Our main findings are that personal BC exposure for inner-city dwellers to a certain extent followed the daily patterns at fixed monitoring stations, that personal exposure on average corresponded well both to urban background measurements and to levels modelled at place of residency, but that monitoring data or time-resolved dispersion modelling data only explained the variability in 24 h exposure to less than 35% a working parent and even less for a parent on parental leave. Neither weather data nor time-activity data made any important improvements. For the weekly average BC exposure, the situation was somewhat better when using modelled BC data, reaching up to 45% explanation for a working parent. When using the personal weekly average to estimate long-term exposure, this was explained at 33% by dispersion modelling data.

Most other studies of personal BC were focused on the role of different microenvironments for individual exposure and inhaled dose, as a basis for policy [5, 6]. We found similar average diurnal patterns with low night-time levels, and for working parents distinct peaks corresponding to commuting, but somewhat less distinct diurnal patterns for the parents on leave. Interestingly the parent groups showed nearly identical patterns on Saturdays, perhaps reflecting joint activities, but not on Sundays, possibly reflecting an exchange of duties between the parents. However, we had no additional information about specific activities from the time-activity diaries and hence we can only speculate on this matter.

It can be noted that the personal BC exposure levels in our study were lower than in most other reports, even in comparison with data from e.g. Birmingham [19], Brisbane [20] or Paris [21], reflecting a comparatively well controlled outdoor environment, and the absence of major BC indoor sources for our study population.

In the perspective of estimating population exposure, it can be noted that in our study the BC personal exposure levels were on weekly average only about 30% higher than the corresponding UB, indicating that UB might be used for estimating average exposure to ambient BC for inner-city dwellers. This is in contrast with other studies that concluded that ambient monitoring did not provide adequate estimates of average population exposure [21, 22]. The reason for this discrepancy might be that our study persons lived close to the UB station, and also reported no smoking or use of open fires. The alternative in this study, using time- and space-resolved dispersion modelling produced even closer estimates, with an average about 10% overestimation of the personal exposure over a one-week period, and might also be better suited for assessing average population exposure in larger areas.

In the perspective of a time-series study using 24 h urban background monitoring data, our results are less promising, as only up to 25% of the temporal variability in personal exposure could be explained by the variability at urban and rural background stations, and that including season or weather data only provided marginal improvements, in contrast to other studies [23]. Data from the street-site monitor did not seem to contribute. Time-series studies of health effects from BC in Stockholm or similar cities are thus to be expected to suffer from substantial bias towards the null.

In total, 24 h dispersion modelling estimates for the outdoor levels at the home address gave similar results, explaining up to 26% of the variability in personal exposure, less than in other similar studies [23, 24]. One reason for this might be the lower general levels of BC in Stockholm. For the working parents up to 34% could be explained by adding dispersion modelling estimates for the work address, and air pressure, indicating that for a working population some exposure assessment precision might be gained using dispersion modelling for both home and work locations. This might however not be that relevant for studies of mortality and other health effects that are less common within working populations.

Information on time-activity patterns from the self-administered diary did not help to explain the variability in 24 h personal exposure levels, in contrast to other reports [5, 6, 21,22,23, 25, 26], with the only exception of time spent at work, probably because most workplaces in the region have mechanical ventilation systems with filters decreasing indoor levels of BC, while most inner-city homes have no treatment of inlet air. The indoor/outdoor BC ratio (I/O) for the workplaces was not measured in this study, but the in the central hours of the workday low levels were recorded. We have previously reported that the I/O for the homes in this study population was on average 79% [8].

While in the long-term perspective, differences in health effects between pollutants with similar spatial spread are difficult to discern, short-term studies of acute effects may shed some light, as different pollutants may show different temporal patterns But when the relation between the temporal air pollution metric used—often the level in urban background—and personal exposure, differs between pollutants, also the degree of bias will differ, invalidating a direct comparison between pollutants [27]. Our finding of a low temporal correlation between UB and personal exposure may thus indicate a relative handicap for BC in short-term studies of multiple pollutants.

Our study has the merit of personal BC measurements being performed in a low-level city environment and using not only monitoring station data but also time-resolved spatial estimates both at home and work addresses. One limitation is that the study area was quite small why our results may not be readily applied to large city areas. The summer period was not well covered, why we had limited power to address seasonal differences. Our self-administered diary data were on purpose quite simple in order to mimic a possible large-scale study. Further, different instruments (AE33 versus AE51) used at fixed monitoring sites and for personal exposure could partly contribute to decrease explained variance of the personal exposure. But as shown by e g Alas et al., the AE33 and AE51 are highly correlated if the filter loading effect is under control [28]. Another noteworthy limitation of our analyses based on variable selection approach is that it ignores repeated sampling within individuals, which renders the estimated standard errors and confidence intervals.

Although the majority of people spend most of their time indoors, the focus of both research and policy on air pollution has been on outdoor levels. While this in practice might be the only possibility, it must be acknowledged that the imperfect relation between outdoor levels and personal exposure is an important source of bias. Not only does it tend decrease statistical power in studies of health effects in general, it may also distort the view of the relative importance of different pollutants. We thus believe that quantifying this bias is of great importance both for science and for policy.

In conclusion, one-week continuous personal measurements performed by inner city adult Stockholmers showed exposure levels in the order of 400 ng black carbon/m3, with distinct diurnal and weekly patterns. Average exposure levels were similar to data from routine urban background monitoring or dispersion model estimates, indicating that long-term population exposure and related health effects may be estimated based on such data. The variability of 24 h exposure levels could however only be explained at about 35%, using routine monitoring or dispersion modelling. Any short-term health effects studies using such exposure data are likely to lack power and to be subject to bias.