Implementing MCDA to Determine Eligibility for Pardon

Abstract

Countries celebrating festive national events sometimes declare a wide-scale pardon of prisoners, based on certain criteria. This paper presents an original and effective method for producing a list of prisoners eligible for early release by utilizing a multi-criteria decision analysis (MCDA) approach. The following criteria were used: gender, age, severity of crime, proportion of sentence already served, and behavior in the last three years. Criteria weights were calculated using an analytical hierarchy process (AHP), based on data provided by the authorities. The model produced clear and unbiased results, offering policymakers a practical and user-friendly decision-making tool. It was implemented on a list of 1500 candidates for early release through a wide-ranging amnesty, and several hundred of them achieved scores that made them eligible for a pardon.

1. Introduction

An amnesty is a pardon granted to convicted prisoners. In countries marking events such as Independence Day, an amnesty policy could grant pardons to specific groups of prisoners or reduce their sentences. This is typically done as a goodwill gesture aimed at promoting national unity. The authorities aim to determine the policy of eligibility for clemency in a way that benefits both the prisoners and the public [1, 2]. Typically, the amnesty is preceded by complex bureaucratic and decision-making processes. Two important parameters that must be determined during this phase are the total number of prisoners to be freed and the set of criteria for early release. These criteria may include the severity of the crime, the portion of the sentence a prisoner has already served, their behavior while in prison, and the likelihood that they will re-offend if released. Formerly, the authorities made such multi-dimensional sensitive eligibility comparisons without using any systematic, objective method. Thus, they faced public criticism and anger.

This article demonstrates the use of two-dimension multi-criteria decision analysis (MCDA) as a simple method to help authorities decide which prisoners to pardon.

The MCDA approach is a popular and user-friendly tool for choosing among alternatives. As such, this technique is widely used to make decisions in diverse fields, such as choosing between investment alternatives [3], locating electric vehicle charging stations [4], evaluating the performance of inland ports [5], allocating budgets for political campaigns [6], and even assessing steam boiler alternatives [7]. The approach has also garnered a great deal of attention in the academic literature, as seen in [8, 9, 10]. The criteria weights -- that is, their relative importance -- were calculated using the basic principle of the Analytic Hierarchy Process (AHP), the pairwise comparisons [11, 12].

The method was implemented in Excel and produced clear and straightforward results in a convenient format that is easy to understand and to use.

The rest of the article is organized as follows: the methodology is explained in the next section; then, the results and conclusions are presented.

2. Methodology

Ahead of Independence Day, the authorities in a certain country announced a wide-ranging amnesty for prisoners. The decision-makers defined the following criteria for pardon: gender, age, severity of crime, proportion of sentence already served, and behavior in the last three years. In addition, the authorities wanted to consider the prisoner’s dominance in prison as well as their stature in the local underworld.

Although the criteria were well defined, the method of translating them into a set of unbiased decisions had to be determined in order to produce the list of pardoned prisoners. Thus, the authorities were looking for a simple but reliable method to assist in decision-making, and MCDA was suggested as an effective model. We applied it using three stages: First, we decided on the criteria; second, we applied an AHP, comparing each criteria-pair to calculate the relative criteria weights. Third, we used these weights to produce a list of prisoners eligible for pardon. In the following paragraphs, we detail these stages.

2.1 Stage 1: setting the criteria

The criteria were set in cooperation with the authorities. We asked them to list the factors that they thought should be considered in the pardon process and explain their preferences. This stage entailed a few rounds of discussions and enabled us to finalize the criteria, as follows (in alphabetical order):

Age - the authorities preferred to grant pardons to prisoners that were younger than 21 and older than 67. This is because youngsters are prone to making impulsive decisions that they later regret [13]. As a result, they were seen as more amenable to rehabilitation and a lesser risk to public safety. On the other hand, older prisoners were seen as a lesser threat to public safety due to their advanced age and declining health. They were also seen as less likely to re-offend because of their age and the fact that they have already spent a significant portion of their lives in prison.

Behavior - the authorities preferred to grant pardons to prisoners who demonstrated exemplary behavior while in prison. This is because good behavior was seen as a sign that prisoners accepted responsibility for their actions, are remorseful, and are making an effort to change their ways. As a result, their odds of reintegration into society were considered high. The contrary also holds true: the authorities preferred not to grant early release to prisoners who exhibited unruly behavior in prison. This was interpreted as a failure to accept responsibility for wrongdoing, and a sign that the prisoners are not remorseful and are making no effort to reform [14, 15]. Thus, releasing such prisoners would not promote reintegration into society while posing a heightened risk to public safety.

Crime severity - the authorities were willing to grant early release to non-violent prisoners who committed non-violent crimes (e.g., white collar offenses, theft, etc.). Officials were less inclined to pardon prisoners who committed severe crimes. Specifically, prisoners with “blood on their hands” were excluded from the amnesty campaign. The authorities believed that releasing these prisoners would likely cause a public uproar, and that such offenders are more dangerous to society.

Gender - the authorities showed greater willingness to release females compared to males. This is because the number of female prisoners was much lower, and they were believed to pose a lower risk to public safety due to their limited physical strength and generally lower levels of aggression. In this context, the authorities viewed non-binary prisoners the same way as females.

Health condition – authorities preferred to grant early release to prisoners with serious health problems. This is because their declining health and limited strength usually decreased the threat they posed to public safety [16, 17]. Considerations such as the costs of medical treatment in prison and the risk of sudden deterioration or death while in jail also carried substantial weight for this preference.

Portion of sentence served - the authorities viewed the portion of the sentence a prisoner has already served as a relevant factor when deciding whether to grant them a pardon. Officials pointed out that the larger this portion was, the less they were concerned about the potential risk to public safety. More importantly, public opinion was more favorable toward releasing such inmates.

Hierarchies among inmates are common in prison [2]. Thus, the authorities wanted to add one more criterion: the prisoner’s dominance within the hierarchy. That is, whether the prisoner is a leader who enjoys an elevated status among the other prisoners and has loyalists inside and/or outside the prison who follow his orders. Early release of such prisoners may have a broader effect on society given their leading role and ability to initiate and direct criminal activity. The authorities were initially unsure which policy to apply to this group, so it was essential to distinguish this criterion from the others.

The complete list of criteria, values and preferences is given in Table 1.

Criterion	Values	Preferences	Comments
Age	The prisoners’ age, ranging from youngest to oldest	The authorities preferred to pardon younger and older prisoners	The ages of “younger” and “older” prisoners may vary, but must be below 21 and above 67
Behavior	Good; Average; Bad	Good behavior increased the odds of early release	The assessment was based on the prison records of inmates
Severity of crime	Minor offenses (white-collar, theft); Intermediate (robbery, drugs); Severe (murder, aiding and abetting murder)	The authorities excluded prisoners with “blood on their hands”	Some inmates who aided and abetted serious crimes were eligible for pardon, depending on their specific background
Gender	Female; Other; Male	The authorities preferred to pardon females/other prisoners
Health condition	Healthy; Minor health problems; Major health problems	Severe health problems increased the prisoner’s odds of getting a pardon	The health condition was determined by the prison’s doctor
Portion of sentence served	Ranged from 10–90%	Prisoners who served most of their sentence were more likely to be pardoned	Prisoners who served a negligible portion of their sentences were not eligible for a pardon. Prisoners who nearly completed their sentences were automatically pardoned
Dominance	-High -Medium -Little or none	The preferences were not unequivocal

Table 1.

List of criteria, values, and preferences (in alphabetical order).

2.2 Stage 2: weighing the criteria

The criteria were weighed using AHP. According to this method, the decision-makers determine their preference for each pair of criteria using a numeric scale ranging from 1 to 9 (see Table 2).

Degree of preference	Numeric value
Equal	1
Equal to moderate	2
Moderate	3
Moderate to strong	4
Strong	5
Strong to very strong	6
Very strong	7
Very strong to extreme	8
Extreme	9

Table 2.

Saaty’s scale of preferences [9].

It is common to use odd numbers to distinguish among measurement points. The use of even numbers is adopted only as a compromise between evaluators. The preferences are collected in a matrix, as shown in Figure 1. In this example, there are three criteria: C1, C2, and C3. The decision-maker extremely prefers C1 to C2, moderately prefers C1 to C3, and strongly prefers C3 to C2. The matrix is reciprocal because the lower triangle is the reverse of the upper triangle, and its diagonal is filled with “1”, comparing each criterion to itself.

Figure 1.

Preferences matrix.

After completing the matrix, a consistency check is performed to detect possible contradictions and fix them. Such inconsistencies may occur when the problem is vaguely defined or when the evaluator faces difficulties in maintaining consistency while comparing too many pairs. As the number of pairs to compare quadratically increases with the number of criteria, achieving consistency is indeed a challenge. An acceptable CR (consistency ratio) value should be less than 0.1.

In our case, we had six criteria (excluding “Dominance”); thus, 15 comparisons were required. To simplify the process for decision-makers, we interviewed three relevant officials and asked each to compare 10 pairs, so each pair got two independent preferences. This technique enabled us to simplify the process for time-constrained decision-makers while still recognizing gaps (or points of agreement) in their preferences. The decision matrix (based on adjusted comparison results) is presented in Figure 2. The resulting weights are shown in Table 3. The values were calculated using [18].

Figure 2.

Decision matrix.

Criterion	Rank	Weight
Crime severity	1	37.8%
Gender	2	27.1%
Health Conditions	3	18.6%
Age	4	7.9%
Portion of sentence	5	5.4%
Behavior	6	3.2%

Table 3.

Resulting ranks and weights.

Unexpectedly, the Portion of Sentence and Behavior criteria were marginalized, whereas the Gender criterion was given significant weight. We believe that this stemmed from official priorities. The officials estimated that policymakers wanted the broadest pardon possible, particularly for women. Thus, considerations such as Behavior or Portion of Sentence played a minor role in the decision, whereas gender played a major role.

2.3 Stage 3: selecting prisoners to pardon

Ranking the eligibility of prisoners for early release was performed as follows: First, we defined the possible scores for each criterion based on the preferences of the authorities (see Table 4). Second, each pardon candidate was graded for each criterion. Third, a prisoner’s overall score was obtained by calculating the score for each criterion based on its weight and summing up the scores. The prisoners were then listed in descending order based on their overall scores, thus enabling the authorities to consider this score in tandem with the dominance ranking to make the pardon decision.

Criterion	Values	Possible scores
Age	17 to 75	17 or 75: 100; 17 < x < 75: 100 – min(75-x, x-17)*100/29
Behavior	Good, Average, Bad	100, 50, 0
Crime severity	Modest, Intermediate, Severe	100, 50, 0
Gender	Female, Binary, Male	100, 100, 0
Health conditions	Major problems, Minor problems, No problems	100, 50, 0
Portion of sentence	Ranged from 10–90%	Time already spent in prison/total punishment*100
Dominance	High; Medium; Little or none	The preferences were not unequivocal

Table 4.

Range of scores for the first six criteria.

For example, the maximum score (100) in the Age criterion was given to the youngest and oldest prisoners, who happened to be 17 and 75 years old, respectively. The minimum grade (0) was given to age 46, which was the average between those two edges. Each year below/above 46 increased the grade by 100/29 points.

To illustrate the process, assume a male prisoner named John. John is 55 years old. He was sentenced to 20 years for robbery 12 years ago. In the last three years, he has been well-behaved and prison authorities have a positive opinion of him. Although suffering from diabetes, he manages to balance it. He is not dominant in jail and prefers to remain outside the predominant groups.

Based on John’s data, we can determine his scores: He is 55, so his score for the Age criterion is approximately 69 (= 20*100/29); his score for Behavior is 100; his crime severity score is 50; his Gender score is 0. Similarly, John’s Health Condition score is 50, and his Portion of the Sentence score is 60 (=12/20*100).

John’s overall score is calculated by multiplying his grades by the criteria weights, as follows:

$$69·7.9\%+100·3.2\%+50·37.8\%+0·27.1\%+50·18.6\%+60·5.4\%=40.1\%$$E1

Note that for identical criteria values except for Gender, the score would be 67.2%. This gap clearly shows the authorities’ strong preference for early release of female prisoners.

3. Results

The approach and the tool developed for it highly met the authorities’ expectations. Thus, they decided to examine them on a list of potential prisoners for pardon, which included 1500 names. The highest score was 76.1 and the lowest was 4.2. The results map is shown in Figure 3 and Table 5. These results show two dominant ranges: in the 51–60 range and 21–30 range. Whereas the low range had less significance because such scores were not eligible for early release, the high range posed a dilemma for the authorities: including those prisoners in the amnesty program meant an early release of more than 850 prisoners, far above the original allocation. But not including them meant releasing only 60 prisoners – far below the initial allocation. To the author’s best knowledge, the solution was to set the threshold at a value between 41 to 50 and to deny a pardon to dominant prisoners.

Figure 3.

Overall scores in descending order.

Score range	Frequency	High dominance	Medium dominance
71–80	17	2	2
61–70	12	0	1
51–60	31	2	8
41–50	814	46	115
31–40	39	0	5
21–30	442	29	63
11–20	19	2	1
1–10	126	4	24

Table 5.

Score-dominance map.

4. Conclusions

In this paper, we presented an MCDA approach for determining the eligibility of prisoners for early release. The criteria were set by the authorities based on vast experience and seemingly solid justifications. The values and weights of each criterion were defined in sessions that combined professional and academic insights.

It should be noted that the complexity of calculating the weights, the difficulty of maintaining internal traceability, and the inherent subjectivity of the process are valid concerns voiced by critics of the AHP process [19, 20, 21]. However, most of these pitfalls were avoided in the current analysis because the score matrix was objectively calculated using independent external resources. Moreover, the process was undertaken transparently and systematically to ensure fairness.

The method was implemented on a list of 1500 candidates for early release through a wide-ranging amnesty. Several hundred achieved overall scores that made them eligible for a pardon.

Acknowledgments

The author would like to thank the IDOD for the opportunity to assist in the discussed topic. The prisoners’ data was intentionally left vague to respect the right to privacy. The author also would like to thank Mr. Yigal Walt for his editorial contribution.