Study Completed

The data shown here is presented in its final form. I have moved to a new apartment and no longer commute to the lab.

Introduction

During my first few years in graduate school I had the good fortune to live very close to campus. People able to avoid driving to work every day are usually pleased enough with their situation to write about it on their blogs. In my case I did not realize just how great it is to walk to the lab until I became a true driving commuter. Having moved much further away until I finish school (eyes on the prize), the minutes consumed by traveling in to and home from the lab are adding up.

First, I began to wonder how much time I was spending on the commute. Next, I started keeping track of this time. Finally, a picture is starting to emerge and I can analyze the results.

To place the results below in context, here is some information about my commute:

• Total Distance ≈ 17 miles
• Freeway Taken: US 405 North
• Average Travel Time = 34 ± 7 minutes
• Total Time Driving to the Lab Over 58 Recorded Days = 33.05 hours

Total Drive Time from Home to Lab

The following plot is the entire data set of driving time. This is measured in minutes from the moment I start moving the car away from my apartment until the moment I park at the lab. I am not keeping track of trips home because I leave the lab at different times. Mornings are much more standard and offer a more quantitative data set, though even this is not a rigorous study. I do drive during the “window of opportunity” every day so these results should be considered the absolute shortest commute times possible.

The X's are the actual commute times. Data gaps occur because of my time away from the lab and those days when I make a mistake in recording the travel time or take a different route.

Presently, this plot shows a consistent travel time of under 50 minutes. The plot is not very interesting with only 58 data points. Comparisons by month do not show any significant difference. In a way, it is unfortunate that my commuting days have ended because it would have been worthwhile to see whether the commute times changed during the summer vacation season or as gasoline prices continued to rise.

Average Travel Time by Day of the Week

The next plot displays the travel time averaged according to the day of the week. I typically come in to the lab every day, which is both a common aspect of graduate school and also a good way to collect more meaningful statistics by including the weekends. The error bars on the plot represent ±σ, where σ is the standard deviation of the data set for the given day.

The “error bars” in this case do not represent actual error in my measurement of the driving time, rather, they are an indication of how unpredictable the drive will be on a certain day of the week. For example, say the average drive time was 50 minutes for Wednesday. That could be achieved through any number of combinations. If only two Wednesdays were recorded, then pairs of A = (50 minutes, 50 minutes) or B = (75 minutes, 25 minutes) would give the same result. The standard deviation of A is zero, while that of B is 35.4, so if the actual data set was similar to B then the error bar on Wednesday would extend 35.4 minutes above and below the data point. It would be reasonable to conclude that Wednesdays, if they had such large error bars, are less predictable than days with smaller error bars. The commute on days with very small error bars is essentially the same every time.

I expected the weekend days to provide a drastically reduced commute time but it appears this is not the case. The unfortunate truth is that Los Angeles traffic is always busy.

Wednesday, Friday, and Saturday have the largest error bars and therefore represent the days with the most unpredictable traffic. If the number of measurements for each day were to increase, then it should be expected that these bars decrease because the outlying points (as seen in the first plot) are averaged out as the complete data set grows.

A Closer Look at Freeway Effects

Initially it would seem that calculating the average freeway speed during my commute is equivalent to the total time. In fact, this is not the case because it takes an average of ten minutes just to reach the freeway from home. Using a timer with lap control I can individually record the times to reach the freeway from home, to travel along the freeway, and to park after exiting the freeway. The freeway speed, v405, is calculated as v405 = 13.9 / (tdrive - tpre - tpost), where tdrive is the total drive time and the pre and post subscripts refer to the legs of the trip occurring before and after the freeway, respectively. The value of 13.9 represents the total freeway distance traveled (with all this information you are just an online map away from having a good idea of where I lived).

Armed with the knowledge of how the average freeway speed is calculated, here is an example of what difference it might make. Let the commute of next Monday take 20 minutes. Tuesday's is completed in 30 minutes. These numbers are close and the difference is only one-third. Subtracting out the ten minutes used to reach the freeway each day, the travel times are now 10 and 20 minutes, meaning that Tuesday's commute was twice as long as Monday's. The idea behind the average freeway speed plot is that it will provide a better highlight of the difference in commute according to the day of the week. For this plot it is better to read large values because that means a faster average speed and shorter commute time.

The above plot displays very large error bars. These are due to the sensitive dependence of this freeway speed measure on accidents. Any day there is an accident on the freeway the driving speed will be severely reduced. During the clean up of the accident an entire lane or more might be closed, thereby restricting and slowing traffic. Immediately after the accident traffic is likely to remain slowed because of rubber necking (see 1 & 2).

The freeway speed plot reproduces the results of the time by day plot. Sunday is still the best day and the others fall into the same order as before. This implies that the non-freeway driving is fairly consistent. If the non-freeway drive was a lot worse on a given day, for example, then the total time plot would show a large value (slow commute) but the freeway speed of that day might also be large (fast commute).

Conclusions

The conclusion reached thus far is that commuting in Los Angeles is no fun. Some days just the thought of having to drive home is enough to encourage sleeping at the lab.

Tags: Los Angeles, commuting, data analysis