Using Twitter lists to balance my attention: a slow motion self-experiment

Area chart from top: grey section diminishing is size; thin purple line; blue sction diminishing in size; pink section increasing in size

If I’m going to live in a filter bubble, it might as well be one that I have intentionally constructed for myself.

For a few years now I’ve been using tools like to keep a check on the gender balance of the people I follow.

nonbinarymenwomenno gender, unknown
People you follow1%50%49% 
Guessed from name0537464443
Declared pronouns14130188 results for @mattedgar – Sampled 1776 people you follow, 3000 followers and 110 users from the latest 200 tweets in your timeline. Gender estimate based on 729 Twitter bios with declared pronouns like “she/her” and 2546 genders guessed from first names.

But does this actually make a difference to where I direct my attention? I figured that if people I follow is an input measure, then people I retweet might represent the output, an indicator of what catches my eye, and whose voices I amplify.

Does following a 50:50 split of men:not-men translate into 50% of retweets to women and non-binary people too? As it turns out… not quite.


A couple of Christmases ago I downloaded my Twitter archive, inexpertly hacked the JSON file of my tweets into a spreadsheet, and analysed who I was retweeting.

nonbinarymenwomenno gender, unknown
Share of 2018 retweets2%45%32%21%
Number of 2018 retweets15329235154

Over the course of a year, despite my efforts to follow a balanced group of people, tweets by men still took up a disproportionate share of the things I chose to retweet. (Virtually all of the “no gender, unknown” category are accounts of organisations and other non-humans.)

I could speculate about the reasons for this – More talkative men? A different mix of topics? My own implicit bias about what’s retweet-worthy? – but how to fix it?

I used my spreadsheet to find a sub-Dunbar number group of people whose tweets I knew I always wanted to see. To help correct for the imbalance in my previous year’s retweets, I set different thresholds for men and women and made a Twitter list:

Most retweeted in 2018: 44 women/NB people I retweeted 2 times or more in 2018; 27 men I retweeted 3 times or more in 2018

This became my go-to list when checking Twitter. Using a list this way had the added benefit that I was no longer prey to the algorithmic whims of Twitter’s “home” view.


A year later, I repeated the analysis.

nonbinarymenwomenno gender, unknown
Share of 2019 retweets1%46%37%16%
Number of 2019 retweets8414330147

My number of retweets of women had increased a little, from 32% to 37%, but retweets of men had also gone up. For the next year, I doubled down on the approach, raising the threshold for men to get onto my regular reading list:

Most retweeted in 2019: 56 women/NB people I retweeted 2 times or more in 2019; 20 men I retweeted 4 times or more in 2019


I’ve just completed analysis for year 2 of my experiment:

nonbinarymenwomenno gender, unknown
Share of 2020 retweets1%45%40%14%
Number of 2020 retweets5414366128

Compared to 2019, retweets of women is now up to 40%, compared to 45% retweets of men. While things are moving in the right direction, I’m not there yet.

For 2021, I’m sticking with the same formula for list creation, with a small tweak that brings the black, Asian and minority ethnic (BAME) representation in the list up to 14%, about the same as in the UK population:

Most retweeted in 2020: 67 women/NB/BAME people I retweeted twice or more in 2020; 18 men I retweeted 4 X or more in 2020

Where next?

There are limits to this approach.

  • Like, I’m making assumptions about some people’s genders, so it’s possible that in some cases I’ve got it wrong. For that reason I’m not sharing names by gender, only the overall numbers.
  • My analysis so far focuses only on gender. I ought also to consider other kinds of diversity among the people I follow.
  • I’m missing the toots I boost on Mastodon, the open-source, self-hosted social networking service. If I put more time into this, I would work out a way to include those too.

In the meantime, this is my filter bubble. I plan to re-run the numbers in December 2021.

28 December 2021 update

I re-ran the numbers for 2021 using the same method, and they’re not much different from the previous year:

nonbinarymenwomenno gender, unknown
Share of 2021 retweets0%43%39%17%
Number of 2021 retweets4446399180

Here’s the list I’ll be following for the next 12 months:

Most retweeted in 2021: 73 women/NB/BAME people I retweeted twice or more in 2021; 27 men I retweeted 4 X or more in 2021

One thought on “Using Twitter lists to balance my attention: a slow motion self-experiment

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s