NEG = 10
NCG = 10
NEG, number of subjects in the intervention group; NCG, number of subjects in the control group; M, mean; MEG, mean in intervention group; SDEG, standard deviation in intervention group; TAU, treatment as usual; ES, effect size. Bold values indicate the total sample of each study .
We located a number of high-quality primary studies that we were not able to include due to the defined criteria. Fourteen studies had been excluded due to missing data or other reasons outlined in the Methods section, some of which may be possibly recovered for future analyses. They investigated learning disabilities (Alotaibi et al., 2017 ), children with attention deficit/hyperactivity disorder (Alrazain et al., 2018 ), fornix integrity (Burzynska et al., 2017 ), depression (Cross et al., 2012 ), falls (Duim et al., 2015 ), oncological patients (Ho et al., 2016b ), schizophrenia (Koch et al., 2017 ; Savill et al., 2017 ), Parkinson's disease (Lewis et al., 2014 ), autism (Mateos-Moreno and Atencia-Doña, 2013 ), trauma in unaccompanied minors (Meyer DeMott et al., 2017 ), development of kindergarteners (Stück and Villegas, 2017 ), personal development and increase of emotional intelligence in students (Vancea, 2013 ), and traumatized children (Van Westrhenen et al., 2019 ).
Mainly through the reviews, we also found a number of formerly undetected studies from 2011 and earlier, not yet included into the general meta-analyses. We consider it important to enumerate them here for potential future analyses: Belardinelli et al. ( 2008 ), Burgess et al. ( 2006 ), Chouhan and Kumar ( 2011 ), Connolly et al. ( 2011 ), Coubard et al. ( 2011 ), Hall ( 2011 ), Hwang et al. ( 2010 ), Kaltsatou et al. ( 2011 , 2015 ), Quiroga Murcia et al. ( 2009 ), and Xiong and Li ( 2009 ).
In general, our literature search revealed that there were at least as many studies on physiological changes after DMT and dance interventions as there were for psychological changes in health outcomes in the time frame of January 2012 to March 2018. We found around 50 primary studies with mere physical/physiological outcomes, which we excluded. However, since embodiment approaches such as DMT assume the body–mind unity, we would encourage researchers to include studies focusing on physical changes into future meta-analyses in dance and DMT.
Synthesizing the outcome foci of the primary studies, this meta-analysis differentiates six outcome clusters: (a) quality of life, (b) clinical outcomes (e.g., anxiety, depression), (c) interpersonal skills, (d) cognitive skills, (e) (psycho-)motor skills, and (f) residuals (psychotic symptoms and physiological change). We based the allocation of dependent variables to outcome clusters on the meta-analysis of Koch et al. ( 2014 ) and detailed investigation of the primary trials. Our aim was to synthesize outcome clusters that were as comparable as possible (e.g., by similarity of measurement instruments).
Quality of life is a broad construct, which contains subscales about subjective well-being (e.g., satisfaction with life) and conditions of daily living (e.g., general health, functional capacity and social integration). We decided to further include dependent variables about sleep quality, pain (invert coding), self-esteem, and control beliefs in this cluster. Most measures in this outcome cluster were based on self-report questionnaires (e.g., rating satisfaction with certain life conditions on a Likert scale), with only one study containing observations and ratings from an external person, who in this case was a clinician (Teixeira-Machado et al., 2017 ). We assume that DMT and dance interventions might influence quality of life on various dimensions, for example, movement might improve vitality and fitness, dance might foster joyful experiences (Koch et al., 2007 ), and interpersonal experiences might have a positive influence on social integration (Sandel et al., 1993 ).
Clinical outcomes summarize dependent variables directly related to conditions of mental health, particularly affective disorders (e.g., depression, anxiety, stress, anger). Studies in this cluster were conducted with a clinical (e.g., persons with a diagnosis of depression) or subclinical population (e.g., persons at risk for depression). Usually, the dependent variables were assessed with self-report questionnaires (e.g., HADS, BDI, BSI, STAXI). Two studies contained an interview and one study used external observation and rating. We decided to conduct a sub-analysis of anxiety and depression, because the treatment of these conditions is of broad public interest and has been discussed in the previous literature (Peters, 2012 ; Koch et al., 2014 ). We assumed that DMT and dance interventions improve psychological functions of emotion regulation, which may be mediated, for example, by authentic expression, experienced agency, body–mind integration, and physiological changes.
The term “interpersonal skills” relates to competences persons apply in social interaction (e.g., empathy, synchronization, communication, prosocial behavior, self-other awareness, maintaining a relationship). Most studies in this cluster were conducted with children, some of whom had been diagnosed with developmental disorders (ASD, ADHD). Therefore, researchers used external observations (e.g., by a parent, teacher, or clinician) more frequently. There were also two adult populations (ASD, schizophrenia). Interpersonal experiences in DMT and dance might particularly improve skills allocated to this cluster, for example, the therapeutic relationship, group cohesion, and (non-verbal) communication. In recent years, researchers have started to investigate whether the “mirroring” technique, proposed by dance movement therapist Marian Chace (Sandel et al., 1993 ), fosters empathy and enhances activity of mirror neurons in the brain (McGarry and Russo, 2011 ). A link of mirroring in movement and attachment has recently been established (see Feniger-Schaal et al., 2018 , this issue).
Cognitive skills relate to the set of mental abilities and processes that we need to carry out any task from the simplest to the most complex (e.g., skills of language, memory, and conceptualizing). Because a decrease of these mental abilities is a typical issue in elderly persons, most primary studies focus on this population. Cognitive skills were assessed using psychological tests (e.g., memory tasks, word tasks, calculation tasks, attention tasks) and tests referring to body image or body imagery, which denominates the ability to perceive and visualize bodily charges and changes. Since the operationalizations of the concept in the included studies measure a mainly representational skill, we categorize it under cognitive skills, even though it includes sensorimotor and emotional aspects.
The assessment of (psycho-)motor skills was conducted in (mostly elderly) patients that were diagnosed with Parkinson's disease. It contained tests on walking, turning, balance, and freezing of gait, and self-report measures on daily functioning. Only dance interventions, no DMT, were found in this outcome cluster. Dance improves motor function by training muscular activity, balance, and flexibility. Furthermore, there is a training of cognitive skills associated with movement, for example, in executing imagined movements, following music, and observing bodily changes (Hashimoto et al., 2015 ). Because we focused on psychological changes in this meta-analysis, studies with outcomes on mere physical skills (e.g., exercise capacity, arm range) were excluded. However, in our literature search, we found at least as many studies on physiological changes after DMT and dance interventions as on psychological changes. It is important to note that changes in Parkinson's disease severity were categorized as psychomotor skills , because Parkinson's disease is an extrapyramidal and neurodegenerative disorder (ICD 10), which entails physical and psychological components. The included studies with a focus on Parkinson's disease measured mixed psycho-physiological variables (as outlined above).
Our residual category contained two types of outcomes too small to analyze in separate clusters: positive symptoms in schizophrenia and physiological changes (e.g., blood pressure). In schizophrenia, we distinguish positive symptoms (an overabundance of perceptions compared to average, e.g., hallucinations) from negative symptoms (a void or lack of perception and expression compared to average, e.g., apathy, mood, and blunted affect). While positive symptoms can be successfully addressed by anti-psychotic medication, DMT seems to be particularly useful for addressing negative symptoms (Röhricht and Priebe, 2006 ; Lee et al., 2015 ; Martin et al., 2016 ; Pohlmann et al., 2017 ). We allocated the negative symptoms to the clinical outcomes cluster. Pohlmann et al. ( 2017 ) postulated a concept of “disembodiment” in schizophrenia stressing that schizophrenia is a self-disorder and is characterized by disturbances of ipseity (selfhood). They state that mechanisms of body–mind integration improve self-awareness. Regarding physiological changes, the effects of dance and DMT are also detectable on a neural and hormonal level (e.g., Quiroga Murcia et al., 2009 ; Stück and Villegas, 2017 ; Abraham et al., 2018 ). Studies assessed positive symptoms in schizophrenia with self-report questionnaires or clinical interviews/observations, whereas physiological data were collected using medical examination procedures (e.g., sphygmomanometer).
The following inclusion criteria were used to filter studies for meta-analysis:
Because we wanted to get a broad picture, we also included studies with interventions named “body psychotherapy” or “movement integration.” We decided to classify an intervention as “DMT intervention” if a dance movement therapist conducted the session. When the qualification of therapist was missing (or unclear), we closely analyzed the descriptions of the intervention. If the intervention description suggested that predominantly typical tools of DMT were used (therapeutic use of dance/movement involving mirroring, conscious social interaction in movement, introspection and reflections on movement and body sensations; see also definition of DMT and description of DMT methods in Koch, 2019 ), we categorized the intervention as DMT. Trials that included dance elements but predominantly used methods of other creative arts therapies were excluded (Mateos-Moreno and Atencia-Doña, 2013 ; Jakobsen et al., 2017 ; Van Westrhenen et al., 2019 ). In contrast, we allocated methods to the dance intervention group, if they were dance training sessions conducted by dancers or exercise instructors from various backgrounds (physiotherapists, nurses, fitness instructors). We decided to include group as well as individual therapy sessions.
For this meta-analysis, we used multiple search strategies. First, we systematically searched electronic databases, namely, Psyndex, PsycINFO, ERIC, CENTRAL, and Google Scholar. We used different terms for dance movement therapy (DMT) and dance-related interventions as single keywords and combined search terms, putting together the single keywords and terms related to the study design (see Box 1 ).
• dance movement psychotherapy •dance movement therapy • dance therapy • therapeutic movement • dance-effectiveness • dance • dance-therapy • expressive movement • expressive dance | • dance movement psychotherapy + controlled trial • dance movement therapy + controlled trial • dance therapy + controlled trial • dance movement psychotherapy + random • dance movement therapy + random • dance therapy + random • dance movement psychotherapy + controlled trial • dance movement therapy + controlled trial • dance therapy + controlled trial • dance movement psychotherapy + random • dance movement therapy + random • dance therapy + random • dance movement psychotherapy + controlled trial • dance movement therapy + controlled trial • dance therapy + controlled trial • dance movement psychotherapy + random • dance movement therapy + random • dance therapy + random • dance movement psychotherapy + controlled trial • dance movement therapy + controlled trial • dance therapy + controlled trial • dance movement psychotherapy + random • dance movement therapy + random dance therapy + random |
The number of hits is reported in the flowchart below ( Figure 1 ). Additionally, we conducted a hand search examining professional journals without widespread indexing, sending requests for unpublished and in process EBM studies to national and international professional listservs, and directly wrote to researchers with a history in EBM research on dance therapy, asking for references we may have potentially missed; we also included references that were mentioned in some of the secondary studies and that slipped by our systematic literature search (we received detailed responses from Dr. Vicky Karkou, Dr. Bonnie Meekums, Dr. Iris Bräuninger, Susanne Bender, Indra Majore-Dusele, and others).
Flowchart of selection procedure. Included studies are displayed in the study characteristics overview ( Table 1 ).
The systematic database search yielded 132 records and was supplemented with 59 studies identified through hand search. Because each included research report contained only one study, we did not have to differentiate between the analysis of research reports and studies. After removing duplicates (i.e., the same studies found in different data sources), a total of 183 studies was screened. Fifty-four of these studies (30%) met the inclusion criteria, and their full-article texts were then assessed for eligibility.
We had to exclude 14 studies for the following reasons (more information in Appendix ):
Finally, we were able to include a total of 41 studies into the analysis (see Figure 1 ).
We coded the study characteristics (see Table 1 ) using the following variables:
The analysis was done in SPSS (IBM, Version 25), employing meta-analysis macros by Wilson ( 2005 ). To synthesize data, we decided to use calculations of effect sizes. There were four levels of analysis (see Figure 2 ).
Hierarchy of effect sizes. EG, experimental group; CG, control group; d(c), weighted mean effect size per cluster; d(i), weighted mean effect size per study; d(i,c), weighted mean effect size per cluster within one study.
Most studies were using multiple scales or tests to measure the changes from pre- to post-intervention in one dependent construct (e.g., mental health, general health, and vitality as facets of quality of life). Furthermore, usually more than one dependent construct was observed in one study (e.g., quality of life, affect, and interpersonal skills). Thus, we had to synthesize data on several levels ( Figure 2 ). From Level IV to Level III and from Level III to Level II effect sizes within studies were synthesized. From Level II to Level I effect sizes across studies were synthesized. We started our analysis at the bottom level (Level IV), calculating effect sizes for each dependent variable (e.g., scales, tests). An adjusted formula of Cohen's d was used (Cohen, 1988 ):
This formula incorporates differences between pre- and post-intervention values in the intervention group controlling for changes that occurred in the control group (for abbreviations, see list of symbols in Appendix ). The effect is standardized using the pooled standard deviation, which is formed by the pooled variance of intervention and control group at the pre-intervention level. The natural variance of characteristic values in one sample is assumed to be confounded by the intervention; this is why the pooled variances at pre-time are more precise estimates of variation (Lipsey and Wilson, 2001 ).
Some studies reported median instead of mean and standard error of the mean or interquartile range instead of standard deviation. In these cases, we used the median as a mean and recalculated dispersion measures to approximate the effect sizes (Higgins and Green, 2008 ). In addition, we adjusted the polarization of the scales multiplying with −1, where necessary. Where the polarization of scales or tests remained unreported in the original study, we conducted a literature search to ascertain the direction of the effect. All effects were polarized to ensure that a positive effect size means that the health or resources of the participant improved (e.g., more interpersonal skills, less depression). In total, we calculated 306 effect sizes ranging from d = −0.82 to a maximum of 9.61. Sixteen effect sizes were larger than 3.0 and could be viewed as outlier effect sizes that could have serious impact on the mean effect size and the moderator analyses. Therefore, to avoid an overestimation of the effectiveness (Lipsey and Wilson, 2001 ), we recoded all effect sizes greater than three to d = 3.0, because it seems that those effect sizes are not realistic within intervention studies (Beelmann, personal communication). Next, calculating mean effect sizes, we aggregated the data of multiple dependent variables that measured changes in one dependent construct (outcome cluster). As a result, we obtained 78 effect sizes, with each effect size referring to one dependent construct in one study ( d c,i , Level III). Again, we calculated mean effect size, to synthesize the effect sizes of the dependent constructs into one mean effect size per study ( d i , Level II). Then, we planned to analyze data across studies (Level I) using methods developed by Hedges and Olkin ( 1985 ). Because studies with a greater sample size are generally more conclusive than smaller studies, each study was weighted with:
To get one overall weighted mean effect size, we calculated:
To obtain weighted mean effect sizes for outcome clusters (dependent constructs), we used the same formula replacing the mean effect sizes per study ( d i ) with the mean effect sizes per construct per cluster ( d c,i , Level III). Our analysis did not correct for multiple testing in order not to lose power. Instead, the chosen method is strong in avoiding Type I error. Having conducted 78 tests, it is important to note that four results may have been randomly significant (expecting five randomly significant results out of 100 tests).
Heterogeneity is an important issue in meta-analysis. It deals with the question of whether all included studies measure the same intervention effect. Usually, if there is a more general research question and there are less strict inclusion criteria, as it is the case in our analysis, we expect that we can observe more differences between studies in content- and method-related characteristics. We analyzed the heterogeneity of the effects using Q -statistics. A significant Q means that the heterogeneity is larger than we would expect from sampling error. In this case, we would need to calculate a random instead of a fixed-effect model (Lipsey and Wilson, 2001 ). In a random-effect model, we use a new inverse variance component resulting in greater confidence intervals of the effect sizes. From Q -statistics, we can calculate I -square, which is an estimator for the inconsistency of the results (Higgins et al., 2003 ). It is interpreted as the percentage of variance of the effect that is caused by heterogeneity (25% = low heterogeneity, 50% = moderate heterogeneity, 75% = high heterogeneity; Cohen, 1988 ).
Due to the file-drawer problem (i.e., the problem that studies failing to produce a statistically significant result are less likely to be published than those that do produce a statistically significant result), meta-analyses are in danger of overestimating the effects of an intervention (Rosenthal, 1979 ). We used a funnel plot and trim-and-fill analysis, two methods for assessing publication biases, to explore whether this might be the case in our analysis. We also investigated outliers that might bias the results.
To address issues of heterogeneity and to obtain a better understanding of which study characteristics might influence the assessment of effect sizes, we conducted sensitivity analysis for categorical variables (METAF, Macros from David B. Wilson; Lipsey and Wilson, 2001 ) and for metric variables (METAREG, Macros from David B. Wilson; Lipsey and Wilson, 2001 ). The study characteristics were incorporated as moderators of the effects.
To compare different control group types (e.g., waiting-list control group, physical exercise control groups), we additionally conducted separate analysis of studies with more than one control group, to ensure that all other study characteristics remained constant.
To obtain information about long-term effects, we created a separate file to analyze follow-up data. Using the same procedure as described above, we calculated the effects from pre-test to follow-up values.
Countries of publication.
The incorporated studies came from 14 different countries. Most studies were conducted in Germany (11 studies, 26.8%). Overall, 25 studies (60.98%) were conducted in Europe. In addition, eight studies (19.51%) were conducted in Asia, three studies (7.32%) in North America, two studies (4.88%) in Australia, two studies in South America (4.88%), and one study in Africa (2.44%). Generally, one can observe that most studies stemmed from “Western” countries (Germany, UK, USA, Canada, Australia, Mediterranean, and Scandinavian countries). Most Asian studies were conducted in Hong Kong, Korea, and Japan. The remaining studies stemmed from Malaysia, India, and Nigeria. We found seven studies that had not been published yet (Aweto et al., 2012 ; Ventura et al., 2016 ; De Natale et al., 2017 ; Koch, 2017 ; Pohlmann et al., 2017 ; Abraham et al., 2018 ; Aithal and Karkou, 2018 ). Figure 3 displays the number of studies that met the inclusion criteria found per year. The search dates were January 2012 to March 2018. While dance and DMT studies were equal in numbers in 2012, there were more DMT studies in 2013–2015, and more dance studies in 2016–2018.
Included studies per year with DMT or dance intervention.
Twenty-nine studies (70.73%) implemented the intervention in a clinical sample, and 11 (27.5%) were conducted with a non-clinical (educational) sample. From the clinical populations, 11 were diagnosed with affective disorders (depression), or reported feelings of stress, sadness, or anxiety. Six samples were diagnosed with Parkinson's disease, five with schizophrenia, four with ASD, and two with cognitive impairments. Overall, there were more female than male participants (percentage female participants: M = 65.66%, SD = 27.65). Comparing intervention and control group, the distribution of female and male participants was even.
About half of the studies were DMT intervention studies (21 studies); the other half were dance intervention studies (20 studies). In the DMT group , 16 interventions were provided by a trained dance movement therapist, at least four of them were also qualified clinical psychologists. In four trials, descriptions of qualification of therapists were missing or dance movement therapists in training (Aweto et al., 2012 ; Serrano-Guzmán et al., 2016 ; Guerra-Balic et al., 2017 ; Wiedenhofer and Koch, 2017 ). We allocated them to the DMT group because the intervention was predominantly DMT (distinguishing criterion is described above). In the study by Abraham et al. ( 2018 ), classification of qualification of therapist was difficult. The intervention was provided by a physiotherapist specialized at “dynamic neuro-imagery” intervention, a DMT-like introspection intervention. We decided to allocate the study to the DMT group, because of content-related similarities of the intervention to DMT. In the dance intervention group , seven interventions were conducted by an instructor with dance education. Additionally, seven interventions were conducted by exercise instructors with other backgrounds (physical education, physiotherapy, nursing). Six trials contained no description about qualification of dance instructors and were allocated to the cluster because of content-related similarities to the other dance intervention studies.
The DMT studies differed in the technique, e.g., the technique by Marian Chace (“Chacian approach”, a standard method in DMT; Sandel et al., 1993 ), dance therapy with ethno-elements, DMT modified to suit Chinese culture, manualized body psychotherapy (conducted by dance therapists; manual of Röhricht and Papadopolous, unpublished), and dance therapy with elements of creative movement play (for more extensive description of intervention methods in DMT, see Koch, 2019 ). Dance interventions used various dance styles, mostly traditional folk dance or cultural dance forms (Irish set dancing, Greek traditional dance, Flamenco, Poco-Poco dance, Belly dance, Tango), further couple dance (Latin dance), and contemporary dance (jazz, modern, creative dance). All interventions took place in groups. DMT and dance interventions were modified to the age of the participants. In the studies of Meric and Ilhan ( 2016 ) and Panagiotopoulou ( 2018 ), the researcher and the implementing person were identical. The quality of intervention description differed widely across studies. Fifteen percent of the studies described the intervention very rudimentarily, 22.5% moderately, 22.5% more detailed, and 40% replicable, which is a clear increase in replicable descriptions compared to the seven studies in the former meta-analysis by Koch et al. ( 2014 ).
To control for unspecific effects, about one third of the studies (31.7%) used active control groups [e.g., listening to music, cycling, Pilates, physical education/sports exercises (physical exercises), psychoeducation, meditation, relaxation exercises], and 65.9% of the studies used a passive control group (e.g., waiting-list group or treatment as usual). One control group description was missing (Meric and Ilhan, 2016 ).
Most of the included studies were implementing the intervention over a longer period of time (Mean = 11.41 weeks, SD = 7.89, Min = 1 week, Max = 32 weeks). Only the studies by Campion and Levita ( 2014 ) and Wiedenhofer and Koch ( 2017 ) implemented one single-dance/DMT session and assessed short time effects immediately after the intervention. Their sessions lasted for 5 min in the study by Campion and Levita ( 2014 ) and for 50 min in the study by Wiedenhofer and Koch ( 2017 ). The other sessions were usually carried out two times per week (Mean = 1.85, SD = 1.05, Min = once per week, Max = 4 times per week), and lasted for 1 h or 1 1/2 h (Mean = 71.32 min, SD = 19.02, Min = 40 min, Max = 120 min).
To assess and compare pre–post differences of the dependent variable in the intervention and control group, either t -tests, ANOVAs, or MANOVAs were calculated. Mostly self-report questionnaires were used for assessment (49%), followed by observation rating scales (17.6%), cognitive tests (13.4%), tests of motor functioning (11.1%), interviews (6.5%), and psychophysiological measures (2.3%).
The meta-analysis included studies with large and very small sample sizes ( N: Mean = 57.90, SD = 49.61, Min = 11, Max = 275). In 21.95% of the cases, the samples were smaller than N = 30, 34.15% were N = 30–50, 31.70% N = 50–100, and 12.20% were larger than N = 100. The studies with the largest sample sizes are contributing most to the results (weighted mean effect sizes); these are (Priebe et al., 2016 ) ( N = 275), (Vankova et al., 2014 ) ( N = 162), (Bräuninger, 2012a ) ( N = 162), (Ho et al., 2016a ) ( N = 139), and (Duberg et al., 2013 ) ( N = 101). In general, smaller studies are less conclusive and generalizable than studies with larger sample sizes (unless their methodological quality is significantly better). Therefore, we weighted their results discriminatingly, calculating weighted mean effect sizes (see previous paragraph). Notably, most studies did not meet established criteria or rules of thumb that would define desirable sample sizes to conclude effects for a certain population (Harris, 1985 ; Kraemer and Thiemann, 1987 ; Green, 1991 ). Nevertheless, it is assumed that (apart from publication bias) studies with a small sample size would not significantly bias the results of meta-analyses but contribute important information, for example, about heterogeneity and the effects in sub-groups (Higgins et al., 2003 ; Grainge, 2015 ).
Randomization is an important criterion for the reliability and validity of estimated effect sizes, because it addresses the issue of comparability of the inspected groups. Fifty-two and a half percent (52.5%) of the studies used randomization as group allocation process, 22.5% used quasi-randomization techniques, and 25% of the studies used no randomization techniques (group allocation was based on self-selection or purposive sampling). Of the 10 non-randomized studies, only 5 included extra assessments of baseline differences, showing that there were significant differences in three of them (Pylvänäinen et al., 2015 ; Adam et al., 2016 ; Aithal and Karkou, 2018 ). The inclusion of studies with baseline differences is justified here, because our methodological approach takes those differences into account when calculating effect sizes. The studies from Biber ( 2016 ), Guerra-Balic et al. ( 2017 ), Meric and Ilhan ( 2016 ), and Panagiotopoulou ( 2018 ) used no statistical tests to assess baseline differences.
Thirty-one of the 41 studies reported dropout rates (Mean = 13.55%, SD = 15.18, Min = 0%, Max = 57.38%). Analysis of dropout rates revealed that 7 of the 31 studies reported a dropout that is higher than 30% (Pinniger et al., 2012 , 2013 ; Röhricht et al., 2013 ; and Rios Romenets et al., 2015 ; Hildebrandt et al., 2016 ; Martin et al., 2016 ; Pohlmann et al., 2017 ). This was in many cases due to the fact that severely impaired patients with schizophrenia or depression had to actively travel to outpatient treatment, requiring a strong motivational state, which is precisely one of the problems in these disorders.
Because we applied mild selection criteria, the methodological quality of the included primary studies is considerably heterogeneous, and risk of bias is a concern in most of the included studies. Referring to the Cochrane Collaboration's tool for assessing risk of bias in randomized trials, there are six domains of bias that should be considered: selection bias, performance bias, detection bias, attrition bias, reporting bias, and other sources of bias (Higgins et al., 2011 ). Selection bias relates to the group allocation of participants. As reported above, a quarter of the studies had a great risk of selection bias (or in this respect), because no randomization or quasi-randomization tools were used. Additionally, in 12.2% of the trials, no baseline characteristics were checked. Secondly, performance bias relates to blinding of the participants and staff involved. Any of the investigated trials might be affected by this risk of bias (Rosenthal effect), because blinding is a challenge in any type of intervention study. While it is possible to conceal, which is the experimental vs. control condition in active control group trials, it is more difficult in waiting list designs, such as is the case in many of the DMT and dance interventions here. However, it is plausible that self-report measures and external rating scales are more affected by performance bias than cognitive or motor tests or physiological data. Detection bias addresses blinding of the researcher, who assesses and analyses the results. We were not able to extract this information from most of our primary trials, but we assume that researchers usually know the treatment allocation of the participants. Attrition bias is about the issue of the amount and handling of incomplete outcome data. As we assessed dropout rates (see section above), we can state that about 42% of the studies are in danger for this domain of risk of bias. Because selective reporting (reporting bias) is hard to extract from primary trails, we did not assess this domain. Concerning other risks of bias (e.g., conflicts of interest), in two studies, the researcher was also the implementing person (Meric and Ilhan, 2016 ; Panagiotopoulou, 2018 ). To conduct further analyses, we used dichotomous coding (higher-risk vs. lower-risk studies), including all studies in the higher-risk category that had a total N smaller than 30 or which yielded issues in any of the assessed domains of risk of bias (group allocation, attrition bias, other bias). A more detailed description on study characteristics is provided in Table 1 (study characteristics chart).
Overall, we synthesized data from 2,374 participants from 41 studies (more than twice the number of the last general meta-analysis by Koch et al., 2014 ). According to Cohen ( 1988 ), effect sizes between d = 0.2 and d = 0.5 are small effects, effect sizes between d = 0.5 and d = 0.8 are medium effects, and effect sizes larger than d = 0.8 are large effects. The mean effect sizes per study (Level II, k = 41) varied between one small negative effect size and large positive effect sizes (Unweighted Mean: d = 0.67, Min: d = −0.17, adjusted Max: d = 2.96). The maximum effect size is not bigger than 3.0, because of the adjustments we obtained at Level III. The unadjusted Maximum would be d = 5.07. Further details about the distribution of effect sizes across studies are described in the paragraphs below.
We calculated a fixed-effect model to obtain a weighted mean effect size across studies (Level I). The weighted overall fixed effect was d = 0.48 ( p < 0.001, CI min = 0.40, CI max = 0.57). The analysis of heterogeneity showed a high inconsistency of results ( Q = 127.52, df = 40, p < 0.001, I 2 = 72.62%). Therefore, we calculated a random-effect model as recommended in Lipsey and Wilson ( 2001 ). The estimated weighted overall random effect was d = 0.60 ( p < 0.001, CI min = 0.44, CI max = 0.76). In the following paragraphs, we will only report results that were calculated with the random-effect model.
The weighted mean effects sorted by constructs (outcome clusters) are summarized in the chart below. The analysis yielded a significant effect for each outcome cluster. The effect for interpersonal skills was the largest one, followed by quality of life and (psycho-)motor skills. Furthermore, all effects but the effect for (psycho-)motor skills showed high heterogeneity. Since we calculated a random-effect model, all effects showed broad confidence intervals (see Table 2 ).
Effect sizes according to outcome cluster.
% | ||||||||
---|---|---|---|---|---|---|---|---|
Quality of life | 20 | 0.67 | 0.41–0.99 | 0.133 | <0.001 | 89.30 | <0.001 | 78.72 |
Affect | 23 | 0.56 | 0.34–0.79 | 0.115 | <0.001 | 88.65 | <0.001 | 75.18 |
Interpersonal skills | 9 | 0.85 | 0.41–1.28 | 0.222 | <0.001 | 38.61 | <0.001 | 78.15 |
Cognitive skills | 10 | 0.53 | 0.13–0.93 | 0.204 | 0.009 | 28.51 | 0.001 | 68.43 |
(Psycho-)motor skills | 10 | 0.65 | 0.36–0.96 | 0.152 | <0.001 | 14.61 | 0.102 | 38.39 |
Residual | 6 | 0.47 | 0.06–0.88 | 0.208 | 0.025 | 20.04 | 0.001 | 75.05 |
k, number of studies; ES, effect size; CI, confidence interval; SE, sampling error;
The subanalysis of anxiety and depression revealed that the differentiation of the two constructs did not lead to more homogeneous results. Depression yielded a slightly larger effect than anxiety. Both effects were medium and showed high heterogeneity. The effect for physiological variables was large but heterogeneous; the effect for positive symptoms in schizophrenia reached significance on the p < 0.1 level and was also heterogeneous (see Table 3 ).
Effect sizes according to sub-clusters.
% | ||||||||
---|---|---|---|---|---|---|---|---|
Anxiety | 9 | 0.47 | 0.09–0.84 | 0.192 | 0.015 | 42.88 | <0.001 | 81.34 |
Depression | 18 | 0.54 | 0.30–0.78 | 0.124 | <0.001 | 65.35 | <0.001 | 73.99 |
Physiological variables | 2 | 0.88 | 0.22–10.54 | 0.338 | 0.009 | 2.63 | 0.105 | 61.98 |
Schizophrenia (pos symptoms) | 4 | 0.40 | −0.01–0.79 | 0.205 | 0.05 | 7.99 | 0.046 | 62.45 |
The one-session study from Campion and Levita ( 2014 ) was the only one to show a (small) negative effect ( d = −14). Other studies that were at the lower end of the distribution were Koch et al. ( 2015a , b ) that included an anti-violence training ( d = 0.02) and Ho et al. ( 2016a ) that aimed to improve the quality of life of cancer patients ( d = 0.06). On the other side of the distribution, we could observe large effect sizes ( d = 1.8 up to d = 2.96) in the studies of Cruz-Ferreira et al. ( 2015 ), Biber ( 2016 ), Teixeira-Machado et al. ( 2017 ), and Aithal and Karkou ( 2018 ). If we had not recoded the effect sizes from Chiang et al. ( 2016 ), Pohlmann et al. ( 2017 ), Teixeira-Machado et al. ( 2017 ), and Aithal and Karkou ( 2018 ) at Level IV, they would have been even larger (up to d = 5.07). Possible reasons for the effect size distribution across studies are detailed in the next paragraphs.
To analyze the distribution of effect sizes and to address the issue of publication bias, we created a funnel plot (see Figure 4 ) that shows the distribution of unweighted effect sizes as a function of sample sizes. The distribution would be asymmetrical in case of a publication bias, because then small sample studies with positive effect sizes would be published, whereas small sample studies with no positive effects would remain unpublished (Lipsey and Wilson, 2001 ).
Funnel plot.
The diagram in Figure 4 shows a skewed distribution to the right; that is, the large positive effect sizes on the right have no equivalent on the left. This is an indication of a publication bias. Therefore, we conducted a regression analysis (independent variable: sample size, dependent variable: unweighted effect size). There was a small negative gradient parameter that did not reach significance (β = −0.003, p = 0.147). We can thus assume that the publication bias does not have a great impact on the results. Nevertheless, the fact that the gradient parameter did not reach significance does not mean that there is no publication bias at all. We also conducted trim-and-fill analysis as a more elaborate method to explore publication bias using the random-effect model and an L 0 estimator (Duval and Tweedie, 2000 ). The number of estimated missing studies on the left side, meaning the number of studies that are assumed not to be found due to publication bias, was zero. Thus, the new estimated overall effect did not differ from the original one.
As described above, we conducted analysis of sensitivity using study characteristics as metric and categorical moderator of the effect sizes per study ( d i ).
Regression analyses revealed that no metric variable alone reached significance as a moderator of effect sizes (see Table 4 below). N total was closest to significance, indicating that studies with a greater sample size yielded smaller effects. This can be interpreted as a result of the file-drawer problem, risk of bias of smaller studies, or as an estimator of quality of implementation (we can assume that, in smaller samples, the intervention was implemented more appropriate to the individual) (see Table 4 ).
Metric moderators.
total | −0.0025 | 0.074 |
Percentage women (total) | 0.1739 | 0.592 |
Relation women EG vs. CG | 0.2861 | 0.477 |
Length process (weeks) | 0.0096 | 0.356 |
Length session (min) | −0.0024 | 0.545 |
Frequency (times per week) | 0.1375 | 0.091 |
Dropout percentage EG | −0.0087 | 0.257 |
Dropout percentage (total) | −0.0092 | 0.172 |
EG, experimental group; CG, control group; β, gradient parameter; p, significance parameter .
Overall, 40.84% of variance ( R 2 ) was explained by metric moderators.
We conducted sensitivity analysis with the following categorical moderators: country, publication status, higher vs. lower risk of bias, age range, clinical vs. non-clinical sample, diagnosis, DMT vs. dance intervention, qualification of implementing person, quality of intervention description, one-session vs. process, randomization, dropout rates, type of control group activity, and measurement type.
Two moderators were significant on a p < 0.05 level: DMT vs. dance intervention studies (between variance: Q = 5.54, df = 1, p = 0.019), and qualification of implementing person (between variance: Q = 8.89, df = 3, p = 0.031). Trials with DMT interventions yielded significant but slightly lower effect sizes ( d = 0.35, p < 0.001) than trials with dance interventions ( d = 0.81, p < 0.001). In addition, the results of the group with DMT interventions were consistent ( I 2 = 3.47), whereas the results in the dance intervention group were inconsistent ( I 2 = 77.96). Country, higher vs. lower risk of bias, and age range were significant moderators of effect sizes on a p < 0.1 level. In children, teenagers, and seniors, larger effect sizes were observed than in adults, but the effect sizes in adults were more consistent. Furthermore, lower-risk studies yielded smaller effect sizes than higher-risk studies and were more homogeneous. In clinical trials, effect sizes were smaller than in non-clinical trials, but the results remained more consistent than in non-clinical trials (weighted mean effect sizes sorted by group, see Table 5 ).
Categorical moderators.
, , ) | , , ) | , , ) | , , ) | , , ) | , , ) | , , ) | |
---|---|---|---|---|---|---|---|
Country | 9.68 5 0.085 | Germany (0.46, 0.002, <1%) | North/Middle/West Europe (0.44, 0.028, <1%) | South/East Europe (0.81, <0.001, 61.26%) | Far East (0.37, 0.142, <1%) | USA/Australia/Canada (0.45, 0.065, 20%) | Others (1.17, <0.001, 61.79%) |
Higher vs. lower risk of bias | 3.27 1 0.071 | Lower risk (0.48, 0.000, 8.68%) | Higher risk (0.76, 0.000, 34.48%) | – | – | – | – |
Randomization | 2.27 2 0.320 | No randomization (0.85, <0.001, 44.17%) | Quasi- randomization (0.48, 0.003, <1%) | Randomization (0.57, <0.001, 37.48%) | – | – | – |
Age range | 10.71 5 0.098 | Children (1.40, <0.001, 79.51%) | Teenager (0.93, 0.003, 79.29%) | Young adults (0.55, 0.051, 27.18%) | Younger + older adults (0.46, 0.000, <1%) | Older adults (0.30, 0.172, <1%) | Seniors (0.83, <0.001, 43.75%) |
Diagnoses | 4.17 5 0.654 | Cognitive impairment (0.87, 0.017, 68.75%) | Affective disorders/stress (0.57, <0.001, <1%) | Developmental disorders (0.41, 0.106, <1%) | Schizophrenia (0.40, 0.075, <1%) | Parkinson (0.53, 0.033, <1%) | None (0.90, <0.001, 72.41%) |
df, degree of freedom (df = k – 1) .
Across all studies, the type of control group activity was not a significant moderator of effect sizes. A different picture resulted, when we only included studies with more than one control group into the analysis ( k = 9, active control groups: meditation, physical exercises). For these studies, the weighted mean effect size remained significant when compared to non-active control groups (in most cases, waiting list CGs), but declined to a visible but non-significant effect when compared to active control groups ( d = 0.43, p = 0.009 vs. d = 0.21, p = 0.126). Especially when compared to meditation intervention, the effects declined to almost zero, which means that the two interventions are equally effective.
Type of measurement was a significant moderator of effect sizes (between variance: Q = 97.92, df = 5, p < 0.001). Observation rating scales yielded the largest effect sizes ( d = 1.34, p < 0.001, I 2 = 76.01%). They were followed by physiological data ( d = 0.71, p = 0.001, I 2 = n.s.), motor tests ( d = 0.58, p < 0.001, I 2 = n.s.), self-report-questionnaires ( d = 0.42, p < 0.001, I 2 = n.s.), cognitive tests ( d = 0.30, p = 0.002, I 2 = n.s.), and interviews ( d = 0.25, p = 0.001, I 2 = n.s.). We thus can assume that the type of measurement, especially observation rating scales, is one of the most significant sources of heterogeneity across studies. We also checked whether observation rating scales were more frequently used in dance intervention studies than in DMT studies, which was not the case.
Since sensitivity analysis revealed type of intervention as a significant moderator, we decided to exploratively conduct separate analyses of the DMT and dance intervention trials (the results are reported in the Tables 6 , ,7 7 below; they were also reported briefly in the paragraph above).
Effect sizes only in DMT studies.
% | ||||||||
---|---|---|---|---|---|---|---|---|
Overall | 21 | 0.35 | 0.23–0.46 | 0.059 | <0.001 | 20.72 | 0.414 | 3.47 |
Quality of life | 10 | 0.32 | 0.02–0.63 | 0.155 | 0.036 | 1.50 | 0.997 | n.s. |
Affect | 12 | 0.51 | 0.18–0.85 | 0.171 | 0.003 | 9.49 | 0.577 | n.s. |
Interpersonal skills | 6 | 0.49 | 0.00–0.97 | 0.249 | 0.051 | 0.43 | 0.994 | n.s. |
Cognitive skills | 3 | 0.26* | (−0.48)−0.99 | 0.268 | 0.011 | 0.23 | 0.890 | n.s. |
Motor skills | 2 | 0.30 | (−0.28)−0.88 | 0.297 | 0.315 | 0.14 | 0.712 | n.s. |
Effect sizes only in dance intervention studies.
% | ||||||||
---|---|---|---|---|---|---|---|---|
Overall | 21 | 0.81 | 0.52–10.11 | 0.149 | <0.001 | 86.22 | <0.001 | 77.96 |
Quality of life | 10 | 1.02 | 0.69–1.34 | 0.165 | <0.001 | 23.41 | 0.005 | 61.55 |
Affect | 11 | 0.69 | 0.34–1.04 | 0.177 | <0.001 | 16.63 | 0.083 | 39.86 |
Interpersonal skills | 3 | 1.65 | 0.90–2.40 | 0.383 | <0.001 | 8.71 | 0.013 | 65.57 |
Cognitive skills | 7 | 0.68 | 0.16–1.21 | 0.268 | <0.001 | 10.85 | 0.093 | 44.65 |
Motor skills | 8 | 0.76 | 0.44–1.08 | 0.163 | <0.001 | 8.67 | 0.277 | 19.23 |
We also analyzed weighted mean effect sizes according to outcome clusters separately for the two groups. For quality of life, the effect remained significant in both clusters, but showed consistency only in the DMT group. For clinical outcomes, both effects were significant but heterogeneous. Related to interpersonal skills, there was a significant but inconsistent effect in the dance intervention group and a significant ( p = 0.05) and homogeneous effect in the DMT group. In the cognitive skills cluster, the effect remained significant but heterogeneous in the dance intervention group, but did not reach significance in the DMT group ( k = 3 trials). The effect for motor skills was significant and consistent in the dance intervention group; in the DMT group, it was not significant ( k = 2 trials).
In Tables 6 , ,7, 7 , changes in depression and anxiety are the main outcomes under affect changes . It is notable that the changes in interpersonal skills just barely missed significance in DMT studies. Considering the high heterogeneity of outcome measures in this domain, they need in any case to be further investigated (plus more standardized outcome measured need to be developed). Changes in motor skills are most pronounced and consistent in dance intervention studies, with the most and the most rigorous evidence from studies of interventions for Parkinson patients (e.g., Hackney and Bennett, 2014 ; Sharp and Hewitt, 2014 ; Loetzke et al., 2015 ). Changes in motor skills are usually not in the focus of DMT studies. The non-significant results on motor skills confirm DMT as a psychotherapeutic intervention.
In order to obtain information about the long-term effects of DMT and dance interventions, we analyzed the available follow-up data (see Table 8 ). Eight studies reported follow-up data, the mean period from post- to follow-up assessment was 22 weeks. In most studies, the effect remained constant. The largest decline was observed in the study by Bräuninger ( 2012a , b ). The effect was reduced by half, but remained significant. In Baptista et al. ( 2012 ), Pylvänäinen et al. ( 2015 ) and Priebe et al. ( 2016 ), there was a slight decline, which was close to zero; in De Natale et al. ( 2017 ), there was a slight increase. In the studies from Pinniger et al. ( 2012 ), Duberg et al. ( 2013 ), and Cruz-Ferreira et al. ( 2015 ), there was a large increase (e.g., in Pinniger the effect tripled). The authors explain this unusual increase with an increase in mindfulness (detachment from negative thoughts, ruminations, and worries, which led to a reduction in symptom severity; Ree and Craigie, 2007 ) and a personal bond between the participants as a learning process, which became more effective after the participants had practiced for a while.
Effect sizes follow-up.
d ) | d ) | d ) | ) | ||
---|---|---|---|---|---|
Baptista et al. ( ) -D | 32 | 0.48 | 0.39 | −0.09 | 18.22 |
Bräuninger ( , ) | 24 | 0.64 | 0.38 | −0.26 | 37.12 |
Cruz-Ferreira et al. ( ) -D | 24 | 2.54 | 3.00 | 0.46 | 7.83 |
De Natale et al. ( ) | 8 | 0.25 | 0.39 | 0.14 | 3.91 |
Duberg et al. ( ) -D | 48 | 0.51 | 0.81 | 0.30 | 24.40 |
Pinniger et al. ( ) | 4 | 0.27 | 0.82 | 0.55 | 11.59 |
Priebe et al. ( ) | 24 | 0.19 | 0.14 | −0.05 | 68.43 |
Pylvänäinen et al. ( ) | 12 | 0.74 | 0.65 | −0.09 | 7.18 |
Weighted mean ( ) | 22 | 0.67 (0.001) | 0.79 (0.001) | 0.05 (0.613) | – |
The bottom line displays the weighted mean effect sizes across the eight studies and a significance parameter (p-value); D, dance intervention study .
In this meta-analysis, we investigated the effectiveness of DMT and dance interventions on health-related psychological outcomes. We included 41 primary trails published between 2012 and March of 2018 that contained a total of 2,374 participants. Twenty-one of these trials considered DMT interventions, whereas 20 trials considered dance interventions. We grouped the dependent variables into six outcome clusters: quality of life ( k = 20 trials), clinical outcomes (23 trials; sub-analysis: depression, anxiety), interpersonal skills ( k = 9 trials), cognitive skills ( k = 10 trials), (psycho-)motor skills ( k = 10 trials), and residuals ( k = 6 trials; physiological data, positive symptoms schizophrenia). The overall mean effect size was d = 0.60 ( p < 0.001, CI min = 0.44, CI max = 0.76), which is a significant medium effect (Cohen, 1988 ). Analysis of heterogeneity yielded that an estimated 71.62% of variance of results can be traced back to dissimilarity of results, which makes the interpretation of results more difficult. In the outcome clusters, we also obtained significant medium effect sizes [quality of life: d = 0.67, affect: d = 0.56, cognitive skills: d = 0.53, (psycho-)motor skills: d = 0.65, residuals d = 0.47] and one significant large effect size (interpersonal skills: d = 0.85). This could be due to the fact that in the interpersonal skills cluster, many trials assessed dependent variables with observation rating scales, which yielded larger effect sizes in general. Sub-analyses revealed that the mean effect in the anxiety cluster was as large as the effect in the depression cluster. Physiological data yielded a mean large, significant effect size, whereas improvements of schizophrenia yielded a mean small effect size, which reached the p < 0.1 level. Assessment of heterogeneity in the outcome clusters revealed that all mean effects, except the effect for (psycho-)motor skills, remained inconsistent (i.e., significant Q, I 2 larger 60%). Furthermore, the larger the mean effect, the larger the heterogeneity of results, which indicates that larger effects were produced by outliers rather than by consistently higher effect sizes (see end of the next paragraph).
In order to identify which study characteristics contributed most to dissimilarities of results, we conducted a sensitivity analysis. Most importantly, we found that the type of intervention (DMT vs. dance) was a significant moderator of results. Therefore, it was reasonable to analyze data in two separate groups to obtain more meaningful results:
In the DMT group , we obtained smaller but more consistent effects. The overall medium effect was small, significant, and consistent/homogeneous ( d = 0.30, p < 0.001, non-significant Q, I 2 = 3.47). For the individual outcomes, the effects varied between d = 0.26 and d = 0.51 and were all homogeneous. The effects for quality of life, affect, and cognitive skills remained significant, whereas the effect for interpersonal skills reached the p < 0.1 level. Effects of motor skills were non-significant, thus confirming DMT as a mainly psychotherapeutic intervention.
In the dance cluster , we observed larger but less consistent effects. The overall medium effect was large, significant but non-consistent/heterogeneous ( d = 0.81, p < 0.001, significant Q, I 2 = 77.96). The effects in the outcome cluster varied between d = 0.68 and d = 1.65. They were all significant, but only the effects for (psycho-)motor skills, cognitive skills, and affect were consistent (non-significant Q, I 2 < 50%). Apart from Koch et al. ( 2015a , b ) and Aithal and Karkou ( 2018 ), all outliers we mentioned in the “Analysis of outliers and publication bias” section were part of the dance intervention group. The consistent effects for (psycho)motor skills were carried by the dance for Parkinson studies.
Dance studies seem to uphold certain characteristics that produce a broader range of results, especially in the large-positive spectrum of effect sizes (i.e., greater than d = 1.0) that are not evident in DMT studies. One factor might be different sample characteristics: Most DMT studies were conducted in a clinical setting, whereas most dance studies were conducted in a non-clinical setting. In severely impaired patient samples, such as in most of the DMT primary studies that entered into this analysis, effects are usually smaller than in non-clinical or subclinical populations (we know this, for example, from pretests of designs with student populations; e.g., Koch, 2011 ). This may be a major reason why the dance intervention studies have yielded larger medium effect sizes. Non-clinical samples also contained a broader age range. Additionally, in dance interventions, the implementation and methods were more heterogeneous than in the DMT group. In terms of culture, most DMT studies were conducted in Germany and other Western European countries, whereas the majority of dance intervention studies were conducted in non-Western countries. Dance intervention studies tended to have smaller sample sizes, less randomization, and more missing information in the reporting of results.
In sum, we obtained encouraging results, which indicated that DMT and dance have positive effects on various health-related outcomes. Most studies found evidence on the effectiveness of DMT on clinical outcomes ( k = 12 trials), followed by quality of life ( k = 10 trials) and cognitive skills ( k = 3 trials). There was also a tendency that DMT improved interpersonal skills ( k = 6 trials; p = 0.051). We did not find enough studies on (psycho-)motor functioning, physiological changes, and positive symptoms of schizophrenia in the DMT cluster to draw conclusions. Dance interventions improved (psycho-)motor skills ( k = 8 trials), clinical outcomes ( k = 11 trials), and cognitive skills ( k = 7 trials). The high variety of results, especially in the dance cluster, needs more investigation in future studies.
In our meta-analysis, we also assessed and analyzed various study characteristics to deepen our understanding of factors that influenced the effects of DMT and dance on health-related outcomes. One important issue in most intervention studies is the question about unspecific and specific effects of the intervention . Often the fact that we pay attention to the participants already can change their symptoms (Hawthorne effect: firstly described in Roethlisberger and Dickson ( 1964 ), and reviewed by McCambridge et al., 2014 ). Using alternative interventions in the control groups helps to distinguish attention effects from specific effects. Therefore, we assessed control group activity. In the primary trials, one third of the DMT and dance intervention groups were compared to active control groups. The participants followed the following tasks: listening to music, cycling, Pilates, physical exercises, psychoeducation, meditation, or relaxation exercises. Control group activity was not a significant moderator of effect, which means that the effects of DMT and dance interventions were still evident when we controlled for attention effects. In other words, this indicates that there are specific effects of DMT and dance interventions. Control group activity seemed to matter, when we kept all other study characteristics constant, which was only possible in studies with an active and a non-active control group ( k = 9).
We compared the intervention groups to control groups that participated in physical exercises and meditation exercises. The effects of DMT and dance interventions declined but remained significant. The advantages of DMT and dance interventions were larger compared to physical exercises than to meditation. This indicates that the change mechanisms of DMT and dance possibly lie beyond the pure execution of sports or mindfulness. This is in line with other research on the mechanisms of change in DMT and dance (Bräuninger, 2014 ; Koch, 2017 ). The effect sizes observed in DMT and dance intervention groups were about as large as the effect sizes in meditation interventions. There is plenty of evidence that meditation interventions, for example, training programs in mindfulness-based stress reduction (MBSR; Teasdale et al., 2000 ; Michalak et al., 2008 ), are successful in treating several psychological conditions (e.g., depression, anxiety, stress, obsessive–compulsive disorders; Bohlmeijer et al., 2010 ; Fjorback et al., 2011 ). On basis of this evidence, there is also an increase in promotion and support of mindfulness-based interventions by the health insurance companies, such as free MBSR training programs at work. Thus, observing that DMT and dance interventions seem to be as effective as meditation suggests promoting these kinds of interventions as well. One interesting research issue that we should investigate in more detail is: What do DMT/dance interventions and mindfulness-based interventions have in common? Mindfulness is an important component of DMT. For example, mindful investigation of body sensations is part of DMT, with similar techniques as the body-scan exercise in MBSR training programs (Dreeben et al., 2013 ). Besides that, getting mindfully in contact with other people (therapist or other group members) is an important mechanism of change in DMT. Concerning dance interventions, there is less explicit use of mindfulness, but if we look closer, similar mechanisms of change stand out. One central component frequently mentioned in the literature on mindfulness is the termination of rumination and automatic negative thoughts, which occurs, because there is high concentration, focus on the here-and-now, and a state of mind, which allows the participant to observe feelings and thoughts from a more distant perspective (non-judgmental state of mind). Future research should investigate to what extent these mechanisms are relevant in DMT and dance intervention and how they influence the outcomes of the interventions. Moreover, we assume that an important overarching therapeutic factor in both mindfulness practice and DMT could be introspection (Price and Smith-DiJulio, 2016 ). Another central mechanism of change in dance and DMT might be the experience of flow (Csikszentmihalyi and Csikszentmihalyi, 1975 ). It is based on the same principles as mindfulness, such as high concentration, absorption, focus on the here-and-now, physical presence, and joy (Csikszentmihalyi and Csikszentmihalyi, 1975 ).
In order to find out which populations DMT and dance interventions are indicated for, we investigated the influence of sample characteristics on the effect sizes. Age range and clinical vs. non-clinical populations were two correlating moderators on a p < 0.1 level. We observed larger mean effects in children and elderly than in younger and older adults. At the same time, there was higher heterogeneity of results in the group of children and elderly. The fact that outcome variables in children and elderly were assessed with observation rating scales more often might contribute to these differences. Furthermore, the larger effects in children may also have come about because there were more prevention studies or studies in educational context with children and older adults, whereas the adults were mostly severely impaired clinical populations. Prevention studies yielded larger effects but higher heterogeneity than clinical studies. Again, this means that rather than more pronounced effects, there are outliers at the large-positive spectrum of effect sizes. From these results, we cannot come to recommendations about the indication of treatment in certain populations. Further research is needed to ascertain moderators that produce the variety of results. Regarding outcomes characteristics , DMT is indicated when a psychological change is intended, whereas dance interventions are indicated to improve motor skills (the consistent effect only occurred on those).
The so-called method effect (“any characteristic of measurement procedure contributes to variance of scores”; Maul, 2013 ) is a widely discussed limitation of intervention studies. Thus, we assessed and analyzed methodological characteristics of studies. We found one significant moderator, type of measurement , and two moderators that reached a p < 0.1 level, country of publication and methodological quality of study (risk of bias, randomization, sample size). Observation rating scales yielded the largest effect sizes followed by physiological data, motor tests, self-report questionnaires, cognitive tests, and interviews. Some measurements may be superior to detect changes or more sensitive to certain risks of bias, therefore producing larger effects than others. For example, observation rating scales and self-report questionnaires are more likely to be affected by expectation effects, which will be discussed further in the limitations of the present study paragraph. However, it is important to mention that method effects can also systematically bias the results. There was also a correlation between country of publication and methodological characteristics of the trials (risk of bias, randomization, sample size), with both factors having a slight influence on the observed effects. This is not surprising, because standards and traditions of research vary between countries. In countries with less evidence-based research traditions, methodological standards tend to be less strict and resulting effects tend to be larger. However, the effect of country could also be related to other cultural factors such as experience with the specific dance form, or the (sub-)cultural value system around dance, which may all influence motivation for and responsiveness to treatment.
We also analyzed whether the intensity of treatment had an influence on effect sizes. The intensity of treatment was indicated by three factors: duration of the whole treatment, duration of one session, and frequency of treatment. None of these factors reached significance. Frequency of treatment was closest to significance (β = 0.1375, p = 0.091), indicating that higher frequency of treatment corresponded with slightly higher effect sizes.
In the last paragraphs, we discussed moderators that showed a significant influence on the effects of DMT and dance in our sensitivity analysis. However, we assume that there are more potentially important moderators of effects , which we were not able to address in our analysis. In the literature, there are factors that we can allocate to four clusters: (a) factors that relate to characteristics of the participants, (b) factors that relate to characteristics of the intervention, (c) factors that relate to characteristics of the implementing person, and (d) factors of environment and factors that relate to the relation between persons involved.
Firstly, we will discuss characteristics of participants. Savill et al. ( 2017 ) stated that gender is an important moderator for the effectiveness of body psychotherapy on negative symptoms in schizophrenia. They performed a secondary analysis with the data of a large multicenter randomized controlled trial by Priebe et al. ( 2016 ), the so-called “NESS paper.” The interaction between gender and treatment allocation as a predictor of outcomes was examined in 275 participants (72 women and 203 men) randomized to either a body psychotherapy or a Pilates group (for a critique of the study, particularly its control group selection, see the last paragraph of the discussion on the need for mechanism studies). Negative symptoms in schizophrenia “were found to significantly decrease in women randomized to the body psychotherapy condition in comparison to Pilates, while no such effect was detected in men” (Savill et al., 2017 , p. 1). To approach this issue, we also assessed percentages of female participants in intervention and control groups in our meta-analysis. We conducted a sensitivity analysis using the total percentage of female participants and the relation between female and male participants of intervention group and control group as a moderator. We found no significant influence of gender. This type of analysis has low power, which means that there is a high risk for a β-error. Thus, future research needs to investigate whether gender is an important moderator of effects. Further potential participant-related factors are self-efficacy or outcome expectations (Murrock and Madigan, 2008 ), attitude toward intervention (treatment adherence, motivation), and previous dance experiences. Regarding characteristics of DMT interventions, there is a high variety of methods and therapeutic styles, because only a few of the trials included manualized implementation of treatment (Martin et al., 2016 ; Priebe et al., 2016 ). Dance instructors used many different dance styles and teaching approaches. Although it was not possible in our meta-analysis, it would be useful to investigate differences between the effects of several methods and styles in the DMT and dance spectrum, in order to gain more knowledge about differential indications (Koch, 2019 ) and therapeutic factors. In her meta-analysis, Peters ( 2012 ) found qualification of therapist to be a significant moderator of effects. Concerning environmental factors, she discussed the influence of social support . Murrock and Madigan ( 2008 ) found that social support from friends mediated the effect between culturally specific dance and lifestyle physical activity. Finally, as in most therapeutic interventions, the relationship between the therapist and the participant is important (Grawe et al., 1994 ; Wampold and Irmel, 2015 ). As previously mentioned in the Introduction, the relationship between participants (group cohesion) is also assumed to be an important mediator of effects (Schmais, 1985 , 1998 ; Yalom, 1985 ) in interventions that are conducted in a group setting.
It is always difficult to observe long-term effects in meta-analyses. In our sample, only eight of the included trials reported follow-up data (mean period: 22 weeks). The analyses yielded various results, but most effects remained constant or increased. Regarding the encouraging follow-up findings, we assume that DMT and dance interventions have the potential to initiate a learning process (body access, interoception, insight) that might instigate positive changes several months after the intervention. Additionally, it is plausible that the permanence of effects depends on the participants' behavior after the intervention, such as revising what they have learned, continuing dance or movement classes, or maintaining contact with the other participants. Further research needs to follow to investigate long-term effects and their moderators.
One of the biggest limitations of our present study was the heterogeneity of results, which occurred due to mild inclusion criteria and various study characteristics of the included trials. The heterogeneity mostly concerned the dance intervention studies, whereas the DMT studies were rather homogeneous. Heterogeneity was caused, for example, by dissimilarities in methodological designs of studies, methods of interventions, and sample characteristics. Besides that, we observed effects on several different health-related psychological outcomes. For this reason, we employed a random-effect model, which considers that not all studies measure the same effect. Consequently, we obtained broad 95% confidence intervals, meaning that the “true effects” could also be much larger or smaller than the weighted mean effect sizes reported in this paper (e.g., some of the confidence intervals varied between no effect and medium effects or small effects and large effects). Such results are less conclusive. This leads to a typical critique of meta-analysis as “ comparing apples and oranges ” (Sharpe, 1997 ), which means that non-comparable outcomes are unjustifiably compared with each other. Yet, in cases where there is not a lot of evidence, it might be a useful start, if we want to obtain information about “fruits”—to stay in the metaphor. However, we should be aware that we lose specific information about discrete sorts of fruits and mostly create a starting point to generate useful hypotheses for future, more specific secondary analyses and primary trials.
Our meta-analysis is one of the first in the field to conduct sensitivity analysis , which is one approach to deepen the understanding about the sources of heterogeneity. Nevertheless, it is impossible to detect all important factors and to draw firm conclusions about causal relationships between those factors. The strength of our paper is that it provides a broad overview of current research on the therapeutic use of dance as an orientation for researchers (summarize findings, identify explanatory variables, help identifying research gaps and develop research questions, control standards of research). It informedly transfers knowledge about the effectiveness of DMT and dance interventions to practitioners, clients, and public decision-makers. However, the disadvantage of broad analyses is that we only obtain results for a rough orientation. It is the assignment of secondary analyses with more narrow research questions (such as provided in Cochrane Reviews for single clinical populations) to gain sharp and more detailed knowledge about the effectiveness of DMT and dance interventions and the interdependency with contributing contextual factors.
Another issue is the so-called “garbage-in–garbage-out problem,” meaning that the results are less conclusive if we include primary outcome trials with poor methodological quality. In the present study, we also included studies with considerable methodological constraints (e.g., small N , no randomization, high dropout, deficient report of implementation or statistics, conflicts of interest). The most important question is how much the methodological constraints may systematically bias the results reported in this paper (the weighted medium effect sizes). We approached the methodological variety of primary trials with sensitivity analysis. There was a tendency that more outliers, especially with large effect sizes, were studies with more severe methodological constraints. “Higher vs. lower risk studies” was a moderator of effects on a p < 0.1 level. The medium effect size in “lower risk studies” was d = 0.48 ( p < 0.001, I 2 = 8.68%), which is slightly lower and more homogeneous than the overall weighted mean effect size ( d = 0.60, p = 0.001, I 2 = 72.62%). This might be a hint that studies with more significant methodological constraints tend to overestimate the effects of DMT and dance. Since all of these problems were more pronounced in the dance intervention studies, one resulting recommendation is to separate DMT and dance intervention studies in the next general meta-analysis.
One important source of bias might be the “Rosenthal effect” or expectancy effect , which means that the expectations of the researcher are subtly communicated to the participants. By guessing the goal of the research, participants try to comply with its assumed goal (Rosenthal, 1966 ). Thus, self-report-questionnaires or observation rating scales are more prone to bias than cognitive, physiological, or motor test, which are somewhat less subjective, but still reactive (note that the reactiveness of measurement type ranked differently in our study as indicated above). Furthermore, it is possible that researchers tend to analyze, interpret, and report results in favor of positive effects, because of their own expectancies or potential conflicts of interest.
Another critical element that might contribute to the fact that higher-risk studies yielded larger effect sizes could be publication bias . One criterion of methodological quality of study was sample size. If there is publication bias, smaller studies yield larger effects, on average, because smaller studies, which did not detect positive effects, remained unpublished. In the distribution of our sample of primary trials, there was a small tendency for publication bias, which did not affect our results significantly (see Methods section). Furthermore, smaller studies yielding larger effects could also be explained by the fact that, in smaller samples, the intervention was more tailored to the individual and therefore more focused.
Apart from quality of included trials, quantity of studies is also a matter in meta-analysis. Compared to Koch et al. ( 2014 ), we obtained larger analysis clusters; however, especially for interpersonal skills, cognitive skills, and (psycho-)motor outcomes, more research is needed to obtain more meaningful results. The advantage of a bigger sample per outcome would be that more homogeneous clusters could be considered and addressed with sensitivity analysis. In addition, several included trials were conducted by the same research group (total: 7 of 41 trials by Koch et al., 2019 ), which is a threat to external validity. This is particularly relevant, when it comes to analysis of clusters (DMT group: 7 of 21 trials, clinical outcome cluster: 4 of 12 trials, interpersonal skills trials: 2 of 6 studies by the research group of Koch et al., 2019 ).
Finally, there is a general discussion on the issue of whether quantitative analyses are the appropriate means to evaluate the therapeutic use of dance. Borg ( 1993 ) stated that the dilemma of scientific (positivistic) research in behavioral sciences is that it applies traditional concepts of physical science to the study of living organisms, although living organisms are far more complex than physical objects of study. The authors imply that naturally dynamic, interdependent networks of factors involved in psychological phenomena are sometimes hard to detect when we apply concepts of causality, predictability, and scientific reducibility. This argument is underlined by a vivid discussion on the ecological validity of studies reducing complex processes, such as an aesthetic experience or the impact of art perception and production on health, into its single components (see e.g., Christensen and Jola, 2015 ). While quantitative methods are helpful for generating facts and explanations, qualitative methods might be more suitable for meaning-making and understanding of such (Berrol, 2000 ). In order to overcome the gaps between a non-linear reality and linear means of investigation, and to generate new scientific insights, quantitative research ought to be applied together with qualitative research, ideally in mixed-methods designs that reflect the epistemological background assumptions of the studied processes.
Hervey ( 2000 ) takes the discussion a step further by stating that artistic inquiry is needed to adequately reflect the results of such process-oriented domains as DMT. She introduces the concept of artistic inquiry, as part of arts-based research to DMT, which implies the use of the respective art form (dance) not only as an intervention to help the recipient but also as a form of data assessment, analysis, and presentation that aims at answering the research question. Leavy's textbook (2017) is the basic source for arts-based research today, providing a terminological and historical overview and best practice examples of arts-based research. Best practice examples of specific and particularly well-described arts-based research methods are Jola's embodied neuroscience (Jola, 2013 ), and Eberhard's aesthetic answering (Lange et al., in press ), both participatory approaches with the researcher in an active embodied role, diving into non-verbal processes to inform and answer the research question. However, at this point, final recommendations about arts-based research as a method are difficult, because the field is young and in a dynamic development (Leavy, 2017 ). Generally, non-verbal methods employing DMT or dance may detect changes in psychological outcomes that are not necessarily accessible with traditional methods and thus create innovative knowledge.
This meta-analysis shows that quantitative research on the therapeutic use of dance is augmenting. However, there is still an urgent quest for more trials with rigorous standards in respect to the chosen way of research (quantitative, qualitative, arts-based). Quantitative trials should consider larger sample sizes, randomized controlled designs, and active control groups, which compare DMT and dance interventions to other psychotherapeutic interventions with existing knowledge about therapeutic mechanisms (different types of psychodynamic therapies, cognitive behavioral therapy, pharmacological therapy). Furthermore, a detailed description of the intervention and its implementation is necessary (e.g., for replications). To avoid bias, there should be as much blinding as possible involved in the process. While complete blinding is not possible in therapy studies (and should thus not be part of the quality assessment of therapeutic trials to add this critique here), blinding of the randomization process and the accessor should be standard. Researchers ought to employ assessment tools that are least sensitive to expectation effects (e.g., standardized tests, psychophysiological measurement). All relevant treatment conditions should be reported in as much detail as possible (e.g., sample characteristics, characteristics of therapists, dropouts, structure and content of the treatment, other therapies provided, and interim circumstances, i.e., all “external” events that occurred in the time of the treatment, such as change of partner or job, etc.). Statistical results should be reported in detail including results that were not in line with the central hypotheses of the papers and descriptive statistics (means, standard deviations, and sample sizes in each group). To provide information on the impact of a treatment, long-term effects must be considered. Therefore, we highly recommend including follow-up assessment in intervention studies on the therapeutic use of DMT and dance.
In order to obtain more comparable research across the globe, there should be more communication between researchers, and they should strive for international standards. We recommend researchers conducting future meta-analyses to include more precise assessment of risk of bias, than was possible here. In addition, systematic analyses of moderators of effects should be performed in the future. Because we observed many dissimilarities between DMT and dance intervention studies, we recommend analysing those two types of studies separately in future studies.
Finally, to draw conclusions for practice, there is a need to complement quantitative research inquiry with qualitative and arts-based research (best in mixed-methods designs, reflecting the epistemological framework) and with clinical mechanism studies.
The urgent need for mechanism studies (Kazdin, 2007 ) and their interdependency with outcome studies shall here be exemplified with the debate around the included “NESS paper” (Priebe et al., 2016 ). Priebe et al. ( 2016 ) tested the decrease of negative symptoms in patients with schizophrenia after movement therapy (i.e., body psychotherapy, BPT, conducted by dance movement therapists; for definition of these terms, see Martin et al., 2016 ) in a randomized sample of N = 275 participants. They found that “the adjusted difference in negative symptoms was 0.03 (95% CI −1.11 to 1.17), indicating no benefit from body psychotherapy. Small improvements in expressive deficits and movement disorder symptoms were detected in favor of body psychotherapy. No other outcomes were significantly different.” With the interpretation of these findings, the authors question other relevant studies of the field, including their own earlier work (see Röhricht and Priebe, 2006 ; Lee et al., 2015 ; and Martin et al., 2016 —the latter a study from the same year, employing the same treatment manual (Röhricht and Papadopolous, unpublished). with a TAU control group, showing a significant reduction of negative affect after DMT (measured with the SANS). The NESS paper's data were mixed though, and further analyses in the secondary trial of Savill et al. ( 2017 ) have shown that the null effect discussed in the NESS paper was only true for men, not for women after the intervention.
Three arguments call for the necessity of a reappraisal of the NESS paper: the problematic domain overgeneralization from DMT/BPT to all arts therapies (domain and terminology-related aspect), the control group selection (DMT/BPT vs. Pilates), and the selective reporting and shortfall in conclusions, which do not appropriately reflect what is evidenced in the data (e.g., from the measures of negative symptoms, the PANSS showed no significant difference between groups, whereas the CAINS did; SANS, PANSS, and CAINS are all standardized observational measures to assess positive and negative symptoms of schizophrenia). In the context of the mechanism problem, we will merely discuss the second argument here.
Active control groups are recommended for most studies by the increasing standards of the evidence-based medicine. Priebe et al. ( 2016 ) thus tried to implement a suitable active control group. From the perspective of DMT though, with its present pronounced research on therapeutic mechanisms, Pilates is not a suited control group to DMT. Both interventions, DMT/BPT and Pilates, employ methods that are suited to increase body awareness. In Pilates, the torso is the focus of the work, the muscle tone is actively controlled and altered in specific regions of the torso, the muscles are strengthened and stretched, and the practice includes breath work for bringing the movements of the torso in resonance with the breath. Pilates had the goal of addressing the trinity of body, mind, and spirit in a holistic way (“Return to Life through Contrology,” Pilates and Miller, 1945 ). In DMT theory, the torso is the seat of the emotions, and the breath brings the emotions to the fore: on the basis of DMT core knowledge, the work with the torso and breath is the direct pathway to sensation, experience, and expression of emotions (e.g., Caldwell, 1996 ). Thus, the resulting null findings on the PANSS are not surprising. While we know very little about the mechanisms of BPT and DMT, we know even less about the mechanisms of Pilates and other body practices. Thus, it is very difficult, if not impossible, to select suitable active control groups for DMT studies, without any knowledge of either intervention's main mechanisms. This example shows that good outcome research needs scholarly mechanisms research (Hayes, 2013 ) and that there is a strong interdependency between these two types of clinical research.
Because mechanisms of DMT/BPT are not well-researched, and even less so mechanisms of Pilates, both the experimental group and the control group may have experienced similar working mechanisms, which may have caused the inconclusive results. As long as the major mechanisms of these therapies remain unclear, it is hard to draw any valid conclusions from the according outcome research. Thus, the effect of DMT/BPT on the reduction of negative symptoms needs to be investigated with a range of control groups. Primarily therapeutic mechanisms of DMT/BPT and potential control interventions, respectively, need to be further investigated, before conducting another primary study of the scope of the NESS paper. With this paragraph, we hope to have illustrated the urgent need for mechanisms studies due to their intricate interdependency with outcome research.
In conclusion, the results of our meta-analysis suggest that therapeutic use of dance potentially affects various health-related psychological outcomes. In total, there was a medium significant overall effect based on heterogeneous results. However, since type of intervention was a significant source of heterogeneity, we explored trials on DMT and trials on dance interventions in two separate groups. We found empirical evidence that DMT consistently and with a high homogeneity improved affect-related psychological conditions by decreasing anxiety and depression levels, and increased quality of life and cognitive skills. Concerning interpersonal skills, the effect reached the p < 0.1 level. More high-quality primary studies need to be conducted and included into meta-analyses to expand the evidence. Dance intervention studies consistently improved motor skills, while findings for the other outcomes had a high heterogeneity. Results of this meta-analysis suggest that DMT and dance interventions improve clinical outcomes, cognitive outcomes, and (psycho-)motor outcomes. The high variability of results, especially in the dance cluster, needs further attention. Moreover, this study contributes initial findings that DMT and dance interventions have persistent long-term effects. These encouraging results are limited by methodological shortcomings of the primary studies. Further research is needed that expands on the evidence of effects of DMT and dance interventions on health-related psychological outcomes.
SK conceived, planned, co-wrote and revised the study and supervised the Master's (RR, KT) and doctoral level students (LM, JB). LM and KT did the systematic literature search. LM, KT, and SK did the hand search. RR and AB planned and implemented the methodological approach, AB supervised the methodology of the study. KT, JB, and RR organized the results. RR analyzed the results and wrote the first draft. All authors contributed to the paper and revised it into its final version.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
We would like to thank Simea Schönenberger, Friederike Klodwig, and Katrin Sachau for helping to extract data from the primary studies as part of their DMT student elective hours. We would like to thank all colleagues who have supported us to find relevant literature, and all authors of primary studies who helped to clarify data issues. Thanks to all participants of the primary studies for completing the scales and driving research on embodied interventions further.
1 This term includes the practice of dance movement psychotherapy (UK terminology) and dance/movement therapy (US terminology).
2 For a list of abbreviations, see Appendix .
3 Creative Arts Therapies (European term: Arts Therapies; Karkou and Sanderson, 2006 ) consist mainly of art therapy, music therapy, dance movement therapy, drama therapy, and poetry therapy/expressive writing.
References marked with an asterisk indicate studies that are included in the meta-analysis; references marked with a circle were screened but excluded studies.
Funding. The study was conducted by the Research Institute for Creative Arts Therapies (RIArT) at Alanus University funded by the Software AG Foundation in Darmstadt, in colaboration with the University of Jena, Drexel University, and SRH University Heidelberg. The Faculty of Therapy Sciences at SRH University Heidelberg funded the publication.
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg.2019.01806/full#supplementary-material
Discover the world's research
Related papers.
Showing 1 through 3 of 0 Related Papers
31 Pages Posted: 3 Nov 2023 Last revised: 22 Jun 2024
Harvard business school, jonas hannane.
German Institute for Economic Research (DIW Berlin); Technische Universität Berlin (TU Berlin)
Imperial College Business School
Date Written: October 15, 2023
This paper studies the impact of Generative AI technologies on the demand for online freelancers using a large dataset from a leading global freelancing platform. We identify the types of jobs that are more affected by Generative AI and quantify the magnitude of the heterogeneous impact. Our findings indicate a 21% decrease in the number of job posts for automation-prone jobs related to writing and coding, compared to jobs requiring manual-intensive skills, within eight months after the introduction of Chat-GPT. We show that the reduction in the number of job posts increases competition among freelancers while the remaining automation-prone jobs are of greater complexity and offer higher pay. We also find that the introduction of Image-generating AI technologies led to a 17% decrease in the number of job posts related to image creation. We use Google Trends to show that the more pronounced decline in the demand for freelancers within automation-prone jobs correlates with their higher public awareness of ChatGPT's substitutability.
Keywords: Generative AI, large language models, ChatGPT, digital freelancing platforms JEL No: O33, E24, J21, J24
JEL Classification: O33, E24, J21, J24
Suggested Citation: Suggested Citation
Harvard University Harvard Business School, 25 Harvard Way Boston, MA Boston 02163 United States
Mohrenstraße 58 Berlin, 10117 Germany
Straße des 17 Juni 135 Berlin, 10623 Germany
Imperial college business school ( email ).
South Kensington Campus Exhibition Road London SW7 2AZ, SW7 2AZ United Kingdom 07778963203 (Phone)
Paper statistics, related ejournals, ebusiness & ecommerce ejournal.
Subscribe to this fee journal for more curated articles on this topic
International political economy: globalization ejournal, decision-making in computational design & technology ejournal.
A high-powered CEO puts her career and family on the line when she begins a torrid affair with her much younger intern. A high-powered CEO puts her career and family on the line when she begins a torrid affair with her much younger intern. A high-powered CEO puts her career and family on the line when she begins a torrid affair with her much younger intern.
Related news, contribute to this page.
Recently viewed.
Advertisement
By Leanne Abraham , June Kim , Elena Shao , Julie Walton Shaver , Anjali Singhvi , Christiaan Triebert , Karen Yourish , Lazaro Gamio and Amy Schoenfeld Walker
The Associated Press; Photograph by Doug Mills/The New York Times
Former President Donald J. Trump was whisked off the stage at his rally in Pennsylvania on Saturday after gunshots were fired toward the area where he was speaking. Mr. Trump could be seen bleeding from his right ear, and officials said that the shooting was being investigated as an assassination attempt.
Location of law
enforcement
snipers on roof
Location of gunman’s body
Location of a
wounded rally
Stage where
Trump was speaking
Butler Farm
Show Grounds
Pennsylvania
Location of
gunman’s body
By The New York Times; Aerial image by Doral Chenoweth/USA TODAY NETWORK, via Reuters
The rally took place on the grounds of the Butler Farm Show in western Pennsylvania, about 35 miles north of Pittsburgh. On Saturday afternoon, tens of thousands of rally attendees started trickling in after the doors opened at 1 p.m.
Mr. Trump was set to begin speaking at 5 p.m., but didn’t appear onstage until about an hour later. Here’s how the next 11 minutes unfolded based on footage of the rally.
6:05 p.m. As the song concludes, Mr. Trump approaches the lectern and begins speaking. He spends the first six minutes talking about President Biden and the state of the country before focusing on immigration.
Eric Lee/The New York Times
TMX via Associated Press
Seconds later A gunshot is heard, and Mr. Trump stops midsentence and flinches. He reaches for his right ear, as another two shots are quickly fired, and ducks behind the lectern. One male Secret Service agent is heard yelling, “Get down, get down, get down, get down!”
Photo appears to
capture the path
of a projectile
Blood on Trump’s
hand after he
touches his ear
Photo appears
to capture the
path of a projectile
Photos by Doug Mills/The New York Times
Secret Service agents surround Mr. Trump as a burst of five more shots is fired. Members of the crowd are panicking, screaming and crouching down. More security personnel run onto the stage, including several heavily armed law enforcement agents.
About 42 seconds after shooting began Agents stay crouched over Mr. Trump until an agent can be heard saying, “Shooter down.” The Secret Service confirmed later in a statement that its “personnel neutralized the shooter.”
Doug Mills/The New York Times
A spectator who had been standing just outside the grounds said in an interview with the BBC that a few minutes into Mr. Trump’s speech, he noticed that someone was “bear-crawling up the roof,” clearly armed with a rifle, and that he tried to notify the police. Law enforcement officials later said that the gunman had opened fire from an elevated position outside the rally’s security perimeter.
After the shooting, the gunman’s body was seen on the rooftop of one of the buildings to Mr. Trump’s right. An AR-15-style semiautomatic rifle was recovered at the scene, according to law enforcement officials.
“I was shot with a bullet that pierced the upper part of my right ear,” Mr. Trump said in a post on Truth Social, his social media platform. “I knew immediately that something was wrong in that I heard a whizzing sound, shots, and immediately felt the bullet ripping through the skin.”
Shortly after the shooting, a spokesman for Mr. Trump’s campaign said that the former president was “fine.” Mr. Trump will still attend the Republican National Convention this week, his advisers said in a statement.
One spectator at the rally, Corey Comperatore , a 50-year-old firefighter, was shot and killed. Two others were critically injured.
Photographs by Eric Lee and Doug Mills/The New York Times
IMAGES
VIDEO
COMMENTS
Line dancing has proven to be an effective physical activity for improving the state of depression in late second- and third-age dancers. ... Feature papers represent the most advanced research with significant potential for high impact in the field. A Feature Paper should be a substantial original Article that involves several techniques or ...
Line dancing is a popular form of exercise that can be modified, and is thus feasible for older adults with mobility limitations. Purpose: The purpose of this study was to assess the effects of 8 weeks of line dancing on balance, muscle strength, lower extremity function, endurance, gait speed, and perceived mobility limitations.
The 8 week line dancing intervention was a low impact dance program (i.e., one foot was always in contact with the floor), previ- ously considered appropriate for older adults with mobility limita-
Music stimulates the brain's reward centers, while dance activates its sensory and motor circuits. Studies using PET imaging have identified regions of the brain that contribute to dance learning and performance. These regions include the motor cortex, somatosensory cortex, basal ganglia, and cerebellum. The motor cortex is involved in the ...
University of Northern Colorado
Dance Research Journal is the longest running, peer reviewed journal in its field, and has become one of the foremost international outlets for dance research scholarship. The journal carries scholarly articles, book reviews, and a list of books and journals received. ... London: Routledge. 358 pp., 30 b/w illustrations. $37.56 paper and e-book ...
Li Shi Teachers Campus girls' 18-week line dance training is to test the change of physical indicators before and after learning.Based on that,the impact of line dancing exercise on physical health is analyzed to further investigate fitness value of line dancing exercise.The results show that line dancing improves student body shape,function and coordination.
An Empirical Study on Line Dancing to Promote the Construction of Community Sports Culture . Yu Yan . Xianyang Normal University Shaanxi Xianyang, 712000, China . [email protected] . ... and provides theoretical and methodological basis for the research of this paper. In order to make the content of the paper more effective, I interviewed 15 ...
Line dance is a dance,and also one kind of movement,even a kind of culture.Line dance can not only exercise,but also improve people's artistic accomplishment.In this paper,the classification,characteristics and style are summarized.From the musical beauty,body beauty and beauty of body action,exhibit beauty and beauty of dress and so on,the dance aesthetic value is analyzed.The result shows ...
However, in fact line dance can effectively strengthen college students' physical function, cultivate their music literacy and enhance their psychological quality. Therefore, this paper further strengthens college students' line dance skills through the study of these three elements. 2. Research Objects and Methods.
This paper takes the community sports of a city as the research object and the characteristics of line dance as the axis to discuss the reasons why line dance enters the community and the development countermeasures, so as to provide theoretical basis and practical reference for the sustainable and healthy development of line dance in community ...
For example, participants can select the intensity levels, dance routines/mode, dance music, even dance game partners. In a research study investigating 36 h of dance exergaming lasting 12 weeks, researchers observed a decrease in adiposity and an increase in bone mineral density compared to a non-exercising control group (Staiano et al., 2017b ...
In this paper, line dancing's moving object detection technology based on machine vision is studied to improve object detection. For this purpose, the improved frame difference for the ...
Background: Dance is an embodied activity and, when applied therapeutically, can have several specific and unspecific health benefits. In this meta-analysis, we evaluated the effectiveness of dance movement therapy 1 (DMT) and dance interventions for psychological health outcomes. Research in this area grew considerably from 1.3 detected studies/year in 1996-2012 to 6.8 detected studies/year ...
Dance Research is edited and published in Britain with the assistance of a distinguished group of editorial consultants based in Europe and the USA. The journal provides an international forum for the presentation and discussion of contemporary dance research and contains a section of comprehensive book and journal reviews. Journal information.
Events: The Society holds regular events for the presentation of the findings of dance research and for their discussion. These include the presentation of papers, lectures, demonstrations and research seminars. National Film Archives: Dance film and video from the BBC archives has been made available by a joint project funded by the Society.
Abstract. This chapter stages a conversation about the development of choreographic analysis as a methodology, since Susan Leigh Foster's 1986 book, Reading Dancing. Through her analyses, Foster ...
InResearching Dance, an introduction to research methods in dance addressed primarily to graduate students, the editors explore dance as evolutional, defining i...
In this paper, we propose a new attempt toward online learning, especially for dancing, which can provide timely and substantial feedback for dance novices to improve learning experiences. We design and develop DancingInside , an online dance learning system that supports dance learning through a 'practice, feedback, reflection, and ...
Dance Research, the journal of the Society for Dance Research, is a bi-annual internationally peer reviewed journal.It welcomes high quality original research articles on dance worldwide both historical and contemporary. The journal aims to engage with current debates on dance and across cognate disciplines with dance at the centre of inquiry.
Dance Research, the journal of the Society for Dance Research, is a bi-annual internationally peer reviewed journal.It welcomes high quality original research articles on dance worldwide both historical and contemporary. The journal aims to engage with current debates on dance and across cognate disciplines with dance at the centre of inquiry.
The implementation of the system effectively improves the efficiency of scoring in the line dance competitions and ensures the correctness of the final score. Combined with the actual needs of scoring in line dancing competition, an Android-based competition scoring system is designed. The system is mainly divided into Android terminal and PC terminal. The administrator can enter the main ...
David Leventhal, a former dancer with the Mark Morris Dance Group who is the program director and a founding teacher of Dance for PD, has been a co-author on several recent scientific papers about ...
RESEARCH PAPER Mahesh Dattani's Dance like a Man: A Study of Gender Discrimination *Dr. Vandana Bhartiya Ph.D. from Department of English & MEL, University of Allahabad, Allahabad ... The line shows the notions of the society that demands man should behave like a man. It shows that man cannot even expect to wear a dress of a woman.
Michael Harrigan, a retired F.B.I. special agent, said the image captured by Doug Mills, a New York Times photographer, seems to show a bullet streaking past former President Donald J. Trump.
This thesis, presented as a research paper, seeks to explore the preservation of two main categories: live coding performances and live mixed-media performances. The first chapter will focus on defining digital live performance art (DLPA) and establishing the parameters of this form of artwork. It will then introduce the two aforementioned
This paper studies the impact of Generative AI technologies on the demand for online freelancers using a large dataset from a leading global freelancing platform. We identify the types of jobs that are more affected by Generative AI and quantify the magnitude of the heterogeneous impact.
Across the world, critical businesses and services including airlines, hospitals, train networks and TV stations, were disrupted on Friday by a global tech outage affecting Microsoft users.
Babygirl: Directed by Halina Reijn. With Nicole Kidman, Harris Dickinson, Antonio Banderas, Jean Reno. A high-powered CEO puts her career and family on the line when she begins a torrid affair with her much younger intern.
Shots rang out as former President Donald J. Trump spoke at a campaign event in Butler, Pa., on Saturday. Maps, photos and video show how the shooting unfolded.