So Yitz and I were hanging out, trying to decide whether we should go to my apartment and code, or go to the pub. After a while, we somehow ended up wondering what day of week pubs were the busiest, and being the geeks that we are, we decided to look for an algorithmic solution. And where would you go to collect data on pub patronage by day of week? Why, Flickr, of course!
The premise of our theory was very simple. We guessed that people are bound to take pictures when they go to the pub, then upload and tag those photos on Flickr. Sure enough, there are over 14,000 images tagged "pub" on Flickr. We decided that would give us enough data points, to at least get some info.
Since we couldn't find API calls to extract the data we wanted, and we don't have API keys anyway, we decided to do it the quick and dirty way. Yitz hacked together a Perl script to extract all the photo IDs tagged "pub", then extract the date taken from the details page, and compile all the data into a single file (well, technically, we had up to 10 of these processes running in parallel). I then scratched together a PHP script to collate that data, and this is what we got:
Exhibit A: # of pictures tagged "pub," by day-of-week
Encouraged by these results, we decided to do the same with photos tagged "bar":
Exhibit B: # of pictures tagged "bar," by day of week
Since no experiment can be considered legitimate without pretty graphs, we passed our data to GNUPlot:
Exhibit C: # of photos tagged "bar" and "pub" by hour
Here are some observations we made:
It was close to 2am when we got this far... but there was one more thing we wanted to know. How does all this correlate to when people are the most drunk? Well, check out what we found:
Exhibit D:Pub/bar-crawling vs drunkenness
There's a very clear phase shift... and as it turns out, drunkenness peaks at 7am, when they've all gone home from the pubs and bars!
Conclusions
Also see: Yitz's LJ post
Sun:1857 (13.38%)
Mon:1771 (12.76%)
Tue:1286 (9.26%)
Wed:1507 (10.85%)
Thu:1735 (12.50%)
Fri:2944 (21.21%)
Sat:2778 (20.02%)
The list shows the total number of pictures taken, by day of week, with percentages. As it turns out, Friday seems to be the busiest day at the pub. (We decided that for this experiment, we should define a day as starting and ending at 7am. It just intuitively seemed like a good time when a trip at the pub should end --more on this later).
Sun: 2917 (11.14%) [896]
Mon: 2803 (10.70%) [795]
Tue: 2949 (11.26%) [780]
Wed: 2690 (10.27%) [840]
Thu: 4196 (16.02%) [941]
Fri: 5101 (19.47%) [1185]
Sat: 5540 (21.15%) [1301]
(Square brackets show number of unique users who posted images.)
It's past 3:20am, and I really should go to bed. Besides, everybody knows conclusions are complete BS anyway. So let's just skip it, and let me go to bed...
