The role of affect and emotion in mental and physical health has long been a source of scientific scrutiny. Negative affect (NA) and positive affect (PA) are two broad dimensions of affect, which can be measured as a general trait or as a transient state factor (Watson and Pennebaker 1989). NA, which includes a range of uncomfortable mood states (e.g., anger, disgust, depression, anxiety, fear), reflects a broad dimension of subjective distress (Watson and Pennebaker 1989). Early research emphasized the relation of trait and state NA to poor cardiovascular health and mental health outcomes (DeSteno et al. 2013; Kubzansky and Kawachi 2000). Unlike the deleterious effects of NA on health, PA appears to reduce the risk of developing disease. PA is defined as the experience of pleasurable emotions (e.g., happiness, joy, excitement, enthusiasm, and contentment) (Watson and Pennebaker 1989). Evidence suggests that positive psychological well-being, a broad construct, which includes positive emotions and trait PA, is associated with a reduced risk of developing cardiovascular disease (CVD) and is protective against the progression of CVD (Boehm and Kubzansky 2012).

There are several potential mechanisms that account for the relation between trait PA and physical health. Fredrickson and Levenson (1998) described what they called, “the undoing effect” whereby positive emotions speed cardiovascular recovery from stress that ultimately leads to better cardiovascular health. In recent years the scope of this work has expanded to include effects of positive emotions on other indices of physiological recovery (Monfort 2012; Stroup 2015; Waugh et al. 2012; Yuan et al. 2010). The undoing effect has been extensively studied and includes multiple indices of physiological recovery that permits a comprehensive exploration of the effect of positive emotions on biological systems and processes. Further, a critical review of the undoing hypothesis is necessary to evaluate the consistency of the literature, factors that may affect the undoing hypothesis, and the evidence for the undoing hypothesis as a potential mechanism explaining the relation between positive emotions and health.

Method and Purpose

Search Strategy

A comprehensive review of the literature was conducted for published literature from 1998 to 2017 and included the databases, PsycInfo, Medline, and Google Scholar, using the following key terms: “undoing hypothesis,” “undoing effect,” and “physiological down-regulation.” In addition, reference lists from eligible studies were reviewed. The search strategy for this review excluded articles prior to 1998 as the research study which initially proposed the undoing hypothesis was published in 1998 (Fredrickson and Levenson 1998). Therefore, research examining physiological recovery prior to 1998 was excluded. The first author (CEC) conducted the literature review and identified articles meeting the inclusion criteria. All articles that met the inclusion criteria were included in this review. Studies meeting the following criteria were included: (1) an empirical study; (2) available in English; (3) included cardiovascular measures, (4) NA and PA were experimentally induced or measured. As the undoing hypothesis concerns physiological recovery, studies utilizing subjective measures as the sole outcomes were excluded. Table 1 provides a summary of studies included in this review.

Table 1 Summary of study characteristics
Full size table

Critique of the Literature

A review of the literature identified 24 studies that met inclusion criteria. Eight of these studies provided support for the undoing hypothesis (Fredrickson and Levenson 1998; Fredrickson et al. 2000; Tugade and Fredrickson 2004; Waugh et al. 2012; Yuan et al. 2010); eight studies demonstrated partial support for the undoing hypothesis (Dowd et al. 2010; Gilbert et al. 2016; Kraft and Pressman 2012; Medvedev et al. 2015; Ong and Allaire 2005; Papousek et al. 2010; Steptoe et al. 2007; Sokhadze 2007); eight studies failed to support the undoing hypothesis (Cavanagh 2016; Deiss 2012; Hannesdóttir 2007; Kaczmarek 2009; Monfort 2012; Purdum 2010; Soenke 2014; Stroup 2015). A critical review of this literature identified several areas of strengths and weaknesses, specifically regarding study design and conceptualization and analysis of physiological recovery. Additionally, we evaluated the clinical utility of the undoing effect. The major issues identified are discussed below.

Effects of Individual Characteristics

Previous research has demonstrated that specific individual characteristics (e.g., trait PA) are associated with better cardiovascular health (Boehm and Kubzansky 2012). Further, there is strong evidence that other individual characteristics, such as trait NA, are associated with worse cardiovascular outcomes (DeSteno et al. 2013; Kubzansky and Kawachi 2000). At a minimum, this research suggests the importance of statistically controlling for these factors. Yet with few exceptions (Ong and Allaire 2005; Papousek et al. 2010; Tugade and Fredrickson 2004), the reviewed studies are notable for a consistent failure to account for these characteristics. Interestingly, of the studies controlling for these effects only two studies (Tugade and Fredrickson 2004) supported the undoing effect, whereas the other two studies provided partial support (Ong and Allaire 2005; Papousek et al. 2009).

Other factors, such as resiliency, flourishing, and appraisal style have been hypothesized to be associated with the propensity to experience positive emotions (Tugade and Fredrickson 2004, 2007). However, few studies controlled for or investigated these factors as potential confounding variables (Hannesdóttir 2007; Kaczmarek 2009; Purdum 2010; Tugade and Fredrickson 2004) and of these studies only two studies supported the undoing effect (Tugade and Fredrickson 2004). It is possible that these individual characteristics exert a more powerful effect on physiological recovery than state PA, which may explain why several studies either found partial support or failed to support the undoing hypothesis altogether.

Effects of Stimuli

Research examining cardiovascular reactivity (CVR) has previously demonstrated that the effect on health outcomes is thought to involve the frequency, intensity, and duration of physiological reactivity to stressful circumstances (Brosschot and Thayer 2003; Goswami et al. 2010). Similarly, the effects of the undoing hypothesis are likely to be largely influenced by these factors. However, in this literature, the type, frequency, intensity, and duration of exposure to both the stressor and the positive emotion induction varied greatly across studies. The reviewed studies employed a range of different types of stressor tasks. Even when examining the studies that employed the two most frequently used stressor tasks, speech-preparation tasks and negatively valenced films, there was an inconsistent pattern of findings. Among the eight studies utilizing a speech-preparation stressor task only four studies demonstrated support for the undoing hypothesis (Fredrickson et al. 2000; Tugade and Fredrickson 2004), whereas the other four studies found partial support or failed to support the undoing hypothesis (Deiss 2012; Dowd et al. 2010; Hannesdóttir 2007; Monfort 2012).

This inconsistency is also apparent when comparing the type of tasks used to induce positive emotions. For example, several studies employed positive films, images, or music to induce positive emotions (Deiss 2012; Hannesdóttir 2007; Kaczmarek 2009; Medvedev et al. 2015; Monfort 2012; Sokhadze 2007); however, the undoing hypothesis was only supported in three of these studies (Fredrickson and Levenson 1998, Study 1; Fredrickson et al. 2000). In contrast, other studies induced positive emotions through more active tasks such as the manipulation of facial muscles to evoke smiling, dyadic conversations, or recalling of positive experiences (Cavanagh 2016; Hannesdóttir 2007; Kraft and Pressman 2012; Purdum 2010; Yuan et al. 2010). Only one of these studies demonstrated full support for the undoing hypothesis (Yuan et al. 2010). Still, this inconsistency persists even among studies that did not experimentally induce positive emotions. Of the eight studies (Dowd et al. 2010; Ong and Allaire 2005; Papousek et al. 2010; Steptoe et al. 2007) that did not experimentally induce positive emotions, the undoing effect was only supported in half (Fredrickson and Levenson 1998, Study 2; Tugade and Fredrickson 2004; Waugh et al. 2012). This may indicate that the ability of PA to quicken physiological recovery requires that the stressor and positive emotion to be relatively equivalent in terms of intensity (Kaczmarek 2009).

Research has found differences in physiological responses based upon the type of positive emotion induced. For example, previous research has found an increase in sympathetic responses related to positive emotions and proposed that this response may reflect a more arousing emotion such as excitement (Cavanagh 2016; Hannesdóttir 2007; Yogo et al. 1995). Therefore, it is possible that the discrepant findings in this review can be accounted for by the different types of stressor and positive emotion induction tasks employed, which suggests that the undoing effect may be limited to very specific circumstances.

Effects of Stressor Exposure

Research has demonstrated that physiological response patterns can differ depending on the type of stressor task utilized. In the reviewed literature, the stressor tasks can be categorized as active- and passive-coping tasks. In active-coping tasks, those in which participants’ actions or performance affects the outcome (e.g., mirror tracing task), beta sympathetic nervous system reactions are typically observed (Andreassi 2007). In contrast, in passive-coping tasks, those in which task outcomes are not influenced by participants (e.g., viewing films), alpha-adrenergic stimulation is observed (Andreassi 2007). Of the studies included in this review, more studies used stressor tasks requiring active-coping (Cavanagh 2016; Deiss 2012; Dowd et al. 2010; Fredrickson and Levenson 1998; Gilbert et al. 2016; Hannesdóttir 2007; Kaczmarek 2009; Medvedev et al. 2015; Monfort 2012; Papousek et al. 2010; Steptoe et al. 2007; Stroup 2015; Waugh et al. 2012) as compared to passive-coping stressor tasks (Fredrickson and Levenson 1998; Ong and Allaire 2005; Purdum 2010; Sokhadze 2007; Soenke 2014; Yuan et al. 2010). Additionally, several studies used both active- and passive-coping stressor tasks (Kraft and Pressman 2012; Tugade and Fredrickson 2004). Of the studies that supported the undoing effect three studies used active-coping tasks (Fredrickson et al. 2000; Waugh et al. 2012), three used passive-coping tasks (Fredrickson and Levenson 1998; Yuan et al. 2010), and two used both active- and passive-coping tasks (Tugade and Fredrickson 2004). Given that there was no discernible pattern to the effects from the types of stressor tasks, it is difficult to evaluate if the effects on recovery differed as the result of the use of active- vs. passive-coping tasks.

Operationalization and Analysis of Recovery

Recovery was operationalized and analyzed in multiple ways in the reviewed literature. For example, 7 of 24 studies utilized the same operationalization of recovery, defined as the time it took for participants to return to baseline cardiovascular levels and remain within those levels for 5 of 6 consecutive seconds (Fredrickson and Levenson 1998; Fredrickson et al. 2000; Hannesdóttir 2007; Tugade and Fredrickson 2004). Studies utilizing this time-based index of cardiovascular recovery examined recovery as an aggregate of several cardiovascular measures and interestingly 6 of the 7 studies that used this method found evidence supporting the undoing hypothesis (Fredrickson and Levenson 1998; Fredrickson et al. 2000; Tugade and Fredrickson 2004). Of the remaining 17 studies, which employed a variety of operationalizations of recovery, two demonstrated support for the undoing hypothesis (Waugh et al. 2012; Yuan et al. 2010), whereas 15 studies either demonstrated partial support or found no evidence supporting the undoing hypothesis (Deiss 2012; Dowd et al. 2010; Gilbert et al. 2016; Hannesdóttir 2007; Kaczmarek 2009; Kraft and Pressman 2012; Medvedev et al. 2015; Monfort 2012; Ong and Allaire 2005; Papousek et al. 2010; Purdum 2010; Soenke 2014; Stroup 2015; Sokhadze 2007; Steptoe et al. 2007). Further, only a couple of studies considered how the type of physiological measures utilized may affect conceptualization of recovery. For example, three studies (Monfort 2012; Medvedev et al. 2015; Yuan et al. 2010) included SCL as a physiological measure. Yet only two of these studies examined recovery as the percentage of return to baseline (Monfort 2012; Medvedev et al. 2015). As the rate of recovery differs by physiological measures, such as SCL, examining recovery as a percentage of return to baseline may help to control for these differences (Linden et al. 1997). By comparing the different operationalizations of recovery, the pattern of results suggests that the undoing hypothesis may be strongly related to how recovery is conceptualized and potentially suggests that the undoing hypothesis may be an artifact of this conceptualization and measurement.

Clinical Utility

Considering the relative paucity of research examining physiological recovery as a time-based variable, it is difficult to evaluate the clinical significance of a 30 s recovery from a stressor. Further, there is evidence that time courses of recovery differ by the type of stressor and that recovery to stressors in the lab are unlikely to generalize to naturalistic settings (Goswami et al. 2010; Linden et al. 1997). In one meta-analysis examining cardiovascular recovery from stress, the average recovery period was 13.4 min with a range of 30 s to 4 h (Schuler and O’Brien 1997). Other researchers have noted that participants typically return to baseline physiological levels within 4 min following mental arithmetic and non-verbal stressor tasks (Roskies and Hanley 1986). Critically, these recovery periods reflect natural recovery (e.g., no induction of PA or other techniques to improve recovery).

Stressors were presented for a relatively brief period (e.g., approximately 60–80 s) in several studies (Fredrickson and Levenson 1998, Study 1; Fredrickson et al. 2000). Considering the brief stressor period, the magnitude of CVR was likely relatively small and if the magnitude of CVR was relatively small, then a 30 s recovery time is unlikely to reflect a difference that is clinically meaningful, particularly as a protective factor against CVD. Individuals who demonstrate faster cardiovascular recovery, consistently, following exposure to stressors may develop a protection against the development of CVD. However, considering the research reviewed here, the clinical significance of faster physiological recovery is largely unknown.

Sample Characteristics

The reviewed literature is also notable for a homogeneity of the samples studied. Of the 24 studies reviewed, 20 relied upon college-aged and young adult samples (Cavanagh 2016; Deiss 2012; Dowd et al. 2010; Fredrickson and Levenson 1998; Fredrickson et al. 2000; Gilbert et al. 2016; Hannesdóttir 2007; Kaczmarek 2009; Kraft and Pressman 2012; Medvedev et al. 2015; Monfort 2012; Papousek et al. 2010; Purdum 2010; Soenke 2014; Sokhadze 2007; Stroup 2015; Tugade and Fredrickson 2004). Only 6 of these 20 studies demonstrated support for the undoing hypothesis (Fredrickson and Levenson 1998; Fredrickson et al. 2000; Tugade and Fredrickson 2004). Other studies included middle-aged and/or older adult samples (Ong and Allaire 2005; Steptoe et al. 2007; Yuan et al. 2010) and one studied included a sample of adolescents (Waugh et al. 2012). Two of these studies demonstrated support for the undoing hypothesis (Waugh et al. 2012; Yuan et al. 2010) and two demonstrated partial support (Ong and Allaire 2005; Steptoe et al. 2007). This pattern of findings suggests that a lack of support for the undoing hypothesis was consistent across ages.

Furthermore, while gender representation was seen in 14 of 24 studies reviewed (Deiss 2012; Fredrickson and Levenson 1998, Study 2; Fredrickson et al. 2000; Hannesdóttir 2007; Kaczmarek 2009; Kraft and Pressman 2012; Medevdev et al. 2015; Monfort 2012; Ong and Allaire 2005; Papousek et al. 2010; Soenke 2014; Stroup 2015; Tugade and Fredrickson 2004; Yuan et al. 2010), a majority of these studies failed to examine sex differences in recovery. Examination of sex differences is critical to consider as two studies found that men recovered from stress significantly faster as compared to women (Fredrickson and Levenson 1998, Study 2; Tugade and Fredrickson 2004, Study 2). Additionally, other research has found sex differences in CVR, which suggests that sex should be either experimentally or statistically controlled for when examining cardiovascular responses (Stoney et al. 1992; Stroud et al. 2001).

Finally, in regard to sample characteristics evidence of ethnic and racial diversity in study samples was limited with only two studies including a sample composed of 50% ethnic or racially diverse participants (Fredrickson and Levenson 1998, Study 1; Gilbert et al. 2016). Additionally, only one study examined ethnic differences in cardiovascular recovery (Fredrickson et al. 2000). Thus, sampling methods raise questions concerning generalizability to the broader adult population.

Summary and Future Directions

Summary

In this critical review we identified 24 studies which examined the undoing effect, and of these studies, 8 studies supported the undoing hypothesis, 8 studies provided partial support, and 8 studies found no evidence supporting the undoing hypothesis. As discussed, the design and conceptual issues identified in this review limit the conclusions one can draw regarding the undoing hypothesis at this time. The pattern of results clearly demonstrates inconsistency and therefore, there is insufficient evidence to support the undoing effect at this time. Further, there is insufficient evidence to believe that positive emotions affect health by enhancing cardiovascular recovery from stress. Although there is insufficient evidence to claim positive emotions improve cardiovascular recovery at this time, future research investigating the undoing effect is important for understanding the mechanisms through which positive emotions affect health.

Future Directions

There are several areas which future research should address. Specifically, future research should endeavor to apply standardized conceptualizations and measurement of physiological recovery and examine potential moderators of the undoing effect to determine under what conditions and contexts the effect can be reliably produced.

Two of the primary concerns that future studies should address is how recovery is operationalized. Researchers must define recovery consistently across the literature to evaluate the undoing effect more thoroughly. Conducting recovery research may require more sophisticated measurement methods and more advanced statistical techniques to examine recovery across multiple physiological parameters, including amount, speed, and level of recovery (Christenfeld et al. 2000; Linden et al. 1997; Luecken and Gallo 2008).

An additional concern is the type of physiological measures employed. Recovery studies typically use a short, prespecified time as the recovery period (e.g., 5 min). Therefore, physiological measures that require longer recovery periods (e.g., endocrine and immune measures) are less likely to be measured. Cardiovascular measures (sympathetic and parasympathetic indices) are better suited for recovery studies; however, these measures provide an extremely narrow view of physiological recovery.

In addition, future research should examine potential moderators of the undoing hypothesis. As discussed, a number of individual characteristics, such as trait PA, resiliency, flourishing, and appraisal style, may affect the relation between positive emotions and physiological recovery. These individual characteristics must be controlled for in the design of studies or at a minimum statistically controlled for in analyses.

Addressing the issues highlighted in this review will enable researchers to evaluate the causal relation between state PA and physiological recovery, the clinical utility of the undoing effect, and will allow researchers to evaluate the undoing effect as a potential mechanism through which PA affects physical health.