Evergreen I
Adjusting for Misreporting in Count Data (306502)
James Hardin, University of South CarolinaJames Hussey, University of South Carolina
*Gelareh Rahimi, Carle Foundation Hospital
Mindi Spencer, University of South Carolina
Feifei Xiao, University of South Carolina
Keywords: Count regression, Underreporting, Overreporting, Misreporting, Simulated Likelihood
Any counting system is prone to reporting errors including underreporting and overreporting. Ignoring the misreporting pattern in count data can give rise to bias in the estimation of Poisson regression model parameters. We assumed that observed counts are the result of two consecutive processes and developed a model applicable to potentially misreported data where the true unobserved counts follows either Poisson or Generalized Poisson distribution. We applied the proposed model to the data from the EBAN study to see whether an intervention could be effective in lowering rate of unprotected intercourse acts within HIV-serodiscordant couples and whether there exists a pattern of underreporting/overreporting in individuals’ responses. The results confirmed that younger people and those with concurrent partners are more likely to underreport their high-risk sexual behavior while living with study partner, found to be correlated with a pattern of overreporting. The proposed models adjust for overreporting and underreporting at the same time. In addition, our approach allows users to specify the individual characteristics that contribute to misreporting.