Does Excluding Major Crimes from Open Prisons Make Society Safer?

Origins of the Draft

On August 22, 2022, a major criminal case occurred in Taiwan where Lin Hsin-wu, a fugitive from the Mingde Open Prison, killed a police officer with a knife, shocking Taiwanese society. Public opinion strongly questioned whether the threshold for selecting inmates for open prisons was too low, and why robbers and inmates convicted of major violent crimes were allowed into open prisons. Under public pressure, the Executive Yuan swiftly passed a partial amendment draft to the 'Outlying Prison Act' (hereinafter referred to as the 'Executive Yuan Draft Version') about a month after the incident, and submitted it to the Legislative Yuan for deliberation.

The core amendment of this draft revolves around modifying the selection thresholds for open prisons to reduce the risk of open prison inmates absconding or failing to return on time, which poses a threat to public order and social security. While the purpose of this legal amendment is understandable, the primary method adopted is to exclude inmates with specific offenses, sentences, and crime categories from the right to participate in open prison selections. From a risk assessment perspective, it is questionable whether such an approach helps achieve the goal of maintaining social security.

Directly Determining Absconding or Recidivism Based on Offense, Sentence, and Crime Category: What is the Probability of False Positives?

From a risk assessment perspective, directly excluding inmates from participating in open prison selections based on specific offenses, sentences, and crime categories is essentially concluding that such inmates carry an extremely high risk of absconding or reoffending—so high that none of these inmates should be allowed to participate in open prison selections. All risk assessments entail the possibility of false positives (the assessment tool deems that the inmate will abscond or reoffend, but they actually will not) and false negatives (the assessment tool deems that the inmate will not abscond or reoffend, but they actually will). If tolerance for the risk of absconding or reoffending is extremely low, the cut-off point will be adjusted to classify more inmates into the high-risk category, which may simultaneously sharply increase the probability of false positives.

Understanding the probability of false positives in the open prison selection thresholds of the Executive Yuan Draft is not actually difficult. One simply needs to clearly list the specific offenses, sentences, and crime categories excluded by the Executive Yuan Draft, and calculate how many of these open prison inmates seized opportunities to abscond or failed to return on time (to understand recidivism, one would also need to calculate the crimes committed during the period of absconding or failure to return), versus how many exhibited no such behavior. This is basic data required for decision-making. Unfortunately, as the Executive Yuan Draft entered the Legislative Yuan, we cannot even see such basic data.

In fact, contemporary risk assessment tool designs rarely rely solely on specific offenses, sentences, and crime categories to determine risk, because the probability of false positives from this approach can be very high. For example, even for inmates with the same type of offense, sentence, and crime category, the rate of absconding or reoffending among women may be lower than that among men. If so, the aforementioned risk determination method would carry an even higher probability of false positives for female inmates.

The Problems Brought by False Positives in Open Prison Selection Thresholds: Not Reducing Recidivism, But Potentially Increasing It?

Beyond unduly restricting inmates' rights to receive appropriate treatment (note: outsiders often view open prisons as privileged, comfortable facilities, but they are actually appropriate treatment options for inmates with low-security management needs and important intermediate treatment measures), the greater problem brought by false positives in open prison selection thresholds is that this approach may not reduce recidivism rates, but could instead increase them.

The primary purpose and value of open prisons is to allow selected inmates to gradually adapt to the outside world in an intermediate treatment environment, and to have opportunities to return home for visits in accordance with laws and regulations, thereby strengthening family ties, reducing maladjustment after returning to society, and consequently lowering the possibility of reoffending. High false positives in open prison selection thresholds lead to the worst-case scenario: denying inmates appropriate intermediate treatment, thereby paradoxically increasing recidivism rates. Let us not forget that the vast majority of inmates will eventually return to society. Even if inmates are not sent to open prisons, they will still be paroled; if parole is denied, inmates will eventually complete their sentences and be released. Those released upon sentence completion have the highest recidivism rates. According to the Control Yuan's 2022 Judicial Investigation Report No. 0022 (page 58), the rates of parole revocation after release for open prison inmates from 2017 to 2011 (note: original text states 2011, likely a typo in source referring to 2021) were 3.8%, 2.6%, 1.4%, 0.4%, and 0.15% respectively, whereas the overall parole revocation rate for correctional institutions was 18%-19% (none of the aforementioned data specify the tracking duration). Furthermore, according to the Academy for the Judiciary, Ministry of Justice's Crime Conditions and Analysis in the Republic of China (2020), the recidivism rate for inmates released upon sentence completion from 2016 to 2020 was consistently higher than that for inmates released on parole. For those with a tracking period of two years at the time of statistics (i.e., 2016 to 2019), the recidivism rates within two years after release upon sentence completion versus release on parole are as follows:

Year	Recidivism Rate within Two Years After Sentence Completion Release	Recidivism Rate within Two Years After Parole Release
2016	44.33%	34.48%
2017	43.84%	33.87%
2018	39.03%	33.69%
2019	23.62%	14.83%

Although the aforementioned data remain rough (as it is difficult to distinguish whether it is the effect of the treatment itself or the result of the screening process), they nevertheless remind decision-makers that if major violent offenders are barred from selection into open prisons out of a belief that they pose a risk to public order and social security, the mindset may not result in lower recidivism rates, but rather higher ones. Source: Academy for the Judiciary, Ministry of Justice, Crime Conditions and Analysis in the Republic of China (2020), Table 2-4-11, 'Post-Release Reoffending Conditions of Inmates over the Past 5 Years'.

Lessons from Abroad: Empirical Research on Absconding from 'Open Prisons' in the UK

The UK also has an 'open prison' system similar to Taiwan's open prisons, alongside policy discussions on how to reduce absconding. Unlike Taiwan, however, the UK's policy discussions include empirical research on absconding incidents that can serve as a reference for decision-making. Below, two British studies on absconding from open prisons are cited to illustrate why empirical research is needed as a reference for decision-making when Taiwan seeks to narrow the scope of open prison selections.

Absconding from Open Prisons by Banks et al.

Published in 1975, this research primarily aimed to assist the UK Prison Department in reducing absconding incidents in open prisons and discussing whether the criteria for selecting inmates for open prisons needed adjustment. Banks et al. collected relevant data and found that male inmates in open prisons aged 21 to 24, those convicted of burglary compared to other crimes, and those who had served sentences of borstal training compared to those who had not, faced a higher risk of absconding. In addition, the risk of absconding increased with the number of previous convictions, the length of the current sentence being served, and the length of past sentences served. However, the length of the current sentence being served and the length of past sentences served exhibited an interesting interaction: for those serving short current sentences (18 months or less), the longer the past sentences served, the higher the absconding risk; for those serving moderate current sentences (18 months to 4 years) with past imprisonment experience, the longer the past sentence, the lower the absconding risk.

In 1970, utilizing preliminary research findings by Banks et al., the UK Prison Department conducted the 'Northern experiment', piloting new open prison selection criteria in the Northern prison region. They attempted to exclude inmates who had served borstal training, were convicted of burglary, were younger, and had more previous convictions from the scope of open prison selection (to the extent possible, though not entirely excluded). The result of the experiment was that the difference was not statistically significant.

Therefore, the research conclusion drawn by Banks et al. was that attempting to simply exclude high-risk categories of inmates from open prison selections to reduce absconding risk is not an effective approach. In fact, even among inmates classified as high-risk in the study, those who absconded were a relative minority. Strictly excluding these inmates according to the research findings would result in a considerable number of inmates who perform well in open prisons and are unlikely to abscond losing their eligibility to serve sentences in open prisons. Consequently, Banks et al. suggested not viewing their research findings as capable of 'predicting' the occurrence of absconding incidents, but rather treating high-risk characteristics as warnings of absconding tendencies, prompting more careful and cautious treatment of such inmates.

Analysis on the Impact of Absconding Incident History as a Predictor of Future Absconding Incidents by Mews

Published in 2014, this research was primarily prompted by the UK Ministry of Justice tightening the threshold for transferring inmates to open prisons, stipulating that if an inmate has absconded (unsupervised or supervised), attempted to abscond, or failed temporary release during their current sentence, they shall not be transferred to open facilities to serve sentences or be granted temporary release unless special exceptional circumstances apply.

Through studying absconding incidents prior to the regulatory adjustment on April 30, 2014 (including the various types of absconding mentioned above, same below), Mews found that past absconding incidents were a statistically significant predictor of future absconding incidents, but lost statistical significance when other variables were controlled. However, a record of temporary release failure was a much stronger predictor of future absconding incidents, remaining statistically significant even after controlling for other variables.

Both aforementioned studies pointed out situations that intuitive decision-making fails to uncover: Banks et al. discovered that for those serving moderate current sentences (18 months to 4 years) with past imprisonment experience, the longer the past sentence, the lower the absconding risk. Mews's research revealed that not all absconding incidents have the same predictive power for future absconding incidents; temporary release failure has stronger predictive power, whereas other types of absconding do not. Although the aforementioned research findings lack replication and sufficient theoretical foundation, Taiwan currently lacks even basic data collection, making it difficult to verify whether assumptions in intuitive decision-making align with basic factual realities.

Furthermore, the preliminary research findings of Banks et al. regarding high-risk categories of inmates were even directly adopted by the UK Prison Department to pilot new open prison selection criteria. The pilot results informed British policy-makers that it is difficult to reduce open prison absconding rates by setting simple selection exclusion criteria, and adopting such exclusion criteria would also cause many inmates currently performing well in open prisons and showing no inclination to abscond to lose their eligibility to serve sentences in open prisons. This conclusion is even more enlightening for Taiwan's current draft amendment to the Outlying Prison Act. This amendment draft similarly hopes to reduce the risk of absconding or reoffending in open prisons by simply excluding inmates with specific offenses, sentences, and crime categories from participating in open prison selections. Moreover, the UK had already conducted data analysis before running a pilot, yet the results still fell short of expectations, and review opinions pointed out that many individuals under the same conditions did not abscond at all. In Taiwan, it is entirely unclear what kind of data analysis formed the basis for deciding to exclude the targets listed in the draft from participating in open prison selections.

Conclusion: Post-Passion Review of Open Prison Issues Requires Deepening

Triggered by the open prison fugitive police-killing case, public criticism and questioning of open prisons surged. To respond to overwhelming public pressure, the Executive Yuan swiftly passed a partial amendment draft to the 'Outlying Prison Act' in about a month after the incident, excluding inmates who originally committed major crimes from participating in open prison selections. However, viewed from the design of contemporary risk assessment tools and the purpose of establishing open prisons, the Executive Yuan's amendment draft has major problems, lacks basic data to support decision-making, and has unclear empirical foundations.
We hope that the amendment draft to the Outlying Prison Act, hastily produced under intense public pressure, can return to rational policy discussion and review after the passion subsides, confirming the theoretical and empirical foundations of this draft, while reflecting back on a more important and fundamental question: if we expect inmates returning to society to try their best not to relapse, what should we do? If open prisons are intended for 'implementing phased treatment' (referencing Article 149 of the Enforcement Act of the Prison Act), how exactly should we act to enable open prisons to properly fulfill this function?