臺灣現職工作期間之存續分析——兼具加速失敗時間與比例風險計量的Weibull分析

AWEIBULL ANALYSIS OF THE CURRENT JOB TENURE IN TAIWAN WITH BOTH ACCELERATED FAILURE-TIME AND ROPORTIONAL HAZARDSMETRICS

本文根據人力資本理論，探討「現職工作年資」與「職涯年資」之不同及其影響變數。鑒於現職工作期間資料具有「右邊無法被觀察」和「最小極值分配」等特性，本文乃根據「Weibull 有母數對數線型持續分析」設定現職工作年資持續估計模型，並解構Weibull 的模型設定及其兼具的「加速時間失敗」與「比例風險」等計量機制。採用2012 年5 月行政院主計總處「人力資源與運用調查」的個體資料估計該模型，主要結果如下： 一、現職工作年資隨教育水準、一般經驗水準、大型企業或政府部門、典型就業、全日工作、已婚且配偶存在、30–44歲壯年組等變數正向移動;但隨任職於中小型企業、非典型就業、部分時間工作、無婚姻狀態、45–65 歲中高齡組等變數反向移動。二、加速失敗時間分析顯示，不具共變數之「平均參考主體」，其現職工作年資的失敗或老化速度比具有共變數之「平均主體」快約8 倍。三、在所有能對現職工作期間造成加乘效果、並對現職工作期間的終止風險率和相對風險比率產生減緩效果的變數中，以正式教育年數的效果為最大。反之，在所有會對現職工作期間造成負面效果、並對現職工作期間的終止風險率及相對風險產生加乘效果的變數中，以非典型就業的效果為最大，且不容小覷。

Relationships between current job tenure,which refers to the length of time in the current role at a specific firm, and firm-specific human capital and their determinants are examined theoretically. The approach of estimation for the survival model is explored by associating Weibull’s parametric log-linear dura- tionmodelwith the extreme-value type-I (min.) distribution and the accelerated life model, as well as both proportional and relative hazards as embedded in Cox’s proportional hazards model. By fitting the survival model to household data from the May 2012 Manpower Utilization Survey of Taiwan, the estima- tion results can be briefly summarized as follows: (1) Of all the covariates, for- mal schooling level imposes the greatest positive effect on the time length of current job tenure and hence has the greatest mitigating effect on the risk that the current job tenure fails (terminates); conversely, atypical employment im- poses the largest negative effect on the time length of current job tenure and hence has the largest increasing effect on the failure risk of current job tenure. (2) The current job tenure of the reference subject fails around 8 times earlier (or ages around 8 times faster) than that of the subject with covariates. More inter- estingly, this implies that the risk that the current job tenure of the subject fails at a given survival time equals only around an eighth the risk that the current job tenure of the reference subject fails at about an eighth times earlier than the given survival time. (3) Estimation results fromWeibull do have implications similar to results from both the accelerated failure-time and proportional haz- ardsmetrics.