Question: Twitter follower count distribution
Hey Fellow AoIRistas, Quick question: Do you know of any top-down measures of the distribution of follower numbers for Twitter accounts? I'm sure the curve overall looks like a standard power law distribution, but I'm interested to know whether there are emergent tiers (e.g. <250; 251-2,500; 2501-10,000; 10,001+) or something along those lines. Thanks! Aram -- *Aram Sinnreich <http://www.sinnreich.com>* Associate Professor Chair, Communication Studies American University <http://www.american.edu/soc/> School of Communication <http://www.american.edu/soc/> Author <http://j.mp/sinnreich>| Musician <http://sinnreich.com/music/> (he/him/his)
Dear Aram, that question also interests me. Based on data from two projects, I assume the distribution doesn't follow a power law, but rather a lognormal distribution (maybe MLP, modified lognormal power-law?). Though, I didn't have time to dig into it yet. See below, the first picture is based on follower counts of 350 German news outlets, the second is from a study about 1835 communication scholars (https://journals.sagepub.com/doi/full/10.1177/1461444819863413). It looks to me like there are some tiers or superimposed distributions. Sample size is somewhat small, maybe someone has insight into the distributions on a larger data basis? Cheers Jakob Am 06.05.2020 um 17:28 schrieb Aram Sinnreich:
Hey Fellow AoIRistas,
Quick question: Do you know of any top-down measures of the distribution of follower numbers for Twitter accounts? I'm sure the curve overall looks like a standard power law distribution, but I'm interested to know whether there are emergent tiers (e.g. <250; 251-2,500; 2501-10,000; 10,001+) or something along those lines.
Thanks!
Aram
-- Jakob Jünger University of Greifswald Institute of Political Science and Communication Studies Ernst-Lohmeyer-Platz 3 17487 Greifswald Germany Room: 3.16 (3. floor) Email: jakob.juenger@uni-greifswald.de Web: http://www.ipk.uni-greifswald.de/
Sorry, the images were stripped out (thought I saw image attachments on the list before): News outlets: https://tinyurl.com/y9xsk8s3 Comm scholars: https://tinyurl.com/y7jnx348 Am 06.05.2020 um 21:31 schrieb Jakob Jünger:
Dear Aram,
that question also interests me.
Based on data from two projects, I assume the distribution doesn't follow a power law, but rather a lognormal distribution (maybe MLP, modified lognormal power-law?). Though, I didn't have time to dig into it yet. See below, the first picture is based on follower counts of 350 German news outlets, the second is from a study about 1835 communication scholars (https://journals.sagepub.com/doi/full/10.1177/1461444819863413). It looks to me like there are some tiers or superimposed distributions. Sample size is somewhat small, maybe someone has insight into the distributions on a larger data basis?
Cheers Jakob
Am 06.05.2020 um 17:28 schrieb Aram Sinnreich:
Hey Fellow AoIRistas,
Quick question: Do you know of any top-down measures of the distribution of follower numbers for Twitter accounts? I'm sure the curve overall looks like a standard power law distribution, but I'm interested to know whether there are emergent tiers (e.g. <250; 251-2,500; 2501-10,000; 10,001+) or something along those lines.
Thanks!
Aram
considering the number of followers is a count variable (non-negative discrete numbers), is is either a poisson, negative-binomial, or sero-inflated distribution log-transforming the distribution for further (statistical) analysis is not advised. best Maurice ________________________________________ Van: Air-L <air-l-bounces@listserv.aoir.org> namens Jakob Jünger <jakob.juenger@uni-greifswald.de> Verzonden: woensdag 6 mei 2020 21:44 Aan: Aram Sinnreich; air-l@listserv.aoir.org CC: Marie-Luise von Berg Onderwerp: Re: [Air-L] Question: Twitter follower count distribution Sorry, the images were stripped out (thought I saw image attachments on the list before): News outlets: https://tinyurl.com/y9xsk8s3 Comm scholars: https://tinyurl.com/y7jnx348 Am 06.05.2020 um 21:31 schrieb Jakob Jünger:
Dear Aram,
that question also interests me.
Based on data from two projects, I assume the distribution doesn't follow a power law, but rather a lognormal distribution (maybe MLP, modified lognormal power-law?). Though, I didn't have time to dig into it yet. See below, the first picture is based on follower counts of 350 German news outlets, the second is from a study about 1835 communication scholars (https://journals.sagepub.com/doi/full/10.1177/1461444819863413). It looks to me like there are some tiers or superimposed distributions. Sample size is somewhat small, maybe someone has insight into the distributions on a larger data basis?
Cheers Jakob
Am 06.05.2020 um 17:28 schrieb Aram Sinnreich:
Hey Fellow AoIRistas,
Quick question: Do you know of any top-down measures of the distribution of follower numbers for Twitter accounts? I'm sure the curve overall looks like a standard power law distribution, but I'm interested to know whether there are emergent tiers (e.g. <250; 251-2,500; 2501-10,000; 10,001+) or something along those lines.
Thanks!
Aram
_______________________________________________ The Air-L@listserv.aoir.org mailing list is provided by the Association of Internet Researchers http://aoir.org Subscribe, change options or unsubscribe at: http://listserv.aoir.org/listinfo.cgi/air-l-aoir.org Join the Association of Internet Researchers: http://www.aoir.org/
Thanks for the advice, may I follow up on this? My first thought was poisson, too. But poisson distributions (and gaussian) assume the events occur independently, as far as I know. And that should not be true for follower counts because, I guess, there is some kind of (nonlinear) preferential attachment process behind this (e.g. http://networksciencebook.com/chapter/3#not-poisson). Might be I missed something. Maybe someone knows studies that fit distributions of social media metrics? Am 06.05.2020 um 21:53 schrieb Vergeer, M.R.M. (Maurice):
considering the number of followers is a count variable (non-negative discrete numbers), is is either a poisson, negative-binomial, or sero-inflated distribution log-transforming the distribution for further (statistical) analysis is not advised.
best Maurice
participants (3)
-
Aram Sinnreich -
Jakob Jünger -
Vergeer, M.R.M. (Maurice)