Comparison of statistic methods for censored personal exposure to RF-EMF data

Abstract

Overview

Personal exposure to Radio Frequency Electromagnetic Fields (RF-EMF) has been thoroughly studied due to its potential effects on health. Handling equipment's detection limits poses a particular challenge, as values below these limits are often recorded as nondetects or censored data. This issue complicates the analysis and understanding of RF-EMF exposure data.

Findings

• The study introduces a comparison of four statistical methods to approach the problem of censored data in RF-EMF exposure measurement.
• Three methods attempt to substitute censored data: the Regression on Order of Statistics (Method 1), substituting nondetect values by half of the detection limit (Method 2), and treating the detection limit itself as a valid measurement (Method 3).
• The fourth method (Method 4) regards censored data as missing values.
• Data from 14 different frequency bands ranging from FM to WiMax were analyzed, recorded over 76 days, every 10 seconds by the personal exposimeter Satimo EME Spy 140 in Albacete, Spain.
• Methods 1, 2, and 3 provided similar outputs for mean and median values, though they tended to underestimate means under high nondetect records. Method 4, treating the censored data as missing, compiled significantly different mean values but performed similarly when nondetects were below 20%.

Conclusion

This comparison highlights the influence of method selection on the measurement of RF-EMF exposure and implies that disregarding nondetects when they form a small percentage might lead to a more accurate portrayal of environmental exposure.