The following acute symptoms were assessed three times during the 24 hours measurement using a paper-based diary: headache, irritation, nervousness, dizziness, fatigue and concentration problems.

The diary items were taken from the "Zerssen complaint list" 23 and assessed on a four point Likert scale (heavy, moderate, barely, not at all). The symptoms were considered present if they were reported with an at least barely intensity.

Chronic symptoms during the last six months were assessed during the CAPI. The symptoms were taken from the questionnaire of the HBSC-survey (Health behaviour in School-Aged Children) and consisted of the following ones: headache, irritation, nervousness, dizziness, fatigue, fear and sleeping problems 24.

The list was assessed on a five point Likert scale (nearly daily, several times a week, nearly every week, about once a month, seldom or never) during the CAPI. All symptoms were considered present if they occurred at least nearly every week according to self-report.

The 25 questions concerning mental health behaviours were asked during the CAPI following the German version of the Strength and Difficulties Questionnaire (SDQ). The questionnaire contained 5 scales: emotional symptoms, conduct problems, hyperactivity/inattention, peer relationship problems and prosocial behaviour. Each scale comprised 5 items with three possible forms ("certainly true", "somewhat true", "not true"). The summary score for each scale could range from 0–10 points. The difficulties score was generated by the summation of all four scales except the prosocial scale. The overall score could range from 0–40 points. For each scale as well as for the total difficulties score the participants were classified as "normal", "borderline" or "abnormal". 252627

Due to the small numbers of children and adolescents that could be classified as "borderline", we used two groups for the analysis ("normal" and "borderline/abnormal") and those participants who were in the borderline group were classified as "abnormal".

Sociodemographic data, the four study towns and environmental worries taken from the CAPI were considered as potential confounder.

The sociodemographic data included age, sex and level of education as confounder 28. All variables were dichotomised:

▪ Age group:

8–10 and 11–12 years (children) respectively 13–15 and 16–17 years (adolescents)

▪ Level of education:

at least 12 years of education (parents of the children) respectively at least grammar school ("Gymnasium") (adolescents)

The four study towns Munich, Augsburg, Rosenheim and Landsberg were also considered as potential confounding variables.

Environmental worries were assessed during the CAPI for the parents of the children and for the adolescents themselves with a short form of the environmental worry scale 2930. This scale consists of 12 questions about general and specific environmental worries (e.g. noise exposure or general environmental pollution) and was assessed on a four-point Likert scale (fully agree, partly agree, partly disagree, fully disagree). We a priori decided to dichotomise the variable using the median as cut off. This was done because there is no standard strategy to evaluate the environmental worry scale.

Besides the objective assessment of exposure using personal dosimetry participants were asked to estimate their exposure to mobile telecommunication networks. Participants estimated their exposure during the measurement day using a diary ("How many minutes did you use a mobile phone or DECT phone within the last 8 hours?"). In the baseline interview participants were also asked about the distance to the next mobile phone base stations in the living environment ("How many meters is the next mobile phone base station away from your home?"). The questions were taken from the questionnaire of the annual survey of the Institute for applied sciences (infas) and provided different reply categories 31.

These exposure estimates have been used to compare our results to previous studies on possible associations between self-reported exposure to mobile phone frequencies and health 416173233.

Exposure was measured using the personal dosimeter ESM-140 (Maschek Electronics). For 24 hours the dosimeter was placed on the upper arm of the participants opposite to the side which they usually used to hold the mobile phone or DECT phone during phone calls. During the measurement, subjects were asked to perform their routine daily tasks. At night, the participants placed the dosimeter next to their beds. After handing back the dosimeter the local assistants immediately read out the profiles and checked them for completeness and plausibility.

Exposure was assessed every second resulting in 86.400 measurements over 24 hours. The following frequency ranges were covered 17:

▪ GSM 900 (up and down link [up link: mobile phone; down link: mobile phone base station]; frequency range: 890–960 MHz)

▪ GSM 1800 (up and down link; frequency range: 1710–1880 MHz) including UMTS 2100 (frequency range: 1920–2170 MHz) and DECT (frequency range: 1880–1900 MHz)

▪ WLAN 2400 frequencies (frequency range: 2400–2480 MHz)

For the exposure assessment, a combination of the frequency bands had to be used because the dosimeter has a low selectivity between the up- and down-link channels.

Four dosimeters were selected at random and tested at the Technical Inspection Agency South (TÜV Süd) to confirm the technical data given by the manufacturer and to investigate if the different dosimeters used in the study are comparable. The findings showed a good comparability of the different dosimeters. At a given exposure level of 0.5 V/m the relative standard deviation was between 4% (DECT) and 17% (GSM 1800). Furthermore the findings proved the validity of the technical data provided by the manufacturer. According to the results of these laboratory measurements, the limit of determination was set at 0.05 V/m. To classify the exposure all measured values that were below this limit were replaced previous to the analysis by half of the limit (0.025 V/m). This method is often used in context of environmental epidemiology and the results seem plausible, because all values have to be between 0 and the limit of determination 34.

As a summary of the measured exposure we calculated in a first step the squared field strength, averaged over the waking hours, for the three frequency ranges. This way e.g. the squared field strength Ē² for GSM 900 has been calculated via the following formula:

E ¯ GSM  900 2 = ( ( Σ waking hours E GSM  900  uplink 2 ( t ) ) / duration of waking hours ) + = ( ( Σ waking hours E GSM  900  downlink 2 ( t ) ) / duration of waking hours ) MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=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@D617@

The overall simultaneous exposure to multiple frequency fields was calculated by summation on the squared field strengths, averaged over waking hours, which are weighted by the inverse of squared ICNIRP reference level 35. Extracting the square-root over this sum returned an overall exposure in terms of field strengths percentage of the reference level 21:

Overall exposure_% = square-root ((Ē²_{GSM 900}/limit²_{900 MHz}) + (Ē²_{GSM 1800}/limit²_{1800 MHz}) + (Ē²_WLAN/limit²_{2400 MHz})) * 100

During night the participants were asked to fix the dosimeter next to their bed on a bottle filled with water as a replacement for the participants' arm. The question came up if the measured values at this fixed position could represent the real exposure.

Therefore, the Technical Inspection Agency South (TÜV Süd) also investigated if the stationary measured bedtime exposure was a valid proxy of night time exposure. It was confirmed in the lab measurements that the measurement results of the dosimeter measurements depend on the direction in the field. So valid measurements of the average exposure can only be obtained if the dosimeter is moved. Due to these results bedtime exposure levels were not considered to be a valid proxy of night time exposure. Thus these levels had to be excluded and only exposure levels during individual waking hours were used.

To verify the representativeness of the 24 hours measurements 54 participants carried the dosimeter for five consecutive days from Monday afternoon to Saturday afternoon.

For the analysis the time between 4.p.m of one day until 4 p.m. of the following day (night time values were excluded) was summed up and exposure during waking hours was used. Exposure was classified into quartiles for each study day. The overall agreement of these quartiles between the weekdays was assessed. As a random misclassification of exposure on neighbouring cells always results in an underestimation of the effect 36, the deviation by at maximum 1 category was also considered.

To analyse a possible bias caused by selective non-participation, we compared the data of the short questionnaire of those who participated in the field study to those who did not. Parents of participating children were more likely mobile phone owners (93% of the participants, 90% of the non-participants; p_Chi2 0.02). Parents who had a higher level of education (at least 12 years of education) and those who were concerned about mobile phone exposure were more likely to take part in the study. Furthermore participating parents more often knew the distance to the next base station (don't know: 3%) than those who did not want to participate (don't know: 22%).

Participating adolescents were also more likely to have a higher level of education (at least grammar school) and to be more frequent mobile phone owners (91%) than non-participants (87%). Regarding the self estimated distance from home to the next mobile phone base station, non-participating adolescents (17%) were more likely to choose the "don't know" category than those who participated (3%). Furthermore participating adolescents were also more concerned about mobile phone exposure (at least fairly concerned: 12%) than non-participants (at least fairly concerned: 8%) (Table 2).

21 measurements hat to be excluded from the analysis due to technical errors. All exposure levels were far below the ICNIRP reference level and ranged from a mean of 0.13% (all measurement values below the limit of determination) to a mean of 0.92% of the ICNIRP reference level per second during waking hours. Median exposure was slightly higher for adolescents (0.19) than for children (0.18). The majority of measured values were below the limit of determination (82% of the measured values during waking hour).

Exposure was higher during afternoon hours than during morning hours. This applied for the children (median morning hours: 0.17, median afternoon hours: 0.19) as well for the adolescents (median morning hours: 0.18; median afternoon hours: 0.20) (table 3).

Exposure varied by the size of the town of residence. Median exposure levels during waking hours were highest in Munich and lowest in the smallest town (Landsberg) (figure 2).

Between 20% and 57% of the participants were in exactly the same exposure quartile on two days of the week (perfect agreement). Highest complete agreement was always seen between two consecutive days. However, the weekend (Friday to Saturday) differed considerably from the weekdays. For weekdays, exposure categories differed by at most one exposure quartile for more than 80% of the population indicating that misclassification of exposure might result in an underestimation of the effect (Table 4).

Sixty-seven percent of the parents of the children and 65% of the adolescents estimated that the next mobile phone base station was less then 500 meters away from their home. Regarding the self estimated use of mobile phones during the field study, nearly every child (98%) and 90% of the adolescents stated a very low usage (< 5 minutes) in the morning hours. For the afternoon hours also 98% of the children and 86% of the adolescents reported less than 5 minutes of usage. The participating children reported longer usage of a DECT phone in the afternoon (< 5 minutes: 85%) than in the morning (< 5 minutes: 91%). The same was stated for the adolescents (afternoon: <5 minutes: 63%; morning: < 5 minutes: 86%). In general adolescents had a higher DECT phone use than children. As expected, highest use was seen in the afternoon.