concurring in the judgment:
I find myself in the perplexing position of being bound by a precedent counseling that I need not be bound by a precedent.
Although there is much in the court’s fine opinion with which I agree entirely' — I would have no quarrel with its analysis of California law were the issue one of first impression — I am profoundly troubled by the notion of reaching this result by our “overruling,” as a three-judge panel, the precedent set by an earlier panel in Jones v. Heskett (In re Jones), 106 F.3d 923 (9th Cir.1997).
I
It is a bedrock principle of our court that the published decision of one three-judge panel binds every other panel,1 from *1084that day forward.2 Put another way, one panel may not overrule another; the power to overrule' is confided to the en banc court, and the en banc court alone.3 Panels may distinguish; they may question; they may deploy virtually any of the other verbs in the Shepard’s vocabulary. But they may not overrule.
There are exceptions. We need not convene the en banc court when the Supreme Court reverses us directly. Nor must we do so when that Court, in reviewing a case from another circuit, knocks the props out from under one of our decisions. See, e.g., Le Vick v. Skaggs Cos., Inc., 701 F.2d 777, 778 (9th Cir.1983); Piedmont Label Co. v. Sun Garden Packing Co., 598 F.2d 491, 495 (9th Cir.1979); see also, e.g., Circuit City Stores, Inc. v. Najd, 294 F.3d 1104 (9th Cir.2002) (noting that Duffield v. Robertson Stephens & Co., 144 F.3d 1182 (9th Cir.1998), has likely been implicitly overruled). This practice represents our confidence, as a court, that our three-judge panels are able to tell the difference between a Supreme Court ruling that rips one of our decisions from the Federal Reporter altogether and one that leaves at least a hanging chad behind. But it also represents our confidence that the Supreme Court stands ready to review and to reverse us when necessary (a proposition for which, I think, no citation is required).
We also permit our panels to use the big eraser when the earlier decision is based on state law that has demonstrably changed in the intervening period. Herein lies the rub. A proposition’s demonstrability depends on the audience’s receptivity. How skeptical must we ask our panels to be when they are urged to exercise the power to overrule in light of a supervening change in the underlying law?
To my mind, a panel must not act in contravention of our precedent without being highly certain of its authority to do so. And that certainty is not easily obtained when, as here, the alleged change in state law comes from case law rather than statutory law.
When it is a state statute that has changed, the question is much simpler, particularly in this age of formal codification.4
In most cases, we need no longer hunt through yellowing volumes of the California state session laws; either the state legislature has altered the statutory section relied upon in the prior decision, or it has not.
But with case law, whether pure common law or judicial glosses on statutory law, the question is more difficult. We can be certain that state case law is an authoritative expression of state law only when it comes from the state’s court of last resort.5 *1085Anything less leaves room for doubt — including a decision by an intermediate state appellate court, which, though perhaps weightier authority than a trial court’s ruling, an attorney general’s opinion, or a learned commentator’s pronouncement, is inevitably less than conclusive. And it seems to me that where there is room for doubt, we must stay our erasers.
II
For the contrary position, the court’s opinion relies primarily on Owen ex rel. Owen v. United States, 713 F.2d 1461 (9th Cir.1983). Supra, at 1082-83. I do not disagree with the court’s discussion of Owen, but I find the reasoning of that case troubling. Owen relied on our cases requiring that we “follow the decision of the intermediate appellate courts of the state unless there is ‘compelling evidence that the highest court of the state would decide differently.’ ” Otoen, 713 F.2d at 1464 (quoting Andrade v. City of Phoenix, 692 F.2d 557, 559 (9th Cir.1982) (per curiam)) (internal quotation marks omitted). The Oiven court reasoned that because we must presumptively follow the decisions of state intermediate appellate courts when we decide questions of first impression, we must also revisit our decisions on the same basis. I do not think that the one proposition follows from the other.
Consider a state with an intermediate appeals court whose two divisions do not bind one another with their decisions. Suppose the two divisions simultaneously reach opposite conclusions on the identical question of law. Suppose further that the issue is so thorny and the two decisions so well-reasoned that, despite the diametrically opposed outcomes, the impartial observer can find one more persuasive than the other only by a hairsbreadth — the winner by a preponderance, but not by clear and convicing evidence. Yet we require “convincing evidence” before disregarding the decision of an intermediate appellate court. Owen, 713 F.2d at 1465. In this example, neither decision is weighty enough to justify disregarding the other. A panel of this court confronted with this situation, but blessed with a prior, on-point Ninth Circuit precedent, would likely sigh with relief and apply the law of the circuit, even if one or both of the state eases postdated the earlier panel’s decision. The law in our court and those bound by our rulings remains unchanged, predictable, reliable until the state’s highest court tells us otherwise.
I believe this salutary predictability justifies following circuit precedent even when a state intermediate appellate court subsequently issues a contrary opinion. One never knows, after all, when the other shoe will drop and another state court will take the opposite position. California maintains a dispersed intermediate appellate court, with six independent districts. Cal. Gov’t Code § 69100. Are we to revisit our rulings each time the weight of authority shifts? (If the First District decides on holding A, the Second and Third hand down holding not-A the next year, and the Fourth, Fifth, and Sixth weigh in with holding A the following year, will we have to undertake three overrulings, with a fourth when the California Supreme Court finally settles on outcome not-A?) I would prefer to keep to a minimum the frequency with which we receive a new datum, revise our view of state law accordingly, and reverse ourselves.
*1086But Can we not ascertain from our own reading of the law how likely it is that this parade of horribles will actually march in any given case? To do so here, we would have to evaluate whether the Jones panel’s decision is so out of line with California law that no other Court of Appeal is likely to adopt it. And it is precisely that sort of on-the-merits reexamination of prior precedent that we are supposed to leave to the en banc court.6
HI
One could certainly argue that the mere fact of a panel opinion should not be given this near-conclusive weight. After all, the three judges (or two) who arrived at the precedential holding in question may or may not have had the benefit of thorough briefing, immersion in the pertinent state law, or a nutritious and balanced breakfast. Cf. Payne v. Tennessee, 501 U.S. 808, 834, 111 S.Ct. 2597, 115 L.Ed.2d 720 (1991) (Scalia, J., concurring) (“[WJhat would enshrine power as the governing principle of this Court is the notion that an important constitutional decision with plainly inadequate rational support must be left in place for the sole reason that it once attracted five votes.”).
But that is not our usual rule — and for good reason. Stare decisis is of particular importance in federal courts. We are, after all, courts of limited jurisdiction that do not enjoy the general common lawmaking authority that many state courts do. And the fact that federal judges are not lawmakers is inextricable from the fact that we enjoy the constitutional armoring that secures our independence — appointment (rather than election), life tenure, and salary protection. Cf. Chisom v. Roemer, 501 U.S. 380, 400, 111 S.Ct. 2354, 115 L.Ed.2d 348 (1991); Northern Pipeline Constr. Co. v. Marathon Pipe Line Co., 458 U.S. 50, 59 n. 10, 102 S.Ct. 2858, 73 L.Ed.2d 598 (1982) (plurality opinion). Our status ill suits us to lawmaking; indeed, the Framers’ expectation that we would not be making law secured us our judicial independence in the first place.
Stare decisis provides crucial reassurance on the latter point: it demonstrates that our decisions represent more than the subjective preferences of the concurring judges. The Federalist No. 78, at 471 (Alexander Hamilton) (Clinton Rossiter ed. 1961) (“To avoid an arbitrary discretion in the courts, it is indispensable that they should be bound down by strict rules and precedents which serve to define and point out their duty in every particular case that comes before them.... ”); see also, e.g., Pollock v. Farmers’ Loan & Trust Co., 157 U.S. 429, 652, 15 S.Ct. 673, 39 L.Ed. 759 (1895) (White, J., dissenting) (“The fundamental conception of a judicial body is that of one hedged about by precedents which are binding on the court without regard to the personality of its members.”). In the courts of appeals, we three-judge panels bind ourselves rigorously to this mast and allow only the en banc court to release us. I find it unfortunate that Owen and Stephan v. Dowdle, 733 F.2d 642 (9th Cir.1984), have departed from this rule. See supra at 1083.
But depart they have, and I must respect those holdings, for they cannot be dismissed as dicta. See, e.g., Spears v. Stewart, 283 F.3d 992, 1006 (9th Cir.2002) *1087(Kozinski, J., statement concerning the denial of the petitions for rehearing en banc) (“[S]o long as the issue is presented in the case and expressly addressed in the opinion, that holding is binding and cannot be overlooked or ignored by later panels of this court or by other courts of the circuit.”)- Nor have they been overruled.
Thus, reluctantly applying the Owen rule, I agree with the majority’s conclusion that Jones is no longer viable and must be rejected — although, for the reasons detailed above, I decline to use the term “overruled.” Accordingly, I concur in the judgment.
. Indeed, our three-judge panels are bound even before the litigants themselves are bound. Even when a judgment is not yet enforceable, the opinion remains circuit precedent unless and until a majority of judges vote to take it en banc, at which point it may not be cited. 9th Cir. Gen. Order 5.5(d); see, e.g., Newdow v. U.S. Cong., 292 F.3d 597 (9th Cir.2002), judgment purportedly "stayed” by one-judge order (9th Cir. June 27, 2002) (No. 00-16423).
. And occasionally backward: because a panel’s decision in any given case is controlling over others submitted for decision afterward, occasionally a case is submitted first but decided second, necessitating the amendment or withdrawal of the case submitted second but decided first. See 9th Cir. Gen. Order 4.1(a).
. Although, in this circuit, the en banc court that exercises that authority in the first in- • stance is a limited, eleven-judge panel rather than the full court, see 9th Cir. R. 35-3, the decision to convene the en banc court is made by a majority of the court's active, nonrecused circuit judges, as the governing statute mandates. 28 U.S.C. § 46(c); Fed. R.App. P. 35. And participation on the en banc court' is specifically restricted to judges of this court (and, with one exception, see 28 U.S.C. § 26(c)(1), to judges in regular active service).
. “Formal codification” refers to the consolidation of legislative enactments under readily indexed subject headings, rather than an attempt to reduce all substantive law to the form of code law to the exclusion of common law. Gunther A. Weiss, The Enchantment of Codification in the Common Law World, 25 Yale X Int’l L. 435, 517 & n. 418 (2000).
. Certainty that state case law is an authoritative and enduring expression of state law is at its highest when a state court of last resort construes a statute, rather than the state’con*1085stitution or the common law, due to the customary rule that stare decisis is most strongly applicable in statutory construction cases, see, e.g., Patterson v. McLean Credit Union, 491 U.S. 164, 172-73, 109 S.Ct. 2363, 105 L.Ed.2d 132 (1989). But the possibility that the state's highest court will one day overrule itself does not make its extant rulings any less authoritative for our purposes, if not the historians’.
. I should add that one way to reduce the incidence of this difficult situation is to exercise the utmost restraint in publishing prece-dential decisions in diversity cases. However, that tool is not available to us in cases like this one, where federal law incorporates or intersects with state law. We must apply the same .Erie-derived principles to ascertain the state law, but we must also give precedential effect to our application of the federal law in some cases, thus creating federal court precedent on state law that Erie principles may later undermine.