We must decide whether the showing necessary for an employee to prevail against his employer’s motion for summary judgment in this employment discrimination case is heightened because the person who demoted him had previously appointed and promoted him; if so, we must decide whether the employee’s evidence of discrimination is sufficient to meet the heightened burden.

I

James Coghlan is a resident of Washington and a commercial fisherman. From 1997 onward, he was employed by American Seafoods Company LLC (ASC), which operates fishing vessels off the coast of the Pacific Northwest and in Alaskan waters. Until the American Fisheries Act (AFA) was passed in 1998, ASC was owned and operated by a Norwegian parent corporation. The AFA required certain fishing companies, including those engaged in the Alaska pollock and cod fisheries, to be American-owned. ASC complied with the law, but its management remains largely the same as it was before the AFA was passed and is made up primarily of native Norwegians.

In 1997 Coghlan was working as mas-ter of the Victoria Ann when ASC purchased that vessel. He continued to serve as master of the Victoria Ann until 1998, when ASC took the Victoria Ann out of service because provisions of the AFA effectively required it to reduce its active fleet. Many masters, mates, and other crewmembers were laid off, but ASC retained Coghlan and appointed him master of the Katie Ann, one of its six factory trawlers still in service. The person responsible for the decision to retain Cogh-lan was Inge Andreassen, ASC’s Vice President of Operations and a man of Norwegian birth. According to Andreassen’s declaration, he selected Coghlan for the job despite the availability of at least one Norwegian candidate, Tor Storkersen.

Coghlan continued to serve as the master of the Katie Ann until 2000, when ASC decided to place another vessel, the American Dynasty, back into operation. At that time ASC appointed Coghlan as mate of the Dynasty; the ship’s master was Kri-stjan Petursson, who was born in Iceland. Again, Andreassen was responsible for the decision to transfer Coghlan. Although the transfer technically involved a step down in rank, from master to mate, Cogh-lan’s new position provided an opportunity to make more money and Coghlan saw it as a desirable change. Coghlan remained mate of the Dynasty until November 2001.

On two occasions in September and October 2001, Petursson had to be temporarily absent as master of the Dynasty. On each occasion, instead of appointing Cogh-lan as the “relief master” (i.e., temporary master) as Coghlan would have liked, An-dreassen selected the Norwegian-born Jarl Hogseth to fill the position. Coghlan considered himself more qualified than Hog-seth, especially since he had been serving on the Dynasty for more than a year and knew the vessel. Andreassen stated in his declaration that he made his decision on the basis of a recommendation from Frank Vargas, ASC’s Fleet Operations Manager and a native-born American of Filipino ancestry.

In November 2001, Andreassen was dissatisfied with the Dynasty’s performance. Its production levels were low and its expenses for equipment replacement were high for a boat of its size. Michael Hyde, the president of ASC, stated in his declaration that he instructed Andreassen to change the Dynasty’s leadership and to allow neither Petursson nor Coghlan to serve as its master. After consulting with Frank Vargas as well as Tammy French, ASC’s Vice President of Human Resources and an American of non-Nordic heritage, Andreassen removed Coghlan from the vessel and demoted Petursson to the position of mate. Andreassen stated in his declaration that Vargas’s recommendation carried special weight because he is in day-to-day contact with the ships and had previously served as master of the Dynasty himself. Andreassen offered the master position to an American of non-Norwegian descent, Mike Kraljevich, who was then serving as a mate on another ASC vessel. Kraljevich declined, however, and An-dreassen instead appointed Ole Knotten, a man of Norwegian descent. Knotten had little experience fishing in American waters and only obtained a Coast Guard license shortly before he was to take over as master of the Dynasty, but he had been working on fishing vessels in Russia for more than ten years, including a stint as fishmaster on what Andreassen described as “the most sophisticated factory trawler ever built.” Around the same time in late 2001, Andreassen removed the masters on two other ASC vessels, both of whom were American. Both of their replacements were Norwegian-born men.

With the 2002 fishing season approaching and Coghlan having been removed as mate of the Dynasty, Andreassen offered Coghlan the position of mate on the Katie Ann. (According to the declarations of An-dreassen and Vargas, they first considered Coghlan for the master position, but were unable to reach him and got the feeling that he was avoiding them. Coghlan disputes this.) Coghlan found such offer objectionable, considering that he had previously served as master of the Katie Ann and felt that he should be reappointed to that position. Instead, Andreassen appointed Jarl Hogseth as master. Coghlan declined the offer of the mate position and brought this lawsuit in the Western District of Washington.

Coghlan alleged national-origin discrimination under Title VTI and the Washington Law Against Discrimination. ASC moved for summary judgment supported by declarations and depositions, arguing that the adverse employment actions regarding Coghlan were motivated by legitimate, nondiscriminatory reasons, namely, the poor performance of the Dynasty in 2001 and observed problems in Coghlan’s employment history. The district court granted ASC’s motion, and this appeal timely followed.

II

We analyze Coghlan’s disparate-treatment claim of employment discrimination under the burden-shifting framework outlined by the Supreme Court in McDonnell Douglas Corp. v. Green, 411 U.S. 792 , 93 S.Ct. 1817 , 36 L.Ed.2d 668 (1973). Under that framework, the burden of production first falls on the plaintiff to make out a prima facie case of discrimination. He may do so by showing that (1) he belongs to a protected class, (2) he was qualified for the position he held (or for the position to which he wished to be promoted), (3) he was terminated or demoted from (or denied a promotion to) that position, and (4) the job went to someone outside the protected class. St. Mary’s Honor Center v. Hicks, 509 U.S. 502, 506 , 113 S.Ct. 2742 , 125 L.Ed.2d 407 (1993); McGinest v. GTE Serv. Corp., 360 F.3d 1103, 1122 (9th Cir.2004). The burden of production then shifts to the employer, who must present evidence sufficient to permit the factfinder to conclude that the employer had a legitimate, nondiscriminatory reason for the adverse employment action. St. Mary’s Honor Center, 509 U.S. at 506-07 , 113 S.Ct. 2742 . Finally, if the employer meets that burden, then the McDonnell Douglas framework drops out of the picture entirely, and the plaintiff bears the full burden of persuading the factfinder that the employer intentionally discriminated against him. Id. at 507-08 , 113 S.Ct. 2742 .

We proceed to apply the McDonnell Douglas framework to the evidence in this case.

A

ASC argues — though only in a footnote — that Coghlan failed to make out a prima facie case of discrimination. ASC admits that Coghlan belonged to a protected class (non-Norwegian-born workers); that he was twice not appointed as relief master of the Dynasty, that he was removed as mate of the Dynasty, and that he was not appointed master of the Katie Ann; and that the people chosen instead were Norwegian-born and thus outside the protected class. It argues, however, that he was not qualified for those positions because he was not performing at a level consistent with ASC’s expectations. This argument is not convincing. We have emphasized that “[t]he requisite degree of proof necessary to establish a prima facie case for Title VII and ADEA claims on summary judgment is minimal and does not even need to rise to the level of a preponderance of the evidence.” Wallis v. J.R. Simplot Co., 26 F.3d 885, 889 (9th Cir.1994); see also Aragon v. Republic Silver State Disposal, Inc., 292 F.3d 654, 659-60 (9th Cir.2002) (emphasizing the low threshold for a prima facie case and holding that even an employee’s self-assessment is relevant evidence). Coghlan has presented enough evidence to meet this minimal burden: most notably, he had previously served as master of two of ASC’s vessels, the Victoria Ann and the Katie Ann, and was offered the important position of mate on the Dynasty, suggesting that he was not incompetent to handle major duties on a relatively large ship.

As for the second step of the McDonnell Douglas framework, Coghlan does not dispute that ASC’s articulation of nondiscriminatory reasons for its actions was sufficient to cause the framework to drop away and to place the burden back on Coghlan to show that ASC’s explanations were actually a pretext for discrimination.

A plaintiff may meet the burden to show pretext using either direct or cir- eumstantial evidence. Direct evidence is evidence “which, if believed, proves -the fact[of discriminatory animus] without inference or presumption.” Godwin v. Hunt Wesson, Inc., 150 F.3d 1217, 1221 (9th Cir.1998) (quoting Davis v. Chevron, U.S.A., Inc., 14 F.3d 1082, 1085 (5th Cir.1994)) (alteration in original). Direct evidence typically consists of clearly sexist, racist, or similarly discriminatory statements or actions by the employer. See, e.g., Godwin, 150 F.3d at 1221 (supervisor stated he “did not want to deal with [a] female”); Cordova v. State Farm Ins., 124 F.3d 1145 , 1149 (9th Cir.1997).

Circumstantial evidence, in contrast, is evidence that requires an additional inferential step to demonstrate discrimination. It can take two forms. First, the plaintiff can make an affirmative case that the employer is biased. For example, statistical evidence is circumstantial evidence that could, if sufficiently probative, point to bias. See Aragon, 292 F.3d at 663 . Second, the plaintiff can make his case negatively, by showing that. the employer’s proffered explanation for the adverse action is “unworthy of credence.” Texas Dep’t of Community Affairs v. Burdine, 450 U.S. 248, 256 , 101 S.Ct. 1089 , 67 L.Ed.2d 207 (1981). As the Supreme Court has explained:

Reeves v. Sanderson Plumbing Products, Inc., 530 U.S. 133, 147 , 120 S.Ct. 2097 , 147 L.Ed.2d 105 (2000). The distinction between direct and circumstantial evidence is crucial, because it controls the amount of evidence that the plaintiff must present in order to defeat the employer’s motion for summary judgment. Because direct evidence is so probative, the plaintiff need offer “very little” direct evidence to raise a genuine issue of material fact. Godwin, 150 F.3d at 1221 . But when the plaintiff relies on circumstantial evidence, that evidence must be “specific and substantial” to defeat the employer’s motion for summary judgment. Id. at 1222 (internal quotation marks removed); see also Aragon, 292 F.3d at 661 .

Coghlan does not offer any direct evidence of ASC’s discriminatory intent. Because his case is entirely circumstantial, he would have to present “specific and substantial” evidence of intentional discrimination to defeat ASC’s motion for summary judgment. See Aragon, 292 F.3d at 661 . His burden is especially steep in this case because of the so-called “same actor inference,” to which we now turn.

B

In Bradley v. Harcourt, Brace & Co., 104 F.3d 267 (9th Cir.1996), we held that “where the same actor is responsible for both the hiring and the firing of a discrimination plaintiff, and both actions occur within a short period of time, a strong inference arises that there was no discriminatory action.” Id. at 270-71 . That holding is relevant here because Inge An-dreassen, the man who made all of the challenged employment decisions, was the same man who appointed Coghlan as master of the Katie Ann in 1998 (over at least one viable candidate of Norwegian descent, according to Andreassen’s declaration). He was also the same man who selected Coghlan for the position of mate/fishmate on the Dynasty in 2000, an appointment that Coghlan desired and viewed as a change for the better.

We based our holding in Bradley on the principle that an employer’s initial willingness to hire the employee-plaintiff is strong evidence that the employer is not biased against the protected class to which the employee belongs. Id. Thus, although we phrased the same-actor rule in Bradley in terms of “hiring and ... firing,” its logic applies no less to cases such as this one, in which the plaintiff was not actually fired but merely offered a less desirable job assignment. See Hartsel v. Keys, 87 F.3d 795 , 804 n. 9 (6th Cir.1996) (applying the same-actor inference where the decision-maker had not hired the plaintiff but had previously promoted her).

Coghlan offers several reasons why the same-actor inference should not apply, but none is convincing. First, he suggests that application of the same-actor inference in the summary judgment con text is inconsistent with the holding of the Supreme Court in Reeves . The Supreme Court held in that case that “a plaintiffs prima facie case, combined with sufficient evidence to find that the employer’s asserted justification is false, may permit the trier of fact to conclude that the employer unlawfully discriminated.” Reeves, 530 U.S. at 148 , 120 S.Ct. 2097 . In other words: in many cases where the evidence is sufficient for a rational trier of fact to conclude that the employer is lying about its reason for firing or demoting the plaintiff, summary judgment will be inappropriate on that basis alone because a jury could reasonably view the employer’s lie as evidence of its guilt. That holding has no bearing on the same-actor inference, however, because the point of the same-actor inference is that the evidence rarely is “sufficient ... to find that the employer’s asserted justification is false” when the actor who allegedly discriminated against the plaintiff had previously shown a willingness to treat the plaintiff favorably. Reeves , then, tells us only that if a plaintiff can muster the extraordinarily strong showing of discrimination necessary to defeat the same-actor inference, then the case likely must go to the jury.

Coghlan also argues that the same-actor inference is not relevant here because three years elapsed between 1998, when Andreassen appointed Coghlan master of the Katie Ann, and 2001, when the earliest of the allegedly discriminatory decisions occurred. Bradley did limit its holding to cases where the alleged discrimination took place “within a short period of time” after the favorable action. Bradley, 104 F.3d at 270-71 . We reject Coghlan’s argument, however, for several reasons. First, this length of time would be significant only had Coghlan proffered evidence suggesting that Andreassen developed a bias against non-Norwegians during that period; but he did not. Cf. Coleman v. Quaker Oats Co., 232 F.3d 1271, 1286 (9th Cir.2000) (taking the same-actor inference into account when the positive action occurred more than a year earlier than the negative action); Schnabel v. Abramson, 232 F.3d 83, 91 (2d Cir.2000) (basing affirmance of summary judgment in an employment discrimination case in part on the fact that the plaintiff “was fired by the same man who had hired him three years earlier”).

Moreover, Andreassen did in fact take favorable action toward Coghlan in 2000, only a year before the earliest adverse employment decision, when he appointed Coghlan mate rather than master of the Dynasty. It is undisputed that Coghlan himself viewed the appointment as a favorable employment action; the new job was on a much larger ship and paid significantly more than the old job, and Coghlan was pleased with the change. As Coghlan points out, however, Andreassen declined to characterize the move as a “promotion”; he preferred to call it a “transfer,” and in a literal sense, it was not a promotion because Coghlan went from the rank of master on the Katie Ann to the lower rank of mate on the Dynasty. Coghlan argues that it is the decisionmaker’s perception, not that of the employee, that controls whether the same-actor inference arises.

As an abstract proposition, this is doubtless true; for if the decisionmaker did not perceive an employment action as favorable, there would be no basis to assume an absence of bias toward the employee. But the fact that we must look to the decision-maker’s perception does not mean that we are bound by the decisionmaker’s label. The question is simply whether the nature of the employment action, viewed from the employer’s perspective, is such that it would have been unlikely if the decision-maker were truly biased against the employee’s class. If it is, then the inference fairly arises. In this case, whether or not Coghlan’s appointment was classified as a “promotion” in ASC’s internal discussion, it is clear that Andreassen intentionally chose to appoint Coghlan to a new, better-paid, more demanding position on a larger ship. The favorable nature of the reassignment satisfies us that the same-actor inference should arise.

Finally, Coghlan suggests that the same-actor inference is just one more factor for the jury to consider in making its decision and should not be used to grant summary judgment to the defendant. That is the law in some circuits, and Cogh-lan cites eases to prove it. See, e.g., Williams v. Vitro Servs. Corp., 144 F.3d 1438 , 1443 (11th Cir.1998) (“[I]t is the province of the jury rather than the court ... to determine whether the inference generated by ‘same actor’ evidence is strong enough to outweigh a plaintiffs evidence of pretext.”). But it is clearly not the law in this circuit, since Bradley itself used the same-actor inference to affirm a grant of summary judgment, taking the case away from a jury. See Bradley, 104 F.3d at 272 ; cf. Brown v. CSC Logic, Inc., 82 F.3d 651, 658 (5th Cir.1996) (affirming grant of summary judgment on the basis of the same-actor inference); Lowe v. J.B. Hunt Transp., Inc., 963 F.2d 173, 173-74 (8th Cir.1992) (affirming grant of directed verdict on the basis of the same-actor inference).

The same-actor inference is neither a mandatory presumption (on the one hand) nor a mere possible conclusion for the jury to draw (on the other). Rather, it is a “strong inference” that a court must take into account on a summary judgment motion. Bradley, 104 F.3d at 271 . We must consider, then, whether Coghlan has made out the strong case of bias necessary to overcome this inference. It will be useful to consider separately each of the incidents that Coghlan alleges constituted illegal discrimination.

C

1

In 2001, while Coghlan was serving as mate on the Dynasty, the ship’s master (Petursson) had to be temporarily absent on two occasions. Each time, An-dreassen appointed a Norwegian (Hog-seth) as the relief master (that is, the temporary replacement for the absent master) instead of Coghlan. Coghlan argues that he was more qualified than Hug seth because he had more experience both with the Dynasty and as a factory-trawler master in general. Of course, the quality of Andreassen’s business judgment is only relevant insofar as it suggests that his decisions were explainable only as the product of illegal discrimination. See Villiarimo v. Aloha Island Air, Inc., 281 F.3d 1054, 1063 (9th Cir.2002). But in this case there is another, much more plausible, reason for Andreassen’s decision: Andreassen stated in his declaration that his decision to appoint Hogseth as relief master was based entirely on the recommendation of Frank Vargas, who, in turn, stated in his declaration that he “strongly recommended Jarl Hogseth to serve as Relief Captain” and disfavored Coghlan because Vargas believed he had poor leadership skills. Coghlan does not offer any evidence that Andreassen’s decision was not based on Vargas’s recommendation, and Coghlan testified that he does not believe that Vargas harbors any sort of national-origin bias against him. Thus, Coghlan has presented no evidence that would cast doubt on this legitimate explanation of the decision to appoint Hogseth as relief master, and so he cannot rebut the “strong inference” of nondiscrimination that arises under the same-actor rule.

2

In November 2001, Andreassen removed Coghlan as mate of Dynasty. At the same time, he demoted Petursson to the mate position and appointed a Norwegian-born person (Knotten) as master. Coghlan argues that this is further evidence of pro-Norwegian bias. It is virtually impossible to credit Coghlan’s argument, however, because Andreassen first offered the newly vacant master position to an American of non-Norwegian heritage (Kraljevich). Coghlan argues that this fact is no more than evidence against him to be evaluated by the jury at trial. But we cannot see how any reasonable jury could conclude that Andreassen was motivated by pro-Norwegian or anti-American discrimination when his first choice was to replace a Scandinavian master with an American one. The directive to change the leadership of the Dynasty came originally from Michael Hyde, an American. And An-dreassen made his decision in consultation with Vargas, an American-born American citizen who Coghlan admitted is not biased in favor of Norwegians.

Moreover, the record suggests that ASC had legitimate reasons for demoting Cogh-lan. Andreassen and Vargas viewed the Dynasty as a poor performer, and its repair costs were much higher in proportion to the total tonnage caught than on the other vessels. Indeed, Coghlan himself agreed that the vessel’s performance was poor and that some sort of change was needed, though of course he did not think that his own- demotion was the right solution.

As further evidence of ASC’s discriminatory intent, Coghlan points to the fact that Andreassen simultaneously removed American masters on two other ships and replaced them with Norwegian-born masters. For two reasons, we find this evidence insufficient to rebut the same-actor inference. First, this “pattern” is hardly a pattern at all once one considers the fact that the master position on the Dynasty was first offered to an American. Second, even a pattern of three replacements is, under our precedent, too small a sample to constitute meaningful statistical evidence. See Aragon, 292 F.3d at 663-64 (holding that “the fact that three of the four casuals singled out for lay off that night were white” was not deserving of “much weight” because of the small sample size); Sengupta v. Morrison-Knudsen Co., 804 F.2d 1072, 1076 (9th Cir.1986) (holding that “statistical evidence derived from an extremely small universe ... has little predictive value and must be disregarded” (internal quotation marks omitted)); Shutt v. Sandoz Crop Prot. Corp., 944 F.2d 1431 , 1433 (9th Cir.1991); see also LeBlanc v. Great Am. Ins. Co., 6 F.3d 836, 848-49 (1st Cir.1993) (holding that“a small statistical sample carries little or no probative force to show discrimination” and noting that statistical evidence is generally less relevant in disparate treatment cases than in disparate impact cases because the focus is on the treatment of an individual rather than on an overall pattern); Aragon, 292 F.3d at 663 n. 6 (approvingly citing LeBlanc for that point). We conclude that Coghlan has not presented sufficient evidence to rebut the same-actor inference with regard to his removal as mate of the Dynasty.

3

The final incident that Coghlan claims is discriminatory occurred in 2002, when An-dreassen offered Coghlan the position of mate on the Katie Ann. Coghlan had previously been master on that ship, and he claims ‘ that Andreassen’s decision to appoint the Norwegian-born Hogseth as master was based on Andreassen’s pro-Norwegian bias.

It is questionable whether this should even be considered a separate act of discrimination from Coghlan’s removal as mate of the Dynasty: upon that removal, instead of simply firing him outright, ASC appointed him as mate of the Katie Ann. In any case, however, all the legitimate reasons for removing Coghlan as mate of the Dynasty also apply to not making him master of the Katie Ann. Again, Coghlan has not presented evidence sufficient to meet the burden imposed by the same-actor inference.

Ill

Employment discrimination cases inevitably present difficult problems of proof, precisely because we cannot peer into the minds of decisionmakers to determine their true motivations. All we can do is apply the evidentiary framework developed in the decisions of the Supreme Court and our own court. In this case, Coghlan has not presented evidence sufficient to defeat the same-actor inference with regard to any of the decisions he challenges. The district court was therefore correct to grant summary judgment in favor of ASC.

AFFIRMED.

. Pub.L. 105-277, div. C, tit. II, 112 Stat. 2681 (codified as amended in scattered sections of 16 and 26 U.S.C.). . The top two positions on fishing boats of this sort are, first, the "master” or "captain” and, second, the "mate.” . He also alleged impermissible retaliation and wrongful discharge in violation of public policy, but does not pursue those claims on appeal. . Washington’s employment discrimination law largely parallels federal law under Title VII, and our treatment of Coghlan's Title VII claim thus applies also to his similar claim under Washington law. See Hernandez v. Spacelabs Medical Inc., 343 F.3d 1107, 1112 (9th Cir.2003). . Moreover, it appears that ASC waived this argument by not presenting it to the district court. See Harris v. Pulley, 885 F.2d 1354, 1367 (9th Cir.1988). We need not rest on its waiver, however, because we reject the argument even on its merits. . When the evidence in question is of an employer's statements that do not directly concern the plaintiff, it is true that some inference is necessary to establish discrimination with regard to the plaintiff. Indeed, strictly speaking, little other than an employer’s own admission could establish discriminatory intent without any inference whatsoever. Nevertheless, when evidence establishes the employer's animus toward the class to which the plaintiff belongs, the inference to the fact of discrimination against the plaintiff is sufficiently small that we have treated the evidence as direct. See Cordova, 124 F.3d at 1149 (deeming “direct” evidence that the employer had referred to an employee other than the plaintiff as a "dumb Mexican”). . We have recently suggested that"specific and substantial” evidence may be required even when direct evidence is at issue. See Stegall v. Citadel Broadcasting Co., 350 F.3d 1061, 1066 (9th Cir.2003). We need not decide today whether that is so, because as we will explain, Coghlan presents no evidence that qualifies as "direct” under the relevant definition. . Unfortunately, some confusion may arise because the terms "direct” and "indirect” have occasionally been used, even by the Supreme Court, to distinguish the two varieties of circumstantial evidence — that is, evidence affirmatively establishing bias, and evidence negatively discrediting the employer’s stated rationale. See, e.g., Burdine, 450 U.S. at 256 , 101 S.Ct. 1089 ; Snead v. Metro. Prop. & Cas. Ins. Co., 237 F.3d 1080, 1093-94 (9th Cir.2001) (quoting Burdine). Coghlan thus suggests that evidence such as ASC's choice to pass over Coghlan in favor of an allegedly less-qualified Norwegian qualifies as "direct.” We made clear in Godwin , however, that for purposes of determining the level of evidence necessary to survive a summary judgment motion, "direct” evidence refers only to evidence (such as racist or sexist statements) that proves the fact of discriminatory animus without the need for substantial inference. See Godwin, 150 F.3d at 1221 ; Stegall, 350 F.3d at 1066 . . Cases not involving hiring and firing could arise, no doubt, in which the same-actor inference would be inappropriate. For example, if a plaintiff were alleging that his employer systematically excluded members of a certain class from upper-management positions, then the mere fact that the employer was willing to hire members of that class for lower-level positions would surely not prove otherwise. This is not such a case, though, because the relevant decisionmaker, Andreas-sen, had not merely hired Coghlan but had appointed him to advanced positions: master of the Katie Ann and mate/fishmate of the Dynasty. . Coghlan argues that the district court mistakenly applied the same-actor inference as a "mandatory presumption overriding all other evidence in the case.” The district court, however, was clearly aware that the same-actor inference was not "mandatory,” since it wrote that "Coghlan fails to rebut the presumption,” whereas a mandatory presumption is by definition irrebuttable. It is true that the district court used the term "presumption” rather than the term "inference” that appears in Bradley , and in some contexts, the two terms can have different meanings. It is clear, though, that we did not use the term "inference” in Bradley in its technical sense. The point, as Bradley makes clear and as the district court understood, is simply that when the allegedly discriminatory actor is someone who has previously selected the plaintiff for favorable treatment, that is very strong evidence that the actor holds no dis-criminatoiy animus, and the plaintiff must present correspondingly stronger evidence of bias in order to prevail. . Though Coghlan does not say so, this is necessarily an argument that Bradley is no longer good law, because Bradley itself dealt with the summary judgment context. . The Supreme Court needed to make this explicit in Reeves because some lower courts had effectively held that summary judgment was always appropriate unless the plaintiff presented not only a prima facie case and evidence sufficient to show that the employer’s proffered legitimate rationale was pretex-tual, but also additional evidence sufficient to show actual discrimination. See Reeves, 530 U.S. at 146-47 , 120 S.Ct. 2097 . . We may deal at the outset with one of Coghlan’s contentions. Among the reasons the district court gave for granting summary judgment was the fact that "Andreassen is a naturalized U.S. citizen who has renounced his Norwegian citizenship." Coghlan argues that the district court thus "erred in holding that a decision-maker's citizenship controls whether he discriminated against persons with the same citizenship.” Coghlan, however, drastically overstates the reliance placed on Andreassen’s citizenship by the district court. It did not hold that the decisionmaker's adopted citizenship "controls” the question of discrimination; it merely found citizenship to be one relevant piece of evidence. We need not decide whether it was wrong to do so, because no inference from Andreas-sen's citizenship is necessary to conclude that Coghlan has not met his burden of proof. . The fact that Coghlan’s demotion occurred as part of a more general rearrangement in which his superior was also demoted makes his claim of intentional bias even more difficult to credit. . Coghlan argues that several ships had higher repair costs in absolute terms. As ASC points out, however, it stands to reason that larger and more complex ships will have higher repair costs than smaller, simpler vessels. What is most relevant is the level of repair costs in proportion to the amount of fish caught, and in those terms the Dynasty was easily the most costly vessel. . Coghlan also points us to the testimony of Brooks Stevens, an American-born employee of ASC, who stated that he believes ASC gives preferential treatment to Norwegians. Stevens based his belief on the fact that he makes approximately the same salary as Lars Oter-hals, a Norwegian employee, even though Ot-erhals is only a fishmate whereas Stevens is simultaneously both a mate and a fishmate. Upon further questioning, however, he admitted that he was unsure that the salaries were due to Oterhals’s Norwegian origin. Michael Hyde testified that Brooks Stevens had stated to him that “he wasn't sure the Norwegians as a whole, but certainly certain Norwegians he thought in the past had received special treatment.” Hyde also explained that the salary issue that concerned Stevens was the result of ASC’s policy that "there are some boats where you have a person serving as both master.and fishmaster ... and yet if you serve two functions, you don't get paid significantly more than if you were only serving one function.” Coghlan presents no evidence to suggest that Hyde's description of ASC's salary policy was inaccurate. His evidence, then, reduces to Stevens's mere theory, unsubstantiated by any factual support whatsoever in the record, that his pay may have been related to his American origin. This speculation adds little to Coghlan's case.