1. Introduction
There seems to be little room for controversy with regard to the issue of when legislation changes. In general, we simply know that law is ‘different from what it used to be’ when novel legislation is enacted by the Parliament to replace old regulations—and normally it is expressly stated that the new piece revokes past legislation on the topic. But legislation is not the only source of law there is. Courts also develop the law. This is the core idea of establishing precedents. But are we as sure of when the law changes in a system of precedents as we are in the case of legislation? The continued use of precedent naturally gives rise to novel scenarios that test its boundaries. Deciding a current case based on a not-exactly-alike precedent raises questions about whether and how case law changes through its use. How does case law evolve in response to novelty?
The answer depends on how precedent is understood; legal philosophers debate this issue. There are three influential models, which are based on the ideas of rules, reasons, and analogy. When facing a new case, a court is normally thought of as having three available options: It can either follow the precedent; distinguish the precedent; or, if it has the powers to do so, overrule the precedent. Although overruling is an important decision within a doctrine of precedent, this paper is concerned with how case law develops in response to new cases. Overruling is a corrective power, available only in very limited situations and to certain courts, used to address mistakes—whether of law, justice, or other broad values or consequences. As such, it reflects concerns of correctness rather than a process of adapting to new facts raised in subsequent cases. For this reason, overruling affects case law similarly across different accounts of precedent, and thus falls outside the scope of this paper. Decisions that follow a precedent or distinguish the current case from a precedent, however, may impact the law based on novelties in different ways, depending on the account of precedent adopted (rule-based, reason-based, or analogy-based)—or so we will argue.
We claim different conceptual structures for legal change for each of the approaches to precedent as novel cases are decided. The varying theoretical possibilities raise the empirical question of how people actually perceive legal change through precedents. Precedent-based reasoning is common in everyday life—but how do laypeople perceive the use of precedents for resolving legal cases? When arguing from precedents in ordinary contexts, people must somehow address the transitions to maintain an understanding of what precedents stand for as subsequent cases arise with novelties. The perception of the use of precedents in ordinary contexts may also influence their perception in legal ones. Which model of precedent—rules, reasons, or analogies—best aligns with folk intuition? We conducted experiments to examine which model of precedent best captures how ordinary people perceive legal change through precedents, in the context of courts’ decisions to follow or distinguish them. Empirical data, we contend, is crucial for advancing the debate around precedents. Since people are subject to the law, it is desirable that they be able to grasp its developments. This consideration may ground a normative argument against non-endorsed views, particularly on grounds of publicity and transparency. In addition, our findings may place a burden of justification on those who advance a conceptual theory of social practice that diverges from actual practice. Legal theories should either satisfy a minimal descriptive requirement or explicitly acknowledge the lack thereof. Determining where this justificatory burden lies requires empirical evidence.
2. Precedent and Legal Change
The main conceptual claim underlying the empirical study presented in this paper is that whether—and how—case law changes through decisions to follow or distinguish precedent depends on the concept of precedent one adopts. What causes change, and how it occurs, depends on the nature of what is being changed in the first place. We consider three influential ideas of precedent—rule-based, reason-based, and analogy-based.
The following sections provide an overview of these accounts of precedent, with a focus on legal change that might take place from a decision to follow and also from a decision to distinguish the precedent. The aim is not to provide a comprehensive survey of the literature on approaches to precedent in all its developments, but rather to highlight what is necessary to generate falsifiable hypotheses about legal change through precedent—hypotheses that will be tested in the empirical study we present.
2.1 The Decision to Follow the Precedent
Consider the following ‘Shoplifting’ case (C1):
Alex shoplifted food from Walmart. They were vulnerable, poor, and had not eaten that day. They had never shoplifted before and only took ordinary groceries, such as rice and eggs. The case went to court, where it was decided that the defendant should not be convicted because they stole out of necessity to feed themselves.
Now, suppose that the following case (C2) is brought to court after C1 is ruled:
Sam went to a local convenience store and shoplifted food. They were vulnerable and poor, and did not have a meal that day. They had shoplifted once before. They stole ordinary groceries, such as milk and bread. The court ruled according to the precedent C1.
C1 and C2 are not identical—as no two cases ever are. C1 is a case of a crime committed by a first-time offender, whereas in C2 the defendant was a repeat offender. For this reason, one might think of the felony in C2 as worse than the one in C1—repeating a felony carries greater blameworthiness. Despite the difference, suppose that the court in C2 decided to follow the precedent C1—as described in the vignette. The court concludes that the precedent controls the present case. What does following precedent C1 in case C2 mean to the development of the law established by that precedent?
The rule-based approach.
A precedent is to be understood as laying down rules to be applied in subsequent cases.Footnote 1 The rule is to be found in the ratio decidendi of a case.Footnote 2 So, the development of the law should occur through the development of these rules.
The ratio of C1 is indicated in this passage: “The defendant should not be convicted, because they stole food out of necessity to feed themselves.”Footnote 3 Thus, it is the element of “necessity” (f 1) that forms part of the ratio, as determined by the court. The case can also be thought of as formed by other elements, such as: “never shoplifted before” (f 2); and “ordinary groceries” (f 3)—but f 1 is the one chosen by the court to feature the ratio. In that scenario, the rule established by C1 could be represented as:
R1 = If {f 1}, then S
where
R is the rule
and
S is the solution given to the case (acquittal).
Element f 1 is also present in case C2, as is element f 3, since the groceries stolen in C2 (milk and bread) were also ordinary. The cases differ with respect to the defendant’s criminal history: In C1, it is non-existent (f 2), whereas in C2, there is one prior conviction (∼f 2). Nevertheless, f 1 is present in both cases, and it is this element that forms part of the antecedent of the rule established by the precedent case.
As the court in C2 decided to follow the precedent, we should note that the ruling in C2 does not change the rule established by the precedent at all: After the decision in the new case, the law still is that If {f 1 }, then S. Even though other features are present in cases C1 and C2 (f 2 ,∼f 2 , f 3), they are not incorporated into the law in a decision to follow the precedent that keeps the ratio the same. This reflects a central feature of the rule-based approach as surveyed here: Only features that are explicitly part of the ratio are treated as part of the law for present purposes. On this understanding, decisions that merely follow a precedent are taken not to affect the state of the law, insofar as they leave the ratio unchanged.Footnote 4
The reason-based approach.
Instead of understanding precedents as creating rules in their rationes, this approach proposes that precedents should be understood as decisions on the balance of reasons.Footnote 5 Later courts or judges must decide in consistency with the authoritative treatment that the earlier court gave to the facts of the precedent, as these were reported.
In this line of thought, the law established by the precedent is developed—or changed—whenever there is a shift in the balance of reasons set by the precedent. This alteration does not necessarily need to tip the balance in the opposite direction for a development in the law to occur—which is exactly what happens in the following scenario.
When precedent C1 is set, it is established that:
R1 =
{f 1} provides a sufficient reason for S
where
f 1 = necessity.
At the same time, the decision’s ratio decidendi is also seen as establishing that:
{f 2 , f 3} do not provide sufficient grounds to defeat {f 1}
where
f 2 = never shoplifted before;
f 3 = ordinary groceries.
This ratio represents the balance of reasons provided by the features of the case (f 1 , f 2 , f 3).Footnote 6 Where the facts of a later case fall under the ratio of C1, i.e., where feature {f 1} is present, the court must either follow or distinguish C1—assuming there is no power to overrule. Subsequent cases should require a rebalance of reasons, as no two cases are identical. So when C2 {f 1 , ∼f 2 , f 3} is brought to court, and the court decides to follow the precedent (as we have established in the previous section), it does so because it accepts that {f 1 } provides sufficient reason for S; it accepts that {f 3} does not provide sufficient grounds to defeat S, and, reevaluating the balance of reasons in the context of C2, decides that {∼f 2}—the new feature in C2—also does not offer sufficient reasons to defeat S, either on its own or in conjunction with f 3. When the later court in C2 follows the precedent C1, it further elaborates the legal doctrine based on C1, so that the doctrine can now be stated as:
D1 {C1, C2} =
{f 1} offers sufficient reason for S;
and
{f 2 ,∼f 2 , f 3} do not defeat {f 1},
where
f 1 = necessity;
f 2 = never shoplifted before;
∼f 2 = shoplifted before;
f 3 = ordinary groceries.
So, now it is established that {∼f 2} does not defeat {f 1}, which is a new development. Future courts are bound by that. From now on, judges must accept that the defendant’s prior shoplifting does not provide any reason that could defeat the reason for acquittal (namely, that the defendant shoplifted out of necessity to eat). A judge dealing with a subsequent case where both {f 1} and {∼f 2} are present—and where no new grounds for distinguishing the case are identified—may personally think it is unfair to find someone not guilty if they are a repeat offender. Despite their opinion, the later judge is bound by doctrine D1, and should not condemn the defendant because of the authority of the balance of reasons established by the line of precedent (C1 and C2).Footnote 7
In the reason-based approach, then, the decision to follow a precedent (and not only the decisions to distinguish the current case from the precedent, as in the rule-based approach) changes the law.Footnote 8 It does so by adding to the doctrine features that do not defeat the sufficient reason for the decision. This acknowledges that no two cases are ever exactly the same, even if they should receive the same treatment under the law. Thus, regardless of whether the defendant is a first-time or repeat offender, acquittal is warranted if the theft was committed out of necessity to obtain food.
Moreover, the law may change in more radical or more subtle ways as a result of a decision to follow a precedent. Suppose that there were other features in C2 that are not present in C1—{f 4, f 5 …}. By deciding to follow C1 in this new case, the court authoritatively established that all these new features do not defeat S, making future judges bound to accept it. The doctrine would then be:
D2 {C1, C2} =
{f 1} offers sufficient reason for S;
and
{f 2 ,∼f 2 , f 3 , f 4 , f 5 …} do not provide reasons defeating S.
And we can say that the law after the adoption of D2 has changed more in reaction to the new case C2 than after the adoption of D1, because more features were added to the doctrine. In terms of degree, this means less discretion to judge subsequent cases under the doctrine D2 than under D1.Footnote 9
The analogy-based approach.
If two cases are similar in the relevant ways and there is no relevant difference—meaning they are legally the same—then there should be an additional similarity: The decision in the current case should match the decision in the previous case. When comparing two cases, there are invariably both similarities and differences. The analogy-based approach to precedent relies on charting these to guide legal reasoning. Within such a framework, legal change occurs either when a difference is overlooked or deemed irrelevant in a decision to follow.
Mapping similarities and differences is an influential method of drawing an analogy.Footnote 10 When C2 is brought to court, the analogy with C1 might lead us to this mapping:Footnote 11
C1 and C2:
Walmart <-> local convenience store;
rice, eggs <-> bread, milk;
a shoplifted b to eat.
x shoplifted from y, a small business.
This simple version of a mapping between the cases involves two levels of similarities: surface similarities (Walmart <-> local convenience store; rice, eggs <-> bread, milk) and structure similarities (in ‘a shoplifts b to eat’, a is a slot that can be filled by elements of both cases, Alex and Sam, as can slot b be filled by rice, eggs and milk, bread). The surface level deals with direct similarities. The structure level involves mapping relations, and relations are predicates that have more than one slot. Here, there is also a structural difference: Slots x and y can be filled by elements from C2 (Sam and local convenience store), but not by elements from C1, as Walmart is not a small business like the local convenience store. Thus, any two cases can be viewed as both similar and dissimilar at the surface and structural levels—even C1 and C2. Let us assume that the court assesses that the similarities mapped between the cases are relevant and the difference found is irrelevant.Footnote 12 The court concludes that the cases are analogous—and issues a decision to follow the precedent.
When the court decides to follow a precedent in a current case, the differences between the cases are incorporated into the law. All the novelties brought up by C2 are now part of the law, as they broaden the scope of the factual categories of the precedent. In following precedent C1 in the current case C2, the court establishes, among other things, that Walmart is analogous to a local convenience store, and the relational difference—that C1 involves a big supermarket chain whereas C2 involves a small local business—is incorporated into the law. One might argue that a large supermarket chain differs from a local business, as the former has more resources—including the ability to withstand the impact of shoplifting. However, any such distinctions are ultimately absorbed, as the decision to follow the precedent causes the notion of a large supermarket chain to expand, encompassing that of a local business as well. The same applies to categories like rice and eggs, from C1, which somehow expand to include items such as milk and bread, stolen in C2.
All other differences between cases follow a similar pattern, leading to just such an expansion effect: By determining that the cases are analogous, the judge establishes that the defendant’s prior shoplifting history (as in C2) must be incorporated into the analogy with C1, a case where the defendant had never shoplifted before. This means that, moving forward, more cases than initially expected—before C2—might be resolved based on C1. Since C2 has been incorporated into the law (a case of a second-time offender), even cases involving multiple-time offenders could potentially be resolved by C1 in the future, if the current court considers a second-time offender analogous to a multiple-time offender—an analogy that would have been more difficult if the precedent had been established solely based on the categories of C1, where the defendant had no prior shoplifting history. Thus, following precedents has the effect of expanding the reach of factual categories of the precedent, opening up more possibilities for analogy. When the precedent is followed, the factual categories of the current case are henceforth considered analogous to those of the precedent.
2.2 The Decision to Distinguish the Current Case from the Precedent
Imagine that a third case (C3) arises after C2 as follows:
Tom went to a Tesco store and shoplifted food. They were vulnerable and poor, and did not have a meal that day. They frequently shoplift. They stole luxury groceries, such as truffles and wine. The choice of items was driven by a dare, a challenge, or a bet.
The court in C3 may think that S—acquittal—is somehow unfair. If the current court believes that, even though in both situations the defendants were under necessity, the selection of specific types of items—luxury groceries (e.g., truffles, wine) maliciously chosen by the defendant as part of a dare, challenge, or bet—makes the case significantly different from C1—where ordinary groceries (e.g., bread, milk) were stolen—it may distinguish the present case from the precedent and is then free to decide ∼S. Does distinguishing a case such as C3 from the precedent C1 modify the law established by that precedent—and if so, how?
The rule-based approach.
By distinguishing, the court refuses to follow C1, even though it applies to the case.Footnote 13 In order to make it inapplicable and thereafter be able to avoid it, the court alters the rule established by the precedent. The alteration is based on the element that grounds distinguishing: C1 is a case where ordinary groceries were stolen (f 3), whereas in C3, luxury groceries were stolen (∼f 3). In that scenario, the law changes because the ratio of the precedent becomes:
R2 = If {f 1 , f 3}, then S,
instead of
R1 = If {f 1}, then S.
The resulting rule from distinguishing is the original rule, supplemented by an additional condition in its predicate (f 3), which—despite the modification—still justifies the outcome S reached in the precedent case C1.Footnote 14 By doing so, the court declines to follow the precedent in the present case. R2 substitutes R1, which means that, from now on, cases are under the control of R2, and not R1, which ceases to exist in the legal system.Footnote 15
Hence, distinguishing is a way to change the law via a system of precedent. But note that such decisions should impact the law only to the extent that the condition added to the antecedent of the rule alters it. Other novelties in the current case should not result in a legal change. In other words, only the element of the case that justifies distinguishing should modify the law. Suppose that there are additional features in C3 that are not present in C1—for example, f 4, f 5, etc. By deciding to distinguish C1 in this new case based on the fact that in C3 the luxury groceries were stolen (∼f 3), the only change we would observe in the law is that the ratio would still become:
R2 = If {f 1 , f 3}, then S,
instead of
R1 = If {f 1}, then S.
Other differences between cases that are not considered grounds for distinguishing (f 4, f 5, etc.) do not alter the law; differences between cases that are not relevant to distinguishing do not make the law change further.
The reason-based approach.
For the court to distinguish the precedent, it must present a difference in the facts of the cases that the court assesses as defeating the reason considered sufficient for the outcome by the ratio established in C1. The difference must either “(a) provide[] a type of defeating consideration different from that provided by” {f 2 ,∼f 2 , f 3}—considering the doctrine D1 (made of C1 and C2); or “(b) provide[] a better defeating consideration of the same type as” {f 2 ,∼f 2 , f 3}.Footnote 16
Suppose that the court in C3 believes that there is a difference between the facts of the cases that grounds a different outcome—the defendant chose to steal luxury groceries (∼f 3), which provides a different type of defeating consideration than {f 2 ,∼f 2 , f 3}. This new feature, ∼f 3, presents a reason that the court assesses as defeating the reason provided by the ratio ‘luxury groceries were chosen on a dare’, which may give rise to the presumption that the defendant’s purpose was to cause significant harm to the business. Therefore, the court decides to condemn the defendant. As with C2, the decision in C3 alters the law, albeit in a different way:
D3 {C1, C2, C3} =
{f 1 , f 3} are sufficient reasons for S,
and
{f 2 ,∼f 2} do not provide grounds defeating {f 1 , f 3}
where
f 1 = necessity;
f 2 = never shoplifted before;
∼f 2 = shoplifted before;
f 3 = ordinary groceries.
Once D3 is established, S—the acquittal—becomes the appropriate solution in cases where the defendant was not only in a situation of necessity, but also where the stolen groceries were of an ordinary nature. The conditions are cumulative. This means that the court, by distinguishing the precedent, avoids the acquittal outcome for the present and every subsequent case where the defendant shoplifts luxury groceries, making room for the decision ∼S.
Note that the law could have changed even further in C3. The court might think that ∼f 3 alone does not provide reason enough to defeat S. From the precedent C2 it knows and must accept that ∼f 2 alone also does not offer grounds defeating the justification for S. The court does believe, though, that ∼f 3 in conjunction with ∼f 2 provides reason to defeat the conclusion of the precedent C1. At the same time, suppose that there were other features in C3 that are not present in C1 and in C2—f 4, f 5 … which are taken by the court as not providing sufficient reason to defeat ∼f 3 in conjunction with ∼f 2. In that case, the law would change to:
D4 {C1, C2, C3} =
{f 1} & ∼{∼f 2 , ∼f 3} is a sufficient reason for S,
{ f 4 , f 5 … } do not provide reasons defeating {f 1} & ∼{∼f 2 , ∼f 3}.
where
f 1 = necessity;
∼f 2 = shoplifted before;
∼f 3 = luxury groceries.
Thus, the law can change with a decision to distinguish beyond the feature that grounds distinguishing.
The analogy-based approach.
We may say that C1 and C3 are not analogous because the mapping process is interrupted by differences, such as:
C1 and C3:
rice, eggs ≠ truffles, wine;
a’s sole goal was eating b ≠ a’s goal was to eat b, but a also sought to fulfil a dare by stealing b.
When cases are deemed not analogous, the law established by the precedent case changes. Distinguishing the current case from the precedent narrows the scope of factual categories, limiting the potential for analogy in future cases. If the judge identifies a relevant difference between case C1 and case C3—such as the distinction between ‘rice, eggs’ and ‘truffles, wine’—they might conclude that these two categories are not analogous. Before this decision, one might have assumed that these categories were analogous—after all, they are all types of food. However, the decision establishes that, despite the apparent similarity between the cases, they are actually different. This redefinition restricts the scope of the factual category ‘rice, eggs’, preventing its extension to cases where ‘truffles, wine’ were stolen. Thus, with the decision to distinguish, the factual scope of the precedent—or more precisely, its potential for future analogies—shrinks. It becomes established that, although the cases appear similar because they concern shoplifting of groceries, this is not enough for the analogy. In distinguishing, the law evolves by defining boundaries around what is not analogous, effectively reshaping the legal scope of factual categories for potential analogies.
However, this dynamic seems to diminish as cases become increasingly different. When cases that are not too different receive different outcomes, more lines are drawn to separate what appears similar but is, according to the current decision, actually different. The greater the relevant differences between cases, the less one can argue that the law has changed, because a too different target case will narrow the precedent’s factual scope in less impactful ways than a less different target case. This is because a current case that apparently shares more of its framework with the precedent will narrow its scope more substantially than a case that shares less. Note that the shrinking effect is more impactful between cases involving ‘rice, eggs’ and ‘truffles, wine’ than would be between cases involving ‘rice, eggs’ and, say, ‘trains, boats’, because ‘rice, eggs’ and ‘trains, boats’ have less of a common framework to be shrunk. It establishes more to say that ‘rice, eggs’ are not analogous to ‘truffles, wine’ than to claim that ‘rice, eggs’ and ‘trains, boats’ are not analogous. As cases become less and less analogous, it becomes clearer that different laws should govern them. In such scenarios, the line drawn to restrict the scope of a factual category to potential analogies becomes less impactful or less helpful. The less analogous the cases are, the less meaningful it is to claim that distinguishing the current case has shifted the law, as there is less of a shared framework to narrow and fewer boundaries to establish for potential analogies. If cases become too different, we might conclude that they simply do not relate and do not intervene in one another.Footnote 17 Although this scenario—where the distinguished precedent and the current case are so different that the current case does not change the law of the precedent at all—is rather implausible (after all, why would a judge or court consider such a different previous case in a current one?), it shows the tendency for the degree of change in the law to diminish as the differences between cases grow when distinguishing is in order.
2.3. Summary and Comparison of Legal Change Across the Accounts
When cases receive the same outcome, the law does not change according to the rule-based account. Whereas, in that context, according to the reason- and analogy-based accounts, the law does change—in direct proportion to the number of new features introduced by each new case (on the former account), or in proportion to the number of differences between cases (on the latter account).
When cases receive different outcomes (due to distinguishing—overruling is set aside in this discussion), the law changes according to the rule-based approach, but only to the extent of the element that is added as a condition of application of the rule. In the reason-based approach, in such a scenario, the law changes in direct proportion to the number of new features in the current case—so, the more differences between cases, the more the law changes. On the other hand, in the analogy-based approach, the law changes in inverse proportion to the differences found between cases—that is, the more differences between cases, the less the law (in terms of the rejected precedent) changes. Figure 1 summarises the predictions generated by the three approaches.
Predictions for legal change across models of reasoning by precedent.
Figure 1 Long description
Figure 1 contains three panels showing predicted ratings of whether the law changed, on a scale from 1 to 5, by outcome and difference score. In the rule-based panel, predicted change is low when the result is the same and high when the result is different, regardless of difference score. In the reason-based panel, predicted change increases with difference score when the result is the same; when the result is different, predicted change is high for difference score 0 because it represents overruling, lower for difference score 1, and high again for difference score 2. In the analogy-based panel, predicted change increases with difference score when the result is the same, but decreases with difference score when the result is different, with high predicted change for difference score 0 because it represents overruling.
Note that when the outcomes differ despite the cases being the same (difference score 0), it constitutes overruling, resulting in a change in the law across all models.
3. Experimental Study
To test which—if any—of the models best explains ordinary perception of changes in the law, we have designed an experiment where participants were presented with two subsequent court rulings. The first case always established an explicit ratio (e.g., “the court decided that the consumer WAS entitled to damages, because the company implicitly took the duty to ensure security by hiring security staff to attract consumers”) based on a fixed feature that was also present in the second case (e.g., hiring security staff). The cases would either be identical or there would be one or two different features (e.g., parking was free in the first case, but charged in the second). Participants were then told about the second case’s decision, which would either be to follow (because of the fixed feature, e.g., hiring security staff) or to distinguish the precedent (because of a different feature described, e.g., free versus charged parking service). After reading both cases, participants were told that:
Two friends familiar with both cases disagree about them. Casey thinks that the law established in the first case was left unaltered after the second case. Jamie disagrees, saying that the law established in the first case was changed in the second case.
Finally, we asked participants with whom they agreed more, with 5 options ranging from “I agree more with Casey” to “I agree more with Jamie”, including a midpoint labeled “No preference”.
In order to test the three models described above, we have independently manipulated two different variables: difference score and outcome. First, we manipulated the degree to which the two cases were similar or dissimilar, by randomly varying whether each of two additional features (e.g., whether the parking lot was walled or an open area and whether the parking lot was free or charged) were present or absent in a given case. This meant that some cases had both of these features in common (leading to a difference score of 0), other cases had just one feature in common (leading to a difference score of 1), while others still had no extra features in common (difference score of 2). ‘Difference score’ is the term we use to refer to the number of factual differences between the descriptions of two cases in our vignettes—ranging from 0 to 2. This is intentionally a general term, meant to capture factual differences regardless of the model of precedent adopted.Footnote 18 Second, we manipulated whether the second case produced the same outcome as the first case (e.g., “the court decided that the consumer WAS entitled to damages because, as established by the previous case, the company implicitly took the duty to ensure security by hiring security staff to attract consumers”) or a different outcome (e.g., “the court decided that the consumer was NOT entitled to damages, because, although the company implicitly took the duty to ensure security by hiring security staff to attract consumers, the new case is significantly different”).
Our study follows a 2 (target: order 1, order 2) x 2 (relevant feature 1: present, absent) x 2 (relevant feature 2: present, absent) x 2 (last case outcome: same, different) x 5 (scenario: Donoghue, Parking Lot, Food Theft, Debtor, Car Crash) mixed design, where each participant responded to the five different scenarios in a random order, with target, both relevant features, and last case outcome varying randomly for each trial.Footnote 19
4. Methods
A gender-balanced sample of 250 participants, based either in the US or the UK, was recruited using Prolific.Footnote 20 After excluding eight responses which failed our pre-registered attention check, we were left with a sample of 242 participants who completed our study (120 males, 121 females, one non-binary, mean of age = 41.5).Footnote 21
5. Results
For further analysis of the method and results, see Appendix. Following the pre-registered analysis plan, we built a linear mixed-effects model of perceived change in the law with ‘case order’ (an arbitrary factor encoding which of two possible conducts was described as the paradigm case), ‘difference score’, and ‘target case outcome’, as well as all 2- and 3-way interactions between them as fixed effects, while accounting for random effects of participant and scenario. An ANOVA test based on this model revealed statistically significant main effects of target case outcome (F(1, 1181) = 509.32, p < .001) and the interaction between target case outcome and difference score (F(1, 1186) = 11.99, p < .001). No significant effects were found for case order (F(1, 1183) = 0.46, p = .496); difference score (F(1, 1185) = 1.38, p = .241); or their interaction (F(1, 1191) = 1.38, p = .241). Similarly, there were no significant interactions between case order and target case outcome (F(1, 1170) = 0.02, p = .898) or the three-way interaction between case order, difference score, and target case outcome (F(1, 1179) = 3.50, p = .062). A post-hoc power analysis was conducted using the SIMR package,Footnote 22 based on 5000 simulations, and revealed that the model was adequately powered to detect the observed main effect of target case outcome (96.28% power [95.72%, 96.79%]), but severely underpowered to detect the observed interaction between outcome and difference score (10.88% power [10.03%, 11.78%]).
To decompose the detected interaction, we inspected the simple slopes of difference score separately across outcome conditions: Difference scores had a positive effect over the perception that the law has changed when the result recommended by the paradigm case is followed in the target case (B = 0.23, t = 3.26, p = .001). However, for situations in which the paradigm and target case were decided differently, difference scores led to non-significantly lower perception that the law has changed, as predicted by the analogy model (B = −0.12, t = −1.69, p = .091). These effects are depicted in Figure 2.
Boxplot with overlaid and jittered observations.
Figure 2 Long description
Figure 2 shows observed ratings of whether the law changed, on a scale from 1 to 5, by outcome and difference score, with boxplots and overlaid jittered individual observations. Ratings are generally lower when the target case reaches the same result and higher when it reaches a different result. When results are the same, ratings increase as difference score increases. When results differ, ratings decrease slightly as difference score increases, although this latter trend is non-significant.
In Figure 2, boxes are bound at the first and third quartiles, with a strong line denoting the median. Diamonds represent the means. Participants agree more with the statement that the law has changed when the target case’s result is different from the paradigm. The degree to which cases differ is associated with a perception that the law has changed when the result is the same, but the relationship reverses when the result changes from the target to the paradigm case.
Still in accordance with the pre-registered analysis plan, we encoded, in a separate column, the predictions of each theory as summarised in Figure 1, according to the following scheme:
-
a) The precedents-as-rules theory predicts that there will be no change when the result is the same (i.e., complete disagreement with the statement that the law has changed, coded as 1), but definitive change when the result differs (i.e., complete agreement with the same statement, coded as 5);
-
b) the precedents-as-reasons theory predicts that changes in law will be proportionate to the changes between the two different cases, such that equal cases entail that the law is the same (i.e., 1), one factual difference entails a degree of legal change (i.e., 3), and two factual differences entail that the law has definitely changed (i.e., 5), with the exception of overruling cases, where there are no significant factual differences, but opposite results, where the law has clearly changed according to all models;
-
c) the precedent-as-analogies theory makes the same predictions as the precedents-as-reasons theory when results are the same, but the opposite prediction when the results are different, such that when dissimilar cases are judged differently, there is no legal change (i.e., 1) and factual similarities with the differently-decided paradigm case lead to progressive higher degrees of legal change (i.e., 3, 5).
Adding the predictions of each theory as case-level predictors in a mixed effects linear model of perceived legal change with random effects for case and participant revealed support for the precedents-as-rules theory (F(1, 1195) = 380.82, p < .001) and for the precedents-as-analogies theory (F(1, 1196) = 5.97, p = .015), but no support for the precedent-as-reasons theory (F(1, 1199) = 0.27, p = .60). The predictions of the precedents-as-rules theory explain approximately 24% (partial η2 = .24, 95% CI [.20, .28]) of the observed variance, compared with roughly 0.5% (partial η2 = .005, 95% CI [.00, .02]) for those of the precedents-as-analogies theory.
Although we pre-registered a linear model, our dependent variable was in a five-point ordinal scale. To model judgments of whether the law had changed as an ordinal variable, we fitted a cumulative link mixed model using the Laplace approximation with a logit link function and flexible thresholds (i.e., independently estimated cut-points) from the “ordinal” regression model package,Footnote 23 with the same specification as in the pre-registered model: fixed effects for order, target case outcome, difference score, and all two- and three-way interactions between them while accounting for random effects of participant and case. The significance pattern is roughly the same as for the pre-registered linear mixed effects model: We found significant main effects of target case outcome (χ2(1) = 318.63, p < .001), a target case outcome * difference score interaction (χ²(1) = 12.23, p < .001), and no other significant effects (all other χ²(1) < 3.9, all ps > .05)—the three-way interaction between order, target case outcome, and difference score was marginally non-significant (χ²(1) = 3.84, p = .0502).
6. Discussion
The results provide empirical support for the version of the rule-based approach discussed here and offer very limited backing for the analogy-based approach.Footnote 24 In contrast, the reason-based approach to precedent did not resonate with our participants. What about legal change within the rule-based approach—unlike the reason- and analogy-based alternatives—might align more with laypeople’s intuitions and fit within a common understanding of precedent? Furthermore, do our findings carry normative repercussions?
6.1 On an Explanatory Hypothesis
These results invite reflection on why the rule-based approach, as opposed to the analogy- and reason-based approaches, resonates with lay intuitions and aligns with the ordinary concept of precedent. Disregarding changes in the ‘following precedent’ scenarios and overlooking difference scores as drivers of change in the ‘distinguishing’ scenario could be interpreted in several ways. Here, we would like to discuss one explanatory hypothesis for the results: a tendency toward simplicity.
Despite their differences, the least-supported approaches—analogy and reason—share the characteristic of viewing a system of precedent as cumulative: The more cases are decided by courts and judges, whether by following or distinguishing a precedent, the more we learn about the law on that specific topic, as a more detailed account of the doctrine is developed. In these approaches, legal changes take the original position into account and elaborate on the doctrine further with each new legal case. It seems difficult to talk about the one relevant precedent in the reason and analogy-based approaches; instead, there is a line of precedent that, over time, refines what is considered similar or dissimilar, or adjusts the balance of reasons.
Consider a Parking scenario. The precedent C1 was established in a case of robbery in a private parking lot, where security staff was present; the area was enclosed with access control, and a service fee was charged. The court ruled that, by providing security staff, the provider implicitly assumed the duty to ensure security and, therefore, should be held liable. Precedent C1 was followed in C2, where the crime occurred in a supermarket car park that was protected by security staff, where parking was charged, but located in an open area. C3 then arose and the court decided to distinguish it from the precedent C1. In C3, the robbery occurred in the McDonald’s car park, where there was security staff. As in C2, but unlike in C1, the parking area was open and lacked access control; however, in this case, unlike the others, the parking service was free, and on this basis, the court decided to distinguish the precedent.
In the analogy-based approach, this sequence might be represented in a simplified way (considering only direct similarities and differences) as follows:
C1 =
private parking lot.
{C1, C2} =
private parking lot <-> supermarket’s car park;
walled area <-> open area for parking.
{C1, C2, C3} =
private parking lot <-> supermarket’s car park <-> McDonald’s car park;
walled parking area <-> open parking area;
but charged parking service ≠ free parking service.
In contrast, the reason-based approach represents the same sequence in the following manner:
C1 {f 1, f 2, f 3}
where
f 1= presence of security staff;
f 2 = walled area;
f 3 = charged service.
Ratio {C1} =
{f 1} offers sufficient reason for S;
and
{f 2 , f 3} do not provide grounds defeating {f 1};
C2 {f 1, ∼f 2, f 3}
f 1= presence of security staff;
∼f 2 = open area;
f 3 = charged service.
Doctrine {C1, C2} =
{f 1} offers sufficient reason for S;
and
{f 2 ,∼f 2 , f 3} do not provide grounds defeating {f 1};
C3 {f 1, ∼f 2, ∼f 3}
f 1= presence of security staff;
∼f 2 = open area;
∼f 3 = free service.
Doctrine {C1, C2, C3}=
{f 1 , f 3} offer sufficient reason for S;
and
{f 2 ,∼f 2} do not provide grounds defeating {f 1 , f 3}.
If a decision is made in C2 to follow precedent C1, the analogy-based approach leads us to understand that a private, enclosed parking lot is considered analogous to an open supermarket car park for the purpose of attributing civil liability to the provider—the fact that in the second case the parking service was secondary to the defendant’s main activity is incorporated into the law that establishes liability. In the reason-based approach, we also learn that liability exists regardless of whether the establishment was walled or open, meaning that this aspect of physical structure can no longer be used to argue for exemption from liability in future cases.
When the court decides to distinguish the precedent in C3, the law is further elaborated: Free parking service at McDonald’s is not considered analogous to a charged parking service in a private parking facility or supermarket, according to the analogy-based approach. In the reason-based approach, distinguishing C3 from the precedent C1 changes the law to establish that the combination of security staff in the parking lot and the fact that the parking service was charged provides a sufficient reason for liability. This sequence (C1, C2, C3) demonstrates that, in these approaches to precedent, the law may change in every case, meaning that the entire line of precedent is important for determining what the law currently is.
Things are different in the rule-based approach as defined in this paper. According to the view, there would only be change if a new case leads to a different outcome. Decisions to follow a precedent would leave the law unchanged, making difference scores irrelevant. In contrast, decisions to distinguish the current case from the precedent would affect the law, but only to the extent that the condition added to the antecedent of the rule modifies it. Other novelties in the current case would not constitute legal change and, again, difference scores would not be relevant. That is, only the new feature of the case that justifies distinguishing would alter the law. This ensures stability in the formulation of law set by a precedent. In this account, it makes much more sense to point to the one precedent that established the ruling currently in force. And although that one relevant precedent may be in a line of others in historical terms, if we only wish to know how the law treats a certain situation now, we can probably dismiss decisions that were distinguished or overruled by the precedent currently in force. The same can be said of subsequent decisions that followed it. In the rule approach, new developments either replace the old law—in the exceptional cases of distinguishing (and also overruling)—or have no impact whatsoever. Consider the Parking scenario once more, now according to the rule-based approach:
(C1) R= If {f 1}, then S
(C2) R= If {f 1}, then S
(C3) R’ = If {f 1 , f 3}, then S
where
f 1= presence of security staff;
f 3 = charged service.
Identifying what the law is appears considerably more straightforward under a rule-based approach than under its alternatives. The relative simplicity of the rule-based view, together with the persistence of the original formulation of the precedent’s ratio and the corresponding limits on legal change, stands in sharp contrast to reason- and analogy-based accounts. On the rule-based approach, legal change is tracked by a simple criterion: If the outcome remains the same, the law remains the same; if the outcome changes, the law changes. As a result, tracking legal change is comparatively easy, and participants need not look beyond the most basic contrast between the two cases to determine whether the law has changed.
By contrast, on reason- and analogy-based approaches, new applications of precedent continually contribute to what the law is, such that the law is in a state of ongoing, incremental modification. Legal change on these views is more subtle and cumulative, and correspondingly harder to track. Of course, the incremental character of these approaches may offer advantages from certain normative perspectives. The point here is not to assess those advantages, but rather to highlight a feature of the rule-based approach—its relative simplicity and ease of application—that may help explain why participants tended to favor it.
In sum, the rule-based approach is structurally conservative: It preserves the original formulation of the law so long as new cases yield the same outcome, and it localizes change to the feature that grounds distinguishing when outcomes diverge. By contrast, on reason-based and analogy-based approaches, legal change extends more broadly, incorporating additional reasons or mapped similarities introduced by each new case. The fact that the rule-based approach best fits participants’ responses in our study suggests that its conservative and simplifying structure may resonate with ordinary judgments about precedent, helping to explain its apparent appeal.Footnote 25
6.2 On the Burden of Justification
We now turn to some normative implications of the lack of support for the reason-based approach. The discrepancy between the predictions of the reason-based approach—and, to some extent, the analogy-based approach—to precedent and the perspectives of laypeople in our study warrants closer examination. This issue, of course, cannot be fully addressed here, but we would like to briefly consider it from two distinct angles: a) the normative framework of the approaches, especially the reason-based approach, fails to meet a minimum descriptive requirement; and b) the non-supported approaches hold the burden to argue against costs in publicity and transparency.
The first perspective prompts us to consider methodological issues. Of course, the proponents of the reason-based approach, if arguing from a normative standpoint, may not be unduly troubled by empirical evidence that the folk do not currently employ their model. Similarly so for those defending the limitedly supported analogy-based approach. What is desirable for a system of precedents to be and what a system of precedents actually is from the perspective of laypeople are two different questions. While that observation holds true, especially in matters of common practice, there appears to be a fundamental need for a normative theory to meet a minimum descriptive standard.Footnote 26 Arguing from precedents is a core part of the law in most legal systems—and in many of them it has been so for centuries. But the matter goes further than the boundaries of the legal field. Precedent-based reasoning is commonplace in everyday life; even children, like lawyers, frequently use past decisions to influence current outcomes, whether with parents, teachers or other authorities.Footnote 27 The phrase ‘setting a dangerous precedent’ highlights a widespread recognition of this argumentative practice, both within and beyond legal contexts.
While arguing from precedent is a common practice, our study has revealed a discrepancy between the predictions of—especially—the reason-based account and the perspectives of laypeople. This raises the question: To what extent is the reason-based approach arguing normatively about an existing practice, or setting an idealized framework different to what we ordinarily call ‘precedent’? If the theory is based on precedents, and not on something that does not yet exist as a practice, there should be a minimum descriptive requirement to be met so the theory can get off the ground. Or the theorists’ departure from the idea entrenched in the intuition of the folk should demand a supplementary account of the difficulty of laypeople in understanding precedents the way they should, as a way to address explicitly the lack of that minimum descriptive requirement.
But it is important to note that some defenders of the reason-based approach do not seem to be arguing from a strictly normative perspective. Lamond criticises the rule-based approach for, among other reasons, not aligning with the way common lawyers think.Footnote 28 This is an empirical claim that has yet to be fully investigated in order to be proven true or false. However, the fact that the critique is based on an empirical argument highlights the importance Lamond places on meeting a descriptive requirement. Our study does not specifically target common lawyers. In that regard, it does not shed direct light on Lamond’s hypothesis, which ought to be done by future studies. But it does demonstrate that, in general, it is the reason-based approach that fails to meet a descriptive requirement, whereas the rule-based approach does meet it.
Some might insist that, if anyone’s perception matters at all, it is that of legal theorists or common lawyers, rather than that of laypeople, when it comes to how a system of precedents is viewed. While future work should investigate legal experts’ perception, we nonetheless consider the empirical data collected from laypeople to be an important component of a normative argument against the unsupported approaches, based on grounds of publicity and transparency. People are subject to the law. It is desirable that they have an adequate understanding of how the law develops in order to know where they stand. If the mechanics through which legal change occurs are mysterious or obscure to ordinary people—as seems to be the case with precedents under the reason-based approach—this might be seen as problematic. Arguing for the extension of this problem, however, requires further normative analysis: Publicity and transparency are often in a trade-off with other values or considerations, such as due process, efficiency, and the institutional functioning of the judiciary. So, there are reasons—defeasible, certainly—but reasons nonetheless to prefer legal concepts that are closer to folk concepts. This means that the fact that a theory reflects the common ground in the way legal changes occur poses an advantage.
Naturally, theorists are free to argue that the folk concept is wrong. From a normative perspective, there might be reasons why the reason- and analogy-based approaches are better when it comes to the matter of legal changes in a system of precedents. Theorists remain at liberty to argue that the cost that we perceive—in publicity and transparency—is outweighed by the achievement of some alternative theoretical virtues—and, indeed, we are inclined to align with this perspective in a normative discussion, since the reason-based approach offers advantages related to its incremental character. With the results of the study at hand, we simply know where the burden to argue in spite of the intuition of the folk is allocated.
Nonetheless, these results are preliminary and exploratory in nature. The study has important limitations. Participants were laypeople from the USA and UK, both common law countries where precedent plays an important role not only in legal practice, but also in folk legal beliefs. Confirmatory research should survey people from diverse cultural backgrounds, including civil law countries where precedents play a less prominent role in developing the law, and should also focus on people with legal training to check whether the results generalize to that population. It is also important to note that the tendency toward a rule-based approach in legal change may be exaggerated by the relative ease of comparing outcomes to features. Since outcomes represent straightforward information, participants may have focused their attention on them while neglecting the difference scores. Future research could explore this possibility by presenting features and outcomes in an equally salient, tabulated format. Another restriction has to do with how the vignettes are designed, since real life cases do not have as neatly arranged features as the cases presented to participants in this study. Finally, our post-hoc power analysis revealed that the sample size was too small to detect a moderate-sized interaction. This means that, even in laboratory settings, several of our variables might play off each other in subtle ways that we were not able to capture. Future work should employ larger samples in order to deliver a more fine-grained picture of lay views on legal change in the context of adjudication.
7. Conclusion
Precedent has been of interest to legal philosophers and legal theorists for a long time. However, there is little agreement among those who have theorized about its nature. At a broad level, three conflicting accounts have emerged from this philosophical debate based on the ideas of rules, reasons, and analogy. No uncontroversial account of precedent exists and the discussion is ongoing. One interesting and underexplored aspect of the debate concerns legal change: How does case law change in reaction to novel cases? Leaving the armchair, we conducted an empirical study on the topic and found support for the rule-based approach, and very limited backing for the analogy-based approach. The predictions of the reason-based approach to precedents were not vindicated among laypeople in our study. Although the limitations pointed out above should be kept in mind, these findings represent an important first step toward understanding people’s use of precedents and intuition about legal change. The results might reveal a tendency to view a system of precedents in a simplified, intelligible, and non-incremental way. This should be an important consideration when making design decisions for such a system (e.g., how a doctrine of precedent is quoted, or how to write a case brief). We have also argued for a burden of justification on theories that, despite addressing common practices, fail to meet a minimum descriptive standard. Considering people as subjects of the law, we further contend that there are reasons—related to publicity and transparency—to favor legal concepts that align more closely with folk understandings.
Acknowledgments
We thank Insper for funding Study 1. Noel Struchiner’s involvement was supported by the National Council for Scientific and Technological Development (CNPq; grant no. 308757/2022-0), the Carlos Chagas Filho Research Support Foundation (FAPERJ; grant no. E-26/204.038/2024), and the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES; grant no. 001). Ivar Rodriguez Hannikainen’s involvement was supported by research grant CNS2023-144543, funded by the Spanish Ministry of Science and Innovation (MCIN/AEI/10.13039/501100011033) and the European Union NextGenerationEU/PRTR.
Appendix
The statistical model chosen for the pre-registered analysis reported in the main text can be criticized for a number of different reasons. In this appendix, we address two possible sources of criticism: 1) that alternate specifications of the random or fixed effects structures, such as the inclusion or exclusion of predictors, could lead to different findings; and 2) that the fact that the model specification included random intercepts for case and participant, but no random slopes, could significantly impact our results.
This appendix reports the outcome of several alternative models which largely support the same results observed in the main analysis, showing them to be robust to a number of choices that were made at the pre-registration step. All of the code required to replicate the analyses reported here is available in the online supplementary materials: see supra note 21.
A.1. Alternative fixed-effects structures
The two statistically significant effects explored in this paper were the main effects of target case outcome and the target case outcome * difference score interaction. As a first set of robustness checks, we fitted alternative specifications of the model. The main effects of target case outcome remain significant if we drop the fixed effect term for order and their accompanying interactions (F(1, 1186) = 510.02, p < .001) and if, instead of dropping order, we add case as a fixed effect term (as opposed to a random effect), allowing it to interact with all other fixed effects (F(1, 1152) = 506.73, p < .001). The same is true for the two-way interaction between target case outcome and difference score, which is significant if we drop the fixed effect term for order (F(1, 1190) = 11.53, p < .001) and if, instead of dropping this fixed effect, we add case as a fixed effect in the manner specified above (F(1, 1156) = 9.67, p = .002). For full results, see the online supplementary materials, supra note 21.
Our study also included exploratory questions about how each of the manipulated features impacted their judgments, as well as a question asking participants to rate their level of agreement with the outcome in each of the two cases. These variables could plausibly account for some or all of the variance associated with our two effects of interest. While a model including fixed effects of each of these additional predictors did improve model fit (AICpre-registered model = 3979.62, AICfull model = 3951.50, χ² = 40.11, p < .001), it once more revealed significant main effects of target case outcome (F(1, 1177) = 424.44, p < .001) and of the target case outcome * difference score interaction (F(1, 1179) = 8.17, p = .004), as well as of the perceived importance of the first manipulated feature (F(1, 1179) = 4.40, p = .036) and of perceiving both the precedent (F(1, 821) = 6.86, p = .009) and the target case (F(1, 1180) = 16.69, p < .001) as correctly decided. Thus, while the additional exploratory questions help explain the variance in perceived legal change, target case outcome and its interaction with difference score retain their explanatory power in this more encompassing model.
A different exploratory model focused instead on the perceived difference in the quality of decision-making in the precedent versus the target case. To explore this, we computed an evaluative change score by subtracting ratings that the precedent case was correctly decided from ratings that the target case was correctly decided. This meant that evaluative change scores would be positive when the target case is seen as improving the law and negative when it is seen as deteriorating the law. We then added evaluative change as a fixed effect in substitution of order in our pre-registered specification. This model once more reveals significant main effects of target case outcome (F(1, 1186) = 500.67, p < .001) and significant effects of the target case outcome * difference score interaction (F(1, 1187) = 11.19, p = .001), and no other significant effects (all other Fs < 1.66, all ps > .19).
One final exploration with alternative predictors leveraged the repeated-measures nature of our design. Since participants engaged in five different trials, it could be that their performance in later trials was significantly different than that in early trials. Moreover, this could plausibly interact with our manipulations—making, for instance, target case outcome more relevant for earlier trials than for later trials. We are grateful to an anonymous reviewer for this suggestion. To test it, we added ‘trial number’ as a predictor in the pre-registered model specification, allowing it to interact with all other fixed effects. Once more, there were statistically significant effects of target case outcome (F(1, 1172) = 514.69, p < .001) and of the target case outcome * difference score interaction (F(1, 1176) =11.26, p < .001). While there were significant main effects of trial number (F(1, 953) = 9.72, p = .002) reflecting a small tendency of participants to perceive greater change in later trials, trial number did not interact with any other fixed effects in our model (all Fs < 2.4, all ps > 0.12).
A.2. Random slopes
Another way to model our results would be to allow random slopes for some of our manipulated variables. For instance: It seems plausible that different people would respond differently to the magnitude of difference between different cases. On this alternative specification, the main effect of target case outcome (F(1, 1180) = 499.95, p < .001) and the target case outcome * difference score interaction (F(1, 1087) = 11.27, p < .001) remains significant. Moreover, comparing this model to the pre-registered one reveals comparatively similar goodness of fit (AICpre-registered model = 3979.62, AICrandom slopes difference = 3981.19, χ² = 2.42, p = 0.298). Alternatively, we could also add random slopes for target case outcome on top of the previous specification. This improves model fit compared to the previous model (AICrandom slopes difference = 3981.19, AICrandom slopes difference and target case outcome = 3940.41, p < .001). Once more, the only two significant effects here are of target case outcome (F(1, 239) = 306.67, p < .001) and of the target case outcome * difference score interaction (F(1, 1031) = 16.01, p < .001).
Open access
