The Psychological Effects of Strength Exercises in People who are Overweight or Obese: A Systematic Review

For the literature review, no restrictions were made regarding year of publication, language of the manuscript (although all manuscripts found were in English), or design of the study. Because of the expected limited number of studies in this specific area, three criteria were originally used for inclusion. Initially, we aimed to develop a search strategy to locate all studies in (1) people (all ages and both sexes) who are overweight or obese that reported the effect of (2) strength exercises on (3) at least one psychological construct. However, for the literature search, this last criterion turned out to be not feasible, because the psychological outcomes were too varied depending on the underlying theoretical concept. Given our aim of identifying any effects that strength exercises may have on psychological outcomes, it proved impossible to capture this last criterion in query terms without running a considerable risk of excluding potentially relevant literature. Therefore, we used the first two criteria and then selected papers that mentioned any psychological concepts, first based on title and abstract and later on full text (see Table 1). Only studies that reported the independent effect of strength training on psychological outcomes in overweight or obese people were included. No other restrictions were applied.

Relevant literature was identified using the PubMed and PsycINFO databases (first data search on 19 May 2014). In the first screening round (N = 7860) two screeners (GH and GK) identified 14 papers that met the eligibility criteria (see Fig. 1 for a flowchart showing the literature search progress). In a second and third screening round a total of three additional studies were found. The final number of included studies was 17. In the Supplementary Material at https://​osf.​io/​8jbaz/​ (Open Science Framework), a detailed list of all initial paper titles and abstracts can be found, including why papers were systematically excluded, together with the PRISMA checklist [20].

To acknowledge study quality and simultaneously take into account the intervention administered in the control group [21], we divided studies into five categories (see also Table 2). Studies in category I (a no-intervention control group compared to strength training) can answer the question of whether strength training has an effect on psychological outcomes. Studies in category II (an active control group vs. the same active control group plus strength training) can answer the question of whether strength training has added value over and above the active control group intervention. Studies in category III (an active control group vs. strength training) can answer the question of how strength training performs compared to the active control group intervention (e.g., diet or aerobic training). Category IV (an active control group (e.g., aerobic plus diet) versus strength training plus another active component (e.g., diet) can answer the question of how strength training performs compared to a given active component, when both are combined with another active component. Category V (studies lacking a control group, i.e., pretest–post-test designs) can provide very weak evidence for an effect of strength training over time, and was mainly included for the sake of completeness. To assess study quality, an additional risk of bias assessment was performed using the Effective Public Health Practice Project Quality Assessment Tool [22]; see also the Supplementary Material).

There was great variation in the psychological terminology used in the included studies. To establish which constructs could be aggregated, GtH extracted the variables and their operationalization from the included papers.

To determine which psychological outcome measures could be aggregated, two authors (GK and GJP) indicated which construct they thought was being measured. To establish this, they consulted the papers’ methodology sections where necessary. After this coding phase, two discussion rounds were conducted. In the first, both coders, facilitated by a third (GtH), discussed the terminology used where there were minor deviations [e.g., one author used “mood (inverted)” and the other “negative mood,” but the same variables were coded (90% consensus)]. In the second discussion round, more fundamental differences were discussed and resolved. After consensus was achieved, the psychological outcomes were ordered into eight broad categories: disorders (e.g., anxiety and depression), inhibition, mood, outcome expectations, quality of life, self-efficacy, self-esteem, and stress (see the Supplementary Material for the coding sheets). The resulting spreadsheet was then imported into R [23] for further analysis using metafor [24].

For each study, effect sizes as well as corresponding variance estimates were extracted or calculated for the main effects of strength exercises on strength (as strength interventions are often focused on improvements in strength) and on psychological outcomes. Most studies used split-plot designs where within-subjects pre- and post-tests were combined with a between-subjects manipulation (see also Table 3). In such cases, computation of the effect sizes’ variance estimates requires the correlation between pre- and post-test measures [25], which was not reported by any of the papers. We therefore computed three types of variance estimates, assuming correlations of 0.3, 0.5, and 0.7 (corresponding to the qualitative labels for effect size as tentatively suggested by Cohen [26]). All analyses were therefore conducted three times. The results for the correlation estimate of 0.3 are reported, supplementing these reports with discussion of diverging outcomes where these occur.

Where studies reported multiple effect size estimates for variables that were coded as the same variable [e.g., “disinhibition” and “hunger” in Messier et al. [27] were both coded as “inhibition (inverted)”], these were first aggregated to obtain one estimate per variable per study. For these intra-study meta-analyses as well as the final between-study meta-analyses, random effects meta-analyses were conducted using the metafor package’s restricted maximum-likelihood estimator [24]. Heterogeneity was estimated using τ ² (estimated between-study variance), I ² (the proportion of variability in effect sizes due to heterogeneity rather than error), H ² (total variability compared to sampling variability) and Q (the χ ² test for heterogeneity), and forest and funnel plots were generated for each meta-analysis and are available in the Supplementary Material.

We identified “positive effects” of strength exercises in people who are overweight or obese as occurring when psychological constructs changed in the desired direction (e.g., increase in self-efficacy or decrease in psychological distress).

In total, 17 studies were included in the systematic review (Fig. 1). Based on our risk of bias assessment, the study quality of 13 papers was rated as “moderate” and four papers were rated as “weak” (see the Supplementary Material). Study characteristics are listed in Table 3. The number of participants in the different comparisons ranged from 32 [28] to 304 [29], with one extreme of 10,386 participants [30]. The intervention period ranged from an acute session of strength exercises [31] to 48 weeks’ training [32]. Seven studies included a comparison between strength training and a no-intervention control group (category I; see also Table 2 for examples). Eight studies included comparisons between an active control group (e.g., diet) and the same control group plus strength training (category II). Three studies compared strength training to aerobic training (i.e., an active control group—category III). One study compared strength training plus diet to aerobic training plus diet (category IV). Finally, three studies employed a pretest–post-test design (category V). Thirteen studies were in adults. All studies included a specific group of people who were overweight or obese (see Table 3).

The 17 included studies had many different psychological outcomes. These are summarized in Table 4.

Based on the available data, for two studies [32, 34] no effect sizes could be calculated, and, therefore, these were not included in the meta-analysis. One additional study [36] was excluded for meta-analysis, as this study examined the acute effects of one strength exercise session. For all other studies effect sizes were calculated based on pre- and post-test means, standard deviations (SDs) and n values in both the strength-exercise group and the comparison group. In one study [35] effect sizes were available. Study outcomes were divided into the five major study types and eight major outcome categories. All individual effect sizes and forest and funnel plots can be found in the Supplemental Material. Note that although the literature contained reports of the effect of strength training on eight different psychological variables, few studies were available for each variable; and as the various studies provided data to answer different research questions, few studies were available for meta-analysis. This small number of studies for meta-analysis made heterogeneity hard to assess. Effect sizes seemed quite consistently heterogeneous for the exercises’ effects on strength (see Supplementary Material). Heterogeneity varied from 0–100%, with p values from <0.001 to 1 (see also the Supplemental Material).

The current state of the literature means that it is unclear how results from the meta-analyses should be interpreted. Therefore, the outcomes will be discussed qualitatively. We have, however, used the meta-analysis to generate diamond plots to aid interpretation of the current evidence base.

The diamond plots show that all effects are weak, but most of them are in a positive direction (i.e., strength training has a possible positive influence on psychological outcomes). Some weak effects emerged on self-efficacy, self-esteem, and psychological disorders (e.g., anxiety and depression), but only compared to a no-intervention control group [first diamond plot (category I)]. The second diamond plot (category II) shows that strength exercises have possible favourable additional effects on psychological disorders, self-esteem, and inhibition when combined with another active component, but that they are weak and have no additional effects on stress, self-efficacy, quality of life, or outcome expectations. In the third diamond plot, strength exercises were compared with other interventions [e.g., diet or aerobic exercises (category III)], showing that strength has possible positive effects on self-esteem but no stronger effects than diet or aerobic interventions on psychological disorders, quality of life, or mood. For the fourth study type [an active control group (e.g., aerobic plus diet) vs. strength training plus another active component (i.e., diet)], no data were available [32]. For the fifth study type (pre–post-test design without a control group), positive time-effects for strength training were found for perceived well-being [31], health and life satisfaction [30], and behavioral expectation, self-regulation, and perceived satisfaction [35]. The study examining the acute effects of strength exercises showed some positive effects on well-being, but the results were inconclusive [36]. Subclassification by age (i.e., under 18 years and over 18 years) showed no clear differences in results (see the Supplemental Material).