Emotion regulation is highly relevant to subjective well-being from both hedonic and eudaimonic perspectives. According to the hedonic perspective, well-being is the experience of happiness or the occurrence of positive affect (PA) and absence of negative affect (NA) [1]. In the eudaimonic perspective, well-being arises when individuals live with a sense of growth, meaningfulness, and purpose [2]. People often engage in emotion regulation to obtain hedonic benefits (eg, feel more PA and reduce NA) [3], but eudaimonic motivations have been found to be important as well [4] (eg, downregulation of negative emotions to achieve a sense of growth in one’s ability to handle daily stressors).

An important aspect of emotion regulation is emotional clarity, a person’s ability to lucidly identify the type of emotion they are experiencing [5]. Emotional clarity is highly relevant to the James Gross Model of Emotion Regulation [6]. The model consists of emotion regulation strategies that are either “antecedent-focused” or “response-focused,” which refers to whether the strategy is used before or after an emotional response fully develops. In the updated emotion regulation model of Gross, understanding and identifying one’s emotions accurately (ie, emotional clarity) are precursors to both these types of emotion regulation strategies [7]. Therefore, individuals with low emotional clarity are less likely to use emotion regulation strategies (as they are failing to identify the need for them), which can negatively impact well-being. Lower emotional clarity has often been associated with reduced mental health [8-10], although there are exceptions. For instance, prior research has suggested that higher emotional clarity may be adaptive primarily for individuals who do not have very frequent and strong experiences of negative emotions but maladaptive for those who frequently have strong feelings of NA [11].

Both direct and indirect measures of emotional clarity have been developed [12,13]. Direct assessments involve the metacognitive task of reflecting on one’s emotional clarity level, while indirect assessments measure the performance of a task relevant to emotional clarity (ie, answering emotion items) and do not require self-insight [12]. Emotional clarity is commonly directly assessed with cross-sectional self-report measures such as the clarity subscales of the Trait Meta-Mood Scale [14] and the Difficulties in Emotion Regulation Scale (DERS) [15], but indirect measures have been argued to have potential advantages over self-report assessments [12,13]. For instance, they could help avoid possible issues with subjective reports, including poor insight into emotional clarity and the possibility of social desirability bias (ie, participants not wanting to report uncertainty about their feelings) [12,16]. However, for the potential utility to be realized, further investigations of the validity of indirect measures of emotional clarity are needed.

Response times (RTs) to emotion questions in ecological momentary assessments (EMAs) have been argued to be indirect measures of emotional clarity at the momentary (within-person) level [12]. Theoretically, the greater an individual’s momentary affective clarity, the less time should be needed to provide a rating of momentary affect [12], whereas longer RTs to affect items should be indicative of lower emotional clarity. Evidence supporting this theory has been found, such as shorter RTs to affect items being associated with better momentary emotion regulation and mood [12]. However, emotional clarity may be confounded with emotional intensity [9], and evidence suggests the validity of RTs as a measure of affective clarity is enhanced by controlling for emotional intensity at the within-person (and not between-person) level [17].

There was no strong evidence that the study-long aggregates of emotion item RTs (between-person level) could act as indicators of trait emotional clarity [12,17]. These aggregates were found to not have significant relationships with global measures of emotional clarity and have inconsistent relationships with global emotion regulation measures [12,17]. Suggested reasons for the lack of a relationship between global measures of emotional clarity and the aggregate of emotion item RTs include a modality difference (ie, self-report vs indirect behavior-based assessment), low conceptual correspondence (ie, different forms of emotional clarity are being assessed), and difference in assessment timing (ie, 1 time vs repeated EMA measurements) [12,17]. The relationships between the study-long aggregate of emotion item RTs and subjective well-being variables of relevance to emotional clarity (eg, depression and anxiety) [8,18] were not examined, which could have served as useful additional convergent validity tests. Finally, the possibility that processing speed had a confounding effect on correlations between study-long aggregates of emotion item RTs and other variables was not investigated. In prior research, RTs to emotion EMA items have been found to have moderate correlations with processing speed measures [19], suggesting that individual differences in emotion item RTs may at least in part be attributable to processing speed.

Another indirect measure of emotional clarity was recently proposed, drift rate, which is computed using the drift-diffusion model [20]. This model, which is often used in cognitive psychology, was explicitly developed to disentangle different components of RTs [21]. The drift-diffusion model proposes that decisions (eg, choosing responses on EMA items) are made through the accumulation of information until a threshold of sufficient information (as determined by the individual) is reached [21]. Decisions can be fast if the speed of information accumulation (ie, drift rate) is fast, if the threshold for a decision (ie, boundary separation or response caution) is low, or both. The D-diffusion item response theory (IRT) model is an IRT version of the drift-diffusion model that was specifically designed for the analysis of self-report ratings, such as emotion ratings [22]. In the context of answering mood questions, the D-diffusion IRT model additionally considers that RTs may also be fast if extremes of the emotion are experienced (ie, very high or low levels), making the provision of ratings more straightforward [20]. When the D-diffusion model is applied to EMA responses and their RTs, it can take into account all the aforementioned influences on RTs and output drift rate. In the EMA context, drift rate can be interpreted as the speed of access to information relevant to the question being asked [20,21], which here is information regarding emotions. A more detailed description of how the drift rate parameter was computed can be found in the Emotional Clarity Measure 2: Drift Rate From the Drift-Diffusion Model subsection in the Methods section.

A person would be considered to have high emotional clarity when responding both fast and carefully to emotion items, which differs from how emotional clarity assessed via RTs considers only speed. As the absolute difference between a person’s NA level and the level of NA captured by an item increases, the D-diffusion model assumes that an individual would be expected to have faster RTs as a result of the so-called “distance-difficulty” principle, a well-established phenomenon in the RT literature [23], whereby the more items contrast with a person’s state, the easier they are. For instance, an individual with a very low NA would be expected to quickly respond to an item asking about being scared (an item often associated with a high NA) [24], whereas the same individual would be expected to require a longer RT for an item asking about irritability (which has been associated with a low NA) [24]. According to the D-diffusion model, response caution is low when a person responds similarly quickly (or slowly) to all items regardless of their content (an indication that the person did not answer carefully). Assuming a particular observed item RT, when response caution is lower (ie, less information is gathered), drift rate (ie, the rate of information accumulation or information divided by time) would also be lower. Conversely, for a person with a higher response caution, the estimated drift rate (and emotional clarity) would be greater for the same RT.

Our initial evidence suggested that the drift rate parameter derived from EMA NA items may be an indicator of emotional clarity of low versus high NA, including expected relationships (ie, negative associations) at the between-person level with neuroticism and depression [20], but we did not account for factors aside from emotional clarity that could impact drift rate. For instance, drift rate could be impacted by individual differences in the general speed of responding to questionnaires (eg, due to reading speed, motor behavior, and familiarity with computer use). Studies on emotional clarity measured by survey item RTs had accounted for individual differences in this baseline speed of responding by adjusting for it [12], but this was not done in our prior study [20]. Cognitive processing speed was hypothesized to also potentially impact drift rate but was also not adjusted for, as the measure was not available in the prior study [20].

The analyzed data were from an EMA study focused on investigating the relationships among momentary blood glucose level, emotional state, and functioning in adults with T1D [29]. Participants were recruited from 3 clinical sites through health care provider referrals, mailings, flyers, and emails. Inclusion criteria were having a diagnosis of T1D, being able to speak and read English or Spanish, and the ability and willingness to carry out study procedures (eg, completion of EMAs and cognitive tests on smartphones). Study procedures included the completion of baseline surveys, training in the use of study devices, 2 weeks of 5 to 6 EMAs and ambulatory cognitive tests daily, and follow-up surveys. EMA surveys began at participants’ selected wake-up time each day and were administered at 3-hour intervals after that until sleep time. If a participant reported that they would likely be sleeping by the time of the sixth survey (ie, 15 hours after the first survey), then they were given the option to complete 5 surveys daily instead of 6. To encourage EMA compliance, 3 brief check-in emails or calls were scheduled with study staff. In addition, if EMA survey completion was low, then study staff would contact the participant to query if any support was required. The study procedures are described in greater depth in a protocol paper [29].

All study procedures were approved by the University of Southern California’s Institutional Review Board (proposal #HS-18-01014). Informed consent was provided before study participation. Participants were compensated US $200 via a reloadable debit card for completion of all study procedures. Study data were deidentified before analysis.

Characteristics of the study sample, 196 adults with T1D, are shown in Table 2. The median EMA completion rate over the 2-week study period was 92% (IQR 11%).

Descriptive statistics for the EMA variables are presented in Table 3. Distributions for RTs to individual NA and PA items are shown in Figures S1 and S2 in Multimedia Appendix 1, respectively. Tables S1 and S2 in Multimedia Appendix 1 show the between-person correlations between (unadjusted) study measures.

A unidimensional model was found to fit both the (dichotomized) 4 NA and 4 PA items acceptably, justifying the calculation of drift rates for both types of items. For NA, χ²₂=10.6, P=.005; CFI=0.998; TLI=0.994; RMSEA=0.018; and SRMR=0.018, while for PA, χ²₂=53.3, P<.001; CFI=0.989; TLI=0.966; RMSEA=0.044; and SRMR=0.031. All these values were within commonly suggested ranges for acceptable model fit [44]. The within-person omega estimate for the 4 dichotomized NA items was 0.710, while the between-person omega estimate was 0.938. For the 4 dichotomized PA items, the within-person omega estimate was 0.674, and the between-person omega estimate was 0.939. The within-person omega estimate for the 4 nondichotomized NA items was 0.793, while the between-person omega estimate was 0.955. For the 4 nondichotomized PA items, the within-person omega estimate was 0.773, and the between-person omega estimate was 0.937. For all PA and NA items, the observed RT distributions (histograms) were consistent with the drift-diffusion model–predicted RT distributions as per density plots of these RTs (Figures S1 and S2 in Multimedia Appendix 1), indicating a good fit overall for the D-diffusion IRT model [22].

Figure S3 in Multimedia Appendix 1 shows the between-person reliabilities as a function of the number of EMA prompts completed for each proposed emotional clarity indicator. For all indicators, reliability increased with more prompts, but each of the proposed emotional clarity indicators demonstrated acceptable reliability (ie, ≥0.70) with a relatively small number of EMA occasions. NA RT required 5 EMA prompts for acceptable reliability, and PA RT required 4 prompts. Both average NA and PA drift rates showed 0.70 reliability when 7 EMA prompts were completed.

Relationships between NA RT and other measures were not consistent with our hypotheses. No significant associations were found with subjective well-being or emotion regulation measures (Table 4; P values ranging from .08 to .98). PA RT was not significantly associated with any well-being or emotion regulation variable (Table 4; P values ranging from 25 to 93), except, unexpectedly, for diabetes distress (r=0.17; P=.009). After adjustment for processing speed, the associations between both NA RT and PA RT and diabetes distress were unexpectedly significant in a positive direction (r=0.14; P=.03 and r=0.20; P=.002, respectively). Adjustment for emotional intensity changed effects very minimally at the between-person level, consistent with the findings of a prior study [17], so the results from that model were not reported here. Both NA RT and PA RT had significant associations with processing speed (r=0.25; P<.001 and r=0.21; P=.001, respectively).

The associations between NA drift rate and other measures were consistent with our hypotheses overall. Unexpectedly, NA drift rate was not related to the lack of emotional clarity (P=.13) or 3 of the other DERS subscales (Table 4; P values from .18 to .76). It was also not associated with life satisfaction (P=.17). However, a greater NA drift rate was significantly associated with lower neuroticism (r=–0.18; P=.01), depression (r=–0.27; P<.001), anxiety (r=–0.27; P<.001), and diabetes distress (r=–0.17; P=.005); lower overall difficulties with emotion regulation (total DERS score; r=–0.15; P=.03); less limited emotion regulation strategies (r=–0.15; P=.048); and lower lack of emotion awareness (r=–0.18; P=.009). Adjustment by processing speed resulted in minimal changes. When NA and PA drift rates were computed from mood items that were dichotomized at each person’s mean NA and PA ratings (rather than at the scale midpoint), the results showed mostly nonsignificant correlations with other measures (refer to Table S3 in Multimedia Appendix 1).

Consistent with expectations, PA drift rate was not significantly associated with most subjective well-being or NA-specific emotion regulation measures (Table 4; P values from .09 to .76). Unexpectedly, PA drift rate was not significantly associated with emotional clarity (P=.59) or emotional awareness (P=.20) before or after adjustment for processing speed. Furthermore, contrary to expectations, a greater PA drift rate was associated with greater diabetes distress (r=0.17; P=.009). Neither NA drift rate nor PA drift rate had significant associations with processing speed (P=.18 and P=.78, respectively). Without adjustment for baseline speed, a few of the associations differed, such as the relationship between NA drift rate and life satisfaction (r=0.16; P=.02; Table 5).

The most notable finding from this study was that the average NA drift rate, a proposed indicator of typical emotional clarity, had expected associations overall with validated measures of subjective well-being and emotion regulation, both before and after adjustment for processing speed and emotional intensity. By contrast, NA RT, another proposed indicator of typical emotional clarity, did not have the anticipated associations with the validated measures. Relative to NA drift rate, NA RT may be confounded by a greater number of factors aside from emotional clarity.

Nevertheless, there is no sufficient evidence to conclude that the average NA drift rate is a valid indicator of NA clarity, while the average NA RT is not a valid indicator of NA clarity due to sample size constraints. In post hoc power analyses, with our sample size of 196 participants, there was 80% power to detect a between-person correlation of 0.20. Therefore, the study may have been underpowered to detect small correlations that might still be meaningful. Furthermore, the sample size was not chosen a priori to be sufficiently powered to detect between-person relationships after adjustment for multiplicity of testing [59] because the sample size was conditioned on research questions that were more primary for the diabetes EMA study.

While we cannot conclude that NA drift rate is a valid indicator of NA clarity, our results suggest that researchers in future studies should continue to investigate NA drift rate as an implicit measure of emotional clarity. We applied stringent validity tests by adjusting for both individual differences in the baseline speed of responding and processing speed and found significant associations between NA drift rate and 4 (80%) of the 5 subjective well-being ratings. Correlations with 4 (67%) of the 6 emotion regulation subscales were in the expected directions. Finally, it is notable that NA drift rate had a correlation of 0.63 (P<.001) with NA RT. People with a higher NA drift rate responded faster to the NA emotion items, had greater subjective well-being (eg, fewer depression symptoms), and had fewer difficulties with overall emotion regulation, which are all aligned with the expectation for an emotional clarity measure. Collectively, study results suggest that NA drift rate deserves further attention in future research.

Some of the magnitudes of correlations between NA drift rate and other study measures were comparable to sizes of correlations between formal (self-report) assessments of emotional clarity and other measures found in prior studies. One study found that in a group with generalized anxiety disorder, a self-report assessment of emotional clarity had correlations of –0.29 and –0.33 with depression and anxiety, respectively [60], which are similar to the correlations found in this study. In the same study, the association between emotional clarity and depression or anxiety was not significant in the healthy control group [60], suggesting that the mental health status of the sample may affect the magnitude of the observed relationships. Other correlations found between emotional clarity and depression were –0.24 in a clinical sample [61] and –0.29 in elementary school–aged children [62]. Correlations between self-reported emotional clarity and neuroticism ranged between –0.31 in a sample of adolescents [63] and –0.37 in a sample of college students [64], and correlations between emotional clarity and life satisfaction ranged between 0.31 in adolescents [65] and 0.35 in undergraduate students [36], which are larger than the correlations observed in this study.

When NA drift rate was computed from mood items that were dichotomized at each person’s mean response rather than the scale midpoint, no significant associations were observed with other study measures. The distance-difficulty hypothesis underlying the D-diffusion model that was used to generate the NA drift parameter assumes that people respond faster to items that contrast more with their current state [23]. According to this hypothesis, an individual with very low NA (regardless of their average level of NA) was expected to quickly report not being scared (an item often associated with high NA). Dichotomizing mood at each person’s own midpoint created a variable representing whether their mood was higher or lower relative to their personal average and not higher or lower in absolute terms (which could be approximated by dichotomizing at the scale midpoint). Perhaps, the former was less relevant to the distance-difficulty hypothesis compared to the latter because it captured relative mood and not actual mood, leading to the creation of NA drift parameters with no associations with other study measures.

The reliabilities for the average NA drift rate, PA drift rate, NA RT, and PA RT were all acceptable with a small number of EMA prompts (ie, 4 to 7 EMA prompts or 1 to 2 days of EMA surveys). Therefore, reliable measurement of these proposed emotional clarity indicators would likely be feasible in most EMA studies where affect items are administered and RTs are recorded.

We found preliminary evidence supporting the argument that emotional clarity deficits are valence specific [9]. NA drift rate and PA drift rate had differential associations with self-report measures. Furthermore, they were moderately correlated with each other (r=0.38; refer to Table S1 in Multimedia Appendix 1). Had NA and PA drift rates been redundant with one another, a high correlation would have been expected.

Greater PA drift rate and PA RT were unexpectedly found to be associated with higher diabetes distress and not associated with the awareness or clarity subscales of the DERS. It is unclear why people with greater diabetes distress would have greater clarity of positive emotions. Perhaps, when overwhelmed with burden from diabetes, people had a greater appreciation of positive emotional states and hence greater clarity of PA. One possible reason why PA drift rate and PA RT had far from significant associations with self-reported emotional clarity may have been because, given that other items in the DERS-SF asked questions relevant to the NA context, participants were primed to answer the emotional clarity questions with reference to feeling NA. More assessments of the validity of PA drift rate and PA RT are needed.

More conclusive evidence of the validity of RT-based measures of emotional clarity may come from studies where NA clarity can be manipulated, and the proposed emotional clarity indicators can be compared for sensitivity to these changes. For instance, people who undergo a mindfulness intervention may be expected to have higher NA clarity in the period following the intervention, and this effect should be reflected in changes in drift rates or RTs to NA EMA items.

EMA mood item RT–based measures of emotional clarity have a great potential utility. They can serve as indices of emotional clarity that do not require burdening participants with additional emotional clarity items. Furthermore, they can help avoid possible issues with subjective reports, including poor insight into emotional clarity and the possibility of social desirability bias [12]. Although the validity of the RT-based emotional clarity measures at the within-person level was not investigated in this study, such validity would allow for the investigation of changes in emotional clarity within and across days and the situational factors that contribute to them. For the potential utility of EMA mood item RT–based measures of emotional clarity to be realized, further investigations of the validity of EMA RT–based measures of emotional clarity (ie, at both the between-person and within-person levels) are needed.

We had decided not to adjust for multiple comparisons, but to also acknowledge that any results would require replication by future studies. It has been argued that the need for adjustment for multiple comparisons should be evaluated on a case-by-case basis and P value adjustment should not be used for all analyses [66]. For instance, adjustment for multiple comparisons comes with not only the benefit of lowering type 1 error but also the disadvantage of increasing the chance of type 2 error. Therefore, one factor to consider when deciding whether to adjust for multiple comparisons is the relative cost of type 1 and type 2 errors for a particular research question [67]. In confirmatory studies with results that have implications for changes in clinical practice or the use of a treatment, the cost of a type 1 error may be higher than that of a type 2 error; hence, P value adjustment for multiple comparisons would be sensible [66]. When performing post hoc analyses on existing data as part of theory building and testing (and without direct treatment implications), the relative cost of a type 2 error may be higher; hence, there may be a stronger argument for not using multiple comparisons adjustment [66]. That is, a type 2 error could cause researchers to not detect potentially important findings [68]. If adjustment is not used, there would need to be an acknowledgment that, to account for the possibility of a type 1 error, further research is needed to examine whether results can be replicated [66].

Nevertheless, we still tested the effect of false discovery rate adjustment on the P values of correlations for the different groups of hypotheses that were tested (eg, association between NA drift rate and subjective well-being measures). Adjustments for multiple comparisons are often applied separately for distinct families of hypotheses [69]. Tables S4 and S5 in Multimedia Appendix 1 show the false discovery rate–adjusted P values associated with Tables 4 and 5, respectively. The biggest differences to note were that several of the associations between NA drift rate and emotional regulation measures were no longer significant.

Using the drift-diffusion model had the advantage of reducing the impact of individual differences in response caution (and potential careless responding) from emotion item RTs but had the disadvantage of assuming a 2-choice task (eg, a high vs low NA) underlying people’s emotion ratings. Because the drift-diffusion model required making the continuous PA and NA items dichotomous for data analysis, granular differences in emotional clarity may have been missed with the drift rate parameter.

We were unable to examine the within-person validity of RT-based clarity measures because EMA measures key to such testing (eg, self-reported emotional clarity and mood regulation success) [12] were not administered in this study. Future studies are needed to examine the within-person validity of the drift rate parameter as a within-person indicator of emotional clarity.

The context in which RTs for emotion items were calculated was similar but not identical to that in prior work. For instance, in this study, median RTs for EMA were calculated based on RTs to 4 items. In prior studies, the median RTs for 5 to 8 items were computed [12,17]. In the original paper examining the validity of RTs to emotion items as indicators of emotional clarity, bipolar mood items were used (eg, items with options from “very unhappy” to “very happy”) [12], whereas our study analyzed unipolar mood items. The results of this study may have been impacted to an extent by differences in EMA administration, such as variations in the type of emotion items used.

Additional evidence is needed that study results generalize beyond adults with T1D. Other populations with an increased likelihood of experiencing lower subjective well-being (eg, individuals with various chronic conditions) [70] may be appropriate targets for future emotional clarity studies.

A measure of NA drift rate derived from RTs to momentary NA items had expected associations with validated measures of relevance to emotional clarity, providing initial evidence supporting its validity as an indicator of individual differences in the clarity of negative emotions. The validities of NA RT, PA RT, and PA drift rate were not strongly supported by our results. More studies are needed to investigate the validities of NA and PA drift rate and NA and PA RT with larger sample sizes. The development of passive measures of emotional clarity would help create minimally burdensome measures of emotional clarity that are less vulnerable to possible issues from subjective self-reports, such as poor clarity insight and social desirability bias. Such measures may be useful in investigations relevant to the role of emotional clarity in people’s experience of well-being.