JMP🧵
Social networks help us deal with -ve economic & health shocks. But what if people worry about norms & do not ask for help? My JMP combines a RCT+ survey experiments+ structural model to show how incorrect beliefs about peers lead to silent networks
JMP🧵
Social networks help us deal with -ve economic & health shocks. But what if people worry about norms & do not ask for help? My JMP combines a RCT+ survey experiments+ structural model to show how incorrect beliefs about peers lead to silent networks
Life update: I defended my doctoral thesis this summer and I'm super excited to begin working as a junior research fellow at
@MertonCollege
, University of Oxford starting tomorrow!
My new working paper on learning among idiosyncratic agents in networks is now on SSRN! Individuals can be slow to update their beliefs, may overreact, be frustrated, or spread misinformation. How does this affect social learning? Summary🧵👇
Had a great time presenting my job market paper
@HarvardCID
and attending so many amazing sessions. Many thanks to the organisers!
@aikhwaja
#NEUDC2023
Delighted to discuss my research with
@srajagopalan
on the podcast
@IdeasofIndia
@mercatus
! We spoke about my job market paper that shows how pessimistic beliefs about peers can reduce useful social interactions in slums in urban India. Listen to it here:
In the latest episode of the 2023 job market series of
@IdeasofIndia
, I speak with
@vatsalecon
, Junior Research Fellow JRF
@MertonCollege
@UniofOxford
about the impact of pessimistic beliefs, correcting such beliefs, social norms, support groups, and more.
I just discovered that much like Stata's beep command,
@MATLAB
too has a set of audio files which can be played to signal the end of a code/iteration. So now, after every iteration of my simulation, my code commemorates Handel.
My AEA 5k run in a surprisingly sunny Oxford
#aea5k
The usual (and alternative routes) to uni parks were flooded so I had to backtrack. Turns out that I'm not very good at executing the depth first search algorithm in real time.
I recently spoke to the amazing
@ben_golub
about his paper "Signaling, Shame, and Silence in Social Learning" on the podcast Behavioural Science Uncovered
@bsuncoveredpod
. You can listen to the episode here
How willing are people to engage in dialogue around mental health and financial health?
@vatsalecon
(
@MertonCollege
&
@OxfordEconDept
) tackles this question in relation to communities in New Delhi, India, at today’s CSAE Research Workshop.
JMP🧵
Social networks help us deal with -ve economic & health shocks. But what if people worry about norms & do not ask for help? My JMP combines a RCT+ survey experiments+ structural model to show how incorrect beliefs about peers lead to silent networks
It was an absolute delight to present my job market paper at the CEPR-TCD-TIME conference in Dublin! Thank you so much for the amazing conference
@cepr_org
@tcddublin
@Trinity_TIME
The Oxford summer school in economic networks is in session! For those of you who haven't registered, you can still attend Matt Jackson's public lecture on the dynamics of human networks this evening
#oxnet19
@OxfordEconNet
"Imagine a world where economists [...] read the literature of a country before making their recommendations for it." An important & interesting article by
@kadambari_shah
, though the title certainly is a bit of a stretch. via
@qz
Today, I taught myself how to make aesthetically pleasing presentations using LaTex
@overleaf
. While I may be found looking wistfully at the powerpoint icon on my computer screen every once in a while, I don't think my life will ever be the same again.
I also have immense gratitude for Peyton Young whose palpable excitement while teaching a course on evolutionary game theory inspired me and has reinstated my faith in the research process over the years. And my undergrad professors who saw a doctor in me much before I did.
It is fascinating to revisit simple linear algebra "facts" from an intuitive, geometric (ex-post obvious) perspective. I was prepping for teaching next week and stumbled upon this gem. TL;DR- the determinant of a matrix = the volume of the parallelepiped formed by its vectors.
Re-upping this terrific video about Pampa Dey, who overcame barriers to get her PhD.
#DayOfTheGirl
.
@BruceWydick
called it "the best 2-minute development economics documentary I have seen on aspirations... Will give a boost to your day!"
We focus on the production function of the paper-- the genesis of the idea and how it evolved over the years. Ben also shares helpful advice for junior researchers including his views on coauthoring, working on multiple projects, and writing theory.
I couldn't have done this without the advice, support, and kindness of my amazing supervisors Stefan Dercon and Alex Teytelboym
@t8el
. Many thanks also to
@Oxford_CSAE
and
@OxfordEconDept
for their continuous support.
See evolution of beliefs for a random scale free graph with 100 agents with random low/med/high IDs. Higher IDs are always bad (reduce speed of convg, or make beliefs diverge) but high self weights can compensate for high peer IDs. 8/n
Looking forward to a lot more learning and improving as I continue both my theoretical work and field experiments on social networks and economic development. Watch this space :D
Very helpful (and relevant) post by
@Razan_Amine1
on Stata's "mi" commands that use simulation-based methods to impute missing observations and run analysis on the imputed dataset.
🌟NEW! CSAE
#Coders
' Corner post 'Dealing with Missing Observations: Multiple Imputation'. Experiencing missing observations or attrition in your empirical research? Then look no further...🤓
✒️
@Razan_Amine1
(
@OxfordEconDept
)
⬇️👉
This process gives birth to, as he describes, "a network of super black boxes". Thus, conspiracies, according to him, aren't planned as we think they'd be. Conspirators are often unaware of the fact that they're conspirators. (4/4)
@causalinf
Those of us working on networks (and I assume spatial econ) are forced to move to R or MATLAB because the environment is more hospitable. It isn't just FOLLI (as coined by
@bradfowd1
) because in this case, we'll probably end up looking more intelligent if we use Stata.
An interesting paper by Batista, Fafchamps and Vicente with some very surprising results.
They find that more information sharing takes place under anonymity and even when some attributes are known, people aren't any more likely to share info with those like them: no homophily!
An experiment on information sharing through the use of text messages in rural Mozambique finds that sharing throughout social networks is most likely when the messages are anonymous and simple
I derive bounds for bottlenecks in regular networks as a function of how sq biased/attention constrained agents are. Agent's "current" and "past" selves, both, need to be taken into account. I find that sq bias reduces convergence speed while connectedness increases it. 13/n
Super-interesting paper by
@mrcpangallo
, Torsten Heinrich and Doyne Farmer. They use simulations to show that the no. of best reply cycles & fixed points in a perfect-information, simultaneous game can be used to predict convergence to equilibrium.
I use their approach & find that a rise(fall) in frustration(overreaction) *can* prevent a seq. of growing networks from coming close to the truth. This is because change in IDs can increase the influence of an agent EVEN IF they become less popular in the network. 10/n
@PRSLegislative
has written an overview of the contentious bill okayed by the union cabinet today. It should (hopefully) be declared unconstitutional by the SC but the need for an informed, civil dialogue among the electorate has never been more critical.
They’re not necessarily those with high conventional measures of centrality. See e.g. from a village in the Banerjee et al 2013 data with randomly endowed IDs & degroot weights.The large green circle is the maximal agent- low degree but crucially, v low weight on own beliefs. 6/n
@NicDuquette
@ben_golub
Reminds me of the majority illusion paper by Lerman et al where they find that "..a state that is globally rare in a network may be dramatically over-represented in the local neighborhoods of many individuals." Scarily enough, they also find that it enhances contagion.
Reminds me of this excerpt from Peyton Young's classic "Individual Strategy and Social Structure" (pg 4)- "we need to recognize that the dust never really does settle- it keeps moving about, buffeted by random currents of air."
I've been getting used to gganimate and thought it would be useful to put together some illustrations of what various causal inference methods *actually do to data* and how they work. Here, for example, is what it means to control for a (binary) variable
Amazing article by
@PennyMealy
&
@DianeCoyle1859
on increasing informed dialogue about policy issues by using comics.
@PennyMealy
's The Sipster, for example, is the story of a super heroine who uses research to fight climate change & democratises knowledge in the process.
@KNargish
@sydneydxb
Oh yes he does substantiate with examples. In fact, this entire discussion starts after he describes a rather dramatic meeting with Larry Summers. The book isn't very academic in nature. It's ecopolitical commentary at its best and sensational gossip at its worst.
This (beautiful) paper plots 16th century financial networks and lists their features- some of which, the authors note, render these networks very similar to their modern counterparts!
@cxdig
I've always been astounded by the beauty of the Watts-Strogatz paper but didn't know about the crickets. Thank you
@markdhumphries
for this enthusiastically written blog post!
Building models is probably like finding the correctly sized map. One must not let them get too enmeshed in the details (i.e. models with very few or no assumptions) but must also ensure that they're not entirely devoid of them (i.e. a model with way too many assumptions).
I usually don't enjoy Yanis Varoufakis's histrionic and metaphor-intensive style of writing. However, in the introduction of his book "Adults in the Room", he surprised me by presenting the following fascinating argument. (1/4)
“Historically, pandemics have forced humans to break with the past and imagine their world anew. This one is no different. It is a portal, a gateway between one world and the next.”
This thought provoking article by Jean Drèze lists 4 reasons why the evidence-policy transformation may not be as straightforward. I disagree with what he insinuates about economists in one of the paragraphs but the rest is worth reading. via
@Ideas4India
He says that what may seem like a planned political conspiracy on the outside may actually be the result of a network formation process where politicians/journalists/people in power, form links as they trade secrets. (2/4)
I answer 3 qs- how does an individual’s ID affect (1) whether society learns & reaches a consensus in beliefs, (2) if it learns accurately, and (3) how quickly it learns. I relate these learning outcomes with *both* the network and behavioural aspects of the environment. 2/n
The silver lining, as we show, is that belief correction can be used to generate the demand and funding for these costlier policies i.e. belief correction can allow communities to self finance policies that increase useful network interactions in underfunded settings.
Every once in a while, I remind my academic self of Borges's pithy story titled "On Exactitude in Science"- the tale of an empire whose over-scrupulous cartographers forget that a map, no matter how precise, isn't of any use if it is the size of the very area it describes.
You can find more details in the paper here You can also listen to my podcast about this paper with
@srajagopalan
for Ideas of India in case you're interested!
We find that beliefs about others' willingness to engage are correlated with own frequency of dialogue around financial & mental health concerns, num links in financial & mental health support networks & own ability to smooth shocks. Can we improve outcomes by correcting beliefs?
I also discuss examples of overreacting networks where beliefs may converge but to the wrong level. This happens when agents don’t receive any informative private signals that would otherwise bind them to the truth in the steady state. Example in the picture- 11/n
A calendar of 'mathsy moments' to live in, and share. Daily date-related number play and theory, notable anniversaries and celebrations. From Aug-18 to Dec-19, with links to explore and in various formats to download, edit, import. Enjoy:
#mathschat
#maths
@causalinf
@aaron_hedlund
UxxUyy -Uxy^2 is the determinant of the Hessian which should be >0 for both convex and concave functions (essentially to ensure that there are no saddle points). I don't think concavity can be inferred solely from that condition.
@paulgp
@jhaushofer
Not on health, but also this paper where they compare seeding v/s broadcasting info about India's demonetisation and vary provision of meta-info about who was told at the start -> to see what works if people worry about signalling while seeking information