Processing length bias of time intervals between the last episode and the interview

The paper offers an improved method for respondents' behavior rate (and other parameters) estimates based on data about its last episode. The mathematical model of this behavior is a generalized Poisson stochastic process; the observations are respondents' natural language answers about time intervals between the interview and the last episode of the respondent's behavior. The paper describes the properties of this stochastic process. Our method process length bias based on an assumption that moment of interview is one of the behavior episodes. The paper proposes approach to taking account of data granularity.