Vulnerability Before, During, and After a Major Depressive Episode: A 3-Wave Population-Based Study

Background  Vulnerability as defined by high levels of neuroticism, low self-esteem, and poor coping skills characterizes individuals with a history of major depressive episodes (MDEs).

Objective  To separate postmorbid vulnerability into (1) trait effects (ie, the continuation of premorbid vulnerability); (2) state effects of subthreshold (residual) symptoms on personality or its perception; and/or (3) scar effects (ie, negative personality change that develops during an MDE and persists beyond MDE remission).

Methods  Data come from the Netherlands Mental Health Survey and Incidence Study, a prospective Dutch psychiatric population–based survey. We obtained psychiatric (Composite International Diagnostic Interview) and personality data on neuroticism, depressive coping style, self-esteem, and mastery from 4796 respondents at 3 time points (T1, T2, and T3), 12 and 24 months apart. Between- and within-subjects differences were tested with repeated-measures analysis of variance and adjusted for sex, age, and time.

Results  After T1, 409 respondents developed an MDE, of whom 334 were MDE-free at T3. In comparison with individuals without any lifetime MDE, the 262 subjects with a first MDE had higher premorbid T1 vulnerability scores on each personality measure (0.38-0.83 effect size units). During the MDE, vulnerability scores further increased (0.33-0.52 effect size units) but returned to premorbid levels after MDE remission. We found no scar effects among subgroups with severe or long-lasting MDEs. Subthreshold residual symptoms at T3 biased comparisons between T1 and T3 if the premorbid period of T1 to MDE onset was longer than the postmorbid period of MDE remission to T3, misleadingly suggesting scar effects. We obtained similar results in the 147 subjects with recurrent MDEs.

Conclusions  There was no evidence of a negative change from premorbid to postmorbid assessment for any of the personality traits. Postmorbid vulnerability reflected the continuation of premorbid vulnerability. Pre-post MDE comparisons are sensitive to prodromal and residual symptoms. Our findings suggest 2 independent simultaneous processes: (1) the ongoing expression of vulnerability as a personality deviance; and (2) synchrony of change between severity of depressive symptoms and personality deviance.

Unipolar major depression and personality traits have complex relationships.^1-8 Severalstudies have consistently documented that personality traits in the domainof neuroticism, low self-esteem, and poor mastery are risk factors for theonset of major depressive episodes (MDEs).^2,9-14 Inthis article, we use the concept of vulnerability to indicate the personalitycharacteristics associated with an increased risk of depression. Ample evidenceshows that during an MDE, the scores for neuroticism are increased and thosefor self-esteem and mastery are reduced, an effect usually denoted as thestate effect of depression.^2,15-18 Itis unclear whether the state effect is limited to the MDE or whether it persistsbeyond MDE remission (ie, postmorbidly). Many investigators use the conceptof scarring to indicate postmorbid vulnerability that was not present premorbidly(ie, before the MDE).^2,19-21 Wefollow this definition. The evidence of scarring is limited and contradictory,with some evidence in favor^2,19 andsome against it.^20,21 Postmorbidvulnerability, however, could also be due to the continuation of premorbidvulnerability. In that case, premorbid personality characteristics (as truetrait factors) would still be present after the MDE. We define this as thetrait effect. Finally, postmorbid vulnerability could reflect mild state effectscaused by subthreshold residual symptoms.

To separate postmorbid vulnerability into state, trait, and scar effects,a prospective multiwave population study was required to identify first andrecurrent MDEs with assessments of vulnerability before, during, and afterthe MDE. First MDEs were needed to obtain premorbid vulnerability scores forindividuals who were going to develop an MDE but had no history of depression.Recurrent MDEs were needed to examine whether state, scar, and trait effectsoccurred in recurrent depression. In addition, the timing of premorbid andpostmorbid vulnerability assessment was important. If the premorbid assessmenttook place long before MDE onset but the postmorbid assessment was performedshortly after MDE remission, the pre-post comparison could be biased by residualsymptoms and might wrongly suggest scarring. The opposite could occur if thepostmorbid period was longer than the premorbid period. Such bias might accountfor the inconsistent findings for scarring.

Using the 3-wave Netherlands Mental Health Survey and Incidence Study(NEMESIS)^22-25 conductedin the Dutch general population, we tested, for first and recurrent MDEs separately,3 hypotheses: (1) Vulnerability scores are higher during the MDE than before(state effect); (2) Vulnerability scores are higher after the MDE than before(scar effect); and (3) Premorbid vulnerability scores for those who developeda first MDE after the first wave (time point 1 [T1]) are higher than the vulnerabilityscores of those who never had an MDE (trait effect). Individuals with a historyof depression who developed an MDE after T1 (recurrent MDE) were contrastedwith individuals who also had a history of depression but did not developa recurrent MDE after T1. To examine the state effects of subthreshold residualsymptoms, we had to use an indirect approach because a measure of residualsymptoms at T3 was not available. The indirect approach used the time elapsedfrom remission of the MDE until T3, assuming that residual symptoms generallydecrease with time.

Scarring may not routinely occur.^26-28 Itmight occur only in people who experience a severe or long-lasting MDE sothat the negative effect on personality is sufficiently severe and prolongedto solidify and persist,^26,28 oronly in people with a history of depression^26,27 owingto the demoralizing effects of a recurrent MDE, or only in high-risk subjects^26,28 as a possible result of traumaticchildhood events or a family history of depression. Hence, we examined scarringin various subgroups.

We obtained the data presented in this article from NEMESIS. The design,sample, and instruments of NEMESIS have been extensively described elsewhere.^22-25 Briefly,NEMESIS was a prospective psychiatric epidemiologic survey in the Dutch adultgeneral population (aged 18-64 years) with assessments in 1996 (T1), 1997(T2), and 1999 (T3). It used a multistage, stratified, random-sampling procedure.From each household, we randomly selected 1 respondent. Interviewers madeup to 10 telephone calls or visits to a household at different times of dayand on various days of the week. For each subject, the interviews were 12months (T1-T2) and 24 months (T2-T3) apart and were administered in the samemonth of the year to prevent season-related bias. The fieldwork of each assessmentwave extended from February to December. Each interview included the CompositeInternational Diagnostic Interview (CIDI)^29,30 andself-report personality measures.

In the first wave, we obtained data from 7076 persons, a response rateof 69.7%. At T2, 1458 respondents (20.6%) dropped out; at T3, a further 822(14.6%) were lost. We obtained General Health Questionnaire data from 44%of the T1 nonresponders.²⁵ The T1 nonrespondershad a slightly lower mean score on the General Health Questionnaire (1.16vs 1.22, suggesting better mental health), a lower mean age (40.2 vs 41.2years), and a higher proportion of women (54.4 vs 53.3). Their psychiatricmorbidity (estimated with a logistic regression model) did not differ fromthat of the respondents.^22-25 Attritionat T2 and T3 was weakly associated with younger age, lower level of education,urbanization, living alone, being unemployed, and being born outside the Netherlands.²⁵ After adjustment for these demographic variables,mental health status as defined by absence vs presence of any disorder accordingto the DSM-III-R³¹ wasnot associated with attrition, but major depression at T2 weakly predictedattrition at T3 (odds ratio = 1.37; 95% confidence interval, 1.05-1.54).Our study used the 4796 respondents who participated in all 3 waves.

At T1, 1-month, 12-month, and lifetime prevalence rates of major depressionfor women were 3.4%, 7.5%, and 20.1%, respectively; for men they were 1.9%,4.1%, and 10.9%, respectively.²³

We used the World Health Organization–authorized Dutch versionof the CIDI, version 1.1,^29,30 andits diagnostic computer program to assign diagnoses according to the DSM-III-R. The CIDI has acceptable interrater and test-retestreliability for most nonpsychotic diagnoses, including major depression.^32,33 At T2 and T3, the CIDI lifetime frameworkwas adapted to the 1-year interval from T1 to T2 and the 2-year interval fromT2 to T3. We assigned DSM-III-R diagnoses withoutthe use of hierarchical exclusion rules or application of the functional impairmentcriterion. Persons with bipolar disorder were not included in the MDE group.

The lifetime prevalence of MDEs according to the T1 CIDI indicated whetheror not there was a history of MDEs. The occurrence of an MDE during the 1-yearT1-to-T2 interval was determined by the 12-month prevalence of MDEs at theT2 CIDI and during the 2-year T2-to-T3 interval by the 24-month prevalenceof MDEs at the T3 CIDI. The presence of an MDE at T1 was determined by the1-month prevalence of MDEs at the T1-CIDI. We excluded subjects with a 1-monthprevalence of MDEs at the T1 CIDI from all analyses (see Figure). The presence of an MDE at T1 and T3 was determined by the1-month prevalence of MDEs at the T2 and T3 CIDI, respectively. On the basisof lifetime history, we classified each MDE occurring after T1 as a firstepisode (first MDE) or recurrence (recurrent MDE). An MDE remission was determinedby a negative 1-month prevalence of MDEs at the T2 or T3 CIDI, according toin which interval the MDE occurred.

We constructed several groups from the CIDI data obtained at T1, T2,and T3 (Figure). The “never MDE”group (A in Figure) refers to respondentswho never had an MDE (T1 CIDI) and did not develop one during the T1-to-T3interval according to the T2 and T3 CIDI. The “history of MDE but noMDE after T1” group (B) refers to respondents who had at least 1 MDEbefore T1 (T1 CIDI) but did not develop a recurrence during the T1-to-T3 interval(T2 and T3 CIDI). The “first MDE” group (C) developed a firstMDE after T1 (T2 and/or T3 CIDI). The “recurrence of MDE in those withhistory of MDE” group (D) had a history of depression (T1 CIDI) anda recurrence after T1 (T2 and/or T3 CIDI). The remitted first MDE (E) andremitted recurrent MDE (F) groups consisted of those from groups C and D whodid not meet MDE criteria at T3 (no 1-month MDE prevalence at the T3 CIDI).

Furthermore, we identified the G1 and G2 groups. The G1 group (n = 46)consisted of subjects from the first MDE group who still met MDE criteriaat T3 (1-month MDE prevalence at the T3 CIDI). Group G2 (n = 39)consisted of subjects from the remitted first MDE group who met MDE criteriaat T2 (1-month MDE prevalence at the T2-CIDI). Groups H1 (n = 29)and H2 (n = 25) were the corresponding subjects from the recurrentMDE group.

With regard to time from T1 to MDE onset (the premorbid period), weclassified all MDEs as having a premorbid period of less than 1 year vs morethan 1 year, depending on whether MDE onset occurred before or after T2. Withregard to time from MDE remission to T3 (the postmorbid period), we assignedall individuals with MDEs and a negative 12-month prevalence according tothe T3 CIDI to the category of more than 1 year. Combining the binary premorbidand postmorbid classifications yielded 4 subgroups for remitted first MDE(E1-E4) and 4 for remitted recurrent MDE (F1-F4) (Figure). The mean premorbid and postmorbid periods in the E1, E2,F1, and F2 subgroups were of approximately equal length, whereas the meanpremorbid period in E3 and F3 was longer than the postmorbid period and shorterin E4 and F4.

We categorized MDEs in terms of severity, designated as mild, moderate,or severe according to the CIDI and DSM-III-R criteria(thus, mild MDEs still met DSM-III-R criteria forMDE). We calculated duration using data from the Life Chart Interview (LCI),³⁴ administered at T3. Complete LCI data were availablefor just 54% of all MDEs that occurred in the T1-to-T3 interval, largely becausethe LCI was administered only to those who responded affirmatively at T3 tothe LCI probe question of whether they had felt depressed for more than 2weeks after T1.³⁵ The LCI rated the presenceof MDE symptoms per 3-month period. This measure defined remission as no oronly subthreshold depressive symptoms in a 3-month period. We calculated durationby summing the 3-month periods with MDE symptoms until the 3-month periodoccurred with no or only subthreshold symptoms. We used both a continuousmeasure of duration and a dichotomized version (split at the mean durationof 6 months).

We administered 4 brief personality questionnaires in the same orderat each of the 3 assessment waves. We assessed neuroticism with the shortform of the Neuroticism scale from the Amsterdam Biographic Inventory.^11,36,37 The Amsterdam BiographicInventory, based on the Maudsley Personality Inventory,³⁸ wasone of the 2 most frequently used personality questionnaires in the Netherlandsduring the last decades. A high score indicates high levels of neuroticismor negative affectivity. Protypical items include the following questions:Does it happen often that you make a decision too late? Are you often so burdenedby disappointments that you cannot put them out of your head? Do you everhave nightmares? Do you ever feel terribly unhappy, without knowing why? Theinternal consistency of the 14-item Neuroticism scale was satisfactory (Cronbach α = .79-.83).We assessed depressive coping style with the Utrecht Coping Scale.^39,40 A high score indicates a depressivecoping style in response to stressors; for example, “Problems overwhelmme” or “Problems tend to make me pessimistic” (Cronbach α = .64-.67).We assessed self-esteem with the 10-item Rosenberg Self-esteem Scale.⁴¹ A high score indicates high self-esteem (Cronbach α = .84-.86).We assessed mastery with the 5-item Mastery Scale from Pearlin and Schooler⁴² (Cronbach α = .80-.81). A high scoreindicates an internal locus of control (the perception of being in controlof one’s life); a low score indicates an external locus of control withfeelings of helplessness (eg, “I have no control over the things thathappen to me”). High scores for neuroticism and depressive coping styleand low scores for self-esteem and mastery indicate vulnerability. Althoughthe time frame of the personality questionnaires was not specified, instructionsand wording with frequent terms such as in general and usually make it likely that respondents used a time frameof at least several months.

Major negative childhood experiences and lifetime parental depressionwere assessed at T1 in NEMESIS with 2 brief questionnaires.^22,35 Weconstructed 3 binary variables that categorized childhood experiences beforeage 16 years into “never” vs “more than once” forneglect, emotional and/or physical abuse, and sexual abuse. To establish lifetimeparental depression, NEMESIS asked respondents whether 1 or both biologicalparents had ever had an MDE.

Within-subjects differences (state effects and scar effects) were testedusing repeated-measures analysis of variance (ANOVA) adjusting for sex andage. We also adjusted for time effects across the interval from T1 to T2 toT3 by using the never MDE group (A in Figure)as a control group for first MDEs and the history of MDE group (B) as a controlgroup for recurrent MDEs (ie, within-subjects effects were tested by the group× time interaction using repeated-measures ANOVA). We tested between-subjectsdifferences (trait effects) with ANOVA adjusted for sex and age.

Because measures of subthreshold prodromal T1 and residual T3 symptomswere not available, we used length of the premorbid (less vs more than 1 yearfrom T1 to MDE onset) and postmorbid periods (less vs more than 1 year fromMDE remission to T3) to examine the effects of prodromal and residual symptomson pre-post MDE differences in personality scores. This indirect approachassumes that in general, prodromal symptoms develop gradually and residualsymptoms remit gradually.

We express differences in personality scores using effect sizes. Effectsize is defined as the Cohen d (ie, the mean difference in personality score)÷ the pooled standard deviation (the square root of the sum of the2 variances ÷ 2). An effect size of less than 0.20 is generally considerednegligible, from 0.20 to 0.39 small, from 0.40 to 0.70 moderate, and greaterthan 0.70 large. The effect sizes and confidence intervals are unadjustedfor sex, age, and time effects and are based on t tests.The repeated-measures ANOVA findings (F statistic) yielded similar or evensmallerP values than the unadjusted t tests, so interpretation of the effect sizes remains straightforward.

Of the 262 first MDEs, 36% were mild, 31% were moderate, and 32% weresevere. For the 147 recurrent MDEs, these proportions were 25%, 35%, and 41%.Mean ± SD episode duration for remitted first MDEs (n = 216)was 6.1± 5.7 months and for remitted recurrent MDEs (n = 118)was 4.9 ± 3.7 months. Of those with a first MDE, 64% werewomen and the mean ± SD age was 40.0 ± 11.4years. For recurrent MDEs, these rates were 75% and 37.0 ± 10.1years. In the “never MDE” group (49.7% women; mean±SD age,41.0±12.2 years), 18.2% experienced emotional neglect, 8.5% psychologicalor physical abuse, and 2.6% sexual abuse; 19.2% reported that at least 1 biologicalparent had experienced a depressive episode. These rates were approximatelytwice as high in the first MDE group and about 3 to 4 times higher in therecurrent MDE group.

We evaluated state effects by comparing the premorbid T1 personalityscores of subjects in the first MDE group who still had the MDE at T2 or T3(groups G1 + G2 in Figure) with theirscores at T2 or T3 (hence, during the MDE). We did the same for those in therecurrent MDE group (H1+H2). Table 1 showsthat vulnerability scores were higher during the episode than before for bothfirst and recurrent MDEs. The F statistic indicates that all differences werestatistically significant. The effect sizes ranged from 0.33 to 0.54, suggestingsmall to moderate state effects.

We evaluated trait effects by contrasting the first MDE group (C in Figure) with the never MDE group (A) as wellas the recurrent MDE group (D) with the history of MDE group (B) regardingtheir T1 personality scores. Table 2 presentsthe results. Compared with the T1 personality scores of their respective contrastgroups, those with first and recurrent MDEs had consistently higher T1 vulnerabilityscores. All differences were statistically significant. Trait effect sizeswere moderate to strong for first MDEs but weaker for recurrent MDEs. Thedifferences in T1 vulnerability scores between subjects in the recurrent MDEand never MDE groups were higher than between those in the first MDE and neverMDE groups.

We repeated this comparison for the subgroups of subjects with firstand recurrent MDEs that had started after T2 to reduce confounding by a possiblestate effect of prodromal symptoms at T1 in those with an MDE onset betweenT1 and T2. Such a state effect would inflate the difference when comparedwith the control groups. The difference in T1 personality scores remainedsignificant but dropped about 10% to 20%.

We evaluated scar effects for subjects with remitted first (E in Figure) and recurrent (F) MDEs separately bycomparing premorbid T1 with postmorbid T3 personality scores. Higher postmorbidvulnerability scores would suggest scar effects. The findings were negative:T1 and T3 scores did not differ for subjects with first (Table 3, row E) or recurrent (row F) MDEs.

Although these results suggest that scarring does not routinely occurin first and recurrent MDEs, it might occur in particular subcategories. Weexamined, for first and recurrent MDEs separately, scarring in the subcategoriesof severe MDEs, MDEs lasting more than 6 months, and severe MDEs lasting morethan 6 months. We also examined scarring in the subcategories of individualswho experienced potentially traumatic childhood events or had a biologicalparent with a history of depression. With 1 exception, the results of allsubgroup tests were negative (data not presented). The exception was depressivecoping style in the subcategory of recurrent MDEs lasting more than 6 months(F₁ = 6.95; P = .02).Given the number of subgroup tests, we consider this a chance finding.

Table 3 also presents T1-to-T3effect size differences in personality scores for subgroups of those withfirst remitted MDEs with equal (E1 and E2) and different (E3 and E4) premorbidand postmorbid periods. In E1 and E2, pre-post episode vulnerability scoresdid not differ, but they did in E3 and E4. In subgroup E3, in which the premorbidperiod was longer than the postmorbid period, we found higher levels of neuroticismand depressive coping style (negative effect size) and lower self-esteem andmastery (positive effect size) at T3 than at T1. Given the longer premorbidthan postmorbid period, this is probably due to fewer prodromal symptoms atT1 than residual symptoms at T3. The reverse occurred in subgroup E4, in whichthe premorbid period was shorter than the postmorbid period. Subgroups F1to F4 of subjects with recurrent MDEs yielded similar results to the E1 toE4 subgroups of those with first MDEs, although the vulnerability scores inF4 showed only a trend toward lower vulnerability at T3 as compared with T1(data not presented). Collectively, these results suggest that pre-post MDEcomparisons in which the premorbid and postmorbid period differ in lengthmay misleadingly suggest scarring or its opposite.

Whereas personality-related vulnerability (high levels of neuroticismand depressive coping style with low self-esteem and mastery) was presentmore than 1 year before and further increased during the MDE, it returnedto its premorbid level after remission of the MDE. Our results confirm earlierfindings of a state effect^2,15-18 ofdepression on personality traits in the domain of neuroticism and controlas well as a risk factor effect^2,9-14 ofthese traits on depression. Our results, however, do not support the scarhypothesis. The findings are consistent with results reported from the NationalInstitute of Mental Health Collaborative Program on the Psychobiology of Depression²¹ and extend these separately to larger and more representativesamples of first and recurrent MDEs. Our findings point to 2 independent butsimultaneous processes: (1) the ongoing expression of vulnerability for depressionin increased neuroticism and depressive coping style as well as reduced self-esteemand mastery, collectively denoted as personality deviance; and (2) synchronyof change between severity of depressive symptoms and severity of personalitydeviance.

Pre-post MDE personality comparisons may misleadingly suggest scarringif the premorbid period is longer than the postmorbid period (and the oppositeif shorter). We attribute this to a state effect of subthreshold residualsymptoms on self-reported personality. If the premorbid period is longer thanthe postmorbid period, residual symptoms will likely outnumber prodromal symptoms.State effects are not limited to the MDE itself but extend to the prodromaland residual symptom phases. The few studies reporting scar effects^2,19 may have had longer premorbid thanpostmorbid periods on average, favoring an excess of residual symptoms atthe postmorbid personality assessment relative to the level of prodromal symptomsat the premorbid assessment. Nothing is known about possible time lags inthe synchrony of change between personality and symptom severity.

Findings and interpretations should be considered in the context ofseveral limitations. The most serious limitation is the lack of a measureof current symptoms at the times the respondents took the CIDI and personalitymeasures (T1, T2, and T3) because the CIDI version 1.1 has no cross-sectionalsymptom assessment. Hence, we cannot rule out that premorbid and postmorbidpersonality deviance might be entirely due to (chronic) subthreshold symptoms.However, we do not think that this is the case for 2 reasons, and our datasupport the notion of premorbid and postmorbid personality deviance as anexpression of trait vulnerability to depression. First, personality deviancewas already present more than 1 year before the first lifetime MDE. Second,if chronic subthreshold symptoms are present years before and after the MDEand they produce personality deviance or are produced by it, it is more appropriateto consider these symptoms as an integral part of the personality and an expressionof underlying trait vulnerability.

The second limitation is that we could not discriminate between 2 competingexplanations of state effects: real albeit temporary personality changes vsreporting bias due to the effects of depression on self-perception and recallprocesses.^43-45 Otherlimitations include the retrospective assessment at T3 of the duration ofonly 53% of MDEs occurring in the T1-to-T3 interval, the lack of reliabilitydata for the measures of childhood adversity and parental depression, andthe slightly selective attrition. We do not believe that attrition threatenedthe validity of our findings because of the large sample size and the factthat approximately 80% of all individuals with an MDE participated again atthe next assessment wave, including those with first and recurrent as wellas remitted and nonremitted MDEs. Hence, attrition probably did not reducethe variation in course. Finally, because detailed treatment data are lacking,we cannot exclude the possibility that at the group level, effective treatmentmay have reduced the personality deviance of some so much that this neutralizedthe scar effects of nontreatment in others.

Although scarring does not routinely occur after an MDE, it may occurin MDEs that require inpatient treatment. These patients are different fromthe average subject with MDE in a community study. In the NEMESIS study, 2%received inpatient psychiatric care, 22% specialty outpatient mental healthcare, and 23% only primary care in the year following onset. We also cannotrule out that scarring occurs in childhood or adolescent depression or thatforms of scarring exist that self-report personality measures do not pickup.¹⁹ Challenge tests to detect latent scarswould have strengthened the study, but given the large sample size, the costswould have been exorbitant.

The major strengths of our study are its sample size, random populationsample, structured diagnostic assessment procedures, and prospective longitudinaldesign made possible by assessment in 3 waves during a 3-year period. Thisway we could identify 334 individuals with an MDE that started after the firstpersonality assessment and remitted before the last one, avoid referral filterand lead time bias, and assess personality prior to, after, and during theMDE.

The occurrence of MDEs is common, with a 1-year incidence in the Netherlandsof 1.72 for men and 3.90 for women per 100 people at risk.⁴⁶ About50% of these episodes do not remit within a few months.³⁵ Thus,the finding that there is no evidence of negative personality change frompremorbid to postmorbid assessment is good news. The bad news is the continuationof premorbid personality-related vulnerability after the MDE. If treatmentswould not only achieve remission of the MDE but would also normalize levelsof neuroticism, coping style, mastery, and self-esteem, relapse and recurrencerates would probably fall.

Correspondence: Johan Ormel, PhD, Departmentof Psychiatry, University of Groningen, PO Box 30.001, 9700 RB Groningen,the Netherlands (j.ormel@med.rug.nl).

Submitted for Publication: December 11, 2003;final revision received April 1, 2004; accepted April 20, 2004.

Funding/Support: This study was supported bythe Netherlands Ministry of Health, Welfare and Sport and the Medical SciencesDepartment of the Netherlands Organization for Scientific Research (The Hague),and the National Institute for Public Health and Environment (Bilthoven, theNetherlands).

Additional Information: The analyses and writingof this article took place while Dr Ormel was a fellow-in-residence at theNetherlands Institute of Advanced Study in the Humanities and Social Sciences(Wassenaar).

Acknowledgment: We thank the anonymous reviewersfor their critical and helpful comments on the first draft.