This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Mental Health, is properly cited. The complete bibliographic information, a link to the original publication on http://mental.jmir.org/, as well as this copyright and license information must be included.
Behavior and emotions are closely intertwined. The relationship between behavior and emotions might be particularly important in populations of underserved people, such as people with physical or mental health issues. We used ecological momentary assessment (EMA) to examine the relationship between emotional state and other characteristics among people with a history of chronic homelessness who were participating in a health coaching program.
The goal of this study was to identify relationships between daily emotional states (valence and arousal) shortly after waking and behavioral variables such as physical activity, diet, social interaction, medication compliance, and tobacco usage the prior day, controlling for demographic characteristics.
Participants in m.chat, a technology-assisted health coaching program, were recruited from housing agencies in Fort Worth, Texas, United States. All participants had a history of chronic homelessness and reported at least one mental health condition. We asked a subset of participants to complete daily EMAs of emotions and other behaviors. From the circumplex model of affect, the EMA included 9 questions related to the current emotional state of the participant (happy, frustrated, sad, worried, restless, excited, calm, bored, and sluggish). The responses were used to calculate two composite scores for valence and arousal.
Nonwhites reported higher scores for both valence and arousal, but not at a statistically significant level after correcting for multiple testing. Among momentary predictors, greater time spent in one-on-one interactions, greater time spent in physical activities, a greater number of servings of fruits and vegetables, greater time spent interacting in a one-on-one setting as well as adherence to prescribed medication the previous day were generally associated with higher scores for both valence and arousal, and statistical significance was achieved in most cases. Number of cigarettes smoked the previous day was generally associated with lower scores on both valence and arousal, although statistical significance was achieved for valence only when correcting for multiple testing.
This study provides an important glimpse into factors that predict morning emotions among people with mental health issues and a history of chronic homelessness. Behaviors considered to be positive (eg, physical activity and consumption of fruits and vegetables) generally enhanced positive affect and restrained negative affect the following morning. The opposite was true for behaviors such as smoking, which are considered to be negative.
More than half a million individuals are homeless at any given time in the United States [
Mood and emotional reactivity play an important role in both mental and physical health. For example, Gallo and Matthews [
Ecological momentary assessment (EMA) techniques use mobile devices to assess thoughts, feelings, and behaviors in real-time in an individual’s natural setting [
Although EMA has been used to evaluate dynamic changes in mood and behavior, no study to date has examined the relationship between emotions and behavior among adults in PSH. The purpose of this study was to explore the prospective associations between emotions (ie, valence and arousal) and health behaviors among adults residing in PSH using EMA. Considering the relatively high costs associated with physical and mental health disorders in this population, it can be beneficial to identify factors affecting daily emotion patterns in order to predict and intervene with persons who are at risk.
We obtained data for this study from the Mobile Community Health Assistance for Tenants (m.chat) project, a technology- assisted health coaching intervention designed to improve health indicators among PSH residents in Fort Worth, Texas [
The Institutional Review Board of the University of North Texas Health Science Center approved this project, and we assured participants of confidentiality. All participants gave informed consent.
For analysis, we included 155 participants who completed a total of 18,357 daily assessments between May 1, 2016, and April 30, 2017. On average, individuals received 139 daily assessments or prompts (range 14-334) and completed 106 assessments (range 4-322). The sample was split almost evenly between males (n=77) and females (n=78), and the average age was 52 (SD 8) years.
The mobile app alerted participants to complete an assessment 30 minutes after the participant’s self-reported waking time. We asked the participants to complete the assessment within 30 minutes of the initial alert; they had the option to “snooze” an assessment request 3 times each day before the EMA would be counted as missed. Below are the questions that were presented in the daily EMA (we have only presented the questions or response options considered in the analyses).
We measured 9 emotions items on a Likert-type scale from 1 (strongly disagree) to 5 (strongly agree): I feel happy, I feel frustrated, I feel sad, I feel worried, I feel restless, I feel excited, I feel calm, I feel bored, and I feel sluggish.
Participants were asked how many hours they spent sitting, how many minutes they walked or biked to get somewhere, how many minutes they were physically active for fitness (eg, running or sports), and how many minutes they were physically active at work or home (eg, cleaning, lifting, or carrying things) the previous day.
We asked participants how many servings of fruits and vegetables they ate, how many sugar-sweetened beverages they drank, and how many desserts and other sweets they ate the previous day.
We asked the participants about total minutes they spent in meaningful one-on-one conversations with other people and the total minutes they spent in meaningful group interactions (eg, going to church, participating in an exercise class, or other social occasions) the previous day.
We asked participants whether they took all of their medication as prescribed the previous day.
We asked participants whether they used tobacco (cigarettes) the previous day, and if so, how many cigarettes they smoked.
Demographic characteristics such as age, sex, and race (white or nonwhite), collected at baseline, were used as covariates in the analyses.
The circumplex model of affect [
Circumplex model of affect.
Circumplex scores for the emotions considered.
Emotion | Valence | Arousal |
Happy | 0.95 | 0.15 |
Frustrated | −0.50 | 0.40 |
Sad | −0.95 | −0.40 |
Worried | −0.15 | −0.30 |
Restless | −0.15 | 0.30 |
Excited | 0.70 | 0.70 |
Calm | 0.75 | −0.70 |
Bored | −0.40 | −0.80 |
Sluggish | −0.15 | −0.50 |
Circumplex model in Cartesian coordinate system.
Descriptive statistics for the 9 emotion outcomes.
Variable | Mean | Between-subject SD | Mean within-subject SD |
Happy | 3.54 | 0.74 | 0.70 |
Sad | 2.55 | 0.79 | 0.79 |
Restless | 2.59 | 0.83 | 0.70 |
Excited | 3.09 | 0.78 | 0.70 |
Calm | 3.42 | 0.71 | 0.67 |
Sluggish | 2.71 | 0.92 | 0.74 |
Frustrated | 2.55 | 0.79 | 0.79 |
Worried | 2.60 | 0.85 | 0.74 |
Bored | 2.41 | 0.80 | 0.66 |
Descriptive statistics for the quantitative momentary predictors.
Variable | Mean | Between-subject SD | Mean within-subject SD |
Total physical activity | 34.62 | 24.67 | 20.78 |
Minutes of one-on-one interaction | 79.97 | 49.28 | 48.94 |
Minutes spent in group interaction | 44.41 | 32.74 | 36.97 |
Hours spent sitting | 5.47 | 1.93 | 2.13 |
Fruits and vegetables | 2.80 | 1.55 | 1.23 |
Sweets | 2.60 | 1.67 | 1.13 |
Number of cigarettes | 3.49 | 5.12 | 1.66 |
In that sense, the predictors are not strictly momentary, but will be referred to as momentary variables for statistical modeling and analysis. Individual demographic characteristics (ie, age at the onset of the EMA study, sex, and race) were considered time invariant for the duration of the study. Race was dichotomized as white and nonwhite, as 94.5% (147/155) of participants were either white or African American individuals.
In the general statistical model for the analysis, for each outcome, we denote the response on the
All analyses were performed using MIXED procedure in SAS (SAS Institute) with the intercept specified as a random effect and within-subject residuals specified to have a first-order autoregressive correlation.
Since there are eleven predictors in our model, we implemented the popular Bonferroni correction to adjust the reported
Analyses of the associations between momentary variables and valence and arousal were performed, controlling for the 3 demographic predictors of age, sex, and race. The results for the valance and arousal outcomes are presented in
Minutes spent doing physical activity the previous day was a statistically significant predictor of both valence and arousal, with expected higher scores for increased physical activity. Time spent in meaningful group interaction the previous day was not a statistically significant predictor of either valence or arousal. Time spent in meaningful one-on-one social interaction the previous day was a statistically significant predictor of both valence and arousal, with expected higher scores for more interaction time. Hours spent sitting the previous day was a statistically significant predictor of both valence and arousal, with an expected lower score for an increase in time spent sitting. Number of total servings of fruits and vegetables consumed the previous day was a statistically significant predictor of both valence and arousal, with expected higher scores for greater servings. Number of total servings of sugar-sweetened beverages and desserts the previous day was not a statistically significant predictor of either valence or arousal. Adherence to medication the previous day was a statistically significant predictor of both valence and arousal, with higher scores for adherence. Any tobacco usage the previous day was a statistically significant predictor of only valence; on average, smoking a higher number of cigarettes resulted in lower valence scores.
Results for valence with momentary predictors, controlling for demographic characteristics.
Effect | Estimate | SE | ||
Intercept | 2.32 | 1.422 | 1.63 |
.10 |
Age | −0.008 | 0.027 | −0.28 |
.78 |
Male | −0.49 | 0.459 | −1.07 |
.28 |
Caucasian | −1.23 | 0.457 | −2.69 |
.007 (.08) |
Total physical activity | 0.007 | 0.0008 | 8.80 |
<.001 (<.001) |
Minutes of one-on-one interaction | 0.004 | 0.0004 | 9.53 |
<.001 (<.001) |
Minutes spent in group interaction | 0.00008 | 0.0005 | 0.17 |
.87 |
Hours spent sitting | −0.04 | 0.011 | −3.42 |
<.001 (.007) |
Fruits and vegetables | 0.12 | 0.014 | 8.54 |
<.001 (<.001) |
Sweets | 0.02 | 0.016 | 1.34 |
.18 |
Medication | 0.76 | 0.105 | 7.23 |
<.001 (<.001) |
Number of cigarettes | −0.06 | 0.008 | −7.99 |
<.001 (<.001) |
aAll the
bBonferroni-corrected
Results for arousal with momentary predictors and controlling for demographic characteristics.
Effect | Estimate | SE | ||
Intercept | −2.97 | 0.520 | −5.72 |
<.001 |
Age | −0.0001 | 0.010 | −0.01 |
.99 |
Male | −0.17 | 0.167 | −1.04 |
.30 |
Caucasian | −0.42 | 0.167 | −2.50 |
.01 (.13) |
Total physical activity | 0.003 | 0.0004 | 7.40 |
<.0001 (<.001) |
Minutes of one-on-one interaction | 0.002 | 0.0002 | 11.17 |
<.0001 (<.001) |
Minutes spent in group interaction | 0.0003 | 0.0002 | 1.67 |
.09 |
Hours spent sitting | −0.02 | 0.005 | −4.20 |
<.001 (<.001) |
Fruits and vegetables | 0.03 | 0.006 | 4.07 |
<.001 (<.001) |
Sweets | −0.009 | 0.007 | −1.36 |
.17 |
Medication | 0.15 | 0.045 | 3.26 |
.001 (.01) |
Number of cigarettes | −0.006 | 0.003 | −1.88 |
.06 |
aAll the
bBonferroni-corrected
Even though the model controls for demographic characteristics in analyzing the effect of momentary variables on valence and arousal, it is worthwhile to explore how much influence the demographic predictors have on the momentary predictors. A strong influence of demographic predictors on the momentary predictors can make the regression coefficients unstable and hard to interpret. Unlike in a standard multiple regression framework, in our hierarchical model, the influence cannot be measured directly by studying the multicollinearity properties and other standard regression diagnostics. Instead, the amount of influence can be indirectly measured by analyzing two additional models: one with only demographic predictors and one with only momentary predictors. The change in values of the estimated regression coefficients in the full models compared with the two isolated models described above can be used to assess the influence and the robustness of the coefficients.
For the sake of brevity, we did not present the actual results from the two isolated models here, but the results are remarkably consistent with our findings from the combined model in the previous section. Not only do the statistical significances of the momentary predictors match closely but also the individual estimates of the regression coefficients are surprisingly close. The individual estimates of the regression coefficients are very close for the demographic predictors as well. The observed consistency provides fairly strong evidence on the orthogonality of the demographic predictors from the momentary predictors.
These findings provide an important glimpse into factors that affect valence and arousal in a population of individuals residing in supportive housing. To our knowledge, this is the first study to examine the connection between emotions and other factors among people with mental health disorders and a history of chronic homelessness. This underserved population is often excluded from research studies due to co-occurring mental and physical disorders, resulting in substantial gaps in our understanding of their health and health behaviors.
Our analyses provide a number of observations about the relationships between health behaviors and subsequent emotions. First, we found that physical activity was significantly associated with positive emotions the following day. This finding is consistent with the literature showing the association of moderate physical activity with improved and maintained mood [
We also found that smoking cigarettes had a negative effect on valence the following day. Although nicotine may have a calming effect due to the inhibition of negative emotions such as anger [
We also found a strong relationship between the amount of time spent in individual social interactions and emotions. Interestingly, “time spent on meaningful one-on-one social interaction the previous day” was strongly associated with arousal and valence, while the “amount of time spent interacting in a group setting” was not significantly associated with emotions. This finding was unexpected given the substantial evidence that social support predicts the quality of life in many areas [
In the analysis of the effect of demographics on the association between momentary predictors with valence or arousal, demographic variables had minimal effects on the regression coefficients of the momentary predictors, even when statistically significant. Hence, it is reasonable to conclude that the demographic predictors operated almost independently of the momentary variables in terms of influencing emotions.
Finally, it would be possible to study the association of emotional affect with subsequent same-day behaviors, for instance, examining the effects of emotions in the morning on smoking or drinking later in the day. Such analyses are beyond the scope of this paper, but we plan to investigate these associations in a future manuscript.
Our study had a number of limitations. Notably, our protocol included only daily morning assessments. Thus, we were not able to examine within-day variability. However, unlike other EMA studies, which typically run for a few days to, at most, a few weeks, our study ran up to 334 days with an average of 156 days of monitoring among all participants. This allowed us to examine associations for a much longer period than most other EMA studies. In addition, our results are generalizable only to a population of individuals residing in PSH with a history of homelessness and mental health issues. It is unclear whether the findings are generalizable to other groups of people with mental health problems, let alone the population in general. Relatedly, all participants self-reported depression or a mental health condition at baseline, which may have affected emotions and mood independently of other behavioral measures. For instance, the average client reported a score of 12.62 on the Patient Health Questionnaire, indicating that most clients felt at least moderate levels of depression upon admission to the program. Finally, we cannot rule out the possibility that participating in the coaching intervention affected the relationship between behaviors and emotion. Our results must be interpreted in the context of the larger services that people were receiving in this program. Further study with a more diverse population is necessary to make any broader assertions.
Despite the limitations, our study offers an important glimpse into health behaviors that affect daily emotional arousal and valence of persons with a history of chronic homelessness and mental health problems. One of the goals of the m.chat program was to provide individual support and assistance in meeting health goals. Because mood was an important target of the program, identifying factors that predicted positive affect can help improve future iterations of programs like this. To that end, identifying modifiable behaviors associated with negative and positive moods is a first step toward improving stability and preventing future homelessness. Understanding factors associated with mood and behaviors, particularly in vulnerable populations such as formerly homeless individuals, can also help providers design more targeted treatment plans and provide more appropriate referrals to ancillary care services [
Notably, many of our findings are consistent with “common wisdom” drawn from other populations. Behaviors generally considered to be positive (eg, physical activities, consumption of fruits and vegetables, adherence to prescribed medication, and one-on-one social interaction) tended to enhance positive affect and restrain negative affect. The opposite was true for behaviors considered to be negative, such as smoking. In fact, the positive and negative impacts on physical health for most of these behaviors are well established, and it is noteworthy that their effects on positive and negative affect appear to be consistent with previous literature with other populations. In a separate analysis of m.chat program data, Holmes et al [
ecological momentary assessment
permanent supportive housing
Funding for this study was provided through a Medicaid 1115 Waiver to the State of Texas. Centers for Medicare and Medicaid Services had no role in the study design, collection, analysis, or interpretation of the data, writing the manuscript, or the decision to submit the paper for publication.
MSB was a paid consultant on the parent grant for this study.