This article was downloaded by: [Harvard Library]On: 06 October 2014, At: 10:49Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: MortimerHouse, 37-41 Mortimer Street, London W1T 3JH, UK

MemoryPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/pmem20

Impact of current grief on memory for past griefin spousal bereavementNigel Field a , Larry Thompson b & Dolores Gallagher-Thompson ba Pacific Graduate School of Psychology, CA, USAb VA Palo Alto Health Care System, CA, USAPublished online: 22 Sep 2010.

To cite this article: Nigel Field , Larry Thompson & Dolores Gallagher-Thompson (2006) Impact of current grief onmemory for past grief in spousal bereavement, Memory, 14:3, 297-306, DOI: 10.1080/09658210500243747

To link to this article: http://dx.doi.org/10.1080/09658210500243747

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”)contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensorsmake no representations or warranties whatsoever as to the accuracy, completeness, or suitabilityfor any purpose of the Content. Any opinions and views expressed in this publication are the opinionsand views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy ofthe Content should not be relied upon and should be independently verified with primary sources ofinformation. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands,costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly orindirectly in connection with, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial orsystematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distributionin any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found athttp://www.tandfonline.com/page/terms-and-conditions

Impact of current grief on memory for past grief inspousal bereavement

Nigel P. Field

Pacific Graduate School of Psychology, CA, USA

Larry W. Thompson and Dolores Gallagher-Thompson

VA Palo Alto Health Care System, CA, USA

This study examined the impact of changes in level of current grief over time on changes in memory forpast grief. Following from previous work on the impact of current affective state on memory for pastaffectively salient events, bereaved individuals who underwent a greater reduction in grief over time wereexpected to show a similar pattern of changes in memory for past grief. A sample of conjugally bereavedparticipants completed measures of current grief and memory for past grief at the time of the loss of theirspouse at each of the 2, 6, 13, and 30 month post-loss time periods. As predicted, those who reportedgreater reduction in grief over the course of the study were more likely to remember their past grief asprogressively less severe relative to those who underwent less reduction in grief over time. A nonrecursivecausal model analysis verified that changes in memory for past grief were explained by changes in currentgrief. These results were discussed in terms of contemporary reconstructive memory accounts of the effectof affective state on memory.

There is an extensive literature showing that whenindividuals are asked to recall the severity of pastdistress, systematic biases exist (see Safer &Keuler, 2002). Furthermore, different factors havebeen identified that have bearing on the nature ofthese biasesÐmost noteworthy is a person'saffective state at the time of recall (Safer &Keuler, 2002). The present study extends this lit-erature in examining systematic changes in mem-ory for past distress as a function of changes overtime in current level of distress. At different pointsin time following the death of a spouse, bereavedindividuals completed a measure assessing theirmemory for past grief at the time of the death aswell as a measure of their current grief level. Inknowing that conjugal bereavement is a highlystressful life event in which the normative patternof recovery towards pre-loss baseline levels ofdistress occurs gradually over time (Bonanno &Kaltman, 2001; Thompson, Gallagher-Thompson,

Futterman, Gilewski, & Peterson, 1991), this is anideal event in which to examine the impact ofchanges over time in bereavement-related distresson memory for intensity of past distress at the timeof the loss. To our knowledge, this study repre-sents the first multiwave longitudinal investigationof the impact of the trajectory of change in level ofdistress over time on memory for past distress.

Contemporary perspectives on memory forpast events highlight its reconstructive nature.According to this framework, autobiographicalmemory, including memory for past affectivelysalient events, is malleable. Such memories thusare assumed to represent transitory dynamicmental constructions as opposed to being indelible(Bartlett, 1932; Conway & Pleydell-Pearce, 2000).Moreover, a person's current attitudes, goals, andaffective states are believed to play an importantrole in what is recalled (Safer & Keuler, 2002). Infact, as memory for details of the past event fade

MEMORY, 2006, 14 (3), 297±306

# 2005 Psychology Press Ltdhttp://www.psypress.com/memory DOI:10.1080/09658210500243747

Correspondence should be addressed to Nigel P. Field, Pacific Graduate School of Psychology, 935 E. Meadow Dr., Palo Alto,CA, 94303, USA. Email: nfield@pgsp.edu

Dow

nloa

ded

by [

Har

vard

Lib

rary

] at

10:

49 0

6 O

ctob

er 2

014

over time, these factors play an increasinglyimportant role in what is recalled (Higgins &King, 1981).

A widely used method for examining memorybiases in the recall of emotional intensity is the``single occasion'' paradigm in which participantsare administered a measure of distress in theinitial wave of data collection and at a later pointin time are given an incidental recall measure inwhich they are asked to recall their initial distressratings using the identical stress scales. In one suchstudy on recall biases in memory for pre-psy-chotherapy distress, Safer and Keuler (2002) hadpsychotherapy clients complete a measure ofgeneral psychological symptomatology at the timeof entry into treatment, and again immediatelyfollowing the completion of psychotherapy inwhich they were unexpectedly asked to againcomplete the symptom measure exactly as theyhad answered it at intake. Differences betweenthe within-person initial and termination symp-tom measure scores provided a measure of mem-ory bias. Similarly, in a study of memory distortionin recall of prior grief-related distress conductedby Safer, Bonanno, and Field (2001), conjugallybereaved participants were administered grief-related symptom measures at 6 months after thedeath of their spouse and again at 5 years post-loss. At the 5-year point, they were asked to recallhow they felt at the time of the 6-month report bytrying to answer the symptom measures exactly asthey had completed them at the initial 6-monthpoint.

Because in both of these studies participantstypically underwent a reduction in level of distressfrom the initial to follow-up assessments, andgiven that considerable individual differences alsoexisted in the extent of such change, it was possi-ble to determine whether current level of distressmoderated memory for past distress. In both stu-dies, there was a tendency for those who remainedmore distressed at the time of recall to over-estimate their level of initial distress, whereasthose with lower levels of distress at the time ofrecall were more likely to remember accurately orto even underestimate their initial distress. Thispattern of findings has been replicated in otherstudies (Bryant, 1993; Conway & Ross, 1984;McFarland & Alvaro, 2000).

Motivational and cognitive factors have beenconsidered in accounting for these findings. Amotivational explanation offered is that those whoremain more distressed over time may be moti-vated to exaggerate their recalled level of past

distress in order to see improvement. In effect, tothe extent that affect regulation is an importantfactor in memory of past distress, there may be ageneral tendency to overestimate past distress incoping with current distress. Thus, memory of pastdistress as worse than it actually was might serveto make a person feel better about his or hercurrent level of distress (Safer et al., 2001). How-ever, this should only apply for those who con-tinue to experience elevated levels of distress.Those with lower levels of distress should feel lesscompelled to engage in such efforts (Safer et al.,2001).

Construct accessibility provides a com-plementary cognitive explanation for how currentaffective state influences memory of past affectiveepisodes. Affective state is known to play animportant role in the accessibility of mentalrepresentations (Bower, 1981; Forgas, 1995; Hig-gins & King, 1981). Given the reconstructivenature of episodic memory in which judgements ofpast events are heavily influenced by the mostmentally accessible information in memory, to thedegree that an individual's current affective statehas an impact on mood-congruent information inmemory, current affective state should influencememory for past affective states. Specifically,higher levels of current distress are likely to leadto an overestimation bias in memory for pastdistressing events (Bower, 1981; Safer & Keuler,2002). This could explain why individuals whoshow less reduction in distress over time are morelikely to indicate overestimation biases in memoryfor past distress.

PRESENT STUDY

This study builds on previous work by examiningmore closely the trajectory of change in memoryfor the given past distressing event. It involves adifferent emphasis from past research in focusingless on memory accuracy per se and more on thesystematic relationship between current distressand memory for past distress. This study examineschanges in recall of intensity for past grief in aconjugally bereaved sample. In knowing that griefsymptoms decrease gradually as a function of timesince the death (Bonanno & Kaltman, 2001;Thompson et al., 1991), this is a fitting context inwhich to examine the effect of change in currentgrief on changes in memory for past grief. Thisstudy provides a rare opportunity for examiningthis, in having participants rate memory for grief

298 FIELD, THOMPSON, GALLAGHER-THOMPSON

Dow

nloa

ded

by [

Har

vard

Lib

rary

] at

10:

49 0

6 O

ctob

er 2

014

at four different points after the death so that it ispossible to conduct a latent growth curve analysisof the trajectory of change. Because measures ofcurrent grief were also obtained at each of thesefour waves of data collection, it was possible toexamine the relationship between the trajectory ofchange in grief in relation to the trajectory ofchange in remembered grief. This study thereforeextends previous work on in impact of currentdistress on remembered distress, in identifying apossible systematic relationship between the tra-jectory of change in each. Specifically, the studyexamines how changes in level of grief moderatememory for past grief.

In line with the above-mentioned motivationaland cognitive explanations on the effect of currentaffective state on memory for past distress, wepredicted a positive relationship between changein level of grief over time and memory for grief atthe time of the death, such that those whounderwent the greatest reduction in grief overtime would indicate a greater tendency toremember their grief as less severe over timerelative to those who underwent less reduction incurrent grief symptoms over time. Furthermore,given that systematic change over time in memoryfor past grief is attributable to changes in level ofcurrent grief, we predicted a unidirectional causalrelationship between the two, such that changes incurrent grief would explain changes in memory forpast grief.

METHOD

Participants

This study is part of a larger database examiningadjustment to conjugal bereavement in later lifeconducted in the early 1980s. The study involved afour-wave data-collection longitudinal design inwhich participants completed interview andquestionnaire measures at 2, 6, 13, and 30 monthspost-loss. For a more detailed description of thedatabase, see Thompson et al. (1991).

Bereaved individuals over the age of 50 whosespouse had died within the previous 2±4 weekswere located through death certificates at the LosAngeles County Health Department and mailed adescription of the study along with a stampedpostcard to indicate their interest in participating.Of the 2450 mailings sent, 735 responded. Ofthese, 228 (129 women and 99 men) participatedin the study. A series of chi-square and t-tests

confirmed that those who completed the study didnot differ from those who dropped out at anypoint during the study on the demographic vari-ables or grief symptom levels (p > .05).

These participants ranged in age from 53 to 90years (M = 68.08, SD = 8.05), were largely Cau-casian (94.7%), well educated (69.1% greater thanhigh-school education), and of moderate eco-nomic status (53.8% greater than $20,000).

Overview of procedure

Participants completed four interviews and ques-tionnaire packets over the course of the study.Each interview included measures coveringdemographics, religious practices, coping, lifestresses, social support, and marital satisfaction.The questionnaire packet included symptominventories. For the purposes of the present study,only the demographic variables and the TexasRevised Inventory of Grief are focused on in theanalyses.

Measures

Texas Revised Inventory of Grief(TRIG). The TRIG (Faschingbauer, 1981) wasdesigned to assess grief-specific symptoms. Itconsists of two parts: Part 1 focuses on past griefseverity at the time of the death and Part 2addresses present grief level. The items for bothscales are rated on 5-point Likert scales rangingfrom ``completely false'' to ``completely true''.

The Part 1 past grief scale consists of eightitems addressing remembered grief-specificsymptoms at the time of the death. Participantsare given the following instructions in answeringthis scale: ``Think back to the time your spousedied and answer all of these items about yourfeelings and actions at that time''. Having parti-cipants answer the items with respect to the timeimmediately following their spouse's death pro-vides a means for assessing changes in remem-bered grief as a function of time. The past griefscale items address aspects of grief-related distressthat are commonly reported in the period shortlyfollowing the death of a loved one (e.g., ``I found ithard to sleep after this person died''; ``I found ithard to work well after this person died''). Thisscale has good internal consistency as well as dis-criminant validity in distinguishing severity of pastgrief in terms of gender of the bereaved and age ofthe deceased (Faschingbuer, 1981). The internal

MEMORY FOR GRIEF 299

Dow

nloa

ded

by [

Har

vard

Lib

rary

] at

10:

49 0

6 O

ctob

er 2

014

consistency coefficient alphas of the TRIG mea-sure of past grief in the present study sampleranged from .78 to .86 across the four adminis-trations of the measure.

The Part 2 present grief scale is a 13-itemmeasure of current grief severity. Participants aregiven the following instructions in answering thisscale: ``Now answer all of the following items bychecking how you presently feel about this per-son's death''. Participants are also instructed notto look back at how they answered the past griefitems while completing the present grief scale.This instruction served to ensure that participantsclearly distinguished between past and presentgrief in their responses. The present grief scaleincludes items addressing extent of distress (e.g.,``I still get upset when I think about the personwho died''), preoccupation (e.g., ``I can't avoidthinking about the person who died''), missing(e.g., ``Sometimes I very much miss the personwho died''), and non-acceptance (e.g., ``I amunable to accept the death of the person whodied''). These items are different from those usedin the past grief scale in capturing aspects of griefthat are known to continue well on after the death.This scale has good internal consistency (Fas-chingbauer, 1981). Support for its discriminantvalidity was shown with respect to sex differences,time since the death, and relation to the deceased(Faschingbauer, 1981). The internal consistencycoefficient alphas of the TRIG measure of presentgrief in the present study sample ranged from .87to .90 across the four administrations of themeasure.

RESULTS

Change in grief-specific symptomsover time

A latent growth curve analysis was conducted toexamine the trajectory of change in symptomlevels over time during the course of the 30-monthpost-loss bereavement period. In essence, thisinvolves estimating an intercept and slope scorefor each participant based on his or her symptomscores at 2, 6, 13, and 30 months post-loss. Theintercept identifies an individual's estimatedsymptom level score at the time of entry into thestudy, while the slope represents a measure ofchange in his or her symptom levels across thefour waves of data collection. Mean scores are

then estimated for the intercept and slope valuesfor the group as a whole.

The growth curve analysis was undertakenusing the AMOS 4 structural equation modellingapproach statistical software (Arbuckle &Wothke, 1999). This software provides an efficientmeans for estimating model fit with randomlymissing data. A linear growth curve, treating eachinterval between successful waves of data equally,represented the best fit with the data. A pictorialrepresentation of the latent growth model isshown in Figure 1. This growth curve modelrepresents an acceptable fit with the data, basedon an assortment of fit indices: w2(df 6) = 44.22, p =.35; CFI = 1.00; RMSEA = 0.02 (0.00 ± 0.09). Anintercept value of 44.22 (CR = 58.55, p < .001)represents the average grief symptom level atentry into the study at approximately 2 monthspost-loss. The variance value of 95.21 for theintercept indicates that 95% of bereaved partici-pants' grief symptom scores at entry into the studyranged between 34.46 and 53.98. The significantsize of the variance estimate here attests to thenoteworthy individual differences in level of griefseverity at entry into the study. A significantnegative slope of ±2.48 (CR = ±11.81, p < .001)indicates that participants' grief symptomsdecreased on average 2.48 points between suc-cessive waves of data collection. Based on thevariance value of 0.44 for the slope, 95% of indi-viduals' change scores ranged between ±1.15 and±3.81.

Recall of past grief symptoms

A similar latent growth curve analysis to theabove was conducted on the four waves ofremembered past grief symptoms. Again, a modelin which each interval was treated equally repre-sented the best fit to the data. A pictorial repre-sentation of the model is shown in Figure 2. Thisgrowth curve model represents an acceptable fitwith the data, based on an assortment of fit indi-ces: w2(df 6) = 7.66, p = .26; CFI = 1.00; RMSEA =0.04 (0.00 ± 0.10). An intercept value of 19.51 (CR= 41.01, p < .001) represents the average memoryfor grief symptom level at entry into the study atapproximately 2 months post-loss. The variancevalue of 35.83 for the intercept indicates that 95%of participants' scores ranged between 7.78 and31.24. A significant negative slope of ±0.39 (CR =±2.57, p < .01) indicates that participants' memoryfor grief symptoms decreased on average 0.39

300 FIELD, THOMPSON, GALLAGHER-THOMPSON

Dow

nloa

ded

by [

Har

vard

Lib

rary

] at

10:

49 0

6 O

ctob

er 2

014

points across each successive wave of data col-lection. It should be noted here that although thisaverage negative slope value is significantly dif-ferent from 0, noteworthy variation existsbetween individuals in their slope values, asreflected in the large slope variance value of 0.99,which indicates that 95% of participants' changescores range between ±2.34 and 1.56. In effect,although on average bereaved individuals show atendency to remember past grief as less severe as afunction of time, a significant proportion of themremember past grief as more severe as a functionof time.

Combined model of present and pastgrief symptoms

A bivariate latent growth curve model was esti-mated for present and remembered past griefsymptoms simultaneously. This provided a meansfor determining whether a systematic relationshipexists between change in current grief and changein remembered grief. It was necessary to lift the

constraints on T2 and T3 to obtain a permissiblesolution. A pictorial representation of this modelis shown in Figure 3. This growth curve modelrepresents an acceptable fit with the data, basedon an assortment of fit indices: w2(df 24) = 44.22, p= .06; CFI = 1.00; RMSEA = 0.05 (0.00 ± 0.08).

A significant correlation coefficient of 0.71 (p <.001) between the intercept for present grief andthe intercept for past grief indicates a strongpositive relationship between current grief symp-toms and memory for past grief at entry into thestudy at approximately 2 months post-loss. This isnot surprising, given the close temporal proximitybetween current grief symptoms and rememberedseverity of initial grief 2 months earlier. A sig-nificant correlation coefficient of 0.87 (p < .001)between the slope for present grief and the slopefor past grief indicates that a strong positiverelationship exists between the trajectory ofchange in current grief and change in memory forpast grief. Specifically, bereaved individuals whoindicate greater reduction in current grief symp-toms over time are also more likely to indicategreater decrease over time in their memory of past

Figure 1. Latent growth curve model for change in current grief symptoms over time. The rectangles represent the observed currentgrief symptom level scores at each of the successive data-collection time points, from wave 1 (T1) to wave 4 (T4). The intercept valuefor each of these observed scores was fixed at 0. The four residual variance circles, labelled E1 to E4, had their mean values fixed at 0.The path connecting each residual variance circle to the observed rectangle scores was fixed at 1. The intercept path coefficients arefixed at 1 while the slope path coefficients are fixed at 0, 1, 2, and 3 successively. The latent growth intercept values represent theestimated average current grief symptom level and corresponding variance respectively at entry into the study at T1, while the latentgrowth slope values represent the estimated average change in current grief symptom levels and corresponding variance respectivelybetween successive waves of data collection.

MEMORY FOR GRIEF 301

Dow

nloa

ded

by [

Har

vard

Lib

rary

] at

10:

49 0

6 O

ctob

er 2

014

grief relative to those who show less of a reductionin current grief over time. These results thereforesupport the position that the trajectory of changein present grief moderates memory of past grief.

A final analysis was conducted to determinethe directionality of the relationship betweenchanges in current grief and changes in recalledgrief. Under the assumption that level of currentgrief serves as an important context in the recall ofpast grief, changes in recalled grief can beexplained by changes in level of current grief.Specifically, the significant positive correlationbetween the current grief slope and recalled pastgrief slope identified in the previous analysis isunderstood in terms of the impact of the formeron the latter. It therefore follows that a directionalrelationship should be found between currentgrief and recalled past grief, such that changes incurrent grief cause changes in recall of past griefwhereas recalled past grief should not play asimilar causal role in current grief.

Estimating causal relationships between vari-ables in longitudinal designs requires the use ofstatistical methods capable of assessing non-

recursive models. Nonrecursive models are char-acterised by circular feedback loops throughwhich a variable can indirectly affect itself (Burns& Spangler, 2000). Structural equation modelling(SEM) statistical procedures provide a means fortesting the relative viability of different models ofcausal relationships between variables over time.

The AMOS 4 structural equation modellingapproach statistical software (Arbuckle &Wothke, 1999) was applied to the nonrecursivepanel design shown in Figure 4 to investigate thecausal relationship between current grief andmemory for past grief. This model was selectedbecause it spans the time period from entry intothe study to the final wave of data collection. Thisnonrecursive model represents an acceptable fitwith the data, based on an assortment of fit indi-ces: w2(df 1) = 0.24, p = .63; CFI = 1.00; RMSEA =0.00 (0.00 ± 0.14). Consistent with the predictionthat changes in current grief explain changes inmemory for past grief, a statistically significantpositive beta coefficient was found for changes incurrent grief in explaining changes in recalled pastgrief (b = 0.21, CR = 3.56, p < .001). In effect,

Figure 2. Latent growth curve model for change in current grief symptoms over time. The rectangles represent the observedremembered grief symptom level scores at each of the successive data-collection time points, from wave 1 (T1) to wave 4 (T4). Theintercept value for each of these observed scores was fixed at 0. The four residual variance circles, labelled E1 to E4, had their meanvalues fixed at 0. The path connecting each residual variance circle to the observed rectangle scores was fixed at 1. The latent growthintercept path coefficients are fixed at 1 while the latent growth slope path coefficients are fixed at 0, 1, 2, and 3 successively. Theintercept values represent the estimated average remembered past grief symptom level and corresponding variance respectivelyassessed at time of entry into the study (T1), while the slope values represent the estimated average change in remembered past griefsymptom levels and corresponding variance respectively between successive waves of data collection.

302 FIELD, THOMPSON, GALLAGHER-THOMPSON

Dow

nloa

ded

by [

Har

vard

Lib

rary

] at

10:

49 0

6 O

ctob

er 2

014

greater reduction in level of current grief fromtime 1 to time 4 is uniquely predictive of greaterdecrease in memory for past grief from time 1 totime 4. On the other hand, the statistically non-significant beta coefficient in the causal path fromchanges in remembered past grief to changes in

current grief (b = 0.09, CR = 0.59, ns) clarifies thatany effect of the former on the latter is completelymediated by the latter.

As a further check on the validity of thedirectionality of this causal model, an additionalanalysis was conducted in which the causal path

Figure 3. Bivariate latent growth curve model showing the correlation between the latent growth intercept values for current andremembered past grief (.71) and the correlation between the latent growth slope values for current and remembered past grief (.86).The rectangles represent the observed current and remembered past grief symptom level scores at each of the successive data-collection time points, from wave 1 (T1) to wave 4 (T4). The intercept value for each of these observed scores was fixed at 0. The eightresidual variance circles, labelled E1 to E8, had their mean values fixed at 0. The path connecting each residual variance circle to theobserved rectangle scores was fixed at 1. The latent growth intercept path coefficients are fixed at 1. The latent growth slope pathcoefficients are fixed at 0 at time of entry into the study (T1) and at the final wave of data collection (T4), while estimated at thesecond and third time points (T2 and T3 estimated values of 1.48 and 1.99 respectively). ***p < .001.

MEMORY FOR GRIEF 303

Dow

nloa

ded

by [

Har

vard

Lib

rary

] at

10:

49 0

6 O

ctob

er 2

014

from changes in remembered grief to changes incurrent grief was omitted from the model. In theresulting model, the beta coefficient of the causalpath from changes in current grief to changes inremembered grief was b = 0.35 (CR = 5.30, p <.001). The fit of this model was good and com-parable with its more complex Figure 4 non-recursive counterpart: w2(df 2) = 0.55, p = .76; CFI= 1.00; RMSEA = 0.00 (0.00 ± 0.09). On the other,when the causal path from changes in current griefto changes in remembered past grief was omittedfrom the Figure 4 model, the resulting model fitwas much poorer: w2(df 2) = 10.74, p = .005; CFI =0.99; RMSEA = 0.14 (0.07 ± 0.23). This furtherclarifies that changes in remembered past grief areexplained by changes in current grief, not viceversa.

DISCUSSION

Consistent with predictions, a systematic rela-tionship was found between changes in currentgrief and recall of past grief as a function of time.Bereaved widows and widowers on the wholerecalled their past grief as less severe as a function

of time, shown in a significant negative slopeacross the four waves of data collection. Thispattern mirrored the pattern for current grief,which also showed a progressive reduction overtime. More importantly, a strong positive rela-tionship found between the trajectory of currentgrief and the trajectory of remembered past griefsuggested that level of current grief was animportant moderator of past grief, such that thosewho underwent a greater reduction in grief overthe course of the study showed a similar greaterreduction in remembered past grief severity overtime relative to those whose grief underwent lesschange. Finally, the results of the nonrecursivecausal analysis confirmed the directionality of therelationship between current and rememberedpast grief in showing that the changes over time inremembered past grief could be explained bychanges in level of current grief.

The present study's findings on the positiverelationship between changes in current grief andchanges in memory for past grief are consistentwith the results of previous studies indicating thatthose who undergo less reduction in distress overtime are more likely to overestimate past distressrelative to those who show greater reduction (e.g.,

Figure 4. Model of the reciprocal causal effects between changes in current grief and changes in remembered past grief from time ofentry into the study (T1) to the final wave of data collection (T4). The values next to the one-headed arrows represent unstandardisedparameter estimates. ***p < .01.

304 FIELD, THOMPSON, GALLAGHER-THOMPSON

Dow

nloa

ded

by [

Har

vard

Lib

rary

] at

10:

49 0

6 O

ctob

er 2

014

Safer et al., 2001; Safer & Keuler, 2002). However,in contrast to prior studies showing a generaltendency to overestimate affective intensity inrecalling past distress (see Safer & Keuler, 2002),the negative slope for the recall of past grief sug-gests that participants in the present study as agroup indicated a progressive tendency to under-estimate past distress as a function of time. Boththe nature of the study design and participantsample might explain this discrepancy.

Previous studies typically employed a single-occasion paradigm wherein participants weregiven an incidental recall measure in which theywere asked to replicate how they had originallyresponded to the affect scale. In other words,participants received an explicit instruction to beaccurate, whereas in the present study contextthere was no such demand. Instead they weresimply asked to rate how distressed they were.When participants try to recall past ratings, asopposed to being asked to simply rate past dis-tress, they are likely to engage in more effortfulprocessing, given the task demand to be accurate.Differences in processing strategies are known tomoderate the impact of affective state on memoryand judgement (for more on different processingstrategies, see Forgas, 1995).

Another aspect of single-occasion studies isthat participants are asked to recall their pastratings, as opposed to being asked to recall theirpast distress. It is possible that their past ratings donot correspond to their past actual level of dis-tress, because they are being asked to make adifferent judgment than if they were to simply ratetheir level of past grief. In effect, the accuracy oftheir past ratings, as opposed to the accuracy oftheir actual level of distress, is being assessed here.In certain respects, one might argue that the wayof measuring remembered past distress in thepresent study is more in keeping with how indi-viduals typically recall past distress relative to thesingle-occasion paradigm. In this regard, the pre-sent study may possess better ecological validity.

Beyond the study design and method of asses-sing recall bias, the nature of the participantsample may explain differences in the resultsbetween the present study and previous researchexamining memory bias for past distressingevents. Many of the studies employing the single-occasion paradigm were psychotherapy studies.One might expect that such individuals would beparticularly motivated to confirm the benefits ofhaving undergone therapy and therefore mightexaggerate their initial level of distress at entry

into therapy. On the other hand, a sample ofbereaved participants who volunteered for a studyon bereavement, who were not in any way led tobelieve that they would derive personal benefitfrom their participant apart from contributing toresearch, are less likely to exhibit such biases. Infact, in the only other study assessing recall of pastdistress in bereavement, Safer et al. (2001) foundoverestimation effects in only one of three symp-tom measures.

The results of the nonrecursive model analysisprovided support for the causal role of changes incurrent distress in explaining changes in memoryfor past distress. Previous research suggests thatmemory for past distress may also influence levelof current distress (Karney & Coombs, 2000;Safer et al., 2001; Safer, Levine, & Drapalski,2002). In effect, recalling one's past distress asmore severe may undermine current copingefforts, thus inciting greater current distress(Safer et al., 2001). However, because past stu-dies employed ordinary least-squares statisticaltechniques that are not capable of assessing non-recursive models, their results are inconclusive asto the directionality of the causal relationshipbetween past and current distress. The presentstudy, having employed a more sophisticated sta-tistical technique capable of addressing direc-tionality of causal relationships betweenvariables, provided no such support for the bidir-ectionality of the relationship between memoryfor past and current grief.

A limitation of the present study is in nothaving obtained participants' ratings of their griefat the time of death. Because the first wave of datacollection did not take place until 2 months afterthe death, it was not possible to know the accuracyof participants' past grief scores, since it is possiblethat memory biases may be evident even forevents occurring only 2 months earlier. Thus,unlike single-occasion studies, the accuracy of pastgrief ratings could not be determined. However, itis virtually impossible to obtain information onlevel of distress immediately after a death. Itshould be noted that the alternative strategy ofhaving participants recall past ratings of griefobtained at the time of the wave 1 data collectionwould involve having them focus on somethingmuch less salient than having them answer withrespect to their initial grief following the death. Infact, that is a limitation of the Safer et al. (2001)study in which participants were asked to recalltheir ratings of grief at the 6-month post-losspoint.

MEMORY FOR GRIEF 305

Dow

nloa

ded

by [

Har

vard

Lib

rary

] at

10:

49 0

6 O

ctob

er 2

014

The present study results are in keeping withthose of other studies in showing that memory forhighly stressful events is not fixed and that factorssuch as current affective state play an importantrole in recall of such events. These results aretherefore supportive of the reconstructive per-spective on memory (Bartlett, 1932; Conway &Pleydell-Pearce, 2000). A direction for futureresearch might involve assessing the role of per-sonality in explaining the impact of current dis-tress on memory for past distress. There is somesupport for the impact of affective traits onmemory for past distress, mediated through theeffect of the former on current distress (Safer &Keuler, 2002). It would be valuable to extend thisprevious work using a multiwave design such asthat used in the present study. Similarly, becausesituational factors are known to influencebereavement-related distress (Macias, Jones, &Harvey, 2004), they may be expected to also affectmemory for past distress, mediated through cur-rent distress. Another research direction mightinvolve identifying the mechanism underlying theimpact of current distress on memory for pastdistressÐin particular, to what extent motiva-tional versus cognitive factors are involved.

Manuscript received 24 November 2004Manuscript accepted 28 June 2005

REFERENCES

Arbuckle, J. L., & Wothke, W. (1999). AMOS user'sguide. Chicago: Smallwaters.

Bartlett, F. C. (1932). Remembering: A study in experi-mental and social psychology. Cambridge: Cam-bridge University Press.

Bonanno, G. A., & Kaltman, S. (2001). The varieties ofgrief experience. Clinical Psychology Review, 21,705±734.

Bower, G. H. (1981). Mood and memory. AmericanPsychologist, 36, 129±148.

Bryant, R. A. (1993). Memory for pain and affect inchronic pain patients. Pain, 54, 347±351.

Burns, D. D., & Spangler, D. L. (2000). Does psy-chotherapy homework lead to improvements indepression in cognitive-behavioral therapy or doesimprovement lead to increased homework com-pliance? Journal of Consulting and Clinical Psy-chology, 68, 46±56.

Conway, M. A., & Pleydell-Pearce, C. W. (2000). Theconstruction of autobiographical memories in theself-memory system. Psychological Review, 107,261±288.

Conway, M., & Ross, M. (1984). Getting what you wantby revising what you had. Journal of Personality andSocial Psychology, 47, 738±748.

Faschingbauer, T. R. (1981). Texas revised inventory ofgrief manual. Houston, TX: Honeycomb Publishing.

Forgas, J. P. (1995). Mood and judgment: The AffectInfusion Model (AIM). Psychological Bulletin, 117,39±66.

Higgins, E. T., & King, G. (1981). Accessibility of socialconstructs: Information processing consequences ofindividual and contextual variability. In N. Cantor &J. Kihlstrom (Eds.), Personality, cognition, and socialinteraction. Hillsdale, NJ: Lawrence ErlbaumAssociates Inc.

Karney, B. R., & Coombs, R. H. (2000). Memory bias inlong-term close relationships: Consistency orimprovement? Personality and Social PsychologyBulletin, 26, 959±970.

Macias, C., Jones, D., & Harvey, J. (2004). Bereavementin the context of serious mental illness. PsychiatricServices, 55, 421±426.

McFarland, C., & Alvaro, C. (2000). The impact ofmotivation on temporal comparisons: Coping withtraumatic events by perceiving personal growth.Journal of Personality and Social Psychology, 79,327±343.

Safer, M. A., Bonanno, G. A., & Field, N. P. (2001a). ``Itwas never that bad'': Biased recall of grief and long-term adjustment to the death of a spouse. Memory, 9,195±204.

Safer, M. A., & Keuler, D. J. (2002). Individual differ-ences in misremembering pre-psychotherapy dis-tress: Personality and memory distortion. Emotion,2, 162±178.

Safer, M. A., Levine, L. J., & Drapalski, A. L. (2002).Distortion in memory for emotions: The contribu-tions of personality and post-event knowledge. Per-sonality and Social Psychology Bulletin, 28, 1495±1507.

Thompson, L. W., Gallagher-Thompson, D., Futter-man, A., Gilewski, M. J., & Peterson, J. A. (1991).The effects of late-life spousal bereavement over a30-month interval. Psychology and Aging, 6,434±441.

306 FIELD, THOMPSON, GALLAGHER-THOMPSON

Dow

nloa

ded

by [

Har

vard

Lib

rary

] at

10:

49 0

6 O

ctob

er 2

014