Source：Safety Science 50 (2012) 1355–1367
Authors：Vivek V. Khanzode, J. Maiti, P.K. Ray,
真是太感激這三位印度大德先進了，如此有條理與架構性的整理過去近百年的相關理論，讓人反省自己所持觀點與背後的哲學觀(Again, 自己想做的議題被捷足先登，而且做得比我想像還要好)
Methods of hazard identification作者認為可區為以下三大類
1. Biased reactive approach, where information is analyzed after occurrence of an accident event.=事故/虛驚事件調查
2. Biased proactive approach, where identification of hazards and failure modes are based on information from similar systems or historic data from the same system(針對上述或類似事故/虛驚事件，有所本(≒bised)展開的Fault-tree analysis (FTA針對事故進行回溯各項可以預防的節點), event-tree analysis (ETA針對虛驚事件往前追溯可能發展情境), Failure Mode Effect Analysis (FMEA), Failure Mode Effect and Criticality Analysis (FMECA)
3. Unbiased proactive approach, where hazard analysis is carried out without waiting for the events to occur, and without any restrictive assumptions about presence of specific hazards in the work system.(不要存有一些事故發生的情境假設與前提，相對的等於天馬行空的杞人憂天？！)；這部份作者提到一些分析方法：Change analysis(變更管理？), energy analysis(火災爆炸模擬？), deviation analysis（沒用過？）, comparison analysis(=Benchmark？), hazard and operability (HAZOP) studies, job safety and work safety analysis, management oversight and risk tree (MORT)
備註：感覺還所本，而這部份的分類有些怪，不夠MECE
Methods of risk assessment1. Methods based on criterion variables依據特定參數與基準來衡量比較風險的大小：injury rate based on descriptive statistics, Lost-time injury frequency rate, fatal accident rate, cumulative incidence rate, severity rate, severity index, Time between incident occurrences(用這幾個指標的罩門在於：這幾個都是依據過去一段時間結果的落後性指標，無法預測未來！)
2. Methods based on modeling techniques第一招=>矩陣法：priority matrix for risk assessment based on probability of occurrence and severity of consequences.
矩陣法的罩門與批判請參照=>
http://eshmanager.blogspot.tw/2010/09/enterprise-risk-management-2-1.html
第二招=>統計分配參數估計accident/injury risk is modeled through appropriate statistical distributions, by fitting distribution of occurrence probability or consequences；作者提到幾個常用的統計分配
Beta distribution-based model with lost workdays (consequence model) as indicator of injury risk is one of the common models
Acute traumatic injuries occurring to individuals follow Poisson process and the inter-injury periods are exponentially distributed.
Injuries are modeled using mixed Weibull distribution, considering injury as a failure, analogous to reliability concepts.
延伸閱讀資料
Beta distribution=> http://zh.wikipedia.org/wiki/%CE%92%E5%88%86%E5%B8%83(浴缸曲線)
Poisson distribution=> http://zh.wikipedia.org/wiki/%E6%B3%8A%E6%9D%BE%E5%88%86%E4%BD%88
Weibull distribution=> http://zh.wikipedia.org/wiki/%E9%9F%A6%E4%BC%AF%E5%88%86%E5%B8%83(可靠度)
第三招=> soft computing techniques such as neural network-based models and functional block diagram-based models .
延伸閱讀資料：
neural network =>http://en.wikipedia.org/wiki/Neural_network (只是藉重電腦運算來進行FTA/ETA，計算各種情境的可能性，重點在於有沒有發生機率的參數資料，否則估算一樣不準)
另外有提到一些人因方面的風險估算，基本上可謂只是分類而已=類似矩陣法
Accident causation theories作者把這些理論的進展分成四代來看
第一代：事故傾向(千錯萬錯都是當事人的錯)
Person’s traits and unsafe behavior as responsible for accident (Greenwood and Woods, 1919)
1.1 Pure chance hypothesis=>一切都是天意與運氣，沒有理論可言
1.2 Accident proneness theories
1.3 Unconscious motivation theory=> Low scores on positive or socially desirable traits, low opinions on jobs and work environment, and withdrawals from work are found as significant contributors to accidents/injuries
我不太能認同作者這部份的分類，走私了一些工業心理與行為方面的”療法”，而非病因分析，而遲至1950年才有這些療法的觀念
第二代：骨牌理論(單線的因果)
Second generation theories (domino theories) conceptualize a chain of sequential events leading to an accident, and call these events as dominos (Heinrich, 1932).
第三代：流行病理論(人、媒介與環境間的交互作用)
Injury epidemiology models (originated in 1960s) represent third generation of accident research (Haddon et al., 1964).
Injury epidemiology models perceive three factors to explain the injury phenomenon as (Haddon et al., 1964): the host (the person injured), the agent (the energy leading to injury), and the environment (physical, biological and organizational). Among the environmental factors, transient factors (those changing with time) are the most immediate factors causing injury. 這一段作者的觀念，夾帶走私了能量釋放理論
第四代：系統理論(整體的複雜互動)
System approach to accident causation (the fourth generation) emerged in 1970s as a response to the challenge of maintaining safety in increasingly complex work system.
學界把這個部份的觀念用SEM走火入魔地展現….複雜與難以測量的不是人的心理，而是當事人所處的組織與組織背後的脈絡情境
底下作者整理了各種相關因素的研究
補充與延伸閱讀：
以下這張圖或許整理的更有條理
第五代會是複雜理論亦或是回歸統計或然率的觀點？
http://eshmanager.blogspot.tw/2012/05/complexity-and-safety.html
http://eshmanager.blogspot.tw/2011/12/4.html
以上各種理論，都還有學者不斷研究
Factors affecting injury 這一段或許作者背後的企圖，是想陳述各種理論與影響因素的解釋力大小(effect size)
這部份或許作者想暗示的是
Organization-related factors(=安全氣候，Higher management, coworker and supervisory support, higher commitment towards safety, safer perception of system status, and smaller workgroup sizes (less than 15), are found to be associated with low injury risks.)> Job-related factors(occupation, location of work, hazards in work system , activity , shift of working , and workplace factors ) > Individual-related factors（Age, work experience, education and EQ）
目前學術實證上比較確定Job-related factors的effect size> Individual-related factors，至於安全氣候的effect size(藥效大小)，由於安全氣候本身由許多構面(Construct)組成，本身就是一坨觀念的大雜燴，自然很難評估藥效
Injury mechanism modelsDeviation models從行為或程序的脫軌偏誤來看
Energy models從能量釋放的角度來看
Injury severity models可能有不同的結果
Intervention strategies1. Engineering interventions
2. Behavioral interventions
3. Enforcement interventions
感覺了無新意
Critical appraisal and future research directionsthere are methodological gaps in analyzing this sequence of events, which are shown in Fig. 2.
作者認為方法論上，存有以下鴻溝
從實務的角度，我倒覺得Gap 不是存在於可能的Risk與已發生的Accident之間，而在於不同方法有其評估的前提與假設條件，彼此的前提與假設不同，當然沒辦法整合在一起
Any injury causation model, in order to be effective, needs to comprise of four components
First, as injury involves transfer of energy to human body, models that do not consider energy interactions can only be of limited use in understanding injury.
Second, transient risk factors are important in realization of injury risk. These factors are the most immediate to injury incident and vary between the injured and non-injured person, when most of the other system variables are the same.
Third, an injury model needs to establish links of causal factors with energy transfer event. Stronger the relationship, more important is the role of that factor in injury occurrence.
Fourth, injury models use different response variables to measure the injury phenomenon. For the findings to be useful in understanding injury, the response variable should have a good content validity with injury event.
作者認為一個好的工傷事故因果模型需包含以下要點
1. 描述能量(釋放)接觸人體的過程
2. 造成是否受傷差別的因素與原因
3. 能量釋放與是否受傷間的關連性
4. 要具備能夠有效預測的測量變數
第四點最難、難度最高=>事故與意外的發生，要用機率來看比較健康，然而機率的詛咒再於型一與型二錯誤=>機率高不代表會發生、機率低不代表不會碰到。
沒有留言:
張貼留言