Effects of Intensive Mobile Happiness Reporting in Daily Life

Method and Procedure

Participants were 162 university students (41% men), 17–30 years old (M = 19.9, SD = 2.4), who self-identified as Caucasian (81%), Asian (10%), Indian (4%), Māori or Pacific Islander (3%), or another ethnicity (3%). ¹ Participants were recruited through the Psychology Department’s experimental participation program and remunerated with partial course credit and a mobile phone voucher to cover the costs of texting.

In an initial laboratory session, participants individually completed informed consent and computerized measures of demographics (gender, age, and ethnicity), current depressive symptoms: 20-item Center for Epidemiologic Studies Depression scale ([CES-D] Radloff, 1977; α = .89), neuroticism: 12-item scale from the NEO-Five-Factor Inventory ([NEUR] Costa & McCrae, 1985; α = .86), trait happiness: 4-item Subjective Happiness scale ([SHS] Lyubomirsky & Lepper, 1999; α = .85) and self-esteem: 10-item Rosenberg Self-Esteem scale ([RSES] Rosenberg, 1965; α = .88). They also reported the number of text messages they received typically per day as a measure of texting experience.

Next, they received training for the texting procedure, which began the next day. Participants received 1, 3, or 6 text messages each day for 13 days based on their randomly assigned condition. Each text contained three questions to tap the hedonic and cognitive aspects of happiness. The questions were presented in a format suitable for text messaging: “RIGHT NOW: How happy do U feel? R U enjoying what U R doing? How positive do U feel about life? 1 = not at all … 9 = extremely.” Participants responded by sending three numeric answers in a single reply text (e.g., “536”), which were averaged for analysis (α = .66; Nezlek, 2012). The three questions were always presented in the same order. Texts were delivered to each participant’s personal mobile phone semi-randomly 1, 3, or 6 times daily between 9 a.m. and 9 p.m. on weekdays and between 12 p.m. and 10 p.m. on weekends using www.message-media.com. Participants were told to reply to the texts “as soon as they could” after receiving them. After 13 days of texting, participants returned to the laboratory the next day for debriefing and remuneration.

Results

Main Effects of Mobile Happiness Reporting

Figure 1 shows the changes in momentary happiness over time for each of the three happiness reporting groups. A visual examination suggested no group differences in changes in happiness. The box plot in the inset of Figure 1 also suggested no group differences in the linear trends over time. Moreover, inferential tests using a series of multilevel models confirmed no main effect of linear time, G10 b(SE) = .006(.006), p = .36; no main effect of group, coded 0, 1, 2; G01 b(SE) = .083(.098), p = .40; and no critical Time × Group interaction, G11 b(SE) = −.008(.008), p = .31; HLM v 6.08; Raudenbush, Bryk, & Congdon, 2004. Follow-up tests using dummy coding with 1 text per day as the reference group also revealed no significant group differences in changes in momentary happiness over time between the 1 versus 3 texts/day group, G11 b(SE) = .005(.016), p = .78, or between the 1 versus 6 texts/day group, G12 b(SE) = −.016(.016), p = .33. The effects remained the same when excluding the three outliers identified in the Figure 1 box plot. Effects also remained nonsignificant when analyzing the three happiness items separately. These null effects occurred despite the design being highly powered (.94) to detect a modest linear trend difference of b = .041 happiness points per day and adequately powered (.80) to detect an even smaller linear trend difference of b = .033 points per day (Optimal Design; Raudenbush, Spybrook, Congdon, Liu, & Martinez, 2009). Thus, the decision to retain the null hypothesis is not likely due to the design of the study.

OPEN IN VIEWER

Figure 1.

Changes in daily averaged momentary happiness ratings for each experimental group. The inset graph shows the box plots of the linear trends for individuals in the three experimental groups. The horizontal lines represent the medians; the boxes represent the interquartile range (IQR) between the first and third quartiles reflecting 50% of cases; the vertical whisker lines represent upper and lower acceptable range of trend values (± 1.5 × IQR). The three cases outside these whisker lines indicate outliers. Graph formatting based on recommendations by Lane and Sándor (2009).

Individual Differences in the Effects of Mobile Happiness Reporting

While no overall group differences were found, changes in happiness as a function of reporting group were moderated by depressive symptoms, Time × Group × CES-D G13 b(SE) = −.0024(.0009), p = .011; and, separately, by neuroticism, Time × Group × Neuroticism G13 b(SE) = −.035(.013), p = .010. Using dummy codes for greater resolution, depressive symptoms was a trend for moderating the effect of responding to 1 versus 3 texts per day, Time × Dummy1 × CES-D G14 b(SE) = −.0033(.0018), p = .072; and significantly moderated the effect of responding to 1 versus 6 texts per day, Time × Dummy2 × CES-D G15 b(SE) = −.0048(.0019), p = .008. Neuroticism significantly moderated the effect of responding to both 1 versus 3 texts per day, G14 b(SE) = −.074(.026), p = .006; and to 1 versus 6 texts per day, G15 b(SE) = −.074(.027), p = .008.

These patterns of moderation are displayed in Figure 2 . As shown in the top panel (Figure 2A), the effect of reporting frequency on changes in momentary happiness was different among those lower and higher in depressive symptoms. Although none of the simple slopes reached statistical significance, the patterns of changes for individuals in the 1, 3, and 6 texts per day groups were nearly reversed as a function of depressive symptoms. Increases in the frequency of reporting corresponded with increases in happiness trajectories among those lower in depression (Figure 2A, left) but corresponded with decreases in happiness trajectories among those higher in depression (Figure 2B, right). Depressive symptoms accounted for 7.1% of the variance in the changes in happiness as a function of reporting group. This percentage reflects a Cohen’s d of .55 (Ellis, 2009), which is considered a medium effect size (Cohen, 1988).

OPEN IN VIEWER

Figure 2.

The top panel (A) shows the changes in happiness over time for the three experimental groups among those low (−1 SD) and high (+1 SD) in current depressive symptoms (Center for Epidemiologic Studies Depression scale [CES-D] scores). The bottom panel (B) shows similar changes among those low (−1 SD) and high (+1 SD) in neuroticism. All values were estimated from multilevel regression models. Significant simple slopes are noted: t (trend), p < .10; * p < .05.

Patterns for neuroticism were similar and stronger, as shown in the bottom panel (Figure 2B). Again, the slope patterns were reversed as a function of neuroticism. Individuals lower in neuroticism generally became happier with more frequent reporting (Figure 2B, left), whereas individuals higher in neuroticism became less happy with more frequent reporting (Figure 2A- right). Here, the effect size was nearly double that of depressive symptoms, with neuroticism accounting for 12.8% of the variance in the changes in happiness as a function of reporting group (Cohen’s d = .77, a “large” effect). Moreover, for neuroticism, two of the six simple slopes were significant. First, as shown in Figure 2B (left), among those lower in neuroticism, reporting happiness three times a day was associated with significant increases in happiness over time, simple slope b(SE) = .056(.011), p = .02. Interestingly, doubling the number of reports to six times per day yielded no significant increases in happiness, simple slope b(SE) = .011(.013), p = .44. Second, as shown in Figure 2B (right), surprising effects were shown among people higher in neuroticism who reported their happiness once per day. Once-daily reporting significantly increased their happiness over time, simple slope b(SE) = .040(.015), p = .01.2

None of the other moderators were significant. Self-esteem exhibited patterns similar to depressive symptoms and neuroticism but did not reach statistical significance, linear results: Time × Group × RSES G13 b(SE) = .022(.014), p = .11; dummy results: Time × Dummy1 × RSES G14 b(SE) = .048(.026), p = .07; Time × Dummy2 × RSES G15 b(SE) = .045(.027), p = .09. Trait happiness was not a significant moderator, linear results: Time × Group × SHS G13 b(SE) = .008(.008), p = .27; dummy results: Time × Dummy1 × SHS G14 b(SE) = .018(.014), p = .20; Time × Dummy2 × SHS G15 b(SE) = .016(.015), p = .29. And experience with texting was not a significant moderator, ³ linear results: Time × Group × Texting Experience G13 b(SE) = −.008(.0080), p = .34; dummy results: Time × Dummy1 × Texting Experience G14 b(SE) = −.002(.001), p = .21; Time × Dummy2 × Texting Experience G15 b(SE) = .003(.002), p = .10.

Discussion

Was intensive mobile self-reporting of happiness detrimental to happiness? Not when the sample was examined as a whole. In fact, we found no evidence for unqualified reactivity in the intensive mobile reporting of happiness. Young adults randomly assigned to report their happiness 3 or 6 times per day showed similar patterns of happiness over the 13 days compared to those who only reported their happiness once per day. However, we did find evidence for qualified reactivity. People at the extremes of negative affectivity—those either low or high in depression or neuroticism—were affected by the frequency of reporting. As happiness reporting frequency increased from 1 to 3 to 6 per day, young adults lower in depressive symptoms or neuroticism showed increasing happiness trajectories, whereas those higher in depressive symptoms or neuroticism showed decreasing happiness trajectories.

There could be a number of mechanisms underlying these patterns. First, it is unlikely that changes are simply due to the act of texting. Students in this sample, including those high in depressive symptoms or neuroticism, were experienced with text messaging, replied quickly to the texts, and showed excellent response rates, suggesting that texting was fairly natural and not particularly burdensome. Moreover, participants’ self-reported experience with texting was not a significant predictor of changes in happiness as a function of reporting group. Instead, the pattern of results—which were strongest for depression and neuroticism, weaker for self-esteem, and not at all significant for trait happiness—suggested that effects were driven more by factors related to negative emotional vulnerabilities and less by factors linked to the self-concept or to trait positivity. Such factors could include self-focus mechanisms, distress over daily mood, rumination tendencies, or some combination thereof. Frequent texts about happiness likely evoked attention to the self, which previous research suggests can impair mood among those higher in depression, neuroticism, and lower in self-esteem (Field, Joudy, & Hart, 2010; Mor et al., 2010; Sakamoto, 1998; Phillips & Silva, 2005). Moreover, texting about happiness seemed to be particularly distressing among those with more negative emotional tendencies (depression and neuroticism). For these individuals, frequent happiness texts may have drawn attention to their typically unpleasant emotional state, thus bearing the potential of perpetuating a downward spiral of satisfaction. For people without such vulnerability, frequent texts may have evoked attention to their typically positive states, enabling them to capitalize on the benefits of their positive emotions, thus creating a potential for an upward spiral of satisfaction. Intensive self-reporting may have also stimulated rumination processes among those higher in depression and neuroticism (Nolen-Hoeksema et al., 2008) leading to more negative changes in happiness. Lastly, any mechanism would need to explain why emotionally negative participants got happier over time when asked to report their happiness once a day. It is possible that a single randomly occurring text simply did not provide enough feedback to vulnerable individuals that they were unhappy. This lack of negative feedback could have had a contrast effect of increasing their happiness over time. This effect warrants replication.

There were several limitations of the study. First unlike some happiness-tracking applications (e.g., Mappiness), we did not provide feedback to participants about their happiness levels or the conditions and circumstances in which they are happiest. Instead, we tried to isolate the effects of tracking happiness, which is core to all mobile happiness applications. Explicit feedback may change the nature of the observed effects either by accelerating them (e.g., by making depressive or high neuroticism participants even more aware of their unpleasant states) or by ameliorating them (e.g., by enabling depressive or high neuroticism people to select activities and environments beneficial to well-being). Future research should examine the effects of feedback on potential reactivity. We also did not include a comparison condition in which people reported their unhappiness or other emotional states. While this design feature allowed us to focus participants’ attention on happiness, without testing other states, it is not possible to generalize our findings to other valenced emotional states.

This study has several implications. Foremost, our findings shed light on the potential reactive effects of emotion tracking by demonstrating that reactivity depends on individual difference factors. Future experiments aiming to replicate prior research (e.g., Mauss et al., 2005; Schooler et al., 2003) might consider including trait emotion factors as moderators, which may lead to greater consistency in findings. Our results also suggest that testing for reactivity in experience sampling data by examining main-effect changes in states, symptoms, or behaviors may underestimate the extent of reactivity. Thus, like Barta, Tennen, and Litt (2012), we also encourage more systematic investigation into “moderated reactivity” that takes into account individual differences.

From a technology standpoint, we have also demonstrated that text messaging through personal mobile phones is a feasible, efficient, and low-cost platform for collecting brief self-report data in large samples (see also Kemp, Burt, & Furneaux, 2008; Kuntsche & Robert, 2009). Participants responded relatively quickly to most of the texts they received (< 2 minutes), showed excellent response rates (96%), and reported feeling very comfortable with the procedure. Thus, text messaging can be added to a list including web-enabled smartphones and the Day Reconstruction Method as tools for conducing large-scale experience sampling (see Mehl & Conner, 2012). Importantly, use of these tools will require continued research into the potential reactive effects of self-tracking.

Our findings also suggest some caution in the use of happiness-related mobile technology—but only for some individuals, under some conditions. Specifically, we observed that depressive or high-neuroticism people showed downturns in their happiness as tracking increased from 1 to 3 to 6 reports per day. Given these patterns, we might (tentatively) suggest that answering more than 3 happiness reports per day for 2 weeks or more may not be beneficial for this population. While it is not known whether depressed or neurotic individuals are more or less likely to use mobile happiness-tracking tools, the current findings suggest that excessive happiness tracking—without proper restraint, support, or guidance—might not be a good strategy of self-improvement for individuals with emotional vulnerabilities. Thus, for some people, there may be some truth to the observation made by the utilitarian philosopher John Stuart Mill: “Ask yourself whether you are happy and you cease to be so” (1873/1989, p. 94, cf. Kurtz & Lyubomirsky, 2012). Although excessive self-reflection may be ill-advised, as noted by Kesebir and Diener (2008), engaging in activities like meditation and counting blessings may be suitable strategies for boosting happiness. Conveniently, there are mobile applications for enacting these happiness-boosting activities in daily life (e.g., Live Happy™).

More broadly, our findings suggest that a hyperfocus on happiness common in today’s “happiness climate” may have negative consequences to some individuals. In this way, our results parallel recent research showing that people with low self-esteem feel worse after making positive self-statements (“I am a lovable person”; Wood, Perunovic, & Lee, 2009). Just like Wood and colleagues found that positive thinking backfired for those with negative self-concepts, we found that evidence that intensive reflection on happiness backfired for those with negative emotional tendencies. These findings also complement those of a recent laboratory study, suggesting that people who value happiness, paradoxically, may experience less happiness after positive events because they are disappointed in their experience (Mauss, Tamir, Anderson, & Savino, 2011). Given the growing awareness of the boundary conditions and potential costs of excessive positivity (Gruber, Mauss, & Tamir, 2011), perhaps these studies suggest the time is right for a healthy correction in the pursuit of happiness.