[abdullahkhalids]

That data is biased by the small minority of super active users who spend 4-5 hours/day on Twitter.

The relevant number would be to condition on only the 20th to 80th percentile users (by time spent/day) and see their breakdown. I am going to bet that number is more biased towards desktop, while both the 0-20% (occasional users) and 80-100% percentiles will be mobile focused.

The other confounding effect is the bots and the pseudo-bots (humans operating many accounts). I don't know how they change these numbers.