Technical Comments

Comment on “Morality in everyday life”

See allHide authors and affiliations

Science  15 May 2015:
Vol. 348, Issue 6236, pp. 767
DOI: 10.1126/science.aaa2409


In examining morality in everyday life, Hofmann et al. (Reports, 12 September 2014, p. 1340) conclude that being the target of (im)moral deeds impacts happiness, whereas committing them primarily affects one’s sense of purpose. I point to shortcomings in the analyses and interpretations and caution that, based on the methodological approach, conclusions about everyday life relationships between morality and happiness/purpose are premature.

Hofmann et al. (1) take the study of morality out of the lab and into real life. They do so by using smartphones to assess momentary happiness and sense of purpose five times a day across 3 days, along with participants’ reports on whether they committed, were the target of, witnessed, or learned about moral or immoral acts within the past hour after being signaled. The aim of providing a more ecologically valid picture of moral life in a large sample of N = 1252 participants is important and deserves close attention by the scientific community.

The white bars in Fig. 1 show the findings presented in the original article, after correction of a computational error (also corrected online in the original paper). With Cohen’s d (2) ranging between 0.57 and 1.34, the effects of moral and immoral acts on everyday happiness are impressive and easily surpass effects sizes typically observed in studies on the impact of major life events on subjective well-being (3), happiness (4), or the effectiveness of positive psychology interventions (5). How can simple (im)moral acts, such as assisting a tourist with directions or disrespecting one’s mother, have such strong effects on everyday happiness and sense of purpose?

Fig. 1 Morality, happiness, and purpose.

Original (white): Reconstructed results from Hofmann et al. (1) after correction for computational errors. Continuous time (black): Accounting for temporal dynamics of happiness/purpose by means of a continuous time analysis. In contrast to the original findings presented by Hofmann et al. (1), all effects are centered at the baseline for reasons of comparability.

To understand this, a closer look at the data analysis and interpretation of results by Hofmann et al. (1) is necessary. The analysis was carried out in two steps. First, about 70% of the total responses, in which no morally relevant events were reported, were deleted from the data set. Second, a two (moral versus immoral) by four (perspective: committed, target of, witnessed, learned about) factorial linear mixed-effects analysis was carried out on the remaining data. This approach is problematic for two reasons.

First, the deletion of such a large portion of data is unnecessary. Although researchers are often trained to think in terms of fully crossed factorial designs (here, a two by four design), the general linear model is not limited to such designs. By using morally irrelevant situations as baseline, the eight different conditions could have been represented by eight dummy variables, making use of all available information. Furthermore, by limiting the analysis to moral and immoral acts, the authors created a situation that has little to do with morality in everyday life, where most situations encountered by individuals are morally irrelevant. In fact, only 104 participants (9%) ever committed an immoral act directly after a moral act, and only 18 participants (1%) were ever the target of a moral act followed immediately by an immoral act (4% for witnessed; 8% learned about). The large effect sizes reported in the study result from comparing these extreme conditions against each other. The effect sizes thus reflect the fact that, if a group of people encountered a moral situation, we can expect them to be happier than a different group who encountered an immoral situation. For the study of morality in everyday life, however, a measure of the expected difference in happiness/purpose between different groups is misleading, because (i) it ignores the low base-rate of the event, and (ii) it describes an unrealistic scenario because any given individual is most unlikely to ever experience a direct transition from a moral to an immoral situation. Including morally irrelevant situations in the analysis, via the dummy variable approach proposed above, provides a more realistic picture. As is apparent from Table 1, the two dummy variables for moral versus immoral acts explain about 2% of the fluctuations in happiness across all conditions (baseline). According to Cohen’s recommendations (2), these are consistently small effect sizes, which not only provide a more realistic picture of the effect of everyday morality on everyday happiness but also are better in line with the existing literature (35).

Table 1 Morality and happiness.

Comparison between effect sizes reported by Hofmann et al. (1) and effect sizes based on a reanalysis that also included morally irrelevant situations. Cohen’s (2) recommendations on effect size interpretation are provided in parentheses. For these comparisons, effect sizes (f2) were computed and rounded up or down to the nearest cut-off value. *Parameter estimates were obtained via the plm package (11) in R (12).

View this table:

Second, the linear mixed-effects analysis does not account for temporal dynamics. The assumption that happiness/purpose at time point t is independent of happiness/purpose at time point t – 1 seems unrealistic and counter to the authors’ own theory and analyses on moral dynamics (1). Although different methods have been developed to account for such dynamics (6, 7), continuous time models (8, 9) seem particularly suited because they generalize readily to designs with arbitrary measurement occasions at the individual level. Results of the continuous time analysis are depicted by the black bars in Fig. 1. When comparing different methods for data analysis, it is important to keep in mind that the primary question is not whether parameter estimates are significantly different from each other but whether the methods result in (significantly) different conclusions. This is the case. For example, in absolute terms, the (corrected) effect of committing a moral act on sense of purpose is lower than the effect of being the target of a moral act, which stands in contrast to the conclusion of the original article. Interestingly, though, although not supported by the authors’ own analysis, this conclusion is remedied by the continuous time analysis (see Fig. 1). Furthermore, whereas the original analysis suggests a significantly stronger reduction in happiness for people who committed rather than witnessed an immoral act, the continuous time analysis suggests a nonsignificant difference.

The strength of the study by Hofmann et al. (1) is the ecological approach combined with a panel data structure with individually varying measurement occasions. In principle, such a design holds the potential to use temporal information to uncover causal dynamics of everyday morality at an average within-person level. Unfortunately, however, Hofmann et al. (1) do not model temporal contingencies at the within-person level. Hence, the data do not show whether “committing moral deeds…boost momentary happiness and sense of purpose” (1), and care must be taken to not generalize beyond the observed group mean difference to within-person processes when interpreting results (10). However, pursuing Hofmann et al.’s (1) ecological approach with a higher temporal resolution regarding the ordering of events in time and better methods to analyze the resulting dynamics, as outlined in this comment, may provide a promising avenue for future research on everyday morality.

References and Notes

  1. Acknowledgments: I thank U. Lindenberger for his comments on an earlier version of this article. I also thank the members of the Intra-Person Dynamics and Formal Methods project at the Max Planck Institute for Human Development for their input and inspiring discussions during the preparation of this commentary.
View Abstract

Navigate This Article