The two analyses presented above primarily evaluated the impact that relaxation training has on anxiety in general. Certainly, there is a methodological difference between the two types of data. In the between groups analysis, effect sizes are computed from the difference between experimental and control group. Thus, it is possible to distinguish the effects produced by the relaxation treatment from those caused by the simple passing of time. This is not possible when the evaluation of the treatment depend on the differences between the score measured before and after the training. Any change observed would depend partially from treatment, from the simple passing of time and from others uncontrolled variables.
In any case, both meta-analyses indicate a good efficacy of relaxation training in the reduction of anxiety, both in comparison with a control group and with the participants as controls for themselves.
This result is aligned with the research literature [33
] and with relaxation manuals indications [43
There is a great heterogeneity of effect sizes. In order to reduce this variability, some distinctions have been made. All the relaxation techniques considered show a good potential in the reduction of anxiety. Applied relaxation and meditation have very high effect scores both in within and between analyses. However, in latter analysis, applied relaxation is used only in one study, making this result not valid. Progressive relaxation produced high effect sizes, with a within group reduction superior to the other techniques. The decrease in anxiety obtained with autogenic training is a little lower (but still positive) than other techniques in the between groups comparison, but its within group effect size is aligned with the general average. A multi-techniques approach does not increase relaxation training efficacy on anxiety reduction, showing an effect size level relatively low in both analysis. Non codified techniques, alias "other techniques", represent the category with the lowest score, especially in the within group analysis.
The selection of the best relaxation technique is quite hard. The high effect size levels reached by meditation, applied relaxation and progressive relaxation may indicate a good efficacy in the reduction of anxiety from all of them. An indication that seems to rise from those data is to apply just one model, avoiding the use of more techniques together.
There is a difference between the two analyses concerning the typology of participants. In between groups analysis, volunteers and students have an higher reduction of anxiety. Within group analysis indicates a good efficacy for this category, despite a lower score than the former one. Patients with psychological or psychosomatic diseases present different results between the two analyses. In controlled studies, the average effect size is medium-low, while open trials without control group indicate a really higher effect size. Globally, at baseline, participants with psychological or psychosomatic diseases show higher anxiety levels in comparison with the other and this can explain greater differences between pre and post assessment in within group studies. Control groups of studies with psychological diseases may help to understand the data of the between group analysis. In fact, there seems to be a waiting list effect [46
], because people often improve just by being in a waiting list. Moreover, some people could have been under an unknown treatment (psychotherapy, pharmacological...) leading to an uncontrolled anxiety decreasing. Differently, people without a particular disease (students or volunteers) present a stable level of anxiety along time and treatment effect is "pure", because not related to an expected "physiological" decreasing of anxiety from higher levels in clinical samples.
An opposite correlation between effect size and percentage of women emerged between the two meta-analyses. This correlation is negative in the between group analysis, while it's positive in the within group one. This result is hard to explain. Maybe it is related to an heterogeneous percentage of women in the different groups of subjects. For example, in the samples with psychological or psychosomatic problems there is a significantly greater presence of women. So, this result may depend mostly on samples composition, and must be taken with caution. Further research is needed.
Patients with medical problems presented the lower effect size, both in within group and between group analyses, with medium-low efficacy. However, for this patients, the objective of relaxation is not the reduction of anxiety. More often relaxation techniques are used to reduce perceived pain or somatic symptoms (i.e. nausea, hypertension).
There is a negative correlation also between the efficacy of relaxation training and the mean age in both meta-analyses. Older people have less benefits than younger. Older people may have also more difficult in the practice of physical exercises (i.e. in the Jacobson's progressive relaxation training) or, maybe, a lesser understanding of instructions.
The context of application doesn't moderate the reduction of anxiety, in contrast to what found by Carlson and Hoyle [28
], who indicate individual treatment as more effective (but their analysis was about the effects of progressive relaxation on various pathology).
The efficacy of the treatment increases with the duration of the protocol, in both meta-analyses. Repetitive training over a long period product significantly higher modification. Maybe there is an expected correlation between the amount of time spent in practicing exercises and their efficacy. In fact, effect size increases significantly with the request of practicing the exercises at home, consistent with past findings [47
Finally, concerning the anxiety questionnaires, studies that used the Trait scale of the STAI show a lower within group effect size compared to those that used the State scale. This result is coherent with the different theoretical constructs measured by the STAI: changing a trait is harder than changing a state. In the between group meta-analysis, there was only one study that used the Trait scale, so no generalization can be done.
The State scale of the STAI is the most used instrument in the present sample of papers and the effect sizes are similar between the two meta-analysis.
The anxiety scale of HADS showed a low between groups effect size and a medium within group effect size. The interpretation of this data is quite complex, because, against the less discriminant result in the intergroup analysis, international literature demonstrated good psychometric properties of the instrument [48
]. A possible explanation deal with the main target of the scale, that assessed usually hospital patients, with a severe physical problem (i.e. cancer).
The higher effect sizes come from the studies that used the BAI, but these are too few in order to make a generalization (one in the between analysis, two in the within one).