38 comments

[ 3.0 ms ] story [ 60.5 ms ] thread
The only thing I think is missing here boarding times for the different carriers. I was reading somewhere (I can't find the link) that southwest has the best method right now. I did find a link to another site that seems to have similar content but doesn't mention Southwests boarding style (because it isn't random it is based on least resistance boarding). http://menkes76.com/projects/boarding/boarding.htm

Edit: Note Dewey found the right article below.

Very cool, maybe I will implement the simulator when I have a chance.
I'd argue that boarding times are generally irrelevant, despite being annoying. If you walk up to the gate a minute before they close it, you'll walk right on. So waiting in that long line for slow boarding is essentially a choice.

Except for Southwest, since they don't have assigned seats.

You might not have room for a bag in the overhead compartments though, which can be a dealbreaker for some people.
(comment deleted)
Looks really cool. One small suggestion, sprinkle over some dc.js [1] and add some interactivity :-)

[1] http://nickqizhu.github.io/dc.js/

This is actually perfect for my next visualization project, where I'm going to start digging into SEC data.

I invest in equities and I find I never have the correct information in the right format.

Indeed - I wish I had know about dc (or dc had existed) when I was working in hedge fund land. Would have made my life a lot easier :-)
Unfortunately, it's fairly difficult to make any concrete conclusions from flight time, especially at a time resolution of one minute. The jet stream and other weather processes affect flight time by more than a minute. This also holds for mechanical reasons as well, since airplanes that have squawked 7700 will almost certainly be given landing priority, leaving other airplanes in the air for longer while they wait their turn.
I agree. I did the best I can by using the median over all flights from the past 12 months.

But yes, there are a lot of irregular operations which can throw any conclusions to the wind. That said, when I was exploring this data I found that delays etc were much less frequent that I perceived.

Interesting.

Riddle me this: http://flightsphere.com/flight-time/from/atlanta/to/chicago

It says United has 6 flights per week from Atlanta to Chicago. They have 6 or 7 per day in reality.

How are airports in metro areas consolidated? Are you using MDW instead of ORD for "Chicago"?

Thanks for finding this, I owe you a beer. Payable any time in the bay area.

It is probably a bug in my data pipeline. I will look into this.

I consolidate all airports in a metro area, MDW and ORD for Chicago.

I'm pretty sure UA has more than 48 flights a week to the NY area as well if you're aggregating JFK, LGA and EWR,the number should be closer to 20 per day. I just arbitrarily picked Friday, 10/17 from SFO => NYC and there are 23 flights. Your number might be right if excluding EWR, which like SFO is a major UA hub.
Appears to be missing data from Alaska Airlines. A lot of flights in the NW are handled by Alaska, so there's a number of missing connections from my local airport.

[EDIT] Not quite correct, just missing all the Alaska airlines flights out of Bozeman. Perhaps missing the turboprop data?

Hmmm. So sometimes these smaller regional flights are subcontracted out to small carriers who don't have $20m of revenue. In that case, I don't have the data since it costs $$$.

I will look into it to figure out the reason. And those beers are still yours.

Maybe it's just missing the carries (like Horizon) that are operated by Alaska?
Very interesting visualization. One thing I want to point out however is the rearrangement of the widgets should not be based on its size, but on relatedness instead. An for that to happen, you need to design the layout instead of using auto re-arrangement based on screen size.
Good point, the tradeoffs between making something visually interesting vs relevant from a visualization point of view were really tough on this project. I will look into some kind of compromise. Related data does not always look good next to each other.

More broadly, when I visualize a set of data I follow Alberto Cairo's guidelines of Organize, Compare, and Correlate. In this project, I moved away from that quite a bit because I was seeking to make something interesting as opposed to useful.

I was scratching an existential need I had to make this data accessible and interesting. :)

For me, I see the following sets of widgets highly related and should be next to each other:

- Flight Time By Airline, .. by Time of Day, Average Timeline

- Flights By Month, .. by Day of Week, .. by Time of Day

But they are not, when I adjust from screen wide from all sizes from 860 to 2740.

Thanks for the feedback. I will play with your suggested arrangements a bit in my next iteration.

I've added it to my basecamp.

I think you need to exclude months or weeks with zero flights, otherwise you get bad data for seasonal routes.

For instance, look at this one: http://flightsphere.com/flight-time/from/hartford/to/las-veg...

It's clearly a daily flight, based on the per-day chart, but it says it is 5 days a week, because they don't run it year-round.

I was looking into temporal filtering algorithms, but clearly your fix is much simpler and better. Sometimes, simplicity trumps sophistication.

Two beers payable in the bay area!!

Looks like you are pretty active around the web. Adding a github/twitter link might help get more feedback and allow other devs to find you.
Nice work,how do you plan to monetize? Affiliate links to online travel agencies?
not really sure yet. This is just a demo of a real time analytics system I'm building.
> We only have domestic, non-stop flights for now

Should mention that it's US only.

London / Edinburgh or Mumbai / New Delhi are still "Domestic" flights.

/PEDANT