New paper (in english) just out in French Journal of Sociology! with T. Louail and I as corresponding authors, and A. Beaumont, J.-S. Beuscart, S. Coavoux, P. Coulangeon, R. Cura, B. Le Bigot, M. Moussalam and C. Roth as co-authors
shs.cairn.info/tap-bj77312m...
hal.science/hal-04448365....
1/10
We present the design and descriptive results of a large-scale research on online music consumption from the ANR research project RECORDS led by T. Louail, in partnership with Deezer R&D. The main specificity of this research is that we combine three types of data, following a nested strategy:
2/10
- The logs of the listening practices of a large sample of Deezer users.
- Socioeconomic infos of a sub-sample of these users via a questionnaire
- Life story and opinions on music from interviews of a sub-sub-sample of these respondents.
Many fun things in this paper:
3/10
1) Survey respondents who agreed for an interview were disproportionately highly educated males aged 35-44 (the modal characteristics of our research team...). To overcome this limitation, we stratified our pool of interviewees by these characteristics (one per cell of table D).
4/10
2)Comparing the streams of a 35-44 y.o. random sample of users with our survey respondents of the same age, we find that our respondents over-listen to rock artists of the past century, and under-listened contemporary French rap ones. More a Radiohead/Muse sample than a Jul/Aya Nakamura one ...
5/10
Mar 4, 2025 22:183) Zooming on streams of our 35-44 y.o. respondents, we find that artists with the most educated audience tend to be in classical and jazz; those with the least are French rap ones. Artists with most masculine audience are men-only artists; those with most feminine audience are more mixed.
6/10
The highlighting on this figure of the rock (triangles) and rap (squares) artists that were strongly over/under listened to in the previous figure provides with further hints of an education-based self-selection of our sample.
7/10
4) Finally, we used the log histories of our interviewees to elicit reactions on songs we predicted they would not like. Here are some of their reactions:
8/10
These different results highlight how such design can contribute to important improvements:
- Learning about distortions in samples,
- Studying, at large scale, the social stratification of recorded practices instead of self reports,
...
9/10
...
- Using less equivocal categories (artists/songs rather than genres)
- Getting "gut" reactions on actual cultural products, rather than abstract opinions.
Cheers!
10/10
