At StumbleUpon we are constantly trying to better understand how our users interact with our product so we can improve our recommendations and the overall stumbling experience. With that in mind, we recently analyzed how different types of content are consumed by our users on a typical day.* Here’s a quick summary of our findings:
- Stumbling increases very quickly in the morning commute hours, and mobile plays a big part in that – during this time, mobile stumbling is 70% higher than the daily mobile stumbling average.
- There is an inverse relationship between what gets stumbled during work hours and what gets stumbled in the evening. Some of the most popular topics during work hours are Cars, Technology, and Health. Some of the most popular topics stumbled at night are Relationships, Philosophy, and Babes.
- There are clear gender differences with many topics on StumbleUpon: e.g., men strongly prefer topics like Babes and Cars, and women strongly prefer topics like Beauty and Fashion.
- Users stumble at work only 20% less than off the clock. But users still stumble once every 3 minutes on average while at work.
- Men stumble later at night than women.
When Do You Stumble?
First, to give you an idea of how big our data sample is and what overall stumbling patterns look like, the StumbleUpon index – i.e., the database from which we draw our recommendations – contains over 60 million distinct URLs (and is growing at over 5% per month). These URLs are categorized across over 500 interests to which Stumblers can subscribe (and the average Stumbler subscribes to around 25-35 topics). And just like TV viewership, the number of active Stumblers varies dramatically through the day:
This chart shows the stumbling pattern over a day, with time slots indicated in different colors for morning, commute, and evening hours.
This chart shows that the number of active Stumblers hits a daily low around the very early morning (when most people are sleeping) and peaks in the late evening (9-10pm). Keeping the shape of this curve in mind, we segmented the day into three major slots: “evening”— the hours between 8pm to 5am, “work”— the hours between 10am and 4pm, and “commute”— the remaining hours of 6am-10am and 4pm to 8pm.
Note how the number of users stumbling during the morning commute hours grows quickly. Considering that during this time we wouldn’t expect people to have ready access to computers, we suspected that mobile usage could help explain some of this early morning activity. While mobile stumbles still only account for 7-8% of the total stumbles we receive on any given day, these numbers are growing very quickly month-over-month. The chart below shows the volume of stumbles coming from mobile devices by time of day:
This chart shows the mobile stumbling activity spread out over a typical day.
This chart tells us that during the morning commute, mobile stumbling is about 70% higher than its daily average.
When Do You Stumble What Topics?
Using these “time-segments,” we then took a look not only at when users are stumbling, but also what they are stumbling. We compared the differences in the percentage of all stumbles served (on web or mobile) from each topic throughout the day. The chart below shows this relationship for the top 50 most popular topics.
What immediately jumps out is that there is an inverse relationship between what gets stumbled during work hours and what gets stumbled in the evening. Some of the topics our community stumbles at night but not at work are what one would expect – e.g. Babes and Adult Topics that are NSFW – but interestingly, we also found that Relationships and Philosophy were stumbled more so at night than during work hours. And the topics that are stumbled during commute hours are a mix of what gets stumbled during work and evening hours. We think this may be because each commute varies in how much private and public space it provides.
Each diagonal line shown here represents a different time slot in a typical day (commute, work, evening). The higher the line is above the x-axis, the more stumbling occurs in a given topic.
In order to get a better sense of what people are stumbling when they should be working (we won’t tell your boss…) and what topics they prefer to stumble at home, we put together the chart below. This chart shows similar data to the chart above, but it limits the data to just work versus evening hours and orders the topics by how much they change between these two times of day. Each concentric circle represents a 5% difference in volume contribution for a particular topic during each of the two time periods.
The green area on the chart above represents the likelihood of a given topic being stumbled during work hours.
Here’s how to read this chart: The green area in the circle represents the likelihood that a given topic will be stumbled during work hours. Start at the top of the circle (i.e., 12 o’clock) with Cyberculture, the topic most likely to be stumbled at work. As you move clockwise, you’ll see less and less green overlapping with the concentric circles. For example, follow the line from Movies to where the green area touches a concentric circle. Then follow that concentric circle all the way to the left side of the circle, where you’ll see numbers and percentages (+15%, +10%, and so on). You’ll notice that the Movies topic has a +5% likelihood of being stumbled at work instead of at night. Relationships, on the other hand, has a -10% chance of being stumbled at work – the least likely topic to be stumbled at work out of any of these top 50 topics we analyzed.
As you can see from the chart, while at work, people prefer to stumble topics like Cars, Health and Technology. There’s also a large portion of our content that gets stumbled at the same rates regardless of the time of day, like Humor, Photography and Music, which are our most popular topics by subscription volume.
What and When Do Men and Women Like to Stumble?
Given the diverse range of pages in our index, it makes sense that men and women would stumble different topics. While some of our most popular topics – like Photography and Humor – don’t skew towards one gender over another, there are a large number of topics that are decidedly gender-biased.
The chart below illustrates the percentage difference between male and female stumbling behavior in our top 50 topics. The Babes topic (to the far left in the graph) is 1300% more popular with men than with women. That is, the percentage of stumbles in the Babes category in an average guy’s stumble stream is about 13 times larger than in the average woman’s stream. On the flip side, stumbles in the Beauty topic, which sits all the way to the right in this graph, is 95% less popular for men than for women; in other words, 20 times fewer Beauty stumbles appear in the average man’s stumble stream than in the average women’s stream.**
This chart shows the difference in topic preferences between male and female Stumblers.
So, if such dramatic differences exist in the topics that women and men prefer, could it be possible that the shifts in topic preference we see throughout the day can be due to differences in when women and men stumble?
At first glance, one wouldn’t necessarily think so. First, if we think back to our comparison of topics stumbled at work versus during the day, we saw that the two relatively more popular “male” topics — Babes and Cars— are stumbled at completely opposite ends of the day. That is, Cars is relatively popular during the day, and Babes is more popular at night. Similarly we see that the “female-preferred” topics Quotes and Fashion are also popular at opposite times of the day. This would imply that users exhibit different content preferences at different times of day.
However, if we look more closely, we also see that a couple of the NSFW topics that are prominent during nighttime stumbling also seem to be “male-preferred” topics. Since we know that our user base is not distributed evenly between both sexes (we have a slight bias towards males, though stumbling by females is growing at a faster rate than stumbling by males), we decided to see if males were stumbling at higher rates during the nighttime than females.
This chart analyzes how much males account for stumbling activity at certain times of the day.
The chart above confirms our hypothesis: Males stop stumbling (and presumably go to bed) later than females, leading them to represent a larger proportion of the stumbles during the evening hours. For example, there are 15% more male stumbles at 3 a.m. versus 9 p.m. This means that we would expect that at least some of the male-biased topics would be relatively more popular during the evening hours than during work hours.
Do You Stumble at Work (or in Class)?
After analyzing all the heavy StumbleUpon activity during the day, we asked ourselves: shouldn’t people be working instead of stumbling during work hours?
Feeling a bit self-conscious about the impact we may be having on employee productivity, we took a look at the average number of sites active users stumble per hour throughout the day. For employers’ sakes, we were hoping to see very low numbers during work hours and higher volumes during the off-work hours.
This chart shows the stumbles per active user rate over a typical day.
The good news is that the stumbles per hour rate is indeed lower during working hours. The bad news is that it’s not by much – the stumbling rate during work hours is only about 20% lower than at the peak stumbling time. Also, the average stumbles per hour for active users is about 20 stumbles. That means that Stumblers are seeing a page, on average, about every 3 minutes while at work!
However, there is one thing we haven’t considered so far: StumbleUpon is particularly popular amongst the 18-24 year-old demographic. We realized that perhaps the “average” Stumbler is not someone that necessarily is sitting at work during “office hours.” So, to be fair, we decided to break down the activity numbers by age bands to see if we could pick up some “responsible” signals from the 25+ workforce demographic. The results are shown in the chart below.
This chart divides active stumblers on a given day into age groups.
In a way, the 25+ crowd is pretty responsible. First off, it appears they go to bed reasonably early: Activity starts winding down after 10pm, about 2 hours earlier than the 18-24 crowd (which also has a chunk of users remaining active through the wee hours of the morning). They also wake up earlier – again about 2 hours earlier than the 18-24 crowd. However, they do stumble throughout the day at roughly the same rate as 18-24 year-olds.
The behavior of Stumblers 17 years old and under is probably the most understandable. During school hours, this demographic’s stumbling rate decreases. After 2pm we start seeing heavy activity, which remains strong until about 10pm, when we assume mom and dad demand lights out.
At StumbleUpon, we’re always thinking of ways to improve how you find content you’ll love. This is not limited to understanding what kind of content you like, but also how you might appreciate content differently depending on your circumstances. Factors like the time of day and the device you’re using may significantly impact your stumbling experience, so we account for these factors in our product design and recommendation algorithms.
Thanks for reading! Please click “Comments” at the end of this post to offer your own ideas on any of the data presented here. I only report the facts!
*For this analysis we looked at over 1B stumbles performed by over 3m users over the 6 week period encompassing all of February and the first two weeks of March 2011. We limited the analysis to the United States and made sure to normalize the time zones].
** For you data geeks out there, here’s the formula I used for the topic/gender comparison chart: ([male contribution-female contribution] / female contribution)
Until next time,
Roberto Sanabria
Stats Ninja
