Ervine Davenport v. Duncan MacLaren

975 F.3d 537

Opinion

RECOMMENDED FOR FULL-TEXT PUBLICATION Pursuant to Sixth Circuit I.O.P. 32.1(b) File Name: 20a0307p.06

UNITED STATES COURT OF APPEALS FOR THE SIXTH CIRCUIT

ERVINE LEE DAVENPORT, ┐ Petitioner-Appellant, │ │ > No. 17-2267 v. │ │ │ DUNCAN MACLAREN, Warden, │ Respondent-Appellee. │ ┘

On Petition for Rehearing En Banc United States District Court for the Western District of Michigan at Grand Rapids. No. 1:14-cv-01012—Ellen S. Carmody, Magistrate Judge.

Decided and Filed: September 15, 2020

Before: COLE, Chief Judge; STRANCH and READLER, Circuit Judges.

_________________

COUNSEL

ON PETITION FOR REHEARING EN BANC: Fadwa A. Hammoud, Jared D. Schultz, OFFICE OF THE MICHIGAN ATTORNEY GENERAL, Lansing, Michigan, for Appellee. ON RESPONSE: Tasha J. Bahal, Reuven Dashevsky, WILMER CUTLER PICKERING HALE AND DORR LLP, Boston, Massachusetts, for Appellant.

The panel issued an order denying the petition for rehearing en banc. STRANCH, J. (pp. 3–7), delivered a separate opinion concurring in the denial of rehearing en banc, in which COLE, C.J., and MOORE, CLAY, WHITE, and DONALD, JJ., joined. SUTTON, J. (pp. 8–11), delivered a separate opinion concurring in the denial of rehearing en banc, in which KETHLEDGE, J., joined. GRIFFIN, J. (pp.12–14), delivered a separate opinion dissenting from the denial of rehearing en banc. THAPAR, J. (pp. 15–28), delivered a separate opinion dissenting from the denial of rehearing en banc, in which BUSH, LARSEN, NALBANDIAN, READLER, and MURPHY, JJ., joined. No. 17-2267 Davenport v. MacLaren Page 2

_________________

ORDER _________________

The court received a petition for rehearing en banc. The original panel has reviewed the petition for rehearing and concludes that the issues raised in the petition were fully considered upon the original submission and decision. The petition then was circulated to the full court. Less than a majority of the judges voted in favor of rehearing en banc.

Therefore, the petition is denied. Judge Readler would grant rehearing for the reasons stated in his dissent. No. 17-2267 Davenport v. MacLaren Page 3

_________________

CONCURRENCE _________________

STRANCH, Circuit Judge, concurring in the denial of rehearing en banc. The en banc petition and the dissents throughout the case argue that we stand alone in how we assess harmless error when, in fact, we stand in the company of our sister circuits and follow the precedent of the Supreme Court. We have long held that an underlying trial error must have had a “substantial and injurious effect or influence in determining the jury’s verdict” for a habeas petitioner to survive harmless error review. Brecht v. Abrahamson, 507 U.S. 619, 637 (1993) (quoting Kotteakos v. United States, 328 U.S. 750, 776 (1946)). Following the Supreme Court, we have also repeatedly (and uniformly) held that applying Brecht exacts AEDPA deference, in part because it is so much more onerous on habeas petitioners than the Chapman v. California standard applied by state courts on direct review, 386 U.S. 18 (1967). The Supreme Court recently reiterated the propriety of this approach, concluding: “[i]n sum, a prisoner who seeks federal habeas relief must satisfy Brecht, and if the state court adjudicated his claim on the merits, the Brecht test subsumes the limitations imposed by AEDPA.” Davis v. Ayala, 576 U.S. 257, 270 (2015). In that scenario, “a federal habeas court need not ‘formal[ly]’ apply both Brecht and ‘AEDPA/Chapman.’” Id. at 268 (quoting Fry v. Pliler, 551 U.S. 112, 119–20 (2007)). The opinions that accompany the denial of rehearing en banc are mistaken to conclude that the panel majority did anything more than apply this test, as authorized by the Supreme Court, to Davenport’s case.

Against the plain language of the Supreme Court, the en banc petition and now the dissenters argue that courts must be required to apply a separate AEDPA test on top of Brecht. But that would make our circuit the outlier. The en banc petition and the original and current dissents, moreover, cannot cite a single case across the circuit courts of appeal in which a habeas petitioner who prevailed under Brecht’s harmlessness inquiry was then required to pass through the gauntlet of a second harmlessness test. This is because, as pointed out at length in the majority opinion, the position advanced in the en banc petition rests on a series of legal errors. There is no dispute that both Brecht and AEDPA must be satisfied for a habeas petitioner to No. 17-2267 Davenport v. MacLaren Page 4

show that a constitutional error was not harmless. The Supreme Court’s teaching in Ayala that the Brecht test subsumes the limitations imposed by AEDPA and our acknowledgment that the Brecht test handles the work of both tests show that the test the panel majority applied accomplishes that principle. Davenport also recognizes that courts may choose to apply AEDPA/Chapman before turning to Brecht’s more demanding inquiry. In sum, Davenport merely reiterates the unremarkable tenet that when the state concedes a constitutional trial error, we can assess harmlessness under Brecht. A closer look at the opinions cited in the en banc petition and the original dissent reveals a consistent interplay between the tests: if a petitioner would lose under AEDPA/Chapman, he necessarily would lose under Brecht; if the petitioner would prevail under Brecht, he necessarily would prevail under AEDPA/Chapman.

The opinions filed with the denial of rehearing en banc either parrot the original dissent, faltering for the reasons explained at length in the majority opinion, or raise peripheral and undisputed issues inappropriate for further appellate review in this case. A critical error, from which others follow, is the notion that our majority decision turns wide of AEDPA’s guardrails or simply ignores them altogether. Following the original dissent’s call for en banc review, pages of the majority opinion were devoted to showing that far from ignoring AEDPA, we apply it—full force—to Davenport’s case. See, e.g., Davenport v. MacLaren, 964 F.3d 448, 458–59 (6th Cir. 2020). The majority also shows that requiring federal habeas tribunals to perform a two-step harmless error analysis contravenes Sixth Circuit and Supreme Court precedent. Id. at 454–59. The majority decision was not waylaid by Fry’s procedural differences; it instead shows that Ayala confirmed the application of Brecht to Davenport’s case. See id. at 456, 458 n.8. In addition, the majority demonstrates that though there may have been a historical circuit split on the standard of review issue, Ayala cleared up the division and specifically authorized the test applied in this case. Id. at 457. And the majority opinion reveals that many of the circuit cases cited by the dissent, read carefully, confirm the precise dynamic between Brecht and AEDPA that the majority applied. See, e.g., Sifuentes v. Brazelton, 825 F.3d 506, 535 (9th Cir. 2016) (citation omitted) (quoting Ayala, 576 U.S. at 270):

In sum, a petitioner “necessarily cannot satisfy” the Brecht requirement of showing that he was “actually prejudiced” by the state court’s error . . . “if a fairminded jurist could agree with the [state appellate court] that this procedure No. 17-2267 Davenport v. MacLaren Page 5

met the Chapman standard of harmlessness.” By the same token, if a petitioner does satisfy the Brecht requirement of showing that an error resulted in “actual prejudice,” then the petitioner necessarily must have shown that the state court’s determination that the error was harmless was objectively unreasonable.

Nor has there been an intra-circuit split in the Sixth Circuit regarding the test to be applied to state court harmlessness assessments. Cases like Stewart v. Trierweiler, 867 F.3d 633 (6th Cir. 2017), and Hollman v. Sprader, 803 F. App’x 841 (6th Cir. 2020), merely show that habeas relief can be denied on AEDPA/Chapman grounds without reaching Brecht—an analytical approach that, as painstakingly explained in the majority opinion, is consistent with Davenport. Davenport, 964 F.3d at 455. Both dissenters from denial of rehearing en banc recently applied the very rule they now decry. Judge Thapar called the harmless error framework “a choice of prompts,” in which one “option—a shortcut of sorts—is to leapfrog AEDPA and jump directly to Brecht.” Hollman v. Sprader, 803 F. App’x 841, 843 (6th Cir. 2020) (citing Ayala, 576 U.S. at 268–70). Judge Griffin wrote: “[t]he Supreme Court and this court have made clear that ‘Brecht is always the test’ for evaluating harmless error on collateral review, even where AEDPA applies.” Reiner v. Woods, 955 F.3d 549, 556 (6th Cir. 2020) (quoting Ruelas v. Wolfenbarger, 580 F.3d 403, 412 (6th Cir. 2009)). The majority decision simply took the unremarkable step of employing our existing circuit standard, which applies Supreme Court precedent. In sum, federal courts across the nation apply Brecht after Ayala. We do too.

My dissenting colleagues also express alarm about the facts of this case in which, different from other cited cases, the Brecht test was satisfied. Admittedly, cases presenting AEDPA issues by their nature contain concerning facts. That is true here. But the manner or type of case does not control what legal standards apply. And the constitutional right to a fair trial cannot depend upon a defendant’s admission to certain underlying—and even egregious— facts.

A few other points are worth mentioning. The majority opinion did not apply a “presumption of prejudice,” as the dissents suppose. That language is not in the majority opinion. Equally misguided is the observation that, notwithstanding the standard of review dispute, the panel majority extended Supreme Court precedent to reach its disposition. My No. 17-2267 Davenport v. MacLaren Page 6

colleagues assert that the majority “relied primarily on Sixth Circuit decisions” in evaluating the state courts’ harmless error decision, (Sutton Concurring Op. at 3), and that it “rel[ied] upon circuit precedent to show what a general Supreme Court standard clearly establishes,” (Thapar Dissenting Op. at 10). True, clearly established law is determined solely by Supreme Court rulings, Stewart v. Erwin, 503 F.3d 488, 493 (6th Cir. 2007), not by circuit precedent, Kernan v. Cuero, 138 S. Ct. 4, 9 (2017). And circuit precedent cannot turn “a general principle of Supreme Court jurisprudence into a specific legal rule” that has not been stated by the Supreme Court. Lopez v. Smith, 574 U.S. 1, 7 (2014) (quoting Marshall v. Rodgers, 569 U.S. 58, 64 (2013)). The majority decision did not violate these dictates. Instead, “in accordance with its usual law-of- the-circuit procedures, [it] look[ed] to circuit precedent to ascertain whether it ha[d] already held that the particular point in issue is clearly established by Supreme Court precedent”—an approach the Supreme Court has explicitly sanctioned. Marshall, 569 U.S. at 64.

The rule is not that the act of citing a Sixth Circuit precedent in this context automatically dooms an opinion. None of the majority’s references to Sixth Circuit precedents overstep the bounds set out by the Supreme Court. The opinion addressed the procedural differences between Davenport’s case and Deck v. Missouri, 544 U.S. 622 (2005). And it then turned to the Supreme Court’s admonition that “[t]he law has long forbidden routine use of visible shackles during the guilt phase” of a criminal trial because “shackling is ‘inherently prejudicial.’” Id. at 626, 635 (quoting Holbrook v. Flynn, 475 U.S. 560, 568 (1986)). Here, the State of Michigan does not dispute that there was no on-the-record justification for the shackling and therefore it was unconstitutional. And the majority opinion did not ignore post-trial juror testimony or speculate about juror bias. Instead, it specifically addressed testimony given by jurors three years after trial that they still remember that Davenport was shackled during trial, leaving them with the impression that he was dangerous. In this context, moreover, the Supreme Court has specified that “[i]f ‘a procedure employed by the State involves such a probability that prejudice will result that it is deemed inherently lacking in due process,’ little stock need be placed in jurors’ claims to the contrary.” Holbrook, 475 U.S. at 570 (citation omitted) (quoting Estes v. Texas, 381 U.S. 532, 542–43 (1965)). Nor does application of the Brecht test make it inappropriate to assess whether shackles branded Davenport as having a violent nature in a case depending on the sharp distinction drawn by Michigan law between first-degree and second-degree murder, which No. 17-2267 Davenport v. MacLaren Page 7

requires that “first-degree murder be committed with premeditation and deliberation.” People v. Morrin, 187 N.W.2d 434, 448–49 (Mich. Ct. App. 1971). In sum, as the majority opinion explicitly noted, the Supreme Court’s shackling jurisprudence was the exclusive basis for its reasoning; it directly applied relevant and specific holdings of Holbrook and Deck. Davenport, 964 F.3d at 466–67.

Ultimately, a majority of this court denied en banc review because it recognized that Davenport is but a recent installment in a chain of Sixth Circuit decisions that follow Supreme Court precedent and apply Brecht to a state court’s harmless error analysis. For the reasons more fully set out in the majority opinion, I concur in the conclusion of our full court that this case does not require en banc review. No. 17-2267 Davenport v. MacLaren Page 8

_________________

CONCURRENCE _________________

SUTTON, Circuit Judge, concurring in the denial of rehearing en banc. This en banc petition implicates a nagging tension between deciding cases correctly and delegating to panels of three the authority to decide cases on behalf of the full court.

I am skeptical that the panel decided this case correctly. At stake is whether Ervine Lee Davenport, a Michigan prisoner convicted of first-degree murder for choking Annette White to death, is entitled to habeas relief for a Fourteenth Amendment Due Process violation caused when the trial court required him to wear shackles during his trial. On direct appeal, the Michigan Supreme Court determined that any constitutional violation was harmless and denied Davenport relief. On collateral review, the Michigan Attorney General’s Office conceded that the trial court’s shackling decision rose to the level of a constitutional violation but defended the state supreme court’s decision that it was harmless. The panel granted the habeas petition on the ground that it was not harmless over a dissent by Judge Readler.

Anyone who has been an Article III judge for five years or more has seen this movie, at least part of this movie, before. What is the correct way to characterize the showing needed to obtain habeas relief when it comes to establishing that the state court’s harmlessness ruling warrants correction after passage of the Antiterrorism and Effective Death Penalty Act of 1996? Debates over the answer center on two pre-AEDPA decisions. Does the Chapman test—which requires the government to “prove beyond a reasonable doubt that the error complained of did not contribute to the verdict”—capture the proper approach together with the AEDPA requirement that any Chapman ruling must be “unreasonable”? Chapman v. California, 386 U.S. 18, 24 (1967); 28 U.S.C. § 2254(d)(1). Or does the Brecht test—which requires a court on collateral review to have deep doubt about whether the error “had substantial and injurious effect or influence in determining the jury’s verdict”—capture the proper approach by itself? Brecht v. Abrahamson, 507 U.S. 619, 637 (1993) (quotation omitted). It has never seemed easy to sort out the difference between these standards, sometimes even to know which one favors whom, and it does not help matters that they both predate AEDPA. More recently, Davis v. Ayala referred to No. 17-2267 Davenport v. MacLaren Page 9

the Brecht formulation, said that it “subsumes” AEDPA, and explained that this language did not make AEDPA irrelevant to harmless-error decisions by state courts. 576 U.S. 257, 269–270 (2015).

For my part, I would answer the problem this way. On direct review of criminal convictions, the Chapman harmlessness standard applies. On collateral review, two main possibilities arise. If the state court issues a ruling on the harmlessness question under Chapman, then habeas claimants may obtain relief only if they can show that the state court’s Chapman ruling sank to the level of an “unreasonable application of” “clearly established” federal law under AEDPA. If the state court does not issue a ruling on the merits of the harmlessness question, then the Brecht standard applies to our review of the decision. This approach seems to account for the Court’s view that Brecht “subsumes” AEDPA and its view that AEDPA remains pertinent to federal court review of state court decisions.

But the standard-of-review question in these sorts of cases—how to capture the ineffable test and the burden of proof in meeting it, to say nothing of the meaning of “subsumes”—can become a distraction, prompting lawyers and courts to overlook core process principles that we must account for under AEDPA before second-guessing state court convictions. The key one overlooked in this case is that we must measure constitutional challenges by the yardstick of United States Supreme Court decisions, not the Chancellor’s foot of our own decisions.

Remember the relevant language of AEDPA. It prohibits us from granting a habeas petition for a claim adjudicated on the merits in state court unless the state court unreasonably applied “clearly established Federal law, as determined by the Supreme Court[.]” 28 U.S.C. § 2254(d)(1). Ayala tells us that a state court’s harmless-error determination constitutes an adjudication on the merits under AEDPA. 576 U.S. at 269. The state courts held that Davenport’s shackling was harmless. That requires us to measure the state court’s decision against holdings of the United States Supreme Court, not holdings of our court. White v. Woodall, 572 U.S. 415, 419 (2014). No. 17-2267 Davenport v. MacLaren Page 10

But that’s not what the panel did. In deciding that the state courts clearly erred in their harmless-error decision, the panel relied primarily on Sixth Circuit decisions. That’s the wrong benchmark. AEDPA forbids it. And the United States Supreme Court has not been reticent about correcting lower courts that violate this core measuring stick under AEDPA. See, e.g., Kernan v. Cuero, 138 S. Ct. 4, 9 (2017) (per curiam); Glebe v. Frost, 574 U.S. 21, 24 (2014) (per curiam).

That brings me to the second process problem. As to the one Supreme Court decision the panel did consider, Deck v. Missouri, 544 U.S. 622 (2005), Deck did not offer a holding on the harmlessness question. While it said that Chapman would apply to any error, see 544 U.S. at 635, it did not provide “clearly established” law about how to apply a harmlessness test to a shackling violation. It simply left matters at a Mt. Everest-level of generality, whether it’s the Chapman formulation mentioned there or the Brecht formulation mentioned in Ayala. So the panel’s reliance on Deck also violates AEDPA and Supreme Court interpretations of it. Nevada v. Jackson, 569 U.S. 505, 512 (2013) (per curiam). Unlike this case, Deck involved the use of physical restraints fully visible to the jury. See Davenport v. MacLaren, 964 F.3d 448, 478 (6th Cir. 2020) (Readler, J., dissenting). And unlike this case, Deck had no reason to apply the harmlessness test to such a violation. Those are distinctions with plenty of differences. Over and over, the Supreme Court has admonished us not to frame its holdings at such a lofty level of generality in deciding what law has been “clearly established.” Lopez v. Smith, 574 U.S. 1, 6 (2014) (per curiam) (quotation omitted); Metrish v. Lancaster, 569 U.S. 351, 367–68 (2013).

These problems with the panel’s decision and its debate over Chapman/Brecht seem to be recurring ones in our circuit and outside of it, suggesting that there is room for clarification by the Supreme Court when it comes to federal court review of state court harmless-error decisions under AEDPA—and the process obligations of lower courts in applying the statute. Just read the eighteen combined pages devoted to the standard of review by this one panel for evidence. It’s been five years since the Court’s most recent contribution to the area and of course many decades since Chapman and Brecht. I suspect every federal judge in the country would welcome guidance in the area. No. 17-2267 Davenport v. MacLaren Page 11

One last note. Query whether future panels need to follow one aspect of the panel’s decision, its use of Sixth Circuit decisions to satisfy AEDPA. That was not a holding of the court on a matter debated by the parties. The Michigan Attorney General’s Office did not raise the point, and as a result it was not joined by the parties or for that matter the panel. A panel’s implicit resolution of an issue that was neither raised by the parties nor engaged by the panel generally does not bind future panels. See United States v. L.A. Tucker Truck Lines, Inc., 344 U.S. 33, 37–38 (1952); Webster v. Fall, 266 U.S. 507, 511 (1925). Because Michigan never raised the point, the panel had no occasion to justify its use of circuit precedent to find a clear violation of federal law. The question thus appears to remain open for future panels to consider in the first instance.

That aspect of the decision makes me reluctant to grant en banc review in this case. So does one other consideration. The problem at hand turns mainly on what to make of language in Supreme Court decisions, particularly the meaning of “subsumes” first mentioned in Fry v. Pliler, 551 U.S. 112, 120 (2007), and repeated in Ayala. Countless inefficiencies arise when a full intermediate court debates the meaning of vexing language from the Supreme Court, the most obvious being this: Not only are we fallible, we are not final either. No. 17-2267 Davenport v. MacLaren Page 12

_________________

DISSENT _________________

GRIFFIN, Circuit Judge, dissenting.

I respectfully dissent from the denial of the petition for rehearing en banc.

By the vote of 8–7, our en banc court has denied respondent’s petition for rehearing en banc. This is most unfortunate for our circuit because the 2–1 panel opinion conflicts with a previous decision of our court and is clearly wrong on a habeas-corpus issue of exceptional importance. While some of my colleagues agree, they nevertheless have opposed the petition in the hope that the Supreme Court will reverse us yet again to clean up our intra-circuit mess.1 This denial of rehearing en banc is reminiscent of CNH Industrial N.V. v. Reese, 138 S. Ct. 761 (2018), wherein we were reversed unanimously by the Supreme Court in a per curiam opinion and admonished that “the en banc Sixth Circuit has been unwilling (or unable) to reconcile its precedents.” Id. at n.2.

The Federal Rules of Appellate Procedure provide an important and necessary remedy for courts of appeals to correct their conflicts and errors of exceptional importance. While en banc hearings or rehearings are not favored, they are authorized when:

1) en banc consideration is necessary to secure or maintain uniformity of the court’s decisions; or 2) the proceeding involves a question of exceptional importance.

Fed. R. App. P. 35(a). This case cries out for rehearing en banc by operation of both 35(a)(1) and 35(a)(2).

First, rehearing en banc is necessary “to secure or maintain uniformity of [our] court’s decisions.” The majority Davenport v. MacLaren opinion is in direct conflict with our prior published opinion, Stewart v. Trierweiler, 867 F.3d 633 (6th Cir. 2017). Specifically, in Stewart,

1See, e.g., Int’l Union v. Kelsey-Hayes Co., 872 F.3d 388, 390–93 (6th Cir. 2017) (Griffin, J., dissenting from the denial of the petition for rehearing en banc). No. 17-2267 Davenport v. MacLaren Page 13

Judge Sutton, writing for our unanimous court, ruled that the habeas petitioner had not sustained his burden of proving that a state court’s ruling of harmless error was an “unreasonable application of clearly established federal law as determined by the Supreme Court of the United States.” Id. at 636–38. Thus by operation of the Antiterrorism and Effective Death Penalty Act of 1996, 28 U.S.C. § 2254 (AEDPA), the petitioner in Stewart was not entitled to habeas relief. 867 F.3d at 638; see also Cullen v. Pinholster, 563 U.S. 170, 181 (2011) (“The petitioner carries the burden of proof” under § 2254(d)).

Opposite to Stewart, the Davenport majority held that respondent Warden had not sustained his burden of demonstrating that the error was not harmless. Davenport v. MacLaren, 964 F.3d 448, 451 (6th Cir. 2020) (“[T]he State has not met its burden to show the restraints did not have a substantial and injurious effect or influence in determining the jury’s verdict.”).

Moreover, contrary to Davis v. Ayala, 576 U.S. 257, 268 (2015) (“AEDPA nevertheless ‘sets forth a precondition to the grant of habeas relief’” (emphasis added and citation omitted)), and in conflict with Stewart, the Davenport panel held that AEDPA does not apply to issues of harmless error raised in habeas corpus petitions. 964 F.3d at 460–63. Davenport viewed the standards of Brecht v. Abrahamson, 507 U.S. 619 (1993), and AEDPA as the same. They are not. While a constitutional error resulting in actual prejudice is sufficient under Brecht, id. at 637, AEPDA requires more.

Specifically, AEDPA requires the petitioner to prove that the error is “an unreasonable application of [] clearly established Federal law, as determined by the Supreme Court of the United States.” § 2254(d)(1). This standard is materially different from proving actual prejudice. As Judge Readler emphasized in his Davenport dissent, while “[i]t may be that a federal court can deny habeas relief by ‘go[ing] straight to Brecht[,]’” 964 F.3d at 469 (Readler, J., dissenting) (citations omitted), a court may not grant a writ of habeas corpus unless the petitioner also sustains his burden under AEDPA of proving that the state court ruling was “an unreasonable application of[] clearly established Federal law, as determined by the Supreme Court of the United States.” § 2254(d)(1); see also Johnson v. Acevedo, 572 F.3d 398, 403–04 (7th Cir. 2009). No. 17-2267 Davenport v. MacLaren Page 14

In addition, because Stewart was an earlier published decision of our court, Stewart should have been precedentially binding for Davenport. 6 Cir. Rule 32.1(b); Miller v. Caudill, 936 F.3d 442, 447–48 (6th Cir. 2019). However, I also acknowledge that other decisions appear to conflict with Stewart, see, e.g., O’Neal v. Balcarcel, 933 F.3d 618, 625 (6th Cir. 2019); McCarley v. Kelly, 801 F.3d 652, 665 (6th Cir. 2015), thus compounding our intra-circuit conflict.

Second, the issue of harmless error for habeas corpus petitions is of exceptional importance, particularly for our court because we review habeas corpus death penalty petitions arising from three of the four states in our circuit. In this regard, the issues of which party has the burden of persuasion on the issue of harmless error for habeas review and whether AEDPA applies are fundamental to our judicial system.

I note that our sister courts of appeals are on the other side on these issues. It appears that we are alone in taking a “straight to Brecht approach” in determining on habeas review whether a state court’s constitutional error was harmless. See Davenport, 964 F.3d at 475–77 (Readler, J., dissenting) (collecting cases).

Finally, the Davenport holding that AEDPA is inapplicable to harmless error rulings by state courts trammels upon the important federalism principles that undergird AEDPA. See, e.g., Woodford v. Garceau, 538 U.S. 202, 206 (2003).

For these reasons, and for those articulated by Judge Readler in his persuasive panel dissent, I respectfully dissent. Because our litigants, attorneys, and judges need guidance from our en banc court on these issues of exceptional importance, I would grant respondent’s petition for rehearing en banc. No. 17-2267 Davenport v. MacLaren Page 15

_________________

DISSENT _________________

THAPAR, Circuit Judge, dissenting from the denial of rehearing en banc. Thirteen years ago, on a cold night in January, Earl Davenport killed Annette White. He closed his hand around her neck and held it there as she struggled against him. Minutes later, she was dead.

Despite the overwhelming evidence of Davenport’s guilt, a panel majority voted to vacate his conviction. It did so without even applying AEDPA deference to the state court’s harmless-error determination.

This tragic case thus presents a fundamental question of habeas jurisprudence: Must a state court’s harmless-error determination receive AEDPA deference under 28 U.S.C. § 2254(d)(1)? The plain text of the statute says that the answer is yes. But the panel majority held that the answer is no. According to the panel opinion, federal judges can simply ignore AEDPA’s guardrails whenever they find that a petitioner has suffered actual prejudice under Brecht v. Abrahamson, 507 U.S. 619 (1993). This holding casts aside AEDPA and misinterprets Supreme Court precedent. That matters because AEDPA’s procedural rules have bite that Brecht lacks. The holding also deepens an existing circuit split. And what’s more, the panel opinion defies Brecht itself, granting habeas relief based on mere speculation and a thin stack of academic articles, some of which postdate the state court’s decision.

Given these errors and their importance, this case merited the attention of the en banc court. I respectfully dissent.

I.

A.

At trial, Davenport did not contest that he killed Annette White. Instead, he claimed that he acted in self-defense. As Davenport tells it, they were both high on crack cocaine that night, and he was driving her around. They started arguing. He was trying to focus on traffic, and she pulled out a box cutter. At that point, he grabbed her by the neck and pinned her against the No. 17-2267 Davenport v. MacLaren Page 16

passenger-side window. He was choking her. She started kicking, and he pushed against her neck even harder. He says that even though the strangulation “happened quick[ly],” “it seemed like it took forever.”

It did. According to the medical examiner’s testimony, it would have taken at least four minutes for Annette to die. She would have passed out much sooner, perhaps after thirty to forty seconds without oxygen. So maybe at the start, the 6'5", 260-pound Davenport was up against an angry, argumentative, 5'2" Annette—all one hundred pounds of her. Maybe at the start, she had a box cutter. But soon he was up against an unconscious Annette: no more anger, no more knife.

Yet he kept his hand around Annette’s throat. She was unconscious. Seconds passed, then minutes passed, then she was dead. Davenport claims he was driving this entire time.

He then proceeded to dump Annette’s body in a field. After leaving her face down in the dirt—wearing a bra and bloodied underwear and one sock—he went to her house. He took some of her food, stole her stereo, and then met up with friends to smoke more crack. How did he feel about Annette’s death? “Just a bad situation,” he said. “That’s all. Just a bad situation.”

Davenport was familiar with bad situations. Annette wasn’t even the first person he had choked that week. Five days before, he found himself annoyed with another woman, so he came up behind her and circled his hands around her neck. He squeezed, lifting her off the ground. She bit her tongue and urinated on herself as she blacked out. By a stroke of luck, her boyfriend walked in. He saw his girlfriend unconscious, feet off the floor, dangling from Davenport’s hands. Davenport let her go, and she survived. “You know, you’re lucky,” he told her later. “I wanted to squish you like a bug.”

Strangling, it seems, is how Davenport dealt with problems. He would brag about it, flexing his hands: He didn’t have to worry about anyone giving him trouble; he’d just “choke ’em up” if it got bad. That’s what happened to Annette. And the jury concluded it was first-degree murder. No. 17-2267 Davenport v. MacLaren Page 17

B.

Davenport was partially shackled during trial. His right hand was uncuffed so that he could write notes to his attorney. And there was a privacy curtain around defense counsel’s table. Yet the trial court never made on-the-record findings justifying the restraints.

On appeal, the Michigan Supreme Court held that the shackling was unconstitutional. It remanded for a determination of whether the jury saw the shackles, and if so, whether the prosecution could show beyond a reasonable doubt that the error did not affect the verdict. People v. Davenport, 794 N.W.2d 616 (Mich. 2011) (order) (Davenport I).

The trial court then held a two-day evidentiary hearing. All twelve jurors testified. Five saw the restraints; four never noticed them; two heard about them from other jurors; and one couldn’t remember either way. The jurors testified that the restraints were never discussed during deliberations, and that to the extent they thought Davenport was dangerous, it was because he was on trial for first-degree murder. Thus, the trial court ruled that the prosecution had proved beyond a reasonable doubt that the shackling did not affect the verdict.

Davenport appealed again, and the Michigan Court of Appeals affirmed. It held that the trial court “did not err in finding that the prosecution proved beyond a reasonable doubt that the shackling error did not affect the verdict.” People v. Davenport, No. 306868, 2012 WL 6217134, at *3 (Mich. Ct. App. 2012) (per curiam) (Davenport II).

The Michigan Supreme Court denied leave to appeal. In denying leave to appeal, it said that Holbrook should have governed the lower court’s harmless-error analysis. People v. Davenport, 832 N.W.2d 389 (Mich. 2013) (Davenport III) (citing Holbrook v. Flynn, 475 U.S. 560, 570 (1986)). But the result would have been the same either way, the court reasoned, “given the substantial evidence of guilt presented at trial.” Id. Thus, the case did not merit its review.

Next, Davenport filed a habeas petition in federal court, which the district court denied. This court then granted a certificate of appealability, and a three-judge panel reversed over Judge Readler’s dissent. According to the majority, Davenport’s shackling amounted to actual No. 17-2267 Davenport v. MacLaren Page 18

prejudice under the Brecht standard. Davenport v. MacLaren, 964 F.3d 448, 468 (6th Cir. 2020) (Davenport IV). It thus vacated his conviction and ordered Michigan to release or retry Davenport within 180 days. Id. In throwing out Davenport’s conviction, the majority held that courts do not have to apply AEDPA deference once they make a finding of actual prejudice under Brecht. Id. at 457–58. But as Judge Readler’s forceful dissent points out, the majority’s error has real consequences—both for this case and for many cases to come.

II.

The majority’s opinion suffers from two fatal flaws. First, it refuses to apply AEDPA deference to the state court’s harmless-error determination, holding that the Brecht standard obviates any need for doing so. Second, it misapplies Brecht, granting habeas relief based on little more than speculation. And as a result of those two errors, the panel vacates the conviction of a Michigan murderer.

A.

The panel majority first erred in holding that AEDPA deference does not apply to state court harmless-error determinations. See Davenport IV, 964 F.3d at 456–57. In reaching this erroneous conclusion, the majority misconstrues AEDPA, misapplies Supreme Court precedent, and deepens an existing circuit split.

1.

To understand how the panel erred in its application of AEDPA, it is important to understand how a criminal defendant can challenge alleged constitutional violations. First, he must bring his challenge in state court. And if the state court finds a constitutional violation, it then analyzes whether that error was “harmless beyond a reasonable doubt.” Chapman v. California, 386 U.S. 18, 24 (1967).

If the state court finds that the error was harmless beyond a reasonable doubt, the petitioner can then seek habeas relief in federal court. But even before AEDPA, the Supreme Court recognized that habeas is an “extraordinary remedy” designed to protect those who have been “grievously wronged.” Brecht, 507 U.S. at 633–34 (citation omitted). Thus, the Supreme No. 17-2267 Davenport v. MacLaren Page 19

Court held that a habeas petitioner can only obtain relief when he demonstrates “actual prejudice.” Id. at 637 (citation omitted). Actual prejudice exists when a judge finds that an error had a “substantial and injurious effect or influence in determining the jury’s verdict.” Id.

But that wasn’t enough for Congress. So three years after Brecht, Congress passed and the President signed into law the Antiterrorism and Effective Death Penalty Act of 1996. Under AEDPA, federal courts can grant habeas relief only when a state court’s adjudication of a federal claim (1) “resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court,” or (2) “was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding.” 28 U.S.C. § 2254(d). Put differently, if “fairminded jurists could disagree on the correctness of the state court’s decision,” federal courts must deny habeas relief. Harrington v. Richter, 562 U.S. 86, 101 (2011) (cleaned up). And that remains true even if a court also finds actual prejudice under Brecht.

2.

This case presents the question whether a state court’s harmless-error determination must always receive AEDPA deference in accordance with § 2254(d)(1). Under the plain text of AEDPA, the answer is clear—yes.

Habeas requires us to review the final state court decision. Harrington, 562 U.S. at 98– 99. When that decision rests on the harmlessness of an alleged error (as is the case here), federal courts must determine whether the state court’s harmless-error determination “was contrary to, or involved an unreasonable application of, clearly established Federal law.” 28 U.S.C. § 2254(d)(1). That is because federal law—namely Chapman—governs the harmless-error inquiry in state court. Thus, AEDPA’s text requires deference to a state court’s harmless-error inquiry.

Supreme Court precedent confirms this common-sense reading of the statute. In Davis v. Ayala, the Court explained that independent of Brecht, “the limitation on federal habeas relief that § 2254(d) plainly sets out” continues to apply. 576 U.S. 257, 268 (2015). Indeed, the Supreme Court has squarely held that “habeas relief is appropriate only if [a state court has] No. 17-2267 Davenport v. MacLaren Page 20

applied harmless-error review in an ‘objectively unreasonable’ manner.” Mitchell v. Esparza, 540 U.S. 12, 18 (2003) (per curiam) (citation omitted). The panel majority ignores that rule here. Thus, not only does the panel opinion conflict with the plain text of AEDPA, it also violates controlling Supreme Court precedent.

3.

How does the panel majority avoid AEDPA’s limits? By first misapplying Supreme Court precedent and then clouding the important differences between Brecht and AEDPA review.

Start with precedent. The panel’s confusion begins with Fry v. Pliler, 551 U.S. 112 (2007). That case concerned what standard federal courts should apply on collateral review when a state court finds no constitutional error and thus doesn’t make a harmless-error determination: Brecht or Chapman? See id. at 120 & n.2. The Supreme Court held that even when a state court makes no harmless-error determination, federal courts must still apply the Brecht test rather than Chapman’s lower standard. Id. at 118–19. In this context, the Court reasoned, Brecht “obviously subsumes” the AEDPA/Chapman test because it requires a higher showing for prejudice, so it “makes no sense to require formal application of both tests.” Id. at 120. Simply put, Fry stands for the proposition that there is nothing to defer to under AEDPA when a state court does not make a harmless-error determination. Id. at 119–20; see Johnson v. Acevedo, 572 F.3d 398, 404 (7th Cir. 2009) (Easterbrook, J.) (adopting this reading of Fry).

The panel takes Fry’s “obviously subsumes” dictum from its case-specific context and insists that Brecht can always operate as a complete substitute for AEDPA deference. See Davenport IV, 964 F.3d at 458–59. But this misreads Fry. Indeed, a closer look at Fry reveals that—far from suggesting that AEDPA’s guardrails are surplusage—the Supreme Court confirmed that courts must ordinarily apply AEDPA alongside Brecht: “[W]hen a state court determines that a constitutional violation is harmless, a federal court may not award habeas relief under § 2254 unless the harmlessness determination itself was unreasonable.” Fry, 551 U.S. at 119 (citing Mitchell, 540 U.S. 12). No. 17-2267 Davenport v. MacLaren Page 21

If any doubt remains, the Supreme Court’s subsequent decision in Ayala should quash it. There the Court explained that “[t]he Fry Court did not hold—and would have had no possible basis for holding—that Brecht somehow abrogates the limitation on federal habeas relief that § 2254(d) plainly sets out.” Ayala, 576 U.S. at 268. So while there will be cases in which “a federal habeas court need not formally apply both Brecht and AEDPA/Chapman, AEDPA nevertheless sets forth a precondition to the grant of habeas relief.” Id. (cleaned up).

For all of these reasons, state court harmless-error determinations are entitled to AEDPA deference under § 2254(d).

4.

The panel majority contends that its non-application of AEDPA makes sense because Brecht imposes a higher bar than Chapman for granting habeas relief. See Davenport IV, 964 F.3d at 458 & n.6. And since Brecht imposes a higher bar than Chapman, the panel reasons that we can safely ignore AEDPA’s statutory guardrails. But this approach makes little sense from a first-principles perspective. Why? Because Brecht and AEDPA call for distinct inquiries.

Brecht asks whether a criminal defendant was actually prejudiced by a constitutional error. 507 U.S. at 637. AEDPA, on the other hand, asks whether a state court decision was unreasonable based on two factors: (1) clearly established federal law, and (2) evidence presented in state court proceedings. 28 U.S.C. § 2254(d). So while Brecht might impose a stricter substantive standard for relief than Chapman standing alone, Brecht and AEDPA ask different questions and are governed by different procedural rules. It thus makes little sense to say that a finding of actual prejudice under Brecht obviates any need to apply AEDPA’s clear rules.

To be sure, if a habeas petitioner has not suffered actual prejudice under Brecht, a state court’s determination that an error was harmless will always be reasonable under § 2254(d). Thus, a court must deny habeas outright when Brecht is not satisfied. See, e.g., Hollman v. Sprader, 803 F. App’x 841, 843 (6th Cir. 2020). No. 17-2267 Davenport v. MacLaren Page 22

But the inverse is not also true: A showing of actual prejudice under Brecht does not mean that a petitioner has also satisfied AEDPA’s requirements. Taking a closer look at the core differences between Brecht and AEDPA shows why, even if one accepts the panel’s Brecht analysis, AEDPA would require that we still affirm the conviction.

First, under the Brecht standard, a petitioner isn’t limited to clearly established Supreme Court precedent. A petitioner can thus argue for an expansion of Supreme Court precedent when attempting to show Brecht prejudice (at least absent AEDPA). But under AEDPA, extending Supreme Court precedent is improper. See White v. Woodall, 572 U.S. 415, 425–26 (2014); 28 U.S.C. § 2254(d)(1).

This case shows why this rule matters. The panel majority rejected Michigan’s harmless-error determination in part because the state court relied on post-trial testimony from jurors. Davenport IV, 964 F.3d at 466; id. at 478 (Readler, J., dissenting). The majority contends that Holbrook bars Michigan’s approach. 475 U.S. 560. But Holbrook concerned pretrial testimony from prospective jurors during voir dire about what effect uniformed law enforcement officers might have on their perception of the defendants. Id. at 565. That is a far cry from post-trial testimony from jurors about how shackling actually affected their verdict. Thus, the majority had to extend Holbrook to apply it here. Davenport IV, 964 F.3d at 479 (Readler, J., dissenting). And while such extensions might be permissible under Brecht itself, they are strictly prohibited by AEDPA. Woodall, 572 U.S. at 425–26; see also Virginia v. LeBlanc, 137 S. Ct. 1726, 1728–29 (2017) (per curiam).

Second, under Brecht, a petitioner arguably could rely upon circuit precedent to show how Supreme Court cases should be applied. But as the Supreme Court has repeatedly reminded us in the AEDPA context, we may not rely upon circuit precedent to show what a general Supreme Court standard clearly establishes. The Supreme Court’s warning speaks for itself: “As we explained in correcting an identical error by the Sixth Circuit two Terms ago, circuit precedent does not constitute ‘clearly established Federal law, as determined by the Supreme Court.’ It therefore cannot form the basis for habeas relief under AEDPA. Nor can the Sixth Circuit’s reliance on its own precedents be defended in this case on the ground that they merely No. 17-2267 Davenport v. MacLaren Page 23

reflect what has been clearly established by our cases.” Parker v. Matthews, 567 U.S. 37, 48–49 (2012) (cleaned up).

Again, this case shows why AEDPA’s constraints matter. The majority relies on Sixth Circuit and Ninth Circuit cases to extend Supreme Court precedent in the shackling (or, as here, partial-shackling) context. See Davenport IV, 964 F.3d at 465–68 (citing Ruimveld v. Birkett, 404 F.3d 1006, 1018 (6th Cir. 2005); Rhoden v. Rowland, 172 F.3d 633, 637 (9th Cir. 1999)). Based on that caselaw, the majority assumes that a “presumption of prejudice” attaches any time a juror sees a shackled defendant, and that the “duration of the jury’s deliberations” matter. See id. Yet Supreme Court caselaw does not require state courts to consider these factors when conducting a harmless-error analysis. So under AEDPA, the state court cannot be faulted for unreasonably applying “clearly established Federal law, as determined by the Supreme Court.” 28 U.S.C. § 2254(d)(1).

Third, under the AEDPA standard, state courts have broad discretion when Supreme Court precedent speaks at high levels of generality. Renico v. Lett, 559 U.S. 766, 776 (2010) (“Because AEDPA authorizes federal courts to grant relief only when state courts act unreasonably, it follows that the more general the rule at issue—and thus the greater the potential for reasoned disagreement among fair-minded judges—the more leeway state courts have in reaching outcomes in case-by-case determinations.” (cleaned up)). In the shackling context, the Supreme Court has told us generally that Chapman applies. But it has not told us specifically how state courts must conduct their Chapman inquiries. Compare Deck v. Missouri, 544 U.S. 622, 635 (2005) (requiring general harmless-error analysis in the shackling context), with Delaware v. Van Arsdall, 475 U.S. 673, 684 (1986) (identifying more specific harmless-error factors in the Confrontation Clause context). Thus, state courts may adopt any reasonable approach when determining whether shackling the defendant was harmless.

Once more, this case highlights why AEDPA has bite. To determine whether the error was harmless, the Michigan Supreme Court directed the trial court to hold a hearing concerning how the shackling had (or had not) affected the verdict. Davenport I, 794 N.W.2d 616. The trial court then required the jurors to testify at the evidentiary hearing. That legal approach—relying on the jurors’ own post-trial views—is certainly a reasonable application of Chapman’s general No. 17-2267 Davenport v. MacLaren Page 24

standard. And if it’s an acceptable approach, there can be no doubt as to the proper outcome in this case under AEDPA. Then, when denying discretionary review of the trial court’s harmless- error determination, the Michigan Supreme Court suggested that the trial court should have focused on the overall evidence of the defendant’s guilt rather than the jurors’ testimony. (It denied review because the evidence of guilt was overwhelming.) Davenport III, 832 N.W.2d 389. Given Chapman’s general standard, this alternative approach would have also been a reasonable approach to harmless-error analysis in the shackling context.

But the majority, in determining Brecht prejudice, adopted its own harmless-error test— asking whether a jury might have convicted the defendant of a lesser included offense (second- degree murder) instead of his actual crime of conviction (first-degree murder). The majority identifies no clearly established Supreme Court precedent requiring this approach. Indeed, Davenport never made this lesser-included-offense argument to the state court. So the majority’s approach, at least on collateral review, is out of bounds under AEDPA. Thus, even if one were to accept the majority’s Brecht analysis, AEDPA would again require a different result.

Fourth, the Brecht standard—at least on its face—does not seem to bar habeas petitioners from presenting evidence outside the state court record (although federalism and comity counsel against it). But under AEDPA, habeas review is strictly “limited to the record that was before the state court that adjudicated the prisoner’s claim on the merits.” Greene v. Fisher, 565 U.S. 34, 38 (2011) (citing Cullen v. Pinholster, 563 U.S. 170, 182 (2011)). In this case, for example, the panel majority relies on social-science studies about “implicit bias” (published as recently as 2019) to support its finding of actual prejudice. See Davenport IV, 964 F.3d at 466 & n.13; id. at 479 (Readler, J., dissenting). But that new evidence is outside the state court record and thus cannot be considered under AEDPA.

The upshot of all this is that by shifting everything into the Brecht basket, the panel dodges AEDPA’s clear rules—no looking at non-Supreme Court cases, no extending existing Supreme Court precedent, and no relying on evidence not presented in state court. Jettisoning these clear, rule-based requirements will make appellate review in habeas cases more difficult and unpredictable. See generally Antonin Scalia, The Rule of Law as a Law of Rules, 56 U. Chi. L. Rev. 1175 (1989). And by affording federal courts the ability to undo state convictions absent No. 17-2267 Davenport v. MacLaren Page 25

a violation of existing Supreme Court precedent, the panel disrespects the comity and federalism concerns at the heart of AEDPA.

In conclusion, the differences between Brecht and AEDPA matter. If AEDPA applies, the panel decision is plainly erroneous since it extends Supreme Court precedent, relies on circuit precedent, creates a new standard for harmless-error review in the shackling context, and introduces evidence not presented in the state court proceedings. See Davenport IV, 964 F.3d at 479 (Readler, J., dissenting). Thus, this case presents an ideal vehicle for clarifying the relationship between Brecht and AEDPA.

5.

This case also highlights a deepening split among the various federal Courts of Appeals. See Davenport IV, 964 F.3d at 475–77 (Readler, J., dissenting) (discussing this circuit split).

One set of circuits generally holds that courts must apply both Brecht and AEDPA when reviewing state court harmless-error determinations. See, e.g., Johnson v. Lamas, 850 F.3d 119, 133–34 (3d Cir. 2017); Johnson v. Acevedo, 572 F.3d 398, 404 (7th Cir. 2009); Malone v. Carpenter, 911 F.3d 1022, 1029–30 (10th Cir. 2018); Mansfield v. Sec’y, Dep’t of Corr., 679 F.3d 1301, 1307–08 (11th Cir. 2012).

Typical of this approach is the Seventh Circuit’s opinion in Acevedo. There, Judge Easterbrook explained that Fry controls where a state court fails to conduct a harmless-error analysis. Acevedo, 572 F.3d at 404. But when a state court does reach harmless error, “the federal court must decide whether that analysis was a reasonable application of the Chapman standard.” Id. And only after a finding of unreasonableness does AEDPA drop out of the picture and allow a federal court to “make an independent decision” about prejudice under Brecht. Id.

Another set of circuits arguably holds that a finding of actual prejudice under Brecht obviates the need to apply AEDPA deference to state court harmless-error determinations. While the panel’s holding charges to the front of the line, at least three other circuits seem to No. 17-2267 Davenport v. MacLaren Page 26

have taken the same approach. See Connolly v. Roden, 752 F.3d 505, 511 (1st Cir. 2014); Wood v. Ercole, 644 F.3d 83, 94 (2d Cir. 2011); Deck v. Jenkins, 814 F.3d 954, 985 (9th Cir. 2016).

Of course, neatly dividing the circuits into categories at times proves too much. For the question presented here has led to both “intracircuit” and “intercircuit” splits. See Deck, 814 F.3d at 973 (Bea, J., dissenting from denial of reh’g en banc). Indeed, my review of the caselaw reveals subtle (and not so subtle) tension within many circuits. Compare Vazquez v. Sec’y, Fla. Dep’t of Corr., 807 F. App’x 901, 905 (11th Cir. 2020) (“[F]ederal courts need not formally apply both the Brecht standard and AEDPA-deference . . . though ultimately both tests must be met for courts to grant habeas relief.” (citing Mansfield, 679 F.3d at 1307–08)), with Hammonds v. Comm’r, Ala. Dep’t of Corr., 712 F. App’x 841, 849–50 (11th Cir. 2017) (“[I]f a petitioner satisfies the Brecht standard, he necessarily also satisfies the AEDPA standard.”); compare also Lamas, 850 F.3d at 133–34, with Bond v. Beard, 539 F.3d 256, 275–76 (3d Cir. 2008). And as Judge Griffin’s dissent aptly points out, our circuit’s caselaw reflects the same tension. See ante at 12–13 (Griffin, J., dissenting from denial of reh’g en banc).

Given the deep confusion within and among the circuits, the question presented here is ripe for further review.

B.

Even if we cast aside AEDPA and its congressionally mandated guardrails, Davenport is still not entitled to habeas relief. He cannot show that the shackling resulted in “actual prejudice.” Brecht, 507 U.S. at 637 (citation omitted). To do so, the shackling must have had a “substantial and injurious effect or influence [on] the jury’s verdict.” Id.

And here, all twelve jurors testified that the shackling “did not influence the[ir] verdict.” Davenport II, 2012 WL 6217134, at *1; see Smith v. Phillips, 455 U.S. 209, 217 n.7 (1982) (noting that a juror is “well qualified to say whether he has an unbiased mind in a certain matter”). While the Supreme Court has not told us how to undertake the harmless-error analysis in this context, the jurors’ testimony seems important. What’s more, eleven Michigan judges concluded the error was harmless. Davenport II, 2012 WL 6217134, at *3, leave to appeal denied, 832 N.W.2d 389, 390; see also Brecht, 507 U.S. at 636 (“[S]tate courts often occupy a No. 17-2267 Davenport v. MacLaren Page 27

superior vantage point from which to evaluate the effect of trial error.”). And just in case anyone thinks that the Michigan judiciary and jurors were out to lunch (they weren’t), a federal magistrate judge, a district court judge, Judge Readler, and several members of this en banc court agreed with their conclusion.

But even if we accept Davenport’s asserted harm, he did not suffer actual prejudice. Davenport strangled Annette’s limp body after she lost consciousness, dumped her half-naked in an empty field, and then robbed her house. Add to this that another woman testified that Davenport had choked her earlier in the week. And if that isn’t enough, the jury heard about his frequent boasts about choking anyone who gave him trouble. Plus, the prosecution repeatedly impeached Davenport during trial. Simply put, the evidence of guilt was overwhelming. The evidence that the shackles did not influence the jury is uncontroverted—not one juror said otherwise. And the evidence that Davenport should go free is non-existent.

So what did two federal appellate judges see that everyone before them missed? The panel’s conclusion largely rests on speculation about how the shackles must have biased the jurors. See Davenport IV, 964 F.3d at 464–68. But the record does not support that speculation. And while the panel also cites a smattering of academic articles about jury bias, those articles hardly establish that the jurors were biased in this case. Id. at 466 & n.13; see also id. at 479 (Readler, J., dissenting). Rather than heed the Supreme Court’s teaching that trial courts typically have “a superior vantage point from which to evaluate the effect of trial error,” the majority casts aside the prior findings in favor of its own guesswork. Brecht, 507 U.S. at 636. That is error. And given the grave consequences, it is error that should not go unchecked.

***

In sum, the panel erred by holding that a finding of actual prejudice under Brecht obviates the need to consider AEDPA’s guardrails. AEDPA and Brecht ask different questions that sometimes demand different answers. Thus, before a federal court may grant habeas relief, it must find both that a state court’s harmless-error determination violated AEDPA and that actual prejudice resulted under Brecht. To make matters worse, the panel opinion badly No. 17-2267 Davenport v. MacLaren Page 28

misapplies Brecht on its own terms, relying on mere speculation to vacate the conviction of a man who strangled a woman in cold blood.

Because this case warrants further review, I respectfully dissent.

ENTERED BY ORDER OF THE COURT

___________________________________ Deborah S. Hunt, Clerk