ONG_Chap-37.3d 06/16/06 9:36pm Page 530

37

Multilevel Modeling of Motivation

A Self-Determination Theory Analysis

of Basic Psychological Needs

Kirk Warren Brown and Richard M. Ryan

Motivation is, by its very nature, a dynamicprocess. Human motives and desires fluctuatein direction and strength according to innerpromptings and environmental contingenciesthat can facilitate or inhibit the satisfaction ofneeds, and desires also vary as individuals nego-tiate their day-to-day lives. This dynamic char-acter of motivation has implications for researchdesigns and analytic approaches used to investi-gate it. Because traditional one-occasion, experi-mental, and longitudinal approaches typicallyprovide very limited opportunities to model thedynamic variability of motivation, researchershave begun to pair nontraditional methods suchas experience sampling with analytic tools such asmultilevel modeling to extend our understandingof motivation as it operates in day-to-day con-texts. In this chapter, we discuss recent researchusing multilevel approaches to modeling thevariable nature of motivation and its conse-quences. Specifically, we focus on recent efforts tounderstand the expression of three basic psy-chological needs specified by self-determinationtheory (SDT; Deci & Ryan, 2000; Ryan, 1995).

Self-Determination Theory and BasicPsychological Needs

Central to SDT is a view of the human organismas active and growth oriented, with a natureendowed with tendencies toward health andwell-being (Ryan & Deci, 2000b). SDT proposesthat inherent to the adaptive design of the or-ganism is a propensity to exercise one’s capa-cities, to seek connection with others, and toexperience behavior as self-organized and con-gruent. It further suggests that these propensitiesare innate psychological needs—for competence(Deci, 1975; White, 1959), relatedness (Baume-ister & Leary, 1995; Reis & Franks, 1994),and autonomy (deCharms, 1968; Ryan, 1993),respectively—whose fulfillment is essential toongoing personal well-being, optimal behavioralfunctioning, and constructive social development(Ryan & Deci, 2000a).Autonomy concerns volitional, self-endorsed

action. When autonomous, individuals perceivethat their behavior is willingly undertakenand concordant with the self. The opposite of

530

ONG_Chap-37.3d 06/16/06 9:36pm Page 531

autonomy is heteronomy, or a perception ofbeing controlled by forces felt to be alien to theself. Autonomy is not equivalent to indepen-dence or separateness, and in fact dependent actsand connecting behaviors are often highly voli-tional or autonomous. Competence pertains tothe experience of mastery, or the effective per-formance of desired behaviors, which is experi-enced under conditions that support efficacy andprovide optimal challenges. Finally, relatednessis the subjective experience of closeness andconnection with others. It is maximized withinrelationships or groups that convey (usuallythrough acts of empathy and caring) the sensethat one belongs with and is significant to theother person or group.Specifying autonomy, competence, and relat-

edness as basic psychological needs implies thatindividuals cannot psychologically thrive orfunction fully without satisfying them, just asthey cannot physically thrive without the basicnutriments of food and water. Likewise, just asthe physical environment can facilitate or ham-per the satisfaction of physical needs, social en-vironments can afford or hinder the fulfillmentof autonomy, competence, or relatedness, withconsequences for individual functioning andwell-being.When a need is thwarted, invariantlythere is a deleterious impact on growth and well-ness, whether or not that need represents one’svalues, goals, or desire. Thus a thrust of SDT isto examine variations in need fulfillment acrosstime, contexts, and cultures, as these impact fluc-tuations in optimal functioning and well-being.A large literature of empirical research has

investigated the social psychology of psycho-logical need support and hindrance (see Deci &Ryan, 2000; Ryan & Deci, 2000b, for reviews).Multilevel modeling approaches have enhancedthe sophistication of SDT research by permittinga detailed examination of questions that reflectthis person-by-situation character of need ful-fillment, and the implications of this dialecticfor psychological well-being. Primary questionsaddressed by this recent research include thefollowing:

� How are basic psychological needs expressedand fulfilled in daily life and other, objectivebehavioral contexts?

� What consequences does need fulfillment, orlack thereof, have for personal (and collective)well-being across time and contexts?

� What psychological supports exist for needsatisfaction?

In this chapter, we discuss research on thebasic needs of autonomy, competence, and re-latedness within an SDT framework, beginningfirst with a brief historical overview of researchmethodology in this area before turning to in-vestigations using multilevel approaches tomodeling the dynamic nature of need regulationand the outcomes of such regulation. We alsodiscuss very recent research examining the roleof mindfulness (Brown & Ryan, 2003) as a keyinternal support for healthy behavioral regula-tion. Finally, we offer suggestions for enhancingthe sophistication of research using multilevelapproaches as a means to both deepening ourunderstanding of need fulfillment and providinga platform for applied research designed to en-hance healthy regulation.

Traditional Approaches to the Studyof Motivation: Cross-sectional,Longitudinal, and Experimental Designs

As in many areas of psychology, including pos-itive psychology, a dominant form of measure-ment in the study of motivation has consisted ofscales and questionnaires designed for single-occasion administration. Many such measures ofmotivation are retrospective in nature, askingindividuals to reflect back upon their past psy-chological experience or behavior. As individualdifference measures, they are designed to assessgeneral dispositions or traits and are assumedto reflect stable individual characteristics thathave important behavioral consequences. Forexample, dispositional measures like the Gen-eral Causality Orientations Scale (Deci & Ryan,1985), which assesses individual differences inautonomy, have demonstrated criterion validity,assessed both cross-sectionally and predictivelyover short- and longer-term time intervals.Despite this, one-occasion assessment relies

upon an assumption of temporal stability, re-gardless of whether the behavior or character-istic measured is theoretically understood to bestable or variable. Yet individuals do not behaveaccording to their trait descriptions in everysituation or even across broad classes of situa-tions. Intraindividual variability of this kindhas been shown both in research that has testedthe situational specificity of motivation and instudies that have repeatedly sampled behaviorover time. For example, La Guardia, Ryan,Couchman, and Deci (2000) found significantvariations in psychological need fulfillment

CHAPTER 37 . MULT ILEVEL MODEL ING OF MOTIVATION 531

ONG_Chap-37.3d 06/16/06 9:36pm Page 532

across different interpersonal relationships. Inturn, these within-person differences in needfulfillment from social partner to social partnerpredicted variability in security of attachment.Although it is often assumed that security ofattachment is a stable individual difference var-iable, contextual variations in support for psy-chological needs appear to create within-personvariations in this presumably stable trait. Otherresearch (e.g., Reis, Sheldon, Gable, Roscoe, &Ryan, 2000) has revealed a marked degree ofintraindividual variability in need fulfillmentacross days of the week. Fluctuations in day-to-day need fulfillment, controlling for individ-ual differences and prior day effects, predictedchanges in daily psychological wellness. Suchstudies showing both contextual and temporalfluctuations thus belie the assumption of sta-bility upon which one-occasion measures arefounded.A reliance upon single-occasion individual

difference measures has had a practical basis, asmotivational dynamics have, until relativelyrecently, been difficult to measure and modelsystematically. Yet the frequent consequence ofthis dominant approach is that variability inmotivation processes has been treated as error,and the assumption of stasis, over situationsand over both short and long periods of time,is tacitly relied upon in such research endeav-ors (Brown & Moskowitz, 1998). A second as-sumption upon which the reliance on one-occasion, retrospective measures rests is thatresponses to them will be reasonably accurate.The validity of this assumption can also bequestioned, given the inherent limitations tohumanmemory in accurately describing past ex-perience and the variety of sources of error thatenter into retrospective reporting, including pastevent salience, idiosyncratic anchor events, cur-rent psychological states, and social desirability.The incorporation of time into research de-

signs enables the study of whether and how acharacteristic will demonstrate stability andchange—over time, as in traditional longitudinaldesigns, or in response to situational factors, asin experimental studies. Both approaches havecontributed significantly to our understanding ofa variety of motivational issues. Yet the rela-tively simple way in which these designs aretypically employed has imposed limitations onthe depth of understanding to be gained aboutmotivational phenomena. Traditional longitu-dinal designs, which use relatively long inter-measurement intervals, limit investigations to

static or slowly changing phenomena. Re-searchers using such designs implicitly or ex-plicitly expect that if change occurs, it will bepermanent or unidirectional, rather than ongo-ing or repeated. Change of the latter sort may beembedded in apparently slowly changing orprogressive phenomena and the picture providedby measures collected at widely spaced intervalsmay mask the nonlinear and short-term lineardynamics inherent in behavior. The use of ex-perimental designs reflects a recognition thatmotivation can vary quickly in response to socialand other situational forces, but such designshave typically been used to examine simplepre-post change. In all three traditional researchdesigns discussed here, the lived and often non-linear nature of motivational processes can eas-ily be missed.

Multilevel Modeling and the ProcessApproach to Motivation

An important advance in the study of motiva-tional processes came with the introduction ofexpanded longitudinal studies that measure be-havior intensively over time. Following a pro-cess approach to research design (Brown &Moskowitz, 1998; Larsen, 1989), this researchminimizes the difficulties associated with one-occasion, retrospective measures by collectingdata as it occurs in laboratory or natural situa-tions. Also, by expanding the traditional lon-gitudinal design to include many reports orsamples of behavior, stability and change can beexamined more closely and precisely. Finally,the process approach still permits the study ofmanipulated effects on behavior but opens thedoor to examining more complex (but often re-alistic) responses to experimental inputs as theyunfold over time. The process approach providesa frame-by-frame motion picture of behavior(Turk, 1994), rather than snapshots at a singletime or at widely spaced time points. A majorbenefit of this approach is not just that it pro-vides a more detailed and, therefore, accuratelook at behavioral processes, but also in un-covering new motivational and other behav-ioral phenomena, because the focus of researchbroadens to include both the structure and thedynamics of behavior over time, including be-havior that is traditionally viewed as disposi-tional in nature (Diener, 1996).The introduction ofmultilevelmodeling (MLM;

e.g., Bryk & Raudenbush, 1992; Kreft & deLeeuw,

532 PART V . INTER INDIV IDUAL DIFFERENCES IN INTRAINDIV IDUAL CHANGE

ONG_Chap-37.3d 06/16/06 9:36pm Page 533

1998) represents a second important contribu-tion to investigations in motivation, in part be-cause it capitalizes on the process approach toresearch design. As Maxwell and Tiberio discuss(chapter 30, this volume), it permits the studyof inter- and intraindividual variability acrosssituations and across time with a more carefulparsing of between- and within-persons vari-ance. It also permits the study of interactionsacross levels of analysis—between groups andpersons, or between persons and occasions—andit allows for the incorporation of time-seriesvariables into analyses, as is discussed below.As this discussion suggests, multilevel modelingcapitalizes on the strengths of each of the tradi-tional methodological approaches to motivationdiscussed above, in its ability to incorporate in-dividual difference effects, experimental groupeffects, and the study of behavioral variability andchange through repeated measurement acrosssituations and time.

Exemplary MLM Research in Motivation

Need Satisfaction and Day-to-DayPersonal Well-Being

Several studies have examined two of the pri-mary questions noted earlier, namely, ‘‘How areneeds expressed and satisfied in daily life andother, objective behavioral contexts?’’ and‘‘What consequences does need fulfillment, orlack thereof, have for personal well-being?’’Sheldon, Ryan, and Reis (1996) examined therole that two of the three basic psychologicalneeds specified by SDT—autonomy and com-petence—have in creating ‘‘good days.’’ In asample of 60 undergraduates, they collected re-ports of need satisfaction at both trait and statelevels, where the latter was assessed once per dayfor 14 days. At the close of each day, individualswere asked to rate the degree of autonomy andcompetence experienced in the three activitiesin which they had spent the most time. Alsomeasured daily were levels of hedonic well-being(pleasant and unpleasant affect), eudaemonic well-being (subjective vitality), and physical heath(common symptoms of illness) experienced overthe course of the day.Using a weighted least squares approach to

MLM, with persons treated as a random effect,Sheldon et al. (1996) entered both trait (Level 2)and state (Level 1) autonomy (AUTO) andcompetence (COMP) need satisfaction terms

into multilevel models predicting each type ofdaily (Level 1) well-being (WB), as well as awell-being composite. Focusing first on Level 2effects, the following basic model was tested:

WBij ¼ bB-PAUTOi þ bB-PCOMPi

þ bB-PSEXi þ ai þ "ij ð1Þ

where i¼ 1, 2, 3, . . . I and j¼ 1, 2, 3, . . . ni.AUTO and COMP are measured once for

each of the I individuals (Level 2), whereasWB is measured ni times for the ith individual(Level 1). The bB-P terms refer to between-personeffects, ai is the intercept for each person (i.e.,each person’s well-being on an average day), andeij refers to the within-person residuals. Sheldonet al. (1996) found that individuals higher intrait autonomy and competence had better days,on average, than those scoring lower in thesedispositions. This result is consistent with apersonological approach, which assumes thatwell-being is determined by stable, enduringqualities of the person. Sheldon et al. (1996) alsotested a model including effects of Level 2 traits,gender, and their interaction, and a number ofLevel 1 predictors, as follows:

WBij ¼ bB-PAUTOi þ bB-PCOMPi

þ bB-PSEXi þ bIntðAUTOi�SEXiÞþ bIntðCOMPi�SEXiÞþ bW-P,iWBLAGij þ bW-P,iIMPORTij

þ bW-P,iWEij þ bW-P,iTIMEij

þ bW-P,iAUTOij þ bW-P,iCOMPij

þ ai þ "ij ð2Þ

In this model, bW-P,i terms refer to within-personeffects in which, for example, people reporthigher well-being at a time when their need forautonomy is more satisfied. Sheldon et al. (1996)found that after factoring in the Level 2 effectsof traits and gender, and the effects of severalLevel 1 variables, including lagged (prior day)well-being (WBLAG), daily activity importance(IMPORT), day of the week (and specifically,a dummy-coded weekend effect,WE), and time ofday (TIME), the days on which individuals ex-perienced more autonomy and competence rela-tive to their own baseline were also better days.In an effort to extend the Sheldon et al. (1996)

findings, Reis et al. (2000) employed the samemethodological and analytic approach to examinethe trait and state effects of all three basic needs—autonomy, competence, and relatedness—upon

CHAPTER 37 . MULT ILEVEL MODEL ING OF MOTIVATION 533

ONG_Chap-37.3d 06/16/06 9:36pm Page 534

day-to-day hedonic, eudaemonic, and physicalwell-being. Insofar as all three represent basicneeds, the authors reasoned that each shouldevidence a unique effect on well-being at bothbetween- and within-person levels of analysis.The Level 2 results of the multilevel analysesshowed that trait autonomy and relatednesspredicted higher day-to-day positive affect, whileautonomy alone predicted higher vitality. Per-sons higher in dispositional competence showedlower daily negative affect and symptoms. AtLevel 1, daily autonomy and competence wereassociated with higher day-to-day levels of alltypes of well-being examined, while relatednesswas most strongly related to higher positive affectand vitality.As a further extension to Sheldon et al.

(1996), Reis et al. (2000) also tested for inter-action effects between trait autonomy, compe-tence, and relatedness and their correspondingstate expressions in the prediction of daily well-being. Theoretically, this investigation wasrooted in the recognition that while the threeneeds are universal (Deci & Ryan, 2000), someindividuals may be more responsive to dailyvariations in activities or situations that affordor hinder need satisfaction. Reis et al. (2000)tested two competing models that would eachindicate that the day-to-day relation betweenneed satisfaction and well-being varied as afunction of traits. A deprivation model positedthat higher levels of daily need satisfaction wouldenhance well-being to the extent that individualtrait levels were low. That is, when one or moreneeds was chronically or dispositionally unsatis-fied, the opportunity to satisfy those needs wouldprovide a greater well-being benefit than to thosehigher in dispositional need satisfaction. In con-trast, a sensitization model proposed that highertrait need satisfaction reflected a higher valueplaced on it, and a greater likelihood that suchpersons would show well-being benefits. Thismodel was tested:

WBij ¼ bB-PAUTOi þ bB-PCOMPi

þ bB-PRELATi þ bW-P,iWBLAGij

þ bW-P,iAUTOij þ bW-P,iCOMPij

þ bW-P,iRELATij

þ bIntðAUTOi�AUTOijÞþ bIntðCOMPi�COMPijÞþ bIntðRELATi�RELATijÞþ ai þ "ij ð3Þ

The bInt coefficients in this equation estimate theextent to which the day-to-day effects of each typeof need satisfaction on well-being (Level 1) varyby trait levels of need satisfaction (Level 2). EachbW-P,i main effect term now represents the effectof day-to-day need satisfaction on well-being forthe ith person if trait levels of need satisfactionequaled zero. Together, the bW-P,i assess the indi-vidual differences in the effects of day-to-day needsatisfaction that cannot be accounted for by traitneed satisfaction (cf. Schwartz & Stone, 1998).The MLM results offered some support for

the sensitization model, in that persons higher ineach trait variable showed higher daily well-being when daily activities supported the satis-faction of each need. Reis et al. (2000) suggestedthat higher dispositional need satisfaction mightreflect a heightened sensitivity to those needsand therefore a greater responsiveness to envi-ronmental events relevant to them. Conversely,lower trait scorers may be comparatively in-sensitive to such events, which then have lessimpact on psychological well-being. This per-son� situation dynamic surrounding the mani-festation and consequences of need satisfaction isconsistent with both Bem and Funder’s (1978)template matching model and Cote and Mos-kowitz’s (1998) behavioral concordance model,which also recognize that responsiveness to sit-uations, and the outcomes of that responsiveness,are guided by dispositional features. However, itis notable that in none of the interactions testedby Reis et al. (2000) was there a negative slopefor those low in trait need satisfaction; therefore,while those more highly ‘‘traited’’ in need satis-faction showed higher well-being when theirneeds were satisfied on a day-to-day basis, thosedispositionally low in need satisfaction also gen-erally benefited, but simply not as much, whenthey had experiences of autonomy, competence,and relatedness. In sum, sensitization effects ac-counted for some added variance in well-being,atop the main effects expected by a model of basicpsychological needs.1

Both of the studies reviewed in this sectionalso shed light on the weekly cyclicity of well-being, or the noteworthy tendency for personsto exhibit higher positive affect and lower neg-ative affect on weekends. The two studies’ resultssuggested that weekends were characterized byrelative increases in fulfillment of basic psycho-logical needs, and the enhanced satisfaction ofthese needs explained a significant proportion ofthis ‘‘weekend effect.’’

534 PART V . INTER INDIV IDUAL DIFFERENCES IN INTRAINDIV IDUAL CHANGE

ONG_Chap-37.3d 06/16/06 9:36pm Page 535

A third study examining need satisfaction andits relation to ongoing well-being focused onyoung athletes in the high-pressure world ofcompetitive sport (Gagne, Ryan, & Bargmann,2003). While recognizing that psychological andphysical benefits can derive from participation inorganized athletics, researchers have also docu-mented mood disturbances and damaged self-esteem in children and teenagers experiencinghigh performance pressure (e.g., Davis, 1997).Of long-standing interest to motivation re-searchers, as well as parents, teachers, coaches,and other motivators, is how healthy motiva-tion, and, by implication, well-being can befacilitated. Gagne et al. applied multilevel mod-eling to examine whether need satisfaction de-rived from athletic involvement was related towell-being outcomes in young female gymnasts.They also examined whether external motiva-tional influences—that is, the motivationalclimate surrounding athletic participation—affected need satisfaction, motivation, well-being,and the relations between them. Specifically,Gagne et al. conducted a 4-week diary study with33 female gymnasts aged 7 to 18 years to inves-tigate the predictive relations of perceived parentand coach motivational supports to the athletes’enduring and daily motivation and need satisfac-tion. They also examined how daily motivationand need satisfaction during practice affected theathletes’ well-being.At the beginning of the study, the athletes

completed a measure of self-regulation for gym-nastics, which assessed the relative autonomy oftheir involvement in the sport. They also com-pleted measures of perceived parental and coachautonomy support. At the beginning of each of 15practices, the gymnasts rated reasons, whichvaried in degree of autonomy, for attending thepractice. At the start and close of each practice,they completed a measure of positive and nega-tive affect, self-esteem, and subjective vitality.A report on need satisfaction during practice wasalso completed at the end of each practice.Gagne et al. (2003) first established, using

aggregated diary data, that the more autonomy-supportive parents and coaches were perceived tobe, the more autonomously motivated was theathletes’ sport involvement over time. Perceivedautonomy support from coaches also predictedhigher levels of need satisfaction over the courseof the 15 practice sessions. Gymnasts whosesport involvement was more autonomouslyregulated showed better well-being over time.

MLM using maximum likelihood estimationfound that the relative autonomy of day level(Level 1) incoming, or prepractice, motivationwas related to incoming well-being, particularlylower negative affect and both higher vitalityand self-esteem. The strongest predictor of Level1 change in well-being from pre- to postpractice(in which prepractice well-being was included inthe equation to control for initial level) waswhether each of the needs for autonomy, com-petence, and relatedness were satisfied duringpractice. Perceived enduring autonomy supportfrom parents and coaches assessed at the begin-ning of the study (Level 2) did not moderatethese results.The results of the multilevel investigations

reviewed thus far suggest three primary con-clusions: First, need satisfaction varies on a day-to-day basis above and below personal baselinelevels, and second, these fluctuations have sig-nificant consequences for day-to-day well-being.Indeed, Sheldon et al. (1996) and Reis et al.(2000) specifically showed that such variationshad influence above and beyond what traits couldexplain. Finally, the significance of the effects ofautonomy, competence, and relatedness at bothbetween- and within-subjects levels of analysisis consistent with SDT’s position that each ofthese three psychological needs has a uniqueand important impact on well-being, whetherexamined as a trait characteristic, as intraindiv-idual variation across time, or as interindivid-ual differences in intraindividual experience.

Value Orientation and CollectiveWell-Being

The interest in active, growth-oriented processesthat characterizes SDT has been applied not onlyto personal well-being but also to social andcollective well-being. Sheldon and colleagueshave explored the role of individual intrinsic andextrinsic value orientations in the preservationof a natural resource (Sheldon & McGregor,2000) and in prosocial behavior (Sheldon, Shel-don, & Osbaldiston, 2000). Intrinsic values aretheorized to reflect basic psychological needsand include the desire for personal develop-ment, affiliation with others, and communityinvolvement (Ryan, Sheldon, Kasser, & Deci,1996). Extrinsic values include the desire forwealth, attractiveness, and fame, and endorse-ment of such values is believed to reflect a dis-connection from, and lack of fulfillment of, basic

CHAPTER 37 . MULT ILEVEL MODEL ING OF MOTIVATION 535

ONG_Chap-37.3d 06/16/06 9:36pm Page 536

psychological needs. Past research has demon-strated that a greater weight given to extrin-sic relative to intrinsic values is associatedwith poorer psychological adjustment, lowersubjective well-being, and health risk behav-ior (e.g., Kasser, 2002; Kasser & Ryan, 1993,1996).Sheldon and McGregor (2000) investigated

whether these value orientations would haveconsequences for collective (ecological) well-being as well. Groups of four participants withsimilar value orientations (all extrinsic and allintrinsic), and groups with mixed orientations(half intrinsic, half extrinsic) were first formed.Each type of group then engaged in a resourcedilemma task involving timber harvest in a na-tional forest. This task represented a ‘‘tragedy ofthe commons’’ paradigm, wherein the procure-ment of resources in the short term could havelong-term effects. It was expected that thosewith relative intrinsic value orientations wouldshow greater care in preserving the resource,given their stronger sense of community andidentification with prosocial norms. Those moreextrinsically oriented were expected to depletethe resource more quickly, given their focus onpersonal material acquisition.The dependent variable in this study was the

total amount of timber that groups, and partic-ipants within groups, would harvest over thecourse of the dilemma task. Each group membersubmitted a harvest bid each year for 25 years, oruntil the resource was depleted. MLM using aweighted least squares approach was used to testthe hypotheses. Because groups were formedusing a median split on individuals’ value ori-entation scale scores, the use of MLM was cru-cial to separate and thereby ‘‘deconfound’’ thetwo levels of effect. The following equation wasconstructed:

HARVESTij ¼ bB-PVOi þ bW-P,iVOij

þ bIntðVOi�VOijÞ þ ai þ "ij ð4Þ

Interestingly, Sheldon and McGregor (2000)hypothesized, and found, contrasting effects ofvalue orientation at the group level (Level 2,represented by bB-PVOi) and at the individual,within-group level (Level 1, represented bybW-P,iVOij). At Level 1, the study showed that,compared to more intrinsic persons, those moreextrinsically oriented made more total profit,because they harvested more timber than theirgroup mates. In contrast, however, Level 2 re-

sults showed that extrinsic groups harvested theleast over time, followed by mixed-value groups,and then intrinsic groups, who harvested themost over time because their forest resource wasnot as quickly depleted by high profit taking.These multilevel results demonstrated that in-dividual self-restraint, rooted in intrinsic values,can benefit a social group, while self-interest canhelp to maximize personal gain but at the ex-pense of collective well-being.In a second study, Sheldon et al. (2000) per-

mitted participants to assort themselves intogroups, who then participated in an iterated pri-soner’s dilemma (PD) game, in which they couldchoose to cooperate or to get ahead over fiverounds of play. Results showed, first, that indi-viduals tended to assort themselves into groupsof similar value orientation—that is, extrinsicswith extrinsics, and likewise for intrinsics. Theprimary multilevel model was structured iden-tically to Equation 4. The analysis revealed,paralleling the results of the previous study, thatindividuals scoring higher in extrinsic valueorientation made more defection choices duringthe PD game and thus scored more points thanthose less extrinsic or those intrinsic in orien-tation. However, Level 2 results showed thatgroups with high mean levels of intrinsic valuesscored more points overall than did groups withlower average levels of intrinsic values. In nei-ther this study nor the Sheldon and McGregor(2000) study did the relation between individ-ual values and outcome vary as a function ofthe mean level of group value orientation (i.e.,there were no Level 1� Level 2 interaction ef-fects). Thus, the effects of individual-level val-ues and group-level values were independentof each other. These results showed, as before,that selfish gain often came at the expenseof collective loss, while altruists accrued bothpersonal and collective benefits, at least whenmaking choices in the company of other proso-cial actors.Theoretically, the multilevel approach taken

to these data support hierarchical conceptions ofadaptive fitness, in which individuals’ fitnesslevel is determined by their behavior withinsocial groups, as well as by the behavior of theaggregate to which those individuals belong (seeSheldon & McGregor, 2000, for review). Theresults of these studies also support the use ofthe multilevel approach to explore the long-termsocial consequences and viability of individualtraits, and the behavior that follows from them(Sheldon & McGregor, 2000).

536 PART V . INTER INDIV IDUAL DIFFERENCES IN INTRAINDIV IDUAL CHANGE

ONG_Chap-37.3d 06/16/06 9:36pm Page 537

The Role of Mindfulness in SupportingHealthy Motivation

As already noted, there has been long-standingresearch interest in how social and other con-textual supports for autonomy, competence, andrelatedness can facilitate self-motivation andwell-being (e.g., Gagne et al., 2003). Recent re-search using MLM has begun to explore thethird primary question of interest to SDT re-searchers noted earlier, namely, ‘‘What psy-chological supports exist for the self-regulationof need satisfaction?’’ This question is importantbecause even when environments provide anoptimal motivational climate, healthy regulationrequires a reflective consideration of one’s be-havior and its fit with personal values, needs, andinterests (Ryan & Deci, 2004). Several influen-tial organismic and cybernetic theories of be-havioral regulation place central emphasis uponattention and awareness, the capacity to bringconsciousness to bear on present events andexperience (e.g., Carver & Scheier, 1998; Deci &Ryan, 1985; Varela, Thompson, & Rosch, 1991).Recent research in our laboratory has focused onthe concept of mindfulness, a quality of con-sciousness that pertains to an open or receptiveattention to and awareness of what is takingplace in the present (e.g., Brown & Ryan, 2003).Several studies have shown that mindfulness

conduces to autonomous behavior (see Brown &Ryan, 2004a, for review). For example, Brownand Ryan (2003) asked samples of students andworking adults to complete a self-report dispo-sitional measure of mindfulness developed bythe authors, called the Mindful AttentionAwareness Scale (MAAS). Individuals then re-corded the relative autonomy of their behaviorat the receipt of a pager signal three times a dayover a 2-week (students) or 3-week (adults) pe-riod. MLM of unconditional means (see Singer,1998) first established that there was significantbetween- and within-subjects variation in day-to-day autonomy. Next, the effects of four im-portant time-series variables, namely, time ofday (TIME), day of study (DAY), weekly cy-clicity (WKCYCLE), and serial autocorrelation(AUTOCORR), were factored in to produce thefollowing equation:

AUTOij ¼ bB-PMAASi þ bW-P,iTIMEij

þ bW-P,iDAYij þ bW-P,iWKCYCLEij

þ bW-P,iAUTOCORRij

þ ai þ "ij ð5Þ

The bW-P,iWKCYCLEij coefficient is representedby a cosine function, which is an alternative wayto assess the day-of-week effect to the dummyvariable approach used by Sheldon et al. (1996)and Reis et al. (2000). Because Brown and Ryan(2003) were not interested in specific day-of-week(e.g., weekend) effects, the trigonometric approachwas used, which allows for fewer terms in modelequations. Also in Equation 5, the inclusion ofbW-P,iAUTOCORRij permitted an assessment offirst-order lagged effects of within-person auton-omy, and is an alternative to the dependent vari-able lag coefficient used by Sheldon et al. (1996)andReis et al. (2000; see Equations 2 and 3).2 In thepresent study, MLM using restricted maximumlikelihood estimation found that more mindfulindividuals in both samples showed higher levelsof autonomous behavior on a day-to-day basis.This study also included a state measure of

mindfulness, such that individuals rated howattentive they were to what was occurringduring the activities that they also rated theirrelative autonomy. Two additional terms wereadded to Equation 5 to examine the role ofstate mindfulness on day-to-day autonomy:bW-P,iMAASij represented the main effect, andbInt(MAASi�MAASij) represented the trait�state mindfulness interaction. MLM showedthat momentary variation in mindfulness wasrelated to fluctuations in autonomy. Specifi-cally, those who were more mindfully attentiveto their activities also experienced more auton-omous motivation to engage in those activities.The effects of trait (Level 2) and state (Level 1)mindfulness on autonomy were independentin this study, indicating that the regulatorybenefits of mindfulness were not limited to thosewith amindful disposition. Other research (Brown& Ryan, 2004b) examining day-to-day compe-tence and relatedness has found similar results,providing evidence that trait and state mindful-ness offer an important, ongoing internal sup-port for psychological need satisfaction.Moreover,those activities that were associated with greatermindfulness not only fostered autonomy; theyalso were related to within-person enhancementsin well-being.

Future Directions in the Modelingof Motivational Processes

MLM techniques present a number of opportu-nities and challenges for future research in mo-tivation, several of which we briefly discuss here.

CHAPTER 37 . MULT ILEVEL MODEL ING OF MOTIVATION 537

ONG_Chap-37.3d 06/16/06 9:36pm Page 538

More Intensive Sampling ofDay-to-Day Behavior

As is apparent from this review, much of the re-search modeling motivational processes fromwithin an SDT framework has used diary meth-ods of various kinds. Optimally, these methodshave the advantage of capturing psychological andbehavioral events and experiences close to thetime of their occurrence, and the use of multiplerecords of experience greatly enhances measure-ment reliability and statistical power. However,most of the diary research inmotivation reviewedhere has not taken full advantage of opportunitiesto model the dynamics of internal experienceand behavior, and increasing the sophistication ofdiary studies would serve important methodo-logical and conceptual purposes. More intensivesampling—multiple times per day rather thanjust once per day and over longer time intervalsthan a week or two—would permit the study ofdiurnal (daily) and septurnal (weekly) dynamic(e.g., cyclic) patterns. For example, there is someevidence for regular daily (Brown & Ryan, 2003)and weekly (Brown & Ryan, 2003; Reis et al.,2000) cyclicity in autonomy and relatedness.Lagged or autocorrelated effects can also be stud-ied in more detail with the collection of measureson a greater than once-per-day basis. Somemultilevel modeling software (e.g., SAS PROCMIXED; SAS Institute, 1992, 1997) is well-suitedto the incorporation of time series variables liketime of day, day of week, day of study, and serialautocorrelation, all characteristics that frequentlyappear in time-serial data and explain meaningfulvariance in day-to-day outcomes (West & Hep-worth, 1991). Even when interest is not specifi-cally in such variation, time-series variablesshould be included in multilevel models of day-to-day behavior for control purposes.More intensive sampling would also permit the

study of social and other situational influences onday-to-day motivational processes and behavior.For example, the role of supports for need sat-isfaction has been understudied in day-to-daycontexts. Given evidence that needs are oftendependent on social contexts for their expressionand fulfillment, and need fulfillment conduces towell-being, it is important to understand howspecific kinds of social and other daily activities,and the reasons for engaging in those activities,contribute to dynamic person� situation modelsof motivation, motivational supports, and thepositive well-being consequences that followfrom them (Reis et al., 2000).

Effects of Interventions to EnhanceSelf-Motivation

The majority of research using diary methods inthe study of motivation, and a variety of otherdomains of behavior, has been interested in ob-taining detailed pictures of naturally occurringbehavior. For reasons already noted, there hasbeen good reason for this. Yet the combina-tion of diary methods and multilevel modelingpresents a rich opportunity to examine the real-world effects of clinical, educational, organiza-tional, and other interventions on motivationalprocesses (cf. Deci, Connell, & Ryan, 1989;Williams, Gagne, Ryan, & Deci, 2002). In in-terrupted time-series designs, for example, atarget behavior is recorded repeatedly before(baseline), sometimes during, and following anintervention. Multilevel analysis can then testwhether (a) the parameters of the behavior(Level 1) change in response to the Level 2intervention, as compared to control or otherintervention conditions, and (b) whether theeffects of the intervention are conditioned bysubject and contextual factors. Smyth, Soefer,Hurewitz, and Stone (1999) provided an exam-ple of a multilevel analysis of interrupted time-series data in the health domain.

Laboratory-Based Investigationsof Motivational Processes

Research in SDT has long been interested in thestudy of motivation under controlled laboratoryconditions (e.g., Deci & Ryan, 1985). Yet littleattention has been paid to motivational pro-cesses unfolding in real-time laboratory con-texts. Sheldon andMcGregor’s (2000) multilevelanalysis of iterated behavioral choices in a re-source dilemma task represents an importanttouchstone for such work, as it models how be-havior can change over time in response to bothpersonality and social influences. As alluded toin our discussion of future intervention research,examination of the effects of experimental ma-nipulations on motivational dynamics is wellsuited to multilevel analysis.

Maximizing the Use of MultilevelAnalyses

Beyond the research design possibilities thatMLM affords are opportunities to make morecomplete use of multilevel analyses and statistics.For example, aside from using model parameter

538 PART V . INTER INDIV IDUAL DIFFERENCES IN INTRAINDIV IDUAL CHANGE

ONG_Chap-37.3d 06/16/06 9:36pm Page 539

estimates, motivation researchers are encour-aged to use effect variance estimates, which re-flect the estimated variance accounted for bymodeled effects within the population. Covari-ance estimates indicate how much the modelintercepts and slopes vary across subjects (orschools, organizations, etc.). Significant covari-ance in a slope parameter, for example, can pointto meaningful individual differences in the co-variation between a predictor and outcome (seeSinger, 1998). Beyond testing for main effects ateach level of a multilevel model, as well as in-teractions within and across levels, researcherscan also use multilevel models to test for medi-ational effects, using the strategies that are usedin ordinary least squares regression (e.g., Baron& Kenny, 1986; MacKinnon, Lockwood, Hoff-man, West, & Sheets, 2002), with modificationsmade to accommodate multilevel testing. Ofinterest, for example, would be tests of whetherneed satisfaction mediates the relation betweeninternal or external motivational supports and avariety of motivational outcomes examined overtime, including academic or work performance,creative output, social behavior, and well-being.

Conclusion

Because motivation is a dynamic phenomenonthat is sensitive to both person and situationalchanges, it is particularly well suited to multi-level modeling research. In this chapter, we haveattempted to show that there is a natural pairingbetween expanded longitudinal designs, includ-ing daily experience and lab-based methods, andmultilevel modeling approaches to the studyof motivational processes. While this body ofwork is still small, important findings have beenuncovered: The fulfillment of basic psychologi-cal needs fluctuates on a day-to-day basis, andsubjective well-being is robustly affected bythese fluctuations. Other research, discussedonly briefly here, indicates that well-being andoptimal functioning show intraindividual fluc-tuations from relationship to relationship andlife role to life role. SDT research has shown thatthis fluctuation in functioning corresponds tochanges in supports for psychological needsacross these relationships and roles (e.g., LaGuardia et al., 2000; Ryan, La Guardia, Solky-Butzel, & Kim, 2005).Although considerable research has been de-

voted to the main effects of traits and types ofdaily events and situations on wellness, multilevel

research reviewed herein suggests that day-to-day well-being is contingent on finding personalvalue or meaning in everyday activities (Reiset al., 2000; see also Sheldon & Kasser, 1995,1998). Research also suggests that collective orgroup well-being depends on personal values tiedto basic psychological needs. Finally, more recentresearch indicates that the quality of conscious-ness known as mindfulness can serve as a dispo-sitional and situational support for basic needfulfillment.There is still considerable potential to be

tapped in the exploration of substantive andanalytic questions on motivational processes,and we have outlined several possibilities forfuture research in multilevel modeling. Whilebeyond the scope of this chapter, it is also worthnoting that there are rich research opportunitiesin joining major analytic forces, such as multi-level modeling with latent variable modeling(e.g., Chou, Bentler, & Pentz, 2000). A definingfeature of the recent progress in motivation re-search, and in the field of psychology as a whole,has been a rapid increase in the sophistication ofits research methods and analytic tools. Re-searchers willing to take a close look at motiva-tional processes through the lens of multileveland other cutting-edge approaches will do muchto expand our basic and applied understanding ofthis central feature of human behavior.

Notes

1. Reis et al. (2000) also tested gender main ef-fects and interactions with both Level 2 and Level1 need satisfaction. None of the results qualifiedthe results reported here.

2. First-order autocorrelation is one of the sim-plest autoregressive error structures and is com-monly termed AR(1). When observations on somevariable are equally spaced in time, this structureassumes that the residuals for adjacent observationswill be most highly correlated, the residuals forobservations two intervals apart will be less highlycorrelated, and so on. There are also continuous-time versions of the usual AR(1) error structurethat are capable of assessing first-order autocorre-lation when observations are unequally spaced intime (see Schwartz & Stone, 1998).

References

Baron, R.M., &Kenny, D. A. (1986). Themoderator-mediator variable distinction in social psychological

CHAPTER 37 . MULT ILEVEL MODEL ING OF MOTIVATION 539

ONG_Chap-37.3d 06/16/06 9:36pm Page 540

research: Conceptual, strategic, and statisticalconsiderations. Journal of Personality and SocialPsychology, 51, 1173–1182.

Baumeister, R. F., & Leary, M. R. (1995). The needto belong: Desire for interpersonal attachmentsas a fundamental human motivation. Psycho-logical Bulletin, 117, 497–529.

Bem, D. J., & Funder, D. C. (1978). Predictingmore of the people more of the time: Assessingthe personality of situation. Psychological Re-view, 85, 485–501.

Brown, K. W., & Moskowitz, D. S. (1998). It’s afunction of time: A review of the process ap-proach to behavioral medicine research. Annalsof Behavioral Medicine, 20, 109–117.

Brown, K. W., & Ryan, R. M. (2003). The benefitsof being present: Mindfulness and its role inpsychological well-being. Journal of Personalityand Social Psychology, 84, 822–848.

Brown, K. W., & Ryan, R.M. (2004a). Fosteringhealthy self-regulation from within and without:A self-determination theory perspective. In P. A.Linley & S. Joseph (Eds.), Positive psychology inpractice (pp. 105–124). New York: Wiley.

Brown, K. W., & Ryan, R. M. (2004b). Mindful-ness as a facilitator of competence and related-ness. Unpublished data, University of Rochester.

Bryk, A. S., & Raudenbush, S. W. (1992). Hier-archical linear models: Applications and dataanalysis methods. Newbury Park, CA: Sage.

Carver, C. S., & Scheier, M. F. (1998). On the self-regulation of behavior. New York: CambridgeUniversity Press.

Chou, C.-P., Bentler, P. M., & Pentz, M. A. (2000).A two-stage approach to multilevel structuralequation models: Application to longitudinal data.In T. D. Little, K. U. Schnabel, & J. Baumert (Eds.),Modeling longitudinal and multilevel data: Prac-tical issues, applied approaches and specific ex-amples (pp. 33–49). Mahwah, NJ: Erlbaum.

Cote, S., & Moskowitz, D. S. (1998). On the dy-namic covariation between interpersonal behav-ior and affect: Prediction from neuroticism,extraversion, and agreeableness. Journal of Per-sonality and Social Psychology, 75, 1032–1046.

Davis, C. (1997). Body image, exercise, and eatingbehaviors. In K. R. Fox (Ed.), The physical self:From motivation to well-being (pp. 143–174).Champaign, IL: Human Kinetics.

deCharms, R. (1968). Personal causation: The in-ternal affective determinants of behavior. NewYork: Academic Press.

Deci, E. L. (1975). Intrinsic motivation. New York:Plenum.

Deci, E. L., Connell, J. P., & Ryan, R. M. (1989).Self-determination in a work organization.Journal of Applied Psychology, 74, 580–590.

Deci, E. L., & Ryan, R. M. (1985). Intrinsic moti-vation and self-determination in human be-havior. New York: Plenum.

Deci, E. L., & Ryan, R. M. (2000). The ‘‘what’’ and‘‘why’’ of goal pursuits: Human needs and theself-determination of behavior. PsychologicalInquiry, 11, 227–268.

Diener, E. (1996). Traits can be powerful, but arenot enough: Lessons from subjective well-being.Journal of Research in Personality, 30, 389–399.

Gagne, M., Ryan, R. M., & Bargmann, K. (2003).Autonomy support and need satisfaction in themotivation and well-being of gymnasts. Journalof Applied Sport Psychology, 15, 372–390.

Kasser, T. (2002). The high price of materialism.Cambridge, MA: MIT Press.

Kasser, T., & Ryan, R. M. (1993). A dark side of theAmerican dream: Correlates of financial successas a central life aspiration. Journal of Personalityand Social Psychology, 65, 410–422.

Kasser, T., & Ryan, R. M. (1996). Further exam-ining the American dream: Differential corre-lates of intrinsic and extrinsic goals. Personalityand Social Psychology Bulletin, 22, 80–87.

Kreft, I., & deLeeuw, J. (1998). Introducing mul-tilevel modeling. Thousand Oaks, CA: Sage.

La Guardia, J., Ryan, R. M., Couchman, C. E., &Deci, E. L. (2000). Within-person variation insecurity of attachment: A self-determinationtheory perspective on attachment, need fulfill-ment, and well-being. Journal of Personalityand Social Psychology, 79, 367–384.

Larsen, R. J. (1989). A process approach to person-ality psychology. Utilizing time as a facet of data.In D. Buss & N. Cantor (Eds.), Personality psy-chology: Recent trends and emerging directions(pp. 177–193). New York: Springer-Verlag.

MacKinnon, D. P., Lockwood, C. M., Hoffman,J. M., West, S. G., & Sheets, V. (2002). Acomparison of methods to test mediation andother intervening variable effects. PsychologicalMethods, 7, 83–104.

Reis, H. T., & Franks, P. (1994). The role of inti-macy and social support in health outcomes:Two processes or one? Personal Relationships,2, 185–197.

Reis, H. T., Sheldon, K. M., Gable, S. L., Roscoe, J.,& Ryan, R. M. (2000). Daily well-being: Therole of autonomy, competence, and relatedness.Personality and Social Psychology Bulletin, 26,419–435.

Ryan, R. M. (1993). Agency and organization: In-trinsic motivation, autonomy and the self inpsychological development. In J. Jacobs (Ed.),Nebraska symposium on motivation: Develop-mental perspectives on motivation (Vol. 40,pp. 1–56). Lincoln: University of Nebraska Press.

540 PART V . INTER INDIV IDUAL DIFFERENCES IN INTRAINDIV IDUAL CHANGE

ONG_Chap-37.3d 06/16/06 9:36pm Page 541

Ryan, R. M. (1995). Psychological needs and thefacilitation of integrative processes. Journal ofPersonality, 63, 397–427.

Ryan, R. M., & Deci, E. L. (2004). Autonomy isno illusion: Self-determination theory and theempirical study of authenticity, awareness, andwill. In J. Greenberg, S. Koole, & T. Pyszczynski(Eds.), Handbook of experimental existentialpsychology (pp. 449–479). New York: Guilford.

Ryan, R. M., & Deci, E. L. (2000a). Intrinsic andextrinsic motivations: Classic definitions andnew directions. Contemporary EducationalPsychology, 25, 54–67.

Ryan, R. M., & Deci, E. L. (2000b). Self-determi-nation theory and the facilitation of intrinsicmotivation, social development, and well-being.American Psychologist, 55, 68–78.

Ryan, R. M., La Guardia, J. G., Solky-Butzel, J., &Kim, Y. (2005). On the interpersonal regulationof emotions: Emotional reliance across gender,relationships and cultures. Personal Relation-ships, 12, 145–163.

Ryan, R. M., Sheldon, K. M., Kasser, T., & Deci,E. L. (1996). All goals are not created equal: Anorganismic perspective on the nature of goalsand their regulation. In P. M. Gollwitzer & J. A.Bargh (Eds.), The psychology of action: Linkingcognition and motivation to behavior (pp. 7–26). New York: Guilford.

SAS Institute. (1992). SAS technical report P-229,SAS/STAT software: Changes and enhance-ments. Cary, NC: Author.

SAS Institute. (1997). SAS/STAT software:Changes and enhancements through release6.12. Cary, NC: Author.

Schwartz, J. E., & Stone, A. A. (1998). Strategiesfor analyzing ecological momentary assessmentdata. Health Psychology, 17, 6–16.

Sheldon, K. M., & Kasser, T. (1995). Coherenceand congruence: Two aspects of personality in-tegration. Journal of Personality and SocialPsychology, 68, 531–543.

Sheldon, K. M., & Kasser, T. (1998). Pursuingpersonal goals: Skills enable progress, but not allprogress is beneficial. Personality and SocialPsychology Bulletin, 24, 1319–1331.

Sheldon, K. M., & McGregor, H. A. (2000). Ex-trinsic value orientation and ‘‘The tragedy of thecommons.’’ Journal of Personality, 68, 383–411.

Sheldon, K. M., Ryan, R. M., & Reis, H. T. (1996).What makes for a good day? Competence andautonomy in the day and in the person. Per-sonality and Social Psychology Bulletin, 22,1270–1279.

Sheldon, K. M., Sheldon, M. S., & Osbaldiston, R.(2000). Prosocial values and group assortationwithin an N-person prisoner’s dilemma game.Human Nature, 11, 387–404.

Singer, J. D. (1998). Using SAS PROC MIXED tofit multilevel models, hierarchical models, andindividual growth models. Journal of Educa-tional and Behavioral Statistics, 23, 323–356.

Smyth, J. M., Soefer, M. H., Hurewitz, A., &Stone, A. A. (1999). The effect of tape-recordedrelaxation training on well-being, symptoms,and peak respiratory flow rate in adult asth-matics: A pilot study. Psychology and Health,14, 487–501.

Turk, D. C. (1994). Potentials of process mea-surement: Motion pictures versus snapshots.Annals of Behavioral Medicine, 16, 198.

Varela, F. J., Thompson, E., & Rosch, E. (1991).The embodied mind: Cognitive science and hu-man experience. Cambridge, MA: MIT Press.

West, S. G., & Hepworth, J. T. (1991). Statisticalissues in the study of temporal data: Daily ex-periences. Journal of Personality, 59, 611–661.

White, R. W. (1959). Motivation reconsidered: Theconcept of competence. Psychological Review,66, 297–333.

Williams, G. C., Gagne, M., Ryan, R. M., & Deci,E. L. (2002). Facilitating autonomous motivationfor smoking cessation. Health Psychology, 21,40–50.

CHAPTER 37 . MULT ILEVEL MODEL ING OF MOTIVATION 541