Interactive visualization of commute times for all US cities

For example, from downtown Kirkland, WA. You have the bus being faster than driving, there is no HOV lane that would explain this, my real commute choice is driving (10-15 minutes) or bus (35-40 minutes) but your heat map pretty much inverts those numbers.

DC actually seemed like it went the opposite way as it usually takes an hour and a half to get from VA through the beltway into DC while the map showed only an hour for as far out as Dale City. A half hour (or more) from Petworth into Downtown during rush hour is fair, since DC has probably the slowest crappiest drivers in the nation.

Updated theory: these commute times might not be based on actual driver data, it's probably a very rough estimation based on distance, travel time, speed limit, and population. It needs to be weighted based on the curve of traffic based on time and the way locals drive. Boca commuting will probably be slower than Miami commuting on average, unless you're on the highway in which case you're totally screwed in Miami. Not to mention anywhere there's an on-ramp with lots of flow you're going to make a choke point, so times would increase a lot, unless you started after that exit.

tl;dr Traffic is hard.

It's more of a backend question, but what algorithm are you guys using to traverse the maps?

Also, why did you decide to use Google for the display maps, when you used OSM for the backend calculations?

Great work. I checked commuting time for Seattle, WA using public transit and most of your visualization are spot on. Is the public GTFS files you are referring to come from here: http://code.google.com/p/googletransitdatafeed/wiki/PublicFe...?

With regard to some of the earlier replies saying that the map is inaccurate, it might be worthwhile to use Amazon Turk to run some evaluation on how reliable the spatial data (or your algorithm for calculating travel time, rural vs local, .etc) is.

-Sen

Also, every route seems to be treated the same. In the morning, the road I live on is always empty in one direction and congested in the other. I used to commute against traffic and now go the same way as everybody else, and I could go twice as far in ten minutes as I can now. The numbers will never be accurate until you ground them in real traffic data.

However the main problem I have during my real estate search has nothing to do with UI. It seems like for every home I find through the site that I contact a realitor about the home is already "under agreement". If your site could find a way to be more up to date about those things, it would really stand apart.

so that it is normalized by population density? Otherwise it's completely misleading and uninformative.

The maps you've added are awesome when you know where you're looking at, but if you don't yet have a place picked out then knowing where to look is good too.

A naive way to compute the kind of summary map I'm interested in is to just make the color at each point on your summary map be the average color on your map for a specific place. Something smarter could weight points by their desirability (I care more about how fast it is to downtown than to an arbitrary point in a suburb) but the naive version is pretty useful on it's own.

As many commentors have pointed out, our time-accuracy is not exactly 100% yet. :)

This is our first iteration on this concept, so we are using some heuristics to try to get times roughly close, and accurate relative to each other. We're also validating the user experience, value as a product, etc.

I'm still investigating a good set for nationwide traffic data, which will help give us a much more accurate picture of real-world commute times. Traffic varies so very much city to city :)

Just kidding, it's alright, it's kind of off for the SF/PA commute though, by nearly a factor of two for many places in SF, unless you're doing absolute worst case scenario every day of the week.

or if I was on travel, and had a meeting at a client site at point a. but a hotel near my companies office at point b, I could see the travel times, and maybe even go for a hotel closer to point a (or hotels near point a are too expensive, how far would cheaper hotels be for my commute) .

They probably should have focused on getting better data before launching. Garbage in, garbage out I guess. I hope they can improve the integrity of the data because it has a lot of potential.

i.e., if you think about commuting to work in the morning, the heatmap will show you the times it takes to drive from any given location to the marker. If you're thinking about driving home in the evening, you can think of the time from the marker to any given location.

on the macro scale, most streets aren't one-way, and commute times tend to be symmetric.

It really looks algorithmic based on street size, assuming some specific congestion level. I'm not seeing any indication of some of the known congestion points that I used to have to deal with.