A meta-analysis update on the effects of early family/ parent training programs on antisocial behavior and delinquency

Alex R. Piquero1 & Wesley G. Jennings2 & Brie Diamond3 &

David P. Farrington4 & Richard E. Tremblay5,6 &

Brandon C. Welsh7,8 & Jennifer M. Reingle Gonzalez9

Published online: 13 April 2016 # Springer Science+Business Media Dordrecht 2016

Abstract Objective To update Piquero et al.’s meta-analysis on early family/parent training programs. Methods Screening of eligible studies was carried out for the period between January 2008 and August 2015. An additional 23 studies were identified, which were added to the original database of 55 studies, totaling an overall sample of 78 eligible studies. A random-effects model was used to obtain an overall mean effect size estimate. Addi- tional analyses were performed to assess publication bias and moderation. Results An overall average, positive, and significant effect size of 0.37 was calculated, which corresponds to 32 out of 100 in a treated group versus 50 out of 100 in a control group who offended. There was some evidence of publication bias and moderation. Conclusions Early family/parent training programs are an effective evidence-based strategy for preventing antisocial behavior and delinquency.

J Exp Criminol (2016) 12:229–248 DOI 10.1007/s11292-016-9256-0

* Alex R. Piquero [email protected]

1 University of Texas at Dallas, Richardson, TX, USA 2 University of South Florida, Tampa, FL, USA 3 Texas Christian University, Fort Worth, TX, USA 4 Cambridge University, Cambridge, UK 5 University of Montréal, Montréal, QC, Canada 6 University College Dublin, Dublin, Ireland 7 Northeastern University, Boston, MA, USA 8 Netherlands Institute for the Study of Crime and Law Enforcement, Amsterdam, The Netherlands 9 University of Texas School of Public Health, Houston, TX, USA

Keywords Antisocial behavior . Crime . Delinquency . Family . Interventions .

Meta-analysis . Parenting . Parent training

Introduction

Antisocial, delinquent, and criminal behavior, especially when sustained through- out the life-course, exerts a significant financial and human cost (Cohen and Piquero 2009). As a result, intervention in adolescence or even adulthood, while important, may occur a bit too late to thwart early-onset criminal careers. As a result, efforts at developing and evaluating prevention strategies early on in the life-course is of critical importance to building a safer society (Sherman et al. 2002; Tremblay and Craig 1995) and deterring the onset of early and especially persistent criminal careers.

One prominent set of prevention strategies surrounds early family/parent training programs. In brief, these efforts seek to provide families and parents with training and skills that help them better attend to the physical, mental, and social skills of their children. And while a variety of these programs exist, including several evidence-based efforts like The Incredible Years, Triple P Parenting, and Nurse Family Partnerships, they all focus on improving child outcomes via helping parents more effectively socialize their children.

In 2009, we published a meta-analysis of 55 high-quality early family/parent training programs, the results of which provided very strong support for the delivery of these programs in deterring subsequent antisocial behavior (Piquero et al. 2009). In this more recent meta-analysis, we provide an update to that original analysis that includes an additional 23 studies uncovered by our update spanning the January 2008–August 2015 period.

Methods

Criteria for inclusion and exclusion of studies

The primary inclusion/exclusion criteria for the Piquero et al. (2009) and the current review were as follows:

1. Types of studies: randomized controlled experimental design. 2. Types of participants: families with a child under age 5 or the mean age of

the sample was approximately age 5 at the beginning of the intervention. Programs with physically and/or mentally handicapped children were excluded.

3. Type of intervention: parent training was a major component of the intervention. 4. Types of outcomes: child behavior problems such as conduct problems, delinquen-

cy, and/or antisocial behavior. 5. Sufficient data: adequate data was needed for calculating an effect size if one

was not provided (i.e., means and standard deviations, t-tests, F-tests, p-values, etc.).

230 A.R. Piquero et al.

6. No time frame restrictions, except that the search began with the first study identified by Bernazzani et al. (2001).

7. No geographic restrictions. 8. Studies needed to be written in English.

A more detailed description of the methods related to the search strategy for the identification of relevant studies and the criteria for determination of independent findings can all be found in Piquero et al. (2009, pp. 91–94). In short, the current meta-analysis adopted the exact same strategies outlined in Piquero et al. (2009), although with a specific focus on identifying relevant published and unpublished studies that have become available since their 2009 review. Thus, this meta-analysis combines all of the relevant studies available from 2008 to 2015 with the data from Piquero et al.’s (2009) meta-analysis to provide a comprehensive and up-to-date resource on the effectiveness of early family/parent training programs on antisocial behavior and delinquency.

Results

Literature search

Our initial literature search for relevant studies available from January 2008 to August 2015 produced over 3,400 hits. Following this initial search, we reviewed these studies and discarded any duplicates, studies that were not published in English, and studies that did not upon further examination conform to our pre-defined inclusion criteria. This cleaning process left 29 studies. After further reviewing this reduced list, six additional studies were removed because they failed to provide data relevant for the coding of an effect size.1 As such, our final sample of relevant studies identified post-Piquero et al.’s (2009) meta-analysis was 23 studies, which after combining these studies with the 55 studies from Piquero et al.’s (2009) meta-analysis left us with a total sample size of 78 randomized, controlled trials of early family/parent training programs. A complete description of all of these 78 studies can be found in the Appendix.

Types of interventions

Generally speaking, the types of interventions identified in this review can broadly be classified as home visitation programs or parent training programs. The home visitation programs typically involve health professionals such as nurses, doctors, or paraprofes- sionals conducting in-home visits with the mothers to educate them on how to properly care for their children (for example, see Butz et al. 2001; Cullen 1976; Fergusson et al. 2005b; Heinicke et al. 2001; Kitzman et al. 1997; McCarton et al. 1997; Olds, Robinson, Pettitt et al. 2004; Stone et al. 1988). Comparatively, the parent training

1 The descriptive results reported in these six studies indicated that significant and positive effects in terms of a reduction in problem behavior were detected for the treatment group relative to the control group in four of the six studies. These results are consistent with the significant effects noted for the 78 studies that are included in this review, where 67 out of 78 of the effects were positive and 38 of the 67 were significant at the p < .05 level.

A meta-analysis update on early family/parent training programs 231

programs often relied on individual or group-based parent training sessions that were either conducted at a clinic, school, or some alternative community-based setting. The most recognizable of these types of parent training programs are the Incredible Years Parenting Program, the Triple P Positive Parenting Program, and Parent–child interac- tion therapy (PCIT). These types of programs often focus on methods of instruction to strengthen the parent’s competencies in monitoring and disciplining their child’s behavior and promoting the child’s social and emotional competence (Edwards et al. 2007; Gardner et al. 2006; Helfenbaum-Kun and Ortiz 2007; Kim et al. 2007; Patterson et al. 2002; Reid et al. 2007; Scott et al. 2001; Taylor et al. 1998; Tucker 1996; Webster-Stratton 1982, 1984, 1990b, 1992, 1998; Webster-Stratton and Hammond 1997; Webster-Stratton et al. 1988, 2001, 2004). In addition, these programs at times can also attempt to train parents to use positive and nonviolent techniques to manage their children’s behavior (Leung et al. 2003; Markie-Dadds and Sanders 2006; Morawska & Sanders, Sanders et al. 2000a, 2000b), or aim to foster a caring and responsive relationship between the parent and child through modeling and role playing (Brestan et al. 1997; Eyberg et al. 1995; McNeil et al. 1991; Schuhmann et al. 1998; Zangwill 1983).

Effect size

We relied on Cohen’s d (Cohen 1988) for determining the effect sizes for this meta-analysis, and the primary source for calculating Cohen’s d was the standard- ized mean difference. However, in the event where Cohen’s d effect size estimates were not provided or in situations where means and standard deviations were not provided, we relied on t-values, f-values, p-values, correlations, odds-ratios, etc. to calculate the effect sizes (see Lipsey and Wilson 2001 for the relevant formulas). Further, we applied the Hedges and Olkin (1985) adjustment and used inverse variance weights in calculating the effect sizes to be consistent with Piquero et al. (2009). All of the meta-analytic results described here and below were performed using Comprehensive Meta-Analysis (CMA) software, version 2 (Borenstein et al. 2005).

Figure 1 provides the results of a forest plot illustrating the individual study effects sizes that were obtained and/or calculated. These Hedges’s g estimates in the forest plot are organized in descending order of effect size magnitude along with their associated z statistic and corresponding p-values. In addition, the graph on the right side of Fig. 1 displays these estimates with their corresponding 95 % confidence intervals. As can be seen, the overwhelming majority of the effect sizes were positive (67 out of 78) and significant at the p< .05 level (38 of the 67 positive effect sizes were statistically significant). Comparatively, only 11 of the studies yielded a negative effect size estimate, with only one of these negative effect sizes being statistically significant (Weihrauch et al. 2014). Also, as shown in Fig. 1, the overall mean effect size (with random effects) for these 78 studies was 0.37 (z= 8.99, p< .001).2

2 We also calculated the average effect size for the 23 studies identified during the search period January 2009–August 2015. These results indicated a positive and significant average effect size of 0.39 (95 % CI = 0.24–0.54; z = 5.148, p < .001).

232 A.R. Piquero et al.

Fig. 1 Forest plot of the distribution of effect sizes sorted by magnitude (n = 78 studies). Note: * = unpublished study

A meta-analysis update on early family/parent training programs 233

Fig. 2 Funnel plot examining publication bias. Note: Kendall’s tau statistic = 0.18, z = 2.38, p = .02; Eggers regression intercept = 1.49, standard error = 0.41, t = 3.66, p < .001

Table 1 Weighted effect sizes, confidence intervals, z-tests and Q statistics of moderators (with random effects)

Variables N Weighted ES Lower 95 % CI Upper 95 % CI Z-test Q-statistic

Country of publication

US-based 46 .39 .28 .50 7.14*** 239.62***

Non US-based 32 .35 .22 .48 5.41*** 227.05***

Type of program

Parent training 67 .39 .30 .48 8.67*** 328.44***

Home visits 11 .28 .07 .49 2.67** 113.09***

Small vs large samples

N < 100 50 .49 .35 .62 6.86*** 312.19***

N > 100 28 .26 .16 .35 5.33*** 158.41***

Publication bias

Published 73 .37 .29 .45 8.74*** 477.77***

Not published 5 .43 .16 .70 3.11** 1.93

Total 78 .37 .29 .45 8.99*** 481.899***

* p < .05 ** p < .01 *** p < .001

234 A.R. Piquero et al.

In order to remain consistent with Piquero et al. (2009), we included unpublished studies to address the Bfile drawer^ problem. In this vein, we assessed the potential for publication bias through the use of a funnel plot and calculated relevant test statistics (e.g., Kendall’s and Egger’s tests) (see Fig. 2). Upon reviewing the funnel plot (where the larger studies are plotted at the top and the smaller studies are plotted at the bottom) and the associated Kendall’s (z= 2.38, p= 0.02) and Egger’s (t= 3.66, p< .001) test statistics, it appears that there is some evidence that publication bias may be present as the smaller studies do seem to be clustering more to the right. This asymmetry suggests that there seems to be a tendency for smaller studies to be published if they have larger than average effects.

Following Piquero et al. (2009) and our own analysis that revealed that there was not homogeneity in the effect sizes (Q= 481.89, p< .001), we estimated a series of moderator analyses to further explore where some of the source of this heterogeneity may exist. Table 1 presents the results for the categorical ANOVA moderator analyses (with random effects), relying on the exact same moderators evaluated by Piquero et al. (2009). The findings suggest that country of

Fig. 3 Meta-regression (with random effects, maximum likelihood). Note: publication year: regression coefficient (slope) = −.001, standard error = .001; z = −1.14, p = .26

A meta-analysis update on early family/parent training programs 235

publication (US versus not US), type of program (parent training versus home visits), sample size (small sample versus large sample), and publication bias (published versus unpublished) were all statistically significant effect size moder- ators. Finally, as plotted in Fig. 3, the results of a meta-analytic regression model (with random effects and estimated using a maximum likelihood function) did not demonstrate the year of the publication of the study to be a statistically significant effect size moderator.3

Fig. 4 Forest plot of the distribution of effect sizes sorted by magnitude and grouped by Bbrand^ of early/ family parent training programs (N = 33 studies). Note: * = unpublished study. IY = Incredible Years; PCIT = Parent–child interaction therapy; PPP = Triple P Parenting

3 It is important to note that we did also estimate a meta-analytic regression model (with random effects and using a maximum likelihood function) with all four of the categorical moderators and the continuous moderator of year of publication included simultaneously. However, as none of these variables were determined to be a statistically significant effect size predictor, we opted to not present the results of this full regression model.

236 A.R. Piquero et al.

In one final interesting and supplemental analysis, we filtered the 78 studies down to 33 studies that were specifically early family/parent training interventions of the three most popular Bbrands^ of early family/parent training programs, that is, The Incredible Years Program, Parent–child interaction therapy, and the Triple P Parenting Program. Figure 4 illustrates the Hedges’s g estimates of these programs only in the forest plot, which are organized in descending order of effect size magnitude along with their associated z statistic and corresponding p-values and grouped by Bbrand^ on the left side of Fig. 4. In addition, the graph on the right side of Fig. 4 displays these estimates with their corresponding 95 % confidence intervals. As illustrated, the overall mean effect sizes for all three Bbrands^ are positive and statistically significant, with the largest mean effect size being observed for Parent–child interaction therapy (mean effect size = 0.98, p< .01), followed by the Triple P Parenting Program (mean effect size = 0.56, p< .001), and the Incredible Years Parenting Program (mean effect size = 0.31, p< .001).

Discussion

The purpose of this paper was to update the early family/parent training meta- analysis conducted by Piquero et al. (2009) and published in the Journal of Experimental Criminology. Our updated search, from January 2008 to August 2015, yielded an additional 23 methodologically rigorous studies that we added to the existing database of 55 studies, which yielded a total sample size of 78 studies to be used in this update.

Four main findings emerged from our analysis. First and foremost, we replicated the substantive finding of our previous work; namely, that early family/parent training Bis an effective intervention for reducing behavior prob- lems among young children^ (Piquero et al. 2009, p. 83). The overall mean effect size in our analysis of 78 individual effect sizes was 0.37, slightly higher than we obtained in our 2009 analysis (0.35). Second, we did find some evidence of publication bias, in that there was a tendency for smaller studies to be published if they have larger than average effects. Third, we also detected some evidence of moderating effects. In particular, country of publication, type of program, sample size, and publication bias were all statistically significant effect size moderators. Lastly, in one final and supplemental analysis, we determined that the overall mean effect sizes for all three Bbrands^ of the most popular early family/parent training programs are positive and statistically significant, with the largest mean effect size being observed for Parent–child interaction therapy (mean effect size = 0.98), followed by the Triple P Parenting Program (mean effect size = 0.56), and the Incredible Years Parenting Program (mean effect size = 0.31). In short, early family/parent training programs are an important evidence-based strategy that deserves continued application and ex- pansion as part of a more general strategy for building a safer society.

A meta-analysis update on early family/parent training programs 237

T ab

le 2

E ar ly

fa m ily

/p ar en t tr ai ni ng

pr og ra m

ev al ua tio ns

in cl ud ed

in m et a- an al ys is

A ut ho r, pu bl ic at io n da te

L oc at io n

Ty pe

of in te rv en tio

n T im

e of

st ud yb

O ri gi na l sa m pl e si ze

a Ta rg et ed

ag e( s)

A bb ot t- Sh

im m

an d L am

be rt (2 00 3)

S ou th er n U rb an

Se tti ng

U SA

Pa re nt

tr ai ni ng

19 98 –1 99 9

E = 87

C = 86

4 ye ar s ol d

A bi ko ff et al .( 20 15 )

N ew

Y or k,

N Y ,

U SA

H om

e vi si ts

20 08 –2 01 2

E = 67

C = 34

3– 4 ye ar s ol d

A xb er g an d B ro be rg

(2 01 2)

S w ed en

In cr ed ib le Y ea rs Pa re nt in g P ro gr am

Pa re nt

tr ai ni ng

N R

E = 38

C = 24

4– 8 ye ar s ol d

A ze ve do

et al .( 20 13 )

P or tu ga l

In cr ed ib le Y ea rs Pa re nt in g P ro gr am

Pa re nt

tr ai ni ng

20 09 –2 01 1

E = 69

C = 56

3– 6 ye ar s ol d

B ar kl ey

et al .( 20 00 )

W or ce st er ,

M A ,U

SA Pa re nt

tr ai ni ng

19 91 –1 99 6

E = 79

C = 42

K in de rg ar te ne rs

≈5 ye ar s ol d

B ra dl ey

et al .( 20 03 )

M et ro po lit an

To ro nt o

C an ad a

Pa re nt

tr ai ni ng

19 98

E = 11 9

C = 10 9]

3– 4 ye ar s ol d

B re ite ns te in

et al .( 20 12 )

C hi ca go ,I L ,

U SA

C hi ca go

P ar en t P ro gr am

Pa re nt

tr ai ni ng

20 02 –2 01 1

E = 26 7

C = 23 7

2– 4 ye ar s ol d

B re st an

et al .( 19 97 )

U SA

Pa re nt –c hi ld

in te ra ct io n th er ap y

Pa re nt

tr ai ni ng

N R

E = 14

C = 16

M ea n ag e = 4. 54

ye ar s

B ut z et al .( 20 01 )

Tw o U rb an

H os pi ta ls

U SA

H om

e vi si ts

19 94 –1 99 7

E = 59

C = 58

B ir th

C on ne ll et al .( 19 97 )

R ur al So

ut h E as t Q ue en sl an d

A us tr al ia

Pa re nt

tr ai ni ng

N R

E = 12

C = 12

2– 6 ye ar s ol d

C ul le n (1 97 6)

A us tr al ia

H om

e vi si ts

19 64 –1 96 7

E = 12 4

C = 12 2

1 ye ar

ol d

C un ni ng ha m

et al .( 19 95 )

H am

ilt on

Sc ho ol s

U SA

Pa re nt

tr ai ni ng

19 91 –1 99 3

E = 94

C = 56

Ju ni or

K in de rg ar te n

≈4 ye ar s ol d

D ay

et al .( 20 12 )

L on do n

U K

Pa re nt

tr ai ni ng

20 10

E = 54

E = 50

2– 11

ye ar s ol d

E dw

ar ds

et al .( 20 07 )

N or th

an d M id

W al es

U K

In cr ed ib le Y ea rs Pa re nt in g P ro gr am

Pa re nt

tr ai ni ng

N R

E = 86

C = 47

3– 4 ye ar s ol d

A pp

en di x

238 A.R. Piquero et al.

T ab

le 2

(c on tin

ue d)

A ut ho r, pu bl ic at io n da te

L oc at io n

Ty pe

of in te rv en tio n

T im

e of

st ud yb

O ri gi na l sa m pl e si ze

a Ta rg et ed

ag e( s)

E yb er g et al .( 19 95 )

U S A

Pa re nt –c hi ld

in te ra ct io n th er ap y

Pa re nt

tr ai ni ng

N R

E = 19

C = 8

3– 6 ye ar s ol d

Fa nn in g (2 00 7)

* U S A

Pa re nt

tr ai ni ng

20 05 –2 00 6

E = 14

C = 14

3– 5 ye ar s ol d

Fe in fi el d an d B ak er

(2 00 4)

L os

A ng el es ,C

A ,U

SA Pa re nt

tr ai ni ng

N R

E = 24

C = 23

4– 8 ye ar s ol d

Fe rg us so n et al .( 20 05 b)

C hr is tc hu rc h

N ew

Z ea la nd

H om

e vi si ts

20 00 –2 00 1

E = 22 0

C = 22 3

B ir th

Fr an z et al .( 20 11 )

G er m an y

H om

e vi si ts

N R

E = 26

C = 35

3– 6 ye ar s ol d

G ar dn er

et al .( 20 06 )

O xf or d

U K

In cr ed ib le Y ea rs P ar en tin g Pr og ra m

Pa re nt

tr ai ni ng

N R

E = 44

C = 32

2– 9 ye ar s ol d

G ro ss

et al .( 20 09 )

C hi ca go ,I L ,

U S A

C hi ca go

Pa re nt

Pr og ra m

Pa re nt

tr ai ni ng

20 02 –2 00 4

E = 13 5

C = 11 8

2– 4 ye ar s ol d

H ah lw eg

et al .( 20 10 )

G er m an y

T ri pl e P Pa re nt in g P ro gr am

Pa re nt

tr ai ni ng

N R

E = 19 6

C = 94

3– 6 ye ar s ol d

H am

ilt on

an d M ac Q ui dd y (1 98 4)

U S A

Pa re nt

tr ai ni ng

N R

E = 18

C = 9

2– 7 ye ar s ol d

H ei ni ck e et al .( 20 01 )

L os

A ng el es ,C

A ,

U S A

H om

e vi si ts

N R

E = 31

C = 33

B ir th

H el fe nb au m -K

un an d O rt iz (2 00 7)

N ew

Y or k,

N Y ,U

SA In cr ed ib le Y ea rs P ar en tin g Pr og ra m

Pa re nt

tr ai ni ng

N R

E = 23

C = 16

3– 5 ye ar s ol d

H er be rt (2 01 3)

* M as sa ch us et ts

U S A

Pa re nt in g Y ou r H yp er ac tiv e P re sc ho ol er

pr og ra m

Pa re nt

tr ai ni ng

20 09 –2 01 1

E = 17

C = 14

3– 6 ye ar s ol d

H er be rt et al .( 20 13 )

M as sa ch us et ts

U S A

Pa re nt in g Y ou r H yp er ac tiv e P re sc ho ol er

pr og ra m

Pa re nt

tr ai ni ng

20 09 –2 01 1

E = 17

C = 14

3– 6 ye ar s ol d

H is co ck

et al .( 20 08 )

M el bo ur ne ,V

ic to ri a

A us tr al ia

Pa re nt

tr ai ni ng

20 04

E = 32 9

C = 40 4

6– 7 m on th

ol d

H om

en et al .( 20 14 )

Po rt ug al

In cr ed ib le Y ea rs P ar en tin g Pr og ra m

Pa re nt