Judgment coding, 1,860 published judgments, five courts

Legal research poster example

A legal research poster has to show where its cases came from and how they were read before it shows any difference between courts. This A0 landscape example reports a fictional study of sentence length for one offence category across five courts, built from 1,860 coded judgments, with a five entry coding protocol, one chart of the five court medians, a key finding callout and a discussion that keeps publication bias in view.

Create a poster with OneCraftA0 Landscape, printed at 1189 by 841 millimetres

The whole board

The poster at full size, exactly as it prints. Every number, citation and caption on it was written for this example, so the layout is being judged on real content.

Nine months apart: sentence length for one offence category across five courts, 1,860 judgments

S. Kaufmann ¹, B. Ntuli ¹

1 School of Law, Redhaven University

ABSTRACT

Sentencing consistency is asserted in appellate judgments and rarely measured. We coded 1,860 published first instance judgments for one offence category across five courts between 2020 and 2025. Median custodial length runs from 14 months at the lowest court to 23 at the highest, a spread of nine months that survives controlling for the aggravating factors recorded in the judgments themselves.

INTRODUCTION

Every study of sentencing spread runs into the same objection: courts differ because the cases differ. The usual answer is to control for whatever the dataset records, which in most sentencing datasets is the offence code, a prior record flag and the plea. That is far less than a judge sees. The published judgment is a better source, because it states in prose what the judge treated as aggravating and mitigating, and it does so in the judge's own words rather than in an administrative category. We therefore coded the reasons rather than the record. Two coders read every judgment and marked seven aggravating factors and five mitigating ones defined before coding began, along with plea, prior convictions and whether the offence involved a weapon. The cost of that approach is the sample: only judgments that are published can be coded, and the publication rate varies by court, from 61% at the busiest court to 88% at the smallest.

METHODS

Sample. 1,860 published first instance judgments, one offence category, five courts, 2020 to 2025.

Coding. Seven aggravating and five mitigating factors defined before coding, applied by two coders.

Agreement. All judgments double coded; agreement 0.89 on aggravating factors and 0.84 on mitigating.

Outcome. Head sentence in months of custody, non parole period recorded separately and not modelled.

Limits. Publication rate varies from 61% to 88% by court; unpublished sentences cannot be seen.

Figure 1. Median head sentence in months (m) by court, lowest to highest: A Ashvale to E Eastcote.

Source: published first instance judgments, 2020 to 2025.

RESULTS

The five medians rise steadily rather than splitting into two groups, and the interquartile ranges overlap heavily: a sentence of 16 months would be unremarkable at any of the five. What differs is the middle of each distribution and the length of its upper tail. Eastcote passes 33 months or more in a quarter of its cases, against 21 months at Ashvale. The proportion of judgments recording a weapon runs from 22% to 29%, which is not enough spread to account for a nine month gap: reweighting every court to the pooled weapon rate moves the medians by at most one month and leaves the ordering unchanged. The same is true of plea, which is recorded in 78% to 84% of judgments across the five. Nor does time: splitting the six years into two halves moves every court median by at most two months and leaves Ashvale lowest and Eastcote highest in both halves.

DISCUSSION

Three explanations remain after the recorded factors are controlled. The first is that the cases genuinely differ on something the judgments do not state, which cannot be tested with this design and is the reason the paper stops short of calling the spread inconsistency. The second is local sentencing culture: four of the five courts have three or fewer judges hearing this offence, and a single judge with a firm view moves a small court median a long way. The third is publication. A court that publishes 61% of its judgments is choosing which ones to publish, and if the shorter and more routine sentences are the ones left unpublished, the median reported here is too high for that court and too close to the others for comfort.

Key finding

Nine months separate the lowest and the highest court median, 14 months at Ashvale against 23 at Eastcote, across 1,860 judgments.

Medians run from 14 to 23 months across five courts

Interquartile ranges overlap; the tails do not

Weapon rate and plea rate do not explain the gap

Publication rate varies 61% to 88% and limits the claim

1.Ntuli, B. and Kaufmann, S. (2025) Coding reasons rather than records in sentencing research. Journal of Empirical Legal Studies, 29(2), 244 to 271.

2.Farrow, D. (2024) Publication selection and the measurement of sentencing spread. Criminal Law Review Quarterly, 46(3), 188 to 210.

Block by block

What each block on the board is for, in the order a reader walks it.

Title band and two authors: The title gives the finding and the corpus in one line: nine months apart, sentence length for one offence category across five courts, 1,860 judgments. Two authors share one school affiliation. A title built this way tells a passing reader the size of the gap and the size of the evidence before they reach the chart.
Abstract: Three sentences: sentencing consistency is asserted in appellate judgments and rarely measured; the study coded 1,860 published first instance judgments across five courts between 2020 and 2025; and median custodial length runs from 14 months at the lowest court to 23 at the highest, a spread that survives controlling for the aggravating factors the judgments record.
Introduction: coding reasons, not records: The longest text block explains the design choice. Administrative sentencing data usually records only the offence code, a prior record flag and the plea, far less than a judge sees. Published reasons state what the judge treated as aggravating and mitigating, so two coders read every judgment against twelve factors defined in advance. The cost is the sample: only published judgments can be coded, and publication runs from 61 percent at the busiest court to 88 at the smallest.
Methods as a labelled protocol: Five entries: the sample of 1,860 published first instance judgments, the coding frame of seven aggravating and five mitigating factors, double coding with agreement of 0.89 and 0.84, head sentence in months as the outcome with the non parole period recorded but not modelled, and a limits entry that names the publication rate before the results do.
Figure 1: five medians: One bar chart carries the whole comparison: median head sentence in months for courts A to E, keyed in the caption from A Ashvale to E Eastcote so the axis labels stay short. The bars rise steadily rather than splitting into two groups, which is itself a finding. The caption says the courts run from lowest to highest median, A Ashvale to E Eastcote, which is the order a reader needs because the board has no table.
Results: the checks behind the gap: The results block does the work a table would. It says the interquartile ranges overlap heavily, so 16 months would be unremarkable anywhere, and that the tails differ: a quarter of Eastcote sentences reach 33 months or more against 21 at Ashvale. It then tests three objections with numbers: weapon rates of 22 to 29 percent, plea recorded in 78 to 84 percent, and a split of the six years into halves. None moves a median more than two months.
Discussion, key finding and checked points: The discussion leaves three explanations standing: case differences the judgments do not state, local sentencing culture on benches of three or fewer judges, and publication selection. A key finding callout states the nine month gap with both courts named, and four checked points close the board, the last one repeating the 61 to 88 percent publication limit.
References: Two references sit at the foot of the board, both invented for this fictional study because the layout carries a reference block. On your board, cite the sentencing data source, the coding frame you adapted and any published work on sentencing consistency in your jurisdiction.
How to adapt this board: Retype the title, abstract and five method entries, then enter your courts and medians in the chart's own data table. If you need interquartile ranges and counts per court on the board, write them into the results text or choose a table led poster layout, because this one has no table block.

What makes this board work

One chart, one comparison

Five bars, keyed A to E, carry the whole claim. The steady rise from Ashvale to Eastcote reads from across the room, and a reader who wants the spread inside each court finds it in the results text beside the chart.

The confound is measured, not waved away

The results state that weapon rates run 22 to 29 percent across the five courts and that reweighting to the pooled rate moves each median by at most a month. Naming the obvious objection and answering it with a number is what persuades a legal audience.

The limit is stated three times

Publication rates of 61 to 88 percent appear in the methods, in the discussion and in the checked points. A study that can only see published judgments has to say so repeatedly, because it is the strongest reason to doubt the result.

Questions people ask

What goes on a legal research poster?

The legal question, the corpus and how it was selected, the coding frame and who applied it, the main comparison, and the limits of the source. Empirical legal work usually rests on published judgments, so the publication rate belongs on the board rather than in a footnote, as it does here in the methods, discussion and checked points.

How do I show a spread between courts without a table?

Chart the medians and write the spread into the results text. This board charts five medians in one bar chart and puts the quartiles in prose, for example a quarter of Eastcote sentences at 33 months or more against 21 at Ashvale. It works for five groups; beyond seven or eight, a table layout reads better.

How should coder agreement be reported on a poster?

As one number per factor group in the methods protocol, with the fact that every item was double coded. This board gives 0.89 for aggravating factors and 0.84 for mitigating ones. Name the statistic you used in the full paper, and say whether disagreements were resolved by discussion or by a third coder.

What type size should a law conference poster use?

Body text large enough to read at a metre, which on an A0 board printed at full size means roughly 24 points or more, with headings about double. Case citations and references can sit smaller because a reader steps in for them. If a block will not fit at that size, cut words rather than shrinking the type.

How do I adapt this to a different jurisdiction?

Retype the title, the abstract and the five method entries, then edit the chart's data table with your own courts and medians. The chart holds its own categories and values, so changing the number of courts is a data change rather than a layout change. Rewrite the limits entry for how judgments are published where you work.

Can a sentencing poster call a gap inconsistency?

Only when the design rules out genuine case differences, which a study of published reasons usually cannot. This board stops short on purpose: its discussion lists case differences, local bench culture and publication selection as explanations still standing. Say what the gap is and what would test each explanation, and leave the stronger word to the reader.

Build your own in about a minute

The button below opens the generator with this use case already described. Change the wording to match your own, generate, then edit anything you like.

Make one like this

Sources

Written and checked by the OneCraft team. Last checked 29 September 2026.