The plaintiff physician, a psychiatrist, was charged before the defendant board with engaging in sexual conduct with a patient, which, if it occurred, would be a violation of G. L. c. 112, § 5 (c) (1986 ed.) (“gross misconduct in the practice of medicine”). The board referred the case to the Division of Administrative Law Appeals to hold a hearing and then to issue recommended findings of fact and a recommended decision. An administrative magistrate issued a recommended decision in which she concluded that the board had not proved the alleged misconduct. The administrative magistrate believed the physician and not his former patient. More than one year later, the board issued its final decision, revoking the physician’s registration to practice medicine on the ground that he had engaged in sexual activity with his patient.
The principal issue of contention in the physician’s appeal from the board’s decision is whether the board, in the circumstances, was warranted in rejecting the administrative magistrate’s determination of the credibility of the physician and the patient and in deciding that issue differently, based on its reading of the record. A single justice of this court concluded that the board was not so warranted, and he vacated the board’s decision. The single justice concluded that the board did not have substantial evidence on the record to support its conclusion that the instances of alleged sexual misconduct occurred. The board has appealed. We affirm the judgment vacating the board’s decision and order the matter remanded to the board for further proceedings consistent with this opinion.
*105Dr. Morris, who had practiced medicine since 1942 without prior complaint, had an office on Marlborough Street in Boston. The patient began seeing him for psychotherapy on a weekly basis in May, 1977, because of anxiety arising out of her deteriorating marriage and a dispute over the custody of her daughter. The physician participated in support of the patient’s attempt to obtain custody of her daughter. In August, 1979, the patient’s divorce became final, and the former husband obtained custody of their child. He remarried immediately and, contrary to court order, took the child out of the country.
The alleged sexual conduct between the patient and the physician occurred in the physician’s office in the months of September and October, 1979. The board makes no claim that the patient did not consent. The physician in turn concedes that it would be improper and not consistent with accepted medical practice for a psychiatrist to engage even in consensual sexual activity with a patient. At that time the patient was still seeking to regain custody of her child. The board concluded that the physician first engaged in sexual activity with the patient in early September, 1979. He attempted, unsuccessfully, to have intercourse with her. The patient testified that she felt that, if she did not comply, he would abandon her. The board found that during each of the successive sessions with the patient until she left the country in November, 1979, to attempt to regain custody of her child, the patient and the physician engaged in sexual activity, but that sexual intercourse was never consummated.
The board noted that the administrative magistrate’s recommended decision provided no discussion of the evidence in explanation of her decision that no sexual activity occurred during any of the therapy sessions. The magistrate did say that her conclusion was “based in part on my finding that the testimony of the [physician] was more credible than [the patient’s] testimony.” She added that she had had ample opportunity to observe the demeanor of the witnesses, including their appearance and general bearing. She described the physician as a “compelling and credible witness,” and found unconvincing *106the patient’s testimony concerning repeated sexual encounters with the physician.1
In a thorough decision, explaining at length why it believed the patient, the board concluded that the evidence of sexual misconduct was overwhelming. It noted the advantage a psychiatrist has in relation to a patient; the particular vulnerability of the patient at the time the sexual activity allegedly occurred; the scheduling of the patient’s appointments late in the afternoon; her statement on September 22,1979, to an acquaintance (discussed later) which the board viewed as a fresh complaint concerning the physician’s activity; the testimony of the patient’s subsequent phychiatrist; the physician’s disclosure of personal information about himself and his family to the patient; the conducting of therapy sessions away from his office; dining alone with the patient; allowing her bill to accumulate to an average of $4,000 and helping her once financially; and the contents of a letter the patient wrote to the physician from abroad in November, 1979. The board concluded that the physician engaged in inappropriate sexual and other activity with the patient that amounted to a violation of good and accepted psychiatric care, and gross misconduct in the practice of medicine in violation of G. L. c. 112, § 5 (c).
The board would have been warranted in remanding the case to the administrative magistrate for supplemental findings and for more detailed explanations of her reasoning. It would also have been warranted in ordering a new hearing before a different magistrate. We reject, however, the board’s decision that, in the circumstances, the board itself could properly make the finding that the physician had engaged in sexual activity with the patient. For reasons we shall explain, the board owed *107greater deference than it paid to the credibility findings made by the magistrate who heard the witnesses.
We agree with the board that the magistrate’s recommended decision was sparse in explanation of her determination of the central factual issue. She said that she based her finding that no sexual conduct took place “in part on my finding that the testimony of the [physician] was more credible than [the patient’ s] testimony ” (emphasis supplied), but she did not explain what else she relied on in making that finding. She did not discuss certain evidence and possible inferences from the evidence that tended to support the opposite finding. In this case, the magistrate should have made an explicit analysis of credibility and the evidence bearing on it. Although we are not as certain as the board that the magistrate was preoccupied with the absence of eyewitnesses, the board would have been justified in seeking a further explanation from the magistrate of the significance in her view of the absence of eyewitnesses.
The board also concluded that the administrative magistrate’s decision may have been affected by certain evidence she improperly admitted. The board stated that the magistrate should have more closely controlled the admission of evidence of the patient’s mental history. The probative value of this extensive evidence, the board said, “was completely overshadowed by the invasive nature and possible prejudicial effect.” The board thought the unwarranted intrusion might have affected the patient’s demeanor and thus improperly influenced the magistrate’s findings on credibility. The magistrate did not expressly rely on this evidence. The board did not rule that the evidence was irrelevant. The rules of evidence generally do not apply in administrative hearings governed by G. L. c. 30A. G. L. c. 30A, § 11 (2) (1986 ed.).2 Our reading of the record shows no intrusion that was so extensive or invasive on the subject *108of prior mental history as to have been likely to have prejudiced the patient in the eyes of an experienced administrative magistrate. An administrative magistrate must have considerable discretion, particularly early in a hearing, to admit evidence that is arguably relevant, particularly where, as here, counsel makes no objection that the evidence is prejudicial.
We share the board’s concern that the magistrate, over repe'ated objection, permitted “the [physician’s] counsel unfettered latitude to inquire into minute details of the Patient’s sexual history.” The magistrate doubted its relevance when she admitted that evidence. The board’s decision says that the evidence was “of questionable relevance.” We can see now that it had no relevance, and the physician does not argue here that it did. The public policy expressed in the rape-shield statute3 and in our common law decisions 4 requires that an administrative magistrate reject that kind of evidence in a proceeding like this, unless the proponent of the evidence demonstrates that evidence of a patient’s prior sexual conduct is more than marginally relevant to an important issue of fact.
We do not agree with the board, however, that the magistrate ’ s decision to admit the evidence was an abuse of discretion or that she “improperly shifted the focus from the [physician’s] alleged misconduct to the issue of a consensual relationship between the [parties].” At the time the evidence was admitted, counsel who then represented the physician made a representation. He argued that he would show that the patient mixed fact and sexual fantasy and that such a showing would be “extremely important in evaluating the veracity of the witnes*109ses.” That promise seems never to have been fulfilled as to the patient’s statements about her prior sexual conduct.5
Admission of this evidence did not shift the focus of the inquiry to the matter of consent. Everyone agreed that consensual conduct was not the issue. The magistrate was avowedly skeptical of the relevance of the evidence. On the record, we see no indication that the magistrate’s decision was influenced by this evidence or, as the board concluded, that this line of inquiry created “considerable reason to question the Magistrate’s rejection of the Patient’s testimony on the basis of ‘demeanor,’ ‘appearance,’ and ‘general bearing,”’ As best one can tell from the reading of a cold transcript, this evidence did not make this patient ineffective as a witness, a condition that could well develop with a different witness in similar circumstances. (See Globe Newspaper Co. v. Superior Court, 379 Mass. 846, 859, vacated on other grounds, 449 U.S. 894 [1980]. S.C. 383 Mass. 838 [1981], rev’d on other grounds, 457 U.S. 56 [1982]).
Although we agree in certain respects with the board’s concems about the magistrate’s recommended decision, we do not accept the board’s determination that it was entitled to disregard the magistrate’s finding on credibility and make its own determination on the central factual issue in the proceeding. In the absence of a statute directing otherwise, we have required that one who has heard the witnesses in an agency proceeding make an assessment of credibility when there is a material conflict in the testimony. See Salem v. Massachusetts Comm’n Against Discrimination, 404 Mass. 170, 175 (1989) (commissioner who heard witnesses died before making decision; decision on printed record by another commissioner not permissible where testimony conflicted on material issues); Dowd v. Director of the Div. of Employment Sec., 390 Mass. 767, 771 (1984) (review examiner who heard evidence should choose between conflicting versions of material events). Cf. Amherst-Pelham Regional School Comm. v. Department of Educ., 376 Mass. 480, 498 (1978) (where credibility of witnesses is not involved *110hearing officer who did not hear witnesses may issue decision); Perkins v. School Comm. of Quincy 315 Mass. 47, 52-53 (1943) (where statute required attendance at hearing by school committee members who were to vote, reading transcript was not adequate substitute for attendance); G. L. c. 30A, § 11 (7) (1986 ed.) (by implication permitting majority of officials of agency who are to render final decision either to hear or to read evidence), discussed in Board of Appeals of Maynard v. Housing Appeals Comm. in the Dep’t of Community Affairs, 370 Mass. 64, 66 (1976).
We have never held, however, nor does the physician argue here, that the administrative agency responsible for making the final decision may not revise or reject the findings of a hearing officer on conflicting evidence.6 The physician makes no claim that any general constitutional principle, such as the denial of due process of law, always bars an agency from reversing the findings of fact of a hearing officer made on conflicting evidence.7 He does argue, however, that in these circumstances the agency’s decision denied him fairness in a due process sense.
Although this court had not addressed the degree of deference an administrative board must give to subsidiary findings of fact made by an administrative magistrate, the Appeals Court did so in a carefully reasoned opinion by Chief Justice Hale. Vinal v. Contributory Retirement Appeal Bd., 13 Mass. App. *111Ct. 85, 99-102 (1982). That court concluded that, when the subsidiary findings of a hearing officer rest on a “resolution of credibility questions (i.e., that a fact is true because a witness testified to it and that witness is believable), they should be entitled to substantial deference.” Id. at 101. See Universal Camera Corp. v. NLRB, 340 U.S. 474, 496 (1951). The Appeals Court added that, if the board were to reject subsidiary findings of a hearing officer, it must comply with the requirement of G. L. c. 30A, § 11 (8) (1986 ed.), that its decision contain “a considered articulation of the reasons underlying that rejection.” Vinal, supra at 101-102. We accept the substantial deference standard expressed in the Vinal opinion. By using that standard we preserve to the board the right to exercise its judgment and statutory responsibility (see A. Celia, Administrative Law and Practice §§ 347, 349 [1986]), while recognizing the right of a physician to have the proceeding decided on the question whether there was substantial evidence on the whole record supporting the board’s decision. We turn, therefore, to a consideration of the board’s decision to see whether, in articulating its reasons for rejecting the magistrate’s resolution of credibility questions, the board gave substantial deference to the magistrate’s findings.
In certain respects, the board’s reasoning is not persuasive. The board was plainly wrong in concluding that it was aided on the credibility issue by having the benefit of transcripts which were not available to the magistrate. The board’s reliance on the assertion that the patient was scheduled in September and October, 1979, as the last patient in the day in order to avoid detection of the physician’s misconduct by others, is not supported by the facts agreed to by the parties.8 We substantially discount the patient’s statements given to a successor psychiatrist made after the patient had commenced a civil action against the doctor based in part on her claim of the physician’s *112improper sexual conduct toward her. Such self-serving statements do little to provide worthwhile corroboration. We see little support for the board’s reliance on the language used in a letter the patient wrote the physician in November, 1979, referring to herself as “this sweet patient.” In context, the words are not significant to this case.
There was, to be sure, evidence that tended to support the board’s conclusion that the patient was telling the truth. The board noted that a psychiatrist learns a great deal about a patient’s weaknesses, emotions, and thoughts and is in a position to take advantage of a patient. In this case, the alleged advances occurred, not coincidentally the board thought, at a time when the patient was particularly vulnerable because of the loss of custody of her child. There was evidence that the physician had disclosed details of his personal life to the patient and had, in the board’s view, compromised the professional relationship in other ways by letting her bill mount up without any substantial payments and by giving her proceeds from a Blue Shield check; by taking the patient to dinner alone at the University Club, and by having sessions in inappropriate places, such as at her residence and in his automobile parked on a public street. Some corroboration of the patient might be found in the testimony of an acquaintance of the patient, a college professor, who testified to what he described as a “bizarre conversation” he had had with the patient on September 22, 1979, during which the patient said that her psychiatrist, not named, had had intercourse with her. Before the magistrate, however, the patient testified that the physician was unable to have intercourse on any occasion.
If the board had heard the evidence and had issued a decision as carefully crafted and as well directed to the evidence in the record as the one it issued, we would have no difficulty in concluding that there was substantial evidence on the whole record to support the board’s conclusions. Here, however, the boárd did not hear the evidence. It concluded, nevertheless, that what it viewed as overwhelming evidence of sexual misconduct made it unnecessary to obtain corrections and amplifications from the person who did hear and see the witnesses. *113Moreover, in deciding that there was overwhelming evidence, the board relied on its understanding of the “dynamics” of a psychiatrist-patient relationship involving sexual exploitation. The board said that “through our work in the field and handling of disciplinary cases, we have gained an understanding of how to interpret and weigh the particular facts in an case of sexual abuse. It is through this insight and expertise in reviewing such cases that we can identify certain red flags indicative of sexual abuse. Without this appreciation, critical facts can be misinterpreted or worse yet, ignored.”
The board exceeded its proper role in announcing, with no expert evidence in the record to support it, that its special expertise permits it to identify “red flags indicative of sexual abuse.” If there is some method known to medical science for determining whether sexual misconduct has occurred in particular circumstances, that information should be disclosed on the administrative record, presumably in the form of an expert opinion. See G. L. c. 30A, § 11 (5) (1986 ed.). The board failed to give substantial deference to the findings of the magistrate on matters of credibility when it substituted a collective assessment that was based on the undisclosed significance of certain “red flags.” Arthurs v. Board of Registration in Medicine, 383 Mass. 299, 310 (1981). If, as the board indicates, these “red flags” are something more than evidence tending to prove certain facts or tending to corroborate the testimony of certain witnesses (the kind of evidence triers of fact have traditionally relied on), they and their importance must be disclosed on the record so that a physician may challenge that evidence. Id. at 305.
Where the magistrate concluded that the physician was credible and where the record lacked evidence to support the significance of the “red flags” the board relied on, the board’s decision was not supported by substantial evidence on the whole record. We do not conclude, however, that a judgment here or a board decision in favor of the physician is required. The board may wish to remand the matter to the administrative magistrate for further findings or for an explanation of her *114decision, or both.9 The board may choose instead to order a new hearing.
We affirm the judgment of the single justice vacating the decision of the board and remand the matter to the board for further proceedings consistent with this opinion.
So ordered.