Temporal Dynamics of Health and Well Being A.11

ORIGINAL ARTICLE
Temporal Dynamics of Health and Well-Being:

A Crowdsourcing Approach to Momentary
Downloaded from http://journals.lww.com/psychosomaticmedicine by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbs
Assessments and Automated Generation of

IHo4XMi0hCywCX1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 03/26/2024
Personalized Feedback
Lian van der Krieke, PhD, Frank J. Blaauw, MSc, Ando C. Emerencia, PhD,
Hendrika M. Schenk, MSc, Joris P.J. Slaets, PhD, Elisabeth H. Bos, PhD,
Peter de Jonge, PhD, and Bertus F. Jeronimus, PhD
ABSTRACT
Objective: Recent developments in research and mobile health enable a quantitative idiographic approach in health re-
search. The present study investigates the potential of an electronic diary crowdsourcing study in the Netherlands for
(1) large-scale automated self-assessment for individual-based health promotion and (2) enabling research at both the
between-persons and within-persons level. To illustrate the latter, we examined between-persons and within-persons asso-
ciations between somatic symptoms and quality of life.
Methods: A website provided the general Dutch population access to a 30-day (3 times a day) diary study assessing 43 items
related to health and well-being, which gave participants personalized feedback. Associations between somatic symptoms
and quality of life were examined with a linear mixed model.
Results: A total of 629 participants completed 28,430 assessments, with a mean (SD) of 45 (32) assessments per participant.
Most participants (n = 517 [82%]) were women and 531 (84%) had high education. Almost 40% of the participants (n = 247)
completed enough assessments (t = 68) to generate personalized feedback including temporal dynamics between well-being,
health behavior, and emotions. Substantial between-person variability was found in the within-person association between
somatic symptoms and quality of life.
Conclusions: We successfully built an application for automated diary assessments and personalized feedback. The applica-
tion was used by a sample of mainly highly educated women, which suggests that the potential of our intensive diary assess-
ment method for large-scale health promotion is limited. However, a rich data set was collected that allows for group-level and
idiographic analyses that can shed light on etiological processes and may contribute to the development of empirical-based
health promotion solutions.
Key words: ecological momentary assessment, idiographic, dynamic effects, quality of life, person-tailored, self-assessment.
INTRODUCTION participants are sampled, based on the implicit assumption

that such results are informative for individual population
P sychological science and health promotion are rooted in

the nomothetic research tradition in which samples of
participants are investigated to derive aggregated results
members (1), which is known as the “ecological fallacy”
EMA = ecological momentary assessment, MSSD = mean squared

successive difference, VAR = vector autoregressive, VAS = visual
about behavior and psychological processes. Averages are analog scale
subsequently generalized to the population from which the
Supplemental Content
From the University of Groningen (van der Krieke, Blaauw, Emerencia, Schenk, Bos, de Jonge, Jeronimus), University Medical Center Groningen, Uni-
versity Center of Psychiatry, Interdisciplinary Center Psychopathology and Emotion regulation (ICPE), The Netherlands; University of Groningen
(Blaauw), Johann Bernoulli Institute for Mathematics and Computer Science, Distributed Systems Group, The Netherlands; Leyden Academy on Vitality
and Ageing (Slaets), Leiden and University of Groningen, University Medical Center Groningen, The Netherlands.
Address correspondence and reprint requests to Lian van der Krieke, PhD, University of Groningen, University Medical Center Groningen, Uni-
versity Center of Psychiatry, Interdisciplinary Center Psychopathology and Emotion regulation (ICPE), The Netherlands, PO Box 30.001, 9700 RB
Groningen. E-mail: j.a.j.van.der.krieke@umcg.nl
Received for publication September 17, 2015; revision received May 2, 2016.
DOI: 10.1097/PSY.0000000000000378
Copyright © 2016 by the American Psychosomatic Society
Psychosomatic Medicine, V 79 • 213-223 213 February/March 2017

Copyright © 2017 by the American Psychosomatic Society. Unauthorized reproduction of this article is prohibited.
ORIGINAL ARTICLE
(2). The nomothetic approach has been criticized for leading well-being, behavior, and affect, and to (b) automatically
to knowledge that is “true on average” (3). Group-based analyze and visualize dynamic effects as a means to provide
research can be informative for studying variation be- personal information about health and well-being to partic-
tween persons, but aggregated results are not necessarily ipants. We launched a website that enables inhabitants of
applicable at the individual level (1). For example, the Netherlands to assess themselves for 30 consecutive
between-person correlations can diverge from within- days, after which feedback was automatically provided.
person correlations in both magnitude and direction (4). Our aim was to investigate both the potential of our ap-
Eventually, readers of health information may think, proach for enabling research on intensive longitudinal data,
“Well, these means and correlations are interesting, but and for individual health promotion. The latter is examined
what do they mean for me personally?” (5). based on adherence and completion rates, and participant
This transferability (or “generalizability”) of health re- evaluation data. The potential of our approach for col-
search to individuals can be increased by approaches that lecting data and enabling research will be explored by in-
facilitate idiographic (individual-level) analyses, such as vestigating the association between somatic symptoms
ecological momentary assessment (EMA; (6)) techniques, and moment-to-moment quality of life (24). Analysis of
also known as experience sampling (7) or diary studies cross-sectional data has repeatedly shown a negative asso-
(8). Ecological momentary assessment refers to frequent as- ciation between somatic symptoms and quality of life
sessments of variables in (near) real-time during partici- (25,26), but it is unclear whether this association is also
pants' natural flow of life. This reduces recall biases and present at the within-individual level. Analysis of intensive
increases ecological validity (9). Ecological momentary as- EMA data can provide estimates at both the between- and
sessment usually consists of multiple repeated assessments, the within-person level and also provide an estimate of
separated by small time intervals, which allows for fine- the degree of heterogeneity in the within-persons effects.
grained analyses of moment-to-moment fluctuations as We will conduct a between-persons analysis to see whether
well as detection of temporal ordering of effects. Extant earlier group-based results can be replicated with our data,
EMA research unraveled the momentary effects of drug and a within-person analysis to explore associations on an
use, positive or negative effect, quitting smoking, helping individual level, which have not previously been investi-
others, and self-efficacy among others (10–12). gated. We expect that somatic symptoms will be negatively
Within-person associations in EMA data can be identi- associated with quality of life within individuals, but we
fied with, for instance, person-mean–centered multilevel hypothesize that the strength of this association differs be-
models, which yield group-averaged within-person effects tween individuals.
and the degree of variability in these effects (as estimated
by the random slopes) (13). Alternatively, when a sufficient
number of assessment points (>50) is sampled, within- METHODS
person effects can be analyzed by using time series analysis
(14). Individual-based time series analysis generates regression Ethics
The Medical Ethical Committee of the University Medical Center Gro-
coefficients for each participant separately. A specific type of
ningen evaluated the study and judged that it was exempted from review
time series analysis, called vector autoregressive (VAR) model- by the Medical Research Involving Human Subjects Act (in Dutch:
ing (15–17), also reveals the dynamic relationships between WMO) because it concerned a nonrandomized open study targeted at anon-
variables. Vector autoregressive analyses enable researchers ymous volunteers in the general public (number M13.147422).
to identify person-specific sequenced changes in variables
and thus to investigate causal effects (18,19). Participants and Procedure
The current study is part of a larger research project called HowNutsAreTheDutch
Quantitative idiographic research provides a promising av-
(in Dutch “HoeGekIsNL”), henceforth HND. The HND project centers
enue to develop a more person-tailored approach in health around the website www.HoeGekIs.nl (or www.HowNutsAreTheDutch.
promotion (18–21). The increased availability of smartphones com), which was launched on December 19, 2013. HND aims to map
and advanced information and communication technology the mental health of Dutch citizens via online self-assessment on a number
renders self-assessment by means of EMA feasible, rela- of cross-sectional questionnaire modules, including sociodemographic in-
tively cheap, and easy to implement. Nonetheless, the ex- formation, well-being, mood, living situation, physical parameters, person-
ality, optimism, humor, and empathy (see (27) for details). From May 22,
pansive implementation of EMA methods in practical 2014 onward, participants were invited to take part in a diary study,
settings requires nonexperts to be able to deal with the com- consisting of momentary assessments including variables related to quality
plex statistical processes involved in idiographic data anal- of life/well-being, behavior, and affect. This latter part of the HND project
ysis (such as VAR). In the current study, we overcame this is the focus of the present study.
obstacle by using automated analysis (22,23). A crowdsourcing procedure (28,29) was applied to recruit adult inhab-
itants of the Netherlands. The launch of the HND website was announced
The present study reports on a Web-based crowdsourcing on local and national radio broadcast, television, during local podium dis-
approach to (a) collect EMA data allowing for research cussions, in newspapers, and in magazines. The announcement was picked
zooming in and out on associations between quality of life/ up and further disseminated by online blogs, tweets, and other social media.

Intensive EMA and Personalized Feedback
Participants could create a personal account on the website by providing estimated at approximately 3 to 4 minutes per assessment. In a pilot study,
their e-mail address and completing information about their sex, birth year, this duration was found to be acceptable.
birth month (optional), postal code area, and country of residence.
Participants were informed about the procedure and requirements of the
EMA assessments by a video and a digital booklet. To subscribe to the di- Statistical Analysis
ary study, participants had to fulfill the following criteria (and check the ac-
Sample Characteristics and
cording boxes): age 18 or older, availability of a smartphone with data plan,

intention and opportunity to start the study within 5 days, intention to fol- Assessment Adherence
low one's usual living pattern (e.g., not being on holiday or admitted to a Calculation of descriptive statistics and analyses regarding assessment ad-
hospital for scheduled surgery), intention to not miss too many assess- herence were performed in SPSS (version 22, SPSS Inc.). Multiple linear
ments, and approving of one's data being used, anonymously, for scientific regression analyses were performed to predict the number of completed di-
research. Subsequently, participants had to configure their personal settings ary assessments per participant. The predictors used were (a) sex, age, and
for the daily assessments, namely, the start date (within 5 days), the sam- level of education; (b) symptoms of depression, anxiety, and stress (as
pling schedule, and telephone number. Participants could choose the time assessed with the Depression Anxiety Stress Scales (34)), (c) the Ryff
at which they wanted to receive their last measurement of the day, prefera- scales of Psychological well-being (35), and (d) the Big Five personality
bly 30 minutes before their regular bedtime. domains neuroticism, extraversion, openness, agreeableness, and conscien-
Participants had to complete the diary assessments on their smartphone tiousness (assessed with the NEO Five-Factor Inventory-3 (36)). Multivar-
3 times a day for 30 consecutive days, resulting in a maximum of 90 assess- iable logistic regression analyses were fit to test whether participants who
ments. The assessments were prompted on fixed and equidistant moments completed only a few assessments differed from participants who com-
in time (at a 6-hour interval). Participants were asked to fill out the pleted many assessments. Moreover, univariate differences were tested
questionnaire immediately after the prompt or, if this was impossible, with Kolmogorov-Smirnov and t tests. To ensure the robustness of our re-
within 1 hour. After that hour, the questionnaire could no longer be sults, all analyses were bootstrapped (k = 10,000). Results were converted
accessed. At the end of the diary, study participants received personal- to standardized effect sizes Cohen d (37) to enable interpretation and com-
ized feedback on the HowNutsAreTheDutch website, including a per- parison with existing literature. The standardized effect sizes were interpreted
sonal network showing dynamic relationships between variables. Five as small, from 0.20; medium, 0.50 or greater; and large, 0.80 or greater (38).
iPads were allotted among participants who completed at least 85% of
the diary assessments. The present study uses the data that have been
provided before December 13, 2014. Between-Persons and
Within-Persons Associations
A mixed linear model was fitted to investigate between-persons and within-
Measures persons associations between somatic symptoms (“I experience physical
The diary questionnaire contained 43 items. It combined items from existing
discomfort”: not at all (0) to very much (100)) and quality of life (“How
and validated questionnaires and a few newly created items. We assessed sub-
are you doing right now?”: very bad (0) to very good (100)). Long-time
jective well-being, sleep, mood, anxiety, depression, physical activity, physi-
trends were removed for each individual separately (13,39). The person
cal discomfort, self-esteem, worrying, loneliness, mindfulness, context
means were added as a predictor to estimate the between-subjects associa-
(location, social company, activities), and the appraisal of this context, stress-
tions; the person-mean–centered scores were entered as a predictor to esti-
ful events, time pressure, the feeling one makes a difference, laughing, and
mate the within-subject association (13,40). Models with random intercepts
being outdoors (27). All questionnaire items and literature references are pre-
and random slopes were fit to estimate the heterogeneity in the within-
sented elsewhere (27). Additionally, participants could define a personal item
subject effect. A variance-components covariance structure at the first level
that they felt relevant to their situation. This item could be chosen from a list
was found to be optimal, according to the Bayesian Information Criterion.
of options or could be self-created during the configuration of personal set-
Additionally, the temporal instability of somatic symptoms and quality of
tings. Examples of personal items were “I worry a lot” or “I smoked a lot
life was calculated by means of the mean squared successive difference
since the last assessment.” All items except categorical ones were rated on
(MSSD (41); see also Supplementary Table S1, Supplemental Digital Con-
a visual analog scale ranging from zero to 100, with appropriate labels at
tent 1, http://links.lww.com/PSYMED/A314).
the extremes and middle of the scale, and the middle as default positive. To
answer a question, the slider had to be moved.
The diary assessments were preceded by a baseline assessment consisting
of the items of the Positive And Negative Affect Schedule (30,31), the Quick
Per-Person Analysis of Dynamic Relationships
Inventory of Depressive Symptomatology (32) and 2 extra items retrieved Between Variables
from the Inventory of Depressive Symptomatology (33) to assess anxiety/ Vector autoregressive analysis was used to analyze the EMA data for each
panic symptoms. participant separately to generate variable networks for personalized feed-
back (15–17). Vector autoregressive models contain a system of regression
equations in which all variables are treated as endogenous variables, mean-
Design Features of the Diary Assessments ing that they function as both outcome and predictor. Vector autoregressive
Assessments were sent to participants as a hyperlink in a short text mes- analysis can be conducted without a prior hypothesis about the direction of
sage. The hyperlink referred to a website with a responsive design, namely, the association between variables. When variation in one variable (Y ) can
that scaled to the appropriate size corresponding to the resolution of differ- be better explained by previous measures of Yand another variable (X ) than
ent types of smartphones. The diary items were divided over 7 pages, rang- by previous measures of Y alone, variable X is said to Granger-cause vari-
ing from 2 to 8 items per page. Most items could be rated with a slider on a able Y, and this temporal sequence can be tested with the Granger causality test
visual analog scale to maximize the degree of variation in response options, (42). Vector autoregressive analyses can thus elucidate the temporal ordering
as variation is prerequisite for meaningful time series analyses. Categorical of dynamic relationships between 2 or more variables and indicate putative
items had a radio button format such that participants could only choose causal associations. An extensive description of the VAR technique can be
one option. When participants skipped an item, a notification appeared in found elsewhere (15–17). An application called Autovar was crafted to auto-
the form of a red line frame, and all items on the page had to be completed matically estimate and optimize VAR models and to determine the relation-
before one could turn to the next. Questionnaire completion time was ships between variables (22,23). These relationships were visualized in

ORIGINAL ARTICLE
FIGURE 1. Timeline and enrollment of study participants. 1 = Launch (cross-sectional part of ) HowNutAreTheDutch project; 2 =
Publication first newsletter; 3 = Publication magazine of national newspaper (Volkskrant Magazine) dedicated to HowNutsAreTheDutch;
4 = Launch diary study; 5 = Publication second newsletter; 6 = Publication third newsletter; 7 = Presentation in an academic setting for a
general audience. Color image is available only in online version (www.psychosomaticmedicine.org).
network images of the models using the Data-Driven Documents-3 JavaScript Adherence and Completion Rates
library (http://d3js.org). We accommodated for potential level differences due
The diary study participants completed a total of 28,430 as-
to the larger difference between the evening and morning sample by including
a dummy variable for the morning. Missing data were imputed using the sessments; with a mean (SD) of 45 (32) assessments each
EM-imputation algorithm implemented by the Amelia library (https://cran.r- (range, 0–90). Assessments completion time was 3 minutes
project.org/web/packages/Amelia/vignettes/amelia.pdf). (median), and the assessments were completed within 11.6
minutes (median) after the prompt. Figure 2 shows that the
number of completed assessments had a bimodal distribution
RESULTS across participants. Almost half of the participants (n = 278
[44%]) completed less than 45 assessments (mean [SD], 12
Timeline and Study Enrollment [13]), thus quit the study early (henceforth denoted as “early
A timeline of the HND project is presented in Figure 1, with quitters”). The other half (n = 351 [56%]) of the participants
the percentage of participants subscribing to the HND pro- adhered to the study and completed 45 assessments or
ject as a whole and to the diary study in particular. The fig- more (henceforth denoted as “adherers”), with a mean (SD)
ure shows several forms of advertising that influenced the of 73 (11) assessments.
number of enrollments. A multivariable linear regression model was fit to test
whether early quitters differed from adherers with regard
Sample Characteristics to personal characteristics, which included 356 participants
During the 7-month study period, 629 persons participated who had provided data on the variables of interest in the
in the diary study. Of these persons, 532 (85%) had also cross-sectional part of the HND study. Surprisingly, no ef-
participated in the earlier cross-sectional part of the HND fects were observed (all p values ≥ .15; see Supplementary
project, which is 4% of the total cross-sectional sample Table S2, Supplemental Digital Content 1, http://links.lww.
(n = 12,503). The mean age of the diary participants was com/PSYMED/A314). Additionally, a multivariate logistic re-
41 years (range, 18–76) and 82% (n = 517) were women. gression to compare study adherers (n = 228) with early quitters
Compared to the educational level of the general Dutch (n = 128) showed no differences (p ≥ .13; see Supplementary
population (high, 28%; middle, 41%; low education, 31%; Table S2). Finally, univariate differences between early quitters
Dutch Central Bureau of Statistics), the participants were and adherers were tested with Kolmogorov-Smirnov and t tests
highly educated (high, 83%; middle, 13%; and low educated, (see Supplementary Table S3, Supplemental Digital Content 1,
4%). Participants were mainly Dutch and located throughout http://links.lww.com/PSYMED/A314), but no differences were
the Netherlands; 5 participants were Belgian, and one had encountered (all p values ≥ .09).
another nationality. As reported elsewhere (27), diary study Participants could quit the diary study passively (by
participants were on average 5.4 years younger than the ceasing to respond to assessment prompts) or actively (by
cross-sectional sample of HND, more often women, higher unsubscribing from the study), and the active quitters were
educated, and they reported lower well-being. asked to check 1 of 3 prespecified reasons for quitting.

FIGURE 2. Distribution of completed assessments across participants.

Most quitters were “passive quitters.” Of the 79 active quit- physically active; (ii) when A was more physically active,
ters (mean age, 39; 79% women), 46% (n = 36) indicated A spent more time outside; (iii) when A spent more time
that the diary study was “too intensive,” 2% (n = 2) checked outside, A had more humor; (iv) when A reported more hu-
the option “I can no longer comply to the study criteria due mor, A felt less down; and (v) when A felt more down, A
to a change in my daily rhythm,” whereas 52% (n = 41) had felt lonelier. Furthermore, the size of the nodes indicate that
“other reasons” (e.g., quit the study because the time sched- physical activity, spending time outside, humor, and feeling
ule did no longer fit or too many assessments were missed). down had most connections with other variables (also
known as degree), and the thickness of the line indicates
Automatically Generated Feedback that the relationship between physical activity and spending
and Evaluation time outside was the strongest relationship. The relation-
All participants who completed at least 65% (t = 59) of the ships in the concurrent network show that a participant's
diary assessments (n = 302) received basic personalized mood, emotions, and behaviors are related to each to other
feedback consisting of several graphs (time plots, bar at the same moment in time but do not indicate the temporal
graphs, pie charts, and scatter plots) and explanatory text in- ordering of these associations.
cluding information about positive and negative affect, The dynamic network models on the right illustrate the
sleep, location, social company, time pressure, physical dis- temporal ordering of the relationships between variables
comfort, self-esteem, worrying, special events, physical ac- (namely, how variables affect each other across time). The
tivity, and the personal item. An example of the graphs dynamic network for person A shows that (i) an increase
presented in the basic feedback is shown in Figure 3. in physical activity preceded (thus predicted) an increase
Participants who completed at least 75% (t = 68) of the di- in spending time outside; (ii) an increase in humor predicted
ary assessments (n = 247) also received 2 personal networks an increase in physical activity and a decrease in loneliness;
showing concurrent and dynamic relationships between their (iii) an increase in feeling down predicted an increase in
mood, health behaviors, and emotions over time. Initially, physical activity; and (iv) enacting a healthy lifestyle did
our threshold for receiving advanced feedback was 85%, not predict, and was not predicted, by any of the included
but during the study, we lowered the threshold to 75% be- variables.
cause this proved sufficient to create meaningful networks After completion of the diary study, participants who re-
(14). For one participant, no personal networks could be gen- ceived feedback were invited to complete an evaluation
erated because of extremely low variability in variable values form about the study on the HND website, which 102 par-
(namely, the response pattern of this participant was highly ticipants did (mean [SD]age, 46 (14) years and 73,5%
similar across assessments). Example networks of 3 random (n = 75) women). Results are presented in Table 1.
participants are presented in Figure 4.
The network models in the left part of Figure 4 show the Between-Persons and Within-Person Associations
concurrent relationships between variables. For instance, To explore the value of our collected EMA data, a mixed
the concurrent network of person “A” shows that (i) when linear model was fit to test the association between somatic
A was enacting a healthy lifestyle, A was also more symptoms and moment-to-moment quality of life. At the

ORIGINAL ARTICLE
FIGURE 3. Sample of basic feedback. Activities and behavior ranked by perceived pleasantness. The size of the nodes indicates the
frequency of the activity; the bigger the node, the more frequent a participant reported to be engaged with that activity. Color image is
available only in online version (www.psychosomaticmedicine.org).
between-persons level, a significant negative association Potential for Automated Self-Assessment

between somatic symptoms and quality of life was ob- and Health Promotion
served (B = −0.25; p < .001). At the within-person level, We created an application that automates self-assessment
the mixed linear model also showed significant, but slightly protocols, complex statistical procedures, and feedback.
weaker, negative associations between somatic symptoms By using an online platform, we enabled all inhabitants
and quality of life (B = −0.22; p < .001). The random slope of the Netherlands to access our diary study, although only
indicated significant heterogeneity in the strength of the small groups of people were actively informed about the
within-person association between somatic symptoms and study. An active approach seems to have been crucial in
quality of life (variance, 0.02; p < .001). Additionally, the the recruitment, given that the number of subscriptions in-
within-person fluctuation in the experience of somatic symp- creased noticeably after presentations and other advertising
toms was much larger than the within-person fluctuation in activities by the research team. As many of our participants
momentary quality of life (MSSD, 395.8 versus 257.3; see were female, relatively high educated (both >80%) and they
Supplementary Table S1 for the MSSDs of all diary items). enrolled via self-selection, the sample is not representative
of the general Dutch population. Moreover, only a fraction
DISCUSSION of the total Dutch population (17 million) participated. This
Our crowdsourcing approach resulted in the recruitment of indicates that the potential of diary studies for national
629 participants who completed more than 28,000 diary as- health promotion through self-assessment, in the format
sessments covering a range of health-related items. For that we applied, is limited. In fact, our format appealed most
those participants who completed a sufficient number of di- to higher educated, and we may assume highly motivated,
ary assessments, a personal network of dynamic associa- women. A comparison with the broader HND sample indi-
tions between affect, cognitions, and behaviors could be cates that their motivation may have been partly driven by a
successfully generated. lower sense of well-being. The overrepresentation of women

FIGURE 4. Advanced feedback. Concurrent (left) and dynamic (right) networks. Striped (red in online version) nodes represent variables
that tend to be perceived as negative (e.g., loneliness, feeling down). Dotted (green) nodes represent variables that tend to be perceived as
positive (e.g., humor, feeling cheerful). The blank (blue) node represents the personal variable that participants could choose to add to the
diary assessment. This variable could either be negative or positive and could be different for each participant. The size of the node
indicates its relative importance (i.e., the bigger the node, the more relationships that a variable has with other variables; also known as
“degree”). The lines represent the relationship between variables; the thickness indicates the strength of the relationship. A plus refers
to a positive relationship; a minus refers to a negative relationship. The arrowheads (only in the dynamic networks) indicate the
direction of the relationships. Color image is available only in online version (www.psychosomaticmedicine.org).
and higher educated people on health websites and online Adherence to our intensive diary assessments was
programs has been documented before (e.g., (43,44)). Women mixed. On the one hand, there were early quitters who only
tend to be more inclined than men toward seeking health in- briefly took note of the diary study. Barriers to access the
formation in general and therefore also via the Internet (45). study were low, causing many noncommitted participants
Furthermore, higher education has since long been identified to subscribe and drop out, like in many other anonymous
as a characteristic of early adapters to innovations (46,47). Internet-based studies (48). On the other hand, there were

ORIGINAL ARTICLE
TABLE 1. Participants' Evaluation

Item Mean (SD) Range
Overall judgment of the studya 64.5 (14.7) 24–90

Usability (technical/practical)a
64.8 (20.2) 10–100

Comprehensibility of the resultsa 57.1 (24.2) 0–97

To what extent did you benefit from the study?b 44.4 (22.0) 0–89
To what extent did the assessments make you more conscious about what you did/felt/thought?b 61.7 (20.8) 0–100
To what extent did you change your behavior or thinking as a result of the diary study?b 30.9 (22.2) 0–100
a
Rated from very bad (0) to very good (100).
b
Rated from very little (0) to very much (100).
adherers who participated rather conscientiously. For these patients, no longer anonymous participants, can use the di-
participants, the feedback promised upon completion may ary assessments as part of their treatment trajectory to gain
have functioned as a strong incentive; although we cannot insight into their health condition. Their health care pro-
rule out that some participants might have participated to vider may serve as an external motivator.
compete for an iPad. Personal characteristics predicting bet-
ter adherence could not be identified, meaning that we did Potential for Research
not find any personal characteristics that makes people “less From a research perspective, the crowdsourcing approach
suitable” for diary studies. has been very successful. Our largely automated diary study
Personal concurrent and dynamic variable networks has proven to be an important vehicle for sampling experi-
could be generated for all but one participant completing at ences about health and well-being in everyday environments
least 68 diary assessments. A preliminary participant evalua- and circumstances in a substantial number of people. Our
tion suggested that the technical and practical usability and data sample is among the largest ambulatory assessment
the diary study as a whole were adequate, but the compre- samples that we know of. It is a rich source of information
hensibility of the feedback was suboptimal. Arguably, the collected at low cost and low effort.
relatively low comprehensibility scores mainly reflect par- Whereas the biased nature of our sample needs to be
ticipants' judgment of the networks rather than the basic taken into account, it does not necessarily invalidate esti-
feedback. Additional participant feedback and e-mail con- mates of associations between variables. Moreover, the va-
versations indicated that these network models were some- lidity of idiographic models is not touched by selective
times perceived as rather abstract and difficult to grasp. We nonparticipation because in individual-based analyses indi-
are currently working on improvement by redesigning the viduals serve as their own controls. Furthermore, the large
network and modifying the explanation. The evaluation number of assessments for each individual enhances the
data also suggested that the assessments made participants statistical power for eventual group analyses, which may
more conscious about their feelings, actions, and thoughts. also be adjusting for sampling bias using sampling weights.
This is consistent with previous studies (49,50). As diary As to explore the scientific value of our data, we investi-
assessments frequently and repeatedly ask participants for gated the association between somatic symptoms and quality
reflection on their daily life, they may function as a low- of life. On both the between- and the within-person level,
level intervention (50). The latter has not been assessed in more somatic symptoms were associated with a lower qual-
the present study, but recent work suggests that network ity of life, but individuals differed substantially in the
feedback based on diary assessments can improve subjec- strength of this association, as we expected. In addition, the
tive well-being (51). experience of somatic symptoms fluctuated more over time
The uptake of intensive diary assessments combined than quality of life. The presence of the heterogeneity in
with automated and personalized feedback in the HND the within-person association and the differences in temporal
study is not large and broad enough to be used as a general stability of the variables demonstrate the importance of anal-
health promotion approach. Broader usage could possibly yses on the within-person level as a complement to cross-
be realized by using less intensive assessments, with fewer sectional, group-based analyses.
questionnaire items or having sensor technology replacing Idiographic analysis with VAR was performed to create
self-report measurement, as we are currently experimenting dynamic network feedback to each individual participant.
with (e.g., (52)). However, quantifying oneself by means of Extant research has shown that this detailed level can be
diary assessments, in a more or less intensive format, may crucial to derive knowledge about the exact dynamics of
be most promising in a health care context. In this setting, health-related variables that tend to be overlooked in

nomothetic analyses (e.g., (53)). Currently, several idio- for test-retest purposes. Finally, the VAR networks are
graphic studies using the HND diary data are submitted or sometimes hard to interpret when the signs (+ or –) of asso-
in preparation (e.g., (12,54)). To perform a reliable VAR ciations between variables are counterintuitive. Some of the
analysis, we set the lower limit to 68 assessments, and ap- associations may be counterintuitive but still be true (what
proximately 40% of the diary participants met this criterion. holds for most people does not have to hold for the individ-
The data of the other participants can be used for within- ual); in other cases, unmeasured third variables may explain
person analysis with other methods, such as person- the association. Further development of EMA and statisti-
mean–centered multilevel models in which at least 2 completed cal techniques can help to sort this out.
assessments per participant are needed. Moreover, large di-
ary data sets like ours open the possibility for statistical pro- CONCLUSION
cedures such as the Group Iterative Multiple Model We successfully built an application for automated diary as-
Estimation (GIMME) method that enables for the identifi- sessments and personalized feedback on multidirected rela-
cation of both individual-level associations as well as the tionships between health-related variables. The application
identification of commonalities across individuals (55). was used by a small sample of highly educated women,
These methods can help to identify subgroups of people which suggests that the potential of our intensive diary assess-
that are characterized by specific dynamic patterns (thus in- ment method for large-scale health promotion is limited.
duce “general rules”) and still enable for personalized Nonetheless, our crowdsourcing study resulted in the collec-
models that can reveal personal dynamics. tion of a valuable data set that allows for group-level and idi-
Rich and detailed data such as collected in our study can ographic analyses that can shed light on etiological processes
help create an empirical basis for the development of person- and may contribute to the development of empirical-based
alized health promotion solutions. Zooming into the individ- health promotion solutions.
ual level, we may identify key factors driving a person's
symptom patterns, and thus provide clues for targeted inter- Part of the HowNutsAreTheDutch project was realized
vention. Further research should examine whether targeting in collaboration with the Espria Academy. Espria is a health
key variables in individual variable patterns, such as in the care group in the Netherlands consisting of multiple companies
VAR networks, can affect other variables. If so, idiographic targeted mainly at the elderly population. We thank all partic-
data might provide concrete guidance for advice. If it is ipants of HowNutsAreTheDutch for their participation and
known from a participant that he smokes to relax, but the dy- valuable contribution to this research project. We also thank
namic network shows that smoking predicts decrease rather the HowNutsAreTheDutch team, RoQua, Esther Hollander,
than an increase in relaxed mood (see also Person “B” in Ester Kuiper and Chantal Bosman for their contributions.
Fig. 4), he might be offered the advice to perform an activity
Source of Funding and Conflicts of Interest: The study was
that the network does show to have a beneficial effect on re-
funded by a VICI grant (no: 91812607) received by Peter de
laxation (or when the network lacks information about the
Jonge from the Netherlands Organization for Scientific Re-
latter, to try out relaxation exercises). This approach merits
search (NWO-ZonMW) and by the University Medical Center
a perspective of health as people's ability to adapt to their en-
Groningen Research Award 2013, also received by Peter de
vironments and self-manage (56,57).
Jonge. The authors report no conflicts of interest.
EMA Methodology
REFERENCES
There are various methodological issues inherent to EMA
studies including recency effects priming effects, and reac- 1. Molenaar PCM. A manifesto on psychology as idiographic
tions to repeated measurements (58). Because the assess- science: bringing the person back into scientific psychology,
this time forever. Measurement 2004;2:201–18.
ments in the current study were prompted at fixed and 2. Piantadosi S, Byar DP, Green SB. The ecological fallacy. Am
equidistant time intervals, participants knew when a prompt J Epidemiol 1988;127:893–904.
was coming. These time intervals were chosen because 3. Lamiell JT. 'Nomothetic' and 'idiographic': contrasting Windelband's
equidistance is a statistical assumption that needs to be understanding with contemporary usage. Theory and psy-
met to perform VAR analysis. The clear disadvantage is that chology 1998;8:23–38.
4. Tennen H, Affleck G. Daily processes in coping with chronic pain:
participants can anticipate on assessments as a result of methods and analytic strategies. In: Endler NS, Zeidner M,
which the ecological validity can be limited. Moreover, it editors. Handbook of Coping: Theory, Research, Applications.
remains unclear to what extent a 30-day study period is rep- Oxford, England: John Wiley & Sons; 1996:151–77.
resentative for a broader time period, or a comparable pe- 5. Borsboom D, Kievit RA, Cervone D, Hoodet SB. The two
riod later on. Hitherto, we are unaware of studies that disciplines of scientific psychology, or: the disunity of psy-
chology as a working hypothesis. In: Molenaar PCM, Lyra
investigated the stability of diary study results, but this is- MCDP, Chaudhary N, Valsiner J, editors. Dynamic Process
sue is currently studied in the HND project, by inviting par- Methodology in the Social and Developmental Sciences.
ticipants to complete a second series of diary assessments New York: Springer Verlag; 2009. p. 67–97.

ORIGINAL ARTICLE
6. Stone AA, Shiffman S. Ecological momentary assessment depressive disorder seeking primary health care in Spain. Prim
(EMA) in behavioral medicine. Ann Behav Med 1994;16:202. Care Companion J Clin Psychiatry 2008;10:355–62.
7. Csikszentmihalyi M, Larson R. Validity and reliability of the 26. Muñoz RA, McBride ME, Brnabic AJ, López CJ, Hetem LA,
Experience Sampling Method. In: de Vries MW, editor. The Secin R, Dueñas HJ. Major depressive disorder in Latin
Experience of Psychopathology: Investigating Mental Disor- America: the relationship between depression severity,
ders in Their Natural Setting New York: Cambridge Univer- painful somatic symptoms, and quality of life. J Affect
sity Press; 1992. p. 43–57. Disord 2005;86:93–8.

8. Bolger N, Davis A, Rafaeli E. Diary methods: capturing life as 27. Krieke LV, Jeronimus BF, Blaauw FJ, Wanders RB, Emerencia
it is lived. Annu Rev Psychol 2003;54:579–616. AC, Schenk HM, Vos SD, Snippe E, Wichers M, Wigman JT,
9. Conner TS, Barrett LF, Tugade MM, Tennen H. Idiographic Bos EH, Wardenaar KJ, Jonge PD. HowNutsAreTheDutch
personality: the theory and practice of experience sampling. (HoeGekIsNL): a crowdsourcing study of mental symptoms
In: Robins RW, Fraley RC, Krueger RF, editors. Handbook and strengths. Int J Methods Psychiatr Res 2016;25:123–44.
of Research Methods in Personality Psychology. New York, 28. Howe J. Crowdsourcing: a definition. 2006. Available at:
NY: Guilford Press; 2007;79–96. http://www.crowdsourcing.typepad.com/cs/2006/06/
10. Hoeppner BB, Kahler CW, Gwaltney CJ. Relationship be- crowdsourcing_a.html. Accessed October 3, 2015.
tween momentary affect states and self-efficacy in adolescent 29. Brabham DC. Crowdsourcing as a model for problem solving:
smokers. Health Psychol 2014;33:1507–17. an introduction and cases. Convergence: The International Jour-
11. Phillips KA, Epstein DH, Vahabzadeh M, Mezghanni M, nal of Research into New Media Technologies 2008;14:75–90.
Lin JL, Preston KL. Substance use and hepatitis C: an eco- 30. Peeters FPML, Ponds RHWM, Vermeeren MTG. Affectiviteit
logical momentary assessment study. Health Psychol 2014; en zelfbeoordeling van depressie en angst. Tijdschr Psychiatr
33:710–9. 1996:240–250.
12. Snippe E, Jeronimus BF, Aan het Rot M, Bos EH, de Jonge P, 31. Raes F, Daems K, Feldman GC, Johnson SL, Van Gucht D. A
Wichers M. Prosocial people smile: The daily dynamics of an psychometric evaluation of the Dutch version of the responses
upward spiral Submitted. to positive affect questionnaire. Psychologica Belgica 2009;
13. Curran PJ, Bauer DJ. The disaggregation of within-person 49:293–310.
and between-person effects in longitudinal models of change. 32. Rush AJ, Trivedi MH, Ibrahim HM, Carmody TJ, Arnow B,
Annu Rev Psychol 2011;62:583–619. Klein DN, Markowitz JC, Ninan PT, Kornstein S, Manber
14. Box GEP, Jenkins GM, Reinsel GC. Time Series Analysis. R, Thase ME, Kocsis JH, Keller MB. The 16-Item Quick In-
Forecasting and Control. 4th ed. New Jersey: Wiley; 2008. ventory of Depressive Symptomatology (QIDS), clinician rat-
15. Brandt PT, Williams JT. Multiple Time Series Models, Thou- ing (QIDS-C), and self-report (QIDS-SR): a psychometric
sand Oaks, CA: Sage Publications; 2007. evaluation in patients with chronic major depression. Biol
16. Chatfield C. The Analysis of Time Series. An Introduction. Psychiatry 2003;54:573–83.
5th ed. London, UK: Chapman & Hall; 1997. 33. Rush AJ, Gullion CM, Basco MR, Jarrett RB, Trivedi MH.
17. Lütkepohl H. New Introduction to Multiple Time Series Anal- The Inventory of Depressive Symptomatology (IDS): psycho-
ysis, Berlin: Springer Verlag; 2006. metric properties. Psychol Med 1996;26:477–86.
18. Lyubomirsky S, Sheldon KM. Pursuing happiness: the archi- 34. De Beurs E, van Dyck R, Marquenie LA, Lange A, Blonk
tecture of sustainable change. Review of General Psychology RWB. De DASS; een vragenlijst voor het meten van depressie,
2005;9:111–31. angst en stress. Gedragstherapie 2001;34:35–53.
19. Shiffman S, Stone AA, Hufford MR. Ecological momentary 35. Van Dierendonck D. The construct validity of Ryff's scales of
assessment. Annu Rev Clin Psychol 2008;4:1–32. psychological well-being and its extension with spiritual well-
20. Trull TJ, Ebner-Priemer UW. Using experience sampling being. Pers Indiv Diff 2004;36:629–43.
methods/ecological momentary assessment (ESM/EMA) in 36. De Fruyt F, Hoekstra H. NEO-PI-3 persoonlijkheidsvragenlijst,
clinical assessment and clinical research: introduction to the Amsterdam: Hogrefe; 2014.
special section. Psychol Assess 2009;21:457–62. 37. Borenstein M, Hedges LV, Higgings JPT, Rothstein HR. In-
21. Zheng Y, Cleveland HH, Molenaar PC, Harris KS. An alternative troduction to Meta-analysis, Oxford: John Wiley & Sons
framework to investigating and understanding intraindividual Ltd.; 2009.
processes in substance abuse recovery: an idiographic ap- 38. Cohen J. A power primer. Psychol Bull 1992;112:155–9.
proach and demonstration. Eval Rev 2015;39:229–54. 39. Rovine MJ, Walls TA. Multilevel autoregressive modeling of
22. Emerencia AC, Van der Krieke L, Bos EH, De Jonge P, Petkov interindividual differences in the stability of a process. In:
N, Aiello M. Automating vector autoregression on electronic Schafer JL, Walls TA, editors. Models for Intensive Longi-
patient diary data. IEEE J Biomed Health Inform 2016;20: tudinal Data. New York, NY: Oxford University Press; 2006.
631–43. In press. p. 124–47.
23. Van der Krieke L, Emerencia AC, Bos EH, Rosmalen JG, 40. Bolger N, Laurenceau J. Intensive Longitudinal Methods: An
Riese H, Aiello M, Sytema S, De Jonge P. Ecological momen- Introduction to Diary and Experience Sampling Research,
tary assessments and automated time series analysis to pro- New York, NY: Guilford Press; 2013.
mote tailored health care: a proof-of-principle study. JMIR 41. Von Neumann J, Kent RH, Bellinson HR, Hart BI. The
Res Protoc 2015;4:e100. mean square successive difference. Ann Math Stat 1941;
24. Barge-Schaapveld DQ, Nicolson NA, Berkhof J, deVries 12:153–62.
MW. Quality of life in depression: daily life determinants 42. Granger CWJ. Investigating causal relations by econometric
and variability. Psychiatry Res 1999;88:173–89. models and cross-spectral models. Econometrica 1969;37:
25. García-Campayo J, Ayuso-Mateos JL, Caballero L, Romera I, 424–38.
Aragonés E, Rodríguez-Artalejo F, Quail D, Gilaberte I. Rela- 43. Brouwer W, Oenema A, Raat H, Crutzen R, de Nooijer J, de
tionship of somatic symptoms with depression severity, quality Vries NK, Brug J. Characteristics of visitors and revisitors to
of life, and health resources utilization in patients with major an Internet-delivered computer-tailored lifestyle intervention

implemented for use by the general public. Health Educ Res 52. Blaauw FJ, Schenk HM, Jeronimus BF, van der Krieke L,
2010;25:585–95. de Jonge P, Aiello M, Emerencia AC. Let's get physiqual—
44. Van 't Riet J, Crutzen R, De Vries H. Investigating predictors of a platform to integrate sensor technology with ecological
visiting, using, and revisiting an online health-communication momentary assessments. Submitted.
program: a longitudinal study. J Med Internet Res 2010;12:e37. 53. Rosmalen JG, Wenting AM, Roest AM, de Jonge P, Bos EH.
45. Rice RE. Influences, usage, and outcomes of Internet health Revealing causal heterogeneity using time series analysis of
information searching: multivariate results from the Pew sur- ambulatory assessments: application to the association be-
veys. Int J Med Inform 2006;75:8–28. tween depression and physical activity after myocardial in-
46. Rogers EM, editor. Diffusion of Innovations, 3rd ed New farction. Psychosom Med 2012;74:377–86.
York, NY: Collier MacMillan Publishers; 1987. 54. Van Loo H, Jeronimus BF, Wichers M. Differential daily life
47. Brynjolfsson E, Mcafee A. The Second Machine Age. New activities and rewards in subjects who experience high symp-
York, NY: W.W. Norton & Company; 2014. tom levels of anxiety, depression, or both. In preparation.
48. Muñoz RF, Bunge EL, Stephen KC, Schueller M, Bravin JI, 55. Gates KM, Molenaar PC. Group search algorithm recovers ef-
Shaughnessy EA, Pérez-Stable EJ. Massive open online inter- fective connectivity maps for individuals in homogeneous and
ventions. A novel model for delivering behavioral-health ser- heterogeneous samples. Neuroimage 2012;63:310–9.
vices worldwide. Clin Psychol Sci 2015. 56. Huber M, Knottnerus JA, Green L, van der Horst H, Jadad
49. Nikles CJ, Clavarino AM, Del Mar CB. Using n-of-1 trials as a clin- AR, Kromhout D, Leonard B, Lorig K, Loureiro MI, van
ical tool to improve prescribing. Br J Gen Pract 2005;55:175–80. der Meer JW, Schnabel P, Smith R, van Weel C, Smid H.
50. Quenter A, Bludau HB, Friedrich HC, Schild S, Riepe T, How should we define health? BMJ 2011;343:d4163.
Zipfel S. Use of electronic diaries within treatment of obesity 57. Duckworth AL, Steen TA, Seligman ME. Positive psychol-
and binge eating disorder: first experiences. Mobile Comput- ogy in clinical practice. Annu Rev Clin Psychol 2005;1:
ing in Medicine 2002:25–33. 629–51.
51. Roekel van E, Vrijen C, Heininga VE, Masselink M, Bos EH, 58. Stone AA, Shiffman SS, De Vries MW. Ecological momen-
Nederhof E, Oldehinkel AJ. The effects of a lifestyle advice tary assessment. In: Schwarz N, Kahneman D, Diener E,
and tandem skydive on pleasure in anhedonic young adults. editors Well-Being: the Foundations of Hedonic Psychology.
Submitted. New York, NY: Russell Sage Foundation; 1999. p. 26–39.


Temporal Dynamics of Health and Well Being A.11

Uploaded by

Copyright:

Available Formats

Temporal Dynamics of Health and Well Being A.11

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Temporal Dynamics of Health and Well Being A.11

Uploaded by

Copyright:

Available Formats

ORIGINAL ARTICLE

Temporal Dynamics of Health and Well-Being:

Assessments and Automated Generation of

INTRODUCTION participants are sampled, based on the implicit assumption

P sychological science and health promotion are rooted in

EMA = ecological momentary assessment, MSSD = mean squared

Psychosomatic Medicine, V 79 • 213-223 213 February/March 2017

Psychosomatic Medicine, V 79 • 213-223 214 February/March 2017

cording boxes): age 18 or older, availability of a smartphone with data plan,

Psychosomatic Medicine, V 79 • 213-223 215 February/March 2017

Psychosomatic Medicine, V 79 • 213-223 216 February/March 2017

FIGURE 2. Distribution of completed assessments across participants.

Psychosomatic Medicine, V 79 • 213-223 217 February/March 2017

between-persons level, a significant negative association Potential for Automated Self-Assessment

Psychosomatic Medicine, V 79 • 213-223 218 February/March 2017

Psychosomatic Medicine, V 79 • 213-223 219 February/March 2017

TABLE 1. Participants' Evaluation

Overall judgment of the studya 64.5 (14.7) 24–90

64.8 (20.2) 10–100

Comprehensibility of the resultsa 57.1 (24.2) 0–97

Psychosomatic Medicine, V 79 • 213-223 220 February/March 2017

Psychosomatic Medicine, V 79 • 213-223 221 February/March 2017

sity Press; 1992. p. 43–57. Disord 2005;86:93–8.

Psychosomatic Medicine, V 79 • 213-223 222 February/March 2017

Psychosomatic Medicine, V 79 • 213-223 223 February/March 2017

You might also like