View by Day

Filipe Campante, Ph.D.

John F. Kennedy School of Government, Harvard University, USA

________________________________________

Authors: Pedro Bessone, Filipe Campante, Claudio Ferraz, and Pedro CL Souza

Abstract: Recent years have witnessed the remarkable diffusion of social media, such as Facebook, in tandem with the spread of the cell phones that have become the key tool for access to those media. We ask whether this has affected the accountability of politicians, in a context where the coverage of local politicians by traditional media is negligible.

     Using data on the spread of 3G cell phone networks across municipalities in Brazil, we implement a triple-differences strategy to show how legislators respond when municipalities that are part of their electoral base obtain access to the 3G technology. The reaction takes place both in their social media activity – they increase the number of Facebook mentions to the municipality – and in their legislative activity – they decrease the amount of earmark transfers to the municipality and mentions to those municipalities in Congressional speeches. We thus find direct evidence of substitution between the online and offline types of responses. We also show that citizens increase their social media interaction with politicians, as measured by Facebook “likes,” “shares,” or “comments.” Taken together, our results suggest that the combination of social media and mobile phones creates the potential for substitution between online and offline behaviors of the politicians.
________________________________________
Full Text:  WILL BE AVAILABLE LATER

As a part of the Week of Science and Technology, CERGE-EI is opening up to public with guided tours of the Schebek Palace on November 12 and November 13, 17:00 – 18:00. CERGE-EI Library is also open for visits, from November 11 till November 14, 10:00 – 18:00.

For attending the tour, registration is necessary.

The Week of Science and Technology, organized by the Czech Academy of Sciences (11-14 November 2019), is the most extensive science festival in the Czech Republic. The program offers more than 500 events all over the Czech Republic. For more information visit www.tydenvedy.cz

Wednesdays 18:00 - 19:30 
Starting date: 13 November 2019
Finishing date: 4 December 2019
Duration: 4 lessons
Course instructor: Ondřej Zahradník

Registration for this course is closed.

Please read about our selection process. Follow EDH on Facebook for the latest news and tips.

There are many online courses explaining different pieces of ML. Very few of them answer how to really apply ML in production problems. We share some best practices on how to build ML system. All built with simple examples in Python but general enough for non-pythonists too.

This course will cover:

1. Data

Given data it is quite simple to build some models. We go through typical modeling workflow from data exploration, feature engineering, to modeling on a complete worked-out example, discuss options and tradeoffs.

2. ML in Production

We test several models in laboratory conditions. We try to extract some knowledge from the model to help us with stakeholders’ buy-in. We move the best model from messy notebooks into production. We give an overview of techniques used to ease transition from development to production and how to keep the model running well.

3. Experimentation

There is no improvement without failures, we have to know what works and what does not. We give examples of basic techniques to run controlled experiments and learn from them. We help to communicate results in natural language and how to get most of the value from the experiment using Bayesian approach.

4. Deep Learning

Deep learning helps where traditional techniques stop. It does not need to be too difficult and technical to implement. We give an example of a problem solved using deep net, what are common pitfalls and how to evade them.

Participants who attend at least 75% of the sessions can claim a Certificate of Attendance issued by CERGE-EI.

We thank our partners for supporting the Economics Discovery Hub.