A Swedish radon data set, consisting of more than 8000 measurements of residential radon levels in about 50% of the Swedish municipalities were analysed using a multilevel approach.

The results were compared with those of a single-level analysis. We found that there was a significant variability between municipalities. The point estimates of the population mean radon levels were similar (geometric mean 60 Bq/m3 and arithmetic mean 106 Bq/m3).

The analysis shows the advantages of multilevel modeling compared with a single-level OLS model.

A single-level model results in too optimistic standard errors, about 25% of those of the multilevel

model which can lead to erroneous conclusions.

In a multilevel model including house type as a fixed effect (single-family house, row house, or apartment in multi-family house), the estimates of the fixed effect of house type were similar for the single-level and the multi-level models