Positive psychology interventions: a meta-analysis of randomized controlled studies

Background The use of positive psychological interventions may be considered as a complementary strategy in mental health promotion and treatment. The present article constitutes a meta-analytical study of the effectiveness of positive psychology interventions for the general public and for individuals with specific psychosocial problems. Methods We conducted a systematic literature search using PubMed, PsychInfo, the Cochrane register, and manual searches. Forty articles, describing 39 studies, totaling 6,139 participants, met the criteria for inclusion. The outcome measures used were subjective well-being, psychological well-being and depression. Positive psychology interventions included self-help interventions, group training and individual therapy. Results The standardized mean difference was 0.34 for subjective well-being, 0.20 for psychological well-being and 0.23 for depression indicating small effects for positive psychology interventions. At follow-up from three to six months, effect sizes are small, but still significant for subjective well-being and psychological well-being, indicating that effects are fairly sustainable. Heterogeneity was rather high, due to the wide diversity of the studies included. Several variables moderated the impact on depression: Interventions were more effective if they were of longer duration, if recruitment was conducted via referral or hospital, if interventions were delivered to people with certain psychosocial problems and on an individual basis, and if the study design was of low quality. Moreover, indications for publication bias were found, and the quality of the studies varied considerably. Conclusions The results of this meta-analysis show that positive psychology interventions can be effective in the enhancement of subjective well-being and psychological well-being, as well as in helping to reduce depressive symptoms. Additional high-quality peer-reviewed studies in diverse (clinical) populations are needed to strengthen the evidence-base for positive psychology interventions.


Background
Over the past few decades, many psychological treatments have been developed for common mental problems and disorders such as depression and anxiety. Effectiveness has been established for cognitive behavioral therapy [1,2], problem-solving therapy [3] and interpersonal therapy [4]. Preventive and early interventions, such as the Coping with Depression course [5], the Don't Panic course [6] and Living Life to the Full [7,8] are also available. The existing evidence shows that the mental health care system has traditionally focused more on treatment of mental disorders than on prevention. However, it is recognized that mental health is more than just the absence of mental illness, as expressed in the World Health Organization's definition of mental health: Mental health is a state of well-being in which the individual realizes his or her own abilities, can cope with the normal stresses of life, can work productively, and is able to make a contribution to his or her community [9].
Under this definition well-being and positive functioning are core elements of mental health. It underscores that people can be free of mental illness and at the same time be unhappy and exhibit a high level of dysfunction in daily life [10]. Likewise, people with mental disorders, can be happy by coping well with their illness and enjoy a satisfactory quality of life [11]. Subjective well-being refers to a cognitive and/or affective appraisal of one's own life as a whole [12]. Psychological well-being focuses on the optimal functioning of the individual and includes concepts such as mastery, hope and purpose in life [13,14]. The benefits of well-being are recorded both in cross-sectional and longitudinal research and include improved productivity at work, having more meaningful relationships and less health care uptake [15,16]. Wellbeing is also positively associated with better physical health [17][18][19]. It is possible that this association is mediated by a healthy lifestyle and a healthier immune system, which buffers the adverse influence of stress [20]. In addition, the available evidence suggests that well-being reduces the risk of developing mental symptoms and disorders [21,22] and helps reduce mortality risks in people with physical disease [23]. Seligman and Csikszentmihaly's (2000) pioneered these principles of positive psychology in their well-known article entitled 'Positive psychology: An introduction' , published in a special issue of the American Psychologist. They argued that a negative bias prevailed in psychology research, where the main focus was on negative emotions and treating mental health problems and disorders [24]. Although the basic concepts of well-being, happiness and human flourishing have been studied for some decades [12,[25][26][27], there was a lack of evidence-based interventions [24]. Since the publication of Seligman and Csikszentmihaly's seminal article, the positive psychology movement has grown rapidly. The ever-expanding International Positive Psychology Association is among the most extensive research networks in the world [28] and many clinicians and coaches embrace the body of thought that positive psychology has to offer.
Consequently, the number of evaluation studies has greatly increased over the past decade. Many of these studies demonstrated the efficacy of positive psychology interventions such as counting your blessings [29,30], practicing kindness [31], setting personal goals [32,33], expressing gratitude [30,34] and using personal strengths [30] to enhance well-being, and, in some cases, to alleviate depressive symptoms [30]. Many of these interventions are delivered in a self-help format. Sin and Lyubomirsky (2009) conducted a meta-analytical review of the evidence for the effectiveness of positive psychology interventions (PPIs). Their results show that PPIs can indeed be effective in enhancing well-being (r = 0.29, standardized mean difference Cohen's d = 0.61) and help to reduce depressive symptom levels in clinical populations (r = 0.31, Cohen's d = 0.65). However, this meta-analysis had some important limitations. First, the meta-analysis included both randomized studies and quasi-experimental studies. Second, study quality was not addressed as a potential effect moderator. In recent meta-analyses, it has been shown that the treatment effects of psychotherapy have been overestimated in lower quality studies [35,36]. The lack of clarity in the inclusion criteria constitutes a third limitation. Intervention studies, although related to positive psychology but not strictly developed within this new framework (e.g. mindfulness, life-review) were included in the meta-analysis. However, inclusion of these studies reduces the robustness of the results for pure positive psychology interventions.

Present study
The aim of the present study is to conduct a metaanalysis of the effects of specific positive psychology interventions in the general public and in people with specific psychosocial problems. Subjective well-being, psychological well-being and depressive symptoms were the outcome measures. Potential variables moderating the effectiveness of the interventions, such as intervention type, duration and quality of the research design, were also examined. This study will add to the existing literature and the above meta-analytical review [37] by 1) only including randomized controlled studies, 2) taking the methodological quality of the primary studies into account, 3) including the most recent studies (2009 -2012), 4) analyzing not only post-test effects but also long-term effects at follow up, and 5) applying clear inclusion criteria for the type of interventions and study design.

Search strategy
A systematic literature search was carried out in Psy-chInfo, PubMed and the Cochrane Central Register of Controlled Trials, covering the period from 1998 (the start of the positive psychology movement) to November 2012. The search strategy was based on two key components: there should be a) a specific positive psychology intervention, and b) an outcome evaluation. The following MeSH terms and text words were used: "well-being" or "happiness" or "happy*", "optimism", "positive psychology" in combination with "intervention", "treatment", "therapy" and "prevention". This was combined with terms related to outcome research: "effect*", or "effic*", or "outcome*", or "evaluat*". We also cross-checked the references from the studies retrieved, the earlier metaanalysis of Sin & Lyubomirsky (2009) and two other reviews of positive psychological interventions [38,39].
The search was restricted to peer-reviewed studies in the English language.

Selection of studies
Two reviewers (LB and MH) independently selected potentially eligible studies in two phases. At the first phase, selection was based on title and abstract, and at the second phase on the full-text article. All studies identified as potentially eligible by at least one of the reviewers during the first selection phase, were re-assessed at the second selection phase. During the second phase, disagreements between the reviewers were resolved by consensus. The inter-rater reliability (kappa) was 0.90.
The inclusion criteria were as follows: Examination of the effects of a positive psychology intervention. A positive psychology intervention (PPI) was defined in accordance with Sin and Lyubomirsky's (2009) article as a psychological intervention (training, exercise, therapy) primarily aimed at raising positive feelings, positive cognitions or positive behavior as opposed to interventions aiming to reduce symptoms, problems or disorders. The intervention should have been explicitly developed in line with the theoretical tradition of positive psychology (usually reported in the introduction section of an article). Randomization of the study subjects (randomizing individuals, not groups) and the presence of a comparator condition (no intervention, placebo, care as usual). Publication in a peer-reviewed journal. At least one of the following are measured as outcomes: well-being (subjective well-being and/or psychological well-being) or depression (diagnosis or symptoms). Sufficient statistics are reported to enable the calculation of standardized effect sizes.
If necessary, authors were contacted for supplementary data. We excluded studies that involved physical exercises aimed at the improvement of well-being, as well as mindfulness or meditation interventions, forgiveness therapy, life-review and reminiscence interventions. Furthermore, well-being interventions in diseased populations not explicitly grounded in positive psychology theory ('coping with disease courses') were excluded. Apart from being beyond the scope of this meta-analysis, extensive meta-analyses have already been published for these types of intervention [40][41][42]. This does not imply that these interventions do not have positive effects on well-being, a point which will be elaborated on in the discussion section of this paper.

Data extraction
Data extraction and study quality assessment were performed by one reviewer (LB) and independently checked by a second reviewer (MH). Disagreements were resolved by consensus. Data were collected on design, intervention characteristics, target group, recruitment methods, delivery mode, number of sessions, attrition rates, control group, outcome measures and effect sizes (post-test and at follow up of at least 3 months). The primary outcomes in our meta-analysis were subjective well-being (SWB), psychological well-being (PWB) and depressive symptoms/depression.
The methodological quality of the included studies was assessed using a short scale of six criteria tailored to those studies and based on criteria established by the Cochrane collaboration [43]: 1) Adequacy of randomization concealment, 2) Blinding of subjects to the condition (blinding of assessors was not applicable in most cases), 3) Baseline comparability: were study groups comparable at the beginning of the study and was this explicitly assessed? (Or were adjustments made to correct for baseline imbalance using appropriate covariates), 4) Power analysis: is there an adequate power analysis and/or are there at least 50 participants in the analysis?, 5) Completeness of follow up data: clear attrition analysis and loss to follow up < 50%, 6) Handling of missing data: the use of intention-to-treat analysis (as opposed to a completers-only analysis). Each criterion was rated as 0 (study does not meet criterion) or 1 (study meets criterion). The inter-rater reliability (kappa) was 0.91. The quality of a study was assessed as high when five or six criteria were met, medium when three or four criteria were met, and low when zero, one or two criteria were met. Along with a summary score, the aspects relating to quality were also considered individually, as results based on composite quality scales can be equivocal [44]. Table 1 shows the quality assessment for each study. The quality of the studies was scored from 1 to 5 (M = 2.56; SD = 1.25). Twenty studies were rated as low, 18 were of medium quality and one study was of high quality. None of the studies met all quality criteria. The average number of participants in the analysis was rather high (17 out of 39 studies scored positive on this criterion), although none of the studies reported an adequate power analysis. Also, baseline comparability was frequently reported (26/39 studies). On the other hand, independence in the randomization procedure was seldom reported (7/39 studies) and an intention-to-treat analysis was rarely conducted (3/39 studies).

Meta-analysis
In a meta-analysis, the effects found in the primary studies are converted into a standardized effect size, which is no longer placed on the original measurement scale, and can therefore be compared with measures from other scales. For each study, we calculated effect sizes (Cohen's d) by subtracting the average score of the experimental group (Me) from the average score of the control group (Mc), and dividing the result by the pooled standard deviations of both groups. This was done at post-test because randomization usually results in comparable groups across conditions at baseline. However, if baseline differences on outcome variables did exist despite the randomization, d's were calculated on the basis of pre-post-test differences: by calculating the standardized pre-post change score for the experimental group (de) and the control group (dc) and subsequently calculating their difference as Δd= dedc. For example, an effect size of 0.5 indicates that the mean of the experimental group is half a standard unit (standard deviation) larger than the mean of the control group. From a clinical perspective, effect sizes of 0.56 -1.2 can be interpreted as large, while effect sizes of 0.33 -0.55 are of medium size, and effects of 0 -0.32 are small [45].
In the calculation of effect sizes for depression, we used instruments that explicitly measure depression (e.g. the Beck Depression Inventory, or the Center for Epidemiological Studies Depression Scale). For subjective and psychological well-being, we also used instruments related to the construct of well-being (such as positive affect for SWB and hope for PWB). If more than one measure was used for SWB, PWB or depression, the mean of the effect sizes was calculated, so that each study outcome had one effect size. If more than one experimental group was compared with a control condition in a particular study, the number of subjects in the control groups was evenly divided across the experimental groups so that each subject was used only once in the meta-analysis.
To calculate pooled mean effect sizes, we used Comprehensive Meta-Analysis (CMA, Version 2.2.064). Due to the diversity of studies and populations, a common effect size was not assumed and we expected considerable heterogeneity. Therefore, it was decided a priori to use the 'random effects model'. Effect sizes may differ under this model, not only because of random error within studies (as in the fixed effects model), but also as a result of true variation in effect sizes between studies. The outcomes of the random effects model are conservative in that their 95% Confidence Intervals (CIs) are often broad, thus reducing the likelihood of type-II errors.
We tested for the presence of heterogeneity with two indicators. First, we calculated the Q-statistic. A significant Q rejects the null-hypothesis of homogeneity and indicates that the true effect size probably does vary from study to study. Second, the I 2 -statistic was calculated. This is a percentage indicating the study-to-study dispersion due to real differences, over and above random sampling error. A value of 0% indicates an absence of dispersion, and larger values show increasing levels of heterogeneity where 25% can be considered as low, 50% as moderate and 75% as a high level of heterogeneity [46]. Owing to the expected high level of heterogeneity, all studies were taken into account. Outliers were considered, but not automatically removed from the meta-analysis. The procedure of removing outliers which are outside the confidence interval of the pooled effect size is advised when a common effect size is assumed. However, in our meta-analysis, high dispersion was expected and therefore only the exclusion of Cohen's d > 2.5 from the final sample was planned.
Subgroup analyses were performed by testing differences in Cohen's d's between subgroups. Six potential moderators were determined based on previous research and the characteristics of the investigated interventions and studies: 1) Self-selected sample/not selfselected: did the participants know that the aim of the intervention was to make them feel better?; 2) Duration: less than four weeks, four to eight weeks, or more than eight weeks; 3) Type of intervention: selfhelp, group intervention, or individual therapy; 4) Recruitment method: community (in a community center, local newspapers), internet, by referral/hospital, at university; 5) Psychosocial problems (Yes/none): was the data based on a group with certain psychosocial problems or was the study open to everyone?; 6) Quality rating: low (score 1 or 2), medium (score 3 or 4) or high (score 5 or 6). The impact of the duration and quality ratings was also assessed using meta-regression.
Results of meta-analysis may be biased due to the fact that studies with non-significant or negative results are less likely to be published in peer-reviewed journals [47]. In order to address this issue, we used three indices: funnel plots, the Orwin's fail-safe number and the Trim and Fill method. A funnel plot is a graph of effect size against study size. When publication bias is absent, the observed studies are expected to be distributed symmetrically around the pooled effect size. The Orwin's fail-safe number indicates the number of nonsignificant unpublished studies needed to reduce the overall significant effect to non-significance (according to a self-stated criterium) [48]. The effect size can be considered to be robust if the number of studies required to reduce the overall effect size to a nonsignificant level exceeds 5 K + 10, where K is the number of studies included. If asymmetry is found in the funnel plot, the Trim and Fill method adjusts the pooled effect size for the outcomes of missing studies [49]. Imputing missing studies restores the symmetry in the funnel plot and an adjusted effect size can be calculated.
For the reporting of the results of this meta-analysis, we applied Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [50].

Description of studies
The selection process is illustrated in Figure 1. First, 5,335 titles were retrieved from databases and 55 titles were identified through searching the reference list accompanying the meta-analysis by Sin and Lyubomirsky (2009) [37] as well as two other literature reviews of positive psychological interventions [38,39]. After reviewing the titles and abstracts and removing duplicates, 84 articles were identified as being potentially eligible for inclusion in our study. Of these 84 articles, 40 articles in which 39 studies were described, met our inclusion criteria (of these, 17 articles describing 19 studies were also included in the metaanalysis by Sin and Lyubomirsky, 2009). In two articles [29,51] two studies were described, and one study [52][53][54][55] was published in four articles.

Post-test effects
The random effect model showed that the PPIs were effective for all three outcomes. Results are presented in  [51,83]. Removing the outliers reduced heterogeneity substantially (to a non-significant level).

Follow-up effects
Ten studies examined follow-up effects after at least three months and up to 12 months (Table 3). For the purposes of interpretation, we used only those studies examining effects from three to six months (short-term follow-up), thus excluding Fava et al. (2005) [83] which had a followup at one year. The random-effects model demonstrated small but significant effects in comparison with the control groups for subjective well-being (Cohen's d 0. 22 15). Heterogeneity was low for subjective well-being (I 2 = 1.1%) and psychological well-being (I 2 = 26.0%), and high for depression (I 2 = 63.9%).

Subgroup analyses
Subgroup analyses are presented in Table 4. We looked at self-selection, duration of the intervention, type of intervention, recruitment method, application of inclusion criteria related to certain psychosocial problems, and quality rating.
For depression, five out of six subgroups of studies resulted in significantly higher effect sizes. Higher effect sizes were found for 1) interventions of a longer duration (only in the meta regression analysis), 2) individual interventions, 3) studies involving referral from a health care practitioner or hospital, 4) studies which applied inclusion criteria based on psychosocial problems and 5) lower quality studies. For subjective well-being and psychological well-being, there were no significant differences between subgroups, although for the latter there was a recognizable trend in the same direction and on the same moderators, except for quality rating.
Twenty-six out of 39 studies were self-help interventions for which we conducted a separate subgroup analysis. However, there was little diversity within the self-help subgroup: only six studies examined intensive self-help for longer than four weeks, self-help was offered to people with specific psychosocial problems in only one study and more than half of the self-help studies (n=14) recruited their participants via university. Consequently, there were no significant differences between subgroups for self-help interventions.

Publication bias
Indications for publication bias were found for all outcome measures, but to a lesser extent for subjective well-being. Funnel plots were asymmetrically distributed in such a way that the smaller studies often showed the more positive results (in other words, there is a certain lack of small insignificant studies). Orwin's fail-safe numbers based on a criterium effect size of 0.10 for subjective well-being (59), psychological well-being (16) and depression (13) were lower than required (respectively 150, 110 and 80). Egger's regression intercept also suggests that publication bias exists for psychological well-being (intercept=1.18, t=2.26, df=18, p=.04) and depression (intercept=1.45, t=2.26, df=12, p=.03), but not for subjective well-being (intercept=1.20, t=1.55, df=26, p=0.13). The mean effect sizes of psychological wellbeing and depression were therefore recalculated by imputing missing studies using the Trim and Fill method. For psychological well-being, three studies were imputed

Main findings
This meta-analysis synthesized effectiveness studies on positive psychology interventions. Following a systematic literature search, 40 articles describing 39 studies were included. Results showed that positive psychology interventions significantly enhance subjective and psychological well-being and reduce depressive symptoms. Effect sizes were in the small to moderate range. The mean effect size on subjective well-being was 0.34, 0.20 on psychological well-being, and 0.23 on depression. Effect sizes varied a great deal between studies, ranging from below 0 (indicating a negative effect) to 2.4 (indicating a very large effect). Moreover, at follow-up from three to six months, small but still significant effects were found for subjective well-being and psychological well-being, indicating that effects were partly sustained over time. These follow-up results should be treated with caution because of the small number of studies and the high attrition rates at follow-up.
Remarkably, effect sizes in the current meta-analysis are around 0.3 points lower than the effect sizes in the meta-analysis by Sin and Lyubomirsky (2009) [37]. We included a different set of studies in which the design Meta Analysis Figure 2 Post-test effects of positive psychology interventions on subjective well-being. The square boxes show effect size and sample size (the larger the box, the larger the sample size) in each study, and the line the 95% confidence interval. The diamond reflects the pooled effect size and the width of the 95% confidence interval.  Figure 3 Post-test effects of positive psychology interventions on psychological well-being. The square boxes show effect size and sample size (the larger the box, the larger the sample size) in each study, and the line the 95% confidence interval. The diamond reflects the pooled effect size and the width of the 95% confidence interval.  Figure 4 Post-test effects of positive psychology interventions on depressive symptoms. The square boxes show effect size and sample size (the larger the box, the larger the sample size) in each study, and the line the 95% confidence interval. The diamond reflects the pooled effect size and the width of the 95% confidence interval. quality was assured using randomized controlled trials only. Effectiveness research in psychotherapy shows that effect sizes are relatively small in high-quality studies compared with low-quality studies [35] and this might also be true for positive psychology interventions. In addition, we applied stricter inclusion criteria than those used by Sin and Lyubomirsky (2009) and therefore did not include studies on any related areas such as mindfulness and life review therapy. These types of interventions stem from long-standing independent research traditions for which effectiveness has already been established in several meta-analyses [40,41]. Also, the most recent studies were included. This might explain the overestimation of effect sizes in the meta-analysis by Sin and Lyubomirsky (2009). Several characteristics of the study moderated the effect on depressive symptoms. Larger effects were found in interventions with a longer duration, in individual interventions (compared with self-help), when the interventions were offered to people with certain psychosocial problems and when recruitment was carried out via referral from a health care professional or hospital. Quality rating also moderated the effect on depression: the higher the quality, the smaller the effect. Interestingly, these characteristics did not significantly moderate subjective well-being and psychological well-being. However, there was a trend in the moderation of psychological well-being that was the same as that observed in the studies which included depression as an outcome. In general, effectiveness was increased when interventions were offered over a longer period, face-to-face on an individual basis in people experiencing psychosocial problems and when participants were recruited via the health care system.

Favours Control group Favours PPI
Although it is clear that more intensive and face-toface interventions generate larger effects, the effects of short-term self-help interventions are small but significant. From a public health perspective, self-help interventions can serve as cost-effective mental health promotion tools to reach large target groups which may not otherwise be reached [86][87][88]. Even interventions presenting small effect sizes can in theory have a major impact on populations' well-being when many people are reached [89]. The majority of positive psychology interventions (in our study 26 out of 39 studies) are already delivered in a self-help format, sometimes in conjunction with face-to-face instruction and support. Apparently, self-help suits the goals of positive psychology very well and it would be very interesting to learn more about how to improve the effectiveness of PPI selfhelp interventions. However, a separate subgroup analysis on the self-help subgroup revealed no significant differences in the present meta-analysis. There was very little variation in the subgroups as regards population, duration of the intervention and recruitment method. As a result, this analysis does not give firm indications on how to improve the effectiveness of self-help interventions. It is possible that self-help could be enhanced by offering interventions to people with specific psychosocial problems, increasing the intensity of the intervention and embedding the interventions in the health care system. However, more studies in diverse populations, settings and with varying intensity are needed before we can begin to derive recommendations from this type of meta-analysis. Other research gives several additional indications on how to boost the efficacy of self-help interventions. Adherence tends to be quite low in self-help interventions [90,91] and therefore, enhancing adherence could be a major factor in improving effectiveness. Selfhelp often takes a 'one size fits all' approach, which may not be appropriate for a large group of people who will, as a consequence, not fully adhere to the intervention. Personalization and tailoring self-help interventions to individual needs [92] as well interactive support [93] might contribute to increased adherence and likewise improved effectiveness of (internet) self-help interventions.

Study limitations
This study has several limitations. First, the quality of the studies was not high, and no study met all of our quality criteria. For example, the randomization procedure was unclear in many studies. Also, most studies conducted completers-only analysis, as opposed to intention-to-treat analysis. This could have seriously biased the results [35]. However, the low quality of the studies could have been overstated as the criteria were scored conservatively: we gave a negative score when a criterion was not reported. Even so, more high-quality randomized-controlled trials are needed to enable more robust conclusions about the effects of PPIs. Second, different types of interventions are lumped together as positive psychology interventions, despite the strict inclusion criteria we applied. As expected, we found a rather high level of heterogeneity. In the future, it might be wise and meaningful to conduct metaanalyses that are restricted to specific types of interventions, for example gratitude interventions, strengths-based interventions and well-being therapy, just as has already been carried out with, for example, mindfulness and life review. In the present meta-analysis, studies on these specific interventions were too small and too diverse to allow for a subgroup-analysis. Third, the exclusion of non peerreviewed articles and grey literature could have led to bias, and possibly also to the publication bias we found in our study. Fourth, although we included a relatively large number of studies in the meta-analysis, the number of studies in some subgroups was still small. Again, more randomized-controlled trials are needed to draw firmer conclusions. Sixth, the study of positive education is an emerging field in positive psychology [94][95][96][97][98] but school-based interventions were excluded from our metaanalysis due to the strict application of the inclusion criteria (only studies with randomization at individual level were included).

Conclusion
This meta-analysis demonstrates that positive psychology interventions can be effective in the enhancement of subjective and psychological well-being and may help to reduce depressive symptom levels. Results indicate that the effects are partly sustained at short-term follow-up. Although effect sizes are smaller in our metaanalysis, these results can be seen as a confirmation of the earlier meta-analysis by Sin and Lyubomirsky (2009). Interpretation of our findings should take account of the limitations discussed above and the indications for publication bias.

Implications for practice
In mental health care PPIs can be used in conjunction with problem-based preventive interventions and treatment. This combination of interventions might be appropriate when clients are in remission; positive psychology interventions may then be used to strengthen psychological and social recourses, build up resilience and prepare for normal life again. On the basis of the moderator analysis, we would recommend the delivery of interventions over a longer period (at least four weeks and preferably eight weeks or longer) and on an individual basis. Practitioners can tailor their treatment strategy to the needs and preferences of a client and can use positive psychology exercises in combination with other evidence-based interventions that have a positive approach and aim to enhance well-being, such as mindfulness interventions [40], Acceptance and Commitment Therapy [7,99], forgiveness interventions [42], behavioral-activation [100] and reminiscence [41,101].
In the context of public health, positive psychology interventions can be used as preventive, easily accessible and non-stigmatizing tools. They can potentially be used in two ways: 1) in mental health promotion (e.g. leaflets distributed for free at community centers, (mental) health internet portals containing psycho-education), and 2) as a first step in a stepped care approach. In the stepped care model, clients start with a low-intensity intervention if possible, preferably a self-directed intervention. These interventions can be either guided by a professional or unguided, and are increasingly delivered over the internet. Clinical outcomes can be monitored and people can be provided with more intensive forms of treatment, or referred to specialized care, if the first-step intervention does not result in the desired outcome [102].

Recommendations for research
Regarding the research agenda, there is a need for more high-quality studies, and more studies in diverse (clinical) populations and diverse intervention formats to know what works for whom. Standards for reporting studies should also be given more attention, for example by reporting randomized controlled trials according to the CONSORT statement [103]. In addition, we encourage researchers to publish in peer-reviewed journals, even when the sample sizes are small or when there is a null finding of no effect, as this is likely to reduce the publication bias in positive psychology. Furthermore, most studies are conducted in North America. Therefore, replications are needed in other countries and cultures because some positive psychology concepts may require adaptation to other cultures and outlooks (e.g. see Martinez et al., 2010) [68]. Last but not least, we strongly recommend conducting cost-effectiveness studies aiming to establish the societal and public health impact of positive psychology interventions. This type of