Model Thinking

Segregation_and_effects.md

Segregation And Peer Effects

We tend to live with people like us VS We tend to change to looks like people we live with === Sorting vs Peer Effects

• Schelling tipic model: segregation model
• Granovetter model
• Standing ovation model: peer effect
• Identification problem: is it sort effect or peer effect

Equation Based Model

Ex: mark Score = 50 + 5 Hours

Agent Based Model

• inidividuals: people, org, country ...
• behavours
• outcomes

Schelling's Segregation Model

• Dev by Schelling (guy) studying segregation in New York

• It is an Agent base model

• People choosing where to live. => Should I stay or should I move

• Let's look at how many people are like me in the neighbourhood and depending on the percentage, people decides to stay or to move. This percentage is a threshold

• Netlogo model (free software)

• Micro leve 40% similar-wanted => Macro level: 79% are similar in there neighberhood!!!

Micromatives != Macrobehaviour

• Tipping phenomena (basculement)
• Exodus tip: someone like you leave => you leave (reached threshold)
• Genesis tip: someone moves in that isn't like you => you leave (reached threshold)

Measuring Segregation

index of dissimilarity: |b/B - y/Y| where b is the number of B in the block and B is the total number of b (same for y) this is the index per block.

• 0 => no dissimilarity
• 1 => fully segregated

index of dissimilarity: 0.5 * SUM |b/B - y/Y|

Peer effect

• Granovettor's Model
• Hard to predict, to anticipate people ambrassing some cause

Model

• N individuals

• Each has a threshold

• Each person needs that there are at least N person in the movement to get involved

• Collective action is more likely to happen if

• Lower thresholds

• More variation in thresholds accross people

it is hard to predict because you need to know how much people is discontent, and the repartition of discontent and their connection

The Standing Ovations Model

• Peer effect (people stand, so you stand)

• Information (someone seems to know more than you, so you use this person to know how good the show is)

• Threshold to stand T is now linked to the quality Q of the show

• Signal S = Q + E where E stands for error

• if S > T you stand

• Subsequent rule

• if more than X% stand, you stand

• Higher Q, more people stand

• Lower T, more people stand

• Lower X (Larger peer effects), more people stand

• Big X means, you are very secure in what you think

• Low X means you are a follower

S = Q + E

• E = Error
• E = Diversity of the way of judging the quality

Example

1000 People
T = 60    
Q = 50    => 50 < 60 so no one stands

• if now E [-15;15] few people stands

• if now E [-50;50] 40% people will stand (if people are equi reparty in this range)

• Why E could be big:

• Diverse Audience, Unsophisticated Audience

• Multidimensional Performance, too complex ...

• More variation, more people stand

Ovation Advanced

• the theater itself has an influence
• you usually go with a group

Theater influence

Your position in the theater will make you see more or less people in the audience and you are more or less seen. This affects peer effects.

Group influence

If your group is standing, you are more likely to stand.

###How to increase Standing ovation probability:

• Higher Quality
• Lower Threshold
• Larger Peer effects
• More variation
• Use celebrity (guys at the first row in the theater because when they stand, everybody see them standing so it increase peer effects)
• Big groups

Fertility of model / Other applications

You can use this model to understand collective action, academic performance, urban renewal, fitness/Health, online course

The Identification Problem

• Is something happenning due to Scehlling (Homophily) or Peer effect?
• If sorting: people move from one group to another, by actually moving: people doesn't change, they just move.
• In case of peer effect: people change. This harder to detect!

You cannot decide if this is sorting or peer effect based on a snapshot

Use_And_Understand_Data.md

What the point?

1. Understand Patterns
2. Predict Points
3. Produce Bounds: like a range [1%, 3%]
4. Redtrodict: see if your model works on old data
5. Predict other stuff: like data shows that there is something that you cannot measure. It has a consequence on other stuff
6. Inform data collection: gather useful data to build your model
7. Estimate Hidden Parameters: extract from data information the data itself doesn't give you. Like evolution of the data
8. Calibrate: Get estimation of the conditions for something to happen in a precise way

whyModels.md

Why Models

Name the part

Detect everything that will participate to the model

Example

Where do people go to eat

• Restaurant
• cheap/expensive
• type of food
• People
• money
• the time they have to eat
• kind of food they like

Identify relationships between the part

Apply rational thinking to link them

Inductively expore

What is the problem? What happen if I do that? What happen if I change this?

Understand class of outcome

• Equilibrium
• Cycle
• Random
• Complex

Identify logical boundaries

Under which conditions my model is true

Use it to communicate

It can be easier to explain stuff using a model