dissenting.
I dissent from the panel’s decision to certify because I think the injunction should be vacated and nothing the New York Court of Appeals could decide would alter this outcome. While this case highlights some difficult questions of social policy that the Administration for Children’s Services (“ACS”) must resolve, it does not, in my view, present difficult questions of constitutional law. I would lift the preliminary injunction because the evidence cannot support the district court’s findings underpinning the injunction: that ACS had a policy or practice that violated the Fourth and Fourteenth Amendments with respect to the removal of children from homes plagued by domestic violence.
The district court’s central factual finding that ACS has a policy of “regularly separating battered mothers and children unnecessarily” is simply unsustainable. Nicholson v. Williams, 203 F.Supp.2d 153, 212 (E.D.N.Y.2002).
First, the district court claims that “[t]he consistent policy applied by ACS is to remove children of abused mothers ... solely because the mother has been abused.” Id. at 250. By this, I take the district court to mean that it is ACS policy to carry out removals based only upon past domestic violence and not based on a risk of ongoing harm to the child. Absent a risk of future harm, of course, removing a child from its parents would violate both the Constitution and the Family Court Act. See Tenenbaum v. Williams, 193 F.3d 581, 594-95 (2d Cir.1999); In re H./R. Children, 302 A.D.2d 288, 756 N.Y.S.2d 166, 167 (1st Dep’t 2003). But the evidence does not support a finding that ACS has such a policy.
In February 2001, seventeen months before the effective date of the district court’s preliminary injunction, ACS promulgated revised “Principles For Addressing Domestic Violence in Children’s Services” (“Principles”). Nicholson, 203 F.Supp.2d at 219. Under the Principles, the policy of ACS is clearly stated: It is to remove children only when “domestic violence creates an immediate danger of serious physical harm or serious emotional impairment to a child” and when “safety planning with the non-abusive parent and/or criminal justice intervention does not reasonably provide for children’s safety.” Evidence of ACS practice in the field fails to contradict this explicit policy statement.
Although four of the ten removals described by the representatives of subclass-A appear, in hindsight, to have been unnecessary, see id. at 168-70 (Sharwline Nicholson); id. at 177-79 (Ekaete Udoh); id. at 182-85 (Michele Garcia); id. at 185-86 (Michelle Norris), these cases demonstrate at most incorrect discretionary decisions by ACS caseworkers. They do not suffice to establish an agency policy or practice. To amount to an official policy or practice, the conduct must be “persis*178tent and widespread” and “so manifest as to imply the constructive acquiescence of senior policy-making officials.” Sorlucco v. N.Y. City Police Dep't, 971 F.2d 864, 870-71 (2d Cir.1992). Because ACS deals with between 53,000 and 58,000 cases each year, the handful of individual cases presented by plaintiffs cannot begin to demonstrate the breadth of occurrence that could sustain a finding of policy.
The district court also relied on statistical studies of ACS removals, see Nicholson, 203 F.Supp.2d at 208-09, but the statistics cut the other way. In every case where removal occurs, authority must be obtained from the Family Court, either in advance of or shortly after the removal (except in the case of a settlement). See N.Y. Family Ct. Act §§ 1022, 1024, 1026. A study by New York State shows that in 88.2 percent of cases involving domestic violence ACS did not seek to remove the child. Of the removals in 11.8 percent of the cases studied, other causes for removal existed in all but one case. Nothing in these studies precludes the possibility that the domestic violence was ongoing and, therefore, they cannot support the finding, or even likelihood, that ACS removes children when there is no risk of future harm. Finally, these removals are reviewed by the Family Court, which only permits removals where there is risk of future harm. See In re H./R. Children, 756 N.Y.S.2d at 167.
The second basis for the district court’s conclusion that ACS removes children unnecessarily is that ACS ignores the reality that many removals could be avoided through better provision of services to battered mothers or through removal of the batterer. See Nicholson, 203 F.Supp.2d at 210-213. Again, the Principles are to the contrary.
The Principles state: “When domestic violence creates an immediate danger of serious physical harm or serious emotional impairment to a child, every effort should be made to provide for safety without separating the non-abusive parent and child.” Id. at 219. In its factual determination of ACS policy, the district court discounted the foregoing statement in favor of ACS’s general mission statement, which calls for caseworkers to resolve “[a]ny ambiguity regarding the safety of the child ... in favor of removing the child from harm’s way.” Id. at 218-19. The district court fails to explain adequately why an earlier, general mission statement is to be preferred over the subsequent, more detailed and relevant Principles For Addressing Domestic Violence. Surely the latter is the better indication of ACS’s policy in this area. Moreover, the general mission statement embodies the understandable impulse toward prudence to ensure the safety of minors in the uncertain situation first encountered by the caseworkers; it does not contradict the Principles.
Plaintiffs offer no persuasive evidence that the Principles are not the true ACS policy. The district court points to two pilot projects undertaken by ACS, the Zone A and Zone C pilot projects, that resulted in fewer removals than occur normally. Id. at 206-07. However, the pilot projects’ lower removal rates do not demonstrate that ACS policy is constitutionally deficient, only that improved management and greater care may result in fewer removals. Moreover, the very fact that ACS is engaged in such phot projects is evidence that it recognizes that there may be instances where caseworkers effect unnecessary removals contrary to ACS policy and is taking steps to correct the problem.
The district court also points to the testimony of former Family Court Judge Phillip Segal that “[ojften, ACS would remove the children as a first resort, rather than providing services.” Id. at 214. But, *179such evidence is consistent with an effort on the part of caseworkers, unable to arrange services quickly enough to ensure the child’s immediate safety, to secure the safety of the child first and then arrange for services once the child is safe, subject always to prompt review by the Family Court. The district court’s reliance on the Domestic Violence and Child Maltreatment Project is similarly flawed. See id. at 215-16. The fact that in this project the Family Court almost always returned the children to the mother does not demonstrate that a caseworker was following a policy of unnecessary removal; it is fully consistent with success on the part of ACS and the Family Court in working out services or obtaining a restraining order that reduces the risk of harm.
To recap, the evidence in this case points in one direction. There is no ACS policy — formal, informal or sub rosa — to take children from their parents solely because the parent has been a victim of domestic violence. The most that can be said from the four instances presented at trial is that when such conduct occurs, it arises from a “a specific act of a governmental officer” making a decision in a particular case. County of Sacramento v. Lewis, 523 U.S. 833, 846, 118 S.Ct. 1708, 140 L.Ed.2d 1043 (1998). Thus, to the extent that the injunction was based on the district court’s factual finding that ACS removes children solely because the child has been exposed to domestic violence, it cannot stand.
The district judge also based the injunction on its view that it is wrong to even temporarily remove children from homes marred by domestic violence upon the justification that witnessing such violence would be harmful to the child. As a matter of policy preference, the issue is debatable. Compare plaintiffs’ Subclass A expert Dr. Pelcovitz (noting “universal agreement that a certain percentage of children exposed to domestic violence suffer from a variety of behavior and emotional difficulties” and the need for “a relative cost benefit analysis” when deciding if a child should be removed) and Subclass B expert Professor McAllister (harm to children from witnessing domestic violence includes “both short-term symptoms” (sleep disturbance, separation anxiety, aggression or passivity/withdrawal, distractibility, hy-pervigilance), and longer-term problems (modeling behavior perpetuating relationship violence, “pervasive pessimism or sense of foreshortened future)”) with plaintiffs expert Dr. Stark (“the vast majority of children ... who witness domestic violence, nevertheless test psychologically normal in most studies”). Indeed, the district court recognized “[t]he consensus of experts was that the children can be — but are not necessarily — negatively affected by witnessing domestic violence,” Nicholson, 203 F.Supp.2d at 197; it sided, however, with Dr. Stark’s “lengthy and well-substantiated opinion that children rarely experience long-term effects from witnessing domestic violence.” Id. at 198.
While this policy preference may someday be manifest in the statutory and case-law of New York, it is not today. Family Court Act § 1012(f)(i)(B) states that a child is “neglected” and hence subject to the Act’s removal process provisions when his “physical, mental or emotional condition has been impaired or is in imminent danger of becoming impaired as a result of the failure of his parent ... to exercise a minimum degree of care ... in providing the child with proper supervision or guardianship [or] by unreasonably inflicting ... a substantial risk thereof.” New York courts readily hold a parent to this standard even though she herself may be the victim of domestic violence. See, e.g., In re Michael G., 300 A.D.2d 1144, 752 N.Y.S.2d 772, 772 (4th Dep’t 2002) (“The *180exposure of the child to domestic violence between the parents may form the basis for a finding of neglect.”); In re Nichole SSL, 296 A.D.2d 618, 745 N.Y.S.2d 128, 129 (3d Dep’t 2002) (“A child’s exposure to domestic violence in the home standing alone may form the basis for a finding of neglect .... ”); In re Carlos M., 293 A.D.2d 617, 741 N.Y.S.2d 82, 84 (2d Dep’t 2002) (“Evidence of acts of severe violence between parents in the presence of their children is sufficient to show that the children’s physical, mental, or emotional conditions are in imminent danger of becoming impaired within the meaning of Family Court Act § 1012(f)(i)(B).”); In re H./R. Children, 302 A.D.2d 288, 756 N.Y.S.2d 166, 167 (2003) (noting that a history of domestic violence, failure to take steps to prevent exposure to the domestic violence, and likelihood of continued exposure would demonstrate neglect).
The judicial function to be exercised in this case does not embrace taking sides in the policy debate over the efficacy of temporary child removal in domestic violence cases. The New York legislature has spoken and the Appellate Divisions of New York have uniformly upheld removals in such circumstances. The only question before this court is whether or not such removals violate the Constitution.
As the district court acknowledged, in Tenenbaum we held that an official’s conduct in the course of removing a child must be “shocking, arbitrary and egregious” to violate substantive due process. 193 F.3d at 600; see also County of Sacramento, 523 U.S. at 846, 118 S.Ct. 1708. But the district court applied a different standard after concluding that Tenenb-aum’s test was too lenient to the government and was “motivated by an unwillingness to turn the Fourteenth Amendment into a catch-all cause of action in tort.” Nicholson, 203 F.Supp.2d at 243. Instead, it determined that the appropriate test was that normally applied to challenges to the constitutionality of state statutes, as found in Joyner v. Dumpson, 712 F.2d 770 (2d Cir.1983), not challenges to discretionary executive judgments in individual cases.
Putting aside the question of whether the Joyner test is appropriate when the governmental action is attributable to the executive or an executive agency as opposed to the legislature, the test has no applicability in this case in which the government actions are individual and discretionary. In the typical situation, caseworkers, following an emergency referral, are confronted with a domestic violence scenario, often accompanied by indications of child neglect or drug and alcohol abuse. Acting under Family Court supervision, they perceive a risk to a child of either direct physical harm or psychological trauma and, in some cases (relatively infrequently, it turns out) decide to seek removal. Such conduct easily passes muster under either Tenenbaum and County of Sacramento’s “shocks the conscience” test or, although it is unnecessary to so conclude, even under Joyner once one fairly evaluates what the government actors were doing here.
With respect to Tenenbaum, the idea that these temporary, discretionary child removals shock the judicial conscience— the correct test — is fully answered by the fact that such removals have not been found at all “shocking” to the judges of New York’s Appellate Division who uniformly have upheld them. And if the applicable test were Joyner, I could not find the absence of a compelling state interest in such circumstances: The state interest in protecting children justifies infringing (generally temporarily) on a mother’s interest in being with her child.
*181Relying on Stanley v. Illinois, 405 U.S. 645, 92 S.Ct. 1208, 31 L.Ed.2d 551 (1972), the district court also found that the procedures before the Family Court violate the battered mother’s procedural Due Process rights by “presum[ing]” that the mother is unfit. Nicholson, 203 F.Supp.2d at 237. I disagree that this case involves an unconstitutional presumption similar to that described in Stanley. First, Stanley is simply inapposite because New York provides a judicial hearing for these child removals while in Stanley there was not even an administrative hearing at which a parent (the unmarried father in Stanley) could prove his fitness. Stanley, 405 U.S. at 646, 92 S.Ct. 1208. Second, because the evidence does not support a finding that ACS has a policy of removing children when there is no risk of future harm, there is no basis for a finding that removal is based on a presumption that battered mothers are neglectful parents simply by virtue of having been the victim of domestic violence. Third, I disagree with the majority’s suggestion that, in the absence of specific expert testimony that the particular child is being harmed by witnessing domestic violence, a Family Court judge who orders a removal to protect the child from exposure to domestic violence may be unconstitutionally “presuming” that such harm has occurred or might occur.1 See Maj. Op. supra at 174-176. When a Family Court judge, experienced in these matters, hears evidence of particular instances of domestic violence and then, taking into account the age of the child, determines that witnessing such violence could harm the child and warrants removal, the judge is making an individual decision based on evidence of violence in the home. Stanley is inapposite. That case involved a presumption based solely on an element of the father’s status — the fact that he had not married his child’s mother — that was only tangentially related to his relationship to his child. Cf. Stanley, 405 U.S. at 654, 92 S.Ct. 1208. In contrast, the inference made by the family judge in these cases, that the child may be at risk of harm, is based on the child’s exposure to specific acts of domestic violence.
Finally, some observations about the preliminary injunction are in order. First, even if this case presented a likelihood of success on the merits, the preliminary injunction was not needed to prevent irreparable harm. Certainly the plaintiff classes took that position until fourteen months after the complaint in Nicholson was filed when, without a motion for a preliminary injunction having been made, the district court ordered on its own initiative that the parties should be prepared “to argue and present evidence ... on the question of whether a preliminary injunction should issue, and if so, on its contents.” The court then tried the case sporadically over six months, issued the amended injunction in January 2002 and delayed its effective date to June 22, 2002 to see if the ACS would reform the procedures and practices that the court did not like.
Second, the sweeping preliminary injunction goes considerably beyond ordering remediation of the constitutional harm that the district court found to have occurred. As the Supreme Court pointed out in Lewis v. Casey, “the remedy must of course be limited to the [constitutional] inadequacy that produced the injury in fact that the plaintiff has established.” 518 U.S. 343, 357, 116 S.Ct. 2174, 135 L.Ed.2d 606 (1996); see also Missouri v. Jenkins, 515 U.S. 70, 89, 115 S.Ct. 2038, 132 L.Ed.2d 63 (1995) (“The nature of the desegregation remedy is to be determined *182by the nature and scope of the constitutional violation.”) (quoting Milliken v. Bradley, 433 U.S. 267, 280, 97 S.Ct. 2749, 53 L.Ed.2d 745 (1977)). Accompanied by an opinion that fills sixty-eight pages of the Federal Supplement and reads like a management consultant’s report, the injunction tells ACS not just to stop the perceived violation but in detail how it must reform itself to satisfy the district court. The injunction mandates how ACS can and cannot exercise its discretion in responding to domestic violence that threatens a child; requires that ACS distribute pamphlets to parents and children informing them of their rights before carrying out a removal; forces the ACS to change the way it reports the results of its investigations to the state; tells the ACS what the agency can and cannot plead in the Family Court; orders new case-planning conferences; directs the hiring of domestic violence specialists, establishes a Review Committee to oversee compliance with the court’s order; and raises the compensation for appointed counsel. See In re Nicholson, 181 F.Supp.2d 182, 188-93 (E.D.N.Y.2001).
Wholly absent from the court’s reasoning is respect for the principle that an injunction “must take into account the interests of state and local authorities in managing their own affairs.” Milliken, 433 U.S. at 280-81, 97 S.Ct. 2749; see also Lewis, 518 U.S. at 362 & 363 n. 8, 116 S.Ct. 2174 (1996) (criticizing injunction as being “inordinately — indeed, wildly — intrusive” and for not “giv[ing] the States the first opportunity to correct the errors made in the internal administration of their prisons”); Jenkins, 515 U.S. at 102, 115 S.Ct. 2038 (“On remand, the District Court must bear in mind that its end purpose is not only ‘to remedy the violation’ to the extent practicable, but also ‘to restore state and local authority to the control of a school system that is operating in compliance with the Constitution.’ ”); cf. Turner v. Safley, 482 U.S. 78, 92, 107 S.Ct. 2254, 96 L.Ed.2d 64 (1987) (noting that assessing the safety risk from inter-prison correspondence requires “a judgment ‘peculiarly within [prispn official’s] province and professional expertise’ ”). Rather, the injunction limits the City’s capacity to manage ACS and improperly intrudes on the Family Courts’ supervision of the ACS personnel who appear before it.
Such demanding and specific injunctions that go beyond ordering an agency to remediate violations of law and direct the agency on precisely how to do it frequently have unintended, counterproductive consequences that impair the administration of government agencies. See Ross Sandler & David Schoenbrod, Democracy by Decree 139-61 (2003) (describing the problems which arise when consent decrees lead to judicial management of administrative agencies). Even more importantly, such injunctions incur the cost of removing agency conduct from political accountability to democratically elected officials. It is those officials in a system of democratically elected government who are charged with meeting societal priorities and allocating budget and personnel. Id. The further risk is that by taking over policy functions that properly belong to elected officials or their political appointees, the courts themselves become political. As Professor Donald Horowitz explained in his study of courts and social policy a generation ago, “[t]he danger is that courts, in developing a capacity to improve on the work of other institutions, may become altogether too much like them.” Donald L. Horowitz, The Courts and Social Policy 298 (1977).
One can envision that such intrusive intervention could become necessary to remedy constitutional wrongs in situations involving deliberate and flagrant violations of a court order by a state or municipal *183government bent on defying a constitutional mandate. Such interventions in the past were necessary to vindicate the equal protection rights of African-Americans in the wake of Brown v. Board of Education, 347 U.S. 483, 74 S.Ct. 686, 98 L.Ed. 873 (1954), and related statutory rights such as those under the Voting Rights Act of 1965. In some instances, they have been needed to remediate the depraved conditions in which prisoners have been kept. They have been justified when the defendant officials either do not care whether the command of law is followed or, worse, are committed to frustrating it. Cf. Inmates of D.C. Jail v. Jackson, 158 F.3d 1357, 1358-59 (D.C.Cir.1998) (noting that “Miter years of attempting to get D.C. to voluntarily comply, and appointing a Special Master to coordinate with D.C. in an attempt to alleviate conditions, the district court ordered that the jail’s medical and mental health services be placed in receivership in 1995.”). But ACS is not this sort of agency and, accordingly, this sort of remedial micromanagement is not warranted here. ACS, far from being a persistent law-breaker, is a creature of reform and appears committed to improvement.
ACS was created by Mayor Guiliani’s executive order in January 1996 following the highly publicized torture and killing of six-year old Eliza Izquierdo by her mother after the abuse had been reported to the New York City’s Child Welfare Agency. Defendant Nicholas Scopetta, the first Commissioner of ACS, developed a reform plan for the new agency, created a new management structure and sought input from hundreds of experts and specialists in child welfare.
Reforms followed. Professional standards for caseworkers and supervisors were raised through more rigorous job qualifications, enhanced training and incentives. Since 1996, ACS has hired over 1,500 new caseworkers and doubled the number of field managers employed in Child Protective Services. Training for new caseworkers was expanded from four weeks to ten months and was instituted agency-wide for supervisors and today includes an intensive six-day training course to update staff on investigative procedures. The result has been a highly trained staff and the reduction of caseworker caseload from approximately twenty-seven in June, 1996 to approximately twelve in March 2001. ACS has also implemented family conferences, usually not more than seventy-two hours after a child enters foster care. These conferences allow families and their support networks to discuss the case in a non-adversarial context and formulate a service or safety plan.
While the magnitude of the task (at least 53,000 cases per year) and the need for basic structural and managerial reform preoccupied the new agency in its early years, the agency also began to address domestic violence issues specifically. In brief summary, ACS has filled a new position of Domestic Violence coordinator created in 1996 with a person with strong ties to the advocacy community; followed its 1994 Zone C pilot (in which children were removed in four of the thirty-five cases in which domestic violence was identified) with a 1999 Zone A Pilot project (in which children were removed in six of the 197 cases in which domestic violence was identified (although another removal was found in a later review of 20 case records)); Commissioner Scopetta promulgated guiding ACS policy in the form of the Domestic Violence Principles in February 2001, characterized as “excellent” by Dr. Stark, the expert upon whom the district court placed primary reliance; and in January 2001, ACS established a separate office of Domestic Violence Policy and Planning and a month later convened a subcommittee on domestic violence screening and *184assessment procedures, including a Domestic Violence Protocol completed by-caseworkers in each case involving domestic violence, to incorporate the contents and philosophy of the Principles. One result is that the Domestic Violence Protocol clearly states that “[t]he preferred way to enhance children’s safety after the detection of domestic violence in their homes is to support and help the victim protect herself and her children, while engaging batterers in services and holding them accountable for their actions.”
Add to the foregoing: the development of twelve Clinical Consultation teams from private providers comprised of specialists in domestic violence, substance abuse and mental health to advise on specific cases; two major initiatives to improve preventative service performance; considerable ongoing, specific training in domestic violence within the caseworker’s core and ongoing training curriculum; and revisions to Family Court petitions to clarify that “engaging in domestic violence” should not be used “if the case history and evidence indicate that an individual has acted solely in self-defense.”
This brief review of ACS initiatives and reforms should clarify that the district court was not faced with an agency resembling the Departments of Education of Alabama or Mississippi in the 1960s. Rather it indicates the appropriateness of deference to a municipal agency committed to reform and, despite the district court’s conclusions to the contrary in its management review, seemingly quite capable of managing its responsibilities and dealing with the subtle and myriad complexities of policy choice within its demanding, politically charged environment.
The district court has no doubt uncovered important and difficult policy questions that ACS must grapple with: Under what circumstances should a child be removed, even temporarily, from a household plagued by domestic violence where the mother is not the instigator and where the effect on the child is that of witnessing domestic violence. But this question does not present a significant issue of constitutional import. Exposure to domestic violence poses enough risk of harm to a child that the Constitution does not bar a state from carrying out a removal to protect against that harm. Accordingly, I would immediately lift this ill-advised injunction and would not certify to the New York Court of Appeals because no decision by that court on the questions certified should alter this outcome.
I respectfully dissent.
. It is also not apparent from the record that ACS generally fails to provide testimony in Family Court regarding the risk of harm to the child.