Reducing overestimation in reported mobile phone use associated with epidemiological studies

Bioelectromagnetics. 2008 Oct;29(7):559-63. doi: 10.1002/bem.20424.


Case-control studies of mobile phones are commonly based on retrospective, self-reported exposure information, which are often characterized as involving substantial uncertainty concerning data validity. We assessed the validity of self-reported mobile phone use and developed a statistical model to account for the over-reporting of exposure. We collected information on mobile phone use from 70 volunteers using two sources of data: self-report in an interview and network operator records. We used regression models to obtain bias-corrected estimates of exposure. A correlation coefficient of 0.71 was obtained between the self-reported and the network operators' data on average calling time (log-transformed minutes per month). A simple linear regression model, where the duration of calls acquired from network operators is explained with the self-reported duration fitted the data reasonably well (adjusted R(2) 0.51). The constant term was 2.71 and the regression coefficient 0.49 (logarithmic scale). No significant improvement in the model fit was achieved by including potential predictors of accuracy in self-reported exposure estimates, such as the pattern of mobile phone use, the modality of response to the questionnaire or demographic characteristics. Overestimation in self-reported intensity of mobile phone use can be accounted for by the use of regression calibration. The estimates obtained in our study may not be applicable in other contexts, but similar methods could be used to reduce bias in other studies.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Artifacts*
  • Body Burden
  • Cell Phone / statistics & numerical data*
  • Computer Simulation
  • Data Interpretation, Statistical*
  • Epidemiologic Research Design*
  • Finland / epidemiology
  • Humans
  • Models, Statistical*
  • Radiation Monitoring / statistics & numerical data*
  • Regression Analysis