dog with poop emojis

ICP v Lincoln Properties and its “progeny.”

In the last days of the Trump administration HUD promulgated a new regulation concerning disparate impact claims under the Fair Housing Act whose intent was very clearly to make such claims difficult even to plead, let alone prove.¹ Its implementation remains stayed by a federal court, but in the Fifth Circuit that may not matter because earlier Fifth Circuit cases are even more restrictive. Prompted by the discussion in Treece v. Perrier Condo. Owners Assn., Inc., 2021 WL 533720 (E.D. La. Feb. 12, 2021) I decided to take a hard look at the Fifth Circuit’s decision in Inclusive Communities Project, Inc. v. Lincoln Prop. Co., 920 F.3d 890, 895 (5th Cir. 2019), cert. denied, 140 S. Ct. 2506 (2020). That turned out to be a bigger project than I expected, because as interpreted Lincoln Properties is incredibly restrictive, but those interpretations are almost certainly wrong. Clarification seems inevitable even with the existing conservative Supreme Court. Here’s why.

Slicing and dicing disparate impact.

The idea of disparate impact discrimination is simple. A rule or policy that seems on its face to be neutral with respect to race, religion and other categories is discriminatory if it has a disparate impact on housing choice for a protected group (like members of a religion, Blacks, disabled individuals etc.) Whether this discrimination is illegal depends on whether there is a legitimate reason for the rule. No legitimate reason means unlawful discrimination. A sufficiently good reason means no there is no unlawful discrimination. discrimination. Huntington Branch, N.A.A.C.P. v. Town of Huntington, 844 F.2d 926, 933 (2d Cir. 1988), aff’d in part sub nom. Town of Huntington, N.Y. v. Huntington Branch, N.A.A.C.P., 488 U.S. 15 (1988). Note the key elements:

  1. A rule or statute or policy
  2. That when applied has a disparate impact on some protected group
  3. In a way that affects housing choice
  4. Without any good reason for the rule, statute of policy
  5. That cannot be accomplished in a non-discriminatory way.

It’s important to remember all five in any analysis of a case, something that isn’t always clear from the written decisions.

A classic and easy to understand example of disparate impact is landlord policies that exclude tenants with arrest or conviction records. On its face such a policy applies equally to everyone. In practice it has a disparate impact on Blacks because Blacks are much more likely than Whites to be arrested. If a landlord turns away everyone with an arrest record it will turn away a disproportionate number of Blacks which will disproportionately affect their housing choice. In addition, because Blacks are disproportionately turned away, the concentration of Whites in the apartments will continually increase, making the apartments more and more segregated. If you take the policy away then Blacks and Whites should be rejected at rates proportional to their existence in the area and the apartments will continue to reflect the racial makeup of the area. At this point it worth noting there are two bad effects caused by the one policy. It limits housing choice for Blacks and it tends to create segregation. Either should be enough for the “no arrest” policy to violate the Fair Housing Act unless there is some good reason for the policy that can’t be achieved in a different way.

At this point we have to note that there are two kinds of “disparity” at play. There is the disparity in treatment of Blacks, and there is the disparity in housing choice that results from the application of the “no arrest” policy. The first disparity combines with the policy to cause of the second. We will see that confusion about which disparity is part of the cause and which is the effect is critical in later cases.

To understand the idea of a good reason consider conviction records. Blacks are not just more likely to be arrested; they are also more likely to be convicted in most of the United States. If the “no arrest” rule were replaced by a “no conviction” rule the end result would be the same, less housing choice for Blacks and the increasing segregation of the apartments.

There is however a difference between the two policies. People who are arrested are presumed innocent; it is not until they are convicted that they can be called criminals. Rejecting people with arrest records causes rejection of the innocent with the guilty, for which there is really no justification. Rejecting people with conviction records, on the other hand, always rejects the guilty. And because some crimes do reflect a danger to other people and property, rejecting the guilty makes sense. Yes, the “no conviction” policy has a disparate impact that limits housing choice and leads to segregation, but it serves the landlord’s legitimate interest in keeping the apartments safe in a way that makes sense.

But does a “no conviction” rule really achieve a legitimate goal in the least discriminatory way possible? Lots of crimes don’t point to a danger of some kind to the apartment community. There’s no reason to think someone convicted of check kiting is going to do harm to fellow residents. And the older a crime gets the less likely it is that the convicted criminal will repeat the crime or one like it. A twenty year old conviction for armed robbery doesn’t mean a lot if the person who committed the crime has never committed another. Most authorities recommend that a “no conviction” policy be more subtle; no conviction for certain crimes that are not too old, depending on the seriousness of the crime.

So far we have talked about policies that affect housing choice in one apartment complex, but the same general principles can apply to a city zoning ordinance. It may seem neutral but have a disproportionate effect on housing choice. The difference is the geographic scope of the effect, something that becomes important when talking about proof of disparate impact. This doesn’t mean there is a clear distinction between public and private policies. A private policy can have a large geographic effect if the private actor’s influence covers a wide area.

Speaking of proof, which is the area of contention, notice that in the discussion above I’ve been talking about what a policy will do, not what it has already done. Disparate impact claims can come in two flavors, claims about a new policy and what it is likely to do or claims about an old policy and what it has already done. The phrase from many, many cases that captures this difference is “actual or predictable effect.” We will see that part of the confusion comes from not explicitly recognizing the difference between a claim that tries to link an existing policy to an existing effect (like the racial makeup of an apartment complex not matching the racial makeup of the local community) and a claim that predicts a new policy will lead to that kind of effect. This distinction is particularly important when you consider the difference between a disparate impact on housing choice for members of a protected class and the segregative effect of a policy that could, in theory at least, occur even if housing choice was not disparately impacted.

To keep things straight then here’s a summary of considerations important to disparate impact claims under the FHA. First, does it meet the requirements of a disparate impact claim?

  1. A rule or statute or policy
  2. That when applied has a disparate impact on some protected group
  3. In a way that affects housing choice
  4. Without any good reason for the rule, statute of policy
  5. That cannot be accomplished in a non-discriminatory way.

Second, is the impact on housing choice for members of the protected class or a segregatory effect or both? Third, are we talking about an existing actual effect or a predictable future effect. And finally, where is the effect going to take place; a single apartment complex or a neighborhood or city or state or even the entire United States.

The Supreme Court speaks – Texas Dept. of Hous. and Community Affairs v. Inclusive Communities Project, Inc., 576 U.S. 519, 135 S.Ct. 2507 (2015).

Everything I’ve said above could be found in cases of various kinds stretching back for decades and in HUD regulations from as early as 2013, but it was not until 2015 that the Supreme Court answered a fundamental question about the Fair Housing Act. Is disparate impact discrimination one of the kinds of discrimination the FHA outlaws? The answer isn’t obvious because disparate impact discrimination may not be intentional. The premise of disparate impact is a rule or policy that looks even-handed and there may be no reason to think those who adopted it even knew it would have a disparate impact. It was at least possible that the FHA only outlawed intentional discrimination.

The Supreme Court thought the FHA did reach disparate impact discrimination. This was not too surprising since disparate impact discrimination had already been recognized by the Supreme Court in other contexts. However, the Court didn’t stop there. It embarked on discussion of how disparate impact could be proved; that is, how do you prove a particular policy was the cause of some observed or likely disparate effect. The Court summarized its considerations by imposing a “robust causality” requirement:

In a similar vein, a disparate-impact claim that relies on a statistical disparity must fail if the plaintiff cannot point to a defendant’s policy or policies causing that disparity. A robust causality requirement ensures that “[r]acial imbalance … does not, without more, establish a prima facie case of disparate impact” and thus protects defendants from being held liable for racial disparities they did not create.

576 U.S. 519, 542 (2015).

We should note a couple of key points here. First, the Supreme Court refers to a particular kind of proof – statistical proof – that may not be necessary in every case. Second, its rule is really a somewhat wordy version of the axiom that correlation is not proof of causation. “Robust causality” is really just “causality,” because by itself correlation is never proof of causation. For disparate impact it is a necessary, but not a sufficient condition.³ This last point is important because so much discussion after ICP (which is how I’ll refer to this Supreme Court case) turns on a false belief that “robust causality” must be more than plain old causality.²

The Fifth Circuit listens. Inclusive Communities Project, Inc. v. Lincoln Prop. Co., 920 F.3d 890, 895 (5th Cir. 2019), cert. denied, 140 S. Ct. 2506 (2020).

By 2019, when the Fifth Circuit took up a new disparate impact case, there was already a small body of case law interpreting robust causality. (I’ll call this the Lincoln Properties case because Inclusive Communities litigates quite a bit, and their name isn’t a good way to distinguish one case from another.) The decision was unusually controversial within the Fifth Circuit. The original three judge panel split 2-1 on the outcome, and when the loser (“ICP”) asked for the full Fifth Circuit to review the decision there was a 9-7 split in the decision to uphold the decision. The dissent claimed the majority had destroyed disparate impact discrimination claims, a claim that was wrong but infected later interpretations of Lincoln Properties.

Lincoln Properties, like many landlords, does not accept rent vouchers from HUD or other agencies. This means that the recipients of rent vouchers are unable to rent from Lincoln Properties if they need the voucher to afford the rent, which they almost invariably do. Not surprisingly, given the economic disparities associated with race and ethnic distinctions, rent vouchers are used disproportionately by Black renters. It is equally unsurprising that apartments that accept such vouchers are mostly subsidized projects required to accept them, and those subsidized projects are mostly located in predominantly minority neighborhoods. The effect of refusing the vouchers, according to ICP, is that minority renters are forced back into minority neighborhoods, which limits housing choice and perpetuates and increases segregation.

Note that there are three kinds of disparity present in this case. First, there is the disparity between the race and ethnicity of voucher holders, who are disproportionately Black, and the racial makeup of population of Dallas which is mostly White. Second, there is the disparity between the ratio of Blacks to Whites excluded by the no voucher policy (high) and the ratio of Black to White renters in general (much lower). Third, there is the possible disparity between the racial makeup of tenants at the Lincoln Properties apartments and the racial makeup of Dallas. I say “possible” because there didn’t seem to be evidence of the racial makeup of the various apartment complexes.

There is also segregation associated with vouchers. Voucher accepting apartments and voucher holders are predominantly located in minority neighborhoods, and since voucher holders are predominantly Black they are segregated from Whites, who are predominantly located in neighborhoods where vouchers are not accepted. ICP’s claim was that Lincoln’s “no vouchers” policy limited housing choice by disparately excluding Blacks from the White areas in which the policy applied, and that this caused the resulting racial segregation.

Analyzing this claim in light of the “robust causality” requirement from ICP, the Fifth Circuit identified what it called four different approaches in earlier cases:

  • A requirement that an artificial, arbitrary, and unnecessary’ policy caused the problematic disparity based on an Eighth Circuit decision,
  • A requirement that statistical disparities must be sufficiently substantial that they raise an inference of causation based on a Fourth Circuit decision,
  • A requirement that the statistical disparity supposedly caused by the policy not be explained by already existing disparities based on the dissent in the Fourth Circuit decision,
  • A requirement that there be a causal connection between the statistical disparity and the policy or law based on an Eleventh Circuit decision.

These are not really four different approaches to causation. The first, which requires an “artificial, arbitrary and unnecessary policy” is not looking at causation, but at justification, which is a separate inquiry. The second and third really just look a general rule and its specific application. The general rule is that an inference of causation must be reasonable. The specific application is that when the resulting disparity can be is explained by pre-existing disparities it isn’t reasonable to infer a causal relationship. The last just restates the basic requirement of a causal connection without addressing how it can be proved.

What we really have here is just one rule that the label “robust causality” doesn’t really change. Disparate impact liability exists where there is a correlation between a policy and a disparity and there is evidence to support the inference that there is causation as well as correlation. And, with a nod to the Eighth Circuit, when there is no justification for the policy.

ICP produced evidence of a correlation between no voucher policies and racial segregation and a similar correlation with the number of voucher holders in a given area. In several small census blocks the presence of no voucher apartment complexes was matched by disproportionately few Blacks and voucher holders. In others where voucher accepting apartments were present there were proportionately more Blacks and a larger number of voucher holders. The Fifth Circuit rejected this evidence as being sufficient to show a causal relationship between no-voucher policies and segregated neighborhoods with the observation that:

[This position] would effectively mandate a landlord’s participation in the voucher program any time the racial makeup of multi-family rental complex does not match the demographics of a nearby metropolitan area.

920 F.3d 890, 909. Just like the Supreme Court’s pronouncement that “robust causation” is required in FHA cases this statement adds nothing that has not been present in disparate impact cases from the beginning. Correlation is not proof of causation. It is a necessary but not a sufficient condition.

A trip to the heartland of disparate impact discrimination

If the Fifth Circuit’s language above were accepted as is the decision would not be noteworthy. For “robust causality” to require more than mere correlation is only logic. However, a later Fifth Circuit decision interprets Lincoln Property to mean that “robust causation” requires

either: “a change in the defendant’s enforcement of [a] policy” caused a disparate impact; or a challenged policy “caused [the relevant minority group] to be the dominant group” of those affected by the policy.

Inclusive Communities Project, Inc. v. Heartland Community Assn., Inc., 824 Fed. Appx. 210, 217 (5th Cir. 2020)(unpublished). These are regarded as the two exclusive means to prove disparate impact discrimination.

Under this interpretation Lincoln Properties would kill disparate impact liability. As applied in Heartland this means that a new policy can never sustain a disparate impact claim because there is no change in policy. The second way to show robust causality is impossibly circular. Disparate impact occurs when, as a result of an already existing disparity some policy has a disparate impact on some protected group. Take, for example, the no arrest policy. There is an existing disparity in the arrest of Blacks – they are arrested at disproportionately high rates. A “no arrest” leasing policy does not create that disparity, but when combined with that disparity it has a discriminatory effect on Blacks. If Lincoln Properties stands for the idea that a policy must cause the disparity that results in the disparate impact from application of the policy then it is as completely and absurdly wrong as the dissenting Judges in Lincoln Properties suggest.

However, the words quoted in Heartland can only be understood in context . The “change in enforcement” evidence of causation discussed in Lincoln Properties is the Fifth Circuit’s explanation for why a different Court found robust causation in a case that involved a change in enforcement. It is an example of evidence of causation, not one of two exclusive kinds of evidence. The “caused the relevant group to be the dominant group” language comes from an extended discussion that is unique to ICP’s evidence of a three way correlation between a racial disparity in who held vouchers, racial disparity in particular geographic areas and the existence of “no voucher” policies that made it frankly difficult to determine just what discriminatory effect the no voucher policy supposedly had. The Court first eliminates the possibility that the discriminatory effect was on who held vouchers, writing that there was no evidence the ““no vouchers” policy, or any change therein, caused black persons to be the dominant group of voucher holders.” This is where the “caused the relevant group” words originated. In the very next sentence the Court eliminates a second possible discriminatory effect; that is, that the no voucher policy caused segregation. This was in fact ICP’s claim; that is, that the no vouchers policy caused segregation. The problem, as the Court observes in the next paragraph, is that:

it is entirely speculative whether the “no vouchers” policy, as opposed to some other factor, not attributable to Defendants-Appellees, caused there to be less minority habitation in individual census tracts after the policy was implemented.

In other words, you haven’t provided evidence of causation beyond correlation. This is followed by the sentence I quoted above:

Without that information, any landlord who did not accept vouchers would be vulnerable to a disparate impact challenge any time a less than statistically proportionate minority population lived in that landlord’s census tract.

The message in Lincoln Properties is that correlation is not proof of causation. It only tells us what will not work as evidence of causation, it does not in any way limit what will.

Is the Fifth Circuit a murderer, or just misunderstood?

As interpreted in Heartland the decision in Lincoln Properties may well mark the end of disparate impact discrimination in the Fifth Circuit. Heartland concerned a policy that was transparently intended to limit the number of black residents in a community by making it a violation of the community rules to accept rent vouchers at a time when all the voucher users were Black. There was no complicated theory that tried to link an existing policy implemented by one apartment owner to existing segregation across an entire city. It was a straightforward application of the “no arrest” theory. If all the voucher holders are Black then a policy of not allowing vouchers will predictably have a disproportionate impact on Blacks and predictably lead to a segregated community. It was a case, to use the Supreme Court’s phrase from ICP, “at the heartland of disparate impact liability.”(4)

This leads to the most striking feature of Heartland and the decision of the trial court. There  is no discussion of the predictable future discriminatory impact of the challenged policy. The entire discussion is about whether the policy caused existing conditions. Many if not most of the decisions in disparate impact cases, including those that the Supreme Court’s decision in ICP approves, were cases concerning the future impact of a policy, its “predictable” discriminatory effect rather than its actual effect. Why the Fifth Circuit and perhaps the litigants in these cases, ignored the “actually or predictably” language that appears in so many decisions is a mystery, but that mystery proves, I think, that Heartland and other cases interpreting Lincoln Properties fail to thoroughly consider what ICP really stands for. “Robust causality” is just “causality” and the requirement that there be proof of causation beyond mere correlation is a well established feature of disparate impact law.  It isn’t that complicated, and surely the courts will soon figure that out.


¹ See my blog at

² Although disparate impact discrimination always starts with a correlation between some policy or rule and a discriminatory effect I found almost no discussion of the difference between correlation and causation in disparate impact cases under the FHA. The difference is mentioned in Jones v. City of Faribault,  2021 WL 638039 (D. Minn. Feb. 18, 2021) as part of a discussion of statistical evidence, as well as in the District Court opinion in  City of Miami v. Bank of Am. Corp., 13-24506-CIV, 2014 WL 3362348, at *5 (S.D. Fla. July 9, 2014) and El Paso Apt. Ass’n v. City of El Paso, 2008 WL 2641350, at *2 (W.D. Tex. June 24, 2008). This is surprising since the difference between causation and correlation is explicitly recognized in many other kinds of litigation that involve statistical evidence [see, e.g. Christian v. Generation Mortg. Co.,  WL 4494860, (N.D. Ill. Sept. 12, 2014) and the cases cited in note 5] and N. Carolina State Conf. of the NAACP v. McCrory, 182 F. Supp. 3d 320, 415 (M.D.N.C. 2016), rev’d and remanded sub nom. N. Carolina State Conf. of NAACP v. McCrory, 831 F.3d 204 (4th Cir. 2016).

³ Another distinction from the world of logic that isn’t explicitly recognized in FHA disparate impact cases.

(4) “Suits targeting unlawful zoning laws and other housing restrictions that unfairly exclude minorities from certain neighborhoods without sufficient justification are at the heartland of disparate-impact liability.” Texas Dept. of Hous. and Community Affairs v. Inclusive Communities Project, Inc., 576 U.S. 519, 521 (2015).