Representing dimensions within the reason model of precedent

That the set of pro-plaintiff reasons is empty is part of Horty’s example, but inessential. What matters is just that the set of pro-plaintiff reasons, outside of those related to the time abroad, remain unchanged in the past and current case.

The switching point receives a formal definition below.

This is based on a similar dimension with more points from HYPO (Ashley 1991). In general, my approach lies between the dimensions of HYPO and the factors of CATO. HYPO allowed dimensions that were continuous (like the length of time abroad), enumerated (like the security measures), Boolean (only extrema are considered), and unary (only point is considered. I convert these to factors. The conversion for Boolean and unary dimensions is straightforward. For enumerated dimensions, I identify factors as corresponding to points or ranges of points on the dimension. For continuous dimensions my strategy is the same, though a great many factors may be required. The approach is very similar to that found in Bench-Capon and Bex (2015) and Prakken et al. 2015), insofar as the dimensions are all treated as order sequences ranges which can then be converted into factors. I am thankful to a reviewer for stressing this comparison.

I use the letter “v” for points instead of the letter “p” to avoid confusion with the notation for pro-plaintiff factors. I had initially called points on a dimension “values,” hence the “v”, but switched to “points” in order to avoid confusion with the different kind of values discussed in Sect. 4.

Matters can be more complicated if the range of points is dense.

This will be a set of ordered sets. Throughout the article I assume that cases only involve one dimension.

For discussion on how a dimension can be introduced, see Al-Abdulkarim et al. (2013) and Ashley (1991).

If cases were characterized using all the factors favoring the losing side and only the remaining factors that were part of the ratio, then the reason and result model are equivalent. This comes at the cost of representing cases that seem very different as equivalent, as the example with the revised Case 1 showed.

The points here are “weak protective measures,” “moderate protective measures,” and “strong protective measures” for V ₁, V ₂, and V ₃, respectively.

Horty (2012) offers a brief discussion of some of possible justifications.

CATO represents a point on a dimension as strict superset of the factor(s) corresponding to the weaker points. As Prakken et al. put it, “That is to say, as we move along the HYPO dimension we collect more and more CATO factors. The increased degree of support is thus given by the cumulative effect of several factors, rather than distinguishing the difference of degree of support given by different positions on the dimension” (2015).

However, the fact of a 30 month stay also entails the fact of a 2 days stay, but a 2 days stay likely corresponds to a pro-plaintiff factor. You do not want the pro-taxpayer factor of a 30 month stay to entail the pro-tax authority factor of a 2 days stay. You can adjust for this by characterizing the pro-plaintiff factors as “at most a—month stay” while the pro-defendant factors are “at least a—month stay”.

This would be done for each dimension, but I’ve stipulated for simplicity that only one dimension is present in the cases. The extension to additional dimensions is clear.

Interstates in Florida have such limits, see Fla. Stat. 316.183 (2017).

Technically the weakest pro-plaintiff factor will also be a switching point.

You can have cases where the weighing established by the case does not establish any ordering for the unordered point. For example, a case where we are told that an unordered point corresponds to a factor that is outweighed by a novel reason. Still, this provides the basis for a future ordering as other cases involving that previously novel reason are decided.

In Bench Capon (Al-abdulkarim et al. 2016) and (Garner 2004, p. 1592) the statement of the resolution at the final level is called the “verdict.” While “verdict” is loosely used to refer to a judge’s resolution of the issues in a non-jury trial (Garner 2004, p. 1592), strictly speaking this is incorrect. The term “verdict” technically refers to “a jury’s finding or decision on the factual issues of a case” (Garner 2004, p. 858). The proper term for a “court’s final determination of the rights and obligations of the parties in a case” is “judgment” (Bench-Capon 2017). This distinction makes possible that last request of the losing trial lawyer: a judgment notwithstanding the verdict (JNOV).

Much interesting work on this topic can be found in Ashley and Walker (2013) and Walker (2007, 2009).

Sentencing guidelines may work in a similar manner. They fix strict maximums and minimums, but then leave the judge complete discretion within those limits and past cases applying the guidelines do not function to narrow this range of discretion.

See Jury Verdict Research Inc (2016), Klein (1991) and Shannon (2002).

Given the difficulties of modeling moral reasoning and the evidence that moral beliefs are very resistant to revision (Aramovich et al. 2011; Luttrell et al. 2016; Skitka 2010), especially in light of argument (Haidt 2001, 2010), it may be prudent for theories to treat the ordering of values (or whatever it is that is ordered at the highest level of the theory) as fixed before the case is argued. Each judge would have a constant ordering and ideally the theory would be able to apply that ordering to make predictions for cases before each judge. Realistically, one could use the widely accessible information regarding the judge’s party affiliation as a proxy for their value orderings. The theorist would categorize the pool of judges according to political affiliation and construct a partial value ordering for each political affiliation, see Graham et al. (2009, 2011) Hirsh et al. (2010), pp. 655–56, Janoff-Bulman (2009) and Janoff-Bulman et al. (2008) for evidence of such value orderings using very abstract values. The political affiliation of the judge would then be encoded as part of the facts of the current case and the corresponding value ordering would be available, if needed, to generate a prediction for the case outcome.

Political affiliation represented in terms of less abstract attitudes or policy preferences does fairly well in predicting judicial decisions, see Heise (2002, sec. V.A.2) and citations therein, though these tend to be focused on very high level appellate cases that compose a small fraction of all cases. Still, if these attitudes seem resistant to argument, it would make sense to fix them within the theory instead of the more abstract moral values. Such a hybrid approach could allow the argumentative portion of the theory to deal with the mundane cases while the fixed values/attitudes can help resolve hard cases.