Need a header here
Ever wondered how Martin and Bosco’s post traveled through the Tumblr community?
The Daily Notes Plot
Focus Question: What trends emerge in people’s daily engagement?
- Hover over the data to reveal tooltips with dates and note counts.
- Change the view by clicking legend items on or off.
Viewing Tip
The plot is best viewed on a laptop or desktop. Annotations provide valuable context on a time-series graph, but they’re also the first thing to get cramped on small screens! If you are viewing the plot on your phone, landscape orientation is suggested for improved plot display and easier use of interactive features.
- A static screenshot of the plot is available [here].
Making Sense of the Data
1. What are we looking at?
This plot shows note count (Y-axis) over time (X-axis) for Martin and Bosco’s post. Time is grouped by month and year along the X-axis; the range is from July 13, 2022 through October 31, 2025.
The blue line shows how many notes the post received on each day. A note is any like, reblog, or reply.
The purple dashed line shows a 7-day rolling average. The purple line smooths out daily noise by averaging note counts over a centered 7-day window. This reduces the impact of random daily ups and downs, making it easier to spot broader patterns in activity.
The red dots mark spike days, or days with more than 241 notes. The threshold is calculated using the median daily note count plus four times the median absolute deviation (MAD), a method for identifying unusually high activity. You can read more about this calculation in a previous blog post.
Which Time Zone Am I Using?
Tumblr’s API stores timestamps in Unix time, which counts the number of seconds since January 1, 1970 (UTC). In all of my plots, times are displayed in Coordinated Universal Time (UTC) for consistency. For example, 4:00 PM UTC is Noon in Eastern Standard Time (Toronto time).
A reasonable question to ask is why don’t I convert the UTC timestamps into the local time of the people reblogging Martin and Bosco’s photo? Golder and Macy (2011) did this with Twitter, using self-reported time zone fields so they could align [rhythms] [this is weird] to sleep and work schedules. Linnell et al. (2021) also used Twitter’s time-zone field to analyze Daylight Savings effects, though they note that the field was later discontinued for privacy reasons. Tumblr’s API does not expose user time zones, and guessing or sleuthing them from people’s blogs would be unreliable or an infringement on their privacy. Tumblr is one of the few social media sites that does not require a user’s real name, and anonymity is a central tenet to the community.
How does UTC affect the way I analyze events?
Using UTC is not just an esoteric technical detail of interest to data scientists; it’s a decision that shapes the way I represent specific events in my analyses.
Martin and Bosco Day
Tumblr users hilariously celebrate July 13 as Martin and Bosco Day, the date of my original post. Activity for the post is also high on July 14; some July 13 activity in local time registers as the next day once converted to UTC. So, I label the event “M & B Day July 13–14” on the graph to reflect that.
WHAT THE RESEARCH SAYS:
Why label Martin & Bosco Day as July 13–14?
Leypunskiy et al. (2018)
Description line.
Placeholder text goes here. Multi-line supported.
Shah et al. (2019)
Description line.
Placeholder text goes here. Multi-line supported.
Laura’s Insight
Placeholder text goes here. Multi-line supported.
Looking at the literature gives me a lens to think more deeply about what the July 14 traces might represent. Leypunskiy et al. (2018) found that U.S. holidays tend to push posting activity later into the night. Shah et al. (2019) showed that late-night posting looks different from daytime posting, with shifts in tone and sentiment. While Martin and Bosco Day is obviously not a national holiday, these studies suggest that July 14 activity might be more than UTC spillover. If I were drop the 14th, I would not just lower the note count for the event, I might cut out the very slice of the event where tag use or emotional tone diverges. Later, when I examine the tag data, I can explore whether late-night reblogs carry a different emotional flavor than the daytime posts.
Blaze Periods
Tumblr officially defines blaze as a 24-hour promotion. For example, one of my email messages read:
“So your post will now be shown to an estimated 102,500 Tumblr users as a sponsored post from Thursday, January 23, 2025 at 2:20 PM to Friday, January 24, 2025 at 2:20 PM.”
However, after 17 blazes, I am deeply skeptical of the precision of these notices, since the engagement activity on Martin and Bosco’s post often begins before the email time or continues past it. Even with that inconsistency, I adhere to Tumblr’s own definition in my plots and label blaze periods as two days. For example, “Top Blaze 2025 ★ Jan 23–24”. For note totals, I also report the following UTC day, since I cannot tell whether that activity is spillover caused by the UTC boundary or ordinary engagement. This follows a similar logic to Linnell et al. (2021), who widened their observation window across dates to account for temporal effects introduced by time-zone shifts.
Stay Tuned
jjjj