A State Model of Negative Urgency: Do Momentary Reports of Emotional Impulsivity Reflect Global Self-Report?

Abstract

Negative urgency is a trait that is a risk factor for a range of psychopathology. Yet, little research has tested whether global self-report measures of negative urgency truly reflect a heightened association between real-world negative emotions and impulsive behaviors. In a sample of young adults (N = 222) assessed 3 times per day for 10 days, we tested whether negative emotions were associated with multiple facets of impulsivity at the state-level, and whether those associations were moderated by global self-report of negative urgency. Our findings suggest a robust within-person association between negative affect and acting on impulse. However, global self-report of negative urgency did not moderate any emotion-impulsivity association we tested.

1. Introduction

There are several personality traits that are associated with impulsive behaviors, commonly referred to as “impulsigenic traits” (Sharma et al., 2014; Smith et al., 2007; Whiteside & Lynam, 2001). Negative urgency is one such trait, broadly defined as the tendency to act impulsively in the face of negative affect (Whiteside & Lynam, 2001). To date, negative urgency has primarily been measured using the negative urgency subscale of the UPPS-P Impulsive Behavior Scale, a widely used and validated global self-report measure of five empirically-derived facets of impulsivity (Whiteside & Lynam, 2001). Meta-analyses suggest that global self-report of negative urgency is more strongly associated with suicidality and non-suicidal self-injury, aggression, borderline personality disorder, disordered eating, depression, and anxiety (with effect sizes of .25-.58) than are other impulsigenic traits (Berg et al., 2015). Negative urgency is also related to more problematic forms of substance use from childhood through young adulthood, and above and beyond the effects of other impulsigenic traits (see Smith & Cyders, 2016, for a review). This has led many to suggest that trait negative urgency may be an important transdiagnostic risk factor for psychopathology (Johnson et al., 2013; Smith & Cyders, 2016). However, it is unclear whether global self-report measures of negative urgency truly measure the construct of negative urgency as we have come to understand it – as a heightened association between negative affect and impulsivity at the state-level. The current study seeks to validate whether people who report themselves to be high in negative urgency via global self-report actually act ‘urgently’ in day-to-day life.

1.1. Negative urgency as a within-person process

Prevailing theories of negative urgency suggest that there is a strong temporal relationship between affect and impulsivity among those high on negative urgency. For these individuals, impulsivity is expected to be higher in the specific context of negative affect compared to neutral contexts. It is believed that the experience of negative affect biases these individuals towards automatic, impulsive responses and away from more thoughtful, controlled decision making (Carver et al., 2008; Smith et al., 2013; Smith & Cyders, 2016). This emotion-behavior dynamic is represented in global self-report items such as “when I feel bad, I will often do things I later regret in order to make myself feel better now.” Thus, trait negative urgency is theorized to reflect individual differences in the state-level association between negative emotions and impulsive action. Yet, support for the construct validity of negative urgency has been derived largely from trait-level studies, which have shown that global self-report of negative urgency is independent of other impulsigenic traits (Cyders & Smith, 2007; Whiteside & Lynam, 2001), is distinct from trait negative emotionality and trait neuroticism (Smith et al., 2007), and is associated with theoretically consistent psychological outcomes such as psychopathology (Berg et al., 2015). These studies provide support for the predictive validity and nomothetic span of urgency (Whitely, 1983), yet they do not provide information about the construct representation: whether global self-reports of negative urgency accurately measure the state-level association they describe.

Three experimental studies have examined whether global self-report of negative urgency is related to individual differences in the state-level association between negative affect and impulsive behavior. Chester, Lynam, Milich, & DeWall (2017) reported that negative affect induction by social rejection lead to more inhibitory errors in a laboratory impulse control task, but only for individuals with average to high global self-report of negative urgency. Billieux, Gay, Rochat, & Van der Linden (2010) similarly found that global self-report of negative urgency was related to poorer inhibition of prepotent responses in a stop-signal task when emotional stimuli were presented, but not when neutral stimuli were used. Finally, our own research previously found that dietary restraint was associated with consuming more calories after a negative affect induction, but only for individuals high in global self-report of negative urgency (Emery, King, & Levine, 2014). Together, these studies suggest that in a controlled environment, negative affect may impair impulse control specifically for those high in global self-report of negative urgency. While these studies provide initial evidence to support that global self-report of negative urgency reflects a state-level tendency to act more impulsively under negative affect, it is not clear if these associations hold in real-world scenarios of negative affect. Experimental studies by necessity use narrow behavioral measures of impulsive behavior, which may have questionable construct validity (King, Feil, Seldin, Smith, & Halvorson, in press), do not map well on to self-report measures of impulsigenic traits (Sharma et al., 2014) or impulsive behavior (Cyders & Coskunpinar, 2011; de Ridder et al., 2012; King et al., 2014; Sharma et al., 2014), and often have low correlation across different tasks intended to measure the same construct (Duckworth & Kern, 2011; Sharma et al., 2014). Thus, in order to understand whether global self-reports of negative urgency truly reflect the real-world association that it describes, it is necessary to use ecologically valid, state-level methods.

1.2. Negative urgency in ecological momentary assessment

Ecological momentary assessment (EMA) methods sample individuals briefly multiple times per day in order to capture a wide array of real-life situations. In the case of negative urgency, EMA allows an ecologically valid assessment of how negative affect and impulsivity relate to each other moment-to-moment in the lives of participants. EMA reduces the influence of recall bias by asking participants to report on their current or very recent state and does not rely on their ability to make global assessments of their behavioral tendencies. Importantly, it does not require participants to infer cause-effect relationships of their own behaviors the way global self-report measures of negative urgency do. In EMA, these parameters can be derived statistically, greatly reducing possible confounds of global self-report measures.

Some studies using more ecologically valid methods have begun to explore the state-level nature of negative urgency. Sharpe, Simms, & Wright (2019) collected daily diary reports of emotions and impulsive behavior in a clinical sample of patients with heterogeneous personality disorders (N=101). They found that daily negative affect was associated with daily impulsivity, and that a global self-report measure of impulsivity (the Personality Inventory for DSM-5 Impulsivity scale) predicted the magnitude of a participant’s relation between daily negative affect and daily impulsivity. These findings support that there is a variable relationship between affect and impulsivity measured at the daily level that can be captured by global self-report measures of personality. However, this study did not specifically look at global self-report of negative urgency, nor did it use state impulsivity measures that differentiate between distinct types of impulsive behaviors.

A recent study by Sperry, Lynam, & Kwapil (2018) used EMA to examine how the UPPS-P facets of impulsivity manifest in daily life in a nonclinical sample of adults (n=294). They sampled aspects of participants’ emotions and recent impulsive behaviors 8 times per day for 7 days. To measure recent impulsive behaviors, they used 6 items thought to capture the full scope of UPPS-P impulsivity facets and created an ‘impulsivity index’ which was an aggregate of these items. They found that on average, experiences related to negative affect (stress, dysphoria, and irritability) were positively related to impulsive behaviors, and that global self-report of negative urgency moderated this association such that individuals higher in negative urgency had a stronger association between negative emotions and the impulsivity index.

However, the magnitude of the moderation was somewhat modest in comparison to the main effect of global self-report negative urgency on state-level impulsivity. These findings lend some validity to the construct representation of negative urgency, while also suggesting that global self-report of negative urgency may primarily capture trait-level differences in impulsivity. In essence, that people who self-report higher global negative urgency are generally more impulsive at the state level, regardless of context. Because global self-reports of negative urgency correlate highly with measures of negative emotionality and affect-free measures of acting on impulse (Sperry et al., 2018; Sharma et al., 2014), it is necessary to carefully differentiate within-person (state-level) from between-person (trait-level) effects in order to examine how well global self-report of negative urgency truly captures the state-level association described by the construct.

The current study primarily aims to explore whether global self-report of negative urgency measures what we expect it to: the state-level association of negative affect and impulsive behaviors. We then introduce conceptualizations of negative urgency that vary slightly from this primary model in order to explore if any specific emotion-impulsivity relationship is best captured by global self-report. The following sections outline our decision-making process for both the primary and alternative models that we examine.

1.3. Negative affect predicting different impulsive states

The first confirmatory goal of the current study was to use EMA methods to measure negative urgency close to how it is described in the literature: as a dynamic interaction between natural experiences of negative affect and impulsivity that varies in strength from person to person. However, at the trait level, negative urgency describes increased risk for multiple types of impulsive behaviors. For example, many negative urgency items describe increased risk for acting on impulses, feelings, and cravings, while other items reflect risk for poor planning, such as “acting without thinking” when upset.” Negative urgency has also been theorized to represent ‘impulsive inaction’ in the face negative affect, such as avoiding or giving up easily in distress-inducing situations (Carver et al., 2008; King et al., 2018), suggesting it may also predict poor perseverance in the face of negative affect. Although these items demonstrate strong evidence of validity at the trait level, the heterogeneity in impulsive behaviors predicted by negative urgency poses some difficulty when trying to model this process at the state level. For instance, should we expect that, for an individual high in negative urgency, experiences of negative affect equally predict all types of impulsive behaviors? Or that different negative emotions predict all types of impulsive behaviors equally well? Because of these uncertainties, and newer research into EMA methods suggesting that the factor structure of impulsive states is quite similar to the factor structure of impulsigenic traits (Halvorson et al., preprint, 2019), we opted to model negative affect predicting distinct types of state impulsivity separately, instead of using a unidimensional measure of state impulsivity. We used three momentary impulsivity scales representing planning, perseverance (both derived from their respective UPPS-P scales), and ‘acting on impulse,’ which we derived directly from the UPPS-P urgency scales to represent the ‘hot’, impulse-driven behaviors not covered by either planning or persistence but well represented within negative urgency items. Using these scales, we tested whether negative affect better predicts any one of these impulsive states, and whether any of those associations are best represented by global self-reports of negative urgency.

Because our EMA ‘acting on impulse’ scale was derived directly from urgency items, we present the model with negative affect predicting acting on impulse as our ‘best-guess’ model of negative urgency, and hypothesize that this scale will show the strongest relationship with both EMA negative affect and global self-report negative urgency.

1.4. Specific emotions predicting different impulsive states

Although negative urgency is most commonly described as an association between general negative affect and impulsive actions, we recognize that this may be too broad a characterization of the true emotion-behavior relationship at play. “Negative affect” is a broad term that encompasses many different negative emotional experiences, each of which can have distinct behavioral effects (Hussong & Chassin, 1994). Negatively valenced emotions (such as anger, anxiety, and sadness) differ in the degree to which they produce approach- versus avoidance-motivated behaviors (Carver & Harmon-Jones, 2009), and may reasonably impact impulsive behaviors in unique ways. Negative urgency items themselves describe a range of emotional experiences, some specific and some vague, such as: being “upset”, feeling “bad” or “rejected,” and being “in the heat of an argument.” Given this, it is not clear whether a general measure of ‘negative affect’, or specific emotional experiences are more associated with impulsive action at the state-level, and which associations are best captured by global self-report of negative urgency. Thus, as our first exploratory goal, we looked at the state-level associations of specific negative emotions with all three types of impulsive behaviors and compared them to the associations for average negative affect. We then tested whether global self-report of negative urgency predicted the magnitude of these associations.

1.5. Analyses of alternative exploratory models

Lastly, we recognize that certain other aspects of the models we describe may not completely fit with some conceptualizations of negative urgency. For example, it is unclear on what time-scale negative emotions are thought to exert their impact on impulse control (concurrently versus prospectively), or if the characteristic association should only be expected to emerge at sufficiently high levels of negative emotion, or in the presence of other moderators such as time of day or general emotionality. In order to address these slight variations in the way negative urgency is discussed in the literature, we include these “highly” exploratory analyses that test variations on our primary models to guide future pre-registered studies.

2. Methods

2.1. Participants

Participants were high school students at a public high school (n = 61) and college students at a four-year university (n = 161) in the Pacific Northwest. The high school sample was 64% female, and 54% White, 17% Asian, 5% Black, and 15% multi-ethnic, and consisted of students at a single high school who consented to participate in a paid research study (aged 15–18, mean age 16.6). The college student sample was 49% female and 66% White, 21% Asian, and 12% other race/ethnicity, and consisted of students (aged 18–21, mean age 19.5). All participants in the university’s Psychology Subject Pool were invited to complete a general screening survey; those who took the survey and were between the ages of 18–20 at the time of screening, were born in the US or moved before the age of 12, and reported at least weekly alcohol or marijuana were invited to participate in this study.

We considered participants who completed fewer than 3 days’ worth of assessments (9 assessments) to be non-responders, and they were excluded from further analyses. Overall response rates for the EMA surveys were high, with only 7 participants excluded. Of those included, the median number of missed surveys was 4 out of 30. 51% of EMA were completed within 10 minutes of being delivered, and 86% were completed within the first hour.

The current study served as pilot data for a larger funded project; thus, we collected data from as many eligible participants as we could recruit within the time frame of data collection for both samples. As a result, we consider the minimum detectable effect size (MDES), given our sample size, alpha, and desired power, rather than our power to detect hypothesized effects. With 222 participants, we had power (1 - β = .80, α = .05) to detect correlations as small as r = .16, and regression effects as small as f² = 0.028 at the between-person level. To estimate the MDES for the EMA data, we computed an effective sample size (Snijders, 2005) using our sample size of observations and intraclass correlations (ICCs). The effective sample size adjusts the observed sample size to account for interdependence among clustered observations. At the within-person level, our ICCs ranged from .35 to .42, and with the largest ICC, and roughly 5,600 observations, our effective sample size was 501. This would give us sufficient power to detect correlations as small as r = .11, or regression effects at the EMA level as small as f² = 0.050 (R² increase = .047; 1 - β = .80, α = .05). Given one major focus of the present paper was on estimating interactions between trait urgency and state negative emotions, we conducted a Monte Carlo simulation in R to estimate the MDES for such an interaction, assuming an effective sample size of 501, and using the results from our primary models to inform other parameter estimates (such as intercept values, residual variances, etc.). Results suggested we could detect interactions between trait negative urgency and negative emotions as small as b = .173 with power (1 – β) = .80 and α = .05. In other words, we could detect an interaction where a 1-unit difference in trait urgency (i.e. choosing “Strongly Agree” on average rather than “Agree”) would be related to a .173 difference in the slope of negative emotions on acting on impulse.

2.2. Data collection procedures

Study procedures were approved by Institutional Review Board. All participants completed a web-based baseline survey and were oriented to the study and trained in the EMA protocol by study staff. The college sample completed these procedures in-lab, while the high school sample completed them remotely via telephone. After orientation, participants completed 10 days of EMA, beginning on the Friday after their baseline assessment. Three times per day, participants received a text message with a link to a brief web-based survey. The college sample received EMAs randomly within fixed windows (morning (9am-1pm), midday (1pm-5pm), and evening (5pm-9pm). The high school sample received their morning survey at a fixed time during their lunch period, and the final two randomly distributed between the end of the school day and 9pm. Surveys were programmed to be at least two hours apart, and participants had two hours to complete them. Reminder texts were sent after one hour if the survey had not been completed. College participants received course extra credit for participation in proportion to the number of EMA surveys they completed. High school students received $1 for each survey completed.

2.3. Measures

2.3.1. EMA Emotion and Affect

At each EMA, participants were asked to rate how strongly they had felt five negative emotions (irritable, unhappy, anxious, angry, and bored) since the beginning of the last assessment window (or since they woke up, in the case of the morning assessment). Participants responded on a sliding visual analog scale with response anchors ‘not at all’ at the minimum, and ‘very much’ at the maximum. For measures of individual negative emotions, responses were scored between 0–100 based on their placement on the slider bar. To calculate our measure of ‘negative affect,’ we computed an average of all negative emotions at each observation (α = .78).

2.3.2. EMA Impulsivity

At each EMA, we administered a random subset of 8 out of a total of 14 items adapted from existing scales to measure impulsive behaviors. 6 items were adapted from the UPPS-P urgency sub-scales, 4 from the planning scale, and 4 from the persistence scale (Whiteside & Lynam, 2001). At least 2 items from each subscale were administered at each time point, based on a randomization algorithm that presented items at random (i.e. unrelated to participant responses). The UPPS-adapted items were altered to reflect the momentary nature of the assessment. For the urgency items, we removed any affect-related language so that they would reflect only the acting on impulse component of the urgency construct – which is how we refer to these items for the remainder of this paper. The full text of these items can be found in Appendix 1. Participants were asked to report on their experiences since the beginning of the last assessment window (or since they woke up, in the case of the morning assessment). Their responses were reported using a slider bar with a range of 0 – 100, with the anchors “strongly disagree” to “strongly agree” at the extremes. EMA acting on impulse, planning, and persistence were measured as the mean of the relevant completed items at each observation (α = .85, .70, .70, respectively). Using the same data as a current study, and cross-validated in an independent sample, these measures exhibit good structure at both the between and within person level, and exhibit evidence of convergent and criterion validity with both global self-report and EMA measures (Halvorson et al., 2019), suggesting they capture similar heterogeneity in impulsive behaviors at the state levels as they do at the trait level.

2.3.3. Global self-report negative urgency

We administered the UPPS-P Impulsive Behavior Scale at baseline (Cyders & Smith, 2007; Whiteside & Lynam, 2001). The negative urgency subscale is comprised of 12 items rated on a four-point Likert scale from “strongly agree” to “strongly disagree.” Sample items include “I have trouble controlling my impulses’ and “When I am upset, I often act without thinking”. In our analyses we use a mean of the 12 negative urgency items (α = .86).

2.3.4. Covariates

For all analyses, we controlled for biological sex (1 = male, 0 = female), ethnicity/race (1 = non-white, 0 = white), age, and sample (1 = college sample, 0 = high-school sample). We also controlled for time of day of the EMA (coded as the time elapsed since 9am, the earliest possible time of assessment), weekend vs weekday (1 = weekend, 0 = weekday), and observation number (1–30), to control for temporal effects. Finally, we controlled for an individual’s average negative affect reported across all EMAs.

2.4. Data analysis

We used R Core Team (2018) for all our analyses. We tested our hypotheses using multilevel models (MLM) with the statistical package ‘nlme’ (Pinheiro, Bates, DebRoy, Sarkar, and R Core Team, 2018). For all analyses, we centered affect variable within-person by subtracting each person’s average affect score across all observations, so that 0 represented the individual’s average affect over the EMA session. Global self-report of negative urgency was centered so that 0 represented the average level in our sample. Age was centered so that 0 was equal to age 18, an easily interpretable mid-point in our sample.

For our primary models (negative affect predicting facets of impulsivity), we analyzed them both with and without auto-correlated residuals (which account for similarities between temporally adjacent observations) and examined the relative fit of the nested models. The models with autocorrelated residuals exhibited significantly improved fit (p < .05) compared to those without. As a result, we report the results from this and all other models using autocorrelated residuals throughout our analyses, including exploratory models. For all models, we report and examine the coefficients from the ‘full’ model including both the random slope and the cross-level interactions added, regardless of the significance of the interaction. It is important to note that because variables involved in the interaction were centered, coefficients and their standard errors did not substantively differ across models with and without the interaction. We report the full model because it most closely reflects our research question and hypothesized model. Because many of our analyses were exploratory, we emphasize effect sizes and confidence intervals across models. For all models, we examined model residuals to test whether the model’s assumptions were violated. Throughout our models, we did not observe evidence of violations of assumptions.

2.4.1. Confirmatory: Negative affect predicting impulsive statesr

We first tested the association between negative affect and all three facets of impulsivity during the same observation using a multilevel model with random intercepts and fixed slopes. This model included all demographic and sampling-related covariates as well as an individual’s global self-report negative urgency, and average negative affect over the study period. Next, we tested whether the association between negative affect and each outcome varied significantly across the population by testing whether the addition of a random slope improved model fit. We made a determination of improved model fit if there was a significant likelihood ratio test and a reduction in the BIC. Finally, we added the interaction between global self-report negative urgency and EMA negative affect to test whether baseline negative urgency explained individual differences in associations between affect and impulsivity (the random slope). We assess this by considering both the significance level of the interaction term and the improvement in fit as described above. If there was a negative affect-impulsivity association that reflects negative urgency, we expected to observe 1) a random slope that improves fit, and 2) a significant interaction between EMA negative affect and global self-report negative urgency in predicting that facet of EMA impulsivity.

2.4.2. Exploratory: Specific emotion-behavior relations

We next repeated the above model-building procedure with each combination of specific negative emotion (angry, anxious, irritable, unhappy, bored) and all three impulsive behavior outcomes (acting on impulse, lack of planning, and low persistence). If there was a specific emotion-impulsivity association that best characterized negative urgency, we expected to observe 1) a stronger emotion-behavior association compared to the general effect of negative affect on that type of impulsivity 2) that global self-report negative urgency would be a significant predictor between-person variation of this slope. Given the number of coefficients examined for each combination of emotion and impulsive behavior, we used a strict criterion of p < .001 to avoid highlighting potentially spurious associations.

2.4.3. Very exploratory: Alternate models

We tested whether global self-report negative urgency strengthened the association between momentary negative affect and acting on impulse under the following conditions: 1) Only in those observations where negative affect was higher than the individual’s average. 2) Only in the presence of a quadratic effect of negative affect, such that acting on impulse is observed when negative emotions were strong. 3) Only at the next, rather than concurrent, time point, and 4) Only in the presence of other moderators, such as time of day, weekend vs. weekday, average negative emotion, sex, and high-school vs. college sample. We tested 1 – 4 in the context of our primary hypothesis - using momentary acting on impulse as our only outcome. For these analyses, we again used a strict criterion of p < .001 to avoid highlighting potentially spurious associations.

3. Results

3.1. Descriptive statistics

Descriptive statistics for sample characteristics and study variables are reported in Table 1.

Table 1:

Means and Standard Deviations of Key Variables

	Mean	SD
Participant Characteristics
Age	18.75	1.51
Negative Urgency	2.37	0.57
EMA Measures
# of EMA Observations	25.23	4.24
Acting on Impulse	19.59	20.04
Planning	57.74	24.94
Persistence	57.41	25.77
Anger	16.23	21.52
Anxiety	29.96	28.63
Irritability	20.8	24.38
Unhappiness	19.19	23.47
Boredom	26.29	26.65
Negative Affect	22.49	18.20

Figure 1 presents the correlation matrix for all three types of EMA impulsivity (acting on impulse, planning, and persisting) with EMA negative emotions and their average (negative affect). Acting on impulse was weakly negatively associated with both planning and persisting, which were moderately associated with each other. All negative emotions were moderately associated with each other, except for bored, which was weakly associated with the others. Acting on impulse was most strongly associated with negative emotions (Mean = .29) compared to planning (Mean= −.10) and persistence (Mean = −.15). The composite variable of negative affect was more strongly correlated with each facet of EMA impulsivity than any single emotion alone.

Figure 1:

Bivariate correlations between EMA emotions and EMA impulsive behaviors at the observation level. ‘Negative affect’ is the average of all negative emotions at a given observation. Shading of the boxes indicates relative strength of the association]

3.2. Confirmatory analyses: Negative affect predicting different impulsive states

Results of these three models can be found in Table 2. Below we describe the model-building process and the results of the final model for each impulsivity outcome.

Table 2:

Main Effects of Negative Affect Predicting EMA Impulsivity

	Acting on Impulse			Planning			Persistence
Between-persona	β	B	95% CI	β	b	95% CI	β	b	95% CI
Intercept		12.95	3.29 – 10.33		42.06	35.08 – 49.03		47.93	40.62 – 55.23
Sexb	.087	1.74	−0.71 – 4.2	.002	.04	−3.28 – 3.37	−.017	−0.45	−3.93 – 3.03
Ageb	−.033	−0.65	−2.21 – .09	−.025	−.62	−2.73 – 1.49	.026	0.67	−1.54 – 2.87
Sampleb	−.44	−8.85	−14.41– −3.29	.93	23.16	15.61–30.71	.79	20.35	12.44–28.25
Non-Whiteb	.15	3.01	0.63 – 5.38	−.039	−0.98	−4.21 – 2.25	−.18	−4.62	−8.00 – −1.24
Average Negative Affect	.28	0.46	0.34 – 0.57	−.024	−0.05	−0.2 – 0.11	−.099	−0.21	−0.38 – −0.05
Negative Urgency	.17	5.96	3.6 – 8.41	−.11	−4.83	−7.95 – −1.72	−.062	−2.83	−6.10 – 0.43
Within-persona
Time of Dayb	.009	0.14	0.02 – 0.25	.003	0.07	−0.10 – 0.23	.011	0.21	0.04 – 0.37
Weekendb	.042	0.57	−0.33 – 1.48	−.048	−0.82	−2.05 – 0.41	−.016	−0.18	−1.45 – 1.08
Observation #b	.001	0.03	−0.02 – 0.08	.001	0.01	−0.06 – 0.08	−.003	−0.05	−0.12 – 0.02
EMA Negative Affect	.16	0.20	0.15 – 0.24	−.013	−0.01	−0.07 – 0.04	−.049	−0.07	−0.13 – −0.02
Cross-levelc
Neg. Urgency × EMA Neg. Affect		0.06	−0.01 – 0.13		−0.07	−0.16 – 0.03		−0.01	−0.11 – 0.09

Note: effects with a p-value <.05 are presented in bold.

^aBetween-person standardized coefficients were calculated using only between-person variance in impulsivity outcomes, whereas within-person coefficients were calculated using only within-person variance.

^bStandardized coefficients for these covariates were calculated using unstandardized predictors and standardized outcomes because these predictors are not meaningful when standardized.

^cStandardized coefficients for interaction terms were not reported as they are not more meaningful when standardized.

3.2.1. Acting on Impulse

Model Fit

A model including a random effect for the slope of EMA negative affect on EMA acting on impulse improved model fit compared to the fixed-slope model (Δ BIC = −46.9, χ² (2) = 64.09, p < .001). This suggests that individuals’ association between negative affect and acting on impulse varies significantly around the mean. Including global self-report negative urgency as a moderator of this association decreased model fit as measured by BIC, and did not lead to a significantly improved likelihood (Δ BIC = 5.93, χ² (1) = 2.63, p = 0.10). The interaction coefficient was not significant.

Between Person Effects

For our main variables of interest, both average negative affect and global self-report negative urgency were associated with an individual’s between-person level of acting on impulse. Every 1 SD increase in average negative affect across the EMA was associated with a 0.28 SD increase in average acting on impulse. A 1 SD increase in global self-report negative urgency was associated with a 0.17 SD increase in average acting on impulse.

Within Person Effects

Within person negative affect was associated with acting on impulse, such that at observations when a participant reported negative affect 1 SD higher than their average, they also reported 0.16 SD higher acting on impulse. The random slope indicated that this effect varied across people, with 68% of participants’ effects falling in the range of β = −.01 to β = 0.33. In short, for some participants, the association between state-level negative affect and acting on impulse was essentially zero, or even slightly negative, while for others, the association was small to moderate (See Figure 2). However, global self-report negative urgency was not associated with the magnitude of the association between negative affect and acting on impulse.

Figure 2:

Distribution of Random Effects of Negative Affect Predicting Acting on Impulse. Bold lines show the average expected effect for an individual at −1 SD, average, and + 1 SD of the estimated distribution of state-level associations between EMA negative affect and EMA acting on impulse. These slopes are graphed on top of the data points for individuals from our sample whose within-person regression estimates fall within the bottom 16%, the middle 68%, and the top 16% of this distribution, respectively.]

3.2.2. Planning

Model Fit

The model including a random effect for the slope of EMA negative affect on EMA planning improved model fit compared to the fixed-slope model (Planning: Δ BIC = −4.26, χ²(2) = 21.30, p < .001). However, the addition of the interaction between global self-report negative urgency and momentary negative affect did not improve fit (Δ BIC = 6.53, χ²(1) = 2.0, p = 0.16).

Between-Person Effects

Global self-repot negative urgency was associated with an individual’s between-person level of planning. A 1 SD increase in global self-report negative urgency was associated with a .11 SD decrease in average planning. Average negative affect was not associated with planning.

Within-Person Effects

Within-person negative affect was not associated with planning at the same time point when averaged over all participants (a mean effect of approximately 0), yet the random slope model estimates that the magnitude of this effect varied across people, with 68% of participants’ effects falling in the range of β = −.17 and β = .15. Global self-report negative urgency did not predict variation in this effect.

3.2.3. Persisting

Model Fit

The model including a random effect for the slope of EMA negative affect on EMA persistence improved model fit compared to the fixed-slope models (Δ BIC = −19.98, χ²(2) = 37.01, p < .001). The addition of the interaction between trait negative urgency and momentary negative affect did not improve fit (Δ BIC = 8.46, χ²(1) = 0.06, p = 0.81.

Between-Person Effects

Participants who reported more negative affect across the EMAs reported lower levels of persisting. For every 1 SD increase in in average negative affect across the EMA, the model predicted a 0.10 SD decrease in average persisting. Global self-report negative urgency was unrelated to average persisting.

Within-Person Effects

Momentary negative affect was associated with slightly less persisting at the same time point. For a 1 SD increase in negative affect above an individual’s average, we estimated a 0.05 SD decrease in persisting. The random slope model estimated that the magnitude of this effect varies across people, with 68% of participants’ effects falling in the range of β = −0.13 to β = 0.23. Global self-report negative urgency did not predict variation in this effect.

3.3. Exploratory: Specific emotions predicting different impulsive states

There was variability in the associations across different specific emotion-by-impulsivity relations, with angry, irritable and unhappy having relatively stronger effects on acting on impulse than anxious or bored (See Table 3); these effects were not significant for persisting or planning. In the case of acting on impulse, no specific emotion-behavior association was stronger than the association with aggregated negative affect.

Table 3:

Effects of Specific Emotions on Impulsive Behaviors

	Main effect of emotion		Emotion X Neg. Urgency
Acting on Impulse	b	99.9% CI	b	99.9% CI
Negative Affect	0.20	0.12 – 0.27	0.06	−0.06 – 0.18
Angry	0.16	0.11 – 0.22	0.03	−0.07 – 0.12
Anxious	0.06	0.03 – 0.10	0.01	−0.05 – 0.08
Irritable	0.12	0.08 – 0.17	0.03	−0.05 – 0.11
Unhappy	0.13	0.07 – 0.18	0.04	−0.06 – 0.12
Bored	0.01	−0.03 – 0.05	0.04	−0.02 – 0.11
Planning	b	99.9% CI	b	99.9% CI
Negative Affect	−0.01	−0.10 – 0.08	−0.07	−0.23 – 0.09
Angry	−0.02	−0.08 – 0.05	0.03	−0.09 – 0.15
Anxious	0.01	−0.04 – 0.06	−0.05	−0.13 – 0.04
Irritable	−0.01	−0.07 – 0.04	0.00	−0.09 – 0.09
Unhappy	−0.03	−0.09 – 0.03	−0.02	−0.13 – 0.10
Bored	−0.01	−0.05 – 0.04	−0.08	−0.16 – 0.01
Persisting	b	99.9% CI	b	99.9% CI
Negative Affect	−0.07	−0.17 – −0.03	−0.01	−0.18 – 0.16
Angry	−0.05	−0.11 – 0.02	0.02	−0.10 – 0.14
Anxious	−0.02	−0.07 – 0.03	−0.02	−0.11 – 0.07
Irritable	−0.04	−0.09 – 0.02	−0.03	−0.12 – 0.07
Unhappy	−0.05	−0.12 – −0.02	0.01	−0.13 – 0.11
Bored	−0.01	−0.07 – 0.04	0.00	−0.10 – 0.09

Note: All models included sex, age, sample, race, global self-report negative urgency, and average EMA emotion as covariates. Effects with a p-value < .001 are presented in bold.

There was not strong evidence that including negative urgency as a moderator provided improvements in model fit for any specific emotions.

3.4. Highly exploratory analyses of alternative models

We tested whether global self-report negative urgency moderated the association between state-level negative affect and acting on impulse under four alternative models that posit that this association 1) only exists in those observations where negative affect is elevated, 2) only in the presence of a quadratic effect, 3) only when looking at the following time-point, and 4) only in the presence of other moderators. We tested these models using EMA acting on impulse as our only outcome. Across all models, we found no evidence meeting our criteria (p < .001) that global self-report negative urgency might be best expressed by one of these alternative formulations. Indeed, no effects crossed the threshold of even traditional significance tests.

4. Discussion

Trait negative urgency is an important predictor of psychopathology (Berg et al., 2015; Smith et al., 2007), yet very few studies have tested the underlying assumption that individuals high on negative urgency act more impulsively when experiencing negative affect. A few prior studies have suggested that individual differences in global self-reports of negative urgency were associated with a higher propensity towards impulsive behaviors using experimental and EMA methods (i.e. Chester et al., 2017; Sperry et al., 2018), but these studies did not use measures of state impulsivity that directly map onto global self-reports of negative urgency. Our results suggest that there is a consistent link between the momentary experience of negative emotions and acting on impulse and persistence, and that the strength of these associations varies across individuals. However, global self-reports of negative urgency were not associated with individual differences in the state-level association between negative emotions and any impulsive behaviors. In other words, we found no evidence that individuals higher in global self-report of negative urgency were more likely to report acting rashly while experiencing real-world negative affect.

State-level negative affect was weakly associated with both acting on impulse and persisting, but the association with planning was not different from zero. When individuals reported higher levels of negative affect than what was usual for them, they also reported that they acted on impulse more and persisted in the face of difficulty less. This extends prior work which used unidimensional measures of state impulsivity (Sharpe et al., 2019; Sperry et al., 2018), and suggest that the effects of negative emotions on impulsive behavior may be more specific to “hot” forms of impulse control, such as controlling impulses and continuing on in the face of adversity, rather than “cool” impulse control, such as thinking ahead and careful planning. Although these effects are small, they could accumulate over longer periods of time to produce substantial impairments in impulse control. These effects are impressive especially considering that they were detectable within the normal range of emotional experience and through the multitude of uncontrolled variables present in real-world scenarios. It is also important to note that no other combination of negative emotion or impulsivity produced a larger main effect than that of average negative affect and acting on impulse, and none of our highly exploratory variations on our primary model produced meaningfully different results. These findings give us confidence that the association between negative affect and acting on impulse, as hypothesized, is the strongest ‘signal’ in our data that reflects the process described by theories of negative urgency.

We conceptualized the state-level association between negative affect and impulsive behaviors as a reflection of the construct of negative urgency, similar to prior studies (Sperry et al., 2018; Sharpe et al., 2019). Although the association between negative affect and impulsive behaviors varied from person-to-person, we did not find evidence that global self-reports of negative urgency explained variation in the moment-by-moment relations between negative affect and impulsive behaviors, regardless of the model we tested. In the current study, we were powered to detect relatively robust interactions (b = .17), and it is certainly possible that smaller interactions were present that we were unable to detect with the current study design. It may be that the effect we are trying to characterize may be small and difficult to detect in real-world settings where a number of factors other than mood may evoke or constrain impulsive behaviors, such as the opportunity to engage in impulsive behaviors, or the magnitude of mood fluctuations. It may also be that relatively rare instances of acting on impulse in response to very strong negative emotions best characterizes negative urgency; if sufficiently impactful, these events may not need to be frequent or consistent to be highly salient, which makes them in turn more difficult to detect. Clearly, replication with larger samples would help resolve the differences across our study and prior work.

Regardless of whether global self-reports of negative urgency map on to participants’ real-time emotion-behavior relations, we did observe individual differences in these relations. Future studies should aim to understand how EMA ‘signals’ of the negative urgency process relate to psychological outcomes in order to unpack what aspects of the construct increase risk for these issues. It could be that the EMA-measured negative urgency ‘signal’ we observed in our data actually predicts variation in negative outcomes independent of global self-reports of negative urgency. If not, it would suggest that aspects related specifically to global self-report of urgent behavior are driving the connection between negative urgency and psychopathology.

In our data, average EMA reported negative affect and global self-report of negative urgency were both associated with EMA acting on impulse. In short, people who have more negative emotions in general act on impulse more in general. Likewise, people who say they act impulsively when upset are more likely to act impulsively in general. Further, average negative affect and global self-report of negative urgency were moderately correlated in our sample (r = .43). These results paint a possible picture of individuals who are generally more emotional and more impulsive, and who may be more likely to endorse global negative urgency items because they reflect both of these attributes. As the construct of negative urgency has gained research attention, the emphasis on state negative affect leading to impulsive action has strengthened as a proposed explanation for the process that this trait represents. However, our results highlight the potential that the temporal co-occurrence of these experiences may be over-stated, and that global self-report negative urgency may partially reflect a trait-level co-occurrence of poor emotional and behavioral self-regulation, and not solely a process-level interaction between the two.

An alternative explanation for our results is that the construct of negative urgency may capture a more complex process than what we have modeled. Some correlational studies have shown that the relationship between negative urgency and impulsive behaviors may depend on higher-order cognitive processes such as beliefs and expectancies about the function of impulsive actions to reduce emotional distress. Adams, Kaiser, Lynam, Charnigo, & Milich (2012) found that negative urgency’s relationship to drinking problems was mediated by both coping and enhancement motives. Fischer, Anderson, & Smith (2004) found that the relationship between trait urgency and binge eating was moderated by expectancies, such that individuals with high urgency and positive expectancies for binging had the most binge days, and those with high urgency and low expectancies had the fewest. Results from studies like these indicate that an individual’s motivations for engaging in impulsive action are likely an important aspect of how negative urgency creates risk for negative outcomes. Future studies could integrate cognitive constructs such as beliefs, motives, and expectancies into process models of urgency to see if this helps connect the construct of trait negative urgency to real-world reactions to negative affect.

The present study had several key limitations. Our sample was limited to high school and college students, which may not generalize to other populations. Further, our college sample consisted of at least weekly alcohol drinkers or marijuana users. This may have over-selected for individuals higher in baseline negative urgency compared to the general population. As a result, the mean levels presented in our analyses may be higher than what would be expected from a truly random sample. Lastly, it is possible that interactions between negative emotion and impulsivity often occur on a shorter timescale than what we were able to capture. Likewise, within-person patterns may change over time. Future studies may wish to sample more frequently, and to collect data on the same individuals over many weeks to ensure more accurate measurement of state-level dynamics.

5. Conclusion

Overall, we failed to validate that global self-report of negative urgency captures individual differences in how impulsively someone responds to negative emotions. However, we did identify and characterize individual differences in the strength of the effect of negative affect on multiple facets of impulsivity in an ecologically valid context. The potential significance of these individual differences could be a fruitful area of future investigation, as is the further exploration of how trait negative urgency creates risk for psychopathology at the within-person level.

Highlights.

• Negative affect predicts acting on impulse in day-to-day life.

• Negative affect predicts impulsivity better than specific emotions do.

• Negative urgency does not moderate the effect of negative affect on impulsivity

Appendix 1: EMA Impulsivity Items

Acting on Impulse

1. It was hard for me to resist acting on my feelings.

2. I had trouble controlling my impulses.

3. I felt like I couldn’t stop myself from going overboard.

4. I got involved in something I later wished I could get out of.

5. I lost control

6. Others were shocked or worried about the things I did.

Planning

1. I thought carefully before doing anything.

2. I followed a rational, “sensible” approach to things.

3. Before making up my mind, I considered all the advantages and disadvantages.

4. I acted without thinking.

Persistence

1. I saw things through to the end.

2. I concentrated easily.

3. I finished what I started.

4. I gave up easily

Note: all items were responded to on a visual analog slider scale with end anchors of ‘not at all’ to ‘very much.’ Placement of the slider was scored 0–100 based on location, though numbers were not displayed to participants.