364 Bosma et al.
Am J Epidemiol Vol. 153, No. 4, 2001
age and zip code, the sample was randomly drawn from the
municipal population registries. The response rate was 70.1
percent, which resulted in a study population of 18,973
respondents. The response rates were not substantially dif-
ferent by age, sex, marital status, level of urbanization, or
social class (as indicated by zip code) (33).
In Eindhoven, the major city in the study, 8,506 GLOBE
participants who reported on their current or last occupation
(younger people, mainly students, were thus excluded)
could be attributed to 86 administrative neighborhoods
(areas) with relatively homogeneous housing. On average,
there were 99 respondents in a neighborhood (range,
5–386). Eindhoven has a total of 106 areas, but some have
only few or no inhabitants. In 1991, the 86 neighborhoods
contained 191,000 inhabitants (all ages), with an average of
2,221 inhabitants per neighborhood.
Municipal population registers provided information on
all-cause mortality during the follow-up period until mid-
1997. These have a virtually complete coverage of the pop-
ulation (34). During the 6-year follow-up period, 487 men
and women died (6 percent).
Individual and neighborhood socioeconomic status
Similar measures were used to indicate socioeconomic
status on the individual and neighborhood levels. To indi-
cate the socioeconomic status of the neighborhood, we
aggregated individual reports on socioeconomic status to
the neighborhood level. An advantage of this approach is
that the effect of neighborhood socioeconomic status on
mortality can be stringently controlled for equivalent mea-
sures on the individual level. This allows an accurate exam-
ination of genuine contextual effects and, to a large extent,
excludes the possibility that any adverse effect of poor
neighborhoods is fully based on poorer people living in
poorer neighborhoods.
Four indicators of individual socioeconomic status were
used: last attained educational level of respondent, current
or last occupational level of household breadwinner (35),
being disabled or unemployed, and presence of severe finan-
cial problems in the household. Being or not being disabled
or unemployed was based on a question asking for the sub-
jects’ main activity for which they received income or spe-
cial social security benefits. Disability implied long-term
disability. Severe financial problems were self-reported by a
single item about whether or not the household had many
problems with making ends meet.
Neighborhood socioeconomic status was indexed by four
equivalent indicators based upon aggregated individual
GLOBE data: the percent of subjects reporting primary
schooling only (range, 0–44 percent), the percent of subjects
reporting that they were unskilled manual workers (range,
0–39), the percent of subjects reporting that they were
unemployed or disabled (range, 0–28), and the percent of
subjects reporting severe financial problems (range, 0–15).
Both continuous percentage scores and quartiles of neigh-
borhoods based on the percentage in the aggregated data
were used; each quartile contained about 22 neighborhoods
(86 neighborhoods total).
Disability and unemployment were combined because
both groups share the characteristic of being without paid
work and receiving social security benefits. Moreover, they
were combined because it is likely that many disabled per-
sons would have been assigned to the unemployed group in
countries other than the Netherlands. In the Netherlands, the
relatively generous disability benefit scheme has been used
on a large scale by employers to give the unemployed a rea-
sonable level of income compensation. Our findings will not
be confounded by prevalent disease (as the main cause of
disability) because the effect of the neighborhood percent
reported as unemployed or disabled will be controlled for
individual disability or unemployment and indicators of
prevalent disease.
Correlates of neighborhood socioeconomic status
Twelve individual characteristics were explored for their
association with neighborhood socioeconomic status. These
could be classified into four groups: housing conditions, and
social, psychologic, and behavioral factors. The three hous-
ing conditions were based on self-reported cold or draft in
the house (yes, no), condensation (yes, no), or moisture and
damp in the house (yes, no). The three social characteristics
were based on self-reported perception of vandalism in
neighborhood (yes, no), social difficulties with family mem-
bers and neighbors (subjects in the most adverse quintile of
the sum of eight five-point items asking for difficulties in
contacts with eight separate groups of important others,
including neighbors vs. the others), and noise pollution from
neighbors (yes, no). The three psychologic factors were
based on self-reported low control (subjects in the most
adverse quintile of the sum of 11 five-point items asking for
an external locus of control vs. the others), passive coping
(subjects in the most adverse quintile of the sum of eight
five-point items asking for an inactive style of coping vs. the
others), and depressive coping (subjects in the most adverse
quintile of the sum of seven five-point items asking for a
depressive reaction pattern when faced with problems vs.
the others). The three behavioral factors were based on self-
reported current smoking (yes, no), physically inactivity in
leisure time (subjects reporting no time spent on sports, gar-
dening, walking, or cycling during leisure time vs. the oth-
ers), and excessive alcohol consumption (subjects drinking
more than six alcoholic beverages on 3 or more days a week
or more than four beverages on 5 or more days a week vs.
the others). More information on the measurement of these
characteristics can be found elsewhere (36, 37).
Data analysis
Each separate neighborhood socioeconomic indicator was
related to all-cause mortality. The effect of individual
socioeconomic status was also determined. The effect of
neighborhood socioeconomic status (e.g., percent of sub-
jects reporting to be unskilled manual workers) was esti-
mated as both unadjusted and adjusted for its individual-
level equivalent socioeconomic measure (e.g., occupational
level). Any residual confounding by socioeconomic status