- 作者: Kazuo A. Yamaguchi
- 出版社/メーカー: Sage Publications, Inc
- 発売日: 1991/09/25
- メディア: ペーパーバック
- クリック: 12回
- この商品を含むブログ (3件) を見る
State dependence refers to a situation in which the covariate process is influenced only by the states of the dependent process. For example, if subjects' employment statuses are affected by their marital statuses, reverse causation due to state dependence is present when employment status is used as a covariate for the hazard rate of divorce.
On the other hand, rate dependence refers to a situation in which the covariate process is influenced directly by the hazard rate of the dependent event. For example, a psychological covariate, such as depression, in the analysis of divorce may be influenced directly by an increase/decrease in the hazard rate of becoming divorced.
1. Compared with linear regression analysis, unobserved heterogeneity is a more serious problem in event history analyses that employ proportional hazards models and/or their extensions. In linear regression analyses, omitted variables that affect the dependent variable do not cause any bias in parameter estimates if they are uncorrelated with the included explanatory variables. By contrast, omitted variables that influence hazard rates in proportional hazards models do cause bias parameter estimates even if they are initially uncorrelated with the included covariates (Trussell & Richards, 1985).
Hence, marijuana use has an indirect effect on the occurrence of premarital cohabitation by postponing the termination of the risk period of premarital cohabitation. That is, premarital cohabitation is more likely to occur for marijuana users because they remain single for a longer period of time. Similarly, marijuana use--which increases the hazard rates of experiencing the first premarital sexual intercourse--indirectly affects the occurrence of premarital pregnancy by accelerating the timing of entry into the risk period of premarital pregnancy.