Causation or correlation: the chimera in section 11 of the Insurance Act 2015

Abstract

Prior to the Insurance Act 2015, an insurer could refuse payment by relying on terms unrelated to the manner in which the loss occurred, or with the assured's default. Section 11 of the Insurance Act 2015 reverses that and requires ‘the punishment to fit the crime’ by introducing a ‘could have increased the risk’ test, which is said to be a requirement for correlation between an assured's non-compliance with a risk clause and the actual occurrence of the loss. The new test effectively raises more profound causation issues from a practical point of view as regards insurance recoveries and also from a theoretical point of view of causation in the law. It is submitted that the test ‘could have increased the risk’ introduced by section 11 of the Insurance Act 2015 should be interpreted by an approach to recognising a nexus – a general causal relevance (general causation) – and adopting interpretation as a restricting tool, to achieve alignment with the legislative intent. In presenting this argument, this paper explores the difference between correlation and causation in the law, expounding a wider understanding of legal causation than causa proxima in the insurance context.

Introduction

It is widely recognised that the rules by which insurance liability is attributed are mainly those contained in the insurance contract. The legal nature of an insurance contract means that the assured is allowed to recover from the insurer only loss that is caused by an event covered by the insuring clause of the insurance contract and not subject to any exclusion. The inquiry as to causation is therefore a crucial issue for ascertaining whether some loss or damage is covered by an insurance policy. Where insurance is a tool to transfer risk between an assured and an insurer, the insurer agrees to bear legal responsibility to recover a loss or damage according to the agreement and the law of insurance. Where there is a failure to contractually transfer the risk to the insurer, the assured will itself assume the risk and loss. ‘Insurance liability’ will be used throughout the paper to refer to the insurer's duty to indemnify the assured in accordance with the insurance policy and the rules of insurance contract law.

Risk is the very subject-matter of insurance contracts; risk definition and risk mitigation are essential to pricing the transfer of risk. There are therefore clear and important reasons for determining whether non-compliance with the insurance terms could have increased the probability of the risk of loss and the occurrence of the loss in fact. The Insurance Act 2015 contains various remedies pursuant to an assured's default, such as the rules relating to the duty of fair presentation of risks at precontractual stage, non-compliance with policy terms and fraudulent claims. As Davey has maintained, these rules matter because all (or at least some) of the insurance liability can be discharged,¹ and the risks and losses effectively shift back to the assured. The limitations to insurers’ liability on account of the fault of those who are assured do not necessarily link to the question of what is the proximate cause of loss in insurance claims, but ask whether insurers will remain liable, if those assured commit wrongful acts during the formation or performance of insurance contracts.

The Insurance Act 2015 now requires ‘the punishment to fit the crime’, but does so in a manner that, it is submitted, raises profound causation issues. Section 11 of the Insurance Act 2015 provides that if the assured shows that non-compliance with a term tending to reduce some particular risk (as listed under section 11(1)) could not have increased the risk of the loss which actually occurred in the circumstances in which it occurred, the insurer may not rely on the non-compliance to exclude, limit or discharge its liability under the contract for the loss. In this paper, out of convenience of expression, terms falling within the scope of section 11 will be referred to as ‘terms tending to reduce a particular risk’, or more broadly ‘risk clauses’.

Section 11 is said to avowedly avoid a causation requirement mainly based upon the Law Commissions Explanatory Notes,² which appears to be an attempt to introduce a correlation requirement between an assured's non-compliance with a risk clause and the actual occurrence of the loss. However, there has been much uncertainty surrounding how to interpret this correlation requirement. As will be shown, it is a widely held view both in practice and academic literature that there is still a need to explore causation issues in section 11 of the Insurance Act 2015. In this paper, it is argued that the correlation or non-causation requirement is unclear in the law and should not be the correct approach or interpretation when the provision is applied in the future. The aims of this research are twofold: clarity for the purpose of insurance claims and recoveries; and, more importantly, conceptual consistency and coherence in the theoretical edifice of causation in insurance law.

Drawing on the philosophical and general legal notions of causation, notably general causation theories, this paper explores the nature of such correlation requirement in section 11 of the Insurance Act 2015. The thrust of the following analysis may be succinctly embodied as the distinction between two legal questions or tests: (1) ‘could the breach have increased the possibility of materialisation of the risk which in fact caused the loss?’; and (2) ‘is the fact that the assured behaved in some undue risk-creating way causally relevant to the loss, so that the loss is not covered by insurance?’. The former question is taken as an encapsulation of what the court is required to ask when applying section 11 of the Insurance Act 2015, whereas the latter is a classic ex-post causal inquiry, the same as the proximate cause as provided in section 55 of the Marine Insurance Act 1906.

The discussion consists of three parts. First, the meanings of correlation and causation in law are considered. Laying down the groundwork for the later discussion, this part demonstrates a wider meaning of causation in the law than causa proxima in the insurance context. This aims to put insurance causation in the central discussion of causation in the law by first refuting the idea that causation theory in insurance law is confined only to the purpose of identifying the proximate cause of the loss. The second part analyses the language of the rules relating to promissory warranties and ‘terms tending to reduce risks’ under the Insurance Act 2015. It interrogates stated and unstated reasons why the legislator did not adopt causal expressions and tests. Finally, the third part draws on academic literature in seeking to define a correlation requirement in the form of a ‘risk increasing’ test. The test of ‘some bearing’ is compared to the effect of a requirement of causal connection both in theoretical accounts and in practice in the approaches of the law of Australia and New Zealand.

The conclusion will be reached that section 11 of the Insurance Act 2015 should be viewed as having laid down a test of correlation with a general causal relevance (general causation) for the determination of insurance liability, and that this approach is distinct from a spurious correlation as well as the more direct and singular causal approach adopted by Australia and New Zealand.

1. What do correlation and causation mean in insurance contract law?

It is important to first unpack our notions of causation in the law. Hart and Honoré suggested that causal notions should be given the meaning attributed to them by an ‘ordinary man’, resulting in a series of varying notions.³ They identified three causal concepts central to legal causal notions. The first central concept is that of cause and effect, such as ‘causing harm’ by one's own act. This is the most common type of situation in which an ordinary man speaks of cause. A second type of causal concept found in ordinary speech is that of X ‘providing reasons’ for Y to do something – analogous to inducement, as in where X induces Y to act, which causes the harm. A third class of causal concept is where X provides Y with the opportunity or means necessary to do something. For example, where X carelessly leaves the door of his friend's house (or his) unlocked, thereby providing the opportunity for Y, a thief, to enter the house.⁴

In Stapleton's view, different information resulting from these different interrogations is often expressed in the same causal terms.⁵ Stapleton's central claim is that in legal contexts, we should choose an interrogation underlying causal usage in the law that captures all the ways in which the factor might be involved in the existence of the particular phenomenon in issue, rather than on the basis of ‘intuition’ or ‘common sense’.⁶ Stapleton therefore introduces the alternative concept of ‘involvement’ in order to avoid circular causal terminology:⁷

[Involvement] provides the width of coverage that is needed to accommodate smoothly all the many diverse enquiries the Law makes. For example, in the hunters case⁸ one project of the Law might be to consider all possible regulatory strategies for preventing such deaths. Such a project requires the Law to address all involved factors – even those that some might describe as ‘mere conditions,’ such as the walker's presence. This is because the most efficient strategy may be, for example, to ban mountain walking during the hunting season.⁹

Moore also poses the question: ‘when lawyers have at least tried to use “cause” as the label for a natural relation that must exist for liability to be found, what notion of cause is implicit in their usage?’¹⁰ He suggests that, ‘[t]here are some prunings to be done if we are to have a chance of extracting any coherent concept of causation from the thousands of its usages in the law’.¹¹ Based upon his ‘pruned’ concept of causation presupposed by the law, he concludes that causation in the law will have a number of characteristics, one of the foremost of which is that mere correlation is not causation.¹² He characterises correlation as commonsensical and a temporal sequence (no matter how universal and necessary) which may not be a causal sequence, and not peculiar to the law's use of cause.¹³

According to Brett, in the fields of philosophy and science, the notion of cause-and-effect has almost, if not entirely, vanished.¹⁴ It has been replaced by concepts of correlation, or of permanent and invariable association, between apparently independent sets of occurrences.¹⁵ As Honoré also explained, the focus of philosophers when studying causal connection is on causation as a relation between events or between facts; whereas the causal inquiry in the legal context is specific. The law requires a claimant to show that the element that makes the conduct wrongful or creates the undue risk was relevant to the harmful outcome for which the law provides a remedy.¹⁶ As Stapleton puts it, ‘it is by multiple precisely characterised experiments that scientists achieve an understanding of how things are in the world, and then express this understanding in terms of physical laws of nature’.¹⁷ She goes on to characterise the legal project as always focused on whether a specified factor played a role in the existence of the particular phenomenon.¹⁸

In the field of philosophy, general causation and actual causation are often distinguished: general causal claims, for example, ‘smoking causes lung cancer’, typically do not refer to particular individuals, places, or times, but only to event-types or properties. Singular causal claims, such as ‘A's heavy smoking during the 2000s caused her to develop lung cancer’, refer to particular individuals, places, and times.¹⁹ As Wright has argued, general causation is not merely a statement of observed statistical correlation.²⁰ He explains that the essence of general causation is the belief that we attach to the generalisation: the belief in its causal or lawlike character – that the antecedent conditions produce or cause the subsequent event.²¹

In general, correlation is even wider than the scope of general causation. As Moore observes, correlation can be a purely temporal sequence which may not be causal in nature.²² In insurance, correlation is typically present in the actuarial exercise of insurance pricing. When assessing an assured's individual risk level for the purpose of setting the insurance premium, actuarial professionals have long understood that insurance pricing is mostly correlative, not causal. Insurance mechanisms are intrinsically collective, as they are built on the pooling of risks.²³ For example, it may be that higher premiums for people with low credit scores are justified because they are both risky in their financial lives and take risks that lead to insured losses. There may well be purely statistical or actuarial evidence as proof of risk assessment and identification, but no direct causal nexus in this fact.

Insurance has been characterised as the transformation of unknown individual uncertainty, or chance, into a measurable, aggregate risk.²⁴ Paefgen et al describe the pricing process as follows:

In order to differentiate the risk of insurance policies, actuaries use a set of rate factors to separate policies into groups (i.e., tariff classes). The construction of tariff classes is ultimately a clustering task. Each tariff class corresponds to a certain combination of rate factor categories or intervals in the case of continuous rate factors. For each tariff class, actuaries analyse historical claims data to arrive at a reliable estimate of the corresponding pure premium, that is, the minimum required payment per policy to cover the expected losses from its class.²⁵

In the underwriting process, correlation in an ex ante situation relies upon the classificatory approach of insurance and the initial assumption that all members of a class are identical risks prior to some actual loss. Individualisation²⁶ is achieved by measuring correlated factors on an individual basis; thus ‘black box’ insurance (also known as telematics insurance) uses a device to measure and record the insured driver's performance, such as vehicle speed, location, distance travelled, driving frequency, and time of day the car is in motion, how hard the driver applies the brakes, how rapidly the car accelerates, and how sharply the driver takes a corner. The data is converted into a score, which the insurance company can use to set a personalised premium rate for the driver. However, how sharply the insured driver takes a corner is not always causally relevant to a collision that has actually occurred. The causal factor (if any) is swept along in a wide range of correlations and proxies in the pricing matrix in the process of generalisation of risks.

An ex post judicial decision on an insurance claim is different from an assessment of risks ex ante. In legal responsibility, causal notions matter because they are more specific, individualised and rooted in the ordinary person's ideas of when it is fair to punish or seek compensation. When tracing the connection between a given event and the loss or damage, many factors may be considered relevant but not causally relevant in the legal sense. For example, the birth itself of the wrongdoer or the injured party are correlated to the occurrence of the injury, perhaps in both philosophical and scientific terms; but they will not be considered in the determination of legal responsibility as a causal factor in the law. Another example is seen in Malcolm v Dickson,²⁷ a case about remoteness of damage in a negligence claim, where Lord Birnam stated: ‘It is of course logically possible, as every schoolboy knows, to trace the loss of a battle, or even of a kingdom, to … the absence of a nail in a horse's shoe. But strict logic does not appear to me to be a safe guide in the decision of questions such as this’.²⁸ However, legal responsibility in general need not always require a causal connection between the responsible party's conduct and the harm.²⁹ As Moore rightly points out, ‘a common fault of the legal literature on causation has been its credulity with regard to the law's demands on the concept of causation’;³⁰ he argues that ‘the law has mixed too many extraneous elements into what it calls causation for there to be much hope for any metaphysical translation’.³¹ Hart and Honoré analysed the grounds for legal responsibility by reference to three criteria: conduct, causation and fault.³² In establishing certain legal responsibility, these three criteria may be combined in various ways.³³ They considered insurance indemnity as an example of undercutting the notion of fault and causation, in the sense of not requiring that the insurer's conduct or fault causing or occasioning harm be shown; nonetheless, the causal issue that usually arises and is necessary to decide in the insurance context is whether a given loss is caused by something insured or uninsured. ³⁴

Causation and the risk transfer in insurance law have traditionally been viewed as idiosyncratic compared to criminal law or the law of tort. The distinguishing factor is that liability doctrines in criminal law and tort typically require that the defendant has caused harm to the claimant.³⁵ As Moore explains, causation enters into both the prohibitions and the requirements of a typical criminal code and tort law, for such rules either prohibit citizens from causing certain results or require them to cause certain results. In either case, causation is central to liability.³⁶ This contrasts with insurance law, which provides for the payment of a sum of money to an insured for a loss or detriment suffered upon the occurrence of specified events. As encapsulated in LCA Marrickville Pty Ltd v Swiss Re International SE,

In the law of tort and in the criminal law [causation] is concerned with the attribution of personal liability or culpability for a personal act. In both, the boundaries are set by public policy. In insurance, the principles of causation, or at least the concept of proximate cause, are concerned with whether an insurer is obliged to provide indemnity pursuant to the bargain between it and the insured …. it is apt to keep in mind that, it is by the policy's terms that the insurer stakes its money against an actuarially assessed risk that an insured peril will be the proximate cause of the insured's loss. This requires a different causal analysis from that applied in relation to the assessment of a person's culpability for an act … [T]he causation requirement is narrower and has the function of determining whether the insurer must indemnify the insured allowing for any agreed limits placed on the transfer of risk from the insured to the insurer…³⁷

It is submitted that insurance law has narrowed the scope of the causal inquiry to the doctrine of proximate cause which is central to the determination of insurance liability. Section 55(1) of the Marine Insurance Act 1906 embeds the concept of proximate cause as follows:

Subject to the provisions of this Act, and unless the policy otherwise provides, the insurer is liable for any loss proximately caused by a peril insured against, but, subject as aforesaid, he or she is not liable for any loss which is not proximately caused by a peril insured against.

In Hart and Honoré's work, and in much academic literature on causation, ‘proximate cause’ refers to a ‘legal cause’, as opposed to factual causation, and is the legal principle for determining liability in various areas of law.³⁸ This usage differs from the concept of proximate cause in the context of insurance law. In English insurance contract law, the proximate cause is the efficient cause to which the event can be ascribed.³⁹

In section 55(1), ‘the provisions of this Act’ refers to the provisions, non-compliance with which will either affect the validity of the policy, or a single claim. As will be discussed, most of these provisions, if not all, avowedly disregard a causal connection between the assured's breach and the actual occurrence of the loss. Therefore, ‘proximate cause’ is – in the insurance context – a criterion limited to explaining how the harm came about and designating the most efficient cause of the loss.

It is accepted that the schism of causation between cause-based liability in tort, criminal law and in insurance exists because of public policies and purposes behind legal doctrines.⁴⁰ However, this paper will proceed from the assumption that it is not a necessary schism to isolate the analysis on causation in insurance contracts from the central debate surrounding the common question regarding the roles of causation in ascertaining and truncating legal liability. Given how insurance law has sat incongruously separately from other areas of law – and legal theory – over the years, it is submitted that insurance contract law is wrongly confined within the notion that insurance causation is merely about seeking the proximate cause of loss, which thus leads to the unnecessary schism from other areas of law. Instead, we should adopt a more comprehensive causal framework from general theories of law, for instance, in the analysis of section 11 of the Insurance Act 2015.

The following propositions may thus be adopted for the purpose of the discussion that follows. First, correlation and causation are distinct connections. Causal notions are of obvious importance to the determination of legal liability in criminal law, the law of tort and contract law, as well as in insurance law, because causal notions respond effectively to an individualised case of risk transfer and liability ascription. This may be contrasted with disciplines that rely on ex ante and generalised explanations. However, it is not the case that causal notions are essential in the determination of all kinds of legal responsibility and in every rule relating to legal liability. It is accepted and worth emphasising that some legal liability does not have causal notions at its foundation.

2. Breach of promissory warranties and risk clauses

Having established in Part 1 that causation is not mere correlation and that there is a wider meaning of causation recognised in case law and legal literature than causa proxima in the insurance context, in Part 2 we will consider the language of the rules relating to promissory warranties and terms tending to reduce risks under the Insurance Act 2015. It will address semantic uncertainty surrounding section 11 of the Insurance Act 2015, including how to interpret the non-causation requirement of breach of warranties.

Armgardt invokes ‘normatively ideal worlds’ as a tool to determine liability, especially where a causal connection is required.⁴¹ A world is normatively ideal if and only if all agents involved in the causal scenario act according to their legal duties.⁴² A violation of a rule affects an actual event e if and only if e does not occur in the normatively ideal world of the corresponding causal scenario.

Consider the case of a house burning down giving rise to an insurance claim: in ‘normatively ideal worlds’, the house burns down fortuitously, for example, because of lightning, a short circuit, without the assured's fault of failing to manage the risks as agreed in the insurance contracts. In such a ‘normatively ideal world’, insurance claims mainly call for an explanation of how the harm came about and to ascertain the proximate cause of loss for the purpose of insurance indemnity. However, when the assured's conduct has breached a risk clause, it will raise an assessment of the insurance responsibility for the assured's act or omission in accordance with insurance terms and the law. Such assessment is not always required by law to consult our causal notions, nor does it require a causal connection to establish a given consequence. This involves a judgement on how non-compliance did affect the actual circumstances under which the loss occurred. The question can be whether the assured's behaviour caused the loss, or whether the risk of the loss occurring has increased as a result of the assured's non-compliance.

Section 33(1) of the Marine Insurance 1906 defines a promissory warranty as ‘a warranty by which the assured undertakes that some particular thing shall or shall not be done, or that some condition shall be fulfilled, or whereby he affirms or negatives the existence of a particular state of facts’. Promissory warranties (hereinafter warranties) historically have a different role to play compared with the link that is required between a peril and loss. Insurance law previously attached to breaches of promissory warranties a legal effect that disregarded any causal connection. In De Hahn v Hartley,⁴³ Lord Mansfield held that there was a breach of warranty to sail with a complement of 50 hands, when the ship in fact sailed with less and made up the deficiency later. Strict compliance with the warranty is required, and it makes no difference that the non-compliance may not have caused any actual commercial prejudice to the insurer's position, or increased the risk that the insurer was to take.⁴⁴ As Merkin explains,

Warranties in their origin were designed to describe and delimit the risk that insurers were prepared to run. If the risk as described to insurers was not that which they actually faced, it seemed right for them to treat themselves as discharged from future liability. However, the original conception has been abused, and by the middle of the twentieth century it had become common practice for insurers to demand warranties of all manner of matters, many of which would have had no or little impact on the underwriting decision….⁴⁵

The law provided no opportunity for the assured to remedy the non-compliance or to make any argument that its failure was causally irrelevant to the losses.⁴⁶ As for the remedy, prior to the Insurance Act 2015, the breach of a warranty would discharge the insurer from all further liability.⁴⁷

In general, methods of interpretation of warranties in effect avoid the causal question.⁴⁸ This, according to Hart and Honoré, is a perfectly intelligible way of making a rule of law.⁴⁹ The strict compliance rule for insurance warranties is distinguishable from strict liability in tort, as in cases falling under the principle in Rylands v Fletcher.⁵⁰ Once it is proved that the defendant has accumulated a noxious substance on the defendant's land, the defendant is liable if the substance escapes and causes harm, whether or not he could have prevented this escape. As Honoré points out, strict liability is liability not for wrongful conduct, but for engaging in risk-creating activity; there would have been no need to show that the act was causally relevant.⁵¹ However, Honoré suggests that strict liability in tort concerns conduct that is made by the law a basis of liability provided it is causally connected with the harm suffered.⁵² This causal connection, according to Honoré, is different from a cause in an explanatory inquiry.⁵³ In contrast, the strict compliance rule for warranties seems to have completely avoided any sort of causal inquiry. Under the original section 33(3) of the Marine Insurance Act 1906, it was immaterial whether the loss would have still occurred had the warranty not been breached. This is one of the main criticisms of the rules of promissory warranties leading to the reform of the law embodied in sections 10 and 11 of the Insurance Act 2015.

In 2007, in the course of the Insurance Law Reform Project, the Law Commissions proposed the introduction of a causal connection test to the effect of breaching a promissory warranty. This proposal was later rejected on the grounds that a causation test would not be appropriate for all warranties.⁵⁴ The main reason is that a past claim does not cause (or even contribute to) a future claim, but it may be highly relevant to the insurer's assessment of the likelihood of future claims. Similarly, the fact that an employee has past criminal convictions does not cause future misdemeanours, but it is a highly relevant consideration in the underwriting process.⁵⁵

It can be seen that the meaning of warranties is twofold. On an individualised level, many forms of breach of warranty are in fact causally related to the occurrence of a given loss. On a generalised level, the inserting of some warranty may be based only upon the insurer's assessment of the risk in the decision-making process, essentially the same as in the pricing matrix mentioned above. The requirement or fact specified in a warranty is for classifying the assured into groups and calculating the likelihood of losses. The risk of non-compliance bears no causal relationship to any claim ex post, but signifies that contracting with an individual assured is more likely to entail exposure to a type of or a particular risk.

Part 3 of the Insurance Act 2015 (sections 9–11), which is headed ‘Warranties and other terms’, is designed primarily to reform the draconian features of promissory warranties. Section 10 provides that the remedy of automatic discharge from liability is abolished and replaced by a suspensive effect. Section 10(2) stipulates that an insurer has no liability under a contract of insurance in respect of any loss occurring, or attributable to something happening, after a warranty (express or implied) in the contract has been breached but before the breach has been remedied. The sub-section aims to lay down and emphasise the suspensive effect of a breach of warranty during the breach which is in replacement of the automatic termination of the policy entirely from the moment of breach in the old regime.

At first glance, some may insist that section 10 has introduced a causal nexus in terms of the remedies of breaching a warranty because of the causal expression ‘attributable to’, so that the legal remedies and liability should be settled with reference to the causal criteria. However, as Merkin and Gürses rightly suggest, ‘section 10 does not impose a causation test, requiring the breach to be causative of the loss. It merely provides that the insurer cannot rely upon a breach of warranty after it has been cured, and says nothing about a loss which occurs in the period of breach’.⁵⁶ Indeed it is clarified in the Explanatory Notes that ‘[t]he “attributable to something happening” wording is intended to cater for the situation in which loss arises as a result of an event which occurred during the period of suspension, but is not actually suffered until after the breach has been remedied’.⁵⁷ The true interpretation appears to be that, where a loss occurs once the breach has been cured, but the cause of the loss occurred before it was cured, the insurer will not be liable for such loss. In such circumstances, the loss is simply a belated result while the contract has been suspended because of the breach of a warranty. This scenario is to be treated the same as a loss which occurred immediately before the breach is remedied.

Section 10 only replaces the harsh remedy of automatic termination of the insurance policy with the effect of suspending the insurance policy. It does not change the strict compliance rule in the sense that the rule still does not require any inquiry into whether the breach is relevant, causally or in other ways, to the actual occurrence of the loss. This is, in part, left to section 11.

The role of section 11 in the overall scheme is to cure potential problems arising from the absence of a causal inquiry between non-compliance and a loss. Section 11 is entitled ‘Terms not relevant to the actual loss’. Section 11 has limited application to a few specified types of warranties. It may apply to terms other than warranties, as it applies to all terms, other than those which ‘define the risk as a whole’, if compliance with the term would tend to reduce the risk of loss of a particular kind, loss at a particular location, or loss at a particular time.⁵⁸ It is immaterial whether a term is interpreted as a warranty or some other type of term such as a suspensive condition, so long as it satisfies the description in section 11(1).

The Explanatory Notes to the Insurance Act 2015 state that ‘[i]n the event of non-compliance with such a term [under section 11], it is intended that the insurer should not be able to rely on that non-compliance to escape liability unless the non-compliance could potentially have had some bearing on the risk of the loss which actually occurred’⁵⁹ (emphasis added). The Explanatory Notes go on to specify that ‘a direct causal link between the breach and the ultimate loss is not required’.⁶⁰ The Law Commissions explained that

The consciously objective element is intended to allow the court to look at what the effect of compliance might generally be. Importantly, it does not introduce a causal element about whether compliance would have prevented the loss, or whether the breach caused or contributed to it. It is simply whether compliance might usually be thought to reduce the chances of the particular type of loss being suffered.⁶¹ (emphasis added)

In the Consultation Paper from which the quote is drawn, three models were considered for reform of the law of warranties in general insurance law, namely the Law Commission Report of 1980,⁶² the Australian Insurance Contracts Act 1984 and the New Zealand model respectively. The 1980 Report recommended that an assured should be able to challenge the insurer's refusal to indemnify in the absence of links between the breach and the loss. The link suggested in that Report focuses on whether the breach is first ‘material to the risk’ and second ‘increased the risk’.⁶³ However, the 1980 Report noted that a solution depending on the presence of a causal connection between the breach and the loss was inappropriate considering the nature of a warranty, which is to prevent the risk or risks from unexpected increase by clarifying and confining the basis or scope of the contract.⁶⁴

By contrast, an overtly causal limitation on the effect of warranties can be found in Australian insurance law.⁶⁵ For non-marine insurance cases,⁶⁶ section 54 of the Insurance Contracts Act 1984 applies and provides that an insurer has a defence only when the assured's acts or omissions, typically in failing to comply with contractual obligations, have caused or contributed to the loss.⁶⁷ It has been suggested that provisions which relate to the risk itself will generally be caught by section 54(2),⁶⁸ whereas section 11(1) of the 2015 Act confines its application to risk clauses relating to a particular kind, particular time or particular location. In the Australian approach, if the clause is a risk clause, its enforcement rests upon causation.⁶⁹ In contrast, where the non-compliance does not concern a particular risk, the remedy for non-compliance rests upon prejudice to the insurer – a proportional recovery of a single claim depending on the extent to which the insurer's interests were prejudiced as a result of such non-compliance.⁷⁰

Section 11 of the New Zealand Insurance Contracts Act 1977⁷¹ also provides that an assured will not be disentitled to an indemnity when non-compliance with a term is not caused or contributed to by the occurrence of non-compliance. However, it retains the ‘all or nothing’ approach to all non-compliance. Both Acts have created the requirement for a causal link between the assured's act or omission and the loss suffered by the assured, which is not confined to breach of a promissory warranty.

Returning to the UK Insurance Act 2015, there has been scholarly criticism of the causal test adopted by Australia and New Zealand.⁷² Both Law Commissions and the Explanatory Notes to the Insurance Act 2015 have made it clear that a causal connection is rejected for both breaching a promissory warranty and ‘a term tending to reduce certain risk’. Consequently, the only causal inquiry in insurance cases remains the causa proxima as discussed in Part 1. By deliberately avoiding causal language in setting out the rules, the legislator also has fastidiously drawn the distinction between causation and correlation (‘some bearing’), adopting the latter for the purposes of section 11.

As the authors of Arnould explain, this section is not a causation enquiry; however, the need to explore causation issues has not been wholly eliminated and this is a widely held view.⁷³ For example, as mentioned above, Merkin and Gürses believe that this will be a causation enquiry.⁷⁴ Colinvaux reaches a similar conclusion, that some element of causation is necessarily imported into section 11(3), and refers to the dissenting view of Lord Mance that the section would not eliminate causation and of Lloyd's Market Association that it would introduce causation by the back door, in evidence presented on the reformulated clause 11 of the Bill.⁷⁵ The editors of MacGillivray are also not entirely persuaded that section 11(3) does in fact dispense with arguments about causation.⁷⁶ Naidoo suggests that while the wording of section 11(3) may not mean causation in its usual sense, it does indicate some other kind of connection and that is where section 11(3) is unclear.⁷⁷

Given the uncertainty, it in turn raises the question, if an orthodox causal approach as adopted in Australian and New Zealand law is not the rule of English law, where precisely is this elusive line between the Explanatory Notes’ ‘some bearing’ test and causation?

3. Causal relation or correlation?

So far, we have seen three models regulating an assured's breach of a risk-related term: a model free of causal inquiries;⁷⁸ a test of ‘would have increased’ the tendency of some risk and loss;⁷⁹ and a model requiring an actual causal connection.⁸⁰ This part will analyse and distinguish the three models further, reasoning from theoretical and practical perspectives.

We may set out the relation between an assured's non-compliance with a risk clause and the occurrence of a loss on a scale:

• • no required nexus including causal ones (eg section 10 breaching a promissory warranty);

• • a hypothetical and generalised correlation in an ex ante assessment of likelihood. As Moore suggests (above n 10), universal correlation without ‘pruning’ (or in other words, contextualising) cannot be identified as causation;

• • general causation not associated with particular circumstances, such as ‘smoking causes lung cancer’; and

• • actual causation, an ex post and specific investigation of cause-and-effect. The proximate cause in insurance cases (the test of efficiency) falls into the category of actual causation.

Philosophers and legal theorists have considered the phenomenon of ‘actual causation’. This concerns the assignment of causal responsibility for an event, based on how events actually play out.⁸¹ ‘Probabilistic causation’, on the other hand, designates a group of theories that aim to characterise the relationship between cause and effect by focusing on how causes change the probabilities of their effects under certain circumstances.⁸² For example, suppose that A and B each throw a rock at a bottle, and that each has a certain probability of hitting and breaking it. As it happens, A's rock hits the bottle, and B's does not. Evaluating this event, we would say that A's throw caused the bottle to shatter, while B's did not. Nonetheless, B's throw tended to increase the probability that the bottle would shatter, mathematically speaking; but it did not actually cause the bottle to shatter in the particular circumstances.⁸³

Wright has correctly suggested that the concept of causal explanation has been confused with predictive statements based either on general causation or on mere statistical frequency; but the probability judgments that underlie each of these types of statements are fundamentally different.⁸⁴ Only causal explanations (typically actual causation) supply the attributive element⁸⁵ that is essential for systems of liability based on individual responsibility. Mere statements of increased probability, even where based on general causation, are by themselves insufficient to establish that attributive element.⁸⁶

Now we shall apply the extended notion of causation and correlation as shown so far to explore the nature of the ‘could have increased risk’ test as provided by section 11 of the Insurance Act 2015. The key uncertainty or confusion surrounding the (non) causation requirement stems from, as the authors of MacGillivray have rightly pointed out, the wording of section 11(3), which has both prospective and retrospective elements.⁸⁷ The provision says that the assured can recover by showing that non-compliance could not have increased the risk of the loss which actually occurred in the circumstances in which it occurred. Moreover, according to the Law Commissions, the intent is simply whether compliance might usually be thought to reduce the chances of the particular type of loss being suffered, rather than for an attributive or preventive purpose.⁸⁸

In brief, section 11(3) in itself highlights two difficulties in the philosophical notion of causation: (1) the division of ex ante risk assessment into a general sense and an ex post judgment on the cause of loss in a specific scenario; (2) a legal judgment based upon the actual influence of an assured's non-compliance or a mathematical analysis of hypothetical cases.

First, section 11(3) is distinct from two tests aiming at causal explanations: one such non-compliance did (not) cause the loss itself; the other such non-compliance did (not) increase the risk of loss. The former is clearly rejected with the comparisons with the solutions adopted by New Zealand and Australia. The significance of such a distinction in the law is evidenced by New Zealand Insurance Co Ltd v Harris.⁸⁹ Here, a policy insuring a tractor excluded loss while the tractor was ‘let out on hire’. The tractor was destroyed by fire while it was on hire to a third party. The Court of Appeal of New Zealand applied section 11 of the New Zealand Insurance Contracts Act 1977, which adopts a causal approach. The court unpacked the provision into two steps. The first step was to determine whether the insurer's liability had been so defined because the happening of the events or the existence of the circumstances was in the view of the insurer likely to increase the risk of occurrence of the loss. This would assist the insurer in retaining a degree of control and to ensure that the contemplated use of the tractor was met by an appropriate premium. The second step was to ask whether the loss was caused or contributed to by the happening of the events or the existence of the circumstances.⁹⁰ However, the court went on to hold that

In construing a statutory provision of this kind designed for practical application on a day to day basis, refined analysis in terms of metaphysical enquiries into causation should be eschewed. Again, a simplistic ‘but for’ approach would rob s.11 of much of its efficacy and deny its application in examples given by the Contracts and Commercial Law Reform Committee as requiring statutory reform (see para. 29). Rather it is a matter of determining, under a section concerned with exclusion from cover where the limitation has been included because the event or circumstance is likely to increase the risk of loss occurring, whether the loss actually sustained by the insured was caused or contributed to by the relevant event or circumstance. If the existence of the relevant circumstances did not in itself increase the risk of loss, there is no justification either in principle as the Contracts and Commercial Law Reform Committee emphasised, or under the statutory language, for denying the insured the protection of the cover.⁹¹

Secondly, in Harris, ‘would have increased the risk’ of loss appears capable of two meanings: one an assessment of likelihood and the other an actual causal test – ‘did increase the risk of the loss’. These two meanings require further discussion.

The first meaning of ‘would have increased the risk’ is a generalised and possibly non-causal question in the law, and is the step-one question identified in Harris. It assesses whether non-compliance fails to control the risk to a level commensurate with the premium. The second meaning, ‘did increase the risk’, is specific to the circumstances and the proximate cause of the loss, and is a causal test. In section 11(3) of the Insurance Act 2015, the expression ‘could not have increased the risk of the loss which actually occurred in the circumstances in which it occurred’ is avowedly designed to sever links with causation. The focus is said to be on looking forward from when the risk was underwritten (the breach could not have increased the risk of the loss), as opposed to looking backwards from the actual circumstances of the claim (whether the breach contributed to the actual loss).⁹² Therefore, it appears that the ‘could have increased the risk’ test is the same as the generalised, non-causal question as in step one in Harris. Accordingly, instead of showing that non-compliance has no bearing on a risk causally relevant to the loss, either a necessary cause or the proximate cause, the assured should prove that at the time of premium pricing. For example, all lessees of tractors are assumed not to be fire hazards on a generalised level for all fire claims as the fact in Harris. Such an assumption is objective and would normally be tested by actuarial science and verified or disproved, which is not necessarily within the scope of causation in the law.

As Merkin and Gürses argue, the expression used in section 11(3) makes it necessary to look at what actually happened, because the statutory reference to the circumstances in which the loss occurred poses the question of why the loss actually happened.⁹³ Had section 11(3) simply stated that there is recovery only if the assured's breach ‘could not have increased the risk of the loss’, the Law Commissions’ intentions might conceivably have been fulfilled.⁹⁴ They further use the examples given by the Law Commissions to support the argument. One is,

The insured warrants that the insured vehicle will be roadworthy. This is breached because the left front headlight is defective. The vehicle skids on black ice in the dark. Although the faulty headlight did not cause the accident, it is possible that it could have contributed, given the circumstances of the loss (that is, that it was dark, and a headlight was defective). The insurer does not have to pay.

Same warranty; same breach. The vehicle collides with a truck in broad daylight. There is no possibility that the defective headlight contributed to the accident given the circumstances in which it happened (it was daylight and the headlights would not have needed to be on even if they were working properly). The insurer should not escape liability based on the breach.⁹⁵

According to the Law Commissions’ further explanation, ‘neither the insured nor the insurer would have to prove what actually caused the loss, or what would have happened if the term had been complied with’.⁹⁶ However, it seems that any actual circumstance of darkness or daylight can be relied upon by the assured to retain its entitlement to indemnity. It seems to be contradictory: the former poses the question of whether it is generally assumed that non-compliance would increase the likelihood of certain risk such as fire or vehicle collision, immaterial of the specific circumstance of every loss, whereas the latter defence indicates that it would be open to the assured to argue that the black ice might not have been visible even if the headlight had been working.

Merkin and Gürses conclude that the phrasing of section 11(3) is a matter of causation.⁹⁷ The type of causation in their claim appears to be actual causation as identified above – a direct and individualised causal connection between non-compliance and the loss. Similarly, the ‘would have increased the risk’ test can be seen in tort law. The general position in tort, as Wright suggests, is that the courts have perceived the critical distinction between pure statistical evidence and causally relevant particularistic evidence.⁹⁸ In a case he uses as an example, the judge refused to allow the jury to infer that one of the defendant's buses caused a particular accident when the evidence merely showed that the defendant's buses were the most frequent users of the route on which the accident occurred.⁹⁹ It has been argued that mere probabilistic statements of increased risks should not generally work in tort cases, so that individual liability will be imposed on a defendant only if it is believed that the tortious aspect of the defendant's conduct actually contributed to the specified legal injury.¹⁰⁰

From English law, in Fairchild v Glenhaven Funeral Services,¹⁰¹ the claimant had suffered mesothelioma caused by exposure to asbestos dust with several different employers long ago in his youth. Mesothelioma is a cancer that starts by an unknown process in the pleura around the lung. The claimant could not prove which employer exposed him to the fibre that caused the triggering of cancer, so on the ‘but for’ test¹⁰² his claim failed. He could, however, prove that every employer who exposed him to asbestos increased the risk that he would suffer the disease. The House of Lords applied the ‘material increase in the risk’ test and gave judgment for the claimant.

According to Wright, one approach to these cases, when the plaintiff cannot establish that the tortious act more likely than not contributed to the injury, is to reduce the plaintiff's burden of proof to permit recovery even though the degree of confidence in the causal contribution is less than the balance of probabilities. A second approach is to allow recovery for the reduced chance of avoiding the injury, rather than for the injury itself.¹⁰³ Wright suggests that the former somewhat weakens the actual causation requirement (meaning whether the defendant's conduct actually contributed to the plaintiff's injury), whereas the latter preserves the integrity of the requirement for actual causation, while also addressing the basic issues in these cases.¹⁰⁴

In Harris, the court held that the traditional but-for test should not be applied to an assured's non-compliance in determining insurance liability. The legislative intent does not require an assured's non-compliance as a necessary cause to the occurrence of the loss. Akin to Fairchild, the New Zealand statutory provision prescribes a test of ‘did increase the risk’ of the loss. This is pure causation – a specific and actual link.

We can now see that actual causation is the dominant model applied in the law, whereas the Law Commissions intended to introduce a hypothetical correlation test by section 11(3) of the Insurance Act 2015. There exists a third model – general causation – which is the focus of the following discussion.

4. Discussion

Our analysis has demonstrated that section 11(3) is an amalgam of the central ideas of both non-specific correlation and general causation, and actual causation and probabilistic causation. As Merkin and Gürses have highlighted, there is a contradiction within the wording of the provision. While an interesting question would be which approach is to be preferred, the question here is about the true meaning of section 11(3) of the Insurance Act 2015. It is submitted that the ‘would have increased the risk’ test of section 11(3) should be a matter of causation but not in the form of actual causation – hence, falling into the scope of general causation in the philosophical sense.

Upon a purposive interpretation according to the Law Commissions, section 11(3) is different from the New Zealand approach in that it appears to only require generalised and statistical evidence, normally applied in an ex ante situation. If the rule is to adhere to a risk assessment on a generalised level, the Law Commissions’ examples would not and should not be interpreted as opening a question of a specific and actual link between non-compliance to some ex post circumstances. In contrast, a literal interpretation of section 11(3) of the Insurance Act 2015 may well lead one to the path of ‘material increase of risk’ – as in Fairchild – or ‘did increase the risk’ – as in Harris. The two-step approach first asks that a correlation be established between qualifying non-compliance under section 11(1) and the prejudice of the insurer's position on a generalised level, and then consideration of whether the non-compliance is causally relevant to the ‘circumstances’ as opposed to the loss itself. This gives rise to a causal question and the application of a test of ‘did increase the risk’ of the loss, which is a distinct test from both causa proxima and the but-for test.

To summarise, it is submitted here that a proper interpretation of section 11(3) should consider both the generalised issue and a specific link to certain circumstances. It is suggested that the wording of section 11(3) should be interpreted by an approach to recognising a nexus of correlation with a general causal relevance (general causation) and interpretation as restricting tools (as in Moore's term, ‘pruning’), to achieve alignment with the legislative intent.

While recognising the existence of necessary causes and the proximate cause, the starting point in insurance law is to embrace general causation – a causal involvement by occasioning the occurrence of the loss, generally and probabilistically. General causation means, in the context of section 11 of the Insurance Act 2015, increasing a particular risk in a generalised way, but excluding purely spurious actuarial outcomes in a particular context. This test defines borderline cases that are in line with the legislative intent of the Insurance Act 2015.

To test this approach, posit a claim in circumstances where there has been a breach of a burglar alarm warranty; but there has been no unauthorised entry into the premises. The loss itself was caused by a fire. As the first step under section 11(3) of the Insurance Act 2015, we ask if a reasonable insurer would normally put the absence of a burglar alarm as a fire risk-increasing factor. This would tend to show that a burglar alarm represents a higher fire risk in all fire losses claims. What if, on a generalised level, non-compliance regarding installation of a burglar alarm could in fact entail a higher risk of fire? Statistics may evidence a correlation between the number of fire incidents and a lack of burglar alarms, because the kind of people who are risk averse in relation to fire prefer installing alarms, including burglar alarms. The rate of installation of alarms here serves as a proxy for risk aversion, including to fire and theft, that can be detected and controlled by prescribing alarms. However, without a specific circumstance and a requirement of causal relevance, it might be an incidental or temporal correlation.

It is arguably less than ideal to place reliance upon actuarial considerations as the sole ground for the determination of legal liability (in this context, insurance liability) for two reasons. First, pure statistical data generally cannot inform actual and specific legal issues,¹⁰⁵ which instead should go hand-in-hand with a more subjective and specific link in the law. An analogy may be drawn with the test of materiality and inducement as required in the rules of the duty of fair presentation of the assured. An assured may provide incorrect information relating to risks, thereby effectively inducing insurers to agree to accept the risk. The Insurance Act 2015, section 7(3) provides that a circumstance or representation is material if it would influence the judgement of a prudent insurer in determining whether to take the risk and, if so, on what terms. The insurer must be also able to prove actual inducement in order to establish a ‘qualifying breach’ under section 8. Neither the Marine Insurance Act 1906 nor the Insurance Act 2015 explicitly requires a causal connection as an essential element to prove a qualifying breach. Instead, the law requires two independent tests of materiality and inducement. Materiality refers to what needs to be said in order to make the presentation fair, involving an objective assessment viewed from the standpoint of the reasonable and prudent insurer.¹⁰⁶ This is similar to the broad and objective test of ‘would have increased certain risk’. There is then a further test evaluating the actual impact on the insurer, the inducement test.¹⁰⁷ This tests for a specific and actual consequence, as opposed to a generalised and purely probabilistic one.¹⁰⁸ Combining the two, a breach of the duty of fair presentation can be established and the remedies prescribed by law will follow. It is submitted that the language in section 11(3) of ‘could not have increased the risk of the loss which actually occurred in the circumstances in which it occurred’ mirrors this construction.

The causal notions rooted in our mind will urge us to investigate the circumstances in which, for example, a fire occurred. We intend to engage in a much more detailed investigation and explicit description of the actual conditions in order to have any substantial degree of confidence in our empirical prediction. Returning to a hypothetical offered by the Law Commissions to exemplify lack of bearing: breach of a fire alarm or locks warranty where the loss is caused by flood.¹⁰⁹ It appears to show that so long as the risk of a particular kind, location or time in the risk clause is different from the (proximate) cause of the loss, the insurer will be liable for the loss. If a fire was set by a burglar but the assured failed to comply with the risk clause regarding the installation of a burglar alarm, considerations may be different from where a fire was due to some natural cause or employee negligence. If the fire was caused by a negligent contractor present on the premises to install a burglar alarm, the question would arise: does that non-compliance have a causal involvement by occasioning the occurrence of the loss for reducing the fire hazard under the specific circumstances?

Here is another of the Law Commissions’ examples: failure to employ a night-watchman where there is a theft in the middle of the afternoon.¹¹⁰ The different circumstances involving the same cause but at different times serve as a limitation which, in essence, is a matter purely of interpretation rather than causation. Although a general causation is satisfied here, section 11(3) excludes irrelevant factors in the specific context by way of interpretation of the actual circumstances.

Although it is accepted that section 11(3) of the Insurance Act 2015 is not intended to introduce a traditional actual causal test, the above has shown that a purely generalised and ex ante test is not easily adaptable to the determination of insurance liability in an ex post situation, as under section 11(3). It is submitted that unlike section 10, section 11(3) is not and cannot be wholly free from causation in the determination of insurance liability. The true test that section 11(3) was intended to introduce is between the loss and assured's non-compliance; there is a correlation embedded with causal assumptions as defined under the specific circumstances by way of interpretation. Supplementing the general causation test, if actual causation can be established, the insurer is, a fortiori, not liable to pay the indemnity.

Conclusion

Causation is not always a requirement for legal liability, and causation in the law is not an unvarying or unitary concept. Other forms of links or bearing exist and should be recognised for the ascription of liability in law. ‘Could (not) have increased the risk’ has the merit of being an accessible term, and in section 11 of the Insurance Act 2015 is deliberately used to avoid the matter of causation. However, it is submitted that a wider, conceptual analysis demonstrates that the test is not completely free of causation.

In this paper, the distinction has been drawn between the concepts of correlation and causation in the insurance context. Academic and philosophical accounts of causation suggest an extended causation concept in the insurance context, beyond the narrow confines of causa proxima, or a but-for test or any causal connection in a weaker form.

Upon analysis of the straightforward causal approaches adopted by Australian Insurance Contracts Act 1984 and New Zealand Insurance Law Reform Act 1977, comparing the different statutory wordings and legislative intents, it is concluded that section 11(3) should be interpreted as follows. The starting point is that the assured's qualifying non-compliance of a risk clause has prejudiced the insurer's position by probabilistically increasing the risk that is involved in causing the loss on a generalised level. In this case, the insurer will pay nothing. This is, in effect, a correlation nexus, an assumption of ‘some bearing’ that would be tested by actuarial science and verified or disproved in an objective way. However, it is submitted that a mere correlation is not effective to determine insurance liability – only causal assumptions should be relevant, whereas purely spurious proxies as a result of actuarial science should be excluded. Even where a general causal relevance can be established, ‘the circumstances in which the loss occurred’ should be interpreted by courts as a limiting factor to the discharge of insurance liability, in line with the legislative intent. If it is further shown that the non-compliance ‘did increase the risk of the loss’, the insurer, a fortiori, will not pay.