Monday, March 10, 2014

Counting under uncertainty

Lots of Pittsburgh bicycling news of late.  Note the trifecta just today. PG: Pittsburgh will get help developing bike lanes. Trib:  Pittsburgh gets nonprofit help to build more bike lanes. PBT: Pittsburgh to add 5 miles of protected bike lanes. Though the most consequential in the long run was a bit of administrative ephemera that BikePittsburgh exec Scott Bricker has been appointed to the board of the Southwestern Pennsylvania Commission (SPC). Maybe fodder for another day?

But bicycle riding is a great example to use in the proper use of data from the American Community Survey.  Ideally, the ACS should only be used to measure characteristics of a given population, and not be relied on to be the reference for the count of any particular group of the population.  That being said, many do use the ACS to report counts of many things,and even I fall into that trap on occasion, but I really ought not to.

What am I talking about?  The ACS is a sample survey program that for small populations, either in small geographies or small subgroups of a larger population, you really have to be aware of sample error. The program is not designed to be the reference count of the population and wonks out there realize that population counts from the Census Bureau's population estimates program are not the same as what you will get out of the ACS in most cases. Some like to pick and choose which data works best for them and you really need to be careful if you want to be objective.

So I ramble...  counting bicycle usage in Pittsburgh...  

So a routine metric in the city of Pittsburgh is bicycle usage.  A common question in the ACS and other programs is to ask your means of transportation to work, i.e. commuting.  Commuting is not the only reason to travel by bicycle. If you are a student getting to class, or shopping, or just riding to ride, you are not traveling to work and thus not captured in any data about commuting technically. But on the focused question of the trend in bicycle commuting in the city of Pittsburgh, you get a much better picture of the trends if you look collectively at the results from the American Community Survey over multiple years.  Relying on any one year of data, and just using the point estimate itself,  really will throw you off. Some years new data will imply a big jump, while some years will show decline on occasion.

So here is my graphic of the ACS estimates of the number of bicycle commuters residing in the city of Pittsburgh.  Note the actual estimate is not shown at all and the bars represent the range from the estimate + or - the reported margin of error. You get a far different feel for the trend when you look at the data like this, but for sure an upward trend.If I had just plotted the estimates by themselves (granted something the world does often, to include me) you give rise to an illusion of accuracy that is not there.  For those who are not ACS data users, basically the program reports data for different time periods. Some data is reported as an annual (1-year) average. Some data is reported over a 3-year average and some data is reported over a 5-year average. Here are all relevant data on the number of bicycle 'commuters' in all of the ACS datasets.

The uber parsing for those still reading is important as well. I probably should take 2005 out of the data because back then the ACS did not include group quarters in the data at all. So any students in dorms who were working and commuting by bike were not counted, so the trend is confused.  There have been other ongoing issues with measurement of the group quarters population population in the ACS program so it is logical to think some of the trend upward is attributed to better capturing of student/GQ populations. But a big topic that I am sure does not mitigate the overall trend.


Blogger Vannevar said...

I'm sorry to ask, but I've got to ask - do we (meaning, you) have population numbers for the same years, just to see if the change is a population shift or a true change? (Also, I wanted to point out that I did read the whole thing. Thanks!)

Monday, March 10, 2014 10:08:00 PM  
Anonymous MH said...

I think the obvious place for a nice, wide bike lane would be something connecting Frick Park with Schenley Park. By connecting those parks, you'd also be connecting the Jail Trail to downtown with the very long lane on Beechwood.

Tuesday, March 11, 2014 9:22:00 AM  
Anonymous Anonymous said...

MH - That vision is consistent with what Beechwood is supposed to historically be and since it's a pretty broad street (from the blue slide park all the way across the bridge and to Bartlett street) there could well be room to put a lane or two in there. The only wrinkle could be the interchange with the Parkway, especially the east bound exit from which people really come flying off the highway. Assuming that can be worked out, this would be nice.

Tuesday, March 11, 2014 1:19:00 PM  
Anonymous Anonymous said...

Another question is how the ACS determines someone as a "bicycle commuter". In the summer, I bike 3-4 days per week, but I don't ride during the winter months. Would I count as a bicycle commuter or an automobile commuter? Based on my observations of bike racks and trails, there are a lot more than me who do this.

Tuesday, March 11, 2014 2:16:00 PM  
Blogger C. Briem said...

Census data like that is mostly self-reported. So if you call yourself a bicycle commuter, you are a bicycle commuter. I suspect that for biking personal bias increases data on biking prevalence. But your example on seasonality would get captured. If one answers accurately... different answers will come in different parts of the year. Since ACS is an annual average, it should work out. Obviously biking is higher in summer than winter so the ACS number should be taken as an annual average.

Tuesday, March 11, 2014 2:24:00 PM  
Blogger C. Briem said...

I'll get back to Ed's comment.

Tuesday, March 11, 2014 2:24:00 PM  
Anonymous BrianTH said...

Here is a quick primer on what the ACS actually asks about commuting:

Looks like they ask about "last week," so apparently the timing of the survey could in fact matter. I might note others have observed that the "longest distance" rule sort of biases mixed-mode commuters in favor of being identified with the motorized portion of their trip.

Wednesday, March 12, 2014 1:47:00 PM  
Blogger C. Briem said...

and since the questions are asked year-round... it should produce an average.

No enforcement, and I doubt much explanation, of any longest-distance rule. ACS has had folks in the east end commuting by literal rail. Whether that mixed mode as in folks getting downtown and taking light rail, which isn't really what is meant by rail, who knows.

Wednesday, March 12, 2014 1:56:00 PM  

Post a Comment

<< Home