Seminar

Interlocking

Quantitative surveys. Social Science Toolkit

Wednesday 18 December 2024 Wednesday 18 December 2024

From 3 to 6 PM

Danielle Navarro, Generative Art, licence CC BY 4.0

ENS, salle de conférence du centre de science des données

ENS-PSL
45 rue d'Ulm
75005 Paris
France

48.8418371, 2.3440403

Last session of the course "Enquêtes quantitatives. Boîte à outils pour sciences sociales", given by Théo Boulakia.

Any incomplete toolbox runs the risk of turning gold into lead. At the foundation of commonly taught statistical models lies a strong assumption: your observations are independent. In practice, this condition often fails to hold. And that's a good thing. Panel data, for example, which contain multiple observations for each individual, are extremely rich. If you approach them with standard tools, not only will you miss out on their potential, but you'll often get bizarre or counter-intuitive results. Fortunately, there are models tailor-made for this kind of data. These are known as hierarchical, mixed effects, multilevel or random effects models. The principle is straightforward, and the benefits immediately apparent. We'll present a number of applications, and show how they can be implemented in a Bayesian framework.

Wednesday 18 December 2024

Quantitative surveys. Social Science Toolkit

S1 2024-2025.

Data analysis course in social sciences, delivered by Théo Boulakia. The sessions take place on Mondays, from 3 p.m. to 6 p.m., in the conference room of the Data Science Center (ENS-PSL, 45 rue d'Ulm, at the top of stairs B or C).

Goals of the course: This course offers an introduction to various quantitative methods based on social science surveys. Each session is organized around the meeting between a question (sociological, anthropological, historiographical), data and a method: cartography, dimensionality reduction, partitioning, sequence analysis, textual analysis, Bayesian statistics. The objective of the course is to acquire a schematic understanding of the implementation of these methods, their merits and their limitations. How to represent spatial data and temporal sequences? What do Bayesian statistics provide that the frequentist approach lacks? How to analyze the morpho-synthactic properties of a text? How to go from a large number of variables to a small number of classes? These questions will arise in context, in a dynamic of adjustment between data, method and research question (an investigation dynamic). Programming questions will be covered only in broad terms, no experience in this area is required.

Validation: Submit a four-page document applying one of the methods discovered in progress to your own data: presentation of the data, interest of the method, implementation and interpretation. People without programming experience will be assisted with implementation.

Registration : theo.boulakia [a] ens.psl.eu