Taiwan's Recidivism Rates and Prison Rehabilitation Efficacy: A Preliminary Inquiry
Last edited: 2026-07-27
Table of Contents
Summary
- Changes in the rates of recidivism and habitual offending among prison inmates can be attributed to numerous potential criminal policy factors and other causes, with particular attention required for adjustments to the "avoiding imprisonment for first-time offenders" policy and "diversionary treatment" policies. An increase in inmate recidivism rates may precisely be the outcome of implementing these policies.
- Under the understanding of the preceding point, the presence or level of prison rehabilitation efficacy is likely unsuited to be judged by the "inmate recidivism rate," but is more appropriately assessed through the "post-release recidivism rate."
- The "post-release (cumulative) recidivism rate" (rounded) of inmates released in Taiwan between 2011 and 2014 as of the end of 2015: within 6 months (13%), within 1 year (25%), within 2 years (41%), within 3 years (49%), within 4 years (54%). ── Within available data ranges, it does not reach as high as 70% or 80%.
- Judging by the rate of accumulation, the first 2 years after release represent the peak period for recidivism, which is the critical timeframe during which criminal policies should strengthen the allocation of resources to counsel rehabilitated individuals in reintegrating into society.
I. Introduction
Having carefully read the article by Attorney Chou Yu-hsiu published on March 17, 2017, in the ETtoday Cloud Forum, entitled "Throwing the Guilty Straight into Prison, Is That How It Works?" (the same article was published by the Taiwan Association for Human Rights on March 21), which stated: "According to statistical data from the Ministry of Justice, among newly admitted inmates in 2010 [104],... those with criminal records totaled 26,260, accounting for 77.5% of the total," and linked this to "the correction and moral instruction functions of correctional institutions indeed needing further enhancement."
These two statements are similar to the post-meeting records of the second meeting of the fifth division of the National Customs on Judicial Reform, which stated: "After visiting Yilan Prison, this division identified two problems: first, the recidivism and habitual offending rates of inmates are very high, accounting for about 83% of the inmate population at Yilan Prison...".
These statements also bear similarity to a remark in the major article by Judicial Reform Foundation researcher Lin Wei-ting, entitled "Prison Reform Policy Beginning with the End in Mind": "Regarding whether programs for social reintegration are effective... this can basically be understood through the 'recidivism rate.'... Current results lean toward pessimism. This can be indirectly inferred from the fact that 77.5% of inmates newly received by prisons in 2015 had criminal records."
Regarding the appeal put forward in these documents—"the state should invest more in correctional personnel and improve the living environment of inmates"—this author expresses wholehearted agreement. However, with respect to their direct or indirect inference of prison correctional efficacy based on the recidivism rates of newly admitted inmates (or all inmates), this author wishes to offer a conceptual clarification.
※ Note: Researcher Lin mentioned in footnote 2 that using inmate recidivism rates is merely an expedient measure in the absence of better recidivism data, utilized solely to raise preliminary concerns regarding prison rehabilitation efficacy. ── This supplementary explanation is correct.
II. Definition of Recidivism Rate Suitable for Inferring Rehabilitation Efficacy
1. Why the "Inmate Recidivism Rate" Is Inappropriate
This author contends that using the recidivism rates of newly admitted inmates (or all inmates) to infer prison correctional efficacy is an inappropriate indicator. The explanation is as follows:
[Example 1]
Assumptions:
- Prior to time point T1, no one is in prison.
- At time point T1, P individuals (all first-time offenders) are admitted to prison at once, while Q individuals (none with criminal records) remain outside the prison, with P + Q equal to the total population.
- At time point T2, A individuals among the P inmates are released from prison (leaving P - A individuals remaining in prison).
- At time point T3, B individuals among the released A individuals commit crimes again; furthermore, M individuals among the Q individuals originally outside the prison commit crimes (first-time offenses).
- At time point T4, C individuals among the B individuals have their convictions finalized, of whom D individuals enter prison (with criminal records); N individuals among the M individuals have their convictions finalized, of whom K individuals enter prison (without criminal records).
- Assume that apart from this, no other changes have occurred in the population figures either inside or outside the prison.
If we calculate the following two values at time point T4:
- Recidivism rate of newly admitted inmates: R1 = D / (D + K)
- Recidivism rate of all inmates: R2 = D / (P - A + D + K)
Because R1 and R2 incorporate D and K in their definitions, they are inevitably influenced by "diversion policies" or "imprisonment-avoidance policies" (along with the dark figure of crime). That is to say, the employment of measures such as discretionary non-prosecution, suspended prosecution, fines, community service, probation, and mandatory treatment will affect D and K, and consequently affect R1 and R2. Therefore, when we observe R1 and R2 soaring year after year, it does not necessarily mean that prison rehabilitation efficacy is becoming increasingly poor.
One possibility for the soaring of R1 and R2 is: the implementation of certain "diversion policies." For instance, if the policy to "avoid imprisonment for first-time offenders" is rigorously executed, K (located in the denominator) will be suppressed, causing both R1 and R2 to approach 100%. In other words, the increase in R1 and R2 may be entirely consistent with policy expectations.
Another possibility for the rise of R1 and R2 stems from: criminal policies showing no leniency toward recidivism for certain categories of crime that were previously subject to less incarceration; that is, the tightening of recidivism diversion policies for specific offenses. Examples include the 2004 (ROC year 93) amendment mandating custodial sentences for repeat offenders of drug use offenses. Another example is the 2013 (ROC year 102) Ministry of Justice letter ruling that third-time drunk driving offenders shall, in principle, not be permitted to commute sentences to fines. Successive statistical reports by the Ministry of Justice [1] also infer a correlation between these policies and R1 and R2.
Within the formulas for R1 and R2, such policies are equivalent to increasing D. With other items remaining constant, adding the same number to both the numerator and denominator results in the ratio moving toward 100%. This implies that: (1) As with the first possibility, the increase in R1 and R2 may conform to policy expectations; (2) Even if prison rehabilitation efficacy has actually improved during the same period (itself a proposition requiring validation), it might still fail to offset the impact of the "enhanced imprisonment policy for repeat offenses" tied to specific crimes, thereby manifesting as a higher proportion of recidivism among prison inmates.
In the examples of the two possibilities above, Attorney Chou's reference in his article to "the ones who keep coming back are old friends" may be because "policies happened to target certain old friends for prioritized readmission," and not necessarily because the recidivism rate among all old friends has increased.
Naturally, a third possibility also exists: the rise of R1 and R2 could be caused by declining prison rehabilitation efficacy.
The aforementioned three or more possibilities are not necessarily mutually exclusive; several causes may operate simultaneously to drive up R1 and R2. Whichever the case may be, further research is required for confirmation, and conclusions should not be rushed.
2. Why the "Post-Release Recidivism Rate" Is More Appropriate
If R1 and R2 are unsuitable because they are easily susceptible to policy interference, what definition of recidivism rate is comparatively appropriate for inferring prison correctional efficacy?
Due to the varying degrees of difficulty in acquiring data and the different phenomena focused upon by researchers, criminological research employs multiple operational definitions of recidivism. Research on rehabilitation efficacy more commonly utilizes the "post-release (or post-institutional) recidivism rate." If one considers solely prison rehabilitation efficacy, the denominator of this "rate" is the number of released individuals (A). Whether the numerator counts cleared cases (identified perpetrators), arrests, prosecutions, convictions, or imprisonments (all based on official data), or self-reported crimes by respondents, will vary across studies for the reason stated in the first sentence of this paragraph; however, this complexity may be temporarily set aside. ── Let us provisionally assume that all crimes are incorporated into the statistics, then:
Post-Release Recidivism Rate: R3 = B / A (for the meaning of the codes, see "Example 1")
R3 is more suitable than R1 and R2 for evaluating prison correctional efficacy because it possesses the following advantages:
-
R3 conforms to the intuitive definition of recidivism: number of re-offenders / number of released individuals.
-
R3 does not contain D or K in its definition, and thus, in most cases, remains unaffected by the fluctuations of "enhanced imprisonment policies" and "diversionary treatment policies."
-
Even if the operational definition of re-offending is "re-imprisonment" (substituting D for B), which would subject it to the aforementioned two policies, the interference suffered by R3 = D / A remains less than that suffered by R1 and R2. Illustrating this using R1 and R2 as counterexamples:
-
In addition to being affected by the aforementioned two policies through D, R1 and R2 must also account for: the number of first-time offenders (K), and the number of people in prison prior to time point T4 (P - A).
-
Since the number of first-time offenders (K) has never served time in prison, incorporating it into the formula naturally expresses not the rehabilitation efficacy of prisons, but rather other causes influencing crime.
-
In the real world, P will not consist entirely of first-time offenders as in Example 1, but will be a mixture of first-time and repeat offenders. Consequently, when calculating R2, one must determine the number of repeat offenders (let us assume E) within the prison population prior to time point T4 (P - A), turning the numerator of R2 into D + E while the denominator remains unchanged. ── However, because E has continuously resided in prison from T1 to T4, using E to infer the correctional efficacy of prisons during this period lacks meaning.
-
Continuing from the previous point, what E expresses is the prison correctional role from "some period prior to T1," which is the duration of E's previous prison term (assuming T-1 ~ T0). When we calculate R2 at time point T4, it is equivalent to performing a sort of weighted average of the prison correctional efficacy between D's previous prison term (T1 ~ T2) and E's previous prison term (T-1 ~ T0). Although we can calculate R2,t for every time point (t), the numerator (Dt + Et) of each R2,t expresses a weighted average of prison correctional efficacy from different periods, with the weight depending on what proportion of the previously incarcerated repeat offenders at each time point t had their previous prison experience in one period (Dt) versus another period (Et). The reality is even more complex, as each Et can be broken down into individuals whose previous prison experiences span multiple different periods. ── Thus, when we infer whether prison correctional efficacy has improved or worsened by comparing year-on-year changes in R2,t, we are not simply comparing differences across periods; the periods expressed by each R2,t vary in length and frequently overlap.
-
Furthermore, the prison correctional efficacy expressed by each Et is subject to the interference of criminal policies at the time they entered prison (in Example 1, earlier than T1). ── Consequently, each R2,t, in addition to reflecting the criminal policy interference of period t through the inclusion of Dt, also reflects, via a weighted average, the criminal policy interference of numerous prior periods through the inclusion of Et.
-
Similar to the initial prison population (P) which generates the problem of E in reality, the extra-prison population (Q) in reality will not be entirely devoid of criminal records. Therefore, a portion of K will consist of repeat offenders (let us assume L) and must be added to the numerators of R1 and R2. Akin to E, the information carried by L also encompasses a weighted average of prison correctional efficacy and criminal policy interference across multiple periods.
-
Synthesizing the above: because K is located in the denominators of R1 and R2, E is in the numerator and denominator of R2 (E being part of P - A), and L is in the numerator and denominator of R1 and R2 (L being part of K)—whereas R3 does not need to account for K, E, and L even in reality—this results in R1 and R2 carrying greater criminal or non-criminal policy interference than R3 in reality, with the prison correctional efficacy expressed by R1 and R2 (if interference is excluded) referring to complex and difficult-to-compare periods.
-
III. Taiwan's "Post-Release Recidivism Rate"
When calculating post-release recidivism rates in criminology, the convention is to calculate them based on the elapsed duration (or tracking period) following release, such as: the recidivism rate within 6 months post-release, within 1 year, within 2 years, etc. This is based on a common-sense principle: the longer the time elapsed since release, the greater the opportunities to re-offend, whereas individuals newly released have very limited opportunities.
The Ministry of Justice statistical report titled "Circumstances of Re-offending Among Inmates Post-Release" is also basically calculated according to the elapsed duration following release, and the resulting post-release recidivism rates are far below 70% (referring to Table 2 in Attachment 2 and the table below (sourced from Attachment 1)).
| Year of Release |
Number |
Number and Rate of Post-Release Recidivism - Categorized by Elapsed Duration of Recidivism | |||||||
| Total | 6 Months or Less |
Over 6 Months to Under 1 Year |
1 Year to Under 2 Years |
2 Years to Under 3 Years |
3 Years to Under 4 Years |
4 Years to Under 5 Years |
5 Years Or More |
||
| Number of Persons (Persons) | |||||||||
| 2011- 2015 |
175,429 | 67,351 | 21,558 | 17,783 | 17,881 | 6,926 | 2,650 | 553 | -- |
| 2011 | 36,474 | 20,151 | 5,017 | 4,490 | 5,225 | 2,970 | 1,896 | 553 | -- |
| 2012 | 34,373 | 17,584 | 4,621 | 4,183 | 5,278 | 2,748 | 754 | -- | -- |
| 2013 | 34,185 | 14,947 | 4,453 | 4,269 | 5,017 | 1,208 | -- | -- | -- |
| 2014 | 35,439 | 10,944 | 4,483 | 4,100 | 2,361 | -- | -- | -- | -- |
| 2015 | 34,958 | 3,725 | 2,984 | 741 | -- | -- | -- | -- | -- |
| Rate (%) | |||||||||
| 2011- 2015 |
100.0 | 38.4 | 12.3 | 10.1 | 10.2 | 3.9 | 1.5 | 0.3 | -- |
| 2011 | 100.0 | 55.2 | 13.8 | 12.3 | 14.3 | 8.1 | 5.2 | 1.5 | -- |
| 2012 | 100.0 | 51.2 | 13.4 | 12.2 | 15.4 | 8.0 | 2.2 | -- | -- |
| 2013 | 100.0 | 43.7 | 13.0 | 12.5 | 14.7 | 3.5 | -- | -- | -- |
| 2013 | 100.0 | 30.9 | 12.6 | 11.6 | 6.7 | -- | -- | -- | -- |
| 2014 | 100.0 | 10.7 | 8.5 | 2.1 | -- | -- | -- | -- | -- |
Explanation: 1. The number of recidivists in this table refers to those who committed offenses after being released from prison, and whose cases were concluded by prosecutors through indictment, application for summary judgment, suspended prosecution, discretionary non-prosecution, or referral for observation and rehabilitation or compulsory treatment up to the statistical cutoff date, indicating suspected criminal involvement. 2. "Elapsed duration of recidivism" refers to the time from the date of release to the date of newly registered case distribution for investigation. Furthermore, the number of recidivists is limited to those whose investigations were concluded before the statistical cutoff date; hence, the final interval group of the elapsed duration of recidivism for each year may not encompass all actual recidivists within that interval.
However, the figures from the Ministry of Justice suffer from a separate calculation problem, which stems from: including released individuals whose tracking period is insufficiently long in the calculations. Because inmates may be released each year, month, and day, the elapsed duration following release differs for each individual relative to the same point in time when the recidivism rate is calculated. This can be illustrated with an example:
[Example 2]
Assume X is released on January 1, 2010, at 00:00, and Y is released on June 30, 2010, at 00:00; and assume X re-offended prior to January 1, 2011, at 00:00, while Y did not.
When taking January 1, 2011, at 00:00 as the cutoff point for recidivism to calculate the "recidivism rate within 6 months post-release," both X and Y can certainly be included in the calculation. However, when we are calculating the "recidivism rate within 1 year post-release," strictly speaking, only X can be included, because: if Y is included in the calculation, Y will enter the denominator (released persons) but not the numerator (since he did not re-offend); yet if Y is observed until completing 1 full year post-release, based on the aforementioned common sense principle—he still has opportunities to re-offend prior to June 30, 2011, at 00:00—Y will have the opportunity to enter the numerator and elevate the calculation result of the recidivism rate.
── In other words, prematurely including Y in a period he has not actually fully experienced may underestimate the true recidivism rate for that period.
In Table 2 of Attachment 1 and Attachment 2, the Ministry of Justice committed the calculation error described in Example 2. Fortunately, the Ministry of Justice possesses the raw data on the annual number of released persons and recidivism times, which allows for the correct formulas to be applied to recalculate the "post-release recidivism rate." Based on the data of inmates released between 2011 and 2015 provided by the Ministry of Justice, observed through the end of 2015, Taiwan's post-release (cumulative) recidivism rates are as follows:
| Cumulative Recidivism Rate | ||
| Elapsed Duration Post-Release | Methodology of This Article | Methodology of Ministry of Justice |
| 6 Months or Less | 13.2% | 12.3% |
| Within 1 Year | 25.4% | 22.4% |
| Within 2 Years | 40.5% | 32.6% |
| Within 3 Years | 48.7% | 36.5% |
| Within 4 Years | 53.7% | 38.0% |
| Within 5 Years | Tracking period incomplete, unable to calculate | 38.4% |
Figure: Comparison of Cumulative Recidivism Rates Between the Methodology of This Article and the Ministry of Justice
Preliminary observations regarding these figures:
-
As predicted by the analysis in Example 2, the Ministry of Justice's calculation method underestimated the post-release recidivism rate. Moreover, the longer the tracking period, the more severe the underestimation, due to the accumulation of underestimations from preceding periods.
-
This is a "cumulative" recidivism rate, which inevitably accumulates higher as the tracking period lengthens. The Ministry of Justice table utilizes the "marginal" recidivism rate of each period, and summing them yields the cumulative recidivism rate.
-
Judging from the speed of accumulation, the first 2 years post-release represent the peak period of recidivism, which is the critical timeframe during which criminal policy should intensify the investment of resources to counsel rehabilitated individuals in reintegrating into society.
-
Why is the "within 5 years" column in this article's calculation method uncalculable? Because even for the earliest possible release date in the data (January 1, 2011), at the time the Ministry of Justice calculated recidivism (the end of 2015), strictly speaking, there was still a short period remaining before the completion of the 5-year tracking period. That is, there remained an opportunity to re-offend (see "Example 2"), so it should not be factored into the "within 5 years" column, let alone for individuals released after this date. In fact, the Ministry of Justice's algorithm incorporates individuals whose tracking period is incomplete into the denominator during the final year of each tracking period (the bottom-most cell of each column), thereby underestimating the recidivism rate for that column (e.g., individuals released from July 1 to December 31, 2015, were all included in the denominator for "6 months or less," and those released from January 1 to June 30 of the same year were all included in the denominator for "over 6 months, under 1 year").
These figures are not as high as many people assume (e.g., 70, 80, 90%). However, please note:
-
These figures are based solely on data from 2011-2015, and the longer the tracking period, the fewer data points available. Therefore, data spanning more years must be accumulated to achieve a more accurate understanding of recidivism rates in Taiwan.
-
"Recidivism" as defined in the Ministry of Justice data refers to those suspected of criminal involvement who were "prosecuted, subjected to applications for summary judgment, granted suspended prosecution, granted discretionary non-prosecution, or referred for observation and rehabilitation or compulsory treatment," rather than all having finalized convictions of guilt. Therefore, defining recidivism by the "re-conviction rate" might yield lower figures. Similarly, not all individuals are imprisoned following conviction, so the "re-imprisonment rate" might be even lower.
IV. International Comparison
Referring to Fazel and Wolf (2015) (doi: 10.1371/journal.pone.0130390), within the scope of data collected in that article, examples include (with countries and release years indicated in parentheses):
-
Cumulative "Re-conviction Rate" Post-Release
- Within 2 years: Lowest 20% (Norway, 2005); Highest 59% (England and Wales, 2000).
- Within 3 years: Lowest 45% (United States, 2005-2010); Highest 66% (England and Wales, 2000).
-
Cumulative "Re-imprisonment Rate" Post-Release
- Within 2 years: Lowest 29% (United States, 2005-2010); Highest 39% (Australia, 2009-2010).
- Within 3 years: Lowest 24% (South Korea, 2002); Highest 50% (Chile, 2007).
-
Data for more countries and different tracking periods can be found in Fazel and Wolf (2015).
V. From "Post-Release Recidivism Rate" to Prison Rehabilitation Efficacy
Regardless of whether Taiwan's post-release recidivism rate ranks as excellent, moderate, or lagging internationally, it cannot be proven solely through rankings whether the rehabilitation and moral instruction functions of prisons exist or whether they are good or poor. Proving rehabilitation effects requires numerous pieces of evidence; calculating recidivism rates is merely the first step, and it is hoped that there will be ample time to write further on this in the future.
To reiterate in closing: regarding the appeal raised by Attorney Chou Yu-hsiu, Researcher Lin Wei-ting, and the fifth division of the National Congress on Judicial Reform—"the state should invest more in correctional personnel and improve the living environment of inmates"—this author expresses wholehearted agreement.
Footnotes
[1] 〈Statistical Analysis of Recidivism Among Newly Admitted Inmates〉 and 〈Criminal Record Circumstances of Prison Inmates〉
Attachments
This article is an AI-generated translation of the original Chinese text. In the event of any discrepancy, the Chinese version shall prevail.