concurring in part and dissenting in part:
I concur in the majority opinion with one exception. I dissent from the majority’s harmless error analysis of the administrative law judge’s (“ALJ”) adverse credibility finding with respect to Carmickle.5 Maj. Op. at 1159-67.
*1168A. The Majority Applies an Incorrect Harmless Error Test.
An ALJ’s error is harmless if, in light of the record-supported reasons supporting the adverse credibility finding, we can conclude that the ALJ’s error did not “affect! ] the ALJ’s conclusion.” Batson v. Comm’r of Soc. Sec. Admin., 359 F.3d 1190, 1197 (9th Cir.2004); see also Stout v. Comm’r, Soc. Sec. Admin., 454 F.3d 1050, 1054-55 (9th Cir.2006) (describing the harmless error test as whether “the ALJ’s error did not materially impact his decision”); Robbins v. Soc. Sec. Admin., 466 F.3d 880, 885 (9th Cir.2006) (holding that an error is harmless if it was “inconsequential to the ultimate nondisability determination” (internal quotation marks omitted)).
The majority disagrees. Instead, the majority holds that “the relevant inquiry ... is whether the ALJ’s decision remains legally valid, despite such error.” Maj. Op. at 1162. “So long as there remains ‘substantial evidence supporting the ALJ’s conclusions on ... credibility’ and the error ‘does not negate the validity of the ALJ’s ultimate [credibility] conclusion,’ such is deemed harmless and does not warrant reversal.” Maj. Op. at 1162 (ellipsis and alteration in original) (quoting Batson, 359 F.3d at 1197). By quoting selectively from Batson, the majority eviscerates harmless error review and creates an intra-circuit conflict with Batson, Stout, and Robbins.
Until today, Batson was the only case in which we have held that an ALJ’s error concerning an adverse credibility finding was harmless. There, the ALJ gave numerous record-supported reasons for finding the claimant not credible, but also made one assumption that was not supported by the record. Batson, 359 F.3d at 1196-97. In particular, “Batson had said in questionnaires about his daily living activities that he watched six to ten hours of television a day,” and the ALJ erroneously assumed that Batson was sitting continuously while watching television (instead of standing, reclining, or changing positions frequently). Id. Examining the full record, we held that the ALJ’s one minor error was harmless: “In light of the substantial evidence supporting the ALJ’s conclusions on Batson’s credibility, we do not think that the ALJ’s assumption about Batson sitting while watching television affected the ALJ’s conclusion or requires remand.” Id. at 1197.
As the quoted sentence makes clear, we did not ask whether “the ALJ’s remaining reasoning and ultimate credibility determination were adequately supported by substantial evidence in the record.” Maj. Op. at 1162 (emphasis omitted) (citing Batson, 359 F.3d at 1197). Instead, we asked whether, “[ijn light of the substantial evidence supporting the ALJ’s conclusions on Batson’s credibility,” the ALJ’s assumption about sitting while watching television “affected the ALJ’s conclusion.” Batson, 359 F.3d at 1197 (emphases added). The fact that some of the ALJ’s reasons were supported by substantial evidence was a necessary pre-condition for reaching the harmless error test; it was not the test itself. Indeed, if substantial evidence supporting the ALJ’s conclusion were sufficient, there would have been little need for the two full paragraphs discussing whether the ALJ’s error was harmless.
Until today, we have had no difficulty understanding Batson’s holding. In Stout, 454 F.3d at 1054-55, we summarized Bat-son’s holding this way: “[W]e concluded that any error the ALJ committed in assuming [Batson was sitting while watching television] was harmless!,] • • • because the ALJ provided numerous other record-supported reasons for discrediting the claimant’s testimony, which allowed our review to determine the ALJ’s error did not mate*1169rially impact his decision.” (Emphasis added.) In turn, we noted in Robbins, 466 F.3d at 885, that “we explained [in Stout, 454 F.3d at 1055-56] that we have only found harmless error when it was clear from the record that an ALJ’s error was ‘inconsequential to the ultimate nondisa-bility determination.’ ” (Emphasis added.) See also Ford v. Astrue, 518 F.3d 979, 983 (8th Cir.2008) (citing Batson and holding that, “[a]fter careful consideration of the record in this case, we cannot say that it weighs so heavily against Ms. Ford’s credibility that the ALJ would necessarily have disbelieved her absent the erroneous inferences that he drew from the record ” (emphasis added)).
In both Robbins and Stout, we accurately described and applied the harmless error test over a dissent that would have applied a less stringent test. See Stout, 454 F.3d at 1057 (O’Scannlain, J., dissenting); Robbins, 466 F.3d at 889-93 (O’Scannlain, J., dissenting). In Robbins, Judge O’Scannlain, in dissent, would have preferred that the panel apply the very test now embraced by the majority, based on the same misreading of Batson. Compare id. at 889 (O’Scannlain, J., dissenting) (“In Batson, for example, we simply asked whether there remained ‘substantial evidence supporting the ALJ’s decision,’ or whether the error in any way ‘negate[d] the validity of the ALJ’s ultimate conclusion.’ A similar analysis would have been proper in this case.” (citations omitted) (alteration in original)) with Maj. Op. at 1162 (“So long as there remains ‘substantial evidence supporting the ALJ’s conclusions on ... credibility’ and the error ‘does not negate the validity of the ALJ’s ultimate [credibility] conclusion,’ such is deemed harmless and does not warrant reversal.” (ellipsis and alteration in original) (quoting Batson, 359 F.3d at 1197)). Unless and until one of our precedents is overruled or becomes clearly irreconcilable with a Supreme Court holding, however, it is binding on a three-judge panel, even if we disagree with it. Miller v. Gammie, 335 F.3d 889, 900 (9th Cir.2003) (en banc). Contrary to the opinion’s attempted justification, Maj. Op. at 1161 n. 2, the majority here adopts the Stout and Robbins dissents’ interpretation of Batson, which the Stout and Robbins majorities cogently rejected.
B. The ALJ’s Errors Are Not Harmless Under the Correct Test.
The ALJ gave four reasons supporting his adverse credibility finding: (1) Car-mickle testified that he must “ ‘change positions constantly’ ” but, “when pressed by counsel, he admitted that he can sit for ‘[a]bout 15 minutes in one particular position’ (2) Carmickle testified that he can lift 10 pounds only occasionally, but his doctor opined that he can lift up to 10 pounds frequently; (3) Carmickle alleges that he is disabled, but he received unemployment benefits; and (4) Carmickle alleges that he has severe pain, but he took only Ibuprofen and not other pain medications. Maj. Op. at 1160-62. I agree with the majority that the ALJ erred because the last two reasons may not be considered under our precedents. Maj. Op. at 1161-62.
Reviewing the record, I cannot say that the ALJ’s errors did not affect his conclusion that Carmickle’s testimony was not credible. In Batson, we held that the ALJ’s assumption that Batson was sitting while watching television was harmless in light of the “numerous other record-supported reasons for discrediting the claimant’s testimony.” Stout, 454 F.3d at 1055 (emphasis added). Here, the errors are not trivial: the ALJ held against Carmickle the facts that he had received unemployment benefits and that he had declined to take prescription pain medications — facts that may not be considered at all by the *1170ALJ, as the majority properly holds. Additionally, the “other record-supported reasons for discrediting[Carmickle’s] testimony” are neither “numerous” nor particularly compelling. Carmickle testified that he has to change positions “constantly” but then clarified, when pressed by counsel, that he could sit for 15 minutes without changing positions. As the majority recognizes, the ALJ credited Carmick-le’s clarification and included the 15-min-ute limitation in the residual functional capacity assessment. Maj. Op. at 1161. It is not apparent that changing positions every 15 minutes, all day long, necessarily would be perceived or described as something less than “constant” motion. Car-mickle also testified that he can lift 10 pounds occasionally even though his doctor found that he could lift 10 pounds frequently. That difference is significant as a term of art, but is not necessarily a meaningful distinction to a lay person.
In summary, the ALJ erred by considering two factors that the law prohibits. Because only two, relatively minor record-supported reasons buttress the ALJ’s conclusion, while two reasons were wholly improper, I cannot conclude that the ALJ’s errors did not affect his adverse credibility finding. Applying the harmless error test mandated by our precedent, I would therefore hold that the ALJ’s errors were not harmless.
. I agree with the majority that we must reverse and remand because the ALJ erred by not including Carmickle's tendonitis limitation in the ALJ’s assessment of Carmickle’s residual functional capacity. Maj. Op. at 1164-65. I also agree with the majority that we must affirm on the remainder of the issues.