Using Gratitude to Promote Positive Change: A Series of Meta-Analyses Investigating the Effectiveness of Gratitude Interventions

ABSTRACT

Gratitude interventions have been proposed as beneficial practices for improving myriad positive outcomes, and are promoted in self-help literature. The current work examined gratitude interventions’ effects with meta-analytic techniques to synthesize findings of thirty-eight studies, totaling 282 effect sizes. Fifty-six separate meta-analyses examined outcome effects for: gratitude versus neutral comparison at postintervention and delayed follow-up; gratitude versus negative comparison at post and follow-up; and gratitude versus positive comparison at post and follow-up. Results show that gratitude interventions can lead to improvements for numerous outcomes, including happiness, but do not influence others. Their unique benefits may be overemphasized in the literature.

Acknowledgments

Thank you to Judith Hall for providing feedback and guidance on this project, and to Annie Keyes and Selime Salim for assisting with coding.

Notes

To further focus their comparisons for psychological well-being, under the umbrella term of alternative-activity, they considered both matched-activities and psychologically active conditions. Matched-activities had participants engage in a matched listing activity, paralleling the gratitude condition, but where they listed either neutral events (describing what they did that day), negative events (describing what hassles they dealt with that day), or positive events (describing kind acts they had performed). Psychologically active conditions had participants do an activity that presumably could increase well-being, such as the actual practice of acts of kindness or meditating. With this designation, this would still result in ambiguity between neutral, negative, and positive conditions.

Davis and colleagues left out numerous studies from their analyses without explanation, despite mentioning them in their introduction (Flinchbaugh, Moore, Chang, & May, 2012; Owens & Patterson, 2013; Sheldon & Lyubomirsky, 2006), and as a more minor note, they included the same sample as two independent samples (Dickerhoof, 2007; Lyubomirsky, Dickerhoof, Boehm, & Sheldon, 2011).

The Davis et al. meta-analysis, in contrast, included the studies by Gilek (2010) and Ozimkowski (2007).

Analyses using composite affect measures (average levels of affect over the duration of the entire intervention) were included if no other data were supplied (e.g., Emmons & McCullough, 2003; Harbaugh & Vasey, 2014).

However, in Lyubomirsky et al. (2011), the reported data pooled both those who had self-selected into a study about happiness and those who were not given rationale; therefore, both types of participants were left in.

Some studies clarified how much time per week was spent on the practice; others did not. Therefore, I used the overall duration of the intervention as a proxy for time investment. Future work would do well to clearly outline the time/effort expended.

At times, this was a straightforward well-being measure such as Ryff’s Psychological Well-Being Scale (Ryff, 1995) or the Warwick-Edinburgh Mental Well-Being Scale (Tennant et al., 2007). If researchers provided well-being composites (generally consisting of satisfaction with life, positive affect, and reverse-coded negative affect), that composite was used. If the individual elements could be broken apart, the study was not included in the well-being meta-analysis, but rather in the appropriate, more specific meta-analyses.

Some studies used measures that combined positive and negative affect into one score and could not be separated (Killen & Macaskill, 2014). These were left out of the current analyses.

Following Emmons and McCullough’s (2003) variable, only the amount of sleep was analyzed, to maintain consistency.

This variable included measures such as Child–Parent Relationship Scale (Pianta, 1995) and the Teacher–Student Relationship Inventory (Ang, 2005), as well as brief questionnaires similar to those found in Emmons and McCullough (2003).

Important to note, the proportion of imputed zeros did not differ dramatically across the three types of comparison groups and did not bias findings against the positive intervention condition. The highest proportion of imputed zeros (.25) actually occurred in the negative comparison condition (positive: .22, neutral: .10).