Facts, But Not Truth

On January 29, WTOP posted an article to their website about texting-while-driving convictions in Virginia. As reported in the story, for the last six months of 2013, since texting-while-driving (TWD) became a primary offense, police have issued 725 such convictions, with 40 percent of those convictions taking place in Northern Virginia.

The full article is here, but the information that piqued our interest, as data analysts, is quoted below.

WASHINGTON - The first six months of Virginia’s law that gets tough on texting while driving reveal a clear trend: Drivers in Northern Virginia are the most likely to be convicted…. With 168 convictions, Fairfax County had far and away the most convictions for texting and driving, almost a quarter of the total…. Prince William (62) and Arlington (41) counties follow Fairfax in having the most convictions.
— WTOP website, 1/29/14

We took issue with the following two points, the first of which is somewhat less explicit than the second.

  1. Fairfax County has the most TWD convictions in the state, by a large margin. Therefore, it is the worst county for texting while driving.
  2. The counties with the second- and third-highest tallies of TWD convictions are also in Northern Virginia. Therefore, drivers in Northern Virginia are the most likely to be convicted.

The article does not present any data that would contradict these points…but a responsible data analyst would instantly realize that there is not enough information presented to support these statements either.

In order to get a better idea of whether these statements were simply made hastily (i.e., there was no time for real data analysis) or with an agenda (data was misappropriated to insufficiently support an argument), we did an initial examination of the data to see if we could find anything that was misleading. 

There is Not Enough Information Provided to Answer this Question

Facts, Assertions, and Missing Data

The WTOP article accurately states that Fairfax County is the Virginia jurisdiction that has issued the most convictions for texting while driving, with nearly three times as many convictions as the second place county.

What’s more, the three counties with the most convictions are, as highlighted in the article, in Northern Virginia.

For the purposes of this discussion, our definition of Northern Virginia includes 11 counties (Arlington, Clarke, Culpeper, Fairfax, Fauquier, King George, Loudoun, Prince William, Spotsylvania, Stafford, and Warren) and six cities (Alexandria, Falls Church, Fairfax, Fredericksburg, Manassas, and Manassas Park).

The chart above shows all of Virginia’s counties and cities that have at least one TWD conviction…and from this chart, Northern Virginia doesn’t look very good. It had close to half of all TWD convictions statewide, and includes four of the five individual jurisdictions with the highest totals. What’s more, Fairfax County does indeed have far and away the most convictions overall.

However, here is the first instance of the article misleading with information. The original story explicitly limited its assessment to counties only, rather than ranking all jurisdictions (including cities). That way, the author could say that the top three counties all came from Northern Virginia—which is technically true, but misleading: 

#1. The jurisdiction with the second-most convictions was a city, far away from Northern Virginia: Virginia Beach.

Excluding cities from the report made it seem that convictions for TWD are largely confined to Northern Virginia jurisdictions, which it is not:

#2. Four of the ten jurisdictions with the highest TWD counts in Virginia are cities, and all four of those jurisdictions are outside of Northern Virginia: Virginia Beach (2nd), Chesapeake (8th), Roanoke (9th), and Hampton (10th).

The truth is that TWD convictions have taken place in substantial numbers outside of the NoVa area as well:

#3. 13 of the 17 jurisdictions with the most TWD convictions are outside of Northern Virginia.

None of these “discoveries,” each of which weakens the premise of the story, required anything more than simply looking at a single table of data—the very table provided in the article itself.

At this point it seemed clear that the limited data presented in the article was being used to paint a possibly inaccurate picture of the TWD issue in Virginia. We decided to (a) research and gather sufficient data to perform what could be considered a baseline-acceptable analysis, and (b) see if what we found would support or refute the claims that were made in the article.

Plugging the Holes in the Data

To support or refute claims like “data reveals a clear trend that drivers in Northern Virginia are the most likely to be convicted” of TWD, we need more data than the article’s author has (or, at least, provides). Regarding TWD in Virginia, we only know the answer to the question, “How many?” We are trying to answer a different question, which is, “How likely?”

The baseline data we chose to use was the amount of traffic on the roads in each Virginia jurisdiction. (Of course there are many other variables that could validly be considered as well. Average speed could be a variable, since TWD might occur more at low speeds, or in stopped traffic. Police patrols per mile traveled could be another variable. But we did not set out to do an exhaustive analysis here; we simply wanted to find something that would provide meaningful context for the existing data.)

While we could not get data from the same time period, we were able to find comprehensive, averaged data from 2012 on the daily vehicle miles traveled over all roadways in each Virginia jurisdiction, compiled by the state Department of Transportation. (This is available online in both PDF and Excel formats.) 

Finding the data took about ten minutes. Cleaning up the data took another 15-30. So in less than an hour, we had sufficient information to create a much more robust picture of the state’s texting-while-driving pattern. We will be able to see more than just the number of convictions per jurisdiction; we will see how much traffic there was in those jurisdictions, and that will tell us more about which places people are more likely to be caught by the TWD laws.

Using Analysis to Answer Two Lingering Questions

With contextual data in hand, we proceeded to investigate the defensibility of the two assertions that caught our attention in the first place.

Question 1: Is Fairfax County the Worst Jurisdiction For Texting While Driving in Virginia?

We know that Fairfax County had the most TWD convictions from July to December of 2013.  Without context, that seems like a reason to call it “the worst.” But without context—total vehicle miles—it’s premature to make that assessment.

With the new data from VDOT, we also know that Fairfax County roads bore by far the most traffic in the state in 2012—on average, 26.3 million vehicle-miles daily. (VDOT calls this measure “Daily Vehicle Miles Traveled,” or DVMT.)  By comparison, Prince William, the second-most traveled county, registered a mere 9.1 million DVMT. 

Rank Jurisdiction 2012 DVMT
1 Fairfax County 26,379,656
2 Prince William County 9,107,975
3 Chesterfield County 8,307,433
4 Henrico County 8,232,198
5 City of Virginia Beach 7,892,420
6 City of Chesapeake 5,913,362
7 Loudoun County 5,881,859

Bearing the total traffic data in mind, let's ask the question anew: are Fairfax County drivers really the worst?

Is Fairfax the Worst of the Three Counties Mentioned?

Let’s start by comparing Fairfax County solely to the other two jurisdictions highlighted in the original article: Prince William; and Arlington.

The TWD conviction data runs from July 1 – December 31, 2013. That’s 184 days, so we multiplied the 2012 DVMT data for each jurisdiction by 184 to get estimated values to use for comparison. 

Jurisdiction Estimated Vehicle Miles Traveled, Jul-Dec 2012 Convictions, Jul-Dec 2013 Convictions per 100 Million Miles
Fairfax County 4,853,856,704 168 3.90
Prince William County 1,675,867,400 62 4.18
Arlington County 1,528,567,672 41 5.77

Of these three counties, Fairfax is actually the least likely place for a TWD conviction. Using the convictions-per-mile metric, Arlington drivers are 46% more likely than Fairfax drivers to be convicted (5.77 convictions per 100 million miles driven in Arlington vs. 3.90 in Fairfax).

So if Fairfax is not even the worst among these three, it surely can’t be the worst in the whole state. Even though it has the most convictions, it also has by far the most vehicle miles traveled, which should be a key element of this consideration.

Adding contextual data has already refuted one of the two premises of the article: that Fairfax County drivers are the worst for texting while driving.  We don’t yet know what jurisdiction will be saddled with that ignominious title, but as we persist in our analysis, that will surely be revealed. The second premise of the article is more complicated to address.

Question Two: Is There a “Clear Trend” That Drivers in Northern Virginia Are “Most Likely” to be Convicted of Texting While Driving?

With the data we gathered, we chose to define “most likely to be convicted” as “most convictions per mile driven within a jurisdiction,” which is more accurate than “most convictions, regardless of any other consideration, in a jurisdiction.”

What Are the Conviction Rates Per Miles Traveled in Virginia Jurisdictions?

In the chart below, we have placed each city and county in Virginia on a scatterplot, where the X-axis is the total miles traveled over a six-month period, and the Y-axis is the number of TWD convictions per 100 million miles traveled.

With this layout, any jurisdiction in which drivers are “highly likely” to have a TWD conviction would be placed towards the top, regardless of where the marker is horizontally.

Are the Highest Conviction Rates Concentrated in Northern Virginia Jurisdictions?

Even without labeling each jurisdiction, we can determine three important things from the above scatterplot immediately:

  1. The two jurisdictions with the highest rates of convictions per miles traveled are not in Northern Virginia. (They are the cities of Winchester and Williamsburg.)
  2. While more miles traveled, generally speaking, means more TWD convictions, the actual TWD conviction rate is not directly related to miles traveled. If it did, there would be a clear line that all the points on the graph cluster around…but there isn’t.

This is more obvious when we remove Fairfax County, the rightmost point, from the chart entirely:

Lots of jurisdictions are clustered in the bottom left corner of the chart, meaning that there are few convictions and not many miles traveled therein.

However, among the remaining jurisdictions, there’s tremendous variation. Some are low mileage/high offender jurisdictions, while some are the exact opposite.

What’s more, there seem to be Northern and non-Northern Virginia jurisdictions all over the chart. Apparently, there is nothing inherent about being in any particular Northern Virginia jurisdiction that makes one more or less likely to be convicted of TWD.

Where Exactly Are the Worst Jurisdictions Statewide?

To see the geographic distinction more clearly, we created the above map, which shows where drivers are more and less likely to be convicted of TWD. Below, the state’s jurisdictions are colored from light red (0-1 convictions per 100 million miles driven) to dark red (7 or more convictions per 100 million miles).  Additionally, the top jurisdictions in Northern Virginia and outside of Northern Virginia for the dubious mantle of “most convictions per mile driven” are shown in tables.

Using the scatterplots and the map, it becomes even more obvious that specific jurisdictions where drivers are “highly-likely” to be convicted of texting while driving exist throughout the state, not just in the Northern Virginia region.

Where Do Things Stand Now?

At this point, it seems clear that the statements in the original article were overly simplistic and not supported by the data presented. We’ve also answered our two questions:

  1.  Fairfax County is not the worst jurisdiction for TWD. It merely has the most convictions. The worst jurisdiction overall is Winchester City, and the worst county is Arlington County.
  2. On a jurisdiction-by-jurisdiction basis, there is no clear trend that Northern Virginia drivers are most likely to be convicted of TWD. There are jurisdictions all over the state with high conviction rates and with low conviction rates.

Let’s continue on in our analysis, and see if we can find something else interesting in the data.

What Else Could Explain Texting While Driving Rates?

What Happens If We Group Jurisdictions Together?

Let’s return to the original premise—are Northern Virginia drivers, taken as a whole, more likely to be convicted?  We’ve seen that this is not true on a jurisdiction-by-jurisdiction level. However, there may be other ways we can group our data that might turn out to be more telling.

Northern Virginia vs. Other Virginia

When we aggregate all the jurisdictions into “Northern Virginia” and “Other Virginia,” here is what the data shows:

Region Number of Jurisdictions Est. Vehicle Miles Traveled, Jul-Dec 2012 Convictions, Jul-Dec 2013 Convictions per 100M Miles
Northern Virginia 17 12,383,815,123 334 2.70
Other Virginia 117 28,316,705,541 387 1.37

On aggregate, it turns out that the Northern Virginia group has been convicted of TWD about twice as frequently per mile driven than operators in other parts of the Commonwealth. This is actually what the original article said, albeit without sufficient evidence to support such a claim. But we already know that there’s great variation in in conviction rates across the state, so let’s keep looking for an explanation for this disparity.

Counties vs. Cities

One person quoted in the original article speculated that an increased number of convictions could be due to traffic speed…people sitting in slow-moving or idling traffic could be more prone to texting, due to boredom, or perhaps a perception that texting in such situations is not really that dangerous. 

Cities, on average, have slower traffic speeds and more stops per mile driven than counties (particularly rural counties, as are present outside of Northern Virginia). What happens if we group our driving into counties and cities?

Type of Jurisdiction Number of Jurisdictions Est. Vehicle Miles Traveled, Jul-Dec 2012 Convictions, Jul-Dec 2013 Convictions per 100M Miles
City 39 9,797,132,137 218 2.23
County 95 30,903,388,527 503 1.63

While not as big as the difference between Northern and rural Virginia, the difference in conviction rates between city jurisdictions and county jurisdictions is significant, with the former rate being 37% higher.

Region and Jurisdiction Type

When we perform both of the above breakdowns simultaneously, the data table looks like this:

Region/ Jurisdiction Type Number of Jurisdictions Est. Miles Traveled, Jul-Dec 2012 Convictions, Jul-Dec 2013 Convictions per 100,000,000 Miles
Northern Virginia
  Cities 6 832,667,740 18 2.16
  Counties 11 11,551,147,383 316 2.74
  All 17 12,383,815,123 334 2.70
Other Virginia
  Cities 33 8,964,464,397 200 2.23
  Counties 84 19,352,241,144 187 0.97
  All 117 28,316,705,541 38 1.37
  Cities 39 9,797,132,136 218 2.23
  Counties 95 30,903,388,528 503 1.63
  All 134 40,700,520,664 721 1.77

Some key observations from this data would be:

  1. In both Northern and non-Northern Virginia, the conviction rate in cities is remarkably consistent. (Statewide it’s 2.225 per 100 million miles; in NoVa, it’s 2.161; and in non-NoVa, it’s 2.231.)
  2. Northern Virginia counties combined have a conviction rate nearly three times higher than non-Northern Virginia counties (2.74 vs. 0.97).
  3.  “Other Virginia” counties have by far the lowest aggregated conviction rate.

Perhaps it’s not Northern Virginia that’s anomalous, but rather rural Virginia.

What happens if we exclude those counties from our analysis, but retain cities outside of Northern Virginia?

Region/Jurisdiction Type Number of Jurisdictions Est. Miles, Jul-Dec 2012 Convictions, Jul-Dec 2013 Convictions per 100M Miles
Virginia (excluding counties outside of Northern Virginia)
  Cities (statewide) 39 9,797,132,136 218 2.23
  Counties (NoVA) 11 11,551,147,383 316 2.74
  All 50 21,348,279,520 534 2.50

Now a reasonable explanation is starting to emerge:

  • Cities, no matter where they are in the state, have similar conviction rates;
  • Northern Virginia counties, perhaps due to dense populations and more commuter traffic, have rates even higher than cities;
  • Rural counties have decidedly lower rates than either cities or NoVa counties.

What explains the fluctuations in convictions on a jurisdiction-by-jurisdiction basis? That is beyond the scope of what we were investigating here, but it could have to do with the local law enforcement profile (number of patrols, aggressiveness in looking for such offenses, etc.), or with the frequency of traffic backups or slowdowns, or even with the differences in texting frequency overall among jurisdictions.

We also don’t have several years of data, nor do we have very large numbers with which to work. For example, in 2012, Virginia State Troopers (according to the 2012 Facts and Figures report issued by the State Police) made 598,889 traffic arrests, including 179,421 speeding, 73,996 reckless driving and 5,295 for driving under the influence. A rate of 1,440 TWD convictions per year is tiny by comparison.

What is the Point Here?

It is irresponsible to make assertions and assessments in an article that clearly lacks the data and the analysis to support those statements. As published, the WTOP article was merely a press release masquerading as data journalism, and was easily misinterpreted by the audience.

Carelessness in the way data is presented in the media should not be acceptable practice, particularly when the tools and the capacity to inform and enlighten the public through visual or explanatory analysis are so readily available.

WTOP chose not to provide sufficient context for the information presented in the piece from January 29, and additionally chose to use language that was not warranted by the bare minimum of data presented. 

By incorporating baseline data from publicly available sources, and performing some simple to intermediate data analysis, the story could have been informative, accurate, and thorough, presenting a nuanced but enlightening take on a public safety initiative that affects the entire region.

Instead, the article was given only a cursory edit to distinguish it from the press release from whence it came, and proceeded to mislead its audience – in some cases, intentionally – in order to, perhaps, increase its virality.

We hope that by pointing out this disparity, we can illustrate that, at times, it only takes a small amount of critical thinking (in this case, in the form of data analysis) to move from reporting facts to providing the truth.


Mike Cisneros is on Twitter at @mikevizneros.