Rodney Howze fired a gun into the air to disperse a group of unruly teenagers. This was not a good idea: Howze, who had several criminal convictions, was not allowed to possess a gun, let alone fire one on a crowded street. He pleaded guilty to violating 18 U.S.C. § 922(g), which prohibits felons from having guns. The prosecutor proposed to treat three of Howze’s prior convictions as “violent felonies,” which would require a 15-year mandatory minimum sentence under the Armed Career Criminal Act, 18 U.S.C. § 924(e). The district court agreed with the prosecutor and sentenced Howze to 180 months’ imprisonment. He concedes on appeal that one of the priors — making terroristic threats — meets the statutory definition. But he objects to giving the same characterization to his convictions for theft from a person and fleeing from an officer. Howze has other convictions whose significance is not argued, so we limit attention to these two.

For purposes of § 922(g), “violent felony” is a crime punishable by imprisonment for more than a year that:

(i) has as an element the use, attempted use, or threatened use of physical force against the person of another; or

(ii) is burglary, arson, or extortion, involves use of explosives, or otherwise involves conduct that presents a serious potential risk of physical injury to another[.]

18 U.S.C. § 924(e)(2)(B). Taylor v. United States, 495 U.S. 575, 110 S.Ct. 2143, 109 L.Ed.2d 607 (1990), the Supreme Court’s[*921] only decision interpreting this language, sets the stage for analysis.

The issue in Taylor was whether every crime bearing the label “burglary” in every state’s criminal code is “burglary” under § 924(e)(2)(B)(ii). The defendant argued that his particular burglaries should not be counted under subsection (ii) because they did not present a risk of physical injury. The Supreme Court decided, first, that classification depends on the nature of the offense as defined in the criminal code rather than either the label the state applies or the specific acts the defendant committed, and, second, that the best way to determine which offenses count as “burglary” for purposes of federal law is to determine which offenses pose risks that force will be used. Classification, in other words, is categorical, as one might expect for a recidivist statute. See 495 U.S. at 590, 601-02, 110 S.Ct. 2148. The Court concluded that “burglary” in subsection (ii) comprises all offenses that entail entries into (potentially) occupied buildings such as residences and offices, because these present a risk that the burglar will encounter the occupant and violence ensue. Because states may draw the line between burglary and other theft crimes differently, however, the Court observed that it may be necessary to look at the charging papers (the indictment or information) to determine whether a particular crime concerned a building or some other structure, such as a box ear or chicken coop. What follows from Taylor is that courts classify convictions rather than acts — that is, recidivist sentencing under § 922(g) follows charge-offense rather than real-offense principles — -but that the acts alleged in the charge may be essential to determine what offense the accused stands convicted of. Any dispute about the nature of the conviction must be resolved from the text of the charge and not by holding an eviden-tiary hearing.

Now let us take up Howze’s conviction for fleeing from an officer. He pleaded guilty to violating this rule of Wisconsin law:

No operator of a vehicle, after having received a visual or audible signal from a traffic officer, or marked police vehicle, shall knowingly flee or attempt to elude any traffic officer by willful or wanton disregard of such signal so as to interfere with or endanger the operation of the police vehicle, or the traffic officer or other vehicles or pedestrians, nor shall the operator increase the speed of the operator’s vehicle or extinguish the lights of the vehicle in an attempt to elude or flee.

Wis. Stat. § 346.04(8). This statute must be classified under the “serious potential risk” aspect of subsection (ii). We have held that Taylor’s categorical approach applies to that issue. See United States v. Shannon, 110 F.3d 382, 384-85 (7th Cir.1997) (en banc).

Howze insists that one can violate Wis. Stat. § 346.04(3) without creating a “serious potential risk of physical injury to another”. In his view, a motorist who disobeys police instructions and increases speed in an attempt to flee does not threaten the safety of bystanders so long as he does not exceed the speed limit or violate some other law along the way. We rejected a similar argument in United States v. Bryant, 310 F.3d 550 (7th Cir.2002). Bryant held that escape from custody always is a “crime of violence”. Bryant failed to return to a halfway house after a spell of work release; he broke no walls and did nothing illegal other than to remain outside the halfway house beyond the permitted time. We concluded that bust-out escapes and simple failures to return should be classified the same because the crime is the same and because that crime holds the potential for violence[*922] during attempted recaptures. Drawing a distinction based on the way in which the defendant got loose would be inconsistent with the categorical approach required by Taylor, we held. See 310 F.3d at 553-54. Because the statute calls for an assessment of risk rather than actual outcomes, and the risk that someone will get hurt during recapture (or flight to avoid recapture) does not depend on how the offender got away in the first place, all escapes must be classified as crimes of violence. See also United States v. Franklin, 302 F.3d 722, 725 (7th Cir.2002). Every other circuit that has addressed this issue has come to the same conclusion. See United States v. Springfield, 196 F.3d 1180, 1185 (10th Cir.1999); United States v. Houston, 187 F.3d 593, 594 (6th Cir.1999); United States v. Abernathy, 277 F.3d 1048, 1051 (8th Cir.2002); United States v. Hairston, 71 F.3d 115, 118 (4th Cir.1995); United States v. Ruiz, 180 F.3d 675, 677 (5th Cir.1999).

Flight to avoid apprehension is one means through which the risk of escape may be realized. When the crime is flight to avoid apprehension, what is only a risk for escape becomes a certainty. Bystanders are in particular jeopardy. Collisions between fleeing vehicles and pedestrians or others who get in the way are common. See, e.g., Sacramento v. Lewis, 523 U.S. 833, 118 S.Ct. 1708, 140 L.Ed.2d 1043 (1998); Mays v. East St. Louis, 123 F.3d 999 (7th Cir.1997); Campbell v. White, 916 F.2d 421 (7th Cir.1990). Thus, if all escapes are violent crimes, all flights to avoid arrest must be violent crimes. Indeed, flight may be even more dangerous than escape, because many escapes do not entail flight to avoid capture — but all flights involve that risk-creating conduct. Howze observes that the custodial status of a person fleeing from the police is different from that of a prisoner escaping from a place of incarceration. True enough, but not very helpful. To distinguish Bryant and Franklin on the basis of the defendant’s custodial status, Howze would have to present evidence that flight after custody (that is, escape) is more likely to result in a “serious potential risk of physical injury to another” than is flight before custody (the crime defined by Wis. Stat. § 346.04(3)). Howze has not come up with any evidence along these lines, and we are not aware of any.

Robbery is the other conviction that requires classification. Howze was convicted of violating Minn.Stat. § 609.52, which covers a variety of theft offenses— such a large variety, indeed, that it is necessary to consult the charging papers to see which kind Howze committed. He was charged with stealing a bicycle from its rider, in violation of Minn.Stat. § 609.52(2X1) and (3)(3)(d). Here is the statutory language:

Subdivision 2. Whoever does any of the following commits theft and may be sentenced as provided in subdivision 3:

(1) intentionally and without claim of right takes, uses, transfers, conceals or retains possession of movable property of another without the other’s consent and with intent to deprive the owner permanently of possession of the property[.]

Subdivision 3. Whoever commits theft may be sentenced as follows:

(3) to imprisonment for not more than five years or to payment of a fine of not more than $10,000, or both, if:

(d) the value of the property or services stolen is not more than $500, and any of the following circumstances exist:

(i) the property is taken from the person of another or from a corpse,[*923] or grave or coffin containing a corpse[.]

Howze, the prosecutor, and the district court all treat subdivision (3), nominally a sentencing rule, as establishing elements of a distinct offense. The charging papers show, moreover, that Howze took the bicycle from a live person rather than from a corpse, coffin, or grave. Thus Howze has been convicted of theft “from the person of another”, a crime that in Minnesota carries a maximum punishment of five years’ imprisonment. Other examples of this offense are purse snatching, pickpocketing, and most varieties of grab-and-run robbery.

What theft from a person has in common with generic burglary is that both entail a risk that violence will erupt between the thief and the victim. The Court held in Taylor that the risk of meeting someone is enough to include within the statute all burglaries of potentially occupied buildings, even though about 87% of all these occur when the building is empty. See Gertrud M. Fremling & John R. Lott, Jr., The Surprising Finding That “Cultural Worldviews” Don’t Explain People’s Views on Gun Control, 151 U. Pa. L.Rev. 1341, 1345-46 (2003). A meeting between criminal and victim, which occurs in 13% of burglaries, occurs in 100% of robberies from persons. In Minnesota, it is impossible to steal anything “from the person” without physical proximity to the victim. See In re Welfare of D.D.S., 396 N.W.2d 831, 832-33 (Minn.1986); In re Welfare of D.I.S., 2002 WL 378084 at *1-2 (MinnApp. Mar.12, 2002) (a theft is “from the person” when “the perpetrator’s actual or threatened violence causes the victim to flee the scene, thus preventing the victim from protecting the property left at the scene”). Doubtless the chance of running into someone during the commission of a theft gives only a rough estimate for the risk of injury. People may dig in their heels (or frighten more easily) when cornered indoors, so the risk that violence will erupt may be greater when a burglar encounters a householder than when a thief makes off with a purse or bicycle. But the thief is eight times more likely to encounter the victim in these street crimes than in burglaries, so the expected risk is substantial even if the risk per encounter is lower for. street thefts than for thefts from buildings.

It would help to know the likelihood that unarmed street crimes lead to injury. Unfortunately, the FBI’s Uniform Crime Statistics do not include measures of these injuries, and victimization surveys approach the subject only indirectly, by counting how many days of work the victims miss. This proxy is imperfect, for victims may lose time from work for many reasons, including the need to appear in court as well as the need to nurse injuries. Still, while the short-term worktime loss may be caused by the need to complete incident-related paperwork and replace stolen documents, people rarely miss more than a week from work unless they were injured. Thus, the frequency of substantial absence from work may be a good estimate of the injury rate. A study conducted by the Department of Justice shows that about 0.7% of all victims of purse snatching and pickpocketing lose more than 11 days of work; 0.75% lose 6 to 10 days, and 58% lose 1 to 5 days. See Bureau of Justice Statistics, Criminal Victimization in the United States 2001 Table 89 (Selected Personal and Property Crimes) (January 2003), at www.ojp.us-doj.gov/bjs/abstract/cvusst.htm. That is to say, about 1.5% of all victims of the most common varieties of street thefts lose more than 6 days from work, a loss implying injury. The real number may well be higher; most injured victims may not be hurt badly enough to miss more than 6 days of work. And the survey category is dominated by pickpocketing, a stealthy crime about five times more common than[*924] purse snatching and similar acts (see Table 26, Personal Crimes, 2001) but less likely to injure the victim. But injury in 2% or so of all street thefts is enough, we think, to fall within the statutory definition; this is at least as likely (in the aggregate) as injury from burglary or escape.

Every other circuit that has addressed this subject has held that street thefts are crimes of violence under the definition in § 924(e)(2)(B)(ii). See, e.g., United States v. Griffith, 301 F.3d 880, 885 (8th Cir.2002); United States v. Payne, 163 F.3d 371, 374-75 (6th Cir.1998); United States v. Wofford, 122 F.3d 787, 794 (9th Cir.1997); United States v. Hawkins, 69 F.3d 11, 13 (5th Cir.1995); United States v. De Jesus, 984 F.2d 21, 24-25 (1st Cir.1993). This is not, however, our court’s first encounter with an issue of this kind. United States v. Lee, 22 F.3d 736 (7th Cir.1994), dealt with the career offender enhancement in the Sentencing Guidelines, U.S.S.G. § 4B1.2 and Application Note 2, a provision sufficiently similar to the Armed Career Criminal statute that in Shannon we treated the two as interchangeable. Lee held, among other things, that theft from a person is not a crime of violence because “[t]he Government’s claim that pick-pocketing and similar crimes pose a substantial risk of injury is, in fact, little more than conjecture”. 22 F.3d at 741.

The United States asks us to overrule this aspect of Lee and eliminate the conflict. That’s a good reason for reconsideration; a circuit that stands alone against the considered view of coordinate courts should be willing to rethink. See United States v. Carlos-Colmenares, 253 F.3d 276 (7th Cir.2001). Lee neglected to apply Taylor's framework (indeed, did not cite that decision) and did not inquire whether it was significant that every street theft entails a close encounter between criminal and victim, an encounter that creates the potential not only for violence but also for injury caused by the act of taking. A purse snatching may dislocate the victim’s shoulder or elbow, or lacerate her arm; a bicycle theft may injure the owner if the thief blocks his path or shoves the bike over to dislodge its rider; and so on. It was logical arguments of this kind, and not data, that won the day in Taylor and our escape cases. Thus Lee is out of line not only with other circuits, but also with how this circuit has analyzed related issues in later decisions such as Bryant and Franklin. See also United States v. Davis, 16 F.3d 212, 217 (7th Cir.1994) (proximity of thief and victim used as indicator of likely injury from attempted burglary).

In order to bring harmony both within and among the circuits, we now overrule the portion of Lee dealing with the treatment of theft and hold that theft from a person is a violent felony under the Armed Career Criminal Act. The other principal holding of Lee — that classification must be based on the elements of the offense as revealed by the charging papers, rather than on the facts of the particular case — is compatible with Taylor and was adopted en banc by Shannon, 110 F.3d at 384-85. It remains good law. As an overruling opinion, this decision was circulated to all judges in regular active service. See Circuit Rule 40(e). No judge requested a hearing en banc.

Affirmed.