Daily Instrumental and Affective Attitudes About Exercise: An Ecological Momentary Assessment Study

Abstract

Background

Daily decisions to exercise may be influenced by day-to-day changes in affective attitudes (AA) and instrumental attitudes (IA) toward exercise. However, the within-day association between AA, IA, and exercise behavior has received little attention.

Purpose

To examine the effects of more temporally proximal (daily) AA and IA on daily exercise behavior beyond traditionally assessed distal (at the beginning of an exercise program) AA and IA.

Methods

In the context of a 3-month exercise promotion program (N = 50), distal AA and IA were assessed at baseline. Ecological momentary assessment was used to assess proximal AA, IA, and exercise each day.

Results

Between-subject differences in distal AA (OR = 1.28, p = .03) and distal IA (OR = 1.34, p = .01) were predictive of average likelihood of exercise each day over the 3-month period. Within-subject differences in proximal AA (OR = 1.19, p = .007), but not proximal IA (OR = 1.11, p = .18), predicted exercise each day beyond the between-subjects effects of distal AA and IA. Exploratory analysis revealed an interaction, such that the within-subjects impact of proximal AA on daily exercise was most evident among individuals who held more negative distal AA at baseline (OR = 0.80, p < .0001).

Conclusions

Attitude type (affective versus instrumental) and temporality (distal versus proximal) are important to consider in attempts to predict and understand exercise behavior. In addition to targeting change in distal attitudes, exercise interventions should target changes in daily AA to impact exercise later in the same day.

Daily fluctuations in attitudes about whether exercise is pleasant had more impact on exercise behavior than attitudes about whether exercise is beneficial among adults participating in a physical activity promotion program

Introduction

Physical inactivity is a modifiable lifestyle risk factor for the leading causes of chronic disease in the USA [1] and globally [2]. Regular physical activity is associated with reduced risk of cardiovascular disease [3], type 2 diabetes [4], cancers of the breast [5], colon [6], and endometrium [7], and all-cause mortality [8]. Despite known health benefits, fewer than 20% of adults in North America achieve the recommended 150 min of moderate intensity physical activity per week [9, 10]. Better understanding of the factors that contribute to physical activity is needed to inform physical activity promotion interventions.

Attitudes are one construct frequently studied in the context of physical activity behavior [11]. In the Theory of Planned Behavior, attitudes toward a behavior are the degree to which a person holds a positive or negative evaluation of a behavior [12]. Attitudes represent an aggregation of salient behavioral consequences and the corresponding subjective value of the outcome [13] and can be categorized by whether they are affective (i.e., feeling-based) or instrumental (i.e., utility-based) in nature [14, 15]. For example, an affective attitude (AA) may reflect the belief that exercise is pleasant versus unpleasant, or enjoyable versus unenjoyable, and is distinct from more specific affective reactions (e.g., anticipated regret; see [16]). An instrumental attitude (IA) may reflect the belief that exercise is useful versus useless, or beneficial versus harmful. Across a large body of work, AA is a stronger predictor of physical activity than IA [11, 17–20]. For example, in a meta-analysis of 38 studies, AA had a stronger association with physical activity than IA (r = 0.42 and 0.25, respectively) [20]. Thus, the need to conceptualize and assess distinct AA and IA components has been established [21].

Typically, attitudes are measured at the start of a study and used to predict physical activity that takes place over the following weeks or months [22–24]. However, the association between attitudes and behavior may change to the extent that attitudes are context dependent, or change over time [25]. Indeed, a large meta-analysis of 103 prospective studies showed that attitudes were more predictive of physical activity when the latency between attitude and physical activity assessments was shorter (<5 weeks), compared to longer latencies (>5 weeks), with a minimum latency of 7 days among the studies reviewed [11]. One explanation for this finding is that attitudes may be changing over time; therefore their relationship with physical activity depends on the timing of the attitude assessment relative to physical activity behavior. Indeed, other social cognitive beliefs such as outcome expectancies, self-efficacy, and intentions related to physical activity have been shown to fluctuate on a daily basis and predict same-day physical activity [26–28]. Though not examined as yet, it is reasonable to expect that there may also be daily fluctuations in attitudes that are predictive of physical activity.

Daily fluctuations in attitudes may differ for AA versus IA. Current circumstances influence how individuals imagine, or perceive a future event (e.g., exercise) and therefore can influence the desire or importance of that event [29]. There are a number of visceral factors that influence decision-making, such as present emotions (e.g., fear), drive states (e.g., hunger) and physical states (e.g., pain) [24]. Fluctuations in these visceral factors may be more likely to influence AA relative to IA. For example, daily differences in feelings of fatigue may impact how enjoyable a person believes physical activity will be. On the other hand, situational factors may have less of an impact on IA because visceral and emotional factors are less likely to influence the salience and value of utility-based outcomes. In the same example, changes in fatigue may not change a person’s attitude that physical activity is important (e.g., for weight loss or overall health). Nonetheless, some situational factors such as a reminder of the risk for diabetes from a television commercial or conversation with a family member could result in modification of a person’s attitude about the instrumental health benefits of physical activity. Thus, it is important to have a better understanding of the differences in daily fluctuations of AA versus IA and whether those fluctuations predict physical activity behavior each day.

Prior studies have examined the association between attitudes and physical activity on a between-person level, but have not examined within-person associations. Between-person or group-level analyses can be used to examine the extent to which people who hold more positive attitudes also participate in more physical activity. However, between-person effects do not provide information about whether day-to-day changes in a person’s attitudes will predict their likelihood of physical activity on that day. Researchers have argued persuasively for the need to test within-person effects in the physical activity domain [30], among other areas [31], because failure to differentiate between-person and within-person effects may lead to inaccurate models for behavioral prediction [32]. Given the capabilities of technology-based interventions to deliver behavior change assistance at any moment, a better understanding of the within-person associations between attitudes and physical activity can inform future interventions on when it may be best to intervene (e.g., just-in-time intervention).

Mobile technology is a useful tool when the goal is to collect information about attitudes and physical activity as they co-occur in real-time. Ecological momentary assessment (EMA) is one such strategy that is used to collect data about a person’s beliefs and behaviors as they unfold throughout the day. While the use of EMA is rapidly growing throughout the physical activity literature [30, 33], EMA has yet to be used to understand the within-person effects of AA versus IA on physical activity behavior.

The Present Study

This study extends prior work that has compared AA and IA in longitudinal, between-person models [11, 17] by examining between-person and within-person effects of distal (measured at baseline) and proximal (measured daily) AA and IA on physical activity behavior over a 3-month period. EMA was used to assess attitudes each morning and to capture physical activity in real-time via self-report. Physical activity was defined as the daily decision to engage (or not) in a bout of purposeful exercise because we were interested in how daily fluctuations in attitudes relate to intentional exercise behavior.

We first hypothesized that, consistent with prior research [11, 17] distal AA would be a stronger predictor of exercise behavior than distal IA. Second, we hypothesized that proximal attitudes would predict exercise behavior above and beyond distal attitudes and that proximal AA would be more predictive of behavior than proximal IA. Finally, in an exploratory analysis, we considered the effects of four putative interactions on exercise behavior, including interactions across attitude type within assessment time point (distal AA × distal IA; proximal AA × proximal IA) and interactions within attitude type across time point (distal AA × proximal AA; distal IA × proximal IA).

Methods

Participants

Participants were recruited into a 3-month exercise promotion program using flyers and online (e.g., Facebook) advertisements. Eligibility criteria included (a) ages 18–65, (b) body mass index (BMI) 18.5 to 40.0 kg/m², (c) inactive to low-active (<60 min/week) participation in structured exercise, and (d) ability to walk for exercise. Exclusion criteria included chronic conditions of the heart (e.g., hypertension) and lungs (e.g., COPD, uncontrolled asthma), the presence of substance use disorders, and other physical limitations (e.g., orthopedic problems).

Study Design and Intervention

This was a prospective observational study in which the primary purpose of the study was to examine associations between attitudes and exercise behavior. Participants were asked to report their attitudes and exercise behavior at baseline and as part of an EMA protocol delivered through a study-related application on their personal cell phones throughout the 3-month exercise promotion program. Each participant attended two sessions at the Brown University exercise laboratories at the beginning of the study to complete baseline assessments of demographics, height, weight, and distal affective and instrumental attitudes, receive intervention materials and to learn to use the EMA application. The protocol for this study was approved by the Brown University IRB.

All participants received print-based exercise promotion materials for the purpose of facilitating change in exercise behavior and thus improve ability to examine the hypothesized relationships between attitudes and exercise behavior. Specifically, participants received three packets (baseline, beginning of weeks 6 and 11), which included general information on overcoming obstacles to exercise. The first packet is distributed by the Weight Control Information Network within the National Institute of Diabetes and Digestive and Kidney Diseases [34]. The manual includes tips for getting ready to start an exercise program, determining which exercises to do, setting an exercise schedule, preparing for known barriers to exercise, and reminders to reward oneself for achieving exercise goals. The second and third packets were from the American College of Sports Medicine Fit Society® Page [35, 36]. These packets included tips to build motivation for exercise, for strength training and flexibility, making exercise fun, and for making exercise a family activity.

Participants also received three exercise counseling sessions to coincide with the print-based materials at baseline (in-person), beginning of weeks 6 and 11 (via phone call). At each session, research staff encouraged exercise, reviewed goal progress, set new goals, and addressed barriers to exercise. Participants were asked to create exercise goals for the next two weeks based on the SMART criteria (specific, measurable, achievable, realistic and timely; [37]). Regarding mode of exercise, participants were encouraged to explore activities that would be most likely to help them achieve the recommended activity levels depending on their access to equipment (e.g., at-home bikes or weights) and space (e.g., walking paths). Regarding frequency and duration, all participants were instructed to target one 30–60-min moderate intensity exercise session at least 5 days per week (or equivalent), for a goal of 150–300 min/week of exercise. Regarding intensity, all participants were given a heart rate range for moderate intensity: 64–76% of their estimated maximal heart rate (i.e., 220-age) for instances when they could monitor their heart rate (e.g., on exercise equipment). Participants were given two additional ways to monitor their exercise intensity: (a) staying within the 12–14 range of the perceived exertion scale [38], and (b) use of the “talk test” (i.e., should be able to carry on a conversation while exercising). It was recommend that participants start out slowly to achieve an intensity that is at the lower end of their prescribed range and then gradually increase their intensity to the middle of the prescribed range.

EMA Protocol

The present study includes daily EMA reports of AA, IA, and exercise behavior over the 3-month study. AA and IA were assessed in the context of morning reports, which participants were instructed to initiate each morning as soon as they awoke, and received automated reminders until 12:00pm (noon) if they had not yet completed one for that day. Participants were also instructed to initiate begin exercise reports upon starting a bout of exercise, and end exercise reports when they finished exercising. Exercise behavior was also retrospectively assessed in the morning report. Incentives were offered at the end of the study to participants who reached high EMA compliance rates ($20 for >70% of daily morning reports completed). In order to avoid over reporting of exercise, no incentive was provided for completing begin exercise or end exercise reports.

Measures

Demographics

At the baseline visit, participants self-reported demographic information, including age, gender, race, ethnicity, employment and marital status.

Height, weight, and BMI

At the baseline visit, research staff measured height and weight using a stadiometer and a calibrated digital scale (Detecto Pro Doc PD 300). Participants were asked to remove shoes and any objects from their pockets. BMI was computed by dividing weight in kilograms by height in meters squared.

Distal affective and instrumental attitudes

Distal AA and IA were assessed at baseline via a written questionnaire with three instrumental (beneficial/harmful, useful/useless, valuable/worthless) and three affective (enjoyable/unenjoyable, pleasant/unpleasant, invigorating/exhausting) items [14, 39] anchored by not at all (0) to extremely (6) using a 7-point semantic-differential scale [40] following the prompt, For me, regular exercise over the next 3 months would be. The average of the three affective and three instrumental items were used in analyses.

Proximal affective and instrumental attitudes

The same AA and IA items from the baseline questionnaire were assessed each morning during the 3-month study as part of the EMA morning reports with the prompt, For me, exercising today would be. The average of the three affective and three instrumental items were computed for each day and used in analyses.

Exercise behavior

Exercise sessions of at least 10 min were assessed by subtracting the time-stamp of each end exercise report from the time-stamp of the corresponding begin exercise report. We also used items from the morning reports to retrospectively assess exercise in instances where participants did not report their exercise in real time. Specifically, in each morning report, participants were asked whether they had exercised the day before by responding “yes” or “no” to the question, Did you exercise yesterday? If they said yes, they were asked whether they entered the exercise session in real time on their phone app. If they indicated that they did not, they were asked about the time and duration of the previous day’s exercise session. For this study, exercise behavior was operationalized as a dichotomous variable that represented whether or not there was at least one exercise session (≥10 min) each day (0 = no exercise, 1 = exercise).

Data Analysis

In preliminary analyses, a linear mixed model with a random intercept was used to examine the number of exercise reports per week. T-test analyses were used to examine differences between distal AA versus IA. The degree of variability for proximal (daily) AA and IA was computed for each participant as variance (s²_{within-person}). We examined associations between baseline factors (BMI, age, gender) and within-person variability in both attitudes using correlations for continuous variables and t-tests for categorical variables. A linear mixed model with a random intercept was used to describe average changed over time in AA and IA. Finally, we examined univariate relations between AA (distal and proximal) and exercise behavior and between IA (distal and proximal) and exercise behavior prior to testing hypotheses in multivariable analyses.

For the primary analyses, generalized estimating equations were used (PROC GENMOD, SAS v.9.3) to account for the clustered nature of EMA data. A generalized linear model with a logit link function was used, with a binomial distribution for the outcome variable. Data were nested such that daily outcomes were clustered within week within participant with standard errors adjusted for repeated observations over time by participant. An autoregressive covariance structure for residual errors was used to account for the interdependence of the repeated outcome assessments within-person. Distal (baseline) AA and IA scores were grand-mean centered and standardized, and represented a between-person predictor. AA and IA scores from daily assessments were person-mean centered to examine within-person processes, and standardized. Using a person-mean centered approach allows us to examine whether variance from an individual’s average in AA versus IA was more predictive of exercise.

Covariates were tested for inclusion in the model to reduce bias in estimates of the coefficients of interest using QIC (quasi-likelihood under the independence model criterion) [41]. We tested the following covariates in a forward stepwise fashion: previous day exercise (yes or no), day of week (weekend versus weekday), season (winter versus summer, to control for potential effects of weather on activity level [42]), BMI, age, and gender. Age and gender did not improve model fit and thus were excluded from the final models. Therefore, the final model controlled for previous day’s exercise, day of week, season, and BMI, based on improved model fit criteria (lower QIC).

We computed odds ratios (OR) from the regression coefficients (and corresponding 95% confidence intervals) to compare the effects of distal AA versus IA and proximal AA versus IA on exercise behavior. We first entered distal (baseline) AA and IA into the model to examine whether AA were more predictive than IA (hypothesis 1). Next, we added proximal (daily) AA and IA to examine whether they predicted exercise over and above distal AA and IA and whether proximal AA was more predictive of exercise than proximal IA (hypothesis 2). We then individually tested the interactions between attitude types within each assessment time point (distal AA × distal IA; proximal AA × proximal IA) and interactions between assessment time points within each attitude type (distal AA × proximal AA; distal IA × proximal IA).

Results

Of the 144 individuals who completed a phone screen, 28 were initially excluded based on exclusion criteria. Of the remaining participants, 55 attended the first orientation session, 2 were ineligible upon further eligibility screening, 2 withdrew for personal reasons, and 1 participant’s cell phone failed to work with the EMA software. One participant did not complete the follow-up visit but provided EMA data and was included in analyses (see Fig. 1). Thus, the final sample included 50 adults with mean age of 33.8 (SD = 12.3) and BMI of 28.5 (SD = 6.4). Participants were predominantly female (72%), white (77%) and non-Hispanic (82%) (see Table 1 for full demographic characteristics). Of the 50 participants who received the exercise intervention, most were available to complete the telephone counseling calls at weeks 6 (86%, n = 43) and 11 (82%, n = 41).

Fig. 1.

CONSORT flow diagram.

Table 1.

Demographics (N = 50)

Variable
Age, M(SD)	33.8 (12.3)
Gender, %(n)
Female	72% (36)
Male	28% (14)
Race, %(n)
White	77% (37)
Mixed race	18% (9)
Non-white	4% (2)
Ethnicity, %(n)
Non-Hispanic	82% (41)
Hispanic	12% (6)
Not reported	6% (3)
Employment, %(n)
Employed	74% (37)
Marital status, %(n)
Married	60% (30)
BMI M(SD)	28.3 (6.4)

BMI body mass index (kg/m²).

Descriptive Findings

On average, participants were enrolled in the EMA protocol for 90.52 days (SD = 17.73). Out of 4,526 potential person-days (50 participants × 90.52 days) the final dataset included only the 3,160 person-days (69.8% of total person-days) with morning reports, which included daily (proximal) attitudes. On a participant level, this corresponds to an average of 63.2 days of morning reports (SD = 20.9; range 16–97). Participants reported exercise on 871 person-days of the 3,160 person-days included in the analysis. On a participant level, this corresponds to an average of 17.4 exercise reports (SD = 15.3; range 0–69). Across participants, proximal AA and IA were more positive on exercise days (AA: M = 4.1, SD = 1.4; IA: M = 4.8, SD = 1.3) than on non-exercise days (AA: M = 3.4, SD = 1.5; IA: M = 4.3, SD = 1.3).

Preliminary Analyses

The average number of exercise reports per week (M = 1.3, SD = 1.6) decreased over time (b = −.03, SE = .01, p = .007). Distal AA (M = 3.8, SD = 1.4) was significantly lower on average than distal IA (M = 5.5, SD = .8), t = 8.80, p < .0001. The degree of within-person variability for proximal AA and IA were not significantly different (AA: mean s²_{within-person} = .66; IA: mean s²_{within-person} = .54) and were not associated with BMI, age, or gender (all p’s > .05). On average, proximal AA became more positive (b = .004, SE = .0006, p < .0001) and proximal IA became more negative (b = −.002, SE = .0006, p = .0005) over time. In univariate analyses predicting the odds of daily exercise behavior, there were significant associations for all attitude variables: distal (between-person) and proximal (within-person) AA and IA (see Table 2).

Table 2.

Univariate associations between attitudes and average odds of exercise behavior

Variable		Odds ratio	95% CI
Distal (Baseline)	Affective attitude (BP)	1.47**	[1.14, 1.91]
	Instrumental attitude (BP)	1.53**	[1.18, 1.99]
Proximal (Daily)	Affective attitude (WP)	1.26***	[1.12, 1.43]
	Instrumental attitude (WP)	1.25**	[1.07, 1.45]

BP between-person; WP within-person; CI confidence interval.

*p < .05, **p < .01, ***p < .001.

Primary Analyses

Our first hypothesis was not supported as distal AA was not more predictive than IA of the average odds of exercising each day on a between-person level (Table 3, Model 1). Controlling for covariates, individuals who reported a 1 SD more positive AA at baseline had a 28% greater average odds of exercising each day (OR = 1.28, 95% CI [1.02, 1.60], p = .03). Likewise, those who reported a 1 SD more positive IA at baseline had a 34% greater average odds of exercising each day (OR = 1.34, 95% CI [1.07, 1.67], p = .01). Next, we entered proximal AA and IA into the model. In support of our second hypothesis, proximal AA significantly predicted same-day exercise behavior (OR = 1.19, 95% CI [1.05, 1.36], p = .007). That is, on days when participants reported a one standardized unit more positive AA than their average (a within-person effect), they had a 19% greater odds (54% greater likelihood) of exercising that day. Proximal IA did not predict same-day exercise behavior (OR = 1.11, 95% CI [0.95, 1.31], p = .19) above and beyond distal AA and IA. Distal AA and IA remained significant with the addition of the proximal attitude variables (Table 3, Model 2).

Table 3.

Model parameters predicting odds of daily exercise

Variable		Model 1: Distal affective and instrumental attitudes		Model 2: Distal and proximal affective and instrumental attitudes		Model 3: Exploratory interaction model
		OR	95% CI	OR	95% CI	OR	95% CI
	Intercept	0.42	[0.13, 1.36]	0.44	[0.13, 1.50]	0.44	[0.13, 1.53]
	Time (day)	1.00*	[0.99, 1.00]	1.00*	[0.99, 1.00]	0.99*	[0.99, 1.00]
	Exercise yesterday	2.09***	[1.54, 2.83]	1.75***	[1.28, 2.39]	1.73**	[1.26, 2.37]
	Weekend	0.76*	[0.61, 0.95]	0.77*	[0.62, 0.96]	0.77*	[0.63, 0.97]
	Season	1.38	[0.97, 1.97]	1.40	[0.97, 2.04]	1.43	[0.99, 2.06]
	BMI	0.99	[0.95, 1.03]	0.99	[0.95, 1.03]	0.99	[0.95, 1.03]
Distal (Baseline)	Affective attitude (BP)	1.28*	[1.02, 1.60]	1.29*	[1.02, 1.63]	1.34*	[1.05, 1.71]
	Instrumental attitude (BP)	1.34*	[1.07, 1.67]	1.36**	[1.08, 1.72]	1.37**	[1.08, 1.73]
Proximal (Daily)	Affective attitude (WP)			1.19**	[1.05, 1.36]	1.21***	[1.08, 1.35]
	Instrumental attitude (WP)			1.11	[0.95, 1.31]	1.17	[0.99, 1.38]
	Distal affective attitude (BP) × proximal affective attitude (WP)					0.80***	[0.72, 0.89]
	Fit Indices
	QIC	3311.6		3307.5		3291.9

OR Odds ratio represents the odds of a participant engaging in exercise behavior on a given day (0 = no exercise, 1 = exercise); CI confidence interval; Time(day) = number of days since study enrollment; Exercise yesterday = whether the participant exercised on the prior day (0 = no, 1 = yes); Weekend = denotes if current day is on a weekday or weekend (0 = weekday, 1 = weekend); Season = time of year (0 = winter, 1 = summer); BMI body mass index (kg/m²); BP between-person; WP within-person.

*p < .05, **p < .01, ***p < .001.

Exploratory Interaction Analyses

An interaction between distal and proximal AA (OR = 0.80, 95% CI [0.72, 0.89], p < .0001) indicated that the effects of proximal AA on the likelihood of same-day exercise were considerably stronger among those with more negative distal AA. Specifically, participants who reported 1 SD lower than average distal AA (a between-person difference) had a 49% greater odds of exercising on days they reported a 1 SD more positive proximal AA than their average proximal AA (within-person effect). Among those who had 1 SD more positive distal AA, there was no increase in the odds of exercise daily as a result of within-person positive shifts in proximal AA (Table 3, Model 3). The remaining three interaction effects were not significant (distal IA × proximal IA; distal IA × distal AA; proximal IA × proximal AA).

Discussion

Results of the present study add to the current literature on the effects of AA and IA on exercise behavior. As shown previously, distal AA and IA—assessed at baseline—predicted average likelihood of exercise over 3 months. Contrary to expectations, however, distal AA did not have a greater impact on the average odds of exercising each day than distal IA. Instead, we found similar effects of distal AA and IA on daily exercise behavior. Our findings underscore the utility of between-person differences in attitudes at a single point in time for future decision-making.

This study is the first to report the impact of proximal AA and IA—measured daily—on exercise behavior at the within-person level. When examined in univariate analyses, within-person variations in daily AA and IA were associated with exercise behavior. However, consistent with our second hypothesis, when considered simultaneously in multivariable models including distal AA and IA, only within-person variability in AA significantly predicted exercise later that same day, whereas the effect of within-person variability in IA on exercise was rendered non-significant. It is unlikely that limited variability in IA explains the diminished effect in the multivariable model, as there was no significant difference in variability over time between AA and IA. It is worth noting that proximal IA had a negative slope, which could represent an expectancy violation effect [43] whereby initial beliefs about the positive instrumental outcomes of exercise may have been inflated. Inactive individuals have reported that their instrumental expectancies (e.g., weight loss) show greater expectancy violations than affective expectancies after starting a new exercise routine [43]. This is also consistent with findings on planning fallacy that show that people can underestimate the work required to achieve a goal [44]. Unmet expectations can, in turn, lead to degrading one’s beliefs and potentially less exercise. Thus, intervening to adjust inflated beliefs may prevent the detrimental effects of unmet expectations of exercise. While AA and IA are distinct attitude components, they often covary [45, 46]. In this study, the daily ratings of each attitude showed an increasing dissociation over time as proximal AA became more positive whilst proximal IA became more negative. More work is needed to explore the determinants of these patterns over time.

In exploratory analyses, we found that the impact of proximal AA on exercise behavior depended on between-person differences in distal AA. Participants who reported more positive distal AA did not benefit—in terms of increased likelihood of exercise—from an increase in proximal AA on a given day. This could be viewed similar to a ceiling effect, whereby individuals who started out with highly positive distal expectations about the affective outcomes of regular exercise were unaffected by deviations from their average proximal AA on any given day when it came to their decision to exercise. On the other hand, participants who reported more negative distal AA benefitted from positive shifts in proximal AA. As these analyses were exploratory, more research is needed to replicate the findings.

Limitations and Future Directions

The findings of this study should be interpreted in the context of several limitations. First, we were interested in the decision to initiate exercise rather than the duration of exercise because we recognize that there are numerous other factors that influence ongoing decisions to continue to exercise (e.g., intensity of the exercise, time available) that are beyond the scope of the current study. Second, while exercise was tracked in real-time, it was not verified using an objective assessment of activity. Although prior research indicates that people can accurately self-report their exercise in real time [47], some instances of exercise may not have been reported. For this study, we provided incentives for daily morning reports in which participants reported whether or not they forgot to enter exercise sessions from the prior day, and included these “forgotten” exercise sessions in the current analyses, thus attenuating some of the potential missing data as a result of compliance with in-the-moment exercise reports. Third, the findings of this study may not generalize to other behaviors. For example, AA have not been strongly associated with tooth brushing or vitamin use [17]. Fourth, while distinguishing between attitudes using the affective/instrumental characterization is consistent with a large body of work [11, 48], there may be other useful categorizations. Moreover, few items were used to target AA and IA. This reflects an inherent challenge within EMA research to balance the number of questions asked while also minimizing participant burden when responding to frequent assessment prompts. Fifth, these findings may not be applicable to the general population given that the sample was mostly White non-Hispanic females who were motivated to sign up for the exercise promotion intervention and in good health. Finally, while the purpose of the intervention was to produce variability in exercise behavior, it remains an open question as to whether the intervention introduced the variability observed with AA and IA, or if there would have been natural variability in these attitudes if there was no intervention delivered. Future research may examine naturally occurring changes in AA and IA in the absence of an exercise promotion intervention.

From a methodological standpoint, temporal context should be considered a meaningful dimension of the attitude construct. Assessing psychosocial predictors of behavior temporally proximal to behavior may be critical for other constructs that are likely subject to contextual influences, such as self-efficacy, anticipated affective response, and perceived barriers [49–51]. Given that daily variability in AA was a useful predictor of exercise behavior, it is conceivable that even more temporally fine-grained assessments could prove useful if AA vary within-day.

The findings from this study are consistent with the idea that AA may be sensitive to different temporal contexts and visceral factors (e.g., feeling states, drive-states). For example, if a person feels more energized than usual, they may have a more positive AA—they think exercise is going to be more invigorating rather than exhausting—and thus they may be more likely to exercise that day. On a subsequent day, if a person feels more tired than usual, their AA may be less positive and therefore detrimental to exercise. Prior work has identified that daily, within-person shifts in affect predict same-day exercise behavior [52]. Future research is needed to explore possible links between AA and momentary affect.

Exercise intervention strategies that target self-regulation and use persuasive messaging tend to be more influential in changing IA than AA [20]. To target AA, it may be more beneficial to focus on the experiential components of exercise (e.g., music, social and physical environment) given the likely role of mood and emotion in a person’s AA [53]. How people feel during exercise may impact future AA, which plays a role in the decision to exercise again. In prior work, greater perceived competence and feeling more energetic during exercise predicted positive changes in AA [54, 55]. Experiential components of exercise shown to enhance positive affect and reduce negative affect during exercise include outdoor settings [56, 57], social settings [57], listening to preferred music [58], lighter exercise intensities [59] and exercises that lead to greater perceived autonomy (e.g., self-pacing) [60, 61].

A better understanding of the day to day decision-making process is valuable for interventions that use technology to reach people at any time. For example, mobile health (mHealth) interventions could provide additional resources to individuals on days when they report more negative AA than they typically do. Moreover, it may not be cost effective to provide assistance every day, given that people may not need assistance on days when they present with more positive AA about exercise than they usually do. The findings of the exploratory moderator model suggest that exercise promotion strategies targeting proximal AA may be most helpful for individuals who hold more negative AA at the start of a program. To move the field forward, future work must also examine other types of complex models of affective and instrumental constructs, such as mediation and moderated mediation, in health behavior decision-making [62].

It will be critical for future interventions to be tailored to an individual’s baseline characteristics (between-person factors) but also to adapt to changes in behavior, affective states, and beliefs (within-person factors) to deliver appropriate and effective intervention [63, 64]. Interventions that adapt to within-person changes will need to consider that an individual’s average may change over time. What was once slightly more positive than usual may become how a person feels on most days as exercise becomes routine. Adjusting intervention techniques based on changes in attitudes and behavior may enhance engagement and foster long-term behavior maintenance. The findings of this study underscore that precisely timed measurement of affect-related beliefs about exercise behavior may be useful for improving physical activity intervention outcomes.

Compliance with Ethical Standards

Conflict of Interest The authors declare that they have no conflict of interest.