Table of Contents | Previous | Next |
Impacts on Employment and Recidivism for the Full Sample
This section discusses CEO’s impacts on selected measures of employment and recidivism. As discussed earlier, the impacts are assessed by comparing outcomes for the program and control groups during a one-year period after sample members entered the study (referred to as Year 1).
Employment
Table 4 shows CEO’s impact on employment during Year 1. Employment rates are shown for the full year and by quarter, with Quarter 1 defined as the quarter in which each sample member was randomly assigned.1
The top panel of the table, based on data from CEO’s management information system, only includes employment in CEO’s Neighborhood Work Project (NWP) transitional jobs. Consistent with the study design, almost no one in the control group worked in NWP. The previous section noted that about 70 percent of the program group worked in an NWP transitional job, with most people starting shortly after random assignment and remaining in NWP positions for less than three months. Thus, it is not surprising that 63 percent of the program group worked in NWP in Quarter 1 and 43 percent worked in NWP in Quarter 2.
| Outcome (%) | Program Group | Control Group | Difference (Impact) | Standard Error | |
|---|---|---|---|---|---|
| Neighborhood Work Project (NWP) employment | Ever employed in NWP Q1-Q4a | 69.5 | 2.6 | 66.9 *** | 2.383 |
| NWP employment Qtr 1 (quarter of random assignment) | 63.0 | 1.3 | 61.7 *** | 2.461 | |
| NWP employment Qtr 2 | 43.3 | 2.0 | 41.3 *** | 2.499 | |
| NWP employment Qtr 3 | 14.0 | 0.1 | 13.9 *** | 1.702 | |
| NWP employment Qtr 4 | 5.3 | 0.5 | 4.8 *** | 1.122 | |
| All UI-covered employment | Ever employed in Q1-Q4 | 80.1 | 56.1 | 24.0 *** | 2.900 |
| Employed Qtr 1 (quarter of random assignment) | 65.8 | 25.9 | 39.9 *** | 3.000 | |
| Employed Qtr 2 | 59.2 | 34.7 | 24.5 *** | 3.200 | |
| Employed Qtr 3 | 41.6 | 35.4 | 6.1 * | 3.200 | |
| Employed Qtr 4 | 33.3 | 33.8 | -0.6 | 3.100 | |
| Employed all 4 quarters Q1-Q4 | 21.7 | 10.9 | 10.8 *** | 2.400 | |
| Sample size (total = 977) | 568 | 409 | |||
| SOURCES: MDRC calculations from CEO's Network for Information Gathering Evaluation and Learning (NIGEL) database and unemployment insurance wage records from New York State. NOTES: Includes sample members randomly assigned between January 2004 and October 2005. A total of three people were excluded from the UI-covered employment because of invalid social security numbers. Results in this table are weighted by week of random assignment and adjusted for pre-random assignment characteristics. Significance levels are indicated as: *** = 1 percent; ** = 5 percent; * = 10 percent. The significance level indicates the probability that one would be making an error in concluding that there is a difference between research groups for the variable in question. The standard error is an estimate of the sampling variation of the impact and is used to assess its level of statistical significance, as shown by the p-value. a Nine control group members were employed in NWP during the follow-up period. |
The bottom panel of Table 4, drawn from New York State unemployment insurance records, shows all UI-covered employment, including both NWP transitional jobs and non-NWP jobs. The table shows the percentage of sample members who had any UI-covered employment in the year or the specified quarter. New York State did not provide individual-level data on earnings, so no earnings results are shown. MDRC has recently obtained data from the National Directory of New Hires, which include individual-level earnings; analyses based on those data will be presented in later reports.
Table 4 shows that a little more than half of the control group worked in a UI-covered job at some point in the year following random assignment, but the quarterly employment rates were much lower —only 26 to 35 percent — indicating that many control group members did not work steadily, at least not in UI-covered jobs.
The table shows that CEO generated a very large increase in UI-covered employment in the early part of the follow-up period. In Quarter 1, 66 percent of the program group worked in a UI-covered job, compared with 26 percent of the control group. The difference, about 40 percentage points, is statistically significant at the 1 percent level, meaning that such a difference is quite unlikely to have occurred by chance. Taken together, the results in the two panels of the table suggest that NWP transitional jobs accounted for a large share of the program group’s employment in the first two quarters of the follow-up period. However, it is not possible to determine from these data how many program group members worked in non-NWP jobs.2
The results for these early quarters demonstrate the value of CEO’s transitional work model. The low employment rate for the control group during this period shows that relatively few CEO participants would be able to find a UI-covered job without NWP. However, the impact on UI-covered employment was relatively short-lived: it declined from 25 percentage points in Quarter 2 to 6 percentage points in Quarter 3, and then disappeared altogether by Quarter 4. Nevertheless, it is worth noting that the program group was about twice as likely as the control group to work in all four quarters of the year.
The main reason for the declining impact is the declining employment rate for the NWP group. (The control group’s employment rate rose somewhat between Quarter 1 and Quarter 2, but then remained relatively stable for the rest of the year.) The data cited earlier from CEO’s management information system show that less than half of those who worked in NWP were placed in a permanent job, so one would expect a significant fall-off in the overall employment rate for the program group. However, because of the limitations of the UI data described above, it is not possible to determine the extent to which program group members failed to transition from NWP into regular jobs, transitioned and then quickly lost their jobs, or transitioned to jobs not covered by unemployment insurance. Data from the New Hires database and the follow-up survey should allow a more complete analysis of this issue in later reports.
Recidivism
Table 5 shows CEO’s impacts on several measures of recidivism. The data provide a relatively complete picture of arrests, convictions, and prison stays in New York State, but they do not include information on admissions to New York City jails; these data will be incorporated into later analyses.
The first panel of the table shows data on arrests and convictions during Year 1, while the second panel shows one parole status measure — the percent of each group whose parole status was ever classified as “absconded” or “revoked.” The third panel shows data on admissions to state prisons during the year, and the bottom panel shows the status of sample members at the end of Year 1 with respect to parole and state prison.
Given national statistics on recidivism for released prisoners, it may seem surprising that only 24 percent of the control group was arrested in Year 1 of the follow-up period.3 However, it is important to recall that many sample members had already been out of prison for a substantial amount of time before coming to CEO, so Year 1 of the study period often does not correspond to the first year out of prison. Thus, these data are not directly comparable to studies of cohorts of recently released prisoners. Moreover, the sample only includes people who showed up at CEO’s offices after a referral from a parole officer, perhaps a relatively motivated group. Finally, it is worth noting that 37 percent of the control group was either arrested or absconded from parole or had their parole revoked during Year 1 (not shown in the tables).
Table 5 shows that, overall, CEO generated few statistically significant impacts on recidivism measures. However, there are small but statistically significant decreases in the percent convicted of a felony during Year 1 and in the percent incarcerated in state prison for a new crime during that period — the two measures that have the most direct implications for public safety. There are no statistically significant impacts on arrests, parole measures, or overall incarceration.4 It is notable that there is no statistically significant impact on arrests during the first three months after random assignment, a period when the program group was much more likely to be working in a UI-covered job (although not shown in the table, there was also no impact on parole outcomes during this period).5 This suggests that the link between employment and rearrest is not straightforward. The bottom panel of Table 5 shows that more than 80 percent of sample members in both groups were still under parole supervision at the end of Year 1, indicating that the story of CEO’s impact on recidivism is still unfolding.
| Program Group | Control Group | Difference (Impact) | Standard Error | |||
|---|---|---|---|---|---|---|
| Arrests and convictions | Arrested (%) | 22.7 | 24.2 | -1.5 | 2.800 | |
| Average number of arrestsa | 0.3 | 0.3 | 0.0 | 0.046 | ||
| Number of months between random assignment and first arrest (%) | 3 months or less | 5.3 | 5.0 | 0.3 | 1.500 | |
| 4-6 months | 4.8 | 6.0 | -1.2 | 1.500 | ||
| 7-9 months | 8.4 | 6.6 | 1.8 | 1.800 | ||
| 10-12 months | 4.2 | 6.6 | -2.4 | 1.500 | ||
| Convicted of a crime (%)b | 16.7 | 17.8 | -1.1 | 2.500 | ||
| Convicted of a felony (%) | 1.8 | 3.7 | -2.0 * | 1.100 | ||
| Convicted of a misdemeanor (%) | 12.1 | 11.4 | 0.7 | 2.100 | ||
| Other convictions (%) | 3.2 | 3.2 | 0.0 | 1.200 | ||
| Parole outcome | Absconded/revoked from parole (%)c | 20.2 | 20.5 | -0.3 | 2.600 | |
| Incarceration in state prison | Incarcerated in state prison (%)d | 11.1 | 13.9 | -2.7 | 2.200 | |
| Incarcerated for a new crime (%)e | 0.7 | 3.1 | -2.4 *** | 0.900 | ||
| Incarcerated for a parole technical violation (%)f | 8.2 | 9.4 | -1.2 | 1.800 | ||
| Incarcerated for all other/unknown reasons (%)g | 2.2 | 1.4 | 0.8 | 0.900 | ||
| Average number of months incarcerated in prison | 0.5 | 0.5 | -0.1 | 0.098 | ||
| Status at the end of Year 1 | Incarcerated in state prison (%)h | 7.1 | 9.8 | -2.7 | 1.800 | |
| Not incarcerated and on parole (%)i | 82.0 | 80.8 | 1.2 | 2.600 | ||
| Not incarcerated and not on parole (%)j | 10.9 | 9.4 | 1.5 | 2.000 | ||
| Sample size (total = 974) | 565 | 409 | ||||
| SOURCE: MDRC calculations using data from the New York State (NYS) Division of Criminal Justice Services (DCJS). NOTES: Includes sample members randomly assigned between January 2004 and October 2005. A total of three people were excluded from the sample because they did not match to the DCJS data. Results in this table are weighted by week of random assignment and adjusted for pre-random assignment characteristics. Significance levels are indicated as: *** = 1 percent; ** = 5 percent; * = 10 percent. The significance level indicates the probability that one would incorrectly conclude that a difference exists between research groups for the corresponding variable. The standard error is an estimate of the sampling variation of the impact and is used to assess its level of statistical significance, as shown by the p-value. a Each arrest date is only counted as a single event. If there are multiple crimes or charges on the same date, only one event per date is recorded in the analysis. b A total of 21 convictions (less than 1%) were found to be associated with an arrest that occurred prior to random assignment. c Measure includes those that absconded and those with a discharge type of "Revoked," according to data from the NYS Division of Parole. d All prison incarcerations (regardless of parole status) according to data from the NYS Department of Corrections. Due to multiple incarcerations, subcategories are not mutually exclusive and may sum to a total greater than the "Incarcerated" measure. e Incarcerations for those committing new crimes, regardless of their parole status. f Incarcerations for those committing technical violations while on parole. g All other incarcerations, including those for technical violations while not on parole and for other unknown/unspecified reasons. h Incarceration status based on month 12 after random assignment. i "Not incarcerated/on parole" includes those not incarcerated and without any parole discharge data in the 12 months after random assignment. j "Not incarcerated/not on parole" includes those not incarcerated in Month 12 who also have a parole discharge in the 12 months after random assignment. |
Table 6 combines the data on UI-covered employment with the data on incarceration in state prison. In general, these results mirror the employment impacts discussed earlier, with large initial positive effects that diminished over time.6 It is striking that, in the last quarter of Year 1, well over half of the sample in both the program and control groups was neither employed in a UI-covered job nor in state prison. The follow-up survey and data from the New York City jail system may help to describe the circumstances of these individuals.
| Outcome | Program Group | Control Group | Difference (Impact) | Standard Error | |
|---|---|---|---|---|---|
| Employed and not incarcerated (%) | Quarter 1 (quarter of random assignment) | 57.6 | 23.9 | 33.8 *** | 3.000 |
| Quarter 2 | 59.0 | 34.6 | 24.3 *** | 3.200 | |
| Quarter 3 | 41.5 | 35.1 | 6.4 ** | 3.200 | |
| Quarter 4 | 33.2 | 33.6 | -0.4 | 3.100 | |
| Employed and incarcerated (%) | Quarter 1 (quarter of random assignment) | 7.9 | 2.0 | 5.9 *** | 1.400 |
| Quarter 2 | 0.0 | 0.0 | 0.0 | 0.000 | |
| Quarter 3 | 0.1 | 0.3 | -0.2 | 0.300 | |
| Quarter 4 | 0.1 | 0.3 | -0.2 | 0.300 | |
| Not employed and not incarcerated (%) | Quarter 1 (quarter of random assignment) | 26.8 | 61.2 | -34.4 *** | 2.900 |
| Quarter 2 | 38.6 | 62.2 | -23.6 *** | 3.200 | |
| Quarter 3 | 51.6 | 58.6 | -7.0 ** | 3.300 | |
| Quarter 4 | 57.4 | 54.9 | 2.4 | 3.200 | |
| Not employed and incarcerated (%) | Quarter 1 (quarter of random assignment) | 7.6 | 12.9 | -5.2 *** | 1.800 |
| Quarter 2 | 2.4 | 3.1 | -0.8 | 1.100 | |
| Quarter 3 | 6.7 | 5.9 | 0.8 | 1.600 | |
| Quarter 4 | 9.4 | 11.2 | -1.8 | 2.000 | |
| Sample size (total = 971) | 562 | 409 | |||
| SOURCE: MDRC calculations using data from New York State (NYS) unemployment insurance wage records and NYS Division of Criminal Justice Services (DCJS). NOTES: Includes sample members randomly assigned between January 2004 and October 2005. A total of six people were excluded from the sample; three were missing a social security number and three did not match to the DCJS data. Results in this table are weighted by week of random assignment and adjusted for pre-random assignment characteristics. Significance levels are indicated as: *** = 1 percent; ** = 5 percent; * = 10 percent. The significance level indicates the probability that one would incorrectly conclude that a difference exists between research groups for the corresponding variable. The standard error is an estimate of the sampling variation of the impact and is used to assess its level of statistical significance, as shown by the p-value. |
1 For example, for a sample member randomly assigned in February 2004, Quarter 1 is the first calendar quarter of 2004. For someone randomly assigned in May 2005, Quarter 1 is the second calendar quarter of 2005. (back)
2 Without individual-level earnings data or employer ID numbers, it is not possible to identify sample members who worked in both an NWP job and a non-NWP job in the same quarter; this is particularly likely to have occurred in Quarters 1 and 2. Thus, there is no way to determine how many program group members worked in non-NWP jobs (that is, it is not possible to subtract NWP employment from total employment to obtain non-NWP employment). However, data from the New Hires database will be used to examine this question in detail in subsequent reports. (back)
3 National data show that 44 percent of released prisoners are arrested within one year of release. (back)
4 A small number of the convictions in Year 1 were associated with arrests that occurred prior to random assignment. This should not affect the program impacts, but the outcome levels would be slightly lower for both groups if the measure counted only convictions associated with arrests after random assignment. (back)
5 There was a hypothesis that CEO might lead to an increase in parole revocations because program group members working in NWP were, in effect, being monitored more closely than control group members. The results in Table show that this did not occur. (back)
6 The status “employed and incarcerated” may seem counterintuitive. However, most of the people in this status in Quarter 1 were released from prison and began working in an NWP job during the same calendar quarter. (back)
| Table of Contents | Previous | Next |