This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
699
ler{r ^rou{ aM 'Poour Pue rftrlenb uoll)€ra}ur
uaamlaq drqsuorlelar aqt Jo saleurrlse ur saJuaJaJ-JIP IEnP$Pur Jo uollenle^a rllsrlear e s,!\oll€osl€ tJeJelur ,{aqr ruoq,u r{rl^{ Pue
'aJar{l\ 'uaq1v\
asoor{) ot sFnpl^pul Surt}rurrad'fpuerrodrul'stxatuoJ r(epr{rana ut a)ue
-ngur lerordnal rraql aultuexa ot st sa8ueqrxa
Ier)os pue poou ueemlaq drqsuorlela.r lesne) Jql
Jo lsat alqeraJard y 'ay{ i(eprfua,ra o} sa)uangurprordnar SurzqereuaB ur palsaJalur aJE aA\ uor{ir{
Ier,tul lou are 'alduexa ro; '3urilas ro raulrBduorlJEJatur Jo aJror{) lu€dDru€d uo s}urEJlsuoJ
leluarurradxg'-{rrprle,r lerrSolora ur,rno1,{11errd,traJe pooru ;o sa8ueqrxa IeIJos Jo suorlelndtueur'r(lrlesner JaJur ol paJrnbal aruaparard prod-ruar ar{r r{srlqerse suortelndrueur letuaurradxaqSnoqrly 'pootu pue suorlJ€rotul I€rros uaem]-aq drqsuorrelar I€snEl IEuorlrarIPICl e PaleJrPurseq aluaprle leluaurradxa'JaloaJol^'.{rrlenbro ad.{l uorlJeJa}ur pue pooru uaam}eq palrasqo
ueuo aJE suorlelaJJo)'sarPnls leuoll)es-ssoJJu1 '8uraq-11a.n puE poour uo sdrqsuorlelar I€Dos1o nedurr aql ol satnqrJluo) aloJaraqt suortJe-Jatur Jo drr1Bnb aqr pue 'suoItJEJatuI
IeIJos uraru€]sqns pue adeqs alel sdrqsuorlelar IeDoS
5 6 0 P A R T V . I N T E R I N D I V I D U A L D I F F E R E N C E S I N I N T R A I N D I V I D U A L C H A N G E
-tIIII
negative social exchan!That is, frequency of Idicted negative affectlwhen positive exchang(frequency of positive {tive affect and not negalexchanges were held Ihand, longitudinal an{dent crossover effects: ]were held constant, frlchanges at baseline prtnegative affect at folllexchanges, independen]did not predict either Iat follow-up.
The foregoing resusistencies in the valenc{of social interacrions mldiffering temporal dyn]negative exchanges. Vjexamined these tempdseries analyses of conrassociations between p(and time spent in vari(actions. Data consistedthree times daily for 4gative affect were concqualitatively distinct tybut there was no evideciations within days, paffective states and infrthe ability to detect timtthe other hand, the resryses reported by Newsthat negative dspects rimpact on subsequent rthese effects may be mreffects of positive aspecsitive affect.
Our interest is in exiof loneliness moderatrlinking interaction qualsearch indicates that inabilities to benefit fromregulation (reviewed b5Pereg, 2003). Extraverample, are not only mo:interaction partner but iare shy or introvertedtheir interactions as poPietromonaco, 1997; Vcially anxious individualonly perceive interactivironment as poorer
some individuals report dissatisfaction with theirsocial relationships and a feeling of isolation orlack of relational or collective connectednesswith friends or groups (Hawkley, Browne, &Cacioppo, 2005). We have begun to learn thatpotentially adverse health consequences mayensue from these feelings of loneliness and iso-lation (e.g., Cacioppo et al., 2002; Hawkley,Burleson, Berntson, & Cacioppo, 2003). Inmarked contrast, very little research has ex-amined how those who are low in feelings ofloneliness are protected from these con-sequences. Studying these socially connectedindividuals-their cognitions, motives, and in-teractions-could help us understand what itmeans to be a successful social being.
In prior research (Hawkley et al., 2003), wefound that socially connected individuals re-ported less negativity and more positivity infeelings about their interaction partners than didtheir lonely counterparts. Socially connected in-dividuals also reported lower negative affect andhigher positive affect over the course of theweek than did their less connected counterparts.Several explanations could contribute to thesedifferences in affect and interaction quality. First,the social interactions of socially connected in-dividuals could be objectively different than thoseof lonely individuals. For example, socially con-nected individuals may attract or choose thekind of interaction partners that facilitate positivesocial exchanges, and this could subsequentlycontribute to higher positive affect and lowernegative affect among socially connected thanlonely individuals. Second comparable socialinteractions may be perceived more positively bysocially connected than by lonely individuals.The same perceptual bias may contribute to re-ported mood differences between socially con-nected and lonely individuals. Third, sociallyconnected individuals may experience a larger orlonger-lasting boost in mood following positivesocial interactions, or a smaller or shorter-lastingdecrement in mood following negative social in-teractions. This could help to explain why, acrossa typical week, average positive affect was higherand average negative affect was lower amongsocially connected than lonely individuals. Foreach of these explanations, the underlying as-sumption is that affective aspects of everyday lifemay contribute to downstream health ef{ects, andmore interaction positivity or less negativitycould put socially connected individuals at a dis-tinct advantage.
Theoretical Background
Numerous studies have examined the relationshipbetween social interactions and mood, and resultssuggest a reciprocal causal relationship. For ex-ample, in a series of experimental studies, Cun-ningham (1988a, 1988b) induced mood states andfound that a positive mood, relative to a negativemood, stimulated greater interest in social inter-actions and increased conversation quantity andquality (i.e., self-disclosure). The reverse causaldirection has also garnered support, however. In aseries of experimental srudies, Mclntyre and col-leagues found that either spontaneous or arrangedsocial interactions increased positive affect relativeto affect during a neutral control setting (Mcln-tyre, Watson, Clark, & Cross, 1991; Mclntyre,Watson, & Cunningham ,1990). Similarly, when adiary methodology was employed" state positiveaffect was higher when individuals were socializ-ing (Watson, Clar[ Mclntyre, & Hamaker, 1992),engaging in physically active social events (Clark& Watson, 1988), interacting with familiar asopposed to relatively unfamiliar partners (Vittengl& Holt, 1998a), and experiencing fun or necessarysocial interactions (Vittengl & Holt, 1998b).Conversely, negative affect was elevated duringarguments and confrontations (Clark & Watson,1988; Vittengl & Holt, 1998b), interactions markedby the receipt of social support (Vittengl & Holt,1998b), and interactions with poor communicationquality (Vittengl & Holt, 1998a).
Positive and negative affect are largely in-dependent dimensions of mood (Watson, Clark,& Tellegen, 1988) and may be differentially af-fected by positive and negative qualities of socialinteractions. In cross-sectional analyses, Rook(2001) found that number of daily positive socialexchanges was related to greater daily positivemood but was unrelated to negative mood,whereas number of negative social exchangeswas related to dampened positive mood and in-creased negative mood. Similarly, Finch, Okun,Barrera, Zautra, and Reich (1989) showed thatnumber of positive social ties was associatedwith well-being, whereas number of negativeties was associated with well-being and psycho-logical distress. More recently, Newsom, Nish-ishiba, Morgan, and Rook (2003) reportedsignificant correlations between positive andnegative social exchanges and both positive andnegative affect in both cross-sectional and lon-gitudinal analyses. However, independent ofeach other, concurrent effects of positive and
'fpnrs srqr ur esn eA\ qreordde eq] sr pue asaqlse q)ns elEP Psrntrnrts ,{1;e:rqr.re.ralq ez^Jeueor pau8rsap anbruqral e sr 'Surlapou sluaDl,-Jeol uopuer pue 'Suqapour slraga-paxrur €ur-lepou rerurl lerrq)rerarq se uMoul osp'(5661'ralsog
4 srapfus !7697'4f,t9 4 qsnquapnEui866I '.^Anaal ap T Uar) :Z0OZ'xoH 1566I 'urars
-ploD 1l,^{'Ifrl) Surrapotu Io^olrrlnl'slenpt,tlpuluil.lllM PJlseu ere urnt ur q)rq1\{ 'sr(ep urqrr,n
;o aar8ap ur Suua;11p slenprrupul ruor1 {aalvr eroy dep .{rana saulua ,{rerp auru 'tlarerurxorddeparrnbre a,l\ r{)rq1r4. uI '(0002 ''1u re oddone3)srlnpe Sunor{ ur ssaupa}Jauuo) pue uol}Elosr
IerJos Jo ,{pnrs Surtdues-aruarradxa ue ur u8rs-ap srqr palerluulsul a14 'suorlsanb xaldruor;ouorteunuexa aql Suulrurrad erep Jo a)Jnos q)rJe saprnord u8rsap sarnsearu-paleodar parsau V
suollelcod4l lofieuung :qceorddv e4lepuenp eql
'ssauPa})auuo) I€I]os
ur .trol snsral q8rq spnp4lpur Suouu sna;;a8ur1se1-13116qs arreq i(eur dllule8au uorD"Jalur'snq1 'suorDeralur aarte8au Jo urroj aql ur tFes-sE ol tuelsrser alrnb aq.[eur teq] ftrlr}lsod Joapdr e Surtenradrad r(qaraqt 'suorlreralur
lunos.{lqenb-ra11aq ralsol deur (SOOZ ''F n ,{agmeg)slenpr,rpul pal)auuor dllenos dq parrodarnay;e azrrlrsod raluar8 ar{r rEq} slsaSSns drrlenbuon)eJJtur pue lrJJJe uJJMleq drqsuortelaraq] Jo arnlEu lerordnar aqr 'uonrppe uI 'ssau
-pelrruuo) Ier)os ur Mol snsran q8rq slenp4lp-ur ur pooru uo tredurr ue Jo ssal aaeq ,(eusuou)eralur;o slradse anrte8au os '(9967 ''p rodelYvreg) slenpl^lpul parreuuor ssal op uEt{},(lanue8au ssal tuar{l anrarrad sFnpr^rpur palJau-uor ,{.l1uDos 'sDadse arrr}e8au eAEr{ suol})Erel-ur u3r{.t\ u3^1 'sl€nPhIPuI
PsDauuor ssal ro}anJl aq plnoM ueqr suorl)elalul leoos 3uwro11o;poour arrrlrsod Jo luaure)uequa Jo Poour aarteSau
to uournpal ratear8 e aruarradxa ol uaql asod-srp ,{eru sHI '(€002 ''1e ra [a1ry*eg) spnpl^p-ur petJauuo) ssal op ueql .{la,ulrsod aroursuolt)Elelur rraqt a,rra:;ad sIenprnrpur pa]rauuo)dlleoos rer{r pa^rasqo a^eq a,v\'q)r€esar rno ul
'(p661 'sur33rg rB 'smorrll 'uauual 5lraUJV)pooru ,(ep-Xau uo slua^a Jo lJedrur .ralea;3 e z*oqs
uoddns IBDos ssal qrl{,\ sl€np4lpur '^lrEIIruIS'(*gOOf 'lloH rB 13uau1n) dlarxue leDos ur rv\olslenp4lpul op u€qr poour rraq] uo pudurr ralear8
Jo osle lnq frrlenb ur rarood se luaruuoJll-ua IeJnleu aql ul suotl)eralur anrar"rad,(1uo
T DeJJEg) stuela alrusod se suorlreralur Jrar{larrrarrad o] slenprrlrpul patJaloJlur ro dqs areueqr d1a11 aroru oslu ar€ lnq rauued uollf,Eretul8ur1pr* e pur; ot ,{1a4rt aroru r{po tou are 'aldure
-xa roJ 'slenpr^rpul palralerrx:l '(9967 '3ara4
a? 'ra^eqs /ra)uIInlII I ,(q parur.aurar) uotleln8ar
uorlorua Jo saJJnos ler)os ruoJJ rrJauaq ol sallrllqerraqt ur raJJrP slenPl^IPur reql sale)IPUI r{)rBas-or rorJd 'lJ3ilE
PuE ,lrrlenb uorDEJalur SuqurtslJaJJe asaql ;o ,(ue saleJaPoru ssaurlauol Joaar8ap raqtaqrr,r Surururexa uI sr lsalolul rno
'trajlE e^rlrs-od uo suorpeJalur Jo slradse azrrlrsod;o srrallaaqr ueqr Surrsel-8uo1 eroru aq feur slra;ya asaqt
reqr pue 'na;1e anrle8au luanbasqns uo pedurr
ue aleq suoE)eJalur ;o slcadse e,rrle8ag 1Bq1tsaSSns (EOOZ) 'p ta uroslv\aN ,{q parrodar sasd-1eue luurpnrr8uol Jo stlnser eql /pueq raqlo aqluO 'suorleDosse pa83e1-ourrr lrelap ol iftrlrgu agt
parapurq sauerp luanbarJul pue salels alrlraJJeSurlaag asneraq ,(1q1ssod 's[ep urqrrm suorleo-osse pa33e1-arurl Jo a)uaprr'a ou sBA\ arar{} rnq'suoEf,eJalur
IErJos Jo sadrfi lrurlsrp ,{larrrlerrlenbr{lrrv\ PaleDosse dlluerrnruor ararr,r pa#e a.lue3-ou Pue J^rllsod 'SIJ3M
f f i z P A R T V . I N T E R I N D I V I D U A L D I F F E R E N C E S I N I N T R A I N D I V I O U A L C H A N G E
One of the advantages of MLM is that missingdata typically do not pose a serious problem;individuals can be measured for different num-bers of days, and different days can involve asfew or as many measurement occasions as ne-cessary.
In this chapter, we illustrate how MLM hasmade it possible to examine the following kinds oftheoretical questions in incomplete time-seriesdata such as that obtained using an experience-sampling methodology (ESM). For didactic clar-iry, subsequent MLM analyses will address thesequestions in the sequence enumerated here.
First, MLM can be used to evaluate the pro-portion of total variation in a given outcomevariable (e.g., affect) that is attributable to var-iation at a given level of the data hierarchy (e.g.,diary, day, or person). Predictors (e.g., lone-liness) can then be introduced to explain varia-tion at some of these levels. We expectconsiderable variability in affect and interactionquality at each level (diary, day, person). Basedon our prior work (Hawkley et al., 2003), wehypothesize that loneliness will predict lowerlevels of positivity and higher levels of nega-tivity in affect and interaction quality.
Second, MLM permits examination of theconcurrent relationship between affect and in-teraction quality in a way that avoids the biasthat would result if ordinary linear regressionwere employed without regard to the hier-archical structure of the data. Based on oriorresearch (Cunningham, 7988a, 1988b; Mclntyreet al., \990,1991), we expect to find a reciprocaland valence-specific relationship between socialinteraction quality and affect (i.e., positive in-teraction quality and positive affect will bemutually predictive, as will negative interactionquality and negative affect). Concurrent inter-action quality and affect may also exhibitcrossover effects (e.g., Newsom et al., 2003), butprior research has been mixed in this regard.
Third" MLM makes it possible to srudy intra-and interindividual variability in the concurrentrelationship between lowerJevel variables. Spe-cifically, multilevel modeling permits regressionparameters (e.g., slopes) at lower levels to bemodeled as dependent variables in regressionequations at higher levels. Such effects aretermed cross-leztel interactions because they in-volve variables at higher levels predicting slopesat lower levels. In our case, we study the effects ofindividual differences in loneliness on the asso-ciation between interaction quality and affect. Weexpect that individuals low in loneliness (i.e.,
socially connected individuals) may exhibit morerobust positive affect such that negative interac-tion quality may not have as great an impact onconcurrent positive affect as it does among thosehigh in loneliness.
Fourth, MLM can be employed to examine thetemporal relationship between variables. Typi-cally, time-series analyses are used to examinetime-lagged effects. We introduce a novel al-ternative means of assessing temporal relation-ships by testing lagged effects within the contextof MLM. Specifically, we use multilevel model-ing to examine the lagged effects linking inter-action quality and affect. Moreover, by varyingthe lag between predictor and criterion, we canuse the MLM approach to test the duration of theeffects of social interaction quality on affect-that is, we can examine the duration as well as thecausal direction of effects.
Prior experimental research has shown re-ciprocal causal effects relating interaction qual-ity and affect (Cunningham, 1988a, 1988b;Mclntyre et al., 1990, 1991), and we expect thatlagged effects will support this causal strucrure.Namely, we hypothesize that mood will influ-ence subsequent interaction quality and viceversa. In general, negative exchanges have agreater impact on psychological outcomes thando positive exchanges (a negativity effect; Rook,1990). The potency of negative exchanges maytherefore lengthen the duration of their influ-ence on a{fect. If this is the case, we expect thatnegative interaction quality will continue to in-fluence affect when the temporal lag is extended.
Fifth, adding yet another level of complexity,MLM makes it possible to study intra- and in-terindividual variability in the lagged relation-ships between lower-level variables. A temporallag between lower-level variables allows us toask whether the causal structure linking inter-personal interactions and affect is the same forindividuals regardless of degree of loneliness. Inaddition, we can examine whether lonelinessmoderates the duration of the effects of inter-action quality and affect on each other. If socialconnectedness facilitates effective emotionalregulation, then a given degree of positivity in asocial interaction may elicit greater or longer-lasting increases in subsequent positive affectwith or without greater or shorter-lasting de-creases in subsequent negative affect amongindividuals low versus high in loneliness. Inaddition, reciprocal reinforcement between po-sitive interactions and positive affect may leadto more effective buffering of the effects of
negative interactrons atopposed to high in lcconnected individuals I
tive interactions morelonely counterparts.
Sample and Methods
Over 2,000 students weto represent the lower (
(total score >33 and <3>46) quintile of scoreliness Scale (Russell, PThe R-UCLA Lonelinitems that were originexperiences that best inonlonely individuals.are "I feel isolated frcpanionship"; and "Thto." Notably, none oterms lonely or loneliton the R-UCLA Lonrpositively correlated r,rliness and inversely cmeasures of social exptime spent alone eaclfriends; Russell et al.,scores on the R-UCIpresent a high degreebeddedness and objecti
Participants were 13(83% Caucasian;7o/"Asian, Asian Americalother or undeclared), ethe loneliness groups,were equally represerExclusionary criteria lwhere (Hawkley etof recruitment, stude(SD: 1.0) and they hand, on average, 3.2 (
ters; 52"/" were freslmores, 87o were junio
were seniors or fifth-yWe employed ESM
halyi, 1983) to collecchosocial and behavcompleted diaries fordays. A programmablcipants was programmtimes between 9:30 r
each day, subject to tterbeep interval was Iutes. The 134 partici
eql Jo arnDnrts lBrrr.IJrErarq aql PUE sasaqloo^tlJql Jo eJnleu aql ua^t3'(suortdunsse asaql
Iraqr or spol.ltat'u Jo uorssn)srp e roJ '6661'ta1
-sog T srapfiug aas) .(1rreaur1 pue 'iftnrlsepars
-oruor{ FnPrsar rrun-urq{M /s])3JJa- ruoPuEr
Pu€ sl€nPrssr Jo 3)uaPueoaPur PUE Alrlerurou:sarenbs lseal .{reurpro ro; .{ressarau asoq} s€arxes aql d1a8re1 ore tr l'1tr\l Jo asn aql:o; z{ressar-au suorldurnsse a{I '(966I 'ura}sp1o3) s1ar.a1raqSrq or etep I Ia^aT Surre8arSSe fpresserau-un uodn tuBPualte ra.trod rra.o1 Pue uorlelardra]-ur reelrun (q) p"" luorssar8ar reauq -{reurpro
qlrl!{ se 'aJn}JnJts palsou ol pre8ar rnoqrr,tr pazdl-eue oJe ereP uoqM Palalunof,ue sErcl lalaurereoara,ras (e) ;o srualqord spro^e I I'IIAI dq elep
IE)[{rrerarq ;o srsr(1eue 'uor]rPPe u1 'tualqord
snorras u asod tou op ,{11elddr elep Surssnu pue'salqeue^ ruapuadap sE palaporu aq UBJ sluar)-rrJaoJ Io^al-raMol lPql pBJ aql ePnPur I 'll
qtlry\ palenosse sa8eluenpe aqt Jo auos 'JarlJea
ot papnlp arvr sV 'erep drerp sarnseau paleadarrno Jo arnDnJls IBJrq)rEJarq 3ql uo azIIElroB) o1Suqapour Ia^alrrlnru pa^{oldura am '^{pnrs srqr u1
A6elerlg c!1t{leuv eleg
'(eeer''le la uosleM ISVNVd) alnparlJs DaJJV a^p-e8ag pue alrlrsod eqt Jo surall anrte8au 0Ipue anrtrsod 0I aql 01 sasuodsar Sururuns ,{q
Jo uorl)un1 e se lrJ;lp rou plp sulntar ,{rerp yoJrel Jqr reqr palea^Jr aJueueno) 1o srs{1eue uy'aler urnlar "/o1g ve Surluasardar 'sarrcrp
71V'galqrssod e p ZU'9 papnord dpnrs drerp dep-1
aqr paraldtuor oqm sluedorlred 7E1 aql 'sarn
-ullu 0Zl pue gt ueemteq sem le^lalur daaq.rar-ur aqt leqt rurelsuor aqr ot ualqns ',{ep qtearqSruprtu 00:ZI pu€ 'ru'E
0g:6 uaamlaq satutrruopuer auru te daaq ol paturuerSord sem sluedn-rlred ol paprzrord q)]EM alqeuruerSord y 'sr{up
alun)rsuor / Jo letol e roJ sauerp palaJduorsluednrlre4 'satets
IErorler{aq pue Ienosoq)--{sd rnoqe uoueurroJur rrallo) ol (ggef 'dF,{
-nutuazs{rs3 4 uosrel) 1193 par{oldrua a14'sluaPnls rea,{-ql;5 lo slolues JIJM
7og Surureurar aq] pue 'srorunf aJa^\ yo8 'sarouJ
-oqdos araru. 7oZ€ /uauqsarj aJarvr o7 79 lsra]
-renb rrurapere (A'Z:AS) 7'g 'a?vnrte uo 'puu
auo lseal re patalduror peq,(aqr pue (g'1:95)
Z'6I sew a8e ueaur ,stuapnts 'luatutrnJ)al
Joaurl orll lV '(8002 ''lE ra faplmeg) araqtvr-as1a palrodar uaaq aleq ErJalrJ) dreuorsnpxS'dnor8 qrea uqlrm poluasardar ^{11unba arar*salerual pue saluur pue 'sdnor8 ssaurlauol aqlSuoure parnqlrrsrp '(1pnba '(parepapun ro ror{lo
%€ lrapuelsl rlJped ro 'uelrraury uErsv 'uBIsV
o l :u>trauuy ueru;V o/ol :ue:.se)nel %€g)stuapnls alenper8rapun SgI ara^r stuedorued
'luaua8e8ua lunos enrnalqo pue sseupeppaq
-ura IEDos a,rrDafqns ;o aarSap qSrq e ruasard-er ale)s ssau{auo'I vl)n-u aql uo salo)sr"ro1 'dlasra,ruol '(086I ''1e ra llassng ispuauJasolr Jo raqrunu 'r(ep qrea auole luads arutl
Jo ]unorue ''3'a) saruar.radxa IBDos Jo sarnseauanrpafqo rllr^{ pa}EIaJJor dlasranur pue ssaurT-auol jo suodar-;1as r{trrv\ patelarror ,(la,rrrrsod.{1q8rq are al€)S ssaurlauo"l Vlln-U aq} uosJrors 'sseleqt;JlaN 'ssJuli ruol so frpuo1 sutalaqt ot ra;ar srualr aqr Jo auou ',{lqetop
5 6 4 P A R T V . I N T E R I N D I V I D U A L D I F F E R E N C E S I N I N T R A I N D I V I D U A L C H A N G E
-lIIII
the equation for Pql1rq
Pmtr!
The full, conditionaf r{now be: q
Yai.: Yooof Yoro(@d
Models such as th'ationly the tip of the iceber{of MLM to test comp!magnitude and directilarchically stmcrured datllevels. Multilevel mo{include nonlinear effect]and multivariate outc!entered into the model ]error distributions and ibe specified. In ad&tio]times be centered (i.e.,
their means) in order tiestimation and enharjmore involved discussiilimitations of MLM, seor any introductory tex2002; Kreft & de LeerBryk,2002; Snijders &
Software
Facilities for specifyirlevel models are inchsoftware packages, in<LISREL (8.54) and SMLM packages suchbush, Cheong, & Cor(Rasbash et al., 2000'(fcireskog & Srirbom,ease of use. Comple:be easily specifiedLISREL using PRELIS
TABLE 39.1 Pr
1 A '
43.t7-!
32.!
Nofer Numbers*Values for posi
the grand mean fo
present data set, MLM was chosen as the mostappropriate analytical tool.
Three-Level Models
In MLM, the lowest level of the hierarchy (herethe diary level) is commonly denoted Level 1,the next highest as Level 2, and so on. Mostapplications of MLM involve only two levels, forexample, students nested within schools or re-peated measures nesred within individuals.There is theoretically no limit to the number oflevels a multilevel model can contain, but thereare often practical and software-related limita-tions, and few sources discuss MLM with morethan two levels. Our data could be organized intothe three-level hierarchy depicted in figure 39.1.
The three-level, fully unconditional multi-level model can be represented in terms of Level1, Level 2, and Level 3 submodels. The Level 1(diary level) submodel is:
Y;i1, : n01t<l elik e4t-N(0, o2) (1)
where Y4t represents the response at diary i onday I for individual k, nsip represents the meanresponse on day i for individual k, and. eqp re-presents the deviation of Y;;7. from no,k.r TheLevel 2 (day-level) model is:
n6;1 : pe61 *r6ip rsil-N(0, t,) (2)
where Beep represents the mean response forindividual k and 1671 represents the deviation ofns;p from 0oor, or the random effect for day l,The Level 3 (person-level) model is:
Foor: Yooo*root z6e1-N(0, rB) (3)
where 16ss represents the grand mean and a961represents the deviation of goor from yooo.Combining Equations 1,2, and 3 yields the fol-lowing composite model:
Yi ik: losolusu,Irsipl eirk (4)
which explicitly models Xir as the sum of thegrand mean and deviations at Level 3, Level 2,and Level 1, respectively.
Predictor variables may be entered at any le-vel in order to explain variability in randomeffects at lower levels. For example, if a day-levelvariable is entered to explain variation in theLevel 1 intercept, Equation 2 becomes:
no;r : Foor * F uk(wft) -t rolk (5)
In this equation, the predictor ru is hypothesizedto predict intraindividual (intraday or inter-diary) variability in the outcome. The slopeparameter describing the expected increase innsik given a unit change in ra is here specified as a
fixed et't'ect, meaning that it is not permitted tovary across individuals or days. Consequently,
l-r"ror"r.l Tr',tor"tl l-r"ror"t.l l-r"iror"r.lI Diary l 1"""'l oiarv g
| | Diary l 1"""'l oiary e Ift"i-t or"t I fi"- o""L II oiarv t l"-"" 1 oiarv s I
?ooo(''in
ipFigure 39.1 Three-level data structure. Data used in this study could be organized into a three-level hierarchy. Level 1 (diary level) consists of repeated measures of affect (positive andnegative) and interaction quality (positive and negative) nested within days (Level 2), which inturn are nested within individuals (Level 3).
'Zgt'l sr ]u€au ro] ueau pu€r8 aqr PuB €90 9 sr rulsod roJ ueau PuerS eql,slrun ]uauernseau IeulSuo ur os '0I dq parrdrllnu aJaM dlrlenb uoqf,pralur a lle8au puu alltlsod JoJ sanle1*
'sJoJJa PJePUeIS aJe Sasar{luaJPo ur sJaqunN iaioN
uI q)H1'^ '(7 1am1)
pue aurrrrsod)-aaJr{t E olur Pazru
'flluenbasuo3 'sfep roOl PailIurJad lou sr lr Ie se pagnads eraq sr rr,ul as€a'IJur paDJdxaadols ar{I 'aruof,lno au-Jalur Jo deperrur') yenpazrsar{}odfq sr rD rolJTpi
(S) ,trq+?!m)1rod
:saruo)aq z uolaql uI uonerren ureldx:
1ala1-r{ep e;r 'eldurexa rcuroPuEr ur ,tlgrqerrerr r-a1 z(ue le paralua aq de
'ri/Z
la^a'l 'E
1ar'a1 le suoraql Jo urns aql se 1ll r
(f) Ilra +,t!ot+4oo1
't't
-loJ eqr splar,t g pue 'z'.0001, urorJ 100d Jo uclmz pue ueaur puerS ag
pue sartrlclrssod aqr;o suolssmsrp pellolur aJourro1 .'dlqrqelardratur aruequa pue uoueturlsa
;o rtrllqers aq] arrordurr ot JaPJo ur (sueaur rraqtruol; suouernap se tseJeJ ''a'l) pa;aluar eq saurl-aruos .{eru sroprpard 'uoqrppe u1 'par;oads aq,{Bru sarnlrnrls a)uerrelo) pu€ suorlnqrJ}srp JoJJaxalduror pue '1or,ra1 i(ue 1e lapour ar{} o}ur para}uaaq UEJ seleue^o] 'sauroJlno aleue^rllnu puEisla^al aarqr ueqt rrour 'sDeJJr rEeulJuou rPnl)urol papuatxa aq ,{eur slapotu Ia a[rFI 'sla^al
o \t uEq] aloru qtr,l\ sras erep pernlrmls dlerrqrru-ran{ ur spaJJa Jo uoqrarrp pue apntru8eurar{r rnoqu sasaqrodr{q xalduor rsar or IAI'IIAI Jolurruarod aqr;o srural ur Eraqerr aqr yo drr aqr ,(po
tuasardar I uonenbf ur tEW se qrns slapol
Q) {r2 a't[o1 a loon + (1102)0r0i + 000L : 't!!I
:aq ltr'\ou
plno^\ Iaporu la^al-aarq] 'leuortrPuo) 'll.J
"rU(s) oroi:1iod
:se paluasardar aq i(eur rtrod JoJ uortenbe eql
9Nt'1100t,1'60 ulIdvHS
5 6 6 P A H T V . I N T E R I N D I V I D U A L D I F F E R E N C E S I N I N T R A I N D I V I D U A L C H A N G E
As has been not.d in plHolt, 1998a, fg98b; Wlond key question invqinteraction quality arelother words, does pol
predict the positivity ojrnteractron (or vlce r',1
possibility, we specifie{with posint and neginldictors of concurrent a{serving as Level 1. iquality. The Seneral ff
posat't',,u: nsftItl
no*:0oot+1
Foot : Yooo"]
Tlril : Frot
Frot : Troo
The combined equatio
posat't' ;i1:"t ooo * ^l t
+fojk-l (
Code for this model crdix. Results are reportlestimates, standard eciated with'i166.
In every case, thelrelationship betvveenquality. The directiorexpected: positive inthigher positive and Ictive interaction qualiland higher negativeconcurrent relationshcific is consistent witl
TABLE
Outcomc
posaffposaffnegaffnegaffposintposlntnegintnegint
interaction quality variables, but does not alterthe statistical significance of any reported effects.
In the null model, the point estimate of theintercept is an unconditional estimate of thesample mean, which in all four cases is sig-nificantly different from zero (p <.00001). Thefact that the Level 1, Level 2, and Level 3 var-iances also are all significantly different fromzero implies that there exists variation to bepotentially explained in all four outcome vari-ables by adding predictor variables to the model.Variation is additive across levels, such that thevariation which exists at a given level in a givenoutcome variable can be expressed as a propor-tion of the total variation by the intraclasscorrelation. For example, the proportions of var-iability in negaff that exist at the diary, day, andperson levels are, respectively:
Y d i a r v -
o2 f t ^ * rg74.844
t4.844+ 3.014+17.729: 0.502 (10)
'cT
Y d a y - R 2 f r r , ^
3.014 : 0.102 (11)14.844+3.014+\1..729
gauged by noting the magnitude of the regres-sion coefficient associated with loneliness, itssignificance, and any reduction in Level 1, Level2, or Level 3 variation when compared to thenull model. A typical equation for this type ofmodel is:
Posat't' ,,1,: Yooo * Yoot (l onelinessp)
I usu< * rsik I eiik (13)
Results of fitting this model to data for all fouroutcome measures are reported in table 39.2.The y61 coefficient in eacir case represents theslope of the dependent variable regressed onloneliness.
Individual differences in loneliness explainvariabi l i ty in al l four ourcome variables, butcomparison of the variance estimates in tables39.1 and 39.2 reveals that the explanation oc-curred mostly at Level 3 (the person levell. Thismeans that loneliness does not tend to accountfor fluctuations in affect or interaction qualityobserved across diaries or across days withinpersons. In line with expectations, however,loneliness predicted decreased positivity andincreased negativity in affect and interactionquality. Referring to the coefficients in table39.2, a one-unit increase in loneliness producedan increase in negative affect of 0.120 and adecrease in positive affect of 0.130. Values forposint and negint were multiplied by 10 beforeanalysis, so we can say that a 10-unit increase inloneliness produced a 0.151 increase in interac-tion negativity and a 0.140 decrease in inter-action positivity.
Conversely, in mood and social interact-ions, experiential domains that arguably are theprimary determinants of overall well-being, so-cial connectedness provides a distinct advan-tage. Specifically, social connectedness predictshigher positive and lower negative affect, andmore positive and less negative qualities insocial interactions. Not surprisingly, life is
o2
n - T FY P e r s o n - o 2 + r n + r p
1L.729 :0.396 (12)14.844+3.014 +11.729
Most of the explainable variation in all fouroutcome variables occurs at Level 1, or acrossdiaries (within days and persons).
After specifying null models, the next step ina multilevel analysis is usually to introducepredictor variables. Of primary interest to uswas the ability of loneliness to predict variabilityin affect and interaction quality. Loneliness wasmeasured only once per individual, making it aLevel 3 predictor. The success of loneliness inpredicting affect and interaction quality can be
TABLE 39.2 Parameter Estimates for Models With Loneliness as a Predictor
Dependent Variable
posaff negaff poslnt neglnt
ioor62in
i p
Note: Numbers in parentheses are standard errors.
roooo > d (sro o) szg'oIoooo'> d (zro o) zgo o-roooo'>d (zeodwz'o-roooo'> d (ero'o) ree'oroooo >d (sro o) sre oroooo'> d (ooo'o) zzo'o-toooo'> d (szo o) rur'o-roooo'> d (rro'o) rez'o
_ 00lJ
= oori_ 00It
: oori- 00rJ
: oori- 00I.1
: oori
geBauyesodyyeBauyyesodlur8autursodturBaulursod
lurBaulurBauluisodluisody1e3au
1;eBauylesod
;;esod
]Jatt:lroDrPaJdJUroJlno
slaponl strajtl-PaxrJ luarrnruo) roJ solBrurlsS raloruered €'5€ llsvl
'sanlB^ d puE 'sJoJJa pJEpuEts 'salEurrtsa
lurod qlrzvr tr' 6€ alqel ur patrodar are stlnsall 'xrp
-uadde aqr ur puno; aq uu) Iapour srqt ro; epo3
ltL),l!!2 yrl!\4)t00nt,
,thryrsod(,torn a 1/rr + 00r/t) + 00o1 : 't!t ttasod
:sr uorlenba paurqruo ar{J
10tn+001i:101d
'tlqa'torg : t[ty
(Sf) 100r+000i:100d
t/o;1 loog : liog
\!1a +l,t!!lutsod))111 1 1101 : tlt llnsod
:sr uortenba aql Jo ruJo;
leraua8 aq1 'suosrad pue sr(ep qroq ssone r{.1aar;,{ren ol sadols 1 Io^a'I aqr par\olle a/r 'lsrrJ'tr r(q aruerre^ € Ia^a'I aqt pu€ zr r(q paruasard-ar aq III^{ rolrrpard aqr ;o adols atlt qrrmpaleDosse DaJJa uoPuer ar{l Jo a)uerre^ z Ia^a'leql 'srJa;la ruopueJ tuerqru8rsuou Sururerts-uor ,{q paA\olloj 's1azra1 uosrad pue ^{ep aqrqloq re slraJJa uropu€r 3ur,fnads 1o ,(Sarerrsaqr pardope aM 'salqerre^ rorrrpard € Ia^a'I ro Z
Ia^e'I r{rr^{ rfirlrqerren reqr ]]rpard ol SurldrualleroJ aPetu aq ue) as€) e uaqr 'sla^al qroq ro raqlrate pa^rasqo sr r(]rlrqeuen tuergruSrs;1 '(suosrad
sso:re) 1a,re1 uosrad aql ro (suosrad urqrr,u)
1a.r.a1 ,(ep aqt le ,fuerr ,{frTenb uor}JEJalur puElraJJp luarrnruor Surtelar sadols 1 Ia^a'I aql JI sI
{se o1 sn s1t\ollE l/{'lhl tuqt uortsanb xalduror y
sdrqsuorlelaU luaJncuoour Allrqeuen lenpr^rpuualul pue -erlul
'l(rlFnb uorl)€ralura'rrte8au raq8rq pue a.,r.rlrsod ra.trol patrrpardpay;e anrle3au lrtrlenb uorDeralur enue8auramol pue a,rrlrsod raq8rq panrpard na;1e a,rrlrs-od :sna;;a JenossoJ) prordnar puno; am 'JaAo
ur sartrrcnb a,rrleSaupue /DaJJ€ anrreSau rsnrpard ssaupanauuoJ-ue^PP rrurlslP e saPl-os '8uraq-gam
lpra,roaql are dlqenSre reql-DEJalur lelf,os PUE
-ralul ul es€arraP 0tI'-JeJaluI ur asEar)ur I9Iur asear)ur lrun-ol E
aroJaq 0I ^q ParroroJ sanl€A '0€I'0
JoE Pue 0zI'0 Jo rraJJEparnpord ssaulauol u]elq€r uI sluaDgJaoruollJ€Jalur Pue lraJJ€
'Jalair{oq 'suorleDad
UIUIIIII SAED SSOJJE IO
,trrlBnb uolDeJalur Jo
lunoJJe ol Pual lou
srql '(1ar'a1 uosrad aqr)-ro uorleueldxa aqlsalqel ur saleuFsa
ureldxa ssaurTauol ur
uo passarSar olqerre^aql sluasardar ase) t{'z'5€ alqer urrno] IIP roJ EleP or
(gf) >t[!2Itt[o1
;o adfr sn{] roJ uorl
eql o1 PeJPduro) uaq
Ie^a'I 'I
Ia a'I ur uorsll 'ssaullauol tllrM-sarSar eqr ]o epnrr
]9NVH
568 P A R T V . I N T E R I N D I V I D U A L D I F F E R E N C E S I N I N T R A I N D I V I D U A L C H A N G E
-'1IIII
rABLE 3el
Outcome I
*-ff;-lIposarrl I
negaltneg3ftr
ry:1'PUrr r r rneglntr
n"gin,,
modets srmilar to thoq
Iag of two diaries. Res39.6, with point estimavalues associated with
These results indicistill in evidence even IFor example, negativdtinued to precipitate nelater. Positive interaclhibit effects of comparwith prior research do<tency of negative intering affect (Rook, L9902003). Notably, rnoodduring valence-specifiquality even two timesitive affect enhancednegative affect exaceriity during interaction:
Predict ing Intra- andDifferences in Lagger
Of primary interest tointra- or interindividurelating lagged predicmight be functionallywe investigated whetl
TABLE 39.4 Parameter Estimates for ConcurrentRandom Effects at the Day and Person Levels
Outcome Predictor Effect
causality. Here we depart from assessing purelyconcurrent effects, and instead focus on ex-amining the relationship between affect and in-teraction quality when these measures areseparated in time. Lagging predictor variablespermits us to investigate whether, for example,neSative interaction quality at diary f - 1 influ-ences affect at time f. Using lagged predictors alsoallows us to make stronger claims about causalitythan have heretofore been possible. Granted, thelag intervals are not constant, but diaries tendedto be around 92.5 (SD:9.5) minutes apart. Thegeneral form of the combined equation is, forexample:
p o s af t',,1,,, : f s6s t u00k I I sit<
* y rcs(p o sint iit,t-) -l e ijr(1e)
The LISREL syntax for this model is the same asfor the concurrent fixed-effects model; the dataare lagged within-day before submitting them toanalysis in LISREL. Results are reported intable 39.5, with point estimates, standard errors,and p values associated with i166.
In virtually every case, there is a strong laggedrelationship between affect and interactionquality. These effects were in the expected di-rection and were not valence-specific. Positiveinteraction quality at time r - 1 positively pre-dicted positive affect and negatively predictednegative affect at the subsequent time point (i.e.,about 90 minutes later), and negative interactionquality at time f - 1 positively predicted nega-tive affect and negatively predicted positive a{-fect at the subsequent time point. These resultssuggest that the crossover effects of interactionquality on positive and negative aspects of moodare not limited to concurrent effects, but extendto influence mood as much as 90 minutes later.
The reverse causal direction linking affect andinteraction quality was also supported, with onlyone exception. Negative affect at time f - 1predicted less positivity and more negativity ininteractions at the subsequent time point, andpositive affect at one time point predicted morepositivity (but not less negativity) in interac-tions at the subsequent time point. Thus, ingeneral, mood had relatively persistent effectson interaction quality.
The strong lagged effects relating interactionquality and affect suggest that their reciprocalinfluence may last even longer than 90 minutes.We investigated the duration of these effects bylengthening the temporal separation betweenpredictor and outcome. We specified eight
tz - 0.o22 (0.008) p:.004i: :0.022 (0.007) p:.001
iz : 0.014 (o.0r4) p -.324i: - 0.041 (0.016) p:.010
iz - 0.01'4 (0.004) p < .0001is : 0.011 (0.003) p:.001
iz :0.062 (0.013) p < .00001i: : 0.015 (0.009) p:.084
iz :0.058 (0.014) p< .0001t: : 0.0L1 (0.009) p: .22L
For every pairing of predictor and outcomereported in table 39.4, slopes vary randomly ateither the day level or the person level, some-times both. But do interindividual differences inloneliness explain some of these intra- and in-terindividual differences in slopes? To addressthis question, we introduced loneliness as a Level3 predictor of slopes. Loneliness was not cen-tered because grand-mean centering would notalter the effect of interest, and it is unclear howgroup-mean centering should be approached orhow the results should be interpreted. The fullthree-level combined equation incorporating across-level interaction is (e.g.):
posat't' ,,1, : yooo * (yroo * r1k I urck)(posintilk)
+ yssl(lonelinessik)
* y rc1(loneline s s it< x p o sint iit<)
1_ruootrlroik* eiik
Code for this model can be found in the appen-dix. However, no cross-level interaction effectswere found to be significant. That is, thestrength of the relationship between affect andconcurrent interaction quality did not differ as afunction of degree of social connectedness.
Lagged Effects
The results reported thus far do not address thetemporal separation (lag) required for establishing
(18)
'168' : d '(roo'o) 100'0- : rori /ssaulTauol uIsa)uaraJJIp IEnpl^pul .{q parrrpard dlluerr;ru8rs]ou sE.,v\ sadols ur ,(trlqerrezr IPnpr^rPurPJrul
iz : -0.001 (0.004) p:./25i: : 0.002 (0.003) p: .334
iz : 0.007 (0.032) p: .8r7ig : 0.014 (0.018) p:.43s
What distinguishesresearch is that weof the co-occurringtures of socialdeparture from typspectrve/ assessmentsdividuals with whompositive or negative1989; Rook, 2001), orsocial exchanges aretive (Newsom et al.,interaction qualityvariance, and this maas9ocratlons we
negative aspects ofHowever, this did noting possible explanatness/loneliness divariables and their
What mightpositivity and lowerteraction qualitydividuals? Our datasocially connected indiindividuals in theirsocial experiences,characterized bytions of positivityne8atlvrty ln
results of our laggedcially connected indilonger-lasting effectsmood, or of positivethan do lonely individoes not account forliness differences inand negativity inknow from past reseathat these same indiconnectedness/lonelispent with others, ruliopportunities as anferences. On thethe possibility thatdividuals haveuon partners/ painclined towardinteraction qualitymaintain positivity iindividual.
In sum, our dataof greater perceivedconnected individualssitivity to be self-reinof affect and i
posaff
negaf{
negaff
poslnt
posint
negint
negint
posrntf, 1
negintr - l
poslntl - 1
ne$rntl 1
posaff, - 1
negaff, 1
posaff, 1
negaff, 1
In additioru the effect of negative affect at timef - 1 on positive interaction quality at time tvaried significantly across people. We investigatedthe ability of loneliness to predict this randomeffect. The combined threelevel eouation is:
posint,,p,,
: Yooo _F (Yroo * r1i1,* ulsk)(negot't' iir,,-)
a y6rr(lonelinessip)
I y .61(I o n e line s s p x n e gat' t',,p,, -r)
I ussk -l rsir< * eiik (27)
Intraindividual variability in slopes was not sig-nificantly predicted by loneliness, ?ror :0.0004
(0.004), p: .e27.Given that there was some evidence for Lag 2
effects (see table 39.6), we were curious to discoverwhether there existed significant intraindividualvariability in dayJevel or person-level slopes forLag 2 effects and, i{ so, whether this variabilitycould be predicted by interindividual differencesin loneliness. We examined day-level and person-level random effects of Lag 2 predictors. None ofthese slopes exhibited significant variance at eitherthe day or person level.
Goncluding Remarks
In this study, MLM afforded theoretical tests andinsights that would not have been visible fromother perspectives. Had we taken an ordinary
least squares regression approach, for example,we would not have seen that variance in affectand interaction quality was evident not onlyacross diaries, but also across days and acrosspersons. Specifically, variability in mood andinteraction quality associated with momentarycircumstances exceeded day-to-day and person-to-person variability in mood and interactionquality. Moreover, MLM enabled us to see thatloneliness/social connectedness was more pow-erful in explaining interindividual variance thanmomentary or daily variance. This finding pro-vides a first clue that loneliness/social con-nectedness is characterized to a greater extent bya pervasive enduring influence over the affectiveexperience of everyday circumstances (includingsocial interactions) than by a transitory influenceon the experience of momentary circumstances.
Second, we saw that concurrent affect and in-teraction quality were reciprocally related acrossvalence domains, as has been suggested by priorresearch (Cunningham, 1988a, 1988b; Mclntyreet al., 1990, 1991; Newsom et al., 2003). Im-portantly, lag analyses revealed potentially causalrelationships linking interaction quality and af-fect: These relationships appeared reciprocal, werein the expected directions, and acted across va-lence domains. In addition, negative aspects ofsocial interactions had a particularly long-lastinginfluence on negative affect, consistent with the"negativity effect" reported by other researchers(Rook, 1990; see also Newsom et al., 2003).
-aI aqr sralso} Alrl€nD uorDeralur Pue r)aJJe Josureruop eql ssoJJe Sunrolurar-;1as aq o1 .{lrrrrlrs-od rol ,{ruapual aq} pue slenpwpur pal)auuor,{11enos Suoure ,(lnrlrsod parrrarrad ralearS ;ouorleurquo) aqr reqr lsaSEns elep rno 'urns uI
'l€nPmPur
pal)auuo) ,(lpDos aqt ur ,{}I^rtrsod ure}ureurpue ralsoJ dlaq dqaraql pue ,,trlenb uonleraturpue DatJE ur .{lrnrlrsod ralear8 pJe,lot paul1Jur,tpepuls aq ,(eur teq] srautred 'srautred uor]-)BJalur Jallaq ol ssarrB ;atuarS a^eq slenpl^p-ur patrauuor .{11enos tet{} fullqrssod aqr
lno olnJ louue) aA /puer{ Jar{}o arl} uO 'saf,uala}
-tp rraJJ€ ro1 uouuuBldxa uB sp sarlrunlroddo
IeDos Jo .{ruanbary tno Surlnr 'sraqlo qlrin tuadsaruu ur sa)uaroJtIP ssaurlauol/ssauPal)auuo)
IeDos lrclrqxa ]ou PP slenPl^lPul aulEs esaql l€ql(EOOZ ''F la {o11ivreg) q)reasar rsed urory moulailr 'uourppe ul 'sur€ruop aseql ur dlrzrrle8au pue,ttrnrlrsod yo aar8ap IIEraAo uI sarueraJ1lp ssaull-auol/ssouPaDauuoJ IerJos roJ luno))e lou saoPuorteueldxa srql os 'slenplrrrpur d1auo1 op u€qr'.frrlenb uor])eratur uo na;1e arrrlrsod;o ro 'pootu
uo suorl)eJatur aurrlrsod Jo sl)aJJa Eurrsel-ra8uolaruarradxa rou op slenprlrpur patrauuor .{11en-os leq] elelrPur sasr(1eue pa33e1 rno to sllnsaraql 't)eJJB pue ,ftrlunb uolDEralur ut dtrzrtleSau
1o suoudorrad pornpar pue ,(lrrrursod Jo suol]-darrad pa)uequa dlluarsrsrad dq pe'{ralrereqrssauPal)euuof, IETJoS qtr1t| 'sefueIJsdxJ
'drqsuorlelorralur rraql pue selqerre^aruo)tno asaqt ur sarueraJJIP ssaulTauol/ssau-patf,auuo) Ierros roJ suorleueldxa alqrssod 3ur-lenlela ruorJ sn lueaard tou pp sql 'rane^\oH'f)aJte puE ,{frlenb uortJeratul Jo s}radse arrtte8aupue aarlrsod eql Suoue pa^Jasqo a.tr suoller)osse3r{r o} Pernqrrluol aAeq ,,{eur srql Pue
'o)ueue^
poqreu rleqs t)eJJe pue {rrlenb uollrerelul
Jo sluaussasse rng '(ggg7 ''p la urosrvra5l) a,rrl-BBau ro arutrsod r(11eraua8 are sa8ueqcxa lenosrplt{Ir\ r.{}r1v\ druanbar; aqr ro '(1997 'IooU 1686I''1e la qrurl ''8'a) saSueqrxa earle8au ro a.trlrsod.{11erauaB azreq stuednrtred uroqrnl qrrru slpnprr'lp-ur Jo Jaqunu oqr Jo sluorussosse 'arrrpads
-ortar ,{lluanberj puB lerrdfr ruorJ a"rnlredapluergruSrs e sr srrll 'suoureJalur
FDos Jo saJnl-eal a,rueSau pue anrlrsod Surrrnrro-or aql 1osluarussesse aur1uo parrnbre ar'l ]Er{} sr r{f,Juasalrorrd ruory sSurpuu asaqr saqs-rnSurrsrp reqM
P A R T V . I N T E R I N D I V I D U A L D I F F E R E N C E S I N I N T R A I N D I V I D U A L C H A N G E
--lIIII
RANDOI4I :CONS; IRANDOM2:CONS; IRANDOI"I3:CONS; ICOVIPAT:DIAG; ICOV2PAT:DIAG; ICOV3PAT:DIAG; I
II
PRELIS Code for Con{I
OPTIONS OLS:YEIn T r T T F - a ^ n a t t r r d
" *- - ''1
} , I ISSING-DAT:-1c v , r - . \ d a t ^ \ d . l
IDl :BEEPNUI" I ; IT n ? Q r r T l n v n a v . I
L v v L v L ' ' ,
I D 3 _ S U B N U M ;RE SPONSE:POSAIFIXED:CONS POS]RANDO} l1 :CONS;RANDOM2:CONS PlRANDOI,I3:CONS P1COVIPAT:DIAG;COV2PAT:1 ! Re!2 3 i
COV3PAT:1 ! Rec2 3 ;
measures and addressed important theoreticalquestions that could not otherwise be addressed.For instance, the relationship between variablescan be studied within nesting units (e.g., days,people), which allows us to ask questionsabout both interindividual variability and in-traindividual variability.
We further demonstrated that longitudinaldesigns need not focus on time as a variable oreven on trends. The capability of testing laggedeffects is very useful, and is an interesting al-ternative to incorporating time as a predictor(which is how most researchers study trends).We were not interested in trends over time(i.e., growth curves), but rather concurrent andlagged effects of other relevant variables. Ourstudy was characterized by repeated measure-ments of the same individuals, yet our predic-tions and hypotheses had little to do withexamining growth or trajectory over time.
Finally, in psychological research, there is anunfortunate tendency to address longitudinal
hypotheses using cross-sectional data. Temporalseparation is a necessary, but not sufficient,condition for causality (Gollob & Reichardt,7987). To address causality, we used laggedprediction in MLM of longitudinal data. Theexamination of lagged effects as illustrated inthis chapter should be undertaken more oftenbecause this modeling strategy recognizes ex-plicitly that independent variables require timeto exert effects on dependent variables. Modelinglagged responses in MLM is a novel but powerfulstatistical approach that overcomes these lim-itations in traditional longitudinal analyses.
Acknowledgments This work was supportedby Program Project Grant PO1 AG18911(social isolation, loneliness, health, and theaging process) from the National Institute onAging and by the Mind-Body Network ofthe John D. and Catherine T. MacArthurFoundation.
Appendix Selected LISREI Gode
PRELIS Code for Nul l Model
OPTIONS OLS:YES CONVERGE:O. OO1 MAXITER:1OO OUTPUT:STANDARD ;O L S - Y E S : U s e O L S e s t l m a t e s a s s t a r t v a l u e sU U N V . t r K G l , - U . U U . L - U O n V e I 9 € n C e C I a ! e I I O n - . U U II i IAX ITER:100 : l l a x imum o f 1OO i t e r a t i ons
TITLE :Nu l l l , I ode I ; ! T i t I e o f ana l ys i sI4 ISS ING_DAT : -999 .000000 ; ! I u l i - ss ing da ta code : -999S Y - ' C : \ d a t a \ d a t a l . P S F ' ; ! L o c a t i o n o f i l a t a , i n P R E L I S d a t a f o r m a tID1 :BEEPNUM; ! Leve l -1 un i t i den t i f i e r - d i a r y numberID2 :STUDYDAY; ! Leve l -2 un l t i den t i f l e r : day o f s tudyID3 :SUBNUM; ! Leve l -3 un i t i den t i f i e r : sub jec t numberRESPONSE-POSAFF ; ! Dependen t va r i ab fe -PANAS pos i t i veF I X E D : C O N S ; ! R e q u e s t s p o i n t e s t i m a t e o f f i x e d e f f e c t
PRELIS Code for Pre
OPTIONS OLS:YE'1'1'-t 'LE:U I O SS-L(M I S S I N G _ D A T : -S Y - ' C : \ d a t a \ d iID1:BEEPNU}I ;ID2:STUDYDAY;ID3=SUBNU]VI ;RE SPONSE:POSA,F I X E D : C O N S P O S
t i o n t e r mRANDO}I1 :CONS;RANDOI42:CONS FRANDOM3:CONS !COVlPAT:DIAG;COV2PAT:1
2 3 ;COV3PAT:1
Nofes
1. The subscript nccients in MLM can btconvention becomes tthird level is added, athe fact that differentsubscripting strategierwe adopt a variation r
f ) i : r r z - l o r z a l i n i - o r n a n l - i c r a n A n m
f ) a r r - l o r r o l i n l - a r c a n l - i q r a n d n m" " f
P e r s o n - l e v e l i n t e r c e p t i s r a n d o mD i a r y - l e v e I c o v a r i a n c e m a t r i x i s d i a g o n a lD a y - l e v e I c o v a r i a n c e m a t r i x i s d i a g o n a lPe rson - l eve I cova r i ance ma t r i x i s i l i aqona f
PRELIS Code for Concurrent Fixed-Effects Model
OPTIONS OLS:YES CONVERGE:O. OO1 MAXITER:1OO OUTPUT=STANDARD ;T I T L E : C o n c u r r e n t F i x e d E f f e c t s ;M I S S I N G - D A T : - 9 9 9 . O O O O O O ;S Y : ' C : \ d a t a \ d a t a 1 . P S F ' ;I D 1 : B E E P N U } 4 ;ID2:STUDYDAY;ID3:SUBNU}4;RE SPONSE:POSAFF ;F I X E D - C O N S P O S I N T ; ! R e q u e s t s p o i n t e s t i m a t e s o f i n t e r c e p t a n d s f o p e
rreJ aql stuasardar oroi ur 0 lsrrj aql ialduExe roJ /9
uorl€nb:l uI suoqenba € la^al pue /ZP^e'I 'I ya,ra1aql ur luaIJIJJao) ar{t Jo uortrsod aq} stuasardarra8alur aqt 'ra8atur ue dq pareldar sr sldrnsqns
asaqr Jo auo uaqM uos.rad pue ',{ep '.{rerp '.{1anrt
-radsar 'alouep lpue
'{'l qrlqm ut '(5,5,51) ralsogpuu sraplrug pup (2002) >1i{rg pue qsnquapne1
dq pa.,{oldura uouelou ar{l Jo uouerr€^ e tdope a.u'raldeqr srql yo sasod.rnd.ro1 'sar8aturts SurldursqnstuareJJlp asn ol Puat sror{tnB rueraJJIP l€qr r)ej aqr,{q parerrtduor raqun1 sr pu€ /pappe sI la^el prrqrE uar{r* Sursnluor aJour uela seruoJaq uor}ualuo)
uorlelou aq1'3ursn;uor aq uer I/{'l}\l ur stuarJ-rJJaoJ roJ padoldura uorlutou rdrnsqns aq1 I
^,{^-^ ̂,.- - ,{^^ -- euv Lr yus +ueJrc
J CUVCNVJS
reuo6erTeu(
Teu06r
IO/\
II
leuroJ efep s:
J CUVCNVJS
rnr{}rvDel^ 'I auryo ryo,nray dpog-
uo alnlrlsuJ IeuorlENaqt Pue
'qrlEeq 'ssa
II6SIDV IOd IUparroddns sem {rom sl
'sasr{1eue leurpnrrSu
-UIIT ASOr{1 SaTUOTJaAO I
p;ramod rnq lalou e sr tr8ur1apo61'salqerr€^ luaparur] arrnbar selqerJel lr-xa sazruSorar rtSalertsuouo aloru ua{ElJaPunur Pat€rtsnll se spatjaaql 'elep
leurpnrr8uolpa33e1 pasn arra, ',{lr1es'lPrErPrau
4 qolioDl'tuarlu;ns lou 1nq 'drr
produral'€lep leuorpas
l9NVH3
saioN
lt zT:JVd€NOJ
" i:rvuznor.9VTC:JVdTNOJ
j .lNrsod sNoJ:elllocNvu: JNISOd SNOC:ZWOCNVU
! SNOJ:TWOONVHu-raJ uorl
-sefaJur pue JNIsod loJ slceJJa pexTJ i !J,NISOdJn JNISOd SNOJ:CSXr,{j ,{,EVSOd:sSNOdS Eu
0001 0NV sN0u-3vulrNl lvlc0s J0 9Nlll00l l '6e ulrdvHJ
5 7 4 P A R T V . I N T E R I N D I V I D U A L D I F F E R E N C E S I N I N T B A I N D I V I D U A L C H A N G E
_lIIII
for Methodology of Ience,75 ,41-56. I
Mclntyre, C. W., WalCross, S. A. (1991).
lrnteracnon on poslJBulletin ot' the Psych
Mclntyre, C. W., Watsoj(1990). The effects olland test stress on p!Bulletin ot' the Psychdi
Mikulincer, M., Shaver]Attachment theory ]dynamics, developn]sequences ot atta!Motiaation and Emo
Myers, D. G. (2000). Thihappv people..4merii
Vy"tr,-n. C., A Oi"n.fPsychological Sciencl
Newsom, J. T., NishistfRook, K. S. (2003). 'three domains of posichanges: A longitudirmeasures. Psychology
Rasbash, J., Browne, WPlewis, L, Healy, M., eMLzoiN. London: Cen
Raudenbush, S. W., &archical linear mod.tanalysis methods (2CA: Sage.
Rook, K. S. (1990). Iadults' social relatior
that this Level 3 fixed effect is ultimately related tothe intercept in the Level 1 equation; the 1 re-presents the fact that it is related to the first slopecoefficient in the Level 2 equation; and the final 0represents the fact that y01s is the first (intercept)coeff icient in the Level 3 equation.
2. Centering sometimes improves the stability ofestimation by reducing collinearity with other pre-dictors, and often renders uninterpretable parameterestimates interpretabie. For example, the intercept intraditional regression is interpretable as the va-lue ofthe dependent variable when all predictors equalzero.If a predictor variable has no meaningful zeropoint, then mean centering allows the intercept to beinterpretable as the predicted value of the dependentvariable at the mean of the predictor. In two- andthree-level models, centering is more complicated(see Kreft & de Leeuw, 1998; Kreft, de Leeuw, &Aiken, 1995; Raudenbush & Bryk, 2002, for generalguidance on centering). In this chapter, we limitanalyses to uncentered data because the effects ofgreatest interest are not altered by the most widelyemployed kind of centering.
3. For the sake of brevity, only those resultsdirectly relevant to the question at hand are re-ported. More details are available from the authorsupon request.
Ret'erences
Affleck, G., Tennen, H., Urrows, S., & Higgins, P.(1994). Person and contextual features of dailystress reactivity: Individual differences in rela-tions of undesirable daily events with mooddisturbance and chronic painintensity. lournal olP er so nality and S o cial P sy ch olo gy, 66, 329 -3 40.
Barrett, L. F., & Pietromonaco, P. R. (1997). Accu-racy of the five-factor model in predicting per-ceptions of daily social interactions. Personalityan d S o ci al P sy ch o I o gy B ull e tin, 23, I77 3-1187 .
Baumeister, R. F., & Leary, M. R. (1995). The needto belong: Desire for interpersonal attachmentas a fundamental human motivation. Psycholo-gical Bulletin, 117, 497-529.
Bryk, A. S., Raudenbush, S. W., Cheong, Y. K., &Congdon, R. (2000). HLMS: Hierarchical linearand nonlinear modeling. Chicago: ScientificSoftware International.
Cacioppo, J. T., Ernst, J. M., Burleson, M. H.,McClintock, M. K., Malarkey, W. B., Hawkley,L. C., et al. (2000). Lonely traits and concomitantphysiological processes: The MacArthur socialneuroscience studies. lnternational Journal olP sy chophy siolo gy, 3 5, 1,43-154.
Cacioppo, J. T., Hawkley, L. C., Crawford, L. E.,Ernst, J. M., Burleson, J. M., Kowalewski, R.8.,
et al. (2002). Loneliness and health: Potentialmechanisms. Psychosomatic Medicine, 64, 407-477.
Clark, L. A., & Watson, D. (1988). Mood and themundane: Relations between daily life eventsand self-reported mood. lournal ot' Personalityand Social Psychology, 54, 296108.
Cunningham, M. R. (1988a). Does happiness meanfriendliness? Induced mood and heterosexualself-disclosure. P ersonality and S o cial P sy chol-ogy Bulletin, 74, 283197.
Cunningham, M. R. (1988b). What do you dowhen you're happy or blue? Mood, expectancies,and behavioral interest. Motioation and Emo-t ion, 12,309-331.
du Toit, M., & du Toit, S. (2001). lnteractiaeLISREL: User's guide. Lincolnwood, IL: Scien-tific Software International.
Finch, J. F., Okun, M. A., Barrera, M., Zaurra, A.J,,& Reich, J. W. (1989). Positive and negativesocial ties among older adults: Measurementmodels and the prediction of psychological dis-tress and well-being. American Journal ofCommunity Psy cholo gy, 17, 585-605.
Fredrickson, B. L., & Joiner, T. (2002). Positiveemotions trigger upward spirals toward emo-tional well-being. Psychological Science, 1.3,172-t75.
Goldstein, H. (1995). Multile'sel statistical models(2nd ed.). London: Arnold.
Gollob, H. F., & Reichardt, C. S. (1987). Takingaccount of time lags in causal models. ChildDezt elopment, 58, 80-92
Hawkley, L. C., Browne, M. W., & Cacioppo J. T.f20051. How can I connect with thee? Let me countthe ways. P sy cholo gi cal S ci enc e, 1, 6, 7 98-A04.
Hawkley, L. C., Burleson, M. H., Berntson, G. G.,& Cacioppo, J. T. (2003). Loneliness in everydaylife: Cardiovascular activity, psychosocial con-text, and health behaviors. Journal ol Person-ality and Social Psychology, 85,105-120.
Hox, J. Q0O2). Multileoel analysis: Techniquesand applications. Mahwah, NJ: Erlbaum.
Jrireskog, K. G., & S<irbom, D. (1995). LISREL Iuser's ret'erence guide. Chicago: Scientific Soft-ware International.
J<ireskog, K. G., Srirbom, D., du Toit, S., & du Toit,M. (1999). LISREL 8: Nero statistical t'eatures.Chicago: Scientific Software International.
Kreft, I., & de Leeuw, J. (1998). lntroducing mul-tileoel modelins. Thousand Oaks, CA: Sage.
Kreft, I. G. G., de Leeuw, J., & Aiken, L. S. (1995).The effect of different forms of centering inhierarchical linear models. Mubiaariate Beha-oioral Research, 30(1), 1-21.
Larson, R., & Csikszentmihalyi, M. (1983). Theexperience sampling method. Neto Directions
III suourartl maN 'Poql{
aqr '(€s6r) 'rl{ 1^F'{l
-oqag awtnalruwrz'r|ur Surralual 1o surro;l'(SeOr) 'S 'T 'uaTr-V T I
'sar8alerls pelEIaJ-tuatupelle Jo saruanbas-uol anrlruSor pue 'luarudola,rap 'srnueudp
aq1 :uorlelnSar pa;Je pue droaqr ruauqJelrv'(6OOZ) 'C '3ara4 ry ''U 'd 'ra^Bqs /'t\l /rarullruI1hl' €rfWI' B Z' n4 a xr o S cruto uo r1t frs 4 a r1l t o uu a 11n g'Dage anne8au pue anrlrsod uo ssaJts lset pue'asr)taxa 'uoq)elalur
Ier)os Jo slralla aq1 '[9661)'1 'y'rueq8uruunf rB "CI'uosluM ",14'3 'ardrup61'Ol-19'62'frranog ltuouoqtfis4 aql lo uqallng'na;;u alue8au puu alltlsod uo uorlreralut