Tweets of a Stable Genius

11 Feb

Notch November 7, 2017 as a watershed tickpoint on your presidential timeline. It was the day on which Twitter doubled its message character allotment to 280, an allowance that induced shudders in both the twittersphere along with vast terrestrial swaths: it mightily enriched the opportunities for the President of the United States to tweet his mind to you and me.

I’ll leave the rest of the editorials to you and me, but the character enhancement begs a few novel questions we could instructively put to the downloadable spreadsheet of Mr. Trump’s communiques offered up (for a small fee) by twdocs.com, the go-to site to which I’ve gone on more than one occasion. My latest requisition of @realDonaldTrump tweets (accept no substitutes) pumped out the incumbent’s latest 2805 entries through February 8, affording us exactly three months’ worth of post-280 streams of consciousness, with the earliest of these bearing the time stamp October 24, 2016. Again, I think you’ll have to pay for the sheet yourself if you want to click along.

A first question, then, an obvious one: has the widened message space encouraged Trump to issue longer tweets? And a second, slightly subtler follow-on: might 280 characters inspire longer, but fewer of them? (Note that this submission considers features of the commander-in-chief’s twitter proclivities over and above his preferred key words, a concern which predominates in my previous posts on the matter).
An answer to the former hinges itself on the LEN function, with which I’ve overwritten the contents of the H column, declaring nothing but the name Donald J. Trump in each of its populated cells. Renaming the field LEN I entered in H7:

=LE

N(h7)

and copied down. (I’m not quite sure what the field name a in the otherwise empty G column is doing. On the one hand, the name suffices to enable the pivot table to enfold the fields following G, but I’m not sure why the field need be there at all.)
The next move is slightly trickier, because the defining November 7, 2017 breakpoint doesn’t lend itself to a graceful pivot table grouping, or at least I don’t think it does. That is, grouping the date data in the A column by Months and Years will naturally turn out conventionally-demarcated months, e.g. Nov and Dec, and won’t group the months beginning with the 7th. A more elegant realization would tap into a pair of array formulas, i.e.: name the LEN field range Length and name the data in Created At (in the A column) Date. Enter the 11/7/2017 boundary date anywhere, say H1, and enter, say in G1:

{=AVERAGE(IF(Date<H1,Length))}

The formula seeks out all dates prior to the benchmark November 7 and averages their tweet lengths. Extended to two decimal points, I get 126.80. Then enter in G2:

{=AVERAGE(IF(Date>=H1,Length))}

This companion expression averages the character length of the President’s tweets dating from November 7, 2017 and evaluates to 193.29, a none-too electrifying differential, I’d allow; ask Twdocs.com to download your tweets, after all, and your stats might look much the same. Survey a few million post-November 7 tweets and then, and only then, might we be able to say something distinctive about Trump’s relative outputs.
But what about the second question posed above – namely, if the heightened loquacity promoted by the 280 characters thus curtails Trump’s need for as many tweets? That finding could again be essayed in a number of ways, including some messier array formulas that would count the available tweets on either side of November 7, 2017 and divide those respective tallies by the number of days elapsed prior to and following that date. In fact, however, a slightly more imprecise but swifter pivot table could do a representative job:

Rows: Created At (UTC + 1h) (The parenthetical notation points to the French provenance of twdocs, at its recording of tweet transmission times per that time in France, an important qualification. The User Created At field, however, merely reports and repeats the date/time at which Trump initiated his Twitter account).

Values: Created At (UTC + 1h) (Count)

I get:

BlogT1

True, the Nov 2017 above conflates the pre and post-November 7 numbers a bit, but the larger thrust of the counts pushes ahead. We see an inconclusive range of distributions: while tweets across the last few months have indeed downturned in relation to October of last year (and that month stands as the most profuse among the table’s holdings), tweet counts in the Feb-Jun 2017 span were decisively lower than the latter-day entries. It’s thus not clear if Trump’s tweeting inclinations have been stunted by the new opportunity to say more at any one time.

And what about Trump’s average retweet count, proposed here as a proxy popularity index? Just substitute Retweet Count for Created At (UTC + 1) in Values and tick Average (with suitable formatting). Here I’d also turn off any subtotals, thus uprooting one higher level of slightly confounding aggregation from the counts. I get:

BlogT2

Commit the above to a rudimentary line chart, and I get:

BlogT3

Is that something worth writing home about, or at least submitting on deadline? Maybe. There’s something at least statistically significant about the line’s fitful locus, and it might be worth investigating April’s retweet dip and/or the more recent spike in RT’s, which does roughly correlate with the opening of the 280-character gateway, though I’m hard-pressed to advance beyond a simple correlation – unless one wants to maintain that longer tweets bespeak deeper policy insights that warrant stepped-up circulation across the twittersphere. Sounds like a reach.

Then add Favorite Count to Values (again, Average):

BlogT4

Favorites correlate most significantly with retweets (I get .8997), an unsurprising resonance. One supposes the prior question, rather a generic one, would ask why favorites reliably outnumber rewteets so enormously (look here, for example, for one take on the matter).

And to fit a final, relatively novel frame atop the Trump tweets, what about the president’s means of transmission? That is:

Rows: Created At (UTC + 1) (again, grouped by Years and Months)

Columns: Source

Values: Source (count)

I get:

BlogT5

Note among other things Trump’s late abjuring of Twitter for Android, and a corresponding uptick in his resort to Twitter for iPhone., his clear source of choice.
Does that mean that Mr. Trump will be recruited to spearhead Apple’s next Think Different campaign?

Advertisements

Leave a Reply

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

WordPress.com Logo

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

Google+ photo

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

Twitter picture

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

Facebook photo

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

w

Connecting to %s

%d bloggers like this: