The plaintiff appeals from a judgment of a Superior Court judge affirming an administrative determination by the Department of Social Services (department), that his hitting or spanking of his minor child constituted “abuse,” as that term is defined by statute and regulation. See G. L. c. 119, § 51 A; 110 Code Mass. Regs. § 2.00 (1996).1 He argues that the department’s decision to support a report of abuse was not adequately *386supported by the factual record. He further contends that the department’s action amounts to an unjustifiable interference by the Commonwealth with his fundamental rights, under both the United States and Massachusetts Constitutions, to the free exercise of his religious beliefs and privacy in child rearing. We granted the plaintiff’s application for direct appellate review and conclude that the department’s decision was not supported by “substantial evidence” in the administrative record. We vacate the judgment of the Superior Court without reaching the constitutional issues.
1. Facts. We begin with a summary of relevant facts.
a. The investigation and decision under G. L. c. 119, § 51A. On March 19, 1997, the department received a report from a mandated reporter, a school teacher, made pursuant to G. L. c. 119, § 51A (51A report), regarding possible abuse and neglect of a nine year old student, the plaintiff’s son. 2 An investigation was conducted by a department social worker, Rena L. Ugol, who separately interviewed the reporter, the plaintiff, the boy, the boy’s mother, and two of the boy’s physicians.3
*387The reporter informed Ugol that there had been no prior concerns about possible abuse of the boy, but that his parents were separated and in the process of divorcing and he had lately appeared “more depressed, angry.” Although she believed the accuracy of the boy’s reports of physical punishment, the reporter stated that the boy would “exaggerate sometimes” and that he might have “perceptual difficulties.”
In his interview with Ugol, the plaintiff admitted to striking the boy on his buttocks with a leather belt on five or six occasions during the preceding school year. He denied being a “spontaneous spanker,” explaining that he only spanked the boy as punishment for reports of misbehavior at school. He described the punishment as follows: The plaintiff would have the boy stand next to him and place his hands on the plaintiff’s outstretched left hand (this latter measure to ensure that the boy would not suffer injury to his hands by attempting to shield his buttocks from the spanking); the plaintiff would grasp the belt buckle in his palm and wrap the belt, which was approximately one and one-half inches wide, around his right hand, leaving approximately one foot of leather strap exposed; he would then hit him on his clothed buttocks once or twice with the strap, explaining to him that it was punishment for bad behavior and that such discipline is required by the Bible. At Ugol’s request, the plaintiff demonstrated the force with which he would spank the boy by striking a couch cushion with the belt. Ugol reported that the belt made a “solid smack.” The plaintiff denied ever having caused any bruising on the boy’s buttocks (although he later admitted that he had never checked for any).
*388Both the boy and his mother confirmed much of the plaintiff’s account. The mother described the plaintiff as nonviolent and controlled, and stated that his disciplining of the boy was never done in anger and “doesn’t escalate” beyond spanking. She reported that, when administering a spanking, the plaintiff would hug the boy, tell him that he loved him, and explain that it was punishment for his misconduct. The boy, although expressing his fear and dislike of the spanking, told Ugol that the plaintiff “wouldn’t hurt me but would spank me.” He confirmed that the spankings were administered as punishment for misbehaving at school, and stated that the plaintiff would hit him once or twice (and occasionally up to five times, if he was “really bad”) with a belt on his fully clothed buttocks, but that the plaintiff “doesn’t whack really hard.” Sometimes when the boy expected a spanking, he would put on sweat pants under his jeans. The boy variously described the physical effects of the spankings as “red marks” on the skin of his buttocks, as marks that were “not red red red” but more like “pink,” and as “a teeny thing of red . . . not really red.” He stated that these marks would last about ten minutes and then fade.
The boy’s pediatrician, Dr. Joel Solomon, informed Ugol that the boy suffers from arthrogryposis, a congenital muscle condition which requires him to wear braces on his back and legs and to undergo regular physical therapy. When asked by Ugol whether, given the boy’s condition, he would have any special concerns about his being disciplined with a belt, he replied that he “sure would” and that it “wouldn’t help the condition,” but he did not specify any particular harmful effects. He reported never having seen bruising or other signs of physical abuse on the boy.
Dr. Michael Erlich, the boy’s pediatric orthopedist, who examined his “whole body” every three to four months, also reported never having seen bruising or other marks on the boy. He stated his opinion that the boy’s parents, and the plaintiff in particular, were “unbelievably devoted” to the boy, noting that it was the plaintiff who made sure that the boy performed all his required physical exercises, without which he would develop muscular deformities.
On the basis of this investigation, Ugol supported the 51A report of abuse and neglect against the plaintiff and the boy’s mother, concluding that the plaintiff’s use of corporal punishment put the boy “at risk of physical hurt/harm, and is not ac*389ceptable.”4 Her report acknowledged that the boy was known to embellish facts, and concluded that the existence of temporary marks left by the spankings was, therefore, only “possible.” Furthermore, although noting the boy’s medical condition, Ugol stated that it was “unclear” whether this condition created any heightened risk of physical harm from the spankings. The department’s northeast region clinical review team reviewed and upheld Ugol’s decision, and the department offered the parents access to counselling services on a voluntary basis. When the parents declined this offer, the department simply closed the case and took no further action.5
b. Administrative and judicial review. Pursuant to its regulations, the department held an administrative hearing at the plaintiff’s request to review Ugol’s decision. See 110 Code Mass. Regs. § 10.06 (8) (1994).6 At this hearing, Ugol testified that she had not found that the boy had actually suffered any bruising or swelling as a result of this punishment, but agreed that her decision to support the abuse report was based on her conclusion that “hitting a child with an object, in this case a belt, puts a child at substantial risk of serious physical injury” or creates the potential for soft tissue swelling and skin bruising. She further testified that her decision was not predicated on any heightened risk of injury created by the boy’s medical condition, which was uncertain, but on her assessment of the risk of injury arising from the nature of the corporal punishment itself.
The hearing officer issued detailed findings summarizing the *390evidence we have reviewed above. On the basis of this record, she concluded that Ugol’s decision was in conformity with the department’s policies and regulations, stating, “[I]t is reasonable to believe that hitting a child with an object in the manner described puts him at substantial risk of physical injury, such as skin bruising or soft tissue swelling. This constitutes physical abuse as defined by Department regulations.”
The department’s decision was subsequently affirmed on appeal by a Superior Court judge, who ruled that there was “substantial evidence” in the record to warrant the decision. The judge concluded that any interference with the plaintiff’s religious and parental rights resulting from the department’s decision was justified by the Commonwealth’s interest in protecting a minor child from harm.
2. Discussion.
a. Standard of review. We may set aside the decision of an administrative agency if it is not supported by substantial evidence. See G. L. c. 30A, § 14 (7) (e); Massachusetts Mun. Wholesale Elec. Co. v. Energy Facilities Siting Council, 411 Mass. 183, 199 (1991). “Substantial evidence,” as defined by statute, is “such evidence as a reasonable mind might accept as adequate to support a conclusion.” G. L. c. 30A, § 1 (6). In conducting this review, we must “give due weight to the experience, technical competence and specialized knowledge of the agency, as well as to the discretionary authority conferred upon it,” G. L. c. 30A, § 14 (7), and should defer to the agency on questions of fact and reasonable inferences drawn from the record. Flint v. Commissioner of Pub. Welfare, 412 Mass. 416, 420 (1992). Significantly, however, that the record may contain some evidence from which a rational mind might draw an inference in support of the agency’s decision does not dispose of our inquiry. New Boston Garden Corp. v. Assessors of Boston, 383 Mass. 456, 466 (1981). Rather, to determine whether an agency’s decision is supported by substantial evidence, we examine the entirety of the administrative record and take into account whatever in the record fairly detracts from the supporting evidence’s weight. See id. See also Daniels v. Board of Registration in Medicine, 418 Mass. 380, 385-386 (1994).
b. Substantiality of the evidence. The substantial evidence standard is thus fairly characterized as a test of rational probability: an agency’s conclusion will fail judicial scrutiny if “the evidence points to no felt or appreciable probability of the *391conclusion or points to an overwhelming probability of the contrary.” New Boston Garden Corp. v. Assessors of Boston, supra, quoting L.L. Jaffe, Judicial Control of Administrative Action 598 (1965). Thus conceived, the substantial evidence test accords an appropriate degree of judicial deference to administrative decisions, ensuring that an agency’s judgment on questions of fact will enjoy the benefit of the doubt in close cases, but requiring reversal by a reviewing court if the cumulative weight of the evidence tends substantially toward opposite inferences. See, e.g., Daniels v. Board of Registration in Medicine, supra at 386, quoting Arthurs v. Board of Registration in Medicine, 383 Mass. 299, 304 (1981) (“as long as there is substantial evidence to support the findings of the agency, we will not substitute our views as to the facts” [emphasis added]).
We do not judge this present case to be a close one. The record contains no affirmative evidence that the boy ever suffered actual “soft tissue swelling or skin bruising” as a result of the plaintiff’s spankings.7 The department argues, however, that its decision to support the 51A report was predicated on there being reasonable cause to believe that the plaintiff’s method of disciplining the boy with a belt created a “substantial risk” that the boy would suffer “soft-tissue injury or swelling,” which “substantial risk” of injury is encompassed by the statutory and regulatory definition of “[a]buse.” See G. L. c. 119, § 51 A; 110 Code Mass. Regs. § 2.00.
We are not persuaded that the factual record lends the weight of probability to the department’s conclusion that a substantial risk of harm was present. The department’s investigator did not observe any physical injuries or marks attributable to the plaintiff’s spankings, nor did the boy’s treating physicians — one of whom, Dr. Erlich, had examined the boy every three to four months from birth — report ever having seen any indicia of inflicted harm. The only evidence pertinent to the physical *392effects of the punishment are the boy’s own statements that the spankings left temporary red or pink marks on his buttocks that would fade after ten minutes or so. Such effects do not by themselves justify a conclusion that the boy is at substantial risk of suffering the sort of injury which the regulations denote as “abuse.” 110 Code Mass. Regs. § 2.00 (1996).
The department argues, however, that its decision to support the 51A report is rendered reasonable by the totality of the circumstances in this case. It cites the following factors as supportive of its conclusion that a substantial risk of harm was present: the regularity of the corporal punishment; the boy’s age and special medical condition; the statement by the boy’s pediatrician, Dr. Solomon, that the use of a belt “wouldn’t help” the boy’s muscular condition; the force the plaintiff used in striking the boy; the plaintiff’s disregard of the boy’s physical well-being, as evidenced by his never having checked to see whether he caused any marks; and the lack of any indication by the plaintiff that he would discontinue his practice of disciplining the boy with a belt.
However, the evidence concerning each of these cited factors is at best inconclusive, and its cumulative weight falls short of the threshold of substantiality required to affirm the department’s decision. We have already noted that the department’s investigator stated in her report, and testified at the hearing, that her decision to support the 51A report was not predicated on any additional risk of harm arising from the boy’s medical condition, as the import of this condition was unclear. We note, further, that Dr. Solomon, although he expressed concern about potential harmful effects of the punishment on the boy’s medical condition, claimed never to have observed any overt signs of abuse on the boy’s body, never identified precisely what sort of harmful effects he had in mind, and voiced his concerns in response to questions from the investigator that (as far as the record discloses) did not accurately describe the nature of the corporal punishment but merely referred loosely to the plaintiff’s “using a belt to discipline” the boy, and the boy’s being “hit with a belt.” Moreover, Dr. Solomon’s concerns were not corroborated by Dr. Erlich, the boy’s orthopedist, who, when asked whether the plaintiff’s use of a belt to discipline the boy posed any particular risks to the boy’s condition, responded that he *393did not “see anything there at all.”8 Dr. Erlich further remarked on the plaintiff’s “devot[ion]” to ensuring that the boy received proper medical treatment and physical therapy, which we view as evidence of the plaintiff’s genuine concern for his son’s physical well-being.
As to the allegation that the plaintiff spanked the boy with great frequency and excessive force, the only pertinent evidence in the record is the plaintiff’s uncontroverted statements that he had spanked the boy five or six times during the preceding seven-month period, the investigator’s report that the plaintiff’s demonstration on a sofa cushion of the force he used on the boy made “a solid smack,” and the boy’s own report of temporary red or pink marks on his buttocks. This evidence, without more, does not make reasonably probable the department’s conclusion that “soft tissue swelling or skin bruising” was substantially likely to result from the plaintiff’s method of corporal punishment.
With respect to the department’s argument that the plaintiff’s unwillingness to abandon his method of corporal punishment is a factor tending to support its conclusion that the boy is at “substantial risk” of physical harm, we note that there is no affirmative statement by the plaintiff in the record regarding his future intentions. Furthermore, we must remark on what seems to us as an anomaly in the department’s handling of this case. When the parents declined to participate in counselling on a voluntary basis, the department closed the case. The department’s regulations require closure of a supported case when a family that is the subject of a supported 51A report refuses further services and there are no grounds for either legal action or a new 51A report. See 110 Code Mass. Regs. §§ 9.02, 9.04 (1994). The commentary to § 9.04 provides that a decision to *394“support and close” is especially appropriate where a supported report of abuse “does not necessarily mean that a child is at ongoing risk” of suffering future abuse. 110 Code Mass. Regs. § 9.04.9 Thus the department’s decision to close the case — a decision which, if we are guided by the department’s own commentary to its regulations, must have been based on the determination that there is no ongoing risk that the child will suffer further abuse — surely undercuts the reasonableness of its prior decision that the boy was at substantial risk of suffering harm in the future if the plaintiff persisted in his method of corporal punishment. Indeed, the department’s willingness to close this case further confirms that its decision to support the abuse report in the first place was not founded on rational inferences drawn from the factual record.
The department argues, however, that it is obligated to support a report of abuse if there is “reasonable cause” to believe that abuse has occurred, and points out that this evidentiary threshold, as defined by regulation and case law, is a minimal one requiring a “relatively low degree of accuracy.”10 But where, as in this case, there is no substantiated claim of actual physical injury, a finding of abuse must be predicated on there being reasonable cause to believe that there is a substantial risk that such injury will occur; i.e., there must be “a collection of facts, knowledge or observations which tend to support or are consistent with the allegations” that a substantial risk of injury is present. See 110 Code Mass. Regs. § 4.32. We conclude that the record does not support a rational inference that a substantial risk of physical injury, as defined by regulation, was present and, therefore, that the department lacked reasonable cause to believe that it was.
Finally, the department urges us to defer to the clinical experience and trained judgment of its social worker and not to disturb its findings on questions of fact. However, we have repeatedly held that, while an agency is free to evaluate the evidence in the record in light of its expertise, it cannot rely on this expertise as a substitute for substantial evidence to support its decisions. See Daniels v. Board of Registration of Medicine, 418 Mass. 380, *395389 (1994), and cases cited. Moreover, the principle of judicial deference to agency judgments on factual issues does not require us to abdicate our responsibility, pursuant to the State Administrative Procedure Act, to review the sufficiency of the factual record. See G. L. c. 30A, § 14 (7) (e), (/). Judicial deference to an agency’s adjudicatory determinations is founded on a recognition of the important role of the administrative agency in the governmental process and a proper respect for the Legislature’s decision to empower an agency with regulatory and discretionary authority. See A. Celia, Administrative Law and Practice § 1576 (1986). But this rationale for judicial deference ceases to apply where, as in this case, we conclude that the agency has failed to adhere to its own statutory mandate and regulatory framework by making a decision without sufficient evidentiary support. In such cases, we are required by the State Administrative Procedure Act to correct the agency’s judgment by means of our own. See G. L. c. 30A, § 14 (7).
3. Conclusion. The department’s regulations, promulgated pursuant to authority expressly granted by the Legislature, see G. L. c. 119, § 51B (8), clearly draw a line between permissible physical discipline and prohibited abuse, specifying the types of physical injuries which may not be inflicted on children and, consistent with the statute, defining abuse as nonaccidental conduct that actually inflicts these injuries or creates the substantial risk that they will result. See 110 Code Mass. Regs. § 2.00. Today, we conclude only that, on the totality of the record presented in this case, the effects of the plaintiff’s physical discipline on his minor child did not satisfy the department’s own regulatory definitions of physical injury and abuse. However, a method of corporal punishment similar to the plaintiff’s could, in different circumstances, rise to a level of severity that would result in the actual infliction of impermissible injuries or, alternatively, warrant a rational inference that it posed a substantial risk that such injuries would result. In these circumstances, this conduct would, at the least, justify the department to support a 51A report of abuse.
The department’s decision to support a 51A report of abuse in this case was not supported by substantial evidence on the administrative record, and must be set aside. See G. L. c. 30A, § 14 (7). Because we resolve this case under the State Administrative Procedure Act, we do not reach or express an opinion on the plaintiff’s constitutional claims.
*396The judgment is vacated. The case is remanded to the Superior Court where a judgment will be entered vacating the department’s decision and ordering the department to notify the plaintiff and any other person, public or private, to whom it conveyed information of its decision, that the 51A report of abuse on his minor son has not been supported.
So ordered.
It is apparent from the transcripts of the administrative hearing that all parties mistakenly applied the definition of “abuse” in G. L. c. 119, § 51 A, as amended through St. 1992, c. 115, § 1, and accompanying regulations, which required a showing of “serious physical or emotional injury” (an error that was repeated in the department’s hearing decision and the Superior Court’s *386memorandum of decision). As the department correctly points out in its brief, in 1993 the Legislature deleted the reference to “serious” injury and expanded the definition of “abuse” to encompass conduct “which causes harm or substantial risk of harm to a child’s health or welfare.” G. L. c. 119, § 51 A, as amended through St. 1993, c. 50, § 23. This discrepancy would ordinarily result in a remand of the matter to the agency for a new hearing under the proper statutory standard. Because we conclude that the factual record in this case is not sufficient to support a finding of abuse under the correct, more expansive definition, we decide the case on the merits.
Persons belonging to a statutorily defined class of mandated reporters have an affirmative obligation to report to the department when they have reason to believe that a minor “is suffering physical or emotional injury resulting from abuse inflicted upon him which causes harm or substantial risk of harm to the child’s health or welfare.” G. L. c. 119, § 51 A. Regulations promulgated by the department pursuant to its enforcement authority, G. L. c. 119, § 5IB (8), define “[a]buse,” in pertinent part, as “the non-accidental commission of any act by a caretaker upon a child under age 18 which causes, or creates a substantial risk of physical or emotional injury . . . .” 110 Code Mass. Regs. § 2.00. “Physical [ijnjury” is further defined as, inter alla, “soft tissue swelling or skin bruising depending on such factors as the child’s age, circumstances under which the injury occurred, and the number and location of bruises.” Id.
On receiving a 51A report of possible abuse or neglect, the department is required to investigate and, if it has “reasonable cause” to believe the report, to “support” it and take remedial action, either offering social services, placing the child in protective custody, or referring the matter to the district attorney for criminal prosecution, depending on the severity of the abuse. See *387G.L. c. 119, § 51B; 110 Code Mass. Regs. § 4.32 (2) (1996). “Reasonable Cause to believe” that abuse has occurred means “a collection of facts, knowledge or observations which tend to support or are consistent with the allegations.” 110 Code Mass. Regs. § 4.32 (2) (1996). See Care & Protection of Robert, 408 Mass. 52, 63 (1990) (“reasonable cause” serves a “threshold function” and means “known or suspected instances of child abuse and neglect”).
A decision to “support” a report means only that the department has reason to believe that an incident of child abuse or neglect has occurred and that some caretaker is responsible; it does not constitute a finding with regard to the identity of the perpetrator. 110 Code Mass. Regs. § 4.32 (2). Somewhat incongruously, however, the regulations further provide that any parent of a subject child or “any caretaker who has been identified in the Department’s records as the person believed to be responsible for the abuse or neglect” has a right to a hearing to challenge the department’s decision to support the report. 110 Code Mass. Regs. § 10.06 (8) (1994).
The department’s finding of neglect pertained to the boy’s mother, and was predicated on her permitting the boy to visit the plaintiff’s home despite knowing about his method of corporal punishment. Because the boy’s mother did not appeal from this finding, we do not address it further.
Testimony given at the administrative hearing by the department’s area program manager indicated that the department was prepared to take steps to close the case, but that this had not yet been done. The department’s records indicate this action was taken later that same day, June 12, 1997.
Following administrative review of a challenged decision to support a 51A report, the hearing officer must determine whether the decision conformed to the department’s policies and regulations and, if it did not, whether it resulted in substantial prejudice to an aggrieved party. 110 Code Mass. Regs. §§ 10.05, 10.06 (8) (1994). The challenged decision may be reversed only if there is no “reasonable basis” in the factual record to support it. 110 Code Mass. Regs. § 10.05. The aggrieved party bears the burden of proving by a preponderance of the evidence that a reversal is warranted. 110 Code Mass. Regs. § 10.23 (1993).
The department states in its brief that the temporary marks left on the boy’s buttocks were “indicative of soft tissue injury” and that the investigating social worker, Rena Ugol, concluded that these marks indicated “the possible presence” of soft tissue damage or swelling. These statements exaggerate the record, however. At the hearing, Ugol testified that she was unable to determine whether there was in fact any physical injury to the boy and that her decision to support was, therefore, based on her judgment that, being hit with a belt “can cause such harm,” creates the “potential” for such harm, and “gives the possibility that there will be the risk of this kind of injury” (emphasis added).
We are mindful that it is for the agency, not the reviewing court, to weigh the credibility of witnesses and resolve factual disputes involving contradictory testimony. See Seagram Distillers Co. v. Alcoholic Beverages Control Comm’n, 401 Mass. 713, 721 (1988). Nevertheless, under the substantial evidence test, we may disregard supporting testimony that cannot reasonably form the basis of impartial, reasoned judgment. See New Boston Garden Corp. v. Assessors of Boston, 383 Mass. 456, 467-468 (1981), citing NLRB v. Pittsburgh S.S. Co., 337 U.S. 656, 660 (1949). We conclude that Dr. Solomon’s expressions of concern were too indefinite and, in light of his own and Dr. Erlich’s statements regarding the absence of any observable injuries, overly speculative to form the basis of a rational inference that the spanking created a substantial risk of physical injury, as defined by statute and regulation.
The second type of case discussed in the commentary, where the family that is the subject of a supported 51A report “has disappeared” and cannot be located by the department, is not relevant here. See 110 Code Mass. Regs. § 9.04 commentary (2) (1994).
See note 3, supra.