Page 145 - Invited Paper Session (IPS) - Volume 1

IPS122 Elise C. et al.
2. How does the SSP Lab function in practice?
2.1. Composition, governance and resources
The SSP Lab team is made up of eight full-time datascientists,
econometricians and IT specialists, including its chief and deputy. The idea is
to group complementary profiles in terms of skills and experience (seniors /
juniors).
The governance of the SSP Lab is ensured by the members of the executive
committee of INSEE and the chiefs of the Ministerial Statistical Services, who
review once a year the performance of the Lab experiments and provide global
directions for future programmes. These directions are flexible enough to
ensure reactivity and innovation on unexpected topics. Consequently, the
investment decisions are made continuously.
The SSP Lab works in close collaboration with the IT innovation unit,
notably through the use of the new IT platform that allows distributed
computing and other treatments adapted to the needs of Big Data
manipulation. The current version of this platform is unfortunately still
incompatible with the security requirements of individual data. Consequently,
only anonymised samples are used in the experiments.

2.2. Activities
The SSP Lab’s activities involve experimental projects, networking activities
and dissemination.

a. Experimentation
The Lab conducts applied research and experimental developments
involving statistical or datascience innovation in partnership with services in
charge of production. The experimental projects are defined as follows. The
subject is proposed and sponsored by a business unit. The project should last
around six months. More generally, the idea is to define several stages that
entail specific deliverables. The approach is exploratory, on a "small" scale, and
similar to the lab testing approach. Consequently, the deliverables may be of
different types: proofs of concept (POC), reports, experimental prototypes,
etc., but cannot be directly integrated into a production process. Ultimately,
they may not even be implemented. Experimental outcomes never being
certain.
Several conditions must be fulfilled to launch an experimental project
involving the SSP Lab and a business unit. As already mentioned, the needs
are expressed by the business unit, should be in keeping with the orientations
previously mentioned, and should concern (at least) one of the Lab's fields of
action (new data sources, new statistical methods or new statistics). In
addition, the SSP Lab should have the appropriate skills or should be able to
acquire them rapidly.

134 | I S I W S C 2 0 1 9