Journal of Pediatric Psychology, 45(4), 2020, 445–453

doi: 10.1093/jpepsy/jsaa007
Advance Access Publication Date: 27 February 2020
Original Research Article

Does Self-Efficacy Mediate the Link Between

Impulse Control and Diabetes Adherence?
Karol Silva ,1 PHD, MPH and Victoria A. Miller,1,2 PHD
1
Department of Pediatrics, Children’s Hospital of Philadelphia and 2Perelman School of Medicine, University of
Pennsylvania
All correspondence concerning this article should be addressed to Victoria Miller, PhD, Children’s Hospital of
Philadelphia, 3500 Civic Center Blvd., Buerger 12th Floor, Suite 12-219, Philadelphia, PA 19104, USA. E-mail:
millerv@[Link]
Received September 13, 2019; revisions received January 30, 2020; accepted January 31, 2020

Abstract
Objective The goal of this study was to test the hypothesis that diabetes self-efficacy mediates
the relationship between impulse control and type 1 diabetes (T1D) management from ages 8 to 18
years, using multilevel modeling. Methods Participants included 117 youth with T1D and their
parents. Youth (aged 8–16 years at baseline) and parents were assessed 5 times over 2 years.
Using a cohort sequential design, we first estimated the growth trajectory of adherence from age
8 to 18 years, then specified a multilevel mediation model using impulse control as the main predic-
tor, diabetes self-efficacy as the mediator, and changes in adherence (both within- and between-
individuals) as the outcome. Results According to youth-reported adherence only, self-efficacy
partially mediated the within-person effect of impulse control on adherence. On occasions when
youth reported increases in impulse control, they tended to report higher adherence, and this was,
in part, due to increases in youths’ perceived self-efficacy. Self-efficacy accounted for approxi-
mately 21% of the within-person relationship between impulse control and youth-reported
adherence. There was no association between impulse control and adherence between-
individuals. Impulse control and self-efficacy were not related to parent-reported adherence.
Conclusion Environments that enrich youth with confidence in their own diabetes-related abili-
ties may benefit self-care behaviors in youth with T1D, but such increases in youths’ perceived
competence do not fully account for, or override, the behavioral benefits of impulse control. Efforts
to improve adherence in youth with T1D will benefit from consideration of both impulse control
and self-efficacy.

Key words: adherence; chronic illness; diabetes; longitudinal research; psychosocial functioning.

Introduction regulatory skills required for successful self-care—

Management of type 1 diabetes (T1D), which entails which include planning ahead, monitoring emotions
checking blood glucose levels several times per day, and behaviors, and controlling the impulse to give into
dosing insulin, monitoring food intake, exercising, immediate rewards—are not fully mature (Casey,
and responding to high or low blood glucose, requires 2015). Although many adolescents are capable of self-
considerable daily effort and can be particularly chal- regulation and impulse control, doing so consistently
lenging in adolescence compared with childhood and can be difficult, especially in the context of competing
adulthood (King, Berg, Butner, Butler, & Wiebe, pressures and demands (Shulman et al., 2016).
2014; Wood et al., 2013). From a developmental per- Cross-sectional and longitudinal studies demon-
spective, a factor that may contribute to the challenges strate that adolescents with good self-regulation en-
of T1D management in adolescence is that the gage in more self-care behaviors (and have better

C The Author(s) 2020. Published by Oxford University Press on behalf of the Society of Pediatric Psychology.
V
All rights reserved. For permissions, please e-mail: [Link]@[Link] 445
446 Silva and Miller

glycemic control; e.g., Vaid, Lansing, & Stanger, control and self-efficacy are known to be indepen-
2018) compared to adolescents with low self- dently associated with adherence in youth with T1D,
regulation (e.g., Bagner, Williams, Geffken, and both constructs have been shown to moderate the
Soilverstein, & Storch, 2007; Berg et al., 2014; Duke relationship between treatment responsibility and
& Harris, 2014; Miller et al., 2013; Wiebe, Berg, T1D management in adolescence.
Mello, & Kelly, 2018). In our own prior longitudinal Social cognitive theory considers impulse control
work, we found that improvements in impulse control, and self-efficacy important person-level factors that
within individuals, coincided with increases in adher- influence behaviors and posits that belief in one’s self-
ence (Silva & Miller, 2019); improvements in impulse efficacy is a common pathway through which psycho-
control appeared to be particularly protective as ma- social variables affect health behaviors and function-
turing youth take on more responsibility for T1D ing (Bandura, 2004). Although several studies point to
management from late childhood to late adolescence self-efficacy as a mediator of the relationship between
(Silva & Miller, 2019). Several qualitative studies of impulsivity and adolescent risk behaviors, such as sub-
adolescents with T1D also illustrate the ways in which stance use (e.g., Hayaki et al., 2011; Patton et al.,
adolescents’ tendency to favor immediate outcomes 2018), few studies have examined the role of these
may interfere with T1D management, particularly in constructs in diabetes management within the same
social or emotionally stimulating contexts (e.g., “If sample. To our knowledge, only one prior cross-
you’re going to a movie to eat popcorn, you don’t sectional study of late adolescents with T1D has ex-
want to be like hold on let me check my blood” plored impulse control and self-efficacy and suggested
[Babler & Strickland, 2015, p. 655]; “When I am out that self-efficacy is a mechanism via which impulse
with my friends a lot of times I’m not as good about control is related to individual differences in youth ad-
checking the things I need to do” [Hanna & Guthrie, herence (Stupiansky et al., 2013). In that study, ado-
2000, p. 170]). lescents with higher impulse control tended to exhibit
Developmental studies demonstrate that impulse better self-management, and this association was
control is a skill that develops across adolescence and largely explained by individuals’ level of self-efficacy.
into early adulthood (Harden & Tucker-Drob, 2011; Based on this finding, the study authors suggested that
Shulman, Harden, Chein, & Steinberg, 2015). As im- diabetes management in adolescence could be amena-
pulse control matures, it may be easier to stop a fun ble to intervention through self-efficacy (Stupiansky
activity to execute important illness-related tasks et al., 2013). However, further longitudinal research is
(Silva & Miller, 2019). Nonetheless, how exactly im- necessary to substantiate such conclusions and to bet-
pulse control facilitates T1D management in develop- ter inform interventions aimed at improving T1D
ing youth remains an understudied subject. management in adolescence.
Since impulse control is a skill that matures over Using multilevel modeling, the present study—
time, maturational gains in impulse control may set based on a secondary analysis of an existing dataset—
the developmental context for other psychological expands existing cross-sectional research by examining
processes to unfold that, in turn, facilitate desirable the longitudinal relationship between impulse control,
health behaviors in maturing youth. One such process self-efficacy, and diabetes adherence in a sample of
that may unfold with improvements in impulse control youth (ages 8–16 years) with T1D. Specifically, the
is self-efficacy. Self-efficacy refers to individuals’ per- goal of this study was to test the hypothesis that diabe-
ceived confidence in their abilities to perform self-care tes self-efficacy mediates the relationship between im-
behaviors in challenging or affect-laden situations— pulse control and diabetes management from late
for example, checking blood glucose levels even when childhood to late adolescence, both between- and
engaged in another task, exercising when one does not within-individuals. Within-person estimates (level 1)
want to, and completing other diabetes-related tasks are based on repeated (time-varying) measures and
despite feeling overwhelmed (Iannotti et al., 2006). represent the extent to which time-specific changes in
Several cross-sectional and longitudinal studies show the predictor(s) (i.e., impulse control and self-efficacy)
that self-efficacy is positively associated with T1D relate to time-specific changes in the outcome (i.e., ad-
management in adolescence (Berg et al., 2011; herence) within each individual. Between-person esti-
Johnston-Brooks, Lewis, & Garg, 2002; Littlefield mates (level 2) are based on sample averages and
et al., 1992; Ott et al., 2000; Wiebe et al., 2014). represent the extent to which individual differences in
Moreover, and similar to impulse control (Silva & impulse control and self-efficacy are related to individ-
Miller, 2019), there is evidence from a longitudinal ual differences in adherence. By distinguishing be-
study of adolescents that improvements in self-efficacy tween- and within-person effects, we are able to
are protective against age-related declines in adher- examine both the extent to which cumulative increases
ence as youth take on responsibility for self-managing in impulse control contribute to increases in adherence
their diabetes (Wiebe et al., 2014). Thus, both impulse (via changes in self-efficacy within individuals) and
Impulse Control and Diabetes Adherence 447

the extent to which youths’ baseline levels of impulse threatening medical condition, unrelated to T1D,
control contribute to individual differences in adher- which required daily treatment for more than
ence (via self-efficacy). At both the between- and 6 months in the last year.
within-person levels, we hypothesized that impulse Five hundred sixteen potential families were identi-
control would be associated with more diabetes self- fied through outpatient clinic lists and schedules. Of
care behaviors, at least in part, because increases in those identified, 48.3% (n ¼ 249) could not be con-
impulse control may enable youth to feel more confi- tacted, and 51.7% (n ¼ 267) were reached by tele-
dent and capable of performing diabetes-related tasks. phone. Of those who were contacted, 18.7% (n ¼ 50)
While we acknowledge that impulse control and could not be reached again, 18.7% (n ¼ 50) declined
self-efficacy may be important predictors of glycemic immediately, and the remaining 62.5% (n ¼ 167)
control in adolescence, we did not include hemoglobin agreed to be screened for eligibility. Of the 167 fami-
A1c (HbA1c, a measure of glycemic control) as an lies screened for eligibility, 9.6% (n ¼ 16) were ineligi-
outcome in this analysis because we were primarily in- ble. Common reasons for ineligibility included that
terested in self-reported behavior, not clinical out- the child was diagnosed with T1D less than 1 year ago
comes. However, HbA1c was included as a covariate or had at least one other illness not related to T1D
in all the analyses conducted and reported here. that required daily treatment for greater than 6
months of the last year. Of the 151 parent–child dyads
that were eligible, 148 (98%) agreed to participate,
Methods
but 10 (6.6%) could not be scheduled or reached
This study is based on secondary analysis of data from again, 14 (9.3%) did not show up for their scheduled
a larger longitudinal study designed to examine pre- appointments, and 1 (0.7%) declined in person. Of
dictors and outcomes of decision-making involvement the 123 (81%) dyads who consented and enrolled in
in a sample of youth with T1D or cystic fibrosis the study, 4 (3.3%) did not complete Visit 1 and an
(Miller & Jawad, 2019). The study used a cohort se- additional two dyads (1.3%) were withdrawn from
quential design, which involves the examination of the study by the principal investigator after finding
different age cohorts over the same time period, and they no longer met eligibility criteria. The final sample
allows the combination of multiple short-term longitu- included 117 parent–youth dyads. Chi-square and in-
dinal data points to estimate a single long-term growth dependent samples t-tests indicated that there were no
pattern using growth curve modeling (Miyazaki & significant differences found between these partici-
Raudenbush, 2000). Nine age cohorts (8-to-16-year- pants and those who were eligible but declined, could
olds) were assessed at baseline and every six months not be scheduled, or did not complete Visit 1 (n ¼ 34)
for two years (for five total assessments). The number with respect to age, duration of diagnosis, child sex,
of participants in each age cohort varied at baseline: child race, or child ethnicity (all ps > .20).
8 years (n ¼ 6), 9 years (n ¼ 11), 10 years (n ¼ 18), All data were collected via self-report question-
11 years (n ¼ 13), 12 years (n ¼ 12), 13 years (n ¼ 10), naires (some of which were completed by youth only
14 years (n ¼ 16), 15 years (n ¼ 18), and 16 years and others were completed by both parents and
(n ¼ 13). The nine age cohorts were linked to form a youth). Research personnel read questionnaires to
common developmental trajectory spanning ages 8– youth ages 8–10 years to promote comprehension.
18. Although the larger study included youth with Approximately 55% of youth in the sample were
T1D or cystic fibrosis, the current study describes the female, and the mean youth age at baseline was 12.87
procedures, measures, sample, and data from youth (SD ¼ 2.53). Overall, approximately 60% of the youth
with T1D and their parents only. were non-Hispanic white (n ¼ 69), 22% were non-
Hispanic black (n ¼ 25), 10% were Hispanic (n ¼ 12),
Participants and Procedures 5% were other (n ¼ 7), and 1 youth was Asian. The
Parents and youth were recruited from an endocrinol- majority (77%) of youth lived in two-parent house-
ogy clinic at a tertiary children’s hospital in a large holds and had been living with T1D for approximately
Northeastern city in the United States. All study proce- 5.63 years (SD ¼ 3.53). At baseline, the mean HbA1c
dures were in accordance with U.S. guidelines for the was 8.76% (SD ¼ 1.49). Approximately 39% of the
ethical conduct of human subject research and ap- youth were using an insulin pump, 40% were on a
proved by the institutional review board. Participants basal-bolus regimen, and 21% were on a pre-mixed
were eligible if they were English-speaking, the parent regimen (70/30).
was the biological or adoptive parent, and youth had
been diagnosed with T1D for at least one year and Measures
lived with the parent at least 50% of the week. Youth Reliability was estimated separately for each variable
were ineligible if they had developmental delay, past- at each time point. We report the range of reliability
year psychiatric hospitalization, or another life- coefficients, as well as the intraclass correlations (ICC)
448 Silva and Miller

across measurement waves to indicate within-subject analyses revealed few notable discrepancies between
reliability. Variable means and standard deviations youth and parent reports of adherence. Overall, a sig-
(for each time point and age cohort) are reported in nificant reporter discrepancy was noted at Time 4
the Supplementary Table. only (t (92) ¼ 2.72, p ¼ .008) and further investigation
revealed that the discrepancy was driven by youth and
Impulse Control parents in the 11-year-old cohort (t(10) ¼ 2.90,
At all visits, youth completed the eight-item impulse p ¼ .016).
control subscale from the Weinberger Adjustment
Inventory (Weinberger, 1997). Participants indicated Chart Review
how accurately a series of eight statements (e.g., “I During visits 1, 3, and 5, study staff completed a re-
stop and think things through before I act;” “I should view of participants’ medical charts to obtain informa-
try harder to control myself when I’m having fun” [re- tion regarding insulin regimen (pump, premixed 70/
verse coded]) described them on a 5-point Likert-type 30, or basal-bolus) and hemoglobin A1C.
scale (ranging from 1 ¼ almost never to 5 ¼ almost al-
ways). Scores were averaged across items such that Attrition and Missing Data
higher scores indicate greater impulse control. Of the 117 dyads enrolled in the study, 78 completed
Cronbach’s alphas across visits ranged from 0.76 to all follow-up visits, 33 completed between 1 and 3
0.84. The ICC was moderately high (0.70). follow-up visits, and 6 did not complete any follow-up
visits. Overall, there were 117 evaluable cases at visit
Diabetes Self-Efficacy 1, 97 at visit 2, 101 at visit 3, 97 at visit 4, and 96 at
At visits 1, 3, and 5, youth completed the Diabetes visit 5. There were no significant differences between
Self-efficacy Scale (DSES), a 10-item measure to assess participants who completed no follow-ups, 1–3
youth self-efficacy in managing illness-related emo- follow-ups, and all 5 follow-ups with respect to base-
tions and procedures associated with the diabetes line age, demographics, or main variables analyzed in
treatment regimen (Iannotti et al., 2006). Youth indi-
the present study (all ps > .05).
cated how sure they felt (on a scale from 1 ¼ not at all The amount of missing data was low and ranged
to 10 ¼ completely sure) that they could do a series of
from 1.3% to 2.6% for each study visit. Across study
10 tasks including choosing healthy foods when they
visits, two youth had missing data on impulse control
go out to eat, doing their blood sugar checks even
and diabetes self-efficacy, one youth had missing data
when they are really busy, and taking care of their dia-
on self-reported adherence, and two youth had miss-
betes even when they feel overwhelmed. Scores were
ing data on parents’ report of adherence. Youth with
averaged across items such that higher scores indicate
missing data on parents’ report of adherence had
higher diabetes self-efficacy. Cronbach’s alphas across
lower self-efficacy at baseline compared to youth
all visits ranged from 0.80 to 0.86. The ICC was 0.54.
without missing parent data (p < .05). No differences
were found between youth with versus without miss-
Diabetes Adherence
At all visits, youth and parents completed the Self ing data on self-reported adherence, impulse control,
Care Inventory (La Greca, Follansbee, & Skyler, or self-efficacy (all ps > .05). Missing data were
1990), a 14-item measure to assess past-month adher- addressed using full Information Maximum
ence to multiple aspects of the diabetes treatment regi- Likelihood estimation.
men (e.g., glucose testing, administering and adjusting
insulin, eating regular snacks). Responses range from Data Analysis
1 (never do it) to 5 (always do this without fail). Since the current study was based on secondary data
Scores were averaged across items such that higher analysis, we ensured we had a sufficiently large sample
scores indicate better adherence. Cronbach’s alphas size to achieve 80% power in the estimation of longi-
across all visits ranged from 0.69 to 0.83 for youth re- tudinal mediation effects. Assuming an ICC of 0.90
port and 0.70 to 0.81 for parent report. The ICC was (given the high ICC for child-reported adherence) and
0.95 and 0.63, for youth and parent report, a study design of at least three observations per subject
respectively. (since data on youth self-efficacy were collected at
Although the main predictor (impulse control) and three-time points only), the estimated number of sub-
proposed mediator (self-efficacy) reflect youth- jects required to achieve a medium-sized mediation af-
reported data, we used both youth and parent reports fect is 73 (Pan, Liu, Miao, & Yuan, 2018). Based on
of adherence because a multi-informant approach is these estimates, we concluded that the current sample
typically better than single-informant (De Los Reyes size was sufficient to achieve 80% power in a longitu-
et al., 2015). Moreover, descriptive and preliminary dinal mediation analysis that treats all path
Impulse Control and Diabetes Adherence 449

coefficients as fixed-effects (i.e., model estimates were Results

constrained to be equal across individuals). Youth-Reported Adherence
The total effects model revealed that impulse control
Model Building was a significant positive predictor of within-person
In the present analysis, we built on two previously de- variability in youth-reported adherence (total effect at
scribed models that estimated age-related changes in level 1: B ¼ .14, SE ¼ .03, p < .001), controlling for
child- and parent-reported adherence in the current HbA1c. This finding indicated that on occasions when
study sample (Silva & Miller, 2019). These models youth experienced increases in impulse control (rela-
were estimated using latent growth curve modeling, tive to baseline), they exhibited concurrent increases
with age (8–18) as the marker of time, centered at the in adherence. At the between-subjects level, impulse
grand mean (13.77 years). The process for identifying control was not a significant predictor of individual
potential covariates of changes in adherence is de- differences in overall adherence at age 13.77 (inter-
scribed in Silva & Miller (2019). For models estimat- cept: B ¼ .04, SE ¼ .03, p ¼ .16), or rate of decline in
ing parent-reported adherence, no significant adherence from ages 8–18 years (slope: B ¼ .001,
covariates were identified. Models estimating youth- SE ¼ .01, p ¼ .94), controlling for baseline HbA1c,
reported adherence included baseline age, baseline ad- baseline age, baseline adherence, race/ethnicity, and
herence, race/ethnicity, and income as covariates at income.
level 2 (to account for differences between Results from the multilevel mediation model indi-
individuals). In this analysis, we retained the cated a significant within-person relationship between
previously identified level 2 covariates and, in impulse control and self-efficacy (Table I; level 1, a
addition, controlled for HbA1c at both level 1 path: Ba¼0.32, SE ¼ 0.10, p < 0.001), suggesting that
(centered at baseline values) and level 2 (baseline on occasions when youth reported increases in impulse
values centered at the grand-mean) to estimate the to- control, they also reported increases in self-efficacy.
tal effect of impulse control on adherence. At level 1, Time-specific increases in self-efficacy were, in turn,
impulse control was centered at participants’ baseline associated with concurrent increases in adherence
score so that the estimate represents within-person de- (level 1, b path: Bb¼0.10, SE ¼ 0.02, p < 0.001).
viation, or cumulative change, from the first time Moreover, there was a significant indirect effect of im-
point (Hoffman & Stawski, 2009). At level 2, baseline pulse control on adherence via self-efficacy within
values of impulse control were grand-mean centered individuals (Table I; B ¼ 0.03, 95% CI ¼ 0.01, 0.05,
p ¼ 0.005). The within-person effect of impulse con-
to reflect individual differences.
trol on adherence remained significant after account-
To test whether self-efficacy mediated the relation-
ing for associations with self-efficacy (level 1, c’ path:
ship between impulse control and adherence, we speci-
Bc’ ¼ 0.12, SE ¼ 0.03, p < 0.001), suggesting a
fied further a multilevel mediation model (Krull &
partially mediated effect at level 1 (Figure 1).
MacKinnon, 2001; Preacher, Zyphur, & Zhang,
Specifically, self-efficacy accounted for 21% of the
2010; Zhang, Zyphur, & Preacher, 2009), whereby
within-person association between impulse control
the main predictor (i.e., impulse control) and mediator
and adherence.
(i.e., self-efficacy) were entered simultaneously at level
At level 2, there was a significant positive associa-
1 and level 2. At level 1, both impulse control and self-
tion between impulse control and self-efficacy
efficacy were person-centered at individuals’ baseline (Table I; level 2, a path: Ba ¼ 0.50, SE ¼ 0.15,
value. We specified random intercepts and fixed p ¼ 0.001), suggesting that, on average, youth with
slopes. At level 2, baseline values of impulse control higher impulse control at baseline also reported higher
and self-efficacy (centered at the grand-mean) were self-efficacy. There was also a positive association be-
used to test mediation at the sample average (similar tween self-efficacy and adherence (Table I; level 2, b
to cross-sectional estimates). Separate multilevel medi- path: Bb ¼ 0.05, SE ¼ 0.02, p ¼ 0.03), suggesting that
ation models were run for parent and youth report of youth with higher self-efficacy at baseline reported
adherence, and all paths controlled for HbA1c at level higher adherence. However, self-efficacy did not medi-
1 and level 2. Analyses were conducted in Mplus 7.1 ate the association between impulse control and ad-
(Muthén & Muthén, 1998–2012). The MODEL herence at level 2 (B ¼ 0.03, SE ¼ 0.01, 95% CI ¼
CONSTRAINT function was used to estimate the in- 0.002, 0.05, p ¼ 0.07).
direct effects of impulse control on adherence (via self-
efficacy) at both levels. To estimate what proportion Parent-Reported Adherence
of the relationship between impulse control and adher- Impulse control was not associated with parents’ re-
ence was accounted for by self-efficacy, we divided the port of adherence at level 1 (Table I, c’ path: Bc’
coefficient capturing the indirect effect by the total ef- ¼0.01, SE ¼ 0.03, p ¼ 0.67) or level 2 (Bc’ ¼ 0.02,
fect coefficient, and then multiplied by 100. SE ¼ 0.03, p ¼ 0.48), controlling for HbA1c. Likewise,
450 Silva and Miller

Table I. Results from Multilevel Mediation Models

a path b path c’ path Indirect effect

Impulse control! Self-efficacy! Impulse control!adherence, (Ba*Bb)
self-efficacy adherence accounting for self-efficacy

Youth-reported adherence Ba (SE) Bb (SE) Bc’ (SE) B [95% CI]

Level 1 .32 (.10)**** .10 (.02) **** .12 (.03) **** .03 [.01, .05]***
Level 2 .50 (.15) **** .05 (.02)** .03 (.03) .03 [.002, .05]*
Parent-reported adherence
Level 1 .32 (.10) **** .01 (.02) .01 (.03) .001 [.02, .02]
Level 2 .50 (.15) **** .002 (.02) .02 (.03) .004 [.01, .01]

Note. CI ¼ confidence interval; SE ¼ standard error.

*p < 0.10.
**p < 0.05.
***p < 0.01.
****p < 0.001.

Self-Efficacy
***
Ba=0.32 (.10) Bb=0.10 (.02)***

Impulse Control Total effect/Bc=0.14 (.03)*** Adherence

(Youth-report)
Direct effect/Bc’ =0.12 (.03)***
Indirect (mediated) effect= BaBb=0.03 (.01)**

Figure 1. Self-efficacy partially mediated the within-person effect of impulse control on youth-reported adherence. Indirect
(mediated) effect ¼ BaBb ¼ 0.03 (.01)**. Note. **p < 0.01; ***p < 0.001.

self-efficacy was not associated with parents’ report of discrepancy in findings. Despite the fact that both
adherence at level 1 (Table I, b path: Bb¼0.01, studies were based on secondary analysis and relied on
SE ¼ 0.02, p ¼ 0.51) or level 2 (Bb ¼ 0.002, self-reported data, our study sample was relatively
SE ¼ 0.02, p ¼ 0.92), even after controlling for younger and included a wider age range (8–16 years,
HbA1c. There was no evidence of a mediated effect mean ¼ 12.9) compared to the study by Stupiansky
between impulse control and parent-reported adher- and colleagues (ages 17–19, mean ¼ 18.3). Moreover,
ence via self-efficacy (Table I). our test of mediation at the between-person level
specified the intercept, or adherence at age 13.77, as
Discussion the outcome. Therefore, it is possible that self-efficacy
is a mechanism via which impulse control is related to
In the present study, we tested the hypothesis that dia-
individual differences in adherence in late, but not
betes self-efficacy would mediate the relationship be-
mid, adolescence.
tween impulse control and T1D management from
Although aspects of self-regulation—including im-
late childhood to late adolescence using multilevel
pulse control—are often targets of behavioral inter-
modeling. We found that at least some (21%) of the
within-person association between regulatory ability ventions for adolescents, they are challenging factors
(i.e., impulse control) and youth-reported adherence to change because the neural processes that underlie
was accounted for by the degree to which youth felt regulatory abilities are still being refined (Casey,
confidence in their ability to carry out diabetes-related 2019; Steinberg, 2018). Given this developmental con-
tasks. This pattern of findings did not hold for text, it may be appealing to attempt to improve diabe-
parent-reported adherence, which was surprising given tes management by helping adolescents gain
the lack of reporter discrepancy in the current sample. confidence rather than helping adolescents control
Contrary to prior findings (Stupiansky et al., 2013), impulses in socially- or emotionally-laden situations.
we did not find evidence that, on average, youth with Our findings suggest that maturational gains in im-
higher impulse control reported higher adherence, nei- pulse control and increases in self-efficacy are both
ther directly nor indirectly through self-efficacy. positively associated with adherence in youth with
Differences in methodology, including study design T1D. Specifically, time-specific improvements in self-
and analytic approach, may account for the efficacy did not entirely account for the behavioral
Impulse Control and Diabetes Adherence 451

benefits associated with improvements in impulse con- even though in the real world the strength of the asso-
trol. Throughout adolescence, there may be occasions, ciations between impulse control, self-efficacy, and
or periods of time, during which youth successfully adherence is likely to vary across individuals. For ex-
and consistently self-manage T1D because they have ample, impulse control may not develop at the same
made strides in their ability to control impulses. rate for all individuals and this variability is likely to
Likewise, there may be periods of time during which relate to the relative changes in self-efficacy and T1D
an adolescent’s ability to regulate impulses is stagnant management that adolescents may experience.
but the adolescent has gained enough diabetes-related Replicating the current findings in a larger sample and
confidence to motivate increases in self-care behav- exploring whether the mediated effects reported here
iors. Thus, long-term intervention efforts to increase vary across individuals would be a valuable contribu-
both self-efficacy and impulse control may yield nota- tion to this line of research. Lastly, we cannot make
ble benefits within individuals over time. Overall, this causal inferences from the concurrent associations ob-
study shows that cumulative increases in impulse con- served in the current study, and it will be important
trol and self-efficacy coincide with a temporary boost for future research to consider the bidirectionality of
in adherence, in spite of the underlying age-related de- the relationships explored reported here. For example,
cline in T1D management that is observed across this since self-efficacy entails a sense of mastery that often
developmental period (ages 8–16). These temporary results from experience, it is possible that youth feel
successes in youth-reported adherence may be mean-
more self-efficacious following successful management
ingful experiences for an adolescent’s adjustment in
of their illness.
both T1D-related and other domains.
The current study findings underscore the impor-
While the present study has strengths, including its
tance of both impulse control and self-efficacy for
longitudinal design, we interpret our findings with
T1D management in adolescence. Efforts to improve
caution, as they were limited to youths’ report of ad-
adherence in youth with T1D will benefit from consid-
herence. Because the associations only held for youth-
eration of youth impulse control and self-efficacy, as
reported adherence, shared method variance cannot
be ruled out as an explanation for the study findings. both of these factors are positively associated with
Despite the importance of this limitation, our reliance changes in T1D management from late childhood to
on youth-reported predictors is nonetheless supported late adolescence. Environments that consistently foster
by a prior study showing that youth self-reports of im- improvements in impulse control and enrich adoles-
pulsivity were better predictors of various child- and cents with confidence in their own abilities may gradu-
parent-reported youth outcomes (e.g., risk-taking, ag- ally contribute to improvements in self-care behaviors
gression, attentional problems) than were parent within-individuals.
reports (Zapolski & Smith, 2013). In that study,
Zapolski & Smith (2013) showed that even when Supplementary Data
there is convergent validity between reporters, relying
Supplementary data can be found at: [Link]
on youth self-reports of constructs like impulse control
com/jpepsy.
may be more valuable than parent reports for some
youth behaviors. Future replications of this study
should nonetheless include parents’ perceptions of Acknowledgments
their child’s regulatory abilities as predictors of The authors thank the children and parents who participated
youths’ self-care behaviors (e.g., Miller et al., 2013). in this study. They also thank the Diabetes Center for
Another limitation of the present study is that our Children and the Cystic Fibrosis Center at Children’s
measure of regulatory abilities was limited to impulse Hospital of Philadelphia for supporting this program of
control, which is one of several skills that encompass research.
self-regulation (Nigg, 2017). Future research in this
area should incorporate multiple indicators of self-
Funding
regulation, including emotional regulation (Hughes,
Berg, & Wiebe, 2012). It may also be important for This work was supported by T71MC30798 from the
future studies to include indicators of self-regulation Maternal & Child Health Bureau (MCHB) and grant
that are specific to diabetes, such as the Diabetes- #1R01HD064638-01A1 awarded to the senior author from
the Eunice Kennedy Shriver National Institute of Child
Related Executive Functioning scale (Duke, Raymond,
Health and Human Development (NICHD). The content is
& Harris, 2014). The current findings are also limited
solely the responsibility of the authors and does not neces-
by the relatively small sample which did not allow a
sarily represent the official views of the MCHB, NICHD, or
more nuanced investigation. For example, the fixed- the National Institutes of Health.
effects models specified in the current study assumed
that associations would be similar for all individuals, Conflicts of interest: None declared.
452 Silva and Miller

References Hughes, A. E., Berg, C. A., & Wiebe, D. J. (2012). Emotional

processing and self-control in adolescents with type 1 dia-
Babler, E., & Strickland, C. J. (2015). Moving the journey to-
betes. Journal of Pediatric Psychology, 37, 925–934.
wards independence: adolescents transitioning to success-
Iannotti, R. J., Schneider, S., Nansel, T. R., Haynie, D. L.,
ful diabetes self-management. Journal of Pediatric
Plotnick, L. P., Clark, L. M., . . . Simons-Morton, B.
Nursing, 30, 648–660.
(2006). Self-efficacy, outcome expectations, and diabetes
Bagner, D. M., Williams, L. B., Geffken, G. R., Silverstein, J.
self-management in adolescents with type 1 diabetes.
H., & Storch, E. A. (2007). Type 1 diabetes in youth: the
Journal of Developmental and Behavioral Pediatrics, 27,
relationship between adherence and executive functioning.
98–105. doi:10.1097/00004703-200604000-00003
Children’s Health Care, 36, 169–179.
Johnston-Brooks, C. H., Lewis, M. A., & Garg, S. (2002).
Bandura, A. (2004). Health promotion by social cognitive
means. Health Education and Behavior, 31, 143–164. Self-efficacy impacts self-care and HbA1c in young adults
Berg, C. A., King, P. S., Butler, J. M., Pham, P., Palmer, D., with type I diabetes. Psychosomatic Medicine, 64, 43–51.
& Wiebe, D. J. (2011). Parental involvement and adoles- King, P. S., Berg, C. A., Butner, J., Butler, J. M., & Wiebe, D.
cents’ diabetes management: the mediating role of J. (2014). Longitudinal trajectories of parental involve-
self-efficacy and externalizing and internalizing behaviors. ment in type 1 diabetes and adolescents’ adherence. Health
Journal of Pediatric Psychology, 36, 329–339. doi:10. Psychology, 33, 424–432.
1093/jpepsy/jsq088 Krull, J. L., & MacKinnon, D. P. (2001). Multilevel model-
Berg, C. A., Wiebe, D. J., Suchy, Y., Hughes, A. E., ing of individual and group level mediated effects.
Anderson, J. H., Godbey, E. I., . . . White, P. C. (2014). Multivariate Behavioral Research, 36, 249–277. doi:10.
Individual differences and day-t-day fluctuations in per- 1207/S15327906MBR3602_06
ceived self-regulation associated with daily adherence in La Greca, A. M., Follansbee, D., & Skyler, J. S. (1990).
late adolescents with type 1 diabetes. Journal of Pediatric Developmental and behavioral aspects of diabetes man-
Psychology, 39, 1038–1048. agement in youngsters. Children’s Health Care, 19,
Casey, B. J. (2015). Beyond simple models of self-control to 132–139.
circuit-based accounts of adolescent behavior. Annual Littlefield, C. H., Craven, J. L., Rodin, G. M., Daneman, D.,
Review of Psychology, 66, 295–319. doi:10.1146/ Murray, M. A., & Rydall, A. C. (1992). Relationship of
annurev-psych-010814-015156 self-efficacy and binging to adherence to diabetes regimen
Casey, B. J. (2019). Healthy development as a human right: among adolescents. Diabetes Care, 15, 90–94. doi:10.
lessons from developmental science. Neuron, 102, 2337/diacare.15.1.90
724–727. Miller, M. M., Rohan, J. M., Delamater, A., Shroff-Pendley,
De Los Reyes, A., Augenstein, T. M., Wang, M., Thomas, S. J., Dolan, L. M., Reeves, G., & Drotar, D. (2013).
A., Drabick, D. A. G., Burgers, D. E., & Rabinowitz, J. Changes in executive functioning and self-management in
(2015). The validity of the multi-informant approach to adolescents with type 1 diabetes: a growth curve analysis.
assessing child and adolescent mental health. Journal of Pediatric Psychology, 38, 18–29.
Psychological Bulletin, 141, 858–900. Miller, V. A., & Jawad, A. F. (2019). Decision-making in-
Duke, D. C., & Harris, M. A. (2014). Executive function, ad- volvement and prediction of adherence in youth with type
herence, and glycemic control in adolescents with type 1 1 diabetes: a cohort sequential study. Journal of Pediatric
diabetes: a literature review. Current Diabetes Reports, Psychology, 44, 61–71.
14, 532–542. Miyazaki, Y., & Raudenbush, S. W. (2000). Tests for linkage
Duke, D. C., Raymond, J. K., & Harris, M. (2014). The dia- of multiple cohorts in an accelerated longitudinal design.
betes related executive functioning scale (DREFS): pilot Psychological Methods, 5, 44–63. doi:10.1037/1082-
results. Children’s Health Care, 43, 327–344. doi:10. 989X.5.1.44
1080/02739615.2013.870040 Muthén, L.K., & Muthén, B.O. (1998–2012). Mplus user’s
Hanna, K. M., & Guthrie, D. (2000). Parents’ and adoles- guide (7th ed.). Los Angeles, CA: Author.
cents’ perceptions of helpful and nonhelpful support for Nigg, J. T. (2017). Annual research review: on the relations
adolescents’ assumption of diabetes management responsi- among self-regulation, self-control, executive functioning,
bility. Issues in Comprehensive Pediatric Nursing, 24, effortful control, cognitive control, impulsivity,
209–223. doi:10.1080/014608601753260317 risk-taking, and inhibition for developmental psychopa-
Harden, K. P., & Tucker-Drob, E. M. (2011). Individual dif- thology. Journal of Child Psychology and Psychiatry, 58,
ferences in the development of sensation seeking and im- 361–383. doi:10.1111/jcpp.12675
pulsivity during adolescence: Further evidence for a dual Ott, J., Greening, L., Palardy, N., Holderby, A., & DeBell,
systems model. Developmental Psychology, 47, 739–746. W. K. (2000). Self-efficacy as a mediator variable for
Hayaki, J., Herman, D. S., Hagerty, C. E., de Dios, M. A., adolescents’ adherence to treatment for insulin-dependent
Anderson, B. J., & Stein, M. D. (2011). Expectancies and diabetes mellitus. Children’s Health Care, 29, 47–63.
self-efficacy mediate the effects of impulsivity on mari- doi:10.1207/s15326888chc2901_4
juana use outcomes: an application of the acquired pre- Pan, H., Liu, S., Miao, D., & Yuan, Y. (2018). Sample size
paredness model. Addictive Behaviors, 36, 389–396. determination for mediation analysis of longitudinal data.
Hoffman, L., & Stawski, R. S. (2009). Persons as contexts: BMC Medical Research Methodology, 18, 32–43. doi:10.
evaluating between-person and within-person effect in lon- 1186/s12874-018-0473-2
gitudinal analysis. Research in Human Development, 6, Patton, K. A., Gullo, M. J., Connor, J. P., Chan, G. C., Kelly,
97–120. A. B., Catalano, R. F., & Toumbourou, J. W. (2018).
Impulse Control and Diabetes Adherence 453

Social cognitive mediators of the relationship between im- Weinberger, D. A. (1997). Distress and self-restraint as meas-
pulsivity traits and adolescent alcohol use: identifying ures of adjustment across the life span: confirmatory factor
unique targets for prevention. Addictive Behaviors, 84, analyses in clinical and nonclinical samples. Psychological
79–85. Assessment, 9, 132–135.
Preacher, K. J., Zyphur, M. J., & Zhang, Z. (2010). A gen- Wiebe, D. J., Chow, C. M., Palmer, D. L., Butner, J., Butler,
eral multilevel SEM framework for assessing multilevel J. M., Osborn, P., & Berg, C. A. (2014). Developmental
mediation. Psychological Methods, 15, 209–233. doi:10. processes associated with longitudinal declines in parental
1037/a0020141 responsibility and adherence to Type 1 diabetes manage-
Shulman, E. P., Harden, K. P., Chein, J. M., & Steinberg, L. ment across adolescence. Journal of Pediatric Psychology,
(2015). Sex differences in the developmental trajectories of 39, 532–541.
impulse control and sensation-seeking from early adoles- Wiebe, D. J., Berg, C. A., Mello, D., & Kelly, C. S. (2018).
cence to early adulthood. Journal of Youth and Self- and social regulation in type 1 diabetes management
Adolescence, 44, 1–17. during late adolescence and emerging adulthood. Current
Shulman, E. P., Smith, A. R., Silva, K., Icenogle, G., Duell, Diabetes Reports, 18, 23–32. doi:10.1007/s11892-018-
N., Chein, J., & Steinberg, L. (2016). The dual systems 0995-3
model: review, reappraisal, and reaffirmation. Wood, J. R., Miller, K. M., Maahs, D. M., Beck, R. W.,
Developmental Cognitive Neuroscience, 17, 103–117. DiMeglio, L. A., Libman, I. M., . . . Woerner, S. E; T1D
Silva, K., & Miller, V. A. (2019). The role of cognitive and Exchange Clinic Network. (2013). Most youth with type 1
psychosocial maturity in type 1 diabetes management. diabetes in the T1D Exchange Clinic Registry do not meet
Journal of Adolescent Health, 64, 622–630. American Diabetes Association or International Society
Stupiansky, N. W., Hanna, K. M., Slaven, J. E., Weaver, M., & for Pediatric and Adolescent Diabetes clinical guidelines.
Fortenberry, D. (2013). Impulse control, diabetes-specific Diabetes Care, 36, 2035–2037. doi:10.2337/dc12-1959
self-efficacy, and diabetes management among emerging Zapolski, T. C., & Smith, G. T. (2013). Comparison of par-
adults with type 1 diabetes. Journal of Pediatric Psychology, ent versus child-report of child impulsivity traits and pre-
38, 247–254. doi:10.1093/jpepsy/jss110 diction of outcome variables. Journal of Psychopathology
Vaid, E., Lansing, A. H., & Stanger, C. (2018). Problems Behavioral Assessment, 35, 301–313.
with self-regulation, family conflict, and glycemic control Zhang, Z., Zyphur, M. J., & Preacher, K. J. (2009). Testing
in adolescents experiencing challenges with managing type multilevel mediation using hierarchical linear models:
1 diabetes. Journal of Pediatric Psychology, 43, 525–533. problems and solutions. Organizational Research
Steinberg, L. (2018). Regulating our enthusiasm for Methods, 12, 695–719. doi:10.1177/1094428108327450
self-regulation interventions. JAMA Pediatrics, 172,
520–522. doi:10.1001/jamapediatrics.2018.0376