Implementation intentions are “if-then” plans that specify when and where an individual will enact particular behaviors in the service of a given goal intention [1]. They are thought to facilitate goal achievement by invoking a behavioral response (“I will go to the gym”) when a specified cue is encountered (“when I drive by the fitness center on the way home”). In this way, implementation intentions are hypothesized to strengthen intention–behavior relationships through the delegation of behavioral control to a situational cue that impels behavior in a relatively automatic fashion [2]. As goal-attainment devices, implementation intentions have been shown to be effective in facilitating several health-protective behaviors, including smoking cessation [3], promoting cancer screening [4, 5], skin cancer prevention [6], decreasing alcohol consumption [7], and dietary behavior [8, 9]. Likewise, a growing body of literature has also spoken to the potential utility of implementation intentions for promoting physical activity [10, 11], though effects in community-dwelling older adult samples are less well documented.

Given their role in facilitating automaticity—and therefore reducing the demand for effortful control—there has been considerable interest in the use of implementation intentions to enhance intention–action concordance for those who experience difficulties with behavior implementation [12]. Indeed, two prior studies have indicated that implementation intentions and other planning augmentation interventions might facilitate health behavior execution among those with weaker executive controlFootnote 1 in young adult samples [13, 14]. On the other hand, it is possible that a minimum level of cognitive aptitude is required in order to sufficiently implement behaviors even when cued, and so there is an argument to be made that intact executive resources might be required to make effective use of implementation intentions, but the restricted range of executive function in undergraduate samples precludes close examination of this. Along these lines, at least one study involving older adults found that those with relatively more intact executive resources may benefit more from planning interventions than their less intact counterparts, though the analysis was under powered to test the effect [15].

Currently, there is uncertainty as to who benefits most from implementation intentions, yet understanding such moderational effects might be crucial for effective health behavior programming and intervention, particularly for older adults who typically have a wider range of intraindividual variability in cognitive capacity, in part due to normal age-related cognitive decline. Accordingly, in the present investigation, we sought to examine two research questions: (1) “do implementation intentions enhance physical activity in older adults?” and (2) “are such effects more pronounced for those with stronger or weaker executive function?” To answer these questions, we conducted a small-scale randomized controlled trial using a sample of community-dwelling women who were above age 60 years but free of major functional impairments and free of major neurological difficulties. It was hypothesized that implementation intentions would be effective overall as per prior meta-analyses [10] but that their effects would be strongest among those older participants who have relatively intact executive control resources because they have more capacity to implement cued behaviors.

Methods

Participants

Seventy-five adult women between 61 and 89 years of age were recruited from a university research participant database for older adults and from the surrounding community (see Table 1). All participants were functionally mobile, had corrected-to-normal vision, and were not suffering from major cognitive, neurologic, or motor impairments.

Table 1 Sample characteristics
Full size table

Procedure

Prior to scheduling the initial laboratory session, a random number sequence was used to allocate consecutive participants to conditions. Upon arrival to the lab, participants provided informed consent, and waist circumference was assessed, followed by completion of three tasks of executive function: go/no-go (GNG; behavioral inhibition), number–letter task (NL; task switching), and keep track task (KT; working memory). The GNG was used previously [14], whereas the NL and KT tasks were constructed in accordance with Miyake et al. [16]. Finally, participants were provided with a hip-mounted Actical accelerometer and instructed as to its proper usage. Participants also provided information on demographics, habit strength for physical activity (using the Self-Report Habit Index; [17]), and intention strength using the item, “To what extent do you intend to engage in physical activity on a daily basis over the next month?” (responses given on 1 = “not at all” to 7 = “very strongly” Likert scale). During each of the four follow-up telephone sessions spaced at weekly intervals following the baseline lab visit, participants provided self-reports of physical activity behavior for each week prior. Those in the experimental and control groups also refreshed implementation intentions for each upcoming week.

Weekly Physical Activity

During weekly telephone contacts, participants verbally reported their physical activity by indicating how many hours (to the nearest half hour) of vigorous physical activity they engaged in over the “past 7 days” (examples provided: “jogging or running, swimming, strenuous sports such as singles tennis or racquetball, digging in the garden, chopping wood, etc.”) using a reworded version of the Stanford 7-day Recall [18]. The original self-reported version of the physical activity recall (PAR) possesses good test–retest reliability (r = .83, p < .001) and criterion validity (r = .73, p < .001; [18, 19]). The version of the recall utilized here refers to only vigorous activity as this is significantly more reliable than moderate and light activity in prior studies, and utilizes estimation to the nearest half hour rather than minutes, given that older adults are less likely to be able to accurately recall minutes.

Baseline Activity

To assess baseline physical activity, participants wore a hip-mounted Actical monitor [20]. The device was worn for 7 days during waking hours, and bodily acceleration data were stored every 15 s as activity counts. From this, average daily activity counts were calculated to represent the measure of physical activity in those that wore the device for at least 4 of the 7 days (n = 62); maximum likelihood imputation was used to calculate the predicted scores of any not meeting this criterion (n = 16). Prior studies have confirmed that accelerometers provide reliable estimates of energy expenditure with a 3–4-day sampling of activity counts [21]. The Actical accelerometer used in the present investigation has demonstrated superior intra-instrument and inter-instrument reliability when compared to other accelerometers in common use [22]. The accelerometer measure was supplemented with a baseline self-report measure of physical activity identical to the weekly measure described above; accelerometer and self-reported activity were correlated (r = .310, p = .054, n = 28). The self-report and accelerometer measures were standardized and combined into a single index score, with higher scores indicating higher levels of baseline physical activity.

Executive Function

Three measures of executive function were administered (in fixed order, as they appear below), each representing one of three facets of executive function as outlined in Miyake et al. [16]. Following a factor analysis using the current data, scores on these three measures all loaded on a single factor, and factor scores for executive function were used as the primary predictor variable:

Go/No-go Task

The go/no-go (GNG) task is a reaction time task that assesses the ability to inhibit prepotent responses to stimuli. For the current version of the task, participants were seated at a desktop computer and instructed to press a button on a response box if a lower case letter was shown, but refrained from pressing the button if an upper case letter was shown. Participants were asked to respond as quickly as possible without making mistakes. After completing a practice block of 12 trials, participants encountered eight more blocks: four with a preponderance of upper case trials (50 upper; 10 lower) and four with a preponderance of lower case trials (10 upper; 50 lower). Shorter reaction times on correct trials were taken to indicate stronger inhibitory abilities. Performance on the GNG task has been shown to be correlated with prefrontal cortex function [23, 24], and this specific version has previously been shown to correlate with other measures of inhibition and health behavior [14].

Number–Letter Task

The number–letter task (NL) described by Miyake and colleagues [16] was employed to assess task switching. Participants were shown a number–letter pair (i.e., “7R”) in one of four quadrants. When the number–letter pair was presented in the lower two quadrants, participants were required to specify whether the letter was a vowel or a consonant by pressing the appropriate button. Similarly, when the number–letter pair was presented in the upper two quadrants, an indication of whether the number was odd or even was required. The task included three blocks: one block of 32 trials in which the number–letter pairs were only presented in the upper two quadrants, one block of 32 trials in which the number–letter pairs were only presented in the bottom two quadrants, and one block of 128 trials in which the number–letter pairs were presented in each of the four quadrants sequentially in a clockwise order. Performance on the NL task was assessed by determining the difference between reaction times of trials in the third block and the first two blocks which provided a measure of shift cost.

Keep Track Task

The keep track (KT) task from Miyake et al. [16] assessed the ability to update and monitor working memory. Participants were asked to hold in memory relevant information while new information was presented. In each trial, participants were presented with a random sequence of words (e.g., bird, green, aunt, Canada, nickel, near). Once the words had been presented on the screen, they were asked to try to recall the most recent words shown that fell within different categories (e.g., animals, colors, relatives, countries, metals, distances) and write them down on a response sheet. In the initial two practice trials, participants were asked to recall the most recent words presented that fall within each of three categories. In the next three trials, participants were asked to recall the most recent words presented from each of four categories. The last trial required participants to try to recall the most recent words presented falling within five categories. The proportion of correct responses was used to assess working memory, with higher values indicating stronger working memory.

Implementation Intentions Manipulation

Those randomly assigned to the experimental or control conditions formed implementation intentions for physical activity goals and reading goals, respectively. At baseline, participants were asked to state their goal and then describe their goal plan (i.e., “In the space below, please think about and describe when, where, and how you will achieve your goals.”). To facilitate thinking pertaining to possible obstacles to goal achievement, participants were next asked to write down what may prevent them from achieving their goals (i.e., “Think of any obstacles that may prevent you from achieving your goals.”). Participants were subsequently instructed to generate solutions to any reported obstacles (i.e., “What are some ways you may be able to overcome those obstacles?”). During each follow-up telephone survey, participants were asked to refresh their weekly goal plans with the researcher by writing down again when, where, and how they would achieve their goal for the next week.

Results

Sample Characteristics and Handling of Missing Data

Participants were between the ages of 61 and 81 years and were generally well educated (Table 1). Thirty-eight percent of participants did not complete behavioral self-report at one point across the four waves of assessment. Thus, to determine the pattern of missingness for physical activity, a dummy variable was created (0 = data absent; 1 = data present). Next, this variable was used to predict missing status on baseline demographic, behavioral, and cognitive variables via zero-order correlations and χ 2 analyses. Results showed that missingness was not significantly (p < .05) related to any variables in our data. Therefore, it was assumed that the data were missing completely at random [25] and missing values were imputed using the expectation maximization algorithm in LISREL 8.8 [26, 27]. Non-imputed means were similar to those gleaned from imputed variables but with more marginal p-values (in two instances only) due to reduced power. As such, to improve power, imputed data were used for all analyses.

The experimental, control, and no-treatment groups did not differ significantly in terms of baseline physical activity behavior, habit strength for physical activity behavior, or intention strength for physical activity behavior. The groups also did not differ in waist circumference in inches, age in years, or years of education completed (Table 1).

Basic Treatment Effects

Hierarchical linear regression analysis were undertaken to quantify the primary treatment effects. Effect coding was used to compare the effects of the experimental and control treatments to the grand mean, using weekly hours of physical activity summed across the 4 weeks of follow-up as the primary dependent measure and baseline activity as a covariate. Following this, we examined the treatment effect separately for each week.

Using the 4-week aggregate measure, there was a significant treatment effect for the experimental group (b = .920 (SE = .470), β = .241, p = .054; d = .536). Specifically, those in the experimental group (M = 3.486, SE = .579) engaged in more activity than those in the control (M = 2.146, SE = .591) and no-treatment groups (M = 2.065, SE = .534). Examination of the experimental group effect on a week-by-week basis confirmed a significant treatment effect for the experimental group in week 1 (b = .308 (SE = .140), β = .283, p = .031), week 3 (b = .279 (SE = .134), β = .260, p = .041), and week 4 (b = .284 (SE = .129), β = .277, p = .031). The effect was not significant in week 2 (b = .050 (SE = .140), β = .045, p = .724). The treatment effect for the control intervention did not attain significance in any study week (all ps < .10).

Moderation Effects

Next, we examined two-way interactions between executive function and treatment effects. Again, we used hierarchical linear regression models with group coded using effect coding and adjusting for baseline physical activity; main effect variables were mean centered prior to combination in the interaction term. Examination of the moderating effect of executive function on aggregated 4-week hours of physical activity revealed a marginal interaction between the treatment effect for the experimental group and the executive function strength (b = .986 (SE = .509), β = .265, p = .057). When baseline habit strength was added as an additional covariate, the interaction was significant (b = 1.150 (SE = .516), β = .312, p = .030); as depicted in Fig. 1, the effects of the experimental group were significantly stronger among those with stronger executive function. Examining the treatment effect on a week-by-week basis revealed a significant interaction with executive function for week 2 (b = .363 (SE = .151), β = .341, p = .020), week 3 (b = .296 (SE = .143), β = .296, p = .043), and marginally for week 4 (b = .266 (SE = .151), β = .063, p = .057). The treatment effect did not attain significance for week 1 only (b = .225 (SE = .157), β = .215, p = .157). No main effects or interactions involving the control condition were significant (all ps < .10).

Fig. 1
figure 1

Physical activity as a function of treatment group and executive function. Physical activity is reported in hours to the nearest half hour, summed across the 4-week follow-up interval. Error bars represent standard errors for each mean estimate. Covariates = baseline physical activity behavior and habit strength for physical activity

Full size image

Discussion

Implementation intentions significantly enhanced physical activity among older adult women, and the effect was medium in size. The beneficial effect of the intervention was modified significantly by executive function: Those with the strongest executive function benefitted the most from the formation of implementation intentions. The current findings reinforce the utility of implementation intentions as a means of enhancing physical activity in older adults, and the importance of executive function as a potential moderator of treatment effects and other outcomes in the health domain [28, 29].

Of special interest is the fact that implementation intentions were effective at increasing activity in this older sample even though participants were not selected specifically for interest in becoming more active. This suggests that implementation intentions may have benefit even for those who only have very minimal goal intentions to be active.

Our finding that implementation intentions are effective for enhancing physical activity corresponds with the findings of many prior studies, though it is among the first to demonstrate such effects in an older adult community sample [10]. The current moderational effect of executive function on implementation intentions also corresponds with at least one prior study documenting similar effects among older adults involving the use of implementation intentions to assist with prospective memory [15]. In fact, it is possible that these findings give some hint as to why implementation intentions are more effective for those with more intact executive function—that is, it could be that plans are more easily remembered and/or processed when cued among those with more preserved cognitive abilities.

Our findings appear, on the surface, to be somewhat divergent from two prior studies involving younger adults [13, 14]. However, there are methodological differences that explain such divergence, including the fact that Hall and colleagues [14] studied intention–behavior relationships as an outcome rather than behavior per se and both studies [13, 14] involved young adult university samples, which would differ substantially in range of executive function scores from older adult samples used here and in the McFarland study [15]. For example, the interquartile range for the GNG measure—the only executive function measure common to both studies—was 84.09 in the current study, but only 47.18 and 52.82 in studies 1 and 2, respectively, of Hall and colleagues [14]. The effects of restricted range are difficult to predict but could include setting a lower limit on the effects observed for the executive function variable or misclassifying individuals as “low” or “high” on executive function in absolute terms; that is, those considered low in undergraduate samples could be moderate in older adult samples, and true “lows” of the nature classified here or in McFarland could be rare in undergraduate datasets. Along these lines, the mean value for the GNG reaction time was 410 ms for the undergraduates in the Hall and colleagues study but much slower (535 ms) for the current study, such that those considered “strong executive function” on this measure in the present study would be considered “weaker executive function” on this measure in the undergraduate sample.

Generally speaking, it is possible that those with less intact cognitive ability may exhibit a mild form of goal neglect, wherein the goal remains salient (because it was either remembered spontaneously or recently cued) yet the individual fails to perform the task [30]. Under such circumstances, additional environmental structuring may be required in order to encourage implementation, as is commonly attempted for those with milder forms of cognitive impairments in order to facilitate activities of daily living. However, few health behavioral interventions currently include ecological structuring activities; this remains an important direction for future research, and one that might particularly benefit those older adults with less intact cognitive resources.

Strengths and Limitations

Some of the strengths of this investigation include the combined use of accelerometry and self-report for assessing baseline behavior, and use of an older adult community sample, which likely has more variability in important predictor variables (specifically, executive function) than younger adult samples. Limitations include the brief follow-up window of 4 weeks, which may not be long enough to establish the long-term efficacy of implementation intentions. However, excessively long windows would begin to conflate adherence to the formation of implementation intention and actual performance of the target behavior, so ultimately, the shorter window may be a necessary limitation to prevent missing data that correlate with some other important predictors in the study.

It could be argued that the control condition (which involved implementation intentions for reading) in the current study might reduce opportunities for active behavior, thereby amplifying differences between experimental and control conditions. However, this is unlikely to be the case for several reasons. First, our data show modest increases in physical activity even in the control condition, rather than reductions. If competition between exercise and reading was taking place, we would expect that the amount of activity reported in the control group would be lower than in the no-treatment group, and this was not the case. Second, given that the majority of our participants would be retired, there is presumably less competition between exercise and reading within limited non-work hours. So although it is possible that the two activities may compete in working populations (particularly for those who work full time plus have parenting roles), this is less likely to be the case with the current population.

It could be argued that executive function predicts activity because executive function itself is enhanced by activity as has been found previously [31]. However, the notion that executive function is improved by exercise does not preclude its role in exercise adherence, and in fact, participation and adherence may have a reciprocal relationship with each other [32]. Also, the present findings are not accounted for by baseline differences in activity, as we controlled for these stringently by co-varying accelerometer, self-report, and habit strength measures.

Finally, the version of the implementation intentions exercise tested here has some additional components (vis-à-vis the use of anticipatory barrier coping) and could be more in line with the action planning/coping planning approach of Sniehotta and colleagues [33]. Additional studies may help to determine whether or not a more “pure” approach to implementation intentions has similar direct and moderated effects on physical activity in this age group.

Conclusions

In conclusion, we found evidence that implementation intentions are effective for enhancing weekly physical activity among older adults, for whom such activity has many proximal health benefits. Importantly, we found that the beneficial effects are especially pronounced for those who have relatively intact executive control. The effects were independent of prior activity level and habit strength for physical activity. Implementation intentions are therefore a useful physical activity promotion tool for older adults, but especially for those with intact executive function. Additional research will be required to more conclusively determine the responsiveness of adults to implementation intentions across the lifespan and across multiple behaviors.