Control groups means and impact estimates in U.S.-based job training experiments

Experiments incorporating training

Those with training primary

Sector programs

Control mean

Impact

N

Control mean

Impact

N

Control mean

Impact

N

Program take-up (%)

10.7
(2.8)

58.3
(5.8)

140

8.6
(3.5)

66.1
(5.7)

78

13.5
(6.5)

55.5
(8.6)

17

Any training take-up (%)

40.3
(2.9)

19.2
(4.4)

95

49.1
(2.8)

28.1
(4.9)

58

53.8
(4.6)

24.5
(4.2)

8

Employment, short-term (%)

56.0
(3.0)

4.5
(1.4)

93

59.5
(4.3)

4.8
(1.7)

41

67.5
(6.0)

3.7
(1.4)

12

Employment, medium-term (%)

56.9
(2.8)

2.6
(0.4)

87

62.6
(4.0)

2.8
(0.6)

40

73.2
(4.8)

2.9
(0.9)

10

Employment, long-term (%)

59.8
(4.4)

1.8
(0.5)

64

63.4
(5.1)

1.7
(0.5)

36

80.3
(0.6)

1.3
(0.8)

6

Earnings, short-term (2025 $)

10,575
(1,194)

640
(244)

119

13,834
(2,038)

835
(553)

53

16,119
(1,048)

1,169
(1,092)

13

Earnings, medium-term (2025 $)

12,577
(1,020)

984
(255)

121

14,645
(1,473)

1,139
(446)

67

19,447
(938)

3,149
(1,260)

12

Earnings, long-term (2025 $)

14,361
(1,802)

657
(185)

81

16,148
(2,443)

791
(302)

53

25,137
(1,664)

3,703
(1,545)

9

Notes: Impacts are estimated with random-effects maximum likelihood (REML) meta-analysis. Control-group statistics are computed using the corresponding REML weights. Figures in parentheses below each estimate are standard deviations of control-group values and standard errors of impact estimates. Sample sizes vary because of missing data. Short-term outcomes have a ~1-year fol-low-up, medium-term ~2 years, and long-term ~3-5 years.