POSNER, Circuit Judge.

We have consolidated for decision three challenges, argued the same day, to district court decisions affirming denials by the Social Security Administration of disability benefits and (for persons who lack social security insurance) supplemental security income benefits, which are similar. Recently, in Spiva v. Astrue, 628 F.3d 346 (7th Cir.2010), and Parker v. Astrue, 597 F.3d 920 (7th Cir.2010), we criticized the Social Security Administration's handling of disability claims in several respects summarized in Spiva:

628 F.3d at 348. We noted that similar criticisms could be found in a number of other court of appeals opinions, in this and other circuits. Id. In two of the cases before us, infirmities similar to those that we found in Spiva and Parker require us to reverse; in the third, the administrative law judge's opinion was thorough and well supported.

We are mindful of the difficulties that the Social Security Administration's administrative law judges labor under. They have a very heavy caseload—the median annual number of disability hearings conducted by an administrative law judge is almost 400 (the average is 434). Social Security Advisory Board, "Improving the Social Security Administration's Hearing Process" 11-12 (Sept.2006), www.ssab.gov/documents/HearingProcess.pdf (visited Dec. 30, 2010, as were all the websites cited in this opinion); Social Security Advisory Board, "Disability Decision Making: Data and Materials" 75 (May 2006), www. ssab.gov/documents/chartbook.pdf; "Social Security Disability: Management of Disability Claims Workload Will Require Comprehensive Planning" 6 (General Accounting Office, Rep. No. 10-667T, Apr. 2010). Staff support is inadequate. "Improving the Social Security Administration's Hearing Process," supra, at 14. The large number of administrative law judges (more than 1400), combined with limited administrative appellate capacity, has resulted in great uncorrected variance in denial rates across administrative law judges. See "ALJ Disposition Data FY 2011," Social Security Online, www.ssa. gov/appeals/DataSets/03_ALJ_Disposition_Data.pdf; "Improving the Social Security Administration's Hearing Process," supra, at 4-5. This in turn implies frequent inconsistency, id. at 6-7, 25; "Social Security Administration: More Effort Needed to Assess Consistency of Disability Decisions" (General Accounting Office, Rep. No. GAO-04-656, July 2004), and quality problems generally. The Lewin Group, Inc., et al., Evaluation of SSA's Disability Quality Assurance (QA) Processes and Development of QA Options That Will Support the Long-Term Management of the Disability Program 16-24 (Final Report, Mar. 16, 2001), www.lewin.com/content/publications/1325.pdf. There thus are ominous parallels to the much-remarked inadequacies in the administration of the immigration laws by immigration judges, see, e.g., Benslimane v. Gonzales, 430 F.3d 828, 829-30 (7th Cir.2005), a type of administrative law judge.

A study found that the district courts (the first-line reviewers of denials of social security benefits) reversed 6.15 percent of denials outright and ordered benefits awarded in those cases, and remanded 48 percent of the denials, and 60 percent of the remands eventuated in a grant of benefits. This means that a total of 34.95 percent (.48 × .60 = .288 + 6.15) of all appeals to the district courts from denials of benefits resulted in their eventual grant. Paul R. Verkuil & Jeffrey S. Lubbers, "Alternative Approaches to Review of Social Security Disability Cases," 55 Admin. L.Rev. 731, 761-62 (2003). (In 2004, the 60 percent figure rose to 67 percent. Disability Decision Making: Data and Materials, supra, at 89.) And that is apart from reversals by the courts of appeals of district courts' affirmances of denials of benefits, as in two of the three cases we decide today.

Since we don't see cases in which social security disability benefits are granted in error, because the government cannot appeal from a grant at the final administrative level, we cannot (quite apart from the nonnegligible possibility of judicial error) conclude that administrative law judges have a 35 percent error rate. Moreover, the high reversal rate may simply reflect caution on the part of claimants' lawyers, since they are unlikely to obtain a significant (perhaps any) fee if the appeal fails; their clients invariably are impecunious. But approximately 20 percent of denials at the highest administrative level (computed from Verkuil & Lubbers, supra, at 760, reporting statistics from 2000), are appealed, which is a high appeal rate; and the higher an appeal rate, the lower the expected reversal rate if the tribunal appealed from is doing a good job.

So much for background; on to the cases, beginning with Martinez (No. 10-1957). Anita Martinez, age 35 at the time of her hearing before the administrative law judge, lives in the basement of her mother's home with her five children. She suffers from severe depression, has symptoms of bipolar disorder (often associated with depression), and has severe arthritic joint and bone pain throughout her body and a swelling of the hands that makes it very difficult for her to carry things, open packages, wash dishes, or write. Physicians have observed worsening signs and symptoms of severe musculoskeletal pain, and have treated her by prescribing ever more potent drugs (plus splints to wear on her wrists), but with mixed results. She also takes drugs for her mental conditions. Unsurprisingly she has difficulty concentrating. Her mother and her three eldest children help her with most household tasks, as she has to rest frequently throughout the day. Daily she thinks about committing suicide, and the children are instructed to summon their grandmother if their mother seems unusually depressed or suicidal.

The administrative law judge found that Martinez indeed has severe arthritis and severe depression, but also found that she can "use her hands for fine and gross manipulations," has only "moderate difficulties" with "concentration, persistence, or pace," and is not disabled, because she is capable of doing her former work (she last worked in 2003) as a car-wash attendant or fast-food cashier, and could also work as a "hand assembler."

The administrative law judge's opinion is perfunctory. The analysis portion begins with the boilerplate recital that "the claimant's medically determinable impairments could reasonably be expected to produce some symptoms, but ... the claimant's statements concerning the intensity, persistence and limiting effects of these symptoms are not entirely credible." There is no explanation of which of Martinez's statements are not entirely credible or how credible or noncredible any of them are. The next sentence is that "the claimant's longitudinal medical history is not necessarily consistent with his [sic] allegations of disability." We don't understand the significance of "necessarily." Did the administrative law judge think Martinez's medical history consistent or inconsistent with her claims?

With regard to pain, all the opinion states is that a rheumatologist "indicates that he detected very subtle evidence of an inflammatory process. He describes it as not well developed, and that [t]he laboratory tests and physical examinations were normal. He suggested that she take a long acting anti-inflammatory." In fact the suggestion for such a drug came from Martinez, who wanted a long-acting drug because she was afraid to have a large quantity of strong drugs on hand given her depression and suicidal thoughts. And the rheumatologist, while stating that the results of a general physical examination of Martinez were normal, observed "definite evidence of a true inflammatory process" and diagnosed her with inflammatory polyarthritis—an inflammation of more than one joint. And his report was only part of the evidence of the severity of Martinez's joint and bone pain, having been issued years before another physician treated her musculoskeletal pain, and also symptoms of carpal tunnel syndrome in both hands, with strong drugs and the wrist splints.

The administrative law judge discussed so little of the evidence that her conclusion that Martinez's complaints of severe pain and fatigue "are out of proportion to the objective physical findings" is suspended over air. She did not so much as mention the difficulty that Martinez has in using her hands, which makes the suggestion that she could be a "hand assembler" amount to a bad joke. Ignored was the requirement that administrative law judges carefully evaluate all evidence bearing on the severity of pain and give specific reasons for discounting a claimant's testimony about it. Social Security Administration, "How We Evaluate Symptoms, Including Pain," 20 C.F.R. § 404.1529; "Policy Interpretation Ruling Titles II and XVI: Evaluation of Symptoms in Disability Claims: Assessing the Credibility of an Individual's Statements," Social Security Ruling 96-7p; Lopez v. Barnhart, 336 F.3d 535 (7th Cir.2003) (per curiam); Lingenfelter v. Astrue, 504 F.3d 1028, 1035-38 (9th Cir.2007); Patrice Rusconi et al., "Taking into Account the Observers' Uncertainty: A Graduated Approach to the Credibility of the Patient's Evaluation," 33 J. Behav. Med. 60, 68 (2010). "The etiology of pain is not so well understood, or people's pain thresholds so uniform, that the severity of pain experienced by a given individual can be `read off' from a medical report." Johnson v. Barnhart, 449 F.3d 804, 806 (7th Cir.2006).

The administrative law judge found that Martinez's severe depression is well controlled by drugs—when she takes them— but ignored the fact that during manic spells Martinez had stopped taking her medications (a common consequence of mania). As we noted in Spiva v. Astrue, supra, 628 F.3d at 351, antidepressant drugs often produce serious side effects that make patients reluctant to take them. George I. Papakostas, "Limitations of Contemporary Antidepressants: Tolerability," 68 J. Clin. Psychiatry 11 (2007); M. Robin DiMatteo et al., "Depression Is a Risk Factor for Noncompliance with Medical Treatment: Meta-analysis of the Effects of Anxiety and Depression on Patient Adherence," 160 Archives of Internal Med. 2101 (2000); see also Gary D. Tollefson, "Antidepressant Treatment and Side Effect Considerations," 52 J. Clin. Psychiatry 4 (1991); Vivek Kusumakar, "Antidepressants and Antipsychotics in the Long-Term Treatment of Bipolar Disorder," 63 id. 23 (2002). Anyway people with serious psychiatric problems are often incapable of taking their prescribed medications consistently. The administrative law judge said that Martinez "has not required the services of a psychiatrist or outpatient psychotherapy for many years"—ignoring the fact that at the time of the hearing Martinez was on a waiting list for a consultation with a psychiatrist.

Absurdly, the administrative law judge found that Martinez "is able to maintain a very active lifestyle in the process of caring for her five children," and concluded from this that she can work—ignoring the fact that Martinez's mother plus three of the children are caring for her. She would not receive that care at a car wash.

Having concluded that Martinez was able to do her past work (plus assemble products by hand), the administrative law judge asked a vocational expert how many jobs there were in Illinois that she could fill—plenty, of course, if she is able to do the work she has done in the past, before her physical and mental conditions worsened. The vocational expert was not asked to consider how many jobs a person with Martinez's handicaps could perform. Nor did the administrative law judge ever consider the interaction of Martinez's many physical and mental problems. Even if each problem assessed separately were less serious than the evidence indicates, the combination of them might well be totally disabling. See, e.g., Parker v. Astrue, supra, 597 F.3d at 923; Villano v. Astrue, 556 F.3d 558, 563 (7th Cir.2009); Gentle v. Barnhart, 430 F.3d 865, 868 (7th Cir.2005); Golembiewski v. Barnhart, 322 F.3d 912, 918 (7th Cir.2003); see also Matthew J. Bair, et al., "Depression and Pain Comorbidity: A Literature Review," 163 Archives of Internal Med. 2433 (2003); Bruce A. Arrow et al., "Comorbid Depression, Chronic Pain, and Disability in Primary Care," 68 Psychosomatic Med. 262 (2006).

As in previous cases, the lawyers for the Social Security Administration, in defending the denial of benefits to Martinez in their brief in this court, milked the record for other evidence that might support the denial, besides evidence discussed by the administrative law judge. The evidence they found and the arguments they make (such as that a 12-year-old child is too young to assist her mother with household tasks) are unconvincing, and in places desperate.

Rider (No. 10-2603). Francis Rider, who is 61, has not worked since 2001, shortly after she injured her right knee and leg. She seeks disability benefits on the basis of a combination of arthritic knees, back pain, and obesity. She is only 5 feet 1 inch tall, but her weight varies between 205 and 220 pounds. That is an average of 212.5 pounds, which translates to a body mass index of 40.1. A person with a BMI of 30 is deemed obese, and a person with a BMI of 40 is deemed extremely obese. "Policy Interpretation Ruling Titles II and XVI: Evaluation of Obesity," Social Security Ruling 02-1p. Dr. John Cragg, an orthopedic surgeon, examined Rider, noticed that she was limping, examined x-rays of her right knee, diagnosed arthritic bone degeneration, and concluded that the knee was "well beyond" minimally invasive surgery and "heading for" knee replacement. His evaluation of the severity of her knee problem was corroborated by Dr. Eric Carlsen, the only state-agency physician who examined Rider; he reported that she was incapable of prolonged standing or walking. The administrative law judge, perhaps forgetting that he was required to give "good reasons" for not giving the well-supported opinion of a treating physician "controlling weight," 20 C.F.R. § 416.927(d)(2); Campbell v. Astrue, 627 F.3d 299, 306 (7th Cir. 2010), relied without explanation on the contrary views of two physicians who did not examine Rider though they consulted her medical records. He attached considerable weight to the fact that Rider had not undergone knee replacement, disregarding her uncontradicted testimony that she could not afford it. He also found that having decided not to have the knee replacement she underwent physical therapy that greatly improved her condition, but the finding is not consistent with the evidence. He remarked inconsequently that she enjoyed watching her grandchildren, from which the Social Security Administration's counsel at the argument in our court inferred that she must be highly mobile, because the grandchildren must run around a lot, so that she would have to be in continual motion to watch them.

The gravest error in an opinion with a number of contenders for that distinction is the failure to consider the bearing of Rider's extreme obesity. The administrative law judge mentioned in passing that it is a severe impairment but did not consider its significance in relation to Rider's knee. It is one thing to have a bad knee; it is another thing to have a bad knee supporting a body mass index in excess of 40. We repeat our earlier reminder that an applicant's disabilities must be considered in the aggregate.

Enough said.

Pound (No. 10-2080). The applicant, a 60-year-old woman, suffers from a variety of chronic illnesses, including coronary artery disease, and has neurological symptoms, such as vertigo, blurred vision, and pulsation in the ears, caused by carotid artery disease. She has experienced a transient ischemic attack (a "warning stroke") as a result of the latter disease, and a variety of symptoms from the former, such as chest pains, palpitations, and shortness of breath. She has back pain resulting from spinal disk disease, and a mysterious ailment called "restless leg syndrome," which refers to experiencing a series of muscular jerks at the moment of falling asleep.

Pound had social security disability coverage only until the end of 2003; if she was not disabled by then, she cannot obtain benefits even if she is disabled now, 42 U.S.C. § 423(c); 20 C.F.R. § 404.140—as she appears to be. Unfortunately for her claim, the symptoms of her coronary artery disease and carotid artery disease had, as a result of excellent medical treatment, largely disappeared by the time she applied for benefits. She applied in 2004 and went downhill rapidly thereafter. Until that happened she had, by her own account, been doing a lot of farm work. She reported experiencing pain from 2000 to 2003 but relied on mild pain medication to deal with it. The only medical opinion she submitted to the Social Security Administration was a 2005 report from her physician that relied primarily on examinations after 2003. The administrative law judge conducted a thorough analysis and concluded that as of the date on which Pound last had social security insurance coverage, the combination of her conditions merely limited her to performing sedentary work, which was the kind of work she had performed before she had first become seriously ill. (She had been an accounting manager.) In the brief but critical window (because of the expiration of her insurance coverage) between the time before her diseases were brought under control and the time that they flared up again, she was not disabled.

To summarize, in Nos. 10-1957 and 10-2603 the district courts' denial of relief is reversed and the cases remanded with instructions to return the matters to the Social Security Administration for further proceedings consistent with this opinion. In No. 10-2080 we affirm.