The simulated data were assigned FC values calculated with a function of the frequency distribution of the measured FC data and LAZ values were generated with random Gaussian variation from a line of linear association of LAZ and FC.
In order to minimise the amount of missing data for each item, a 'best estimation' method was used whereby respondents with missing data were assigned the item-score of a respondent with a similar age, gender, income, and education level [ 37- 39].
Ørsted and CHAMP data were assigned a relative 1/σ2 weight, with σ equal to 2 nT and 3 nT for CHAMP and Ørsted, respectively.
For the other crops (colza, peas, potatoes and fresh vegetables), municipalities with confidential data were assigned to the category "present" (respectively, n = 142, 142, 139, and 156 municipalities, home to n = 488, 712, 719 and 747 pregnant women).
Cox proportional hazards regression was used to estimate the average age- and sex-adjusted relevance of each baseline characteristic to the risks of ESRD and death (the small numbers of participants with missing data were assigned the mean or median value, as appropriate).
Thus, for corn and wheat crops, the few municipalities (respectively n = 16 and n = 12) with confidential data were assigned to the low tertile, on the assumption that the percentage of area dedicated to these crops by only one farmer in a municipality was small.
Reeves et al. [ 3] considered the impact of measurement error in a situation where individual radon exposure was measured with additive classical error but where subjects with missing radon data were assigned an area average.
Infants with missing outcomes data were excluded from the analyses, while those with missing covariate data were assigned a zero in the case of binary indicators.
