You are on page 1of 9

Journal of Consulting and Clinical Psychology Copyright 1999 by the American Psychological Association, Inc.

1999, Vol. 67, No. 5, 746-754 0022-006X/99/S3.00

Multilevel Daily Process Designs for Consulting and Clinical Psychology:


A Preface for the Perplexed

Glenn Affleck Alex Zautra


University of Connecticut Health Center Arizona State University

Howard Tennen and Stephen Armeli


University of Connecticut Health Center
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

The next 3 articles in this issue use multilevel statistical procedures to analyze data collected in daily
process studies of (a) stress and coping, (b) binge eating, and (c) chronic pain experience. Important
differences in the methods and procedures of these studies illustrate the many options available to
investigators and data analysts. This article serves as a preface to help readers who are new to these
studies' methodology appreciate their novel contributions to the literature in consulting and clinical
psychology. Four frequently asked questions are addressed concerning the design of daily process
studies, the distinctive meaning of a within-person finding, the possibility that self-monitoring studies are
measurement reactive, and complexities in the use of multilevel statistical procedures for analyzing
person-day data sets.

Patient readers of the next three articles (Feldman, Downey, & becoming increasingly popular options for the statistical treatment
Shaffer-Neitz, 1999; Marco, Neale, Schwartz, Shiftman, & Stone, of experimental (Kenny, Kashy, & Bolger, 1997) and longitudinal
1999; Steiger, Gauvin, Jabalpurwala, Seguin, & Stotland, 1999) (Bryk & Raudenbush, 1987) data.
will come to share their authors' enthusiasm for the fresh insights To hedge our bet, we offer this preface on the multilevel
to be gained from multilevel analyses of daily process studies in analysis of data from daily process designs. It keeps our promise
clinical and consulting psychology. We dare this prediction not to the inviting editor to provide a nontechnical commentary on
merely because we have been cultivating the same field and basic issues. Those wanting a more advanced treatment of this
welcome this chance, however indirectly, to praise our own labors. topic can read several outstanding commentaries and guides on
We promise even those new to these studies' methods and proce- these methods written for and by behavioral scientists (e.g., Jac-
dures that they will acquire novel information about topics that are card & Wan, 1993; Kenny et al., 1997; Schwartz & Stone, 1998;
well represented in the annals of this journal: stress and coping Stone, Kessler, & Haythornthwaite, 1991; West & Hepworth,
(Marco et al., 1999), chronic pain (Feldman et al., 1999), and 1991). We can also recommend several well-informed articles and
eating disorders (Steiger et al., 1999). Readers will find excellent books on multilevel applications (Hofmann, 1997; Kreft & de
illustrations of data-collection strategies and methods that are Leeuw, 1998; Littell, Milliken, Stroup, & Wolfinger, 1996; Singer,
rapidly gaining currency in the literature. They will also be treated 1998). To these authors, we are indebted for helping us reduce at
to the kinds of findings that make these labor-intensive studies least some of our own perplexities about these approaches. We
worthwhile. The articles themselves also provide an introduction borrow our organization from Kreft and de Leeuw's method of
to methods for multilevel, mixed-model data analysis, which are posing frequently asked questions about multilevel models. The
questions we have chosen are ones we have asked ourselves
repeatedly and heard repeatedly from colleagues and students.
Glenn Affleck and Howard Tennen, Department of Community Medi-
They concern the design of daily process studies, the distinctive
cine, University of Connecticut Health Center; Alex Zautra, Department of meaning of a "within-person" finding, the possibility that self-
Psychology, Arizona State University; Stephen Armeli, Department of monitoring studies are "measurement reactive," and complexities
Psychiatry, University of Connecticut Health Center. in the use of multilevel statistical techniques for analyzing daily
Partial support for the preparation of this article came from the National process data.
Institute of Arthritis, Musculoskeletal, and Skin Diseases (Grant
R0141687) and the National Institute on Alcohol Abuse and Alcoholism
(Grants T32-AA07290 and P50-AA03510). How Should a Daily Process Study Be Designed?
We are grateful to Charles Hall and Bruce Smith for their help with
software applications for multilevel modeling. Feldman et al.'s (1999) daily diary study, Steiger et al.'s (1999)
Correspondence concerning this article should be addressed to Glenn experience-sampling study, and Marco et al.'s (1999) ecological
Affleck, Department of Community Medicine (MC-6205), University of momentary assessment study each qualifies as a daily process
Connecticut Health Center, Farmington, Connecticut 06030. Electronic study. This means that their repeatedly measured dependent vari-
mail may be sent to affleck@nsol.UCHC.edu. ables are thought to change in meaningful ways from day to day
746
MULTILEVEL DAILY PROCESS DESIGNS 747

(or within a day) and are measured prospectively at daily (or pain ratings by having participants compare their own pain to their
within-day) intervals. Each major dependent variable in the fol- average pain level. Unfortunately, because this report invokes
lowing investigations is arguably a daily process; they include temporal comparison processes, which themselves are implicated
mood, stressful occurrences and their appraisals, chronic pain in coping with pain (Tennen & Affleck, 1997), its meaning is
intensity, social interactions, self-critical thoughts, disordered eat- complicated. Echoing previous findings (Affleck, Urrows, Tennen,
ing episodes, and coping strategies. Together, the authors submit a & Higgins, 1992), Marco et al. (1999) found that some important
compelling rationale for the frequent measurement of these pro- coping responses did not vary for many study participants, who
cesses: (a) to capture as closely as possible their "real-time" used them too often or too rarely to permit within-person analyses.
occurrences or moments of change; (b) to reduce recall bias; (c) to Further, with such a short study time frame, their study participants
mitigate some forms of confounding by using participants as their did not experience enough "stress episodes" corresponding with
own controls; and (d) to establish temporal precedence to their chronic strain classification (e.g., marital events for partici-
strengthen causal inference, to name a few reasons (see also pants reporting chronic marital problems) to allow what would be
Tennen & Affleck, 1996). a most revealing analysis of the implications of strain-stressor
The procedural differences among the studies illustrate options matching. And they must of course qualify their "null effects" of
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

available to daily process researchers.1 The studies differ first in coping by keeping open the possibility that coping takes longer
This document is copyrighted by the American Psychological Association or one of its allied publishers.

how they scheduled participant responding. Feldman et al. (1999) than a few hours to work its consequences.
use an interval-contingent method, asking participants to describe In this vein, we remain as concerned as two of us were nearly a
the day once each night. Marco et al. (1999) use a signal- decade ago (Tennen, Suls, & Affleck, 1991) that daily process
contingent method, which asks participants to respond to randomly researchers will forsake descriptive inquiry in favor of a rush to
scheduled "beeps" emitted by a palmtop computer. Steiger et al. sophisticated hypothesis testing. Descriptive studies of the tempo-
(1999) use an event-contingent method, which requires partici- ral organization of daily experiences are needed not only to inform
pants to supply information whenever a social interaction occurs. design decisions but also to serve as a starting point for theoretical
Wheeler and Reis (1991) and Stone et al. (1991) have offered formulations. Yet we still know very little about the mundane
useful guidelines for selecting among these options. events and experiences in people's daily lives or how they are
A second option concerns the actual method by which partici- patterned over time. Techniques of exploratory data analysis
pants record their responses. Both Steiger et al. (1999) and Feld- (Tukey, 1977) can be insightfully applied to the description of
man et al. (1999) use the conventional "low-tech" paper-and- daily experience data. Our illustration of how chronic pain inten-
pencil method. In the Marco et al. (1999) study, participants enter sity ratings are distributed and temporally organized (Affleck,
their responses in a PSION palmtop computer, which time-stamps Tennen, Urrows, & Higgins, 1991; Tennen & Affleck, 1996) could
the data entry.2 The many advantages of computer-assisted daily help answer questions concerning the optimal frequency and du-
reporting have been summarized by Stone and Shiftman (1994) ration of measurement of this daily process.
and include both the encouragement of compliance with demand-
ing reporting schedules and the ability to rule out "faked compli- Why Go to All This Trouble to Uncover a
ance" with these schedules (Litt, Cooney, & Morse, 1998). Al-
Within-Person Relation?
though this innovative technology may not be for everyone—
whether because of limited resources, technical expertise, a low Because each group of authors collected data repeatedly and
tolerance for "cybernetic malfunction," or study samples for whom frequently from participants, they were able to examine relations at
the cognitive or real-life demands of this method may not be the within-person level over time (across-days or within-day ob-
suitable—it does undercut many of the "quality control" criticisms servations). The analytic procedures used in these studies, however
of experience-sampling designs. This is not to imply that partici- different, furnish estimates of average within-person relations
pants in Steiger et al. and Feldman et al.'s studies cannot be trusted between the intensively measured study variables while control-
to follow the response protocol, but computer-assisted methods do ling for the influence of between-persons differences in the vari-
allow investigators to "trust but verify." ables. To appreciate the novel contributions made by these articles
A third, and most critical, option that differentiates the three to the literature on stress and coping, chronic pain, and eating
studies is the frequency and duration of self-monitoring. Ideally, disorders requires a firm understanding of the difference between
and with adequate resources to do so, we would all like to measure an across-persons association and a within-person association. We
daily processes as frequently and for as long as possible, but there
are natural limits associated with participant burden and, perhaps,
1
measurement reactivity. Steiger et al.'s (1999) average participant See Wheeler and Reis (1991), Stone et al. (1991), and Stone and
furnished between 3 and 4 daily reports for an average of 18 days. Shiffman (1994) for a more detailed discussion of these alternatives.
2
Feldman et al. (1999) gathered 28 end-of-day reports. Marco et al. Many published studies have demonstrated the feasibility and reliabil-
(1999) followed their participants for 2 days, obtaining 10 reports ity of the PSION Organizer (PSION, Ltd., London, England) as a pro-
grammable self-monitoring device (e.g., Affleck et al., 1998; Affleck,
a day. These authors obviously differ in how they weigh the
Urrows, Tennen, Higgins, & Abeles, 1996; Apter et al., 1997; Carney,
trade-off in measurement frequency and duration. Tennen, Affleck, DelBoca, & Kranzler, 1998; Hyland, Kenyon & Howarth,
Readers can note the implications of the frequency-duration 1993; Kamarck et al., 1998; Stone et al., 1994, 1998; Swendson et al., in
decision for the instrumentation and analysis of the three studies. press; Totterdell, Reynolds, Parkinson, & Briner, 1993). These investiga-
For example, because Feldman et al. (1999) had prejudged their tions include assessments of daily or momentary coping, mood, social
participants' pain intensity to vary so little over a month's time, interactions, drinking, smoking, fatigue, sleep quality, pain, and asthma
they decided, in effect, to "create" within-person variance in daily symptoms.
748 AFFLECK, ZAUTRA, TENNEN, AND ARMELI

and several other investigators have commented extensively on from being stress free to being stressful. Such bivariate pro-
this fundamental distinction (e.g., Larsen & Kasimatis, 1991; cesses simply cannot be ascertained through between-persons
Shiffman & Stone, 1998; Tennen & Affleck, 1996; West & Hep- associations.
worth, 1991). Nevertheless, we continue to witness enough con- This design also provides the means of uncovering important
fusion about it among our colleagues (not to mention the occa- between-persons differences in within-person processes. Each
sional manuscript or grant application reviewer) to warrant another study features the ability to assess individual differences in some
reminder. form of "event reactivity" in its most natural within-person for-
We suspect that the unique contributions of within-person stud- mulation. Feldman et al. (1999) shed light on "pain reactivity"
ies remain underappreciated in part because of our discipline's (i.e., the extent to which an individual's increase in pain is matched
tradition of nomothetic inquiry or because of our search for lawful by a subsequent increase in negative mood) as well as "mood
relations among variables across populations of persons. The no- reactivity" (i.e., the consequences of negative mood for pain in-
mothetic approach can be contrasted with the idiographic ap- tensification). Marco et al. (1999) examine "stressor reactivity,"
proach, which pursues relations among variables within a single that is, an individual's tendency to experience emotional distress
individual over time or across situations and has occasionally been following a stressful occurrence. Steiger et al. (1999) are able to
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

used in some extraordinarily well-documented and revealing case describe reactivity to negative social encounters by assessing their
This document is copyrighted by the American Psychological Association or one of its allied publishers.

studies (Hokanson, Tate, Niu, & Stader, 1994; Potter & Zautra, within-person relations with changes in self-criticism and negative
1997). The overall success of the nomothetic approach to psycho- mood. The capacity to differentiate individuals according to these
logical research has had an untoward effect. Until quite recently, and other reactivity processes is a compelling rationale for within-
within-person inquiries have been eschewed or have at least shown person analysis.
benign neglect. Perhaps the most impressive finding in Steiger et al.'s (1999)
The three studies by Feldman et al. (1999), Marco et al. (1999), study is that eating disorder status accounts for significant varia-
and Steiger et al. (1999) illustrate the hybridization of nomothetic tion in the within-person relation between negative social encoun-
and idiographic inquiry. In mixing these approaches, investigators ters and self-criticism. Individuals meeting criteria for a current
ask, "Are there relations between variables within individuals over eating disorder experienced more negative encounters and self-
time that generalize across individuals or that relate to differences criticism than those with a previous history of eating disorders and
between individuals!" The virtues of combining these approaches those with no eating disorder history. Yet the two eating disorder
were first advanced by Seymour Epstein (1983) and reinforced by groups remained alike in their inclination to exhibit self-criticism
Randy Larsen in his investigations of daily emotions and physical following a negative social interaction, and both of these groups
symptoms (e.g., Larsen & Cowan, 1988; Larsen & Kasimatis, differed from the group with no history of eating disorders. This
1991). suggests a subtle vulnerability associated with having had an
We, like most, have been tempted to draw within-person infer- eating disorder history that could not have been readily detected in
ences from across-persons associations. For example, in early more conventional research designs. Steiger et al. provide an
cross-sectional studies of stressful life events, correlations between intriguing model for research on other psychiatric disorders, in-
the number of events and health problems were taken by many to cluding more fine-grained tests of the hypothesis that a prior major
mean that when a person experiences a stressful event, he or she depressive episode leaves a psychological "scar" (Lewinsohn,
would be more likely to become ill. No such inference can be made Hoberman, & Rosenbaum, 1989).
without observing people when they are under stress and when
they are not. An across-persons correlation, moreover, can depart Are Self-Monitoring Studies Measurement Reactive?
markedly from a within-person correlation. We cannot emphasize
this enough. Tennen and Affleck (1996) and Kenny, Bolger, and How we answer this frequently asked question has important
Kashy (in press) have demonstrated that between-persons and implications for the external validity of daily process studies. Put
within-person correlations can differ not only in magnitude but simply, do the findings we obtain generalize to persons who do not
also in direction and that a statistically significant positive spend their days monitoring their behavior so often and so exact-
between-persons association can emerge when not a single indi- ingly? Each of this issue's daily process studies asks a lot of its
vidual in the group shows a positive within-person association! participants: the nightly completion of questionnaires about one's
Yet it is precisely intraindividual questions that investigators and pain, social support, and mood for a month (Feldman et al., 1999);
clinicians often need to answer. the description, approximately every 40 min for 2 days, of one's
Although it is worth knowing that Feldman et al.'s (1999) stressful problems, ways of coping with them, and mood (Marco et
chronic pain patients who experience more pain also suffer al., 1999); and the recording for 2 or 3 weeks of every instance of
more emotional distress, it is equally important to know if some a social interaction lasting more than 10 min and answering
or all of them are more distressed on a more painful day. questions about its character and one's self-referential thoughts,
Perhaps even more important from the standpoint of causal mood, and disordered eating behaviors (Steiger et al., 1999).
inference is whether they were more distressed the day after a Marco et al. legitimately question whether such intensive self-
more painful day. Indeed, Feldman et al.'s interest in intrain- monitoring of behaviors, thoughts, emotions, and psychologically
dividual lagged relations is what transforms the associations significant events may have "reactive effects." Does it alter the
between these variables into processes; so do Steiger et al.'s very states and experiences that are under study? Although calling
(1999) interest in social interactions occurring prior to, and for studies on the reactivity of self-monitoring methods, they cite
following, a binge eating episode and Marco et al.'s (1999) encouraging evidence from two studies of chronic pain monitoring
interest in mood changes during times of the day that move that show no average changes in pain level over the study period.
MULTILEVEL DAILY PROCESS DESIGNS 749

Steiger et al. admit the possibility of their procedure's "demand 15% and less numerous for 35%. These findings imply that aver-
characteristics" but discount their systematic effect on the results. age trend estimates could obscure possible opposing reactive ef-
We are not so sure. fects of intensive self-monitoring.
Those of us engaged in the study of everyday life are keenly The nature and determinants of individual differences in the
aware of the limits of laboratory investigations; indeed, we use trending of daily experience data require more attention. Formu-
these to our advantage in extolling the virtues of our field research lating hypotheses for such studies can be aided by debriefing
designs. Hence, we raise the problem of ecological validity: the participants about their reactions to these novel procedures. We
danger that "laboratory studies actually exaggerate the contours of have learned that some chronic pain patients, by virtue of keeping
life, to which our theories are therefore more likely to give unrep- track of pain, events, and their coping efforts, were able to discover
resentative prominence" (Duck, 1980, p. 232). their own "response contingencies." Some said they began to
Should we not be equally concerned about the possible effect of assess the efficacy of their coping strategies by recording these
intensive self-monitoring on findings we then use to ground our strategies and their pain at the end of each day. Could daily
theories about the ecology of daily life? After all, are not behavior measurement have a pain-reducing effect for these individuals yet
therapists so confident that recording behavior regularly actually potentiate pain for others? We believe so. Just as downward trends
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

changes the behavior being recorded that they use self-monitoring in pain may reflect changed coping strategies in response to daily
This document is copyrighted by the American Psychological Association or one of its allied publishers.

as a therapeutic method? The success of this technique, (e.g., monitoring, so upward trends may result from an untoward effect
Rehm et al., 1981), even if limited to those who are motivated to of daily monitoring on the availability of preferred coping strate-
change their behavior, underscores the methodological challenge gies. During debriefing, some patients said that they found mon-
of potential reactivity in daily experience studies. itoring extremely aversive precisely because it forced them to keep
There is some evidence that measurement reactivity associated track of their pain. These were individuals who had found that their
with self-monitoring may not be a significant problem (Vuchinich, most effective coping strategy was distraction (Endler & Parker,
Tucker, & Harlee, 1988) and that such effects may be minimal 1990), which is a common daily strategy for reducing the intensity
when more than one behavior is recorded and when participants of chronic pain (Affleck et al., 1992; Keefe et al., 1997). Dispo-
have no opportunity to review their daily recordings (Hayes & sitionally, they might be what Miller (1987) called "blunters." By
Cavior, 1980). This is one more argument for using computerized forcing them to focus on their pain, we may have interfered with
diaries, which conceal prior responses from review. Control a preferred coping strategy, which in turn may have increased
groups have not been included in daily process studies, but it is diary keepers' pain over time.
hard to see how they alone could provide strong evidence for or Such anecdotal evidence suggesting heterogeneous measure-
against measurement reactivity. One could compare the data on ment reactivity cannot be mitigated just by controlling for linear
one nightly questionnaire completed by a group that did not trends in data analysis. We must at least admit the possibility that
self-monitor with the last day's data of a self-monitoring group, by asking people to concentrate so intently on their experiences so
but such a comparison places considerable weight on one day's as to provide the best data possible, we might even create or
experience. consolidate the very relations we then "discover." An emerging
The presence or absence of linear trends in the data set, men- experimental (Croyle & Uretsky, 1987; Salovey & Birnbaum,
tioned by Marco et al. (1999) are a good lead, but just that. Of 1989) and field study (Larsen & Cowen, 1988) literature shows
course, the absence of trends would be inconsistent with a reactive that self-focused attention may itself create linkages between
effect of self-monitoring. Time trends in the data suggest reactiv- stress, somatic symptoms, and emotional distress. For now, our
ity, but these could be explained by other factors, including the hypotheses are no more than informed speculation. We hope that
decay in responding from fatigue or boredom with the procedure future investigators will examine in imaginative ways the possi-
over time (Stone et al., 1991)—not so much a "reactivity" problem bility that time-intensive research designs interact with the ecology
as a problem of differential reliability in self-reports across of daily life and for whom it may or may not.
time—or the partial mapping of a longer cycle of secular change
in the states and events under study (Affleck et al., 1991). In our How Should the Data Be Analyzed?
own daily experience studies, we have found significant linear
trends. For example, 100 moderate-to-heavy drinkers who re- As a starting point, readers need to be familiar with the vernac-
corded "real-time" drinking in a palmtop computer for 30 days ular of multilevel data analysis. Each of the following three studies
evidenced a significant average decline in drinking over time, works with a multilevel data matrix. This means that the investi-
despite having been asked not to change their drinking habits gators collected data from at least two sampling units hierarchi-
during the study. cally arranged such that one is "nested" in the other. With suffi-
Besides addressing the general problem of reactive measure- cient amounts of these data in hand, they could draw reliable
ment, we should also ask whether there are individuals who are statistical inferences about relations at each level of the analysis as
more or less reactive to these procedures. In our chronic pain well as across the levels. In the parlance of multilevel modeling,
studies, we have examined individual differences in the trending of the repeated observations (the daily reports, the replies to the
daily reports of pain, mood, and stressful events (Affleck et al., "beeps," the responses following social interaction episodes) are
1991; Affleck, Tennen, Urrows, & Higgins, 1994). The differences called Level 1 (or lower level) variables. These Level 1 observa-
were intriguing. Pain intensity ratings rose significantly over time tions are organized within Level 2 units, which in these studies
for 25% of the participants but dropped significantly for 30%, constitute persons. Their characteristics are called Level 2 (or
negative mood ratings increased for 10% and decreased for 40%, upper level) variables. A critical feature of a multilevel model is
and reports of stressful daily events became more numerous for that responses from the Level 1 observations are likely to be
750 AFFLECK, ZAUTRA, TENNEN, AND ARMELI

nonindependent for each Level 2 unit. Whereas each study reports independent variable, we could not generalize our findings to
analyses of relations between Level 1 observations, only Steiger et levels other than those selected expressly for study. This is not a
al. (1999) analyzed the effects of a specific Level 2 variable— problem for some designs. When gender is the factor, we don't
eating disorder status—on Level 1 observations (intercepts) and mind limiting our inferences to men and women! However, when
their interrelations (slopes). Their article reveals many instances of "participants" is made a fixed factor, and unless we have the whole
such "cross-level" interactions.3 population of interest under wing, the restricted inference exacts a
Do multiple levels of the data structure need to be taken into toll on our conclusions. We hope it is also clear that because this
account in some way if a sound within-person analysis is desired? approach ignores interindividual differences, it is also limited in its
We can say yes, and without qualification. Each group of authors ability to examine Person X Observation interactions.
did so, albeit using somewhat different approaches. Before at- These limitations are overcome by the more recently developed
tempting to encapsulate the important details of each study's methods that allow data analysts to treat persons and daily obser-
data-analytic strategy, we offer this brief tour of approaches that vations as independent sampling units and as random effects. This
have been used most often by contributors to the daily process general class of statistical methods—which can truly be called
literature. multilevel modeling (MLM)—have been called random-effects
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

One early method, called idiographic regression analysis, cal-


This document is copyrighted by the American Psychological Association or one of its allied publishers.

models (Laird & Ware, 1982), hierarchical linear models (Bryk &
culates separate time-series regression equations for each individ- Raudenbush, 1992), multilevel models (Goldstein, 1987), random
ual in the study (Affleck et al., 1994; Suls, Wan, & Blanchard, coefficient models (de Leeuw & Kreft, 1986), and random regres-
1994; Ward & Leigh, 1993; West & Hepworth, 1991). Although sion models (Bock, 1989).
this procedure quite effectively discloses the range of individual MLM allows one to estimate the Level 1 effects of one daily
differences in the parameters of interest, its capacity to account for observation on another and to model the Level 2 effects of indi-
these differences by Level 2 person variables that do not change vidual differences that may "modify" the strength and direction of
over time (stable traits and characteristics) is a topic of controversy the relation. An important advantage of these statistical methods is
(cf. West & Hepworth's, 1991, advocacy of meta-analysis with their ability to handle "unbalanced" designs in which participants
Jaccard & Wan's, 1993, criticism of this method). Moreover,
provide varying numbers of Level 1 observations (especially crit-
because some Level 1 variables may not change enough for some
ical for Steiger et al.'s, 1999, event-contingent study) and missing
participants (e.g., the individual uses the same coping strategy each
Level 1 observations (especially critical for Marco et al.'s, 1999,
day), it may not be possible to incorporate these in all of the
signal-contingent study and Feldman et al.'s, 1999, interval-
individual regression analyses. Other techniques, which circum-
contingent study).
vent some of these problems, rely on pooling the data, generating
One regrettable shortcoming of most applications of MLM in
a data matrix in which each row represents a single observation
the psychology literature is that the reliability of measurement is
and the number of rows equals the number of persons times the
assumed to be perfect. Kenny and Zautra (1995) have recently
number of observations.
Disregarding the person level and analyzing the pooled obser- developed another method of multilevel analysis that provides a
vations as if they were independent units of analysis would be means of estimating error in repeated measures. This approach,
misleading because it mixes variance due to differences across called TSE (for trait-state-error), uses structural equation model-
persons with variance that is due to changes within persons over ing to partition the variance of measures into three independent
time. To overcome this problem, early studies used a pooled components: (a) trait variance, which identifies stable individual
regression technique known as least squares dummy variable differences in the measures over time; (b) state variance, which
(LSDV) analysis (e.g., Harrington et al., 1993; Suls et al., 1994). identifies differences between persons that change over time; and
This method controls for between-persons differences in the data (c) error in measurement. Zautra et al. (1995) tested the TSE model
pool by entering dummy variables for each person (minus one) to in a longitudinal data set comprising 10 monthly assessments of
identify the source of the observation. This, in effect, centers the myofascial face pain and emotional distress in 110 women. This
pooled data set and furnishes estimates of average within-person approach ignores differences in the slope of the regression line
relations, free of differences between persons in mean levels of the between participants, assuming that the relation between variables
variables (i.e., each person's intercept). A key liability of this is approximately the same for all. How much of a problem this
technique is that it does not allow for an inspection of individual assumption causes in practice has yet to be studied, and the
differences in the within-person relations between variables (i.e., applicability of TSE models in typical daily process studies that
each person's slope). Its ignorance of individual differences in have dozens of repeated measures and some missing data points is
slopes introduces error in the estimates of the within-person uncertain. Statistical innovations in the combination of latent vari-
relations. able modeling with MLM (e.g., MacCallum, Kim, Malarkey, &
There is another key limitation of the LSDV approach that was
not well appreciated in its early applications. The dummy variables 3
used to identify between-persons differences in the data set treated Multilevel models can take other forms and are not limited to Person X
these differences as a fixed factor, meaning that inferences could Observation daily process designs. For example, persons could be Level 1
units nested within Level 2 units of schools, treatment facilities, or any
be made only to the specific sample of persons providing the
other grouping of persons that would make each person's responses more
observations. Those of us who grew up on analysis of variance and like other members of the group than members in other groups. Multilevel
regression techniques did not encounter this problem before, ex- models could include three or more nested levels (e.g., days nested within
cept in reference to levels of an independent variable. From Hays persons and persons nested within settings or days nested within waves of
(1988), we learned that if we did not randomly select levels of the measurement and waves of measurement nested within persons).
MULTILEVEL DAILY PROCESS DESIGNS 751

Kiecolt-Glaser, 1997) will undoubtedly soon find their way into Moreover, Kreft and de Leeuw (1998) cautioned that these
the psychological literature. MLM variance estimation techniques are not as straightforward as
Readers will be treated to some state-of-the-art methods of they might appear, especially when Level 1 slopes are allowed to
MLM in the following articles. Feldman et al. (1999) use a vary randomly. Having occasionally seen evidence in our own
"two-stage" weighted least squares approach to parameter estima- analyses of Kreft and de Leeuw's prediction that the addition of
tion (see Kenny et al., 1997), which uses the more familiar least another random effects variable could actually increase the ran-
squares estimation method. Kenny et al. have demonstrated con- dom variance component in the model, we concur. Such compli-
vergence between the solutions provided by this procedure and cations rarely, if ever, occur with the variance estimates supplied
those reached by hierarchical linear modeling (HLM) maximum by single-level regression analysis.
likelihood estimates (Bryk & Raudenbush, 1992). Marco et al. "Variable centering" is another feature of Steiger et al.'s (1999)
(1999) use the SAS (Statistical Analysis System) PROC MIXED analysis that will require further scrutiny as MLM applications
procedure to model within-person relations as maximum likeli- proliferate in the psychological literature. For OLS regression,
hood estimates. Although intimidating to learn, the SAS applica- Aiken and West (1991) stressed the importance of centering, or
tion has the capacity to eliminate "correlated error" in the residu- subtracting the "grand mean" from the raw score, to avoid incor-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

als, an eventuality that can lead to the misestimation of standard


This document is copyrighted by the American Psychological Association or one of its allied publishers.

rect interpretations of the intercept of a significant interaction term.


errors and hence the parameter's level of statistical significance. This practice, and its implications, is more complicated in MLM
Because time-dependent data are likely to be autocorrelated, mean- because two forms of centering are possible: one using the grand
ing that a score for one observation is correlated more highly with mean across all observations and the other using each individual's
the score on the following observation than with more distant own mean (known as "group" or "cluster" centering). For some of
observations, the appropriate modeling of the residuals is a critical their analyses, Steiger et al. report group centering; this allowed
task for daily process investigators. them to examine how deviations from each person's average
Steiger et al. (1999), using the increasingly popular HLM pro-
interaction tone figured in that person's mood or self-criticism.
gram (Bryk & Raudenbush, 1992), reveal how far a multilevel
Centering interaction tone scores around the grand mean would
modeling strategy can be pushed and introduce a host of novel
have addressed a different question. Several MLM experts (e.g.,
statistical concepts to those trained in traditional methods of data
Hofmann & Gavin, in press; Kreft & de Leeuw, 1998) have rightly
analysis. Particularly important to appreciate are their strategies for
cautioned that centering procedures can influence both the "esti-
estimating "variances explained" by the differences between and
mation and meaning of the hierarchical linear model as a whole"
within persons in the study variables and the incorporation of
(Hofmann, 1997, p. 738).
Level 2 (person) predictors of the variance in individual intercepts
The designation of a variable as "random" or "fixed" figures
and slopes. As is underscored by Kreft and de Leeuw (1998) and
prominently in the data analyses of the three studies. To remind the
Singer (1998), MLM estimates of variance explained should not be
reader, a variable whose effect is "fixed" is not allowed to gener-
confused with the familiar R2 estimate of variance explained in
alize to a "population" of such effects. A variable whose effect is
ordinary least squares (OLS) regression equations. Rather, the
"random" is allowed to generalize to a population of such effects
modeling procedure concentrates on the ability of explanatory
whose mean is zero and whose standard deviation is unknown. In
variables, either lower level or upper level variables, to reduce the
our critique of LSDV regression, we introduced the distinction
residual variances (or the "variance components") of the Level 1
and Level 2 equations. between fixed and random effects and of the importance of allow-
Consider, for example, Steiger et al.'s (1999) finding that eating ing "person" to vary as a random effect. Each study addresses this
disorder status explains 12% of the variance in the relation be- issue and treats "person" as a random variable. This, in itself,
tween interaction tone and self-criticism. They begin by modeling represents progress in the application of MLM techniques in
the variance in the Level 1 self-criticism observations without any psychological research.
predictors and find that roughly 61% of its total variance occurred Marco et al. (1999) and Feldman et al. (1999) also allow in
within persons (and 39% between persons). Then they demonstrate blanket fashion all within-person relations (slopes) to vary as
with a Level 1 model that about 46% of the within-person variance random coefficients, meaning that they too are drawn from, and
in self-criticism could be reduced by knowing the tone of the can be generalized to, a "population" not of persons but of within-
immediately preceding interaction. This value is similar computa- person relations. Steiger et al. (1999) take an empirical approach,
tionally to the R2 estimate in standard regression; that is, it repre- conducting statistical tests of parameter variance to ascertain
sents the difference between the residual variances of the Level 1 whether a random or a fixed effects designation for a given
model without a predictor and the Level 1 model with the predic- variable better fits the data. This is a step in the analysis of
tor, divided by the residual variance for the no-predictor model. multilevel data that can be missed and which may affect the
Note that the 46% figure thus refers to the reduction in the residual results. A fixed effect is more efficient to compute because it does
within-person variance (which itself represented 61% of the total not estimate error in the equations that estimate the size of the
variance), not of the total variance, of self-criticism. Finally they regression coefficient. In doing so, however, the fixed effect may
use a Level 2 model to show that 12% of the slope variance (i.e., misspecify the relationship. As is illustrated by Steiger et al.'s
the variance in self-criticism captured by interaction tone) could be presentation, multilevel analysis software programs such as HLM
explained by eating disorder status. Obviously, the ever-shrinking and SAS PROC MIXED provide the means of testing whether the
denominators in MLM procedures to estimate reductions in vari- fit of the model to the data improves significantly when the
ance components must be followed very carefully to draw the variable or variables are treated as fixed or random. A growing
correct interpretations of effect sizes. literature also is concerned with how estimation of random effects
752 AFFLECK, ZAUTRA, TENNEN, AND ARMELI

for specific variables may be sensitive to minor changes in the way we used to do it, and already understand)?..." Disaffection sets
parameters (e.g., VanLeeuwen, 1997). in. The assumptions are not always met and often seem unjustifiable
One of the many useful suggestions in Singer's (1998) instruc- in the context of a particular problem.... Perhaps the approach
tional guide on how to use the SAS "mixed model" program to becomes identified with a particular school of substantive thought,
and that perspective is found wanting. The high water mark has been
furnish the kinds of estimates provided by HLM software is the
reached and the tide of enthusiasm begins to ebb. But the ark comes
value of writing down the linear equations ahead of time, one for to rest somewhere on the side of the mountain, at an elevation that
each level of analysis. The following articles provide such equa- differs from its former resting place, because in fact, the new meth-
tions, and the importance of doing so prior to analysis of multilevel odology has something genuine to offer, and because it continues to
models cannot be overstated because many problems in estimation evolve as generalizations, extensions, and refinements of understand-
can be avoided early on. One of the advantages of the HLM ing are offered. It is not just the same old thing in a new package,
statistical package is that it requires an explicit statement of the (p. 221)
linear model to be estimated.
We are confident that readers will witness a steady increase in
References
the publication frequency of multilevel daily process studies. The
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

next wave of studies will undoubtedly attempt to transfer these


This document is copyrighted by the American Psychological Association or one of its allied publishers.

Affleck, G., Tennen, H., Urrows, S., & Higgins, P. (1991). Individual
methods to clinical intervention outcome studies with the capacity differences in the day-to-day experience of chronic pain: A prospective
to test genuinely new hypotheses. One illustration of such an daily study of rheumatoid arthritis patients. Health Psychology, 10,
application is a study that three of us have initiated. In that study, 419-426.
we are examining the mechanisms of action in two brief interven- Affleck, G., Tennen, H., Urrows, S., & Higgins, P. (1994). Person and
tions for problem drinkers. Participants are using electronic inter- contextual features of stress reactivity: Individual differences in relations
viewers to describe their fleeting emotions, social pressure to of undesirable daily events with mood disturbance and chronic pain
drink, and interpersonal conflict. After 3 baseline weeks of daily intensity. Journal of Personality and Social Psychology, 66, 329-340.
Affleck, G., Tennen, H., Urrows, S., Higgins, P., Abeles, M, Hall, C.,
self-monitoring, participants are randomly assigned to one of three
Karoly, P., & Newton, C. (1998). Fibromyalgia and women's pursuit of
treatment conditions: (a) coping skills training, which identifies personal goals: A daily process analysis. Health Psychology, 17, 40-47.
new ways of responding to emotions, social pressure to drink, and Affleck, G., Urrows, S., Tennen, H., & Higgins, P. (1992). Daily coping
interpersonal conflict; (b) brief motivational intervention, which with pain from rheumatoid arthritis: Patterns and correlates. Pain, 51,
explores readiness to change problem drinking; or (c) a waiting-list 221-229.
control group. A second 3-week self-monitoring period follows the Affleck, G., Urrows, S., Tennen, H., Higgins, P., & Abeles, M. (1996).
intervention. Although both coping skills training and the motiva- Sequential daily relations of sleep, pain intensity, and attention to pain
tional intervention should result in reduced drinking, we anticipate among women with fibromyalgia. Pain, 68, 363-368.
a greater change in the within-person association between drinking Aiken, L., & West, S. (1991). Multiple regression: Testing and interpreting
and interpersonal conflict, positive and negative emotions, and interactions. Thousand Oaks, CA: Sage.
Apter, A., Affleck, G., Reisine, S., Tennen, H., Barrows, E., Willard, A., &
social pressures to drink for participants treated with the coping
ZuWallack, R. (1997). Perception of airways obstruction in asthma:
skills approach. Another illustration of a multiwave daily process Sequential daily analyses of symptoms, PEFR, and mood. Journal of
design for intervention outcomes is a series of studies in progress Allergy and Clinical Immunology, 99, 605-612.
by Francis Keefe and colleagues on the effects of several inter- Bock, R. (1989). Multilevel analysis of educational data. New York:
vention strategies (e.g., exercise, cognitive-behavioral interven- Academic Press.
tion, and couples-oriented intervention) on chronic pain patients' Bryk, A., & Raudenbush, S. (1987). Application of hierarchical linear
within-person relations among daily pain, mood, and coping (F. models to assessing change. Psychological Bulletin, 101, 147-158.
Keefe, personal communication, November 1998). Such designs Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical linear models.
invite application of three-level hierarchical models, nesting daily Newbury Park, CA: Sage.
observations within multiple occasions of daily data collection Carney, M., Tennen, H., Affleck, G., DelBoca, F., & Kranzler, H. (1998).
within persons (Schwartz & Stone, 1998). Procedures for evalu- Levels and patterns of alcohol consumption using timeline follow-back,
daily diaries, and real-time electronic interviewers. Journal of Studies on
ating within-person changes across time and conditions are high on
Alcohol, 59, 447-454.
the list of emergent statistical applications of MLM (H. Berkoff, Croyle, R. T., & Uretsky, M. (1987). Effects of mood on self-appraisal of
personal communication, November 1998; Raudenbush, Brennan, health status. Health Psychology, 6, 239-253.
& Barnett, 1995). de Leeuw, I., & Kreft, I. (1986). Random coefficient models for multilevel
In closing, readers who are wondering how much more to invest analysis. Journal of Educational Statistics, 11, 57-85.
in learning about MLM applications, or when to do so, may wish Duck, S. (1980). Personal relationships in the 1980's: Toward an under-
to consider its place in Mason's (1995) predictable course of standing of complex human society. Western Journal of Speech Com-
heralded new statistical innovations: munication, 44, 114-119.
Endler, N., & Parker, I. (1990). The multidimensional assessment of
A new statistical formulation is introduced.. .seems to answer a coping: A critical evaluation. Journal of Personality and Social Psy-
widely perceived need. . .and is [then] promulgated as The Answer. In chology, 58, 844-854.
response. . .practitioners accord the methodology high status. Gradu- Epstein, S. (1983). A research paradigm for the study of personality and
ate students learn some portion of it. So do young professors. Older emotions. In M. Page (Ed.), Personality—Current theory and research:
ones consider how close they are to retirement before deciding how 1982 Nebraska Symposium on Motivation (pp. 91-154). Lincoln: Uni-
much to invest in it. Then comes the reaction. It is said that the new versity of Nebraska Press.
method is like some other, previously used method in various ways. Feldman, S., Downey, G., & Shaffer-Neitz, R. (1999). Pain, negative
Critics begin to ask, "Do we really get different answers (from the mood, and perceived support in chronic pain patients: A daily diary
MULTILEVEL DAILY PROCESS DESIGNS 753

study of people with relax sympathic dystrophy syndrome. Journal of unexpected failure to observe effects of coping. Journal of Consulting
Consulting and Clinical Psychology, 67, 000-000. and Clinical Psychology, 67, 000-000.
Goldstein, H. (1987). Multilevel models in educational and social re- Mason, W. (1995). Comment. Journal of Educational and Behavioral
search. New York: Oxford University Press. Statistics, 20, 221-227.
Harrington, L., Affleck, G., Urrows, S., Tennen, H., Higgins, P., Zautra, Miller, S. M. (1987). Monitoring and blunting: Validation of a question-
A., & Hoffman, S. (1993). Temporal covariation of soluble interleukin-2 naire to assess styles of information seeking under threat. Journal of
receptors, daily stress, and disease activity in patients with rheumatoid Personality and Social Psychology, 52, 345-353.
arthritis. Arthritis and Rheumatism, 36, 199-203. Potter, P., & Zautra, A. (1997). Stressful life events' effects on rheumatoid
Hayes, S. C., & Cavior, N. (1980). Multiple tracking and the reactivity of arthritis disease activity. Journal of Consulting and Clinical Psychol-
self-monitoring: II. Positive behaviors. Behavior Therapy, 11, 283-296. ogy, 65, 319-323.
Hays, W. (1988). Statistics (4th ed.). New York: Holt, Rinehart & Winston. Raudenbush, S., Brennan, R., & Barnett, R. (1995). A multivariate hier-
Hofmann, D. (1997). An overview of the logic and rationale of hierarchical archical model for studying psychological change within married cou-
linear models. Journal of Management, 23, 723-744. ples. Journal of Family Psychology, 9, 161-174.
Hofmann, D., & Gavin, M. (in press). Centering decisions in hierarchical Rehm, L., Kornblith, S., O'Hara, M., Lamparski, D., & Romano, J. (1981).
linear models: Theoretical and methodological implications for organi- An evaluation of major components in a self-control therapy for depres-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

zational science. Journal of Management. sion. Behavior Modification, 5, 459-490.


This document is copyrighted by the American Psychological Association or one of its allied publishers.

Hokanson, J., Tate, R., Niu, X., & Stader, S. (1994). Illustration of Salovey, P., & Birnbaum, D. (1989). Influence of mood on health-relevant
concomitant time series analyses in a case of somatoform disorder. cognitions. Journal of Personality and Social Psychology, 57, 539-551.
Cognitive Therapy and Research, 18, 413-437. Schwartz, J., & Stone, A. (1998). Strategies for analyzing ecological
Hyland, M., Kenyon, C., & Howarth, P. (1993). Diary keeping in asthma: momentary assessment data. Health Psychology, 17, 6-16.
Comparison of written and electronic methods. British Medical Journal, Shiffman, S., & Stone, A. (1998). Introduction to the special edition:
306, 487-489. Ecological momentary assessment in health psychology. Health Psy-
Jaccard, J., & Wan, C. (1993). Statistical analysis of temporal data with chology, 17, 3-5.
many observations: Issues for behavioral medicine data. Annals of Singer, J. (1998). Using SAS PROC MIXED to fit multilevel models,
Behavioral Medicine, 15, 41-50. hierarchical models, and individual growth models. Journal of Educa-
Kamarck, T., Shiffman, S., Smithline, L., Goodie, J., Paty, J., Gnys, M., &
tional and Behavioral Statistics, 24, 323-355.
Yi-Kuan Jong, J. (1998). Effects of task strain, social conflict, and
Steiger, H., Gauvin, L., Jabalpurwala, S., Seguin, J. R., & Stotland, S.
emotional activation on ambulatory cardiovascular activity: Daily life
(1999). Hypersensitivity to social interaction in bulimic syndromes:
consequences of recurring stress in a multiethnic adult sample. Health
Relationship to binge eating. Journal of Consulting and Clinical Psy-
Psychology, 17, 17-29.
chology, 67, 765-775.
Keefe, F., Affleck, G., Lefebvre, J., Starr, K., Caldwell, D., & Tennen, H.
Stone, A., Broderick, I., Porter, L., Krupp, L., Gnys, M., Paty, I., &
(1997). Coping strategies and coping efficacy in rheumatoid arthritis: A
Shiffman, S. (1994). Fatigue and mood in chronic fatigue syndrome
daily process analysis. Pain, 69, 35-42.
patients: Results of a momentary assessment protocol examining fatigue
Kenny, D., Bolger, N., & Kashy, D. (in press). Traditional methods for
and mood levels and diurnal patterns. Annals of Behavioral Medi-
estimating multilevel models. In D. Moskowitz & S. Hershberger (Eds.),
cine, 16, 228-234.
Modeling intraindividual variability with repeated measures data:
Stone, A., Kessler, R., & Haythornthwaite, J. (1991). Measuring daily
Method and applications. Englewood Cliffs, NJ: Erlbaum.
events and experiences: Decisions for the researcher. Journal of Per-
Kenny, D., Kashy, D., & Bolger, N. (1997). Data analysis in social
sonality, 59, 575-608.
psychology. In D. Gilbert, S. Fiske, & G. Linzey (Eds.), Handbook of
social psychology (4th ed., pp. 233-265). New York: McGraw-Hill. Stone, A., Schwartz, J., Neale, S., Shiffman, S., Marco, C., Hickcox, M.,
Kenny, D., & Zautra, A. (1995). The trait-state model for multiwave data. Paty, J., Porter, L., & Cruise, L. (1998). A comparison of coping
Journal of Consulting and Clinical Psychology, 63, 52-59. assessed by ecological momentary assessment and retrospective recall.
Kreft, I., & de Leeuw, J. (1998). Introducing multilevel modeling. London: Journal of Personality and Social Psychology, 74, 1670-1680.
Sage. Stone, A., & Shiffman, S. (1994). Ecological momentary assessment
Laird, N., & Ware, J. (1982). Random-effects models for longitudinal data. (EMA) in behavioral medicine. Annals of Behavioral Medicine, 16,
Biometrics, 38, 963-974. 199-202.
Larsen, R., & Cowan, G. (1988). Internal focus of attention and depression: Suls, J., Wan, C., & Blanchard, E. (1994). A multilevel data-analytic
A study of daily experience. Motivation and Emotion, 12, 237-249. approach for evaluation of relationships between daily life stressors and
Larsen R., & Kasimatis, M. (1991). Day-to-day physical symptoms: Indi- symptomatology: Patients with irritable bowel syndrome. Health Psy-
vidual differences in the occurrence, duration, and emotional concomi- chology, 13, 103-113.
tants of minor daily illnesses. Journal of Personality, 59, 387-424. Swendsen, J., Tennen, H., Carney, M., Affleck, G., Willard, A., & Hromi,
Lewinsohn, P., Hoberman, H., & Rosenbaum, M. (1989). Probability of A. (in press). Mood and alcohol consumption: An experience sampling
relapse after recovery from an episode of depression. Journal of Abnor- test of the self-medication hypothesis. Journal of Abnormal Psychology.
mal Psychology, 97, 251-264. Tennen, H., & Affleck, G. (1996). Daily processes in coping with chronic
Litt, M., Cooney, N., & Morse, P. (1998). Ecological momentary assess- pain: Methods and analytic strategies. In M. Zeidner & N. Endler (Eds.),
ment (EMA) with treated alcoholics: Methodological problems and Handbook of coping (pp. 151-180). New York: Wiley.
potential solutions. Health Psychology, 17, 48-52. Tennen, H., & Affleck, G. (1997). Social comparison as a coping process:
Littell, R., Milliken, G., Stroup, W., & Wolfinger, R. (1996). SAS system A critical review and application to chronic pain disorders. In B. Buunk
for mixed models. Cary, NC: SAS Institute. & F. X. Gibbons (Eds.), Health, coping, and well-being: Perspectives
MacCallum, R., Kim, C., Malarkey, W., & Kiecolt-Glaser, J. (1997). from social comparison theory (pp. 263-298). Hillsdale, NJ: Erlbaum.
Studying multivariate change using multilevel models and latent curve Tennen, H., Suls, J., & Affleck, G. (1991). Personality and daily experi-
models. Multivariate Behavioral Research, 32, 215-253. ence: The promise and the challenge. Journal of Personality, 59, 313-
Marco, C. A., Neale, J. M., Schwartz, J. E., Shiffman, S., & Stone, A. 338.
(1999). Coping with daily events and short-term mood changes: An Totterdell, P., Reynolds, S., Parkinson, B., & Briner, R. (1993). Associa-
754 AFFLECK, ZAUTRA, TENNEN, AND ARMELI

tions of sleep with everyday mood, minor symptoms, and social inter- West, S., & Hepworth, J. (1991). Statistical issues in the study of temporal
action experience. Sleep, 17, 466-475. data: Daily experiences. Journal of Personality, 59, 609-662.
Tukey, J. (1977). Exploratory data analysis. Reading, MA: Addison- Wheeler, L., & Reis, H. (1991). Self-recording of everyday life events:
Wesley. Origins, types, and uses. Journal of Personality, 59, 339-354.
VanLeeuwen, D. (1997). A note on the covariance structure in a linear Zautra, A., Marbach, J., Raphael, K., Dohrenwend, B., Lennon, M., &
model. The American Statistician, 51, 140-144. Kenny, D. (1995). The examination of myofascial face pain and its
Vuchinich, R., Tucker, J., & Harlee, L. (1988). Behavioral assessment. In relationship to psychological distress. Health Psychology, 14, 223-231.
D. M. Donovan & G. A. Marlatt (Eds.), Assessment of addictive behav-
iors (pp. 51-83). New York: Guilford Press.
Ward, M., & Leigh, J. (1993). Pooled time series regression analysis in Received December 2, 1998
longitudinal studies. Journal of Clinical Epidemiology, 46, 645-659. Accepted February 3, 1999
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Call for Nominations


The Publications and Communications Board has opened nominations for the
editorships of Behavioral Neuroscience, JEP: Applied, JEP: General, Psychologi-
cal Methods, andNeuropsychology for the years 2002-2007. Michela Gallagher, PhD;
Raymond S. Nickerson, PhD; Nora S. Newcombe, PhD; Mark I. Appelbaum, PhD; and
Laird S. Cermak, PhD, respectively, are the incumbent editors.

Candidates should be members of APA and should be available to start receiving


manuscripts in early 2001 to prepare for issues published in 2002. Please note that the
P&C Board encourages participation by members of underrepresented groups in the
publication process and would particularly welcome such nominees. Self-nominations
are also encouraged.

To nominate candidates, prepare a statement of one page or less in support of each


candidate. The search chairs are as follows:

• Joe L. Martinez, Jr., PhD, for Behavioral Neuroscience


• Lauren B. Resnick, PhD, and Margaret B. Spencer, PhD, for JEP: Applied
• Sara B. Kiesler, PhD, for JEP: General
• LyleE. Bourne, Jr., PhD, for Psychological Methods
• Lucia A. Gilbert, PhD, for Neuropsychology

Address all nominations to the appropriate search committee at the following


address:

[Name of journal] Search Committee


c/o Karen Sellman, P&C Board Search Liaison
Room 2004
American Psychological Association
750 First Street, ME
Washington, DC 20002-4242

The first review of nominations will begin December 6,1999.

You might also like