Author	Type of Study	Sample	Mean 35 Age (SD)	School‐based Programme	Universal/ Indicated 36	% FSM	Cluster 37	Extracted data for effect size calculations	Effect size calculation	Measure of exclusion
1) Allen et al. (1997)	Journal article	695	15.8 (1.13)	Teen outreach	Unclear	Unknown	No	Suspension Baseline T1: 58 cases (17%); N=342 C1: 81 cases (23.8%); N=353 Post treatment (immediately after treatment) T2: 42 cases (13%); N=324 C2: 93 cases (28.7%); N=323	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculations by adding the value of pre/post correlation, assumed to be equal .75. See methods section for further details	Presence/absence
2) Arter, (2005)	Journal article	52	Secondary school	Positive Alternative Learning Support (PALS)	Indicated	40% FSM	No	Suspensions No baseline reported Post Treatment (presumably after treatment) T2: M=.675; SD=.194; N=23 C2: M= 675; SD=.227; N=17	SMD was calculated using equations 3 and 4 in the methods section.	Nº of days
3) Barnes et al. (2003)	Journal article	45	16 (1.3)	Stress reduction	Universal	Unknown	No	Suspension Baseline T1: M=0.8 days; SD=1.8; N=23 C1: M=0.0 days; SD=.0; N=18 During intervention T2(during): M= 0.5 days; SD=1.2; N=23 C2(during): M=1.2 days; SD=3.0 N=18	The principal investigator provided N size for T2. SMD was calculated as the difference between time 1 and time 2, accounting for the covariation between pre‐ and post measures (equations 8 and 9 in the methods section).	Nº of days
4) Berlanga (2004)	PhD Thesis	80	Eighth grade	Grades, Attendance and Behaviour (GAB)	Indicated	Unknown	No	In School Suspension Baseline T1: M=1.15; SD=1.29; N=32 C1: M=.61; SD=1.14; N=31 Post Intervention (immediately after treatment) T2: M=1.12; SD=1.21; N=32 C2: M=1.03; SD=1.88; N=31 Suspension Baseline T1: M=.31; SD=.53; N=32 C1: M=.35; SD=.83; N=31 Post Intervention(immediately after treatment) T2: M=.34; SD=.90; N=32 C2: M=.58; SD=1.11; N=31 Removal/Expulsion Baseline T1: M=.06; SD=.24; N=32 C1: M=.06; SD=.24; N=31 Post Intervention (immediately after treatment) T2: M=.03; SD=.17; N=32 C2: M=.16; SD=.45; N=31	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculations by adding the value of pre/post correlation, assumed to be equal. 75. See methods section for further details	Nº of events
5) Bradshaw et al. (2012)	Journal article	12,334	Elementary school	School‐Wide Positive Behavioural Interventions and Support (SWPBIS)	Universal	49% FSM	Yes	Out‐of‐school suspension Follow‐up (four years) Student level Control: 21 schools (N=5124) Treatment: 16 schools (N=6614)OR= .73; 95% CI .59 and .91	The study reports results using multi‐level analysis. In this case, we have not applied any correction of standard errors. We assume that MLM accounted for clusters and subsequently corrected the bias (see p. e1140). See methods section for further details	Nº of events
6) Bragdon (2010)	PhD Thesis	68		Teach Team Project	Indicated	49% FSM	No	Suspension Baseline T1:M=.06; SD=.34; N= 34 C1:M=.96; SD=2.42; N=34 Post treatment (during) T2:M=.03; SD=.17; N= 34 C2:M=1.10; SD=2.61; N=34 Follow‐up (three months later) T3:M=.09; SD=.51; N=34 C3:M=1.07; SD=3.19; N=34	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculations by adding the value of pre/post correlation, assumed to be equal. 75. See methods section for further details.	Nº of days
7) Brett (1993)	PhD Thesis	126	12‐14 years	Efficacy, DC	Universal	Unknown	Yes	School Suspension Clusters: 3 control (n=66) 3 Experimental (n=60) Post‐treatment (1 Month) T2: M=.53; SD= 1.02; N=40 C2: M=.63; SD=1.21; N=57;	Based on Hedges (2007) and Spier et al. (2013), effect sizes were computed using dT2 , assuming equal cluster sample size, ρ=.05. See methods section for further details	Nº of events
8) Burcham (2002)	PhD Thesis	71	Middle school	Social problem solving skills training	Indicated	38% FSM	No	In‐School Suspension‐Baseline T1:M=8.62; SD=6.44; N=37 C1:M=7.88; SD=4.47; N=32 p=.58 ‐Immediately after treatment T2: f=.18; p= 0.67; N=69 ‐18 months after treatment T3: f=.04; p=0.84; N=38 Out‐of‐school Suspension Baseline T1: M=3.22; SD=3.71; N=37 C1: M=2.56; SD=3.40; N=32 p=.45 Immediately after treatment T2: f=1.09; p=.30; N=69 18 months after treatment T3: f= 1.83; p=.18; N=38	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculations by adding the value of pre/post correlation, assumed to be equal. 75. See methods section for further details	Nº of events
9) Collier (2002)	PhD Thesis	60	5‐14 years	Pro‐social skills training	Indicated	Unknown	No	School Suspension (+) Baseline T1: M=1.93; SD=.4498; N=26 C1: M=1.86; SD=.5074; N=25 Post treatment (presumably after treatment) T2:M=1.15; SD=.6748; N=26 C2:M=2.16; SD=.3742; N=25	Study was identified as an outlier value. It was winsorised as suggested by Wilson & Lipsey 2001. Effect size was calculated as the difference between baseline and post treatment. We corrected final calculation by adding the value of pre/post correlation, assumed to be equal .75.	Nº of days
10) Cook et al. (2014)	Technical report	106		BAM (skills‐training) and MATCH (tutoring)	Indicated	26% FSM	No	Baseline data is reported but incomplete Out‐of‐school suspension (ITT) b=‐.642; SE=.501; (unclear the number of months/weeks of post treatment measured)	Data was entered into CMA by using the option Log OR and its SE. No further corrections. Even if the evaluation was testing two different interventions, data was reported in a composite measure. The original author took that option because they recognised contamination between groups (spill‐over).	Nº of events
11) Cornell et al. (2012)	Journal article	201		Threat assessment	Indicated	Unknown	Yes	Long‐term suspension Post treatment (presumably after treatment) T2: 25 (25%); N=100 C2: 49 (49%); N=101	No corrections	Nº of events
12) Crowder (2001)	PhD Thesis	109		Gang Resistance, Education and Training (GREAT)	Unclear	Unknown	No	Out of School Suspension No baseline measure reported Post intervention (presumably after treatment) T2: M=.1329; SD=.4629; N=53 C2: M=.1429; SD=.1610; N=56 In School Suspension No baseline measure reported Post intervention (presumably immediately after treatment) T2: M=.3584; SD=.7464; N=53 C2: M=.4464; SD=.8464; N=56	SMD was calculated using equations 3 and 4 in the methods section. No further corrections.	Nº of events
13) Dynarski et al. (2003:2004)	Technical report	968	Elementary school	21st Century Community Learning	Unclear	Unknown	No	Post treatment 12 months (2003) T2: 7.1% (38) N=537 C2: 5.2% (16) N=317 Follow up 24 months T3: 60 (6.2%) N=537 C3: 43 (4.4%) N=317	Data was entered into CMA by using a 2x2 table. No further corrections.	Nº students
14) Edmunds et al. (2012)	Journal article	1607	15.3	Early College High School Academic skills enhancing	Unclear	50.6%	No?	% Suspended at least once T2: 6.4%; (57) N=885 C2: 13.3%; (86) N=644	The principal investigator provided measures for effect size calculation (via mail communication).	Nº of events
15) Farrell et al. (2001)	Journal article	626	11.7 (0.6)	Responding in Peaceful and Positive Ways (RIPP)	Universal	Unknown	Yes	In‐school suspension Post intervention (immediately after treatment) OR=5.0 (95%CI 1.5; 17.1) 6 months OR=1.4 (95% CI .7; 2.8) 12 months OR=1.4 (95%CI .6; 3.0) Out‐of‐School Suspension Post‐intervention (immediately after treatment) OR=0.9 (95%CI .5; 1.8) 6 months OR=1.1 (95% CI .6; 2.0 12 months OR=0.9 (95% CI .6; 1.4)	Although the study is based on clustered data, we have not applied any correction of standard errors. The author mentions the use of GEE to calculate robust estimates of standard errors (see Farrell et al., 2001, p. 455). See methods section for further details.	Nº of events
16) Feindler et al. (1984)	Journal article	36	13.8 (.68)	Anger control training	Indicated	Unknown	No	School Expulsion Baseline T1: M=1.45; SD=.71; N=18 C1: M=1.40; SD=.44; N=18 Five‐weeks follow‐up T2: M=.77; SD=.29; N=18 C2: M=.1.2; SD=.46; N=18	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculation by adding the value of pre/post correlation assumed to be equal .75. See methods section for further details.	Nº of events
17) Harding (2011)	PhD Thesis	48	Eighth grade	Over to you	Indicated	Unknown	No	Six‐months follow‐up Fixed term exclusion T2: 5 N=20 C2: 5 N=23 Permanent exclusion T2: 1 N=20 C2: 1 N=23	Data was entered into CMA by using a 2x2 table. No further corrections.	Nº of events
18) Hawkins et al. (1988)	Journal article	160	Seventh grade	Proactive Classroom Management	Indicated	Unknown	No	Times Suspended Post‐treatment (presumably after treatment) T2: M=.48; SD=1.3; N=67 C2: M=.89; SD=2.1; N=75	SMD was calculated using equations 3 and 4 in the methods section. No further corrections.	Nº of events
19) Hirsch et al, (2011)	Technical report	535	15.9	After School Matters	Indicated	86%	No	School Suspension Baseline T1: M=1.27; SD=.63; N=259 C1: M=1.29; SD=.64; N=178 Post treatment (immediately after treatment) T2: M=1.36; SD=.72; N=259 C2: M=1.40; SD=.75; N=178	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculations by adding the value of pre/post correlation, assumed to be equal .75. See methods section for further details.	Nº of events
20) Hostetler & Fisher (1997)	Journal article	317	Third grade	Project CARE (Skill for parents and children)	Indicated	Unknown	No	Suspension Baseline T1: M=0.13; SD=0.56; N=151 C1: M=0.07; SD=0.35; N=140 Post treatment (a few months after treatment, no clear specification) T2: M=0.20; SD=0.53; N=155 C2: M=0.25; SD=0.89 N=141 One year follow‐up T3: M=0.26; SD=0.80; N=90 C3: M=0.15; SD=0.66, N=86 Two year follow‐up T4: M=0.27; SD=0.74; N=30 C4: M=0.09; SD=0.29; N=34	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculations by adding the value of pre/post correlation, assumed to be equal .75. See methods section for further details.	Nº of events
21) Ialongo et al. (2001)	Journal article	678	6.20 (.34)	Two interventions i) Classroom‐centred (CC) ii) Family‐school partnership (FSP)	Universal	62.3% FSM	Yes	Suspension Five year follow‐up Classroom‐centred vs. control OR=.73; (95%CI=.56; .95) ** treatment group less likely to be suspended Family‐school partnership OR=.59 (95%CI .35; .97) Boys: OR=1.13 (95%CI .61; 2.09) Girls: OR=.38 (95%CI .17; .86) ** treatment group less likely to be suspended	Since data was dichotomous and nested in clusters, we corrected standard errors of effect sizes. The design effect was corrected by using the formula suggested by Higgins & Green (2011) expressed by the equation [1+(M‐1) x1]. See methods section for further details.	Presence/absence
22) Johnson (1983)	PhD Thesis	60	Seventh and eighth grade	ATTEND (Counselling and monitoring)	Indicated	Unknown	No	Suspension Baseline T1: M=.76; SD=.85; N=30 C1: M=.83; SD=.87; N=30 Post intervention (after treatment) T2: M=.36; SD=.55; N=30 C2: M=1.5; SD=1.25 N=30	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculations by adding the value of pre/post correlation, assumed to be equal .75.	Nº of events
23) Lewis et al. (2013)	Journal article	624	Elementary school	Positive Action	Universal	Grade 3 84% FSM	Yes	Suspension Baseline (2001) T1: M=40.95; SD=48.13; N=3648 C1: M=65.25; SD=56.15; N=3800 Post treatment (2004) T2: M=55.17; SD=64.84; N=3407 C2: M=77.63; SD=66.8; N=3687 Follow‐up (2005) T3: M=68.08; SD=80.02; N=3367 C3: M=88.96; SD=76.56; N=3539	The principal investigator provided data for calculations. Based on Hedges (2007) and Spier et al. (2013), effect sizes were computed using dT2 , assuming equal cluster sample size, ρ=.05.. See methods section for further details.	Nº of events
24) Mack (2001)	PhD Thesis	20	Fourth to sixth grade	ICAN Kids! Behavioural group counselling	Indicated	95% (for school, no stated for sample)	No	Out‐of‐school Suspension Baseline T1: M=1.5000; SD=.9718; N=10 C1: M=1.9000; SD=.8756; N=10 3 weeks (during) Tduring: M=.8000; SD=.6325; N=10 Cduring: M=.9000; SD=.7379; N=10 Post‐treatment T2: M=.3000; SD=.4830; N=10 C2: M=.4000; SD=.5164; N=10 + 3 weeks T3: M=.0000; SD=.0000; N=10 C3: M=1.0000; SD=.6667; N=10	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculations by adding the value of pre/post correlation assumed to be equal .75. See methods section for further details.	Nº of events
25) Obsuth et al. (2016)	Journal article	738	13.9	Engage in Education (Skills training)	Indicated	32%	Yes	Exclusion Official Records Baseline T1: OR=2.784; SE=.300; p=.001 Post‐treatment (1 month) T2: OR=1.444; SE=.389; p=.344	The study reports results using multi‐level analysis. In this case, we have not applied any correction of standard errors. We assume that MLM accounted for clusters and subsequently corrected bias (see p. 11). The study offered measures of impact based on self‐reporting, teachers’ reports and official records. We extracted from OR based on official records (most of our studies report official records of suspension).	Presence/absence
26) Okonofua et al. (2016)	Journal article	1682	Middle school	Empathic Discipline	Universal	Unknown	Yes	Suspension Post treatment (unclear number of months/weeks) T2: OR=.42; z= ‐3.33; p=.001; N=1449 31 clusters	Since data was dichotomous and nested in clusters, we corrected standard errors of effect sizes. The design effect was corrected by using the formula suggested by Higgins & Green (2011) expressed by the equation [1+(M‐1) x1]. See methods section for further details. Published data did not provide confidence intervals or SE. We tried to contact authors but it was not possible. We calculate an approximate SE=.013.	Presence/absence
27) Panayiotopoulos & Kerfoot (2004)	Journal article	124	10	Home and School Support Project (HASSP)	Indicated	Unknown	No	Exclusion T1: M=9.50; SD=14.81; N=62 C1: M=5.11; SD=7.56; N= 62 Post treatment (After three months) T2: M=4.95; SD=13.11; N=61 C2: M=5.51; SD=11.94; N=62	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculations by adding the value of pre/post correlation, assumed to be equal .75.	Nº days
28) Peck (2006)	PhD Thesis	1050	Fifth to eighth grade	Student Targeted with Opportunities for Prevention (STOP)	Unclear	Unknown	No	Suspension Post‐treatment (unclear number of weeks/months after treatment) T2: 22; N=315 C2: 22; N=321	Data was entered into CMA by using a 2x2 table. No further corrections.	Nº of events
29) Reese et al. (1981)	Journal article	98	Seventh to ninth grade	Preparation through Responsive Education Programs (PREP)	Indicated	Unknown	Matched peers	Suspension During school year T2 vs C2: X2 (1)= 6.58, p<.02	Data was entered into CMA by using X2 originally reported. No further corrections.	Nº of days
30) Russell (2007)	PhD Thesis	61	11.5 (.46)	Coping Power (Skills training for reducing aggression)	Indicated?	Unknown	No	Suspension Post treatment T2(School A): M=.15; SD=.38; N=13; C2(School B): M=.31; SD=.60; N=16; T2(School B: M=.29; SD=.61; N=14 C2(School B):M=.00; SD=.00;N=10	SMD was calculated using equations 3 and 4 in the methods section. No further corrections.	Nº of events
31) Shetguiri et al. (2011)	Journal article	108	14	Violence and drug use reduction	Indicated	100%	No	Suspended or Expelled Baseline T1: 8 (21%) N=40 C1: 10 (22%) N=46 Eight months follow‐up T2: 6 (14%) N=40 C2: 4 (8%) N=46	Effect size was calculated as the difference between baseline and post treatment. We corrected final calculations by adding the value of pre/post correlation, assumed to be equal .75.	Presence/absence
32) Smith (2004)	PhD Thesis	40		The Personal Responsibility Group (Emotional Intelligence skills)	Indicated	Unknown	No	In‐school Suspension Post‐treatment T2: f=11.085; p greater than or equal to .002; Out‐of‐school Suspension Post treatment T2: f= 10.088; p greater than or equal to .003	SMD was calculated based on f‐test. No further corrections.	Nº of events
33) Snyder et al. (2010)	Journal article	544	Elementary school children	Positive Action	Universal	55%	Yes	Suspension (% of students suspended) 2002 T1: M=1.12; SD=1.10; N=5000 C1: M=.98; SD=1.11; N=5000 2006 T2: M=.67; SD=.64; N=5000 C2: M=1.72; SD=1.55; N=5000 2007 T3: M=.84; SD=.61; N=5000 C3: M=2.53; SD=2.80; N=5000	The principal investigator provided sample sizes for calculations. Based on Hedges (2007) and Spier et al. (2013), effect sizes were computed using dT2 , assuming equal cluster sample size. Since the report presented the value of ρ, we used this value in calculations.	Nº of students
34) Sprague et al. (2016)	Unpublished paper	13,498	Middle school children	School‐Wide Positive Behavioural Interventions and Support (SWPBIS)	Universal	Unknown	Yes	Expulsion Baseline T1: M=.002; SD=.004; N=6492 C1: M=.003;SD=.004; N=7006 Post treatment T2: M=.002; SD=.004; N=6492 C2: M=.003; SD=.005; N=7006 Follow‐up (1 year later) T3: M=.003; SD=.006; N=6492 C3: M=.003; SD=.004; N=7006 In School Suspension Baseline T1: M=.071; SD=.094; N=6492 C1: M=.135;SD=.189; N=7006 Post treatment T2: M=.064; SD=.087; N=6492 C2: M=.097; SD=.133; N=7006 Follow‐up (1 year later) T3: M=.058; SD=.060; N=6492 C3: M=.095; SD=.145; N=7006 Out‐of‐School Suspension Baseline T1: M=.082; SD=.063; N=6492 C1: M=.078; SD=.065; N=7006 Post treatment T2: M=.076; SD=.077; N=6492 C2: M=.061; SD=.042; N=7006 Follow‐up (1 year later) T3: M=.073; SD=.064; N=6492 C3: M=.075; SD=.051; N=7006	Principal investigator provided data for calculations. Based on Hedges (2007) and Spier et al. (2013), effect sizes were computed using dT2 , assuming equal cluster sample size, ρ=.05. See methods section for further details.	Nº of events
35) Tilghman (1988)	PhD Thesis	100	12.5	Counsellor Peers	Indicated?	Unknown	No	Suspension T2: 11 (N=46) C2: 26 (N=45) Significance test	Data was entered into CMA by using a 2x2 table. No further corrections.	Nº of students
36) Ward & Gersten (2013)	Journal article	33 schools ≈ 25,000 students record	Elementary school children	Safe and Civil Schools	Universal	90%	Yes	Post treatment (end of intervention) OR=.83; SE=.05 Follow‐up (1 year later) (cumulative impact) OR=.77; SE=.04	Since data was nested in clusters, we corrected standard errors of effect sizes. The design effect was corrected by using the formula suggested by Higgins & Green (2011) expressed by the equation [1+(M‐1)x1].	Nº of days
37) Wyman et al. (2010)	Journal article	226	K ‐ 3rd	Rochester Resilience Programme	Indicated	90%	Yes	Suspension events Post‐treatment (immediately after intervention) TC2: Exp b=−0.57; SE=0.23; z=−2.48; p=0.013 Controlling for suspension T1 59 classrooms, 4 students per classroom	Since data was nested in clusters, we corrected standard errors of effect sizes. The design effect was corrected by using the formula suggested by Higgins & Green (2011), expressed by the equation [1+(M‐1) x1]. See methods section for further details	Nº of events

Abbreviations : T1(treatment group baseline measure); C1 (control group baseline measure); T2 (treatment group post treatment measure); C2 (control group post treatment measure); T3(treatment group follow up measure); C3(control group follow up measure); FSM (free school meals); M (mean); SD (standard deviation); N (sample size); OR (Odds ratio); 95% CI (95% confidence interval); SE (standard error); f (ANCOVA coefficient); p (p‐value); b (beta coefficient); X² (chi‐squared).

³⁵When the mean age was not available in the original study, their grade in school has been reported. Their school grade gives the reader a general idea of the age of the students.

³⁶Universal intervention strategies are those oriented to reach the entire population of students, without regard to individual risk factors. Indicated programmes were defined as those targeting students displaying behavioural problems, punished at school or presenting a specific risk to their educational development.

³⁷Due to the nature of the settings (schools) some studies reported clustered data. We corrected SE errors when it was needed. See methods section for further details.