of degree. When you think of
path dependence, what we mean is things along the path change where we're likely to go. So we move from,
you know, this set of things to this other set of things but in a gradual way. Tipping points mean that there's an
abrupt change, there were a bunch of things that could occur. Now we're likely to move to something entirely
different, or something that was unexpected, or that uncertainty got resolved.