2.5 – Metrics of SHE performance

 2.5.1 - Requirement for SHE performance indicators

 Good SHE management principles prescribe monitoring of SHE performance.  Such performance monitoring requirements are included in SHE management systems guidance such as those written by the HSE [84] and ISO [92] .  The measurement and positive response to SHE performance indicators also makes good business sense, as significant loss-prevention savings can be made [42].

 2.5.2 - Lagging and leading SHE performance indicators

Measurement of SHE performance can be done proactively or reactively.  Proactive measurements are made before an event, and reactive measurements are made after an event.  Proactive measurements are often referred to as ‘leading’ and reactive measurements as ‘lagging’ indicators.  Reactive measurements include the reporting of the number of injuries, accidents and loss incidents.  Proactive SHE measurements include examining the number or quality of audits performed, number of ‘toolbox’ talks, workplace inspections and environmental performance monitoring with statistical process control.  Leading SHE performance measurements have benefits over lagging indicators in that absence of occurrence of lagging indicator occurrences does not necessarily indicate the absence of underlying problems that could give rise to a loss event.  From a business and perhaps ethical standpoint it is therefore more appropriate to build SHE improvement strategies around leading SHE performance indicators.  Reliable and meaningful leading indicators are, however, generally more difficult to identify, construct and manage.  Further information regarding the usefulness of leading versus lagging SHE indicators is given by Mansley [110] .  

It is a generally accepted view that the majority of industries assess the success of their safety programmes using lagging indicators and that the measurements of leading indicators are rarely used as ‘key performance indicators’.  Lagging SHE performance indicators such as environmental incident and accident rates present a direct and unambiguous measure of safety performance.  Thompson et al [163] suggest that these lagging indicators are not effective measurements of safety for the following reasons. 

·         “Generally accidents and incidents occur infrequently.  The resultant low variance means the statistical significance of any results is reduced”.

·         “Accidents do not necessarily occur to those people who are in control or who have an influence over the outcomes”.

·         “No matter how safe and compliant personnel are, extraneous random events and influences can cause or contribute to accidents”.

·         “The way in which accidents are recorded may be inconsistent.  Under- and over-reporting can give rise to errors”.

·        “The severity of accidents may have an influence over whether an incident is or is not reported.  For example, sawdust in the eye requiring eye-wash irrigation may go unreported.  The surgical removal of swarf from an eye will more likely be reported”.

Although the above reasons may affect the reporting of accident statistics they should not preclude their use in research as will be discussed later.  The problem of using accident data such as lost time injuries in culture research has been highlighted by Zohar [177] and Menckel and Carter [116] , who note that lost time injury accidents occur relatively infrequently and therefore it may take many years to collate enough data to be statistically significant. 

The difficulty in identification and selection of suitable SHE lagging performance indicators is well recognised in organisational safety research literature.  Diaz and Cabrera [41] write:

“In our opinion, one of the greatest challenges for organizational safety researchers is precisely the development of a set of reliable measures of organizational safety.  This would imply the use of various measures, not only one”.

If a statistically significant link between lagging SHE performance indicators and SHE culture factors could be shown, then the latter would form the basis of a potentially powerful leading SHE performance indicator.  The identification of a link between organisational culture and lagging indicators and its subsequent use as a leading SHE KPI is one of the principal objectives of this research. 

2.5.3 – Behavioural performance indicators 

It has been claimed that 96% of industrial accidents are caused by unsafe behaviours [144] , [42].  Since unsafe behaviours precede accidents, their measurement provides a leading SHE performance indicator.  Behavioural-based safety programmes such as Safe Unsafe Acts Auditing [125] and B-Safe [144] measure workers’ behaviours.  The scoring of behavioural audits allows for the identification of unsatisfactory behaviours as well as reinforcement of desirable behaviours.  Although most behavioural safety programmes are fundamentally similar in approach [40] , their output and scoring systems differ.  Inter-comparisons between companies using two different systems are not therefore possible.

2.5.4 – Factors affecting the reporting of performance indicators 

Several factors may influence the accuracy of reported SHE performance indicators.  Madsen  [109] has indicated that accidents, incidents and near misses are more likely to be reported in a supportive ‘no-blame’ working environment compared to an environment in which reporting is likely to give rise to punishments or other retribution.  Pidgeon [128] disagrees and suggests that a reporting culture should take into account some degree of responsibility and accountability.  Safety incentive schemes can cause under-reporting of events.  Safety schemes provide monetary or other rewards for good SHE performance.  Rewards may be paid to individuals, groups or charities.   Sawacha et al [146] found that the introduction of bonus payments for good productivity performance reduced reported safety incidents. 

2.5.5 – Self-reported performance indicators    

Several researchers have used self-reported accident data within their research [37, 76, 163].  The use of self-reported accidents is questionable due to the following reasons:

 In terms of accident data, Thompson et al [163] argued that self-reported injuries are preferred to “minor workplace accidents” as the latter often go under-reported.  Thompson et al did however recognise that:  “ … those unreported [minor workplace accidents] may be the best indicator of improving or worsening safety conditions that might eventually lead to serious injury”.

2.5.6 – Calculation of accident statistics

Researchers must be mindful of the different ways in which accident statistics are calculated when comparing data.  For example, accident frequency rates and incident frequency rates are commonly calculated within industry [100] .  Accident and incident rates do not measure the same thing.  The frequency rate is the number of occurrences of a given type of event expressed in relation to the base unit of measure.  For example, number of accidents per number of hours an employee is at work.  The formula for injury frequency rate taken from Ridley [141] is given in Equation 1.

    (1)

For the purposes of the work detailed in Chapters 5 and 6, so long as the number of injuries occurring at a site is divided by the total number of employee hours worked per unit time, the precise equation used to calculate the injury frequency rate is unimportant.     Incident rate calculations derive from the U.S. Department of Labour [172] .  Incident rates indicate the number of incidents per 200,000 man-hours.  

2.6 - Correlation between organisational culture and SHE performance

Although safety culture has been stated to be an important contributory factor that predates accidents such as Piper Alpha [32] and Kings Cross [54] , little research has been done to establish the quantitative link between the two.  Pidgeon [130] noted:

“ … some 10 years on from Chernobyl, the existing empirical attempts to study safety culture and its relationship to organisational outcomes have remained unsystematic, fragmented, and in particular under-specified in theoretical terms”. 

In his discussions regarding the evolution of the terms ‘safety culture’ Sorenson [158] wrote:

 “Statistical evidence that unambiguously links safety culture with the safety of operations is surprisingly rare, especially within the nuclear industry”.

Industry’s enthusiasm to improve safety culture in the absence of statistical evidence linking it to organisational outcomes has been summarised by Sorensen [158] :

“The proponents of safety culture as a determinant of operational safety in the nuclear power industry are relying, at least to some degree, on that indirect assumption [that relatively low accident plant must have a relatively good safety culture]”. 

Even in the arguable absence of a significant statistical link between safety culture and lagging performance indicators, other researchers have commented upon the positive benefits of carrying out organisational culture climate surveys.  Bailey and Peterson [7] concluded that a safety perception survey was useful because:

·         “the effectiveness of safety efforts cannot be measured by traditional procedural-engineered criteria like safety reviews, audits and inspections”.

·         “the effectiveness of safety efforts can be measured with surveys of employee perceptions”.

·         “a perception survey can effectively identify the strengths and weaknesses of elements of a safety system”.

·         “a perception survey can effectively identify major discrepancies in perception of program elements between hourly rated employees and levels of management”.

·         “a perception survey can effectively identify improvements in and deterioration of safety system elements if administered periodically”.

In addition to the above, other benefits result from the administration of safety perception surveys, including: 

·        Raising employee’s awareness of SHE issues.

·        Identification of initiatives and action planning to improve SHE management systems and hardware deficiencies.

·        Inter and intra-organisation benchmarking.

·        Involvement of personnel in all levels of the organisation in SHE who would not otherwise be involved.

·        Promotion of a positive safety climate in which SHE initiatives such as behavioural-based safety initiatives are better received and implemented.

·        Raising discussion topics that would otherwise be perceived to be difficult to discuss.

In his literature review Zohar [176] found that several organisational characteristics were able to discriminate those industrial sites with very good safety performance and those with bad safety performance.  He found that the most common factor that distinguished superior safety performing sites was a strong management commitment to safety.  Those sites with good safety performance were characterised by management being regularly and actively involved with safety issues.  Other relationships that are more obscure were also found, for example, companies with good safety performance were characterised by personnel holding their Safety Officers in higher regard.  This finding was also reported in the 1976 Accident Prevention Advisory Unit report [82] .

Studies attempting to correlate organisational culture with lagging SHE performance indicators are complicated by the potentially very different inherent SHE risk of the units encountered.  This complication results from the Heinrich et al [73] model of accident causation.  In the Heinrich et al model, accidents are caused by the simultaneous occurrence of unsafe acts and unsafe conditions.  One can infer that a greater number of accidents will therefore occur in work areas with a large number of unsafe conditions compared to a similar area with a lesser number of unsafe conditions. 

In his study examining the relationship between accidents and organisational culture, Zohar [177] attempted to take the inherent risk of a workgroup into consideration by taking a risk factor into consideration which was based upon the subjective assessment of the risk each area was subjected to.  Zohar failed however to establish a correlation between perceived risk of a work area and the number of accidents occurring within it.

Shannon et al [151] performed a survey of available literature which examined the relationship between organisational and workplace factors and injury rates.  Examples of organisational and workplace factors included:

·        Characteristics of the workforce such as age, seniority, education, literacy.

·        The presence and effectiveness of a health and safety committee such as senior management presence, numbers of workforce represented and duration of participant training.

·        Managerial style and culture such as encouragement of long-term commitment of workforce, profit sharing, grievance rate, good relations between management and workers.

·        Organisational philosophy on health and safety such as safety incentives, presence of rules and the unsafe behaviours of workforce observed.

Sections 2.2 and 2.3 gave an overview of the multi-dimensional nature and definition of culture.  Potentially, all of the factors contained within the Shannon et al study could be indicative of the underlying culture in an organisation.  The scope of their literature search was restricted to studies that contained at least 20 sites and were of a quantitative nature rather than qualitative.  The literature survey identified 10 studies meeting this criterion   [19, 23, 69, 70, 89, 150, 152, 153, 155, 171].  The research studies identified by Shannon et al shared a common feature in that all of them attempted to correlate single-factor elements of organisational culture with injury rates.  Other examples of attempts to correlate single aspects of organisational culture with SHE outcomes come from the field of behavioural-based safety programmes. 

Much information regarding behavioural-based safety programmes and their associations with improved safety outcomes is available in the literature.  Example accounts of these systems include Ormond [125] , Sulzer-Azaroff et al [162] , McAffe and Winn [112] , all of whom claim significant reduction of accident rates and business losses.  The reduction of accident and loss rates claimed by proponents is significant; for example, Cooper [25] states that a 40-75% year-on-year reduction in accident rates and accident costs represents typical improvements after the introduction of a behavioural-based management system. 

Research has also indicated associations between personal characteristics and accident occurrences.  Ferguson et al [55] found a relationship between educational background and accidents.  Leigh [105] discovered a relationship between gender and accidents.  Leveson and Hirchfield [106] found a relationship between the occurrences of accidents and recent ‘life events’; those having recently experienced an incident such as divorce were shown to be more likely to be involved in an accident.  Melamed et al [115] discovered a link between accidents and job satisfaction.  Dwyer and Raftery [50] have linked accidents with management reward for work rates and overtime.  Cox and Cox [28] linked perceptions of risks and/or attitudes toward safety to safety behaviour.  Researching the links between safety climate factors and accidents, Coyle et al [31] concluded that safety climate factors correlated highly with traditional (lagging) indicators.  Coyle et al’s experiment, however, did not quantify this correlation. 

Based upon a literature review of previous research papers, Zohar [176] formulated a safety climate questionnaire.  The questionnaire consisted of 49 questions designed to measure 7 organisational climate dimensions, namely:

·        Perceived management attitudes toward safety.

·        Perceived effects of safe conduct on promotion.

·        Perceived effects of safe conduct on social status.

·        Perceived organisational status of Safety Officer.

·        Perceived importance and effectiveness of safety training.

·        Perceived risk level at the workplace.

·        Perceived effectiveness of enforcement versus guidance in promoting safety.

The questionnaire was administered to 20 factories in the chemical, metal fabrication, food processing and textile industry sectors in Israel.  The questions measuring the factors were then aggregated to give a single ‘safety climate score’.  A team of four judges assessed the perceived risk of each factory.  Spearman rank correlation coefficients between the safety climate scores and the subjective perceived risk were calculated for 5 of the metal, 4 chemical and 3 food companies.  The resultant Spearman rank coefficients were 0.9 for the metal factories, 0.8 for the chemical factories and 0.5 for the food companies.  A definition and interpretation of Spearman rank coefficient is given in Section 2.9.3.1.  The usefulness of Zohar’s [176] research is debateable.  He did not provide any information regarding the significance of the results (although, based upon the limited number of degrees of freedom, only the metal factories’ Spearman rank correlation exceeds the level of statistical significance at the 95% level).  Subjective risk criteria were used and the safety climate score was one-dimensional.  As Section 2.2 indicated, organisational climate is a multi-dimensional construct.  Using a one-dimensional scale is therefore inappropriate. 

Ostrom et al [126] administered a safety climate survey within the US nuclear industry.  In their paper they provided examples of the usefulness of the organisational climate survey data.  One such example given is a graphical comparison of the accident records of five departments with their relative score of one particular climate attribute.  Although the comparison was made, no numerical analysis was performed. 

By administration of a climate survey, Lee and Harrison [103] measured individuals’ attitudes in three nuclear power stations.  24 climate factors resulting from the climate survey were correlated with one or more of nine self-reported accident criteria.  In his paper, O’Toole [127] proposed a link between employees’ perceptions of management’s commitment to safety and injury frequency rate; the correlation was, however, not quantified. 

Fleming et al [58] measured subjective risk perception in six offshore drilling rigs and examined how these perceptions related to lost time accident frequency records (number of LTAs x 1,000,000/number of man hours worked) and available quantitative risk assessment data.  The risk perceptions of 622 workers across six different UK oil platforms were measured by the application and analysis of a 14-section questionnaire.  The risk perceptions of the workers were plotted against the rank average LTA frequency for the preceding 2-3 years (Table 2.2).  It is noted that Flemming et al did not provide the actual LTA figures within their paper.  It is also noted that the use of the mean feelings for safety is questionable due to the closeness of the resultant values.  Summarising the data in Table 2.2, Fleming et al conclude that:

“… the installations are generally ranked in the same order (with the exception of installation 5) according to respondents feelings of safety, as they are by the frequency of LTAs, although the degree of correlation was not found to be significant”.

Table 2.2 - Fleming et al  [58] – Comparison of safety feeling and the frequency of lost time accidents

A similar study that examined the relationship between safety climate and the number of accidents was carried out at British Steel [17] .   The study correlated the attitudes of workers at sixteen British Steel plants with the numbers of accidents occurring at each plant.  The study found that the correlation between safety climate and accident numbers was stronger than that between a panel of experts’ perception of the inherent safety risks and accident numbers.  Attempting to correlate ‘feelings of safety’ against accident data is potentially problematic.  It is a well-established fact that individuals and groups of individuals have different thresholds of risk.  Adams [2] calls this risk threshold a person’s ‘risk thermostat’.  What one person considers ‘risky’, another may consider ‘acceptable’.  In the absence of negative feedback, repeated exposure to risk gives rise to a person’s risk tolerability threshold being raised.  The consequence of this is that, for a given hazard, the correlation between feelings of risk and actual exposed risk may be weak.  The use of measurements of feelings as a metric to be correlated with SHE outcomes should therefore be used cautiously.

In their paper Fleming et al [58] also compared ‘feelings of safety’ with quantitative risk assessment (QRA) data.  All offshore installations working within UK territorial waters are required to have a temporary refuge (TR) for emergency use.  Five of the six installations provided Fleming et al with QRA data associated with temporary refuge (TR) impairment, i.e. an inability to use the refuges.  Flemming et al noted within their paper that the QRA data provided to them was confidential.  Fleming et al interpreted TR impairment as follows:

 “The TR failure value is a representative measure of how secure the platform is with regard to the reliability of protective systems on the installation”.

 Flemming et al proposed that the TR QRA values represented:

“… a measure of the cumulative failure of the systems, which takes both the probability and consequence of the events into consideration”.   

The author is of the opinion that Flemming et al perhaps have an overly high expectation regarding the ability of QRA to measure the overall risk of an installation.  The rankings of feelings of safety towards hazards at an installation were compared with the order of risk from the QRA calculations for TR failure (Table 2.3).  Comparison between the rankings indicated a high but non-significant correlation (due to the small sample size).  Again the author questions the validity of the survey due to the closeness of the mean feelings for safety.

Table 2.3 - Fleming et al [58] – Comparison between feelings of safety towards hazards to the installation and the likelihood of TR failure

In a study comparing culture factors within organisations, Coyle et al [31] wrote:

“From a proactive viewpoint, it appears that low scores on the safety climate factors identified in the organisations studied here correlated highly with traditional indices such as lost time or accident rate.  While the design of the present study did not enable this to be quantified, clearly this has major implications for planning and managing occupational health and safety.  The relationship between safety climate analysis and other positive performance indicators of occupational safety and health has not been reported and is a major area for future research”.

Torbjørn [165] has examined the relationship between perceived risk, job stress and frequency of accidents and near misses.  Perceived risk and job stress of eight Norwegian oil installations representing five different companies were evaluated by administering a question survey.  The results of the questionnaire were analysed and compared with self-reported injury data.  Although Torbjørn’s study reported correlations between organisational culture factors, no statistical correlations were calculated between any of the factors and injury data.

Canter and Olearnik [17] , Canter and Donald [16] , Donald and Canter [49] developed a ten factor question set to measure safety climate, and administered it to ten chemical plants in the Humberside area of the UK that operated under the Control of Industrial Major Accident Hazards (CIMAH) Regulations.  The results of the study showed that there was a:

“… strong link between the ten climate measures and the number of self-reported accidents”.

It is the author’s opinion that the correlation of self-reported accident rates with climate metrics is inappropriate because self-reported data are unreliable.  The issue regarding the use of self-reported accident data is discussed in Section 2.5.  In their conclusions, Donald and Canter [49] suggest that it would be advantageous to establish the relationship of climate factors with other safety performance indicators, especially those which are not confined to occupational safety.  

Diaz and Cabrera [41] administered a safety climate and attitude survey to two companies and one authority associated with a Spanish airport, namely:

·        A ground handling division of an airline (Total workforce of 247).

·        A fuel company who provided refuelling services at the airport (Total workforce of 45).

·        Personnel from the airport authority (Total workforce of 73).

166 subjects participated in the survey.  The responses were used to assign a safety attitude, safety climate and safety level scale to each of the three companies.  The safety level scale was calculated based upon the response to six questions that measured workers perceptions regarding:

·        Their involvement in an accident in the previous 12 months and the likelihood of them being involved in accidents in the near future.

·        The level of safety involved with work tasks.

·        The compliance with safety standards.

·        The general level of safety of the operators.

Diaz and Cabrera then plotted the resulting average scores for safety attitude, safety climate and safety level of the three companies.  The resultant plot indicated that high safety-level scores correlated with high safety-climate and safety-attitude scores.  Although a correlation was demonstrated, the results of the analysis were not statistically significant due to only three companies being involved in the survey.

According to Zohar [177] , establishing a correlation between organisational climate and safety outcomes has been hampered by the lack of objective safety criteria and data.  Various objective criteria have been used by researchers including self-reported safety behaviour [37] , experts’ rating of safety level [176] , retrospective accident data [13] and frequency of micro-accidents [177] . 

Research looking at how safety climate indicators correlate with lagging safety performance indicators has been recently reported by the HSE [88] .  In their research, safety climate surveys were performed on 13 North Sea oil-drilling platforms during 1998 and 1999.  In 1998, 682 questionnaires from 10 installations were available.  In 1999, 806 questionnaires from 13 installations were available.  The 1998 and 1999 safety climate surveys differed.  Both surveys grouped questions into related themes.  It is noted that the grouping of the questions into themes was done subjectively and was not subject to statistical analysis.      The 1988 survey consisted of grouped questions under the following six themes: 

 The 1999 survey included additional questions that were grouped under the theme of operator-contractor interface. 

 In both surveys the average responses to all of the questions related to each theme were calculated for each installation.  The resultant average responses were given the label of ‘safety climate factor score’.  The values of safety climate factor scores were then correlated with the following four lagging safety indicators: 

·        Number of injuries requiring absence from work greater than three days.

·        Number of dangerous occurrences.

·        Number of visits to the rig medic for first aid.

·        Number of RIDDOR [75] reportable incidents.

 The reported significant correlations (at 95% confidence level) between the climate factors for the 1998 and 1999 data are reproduced in Tables 2.4 and 2.5 respectively. Negative correlations are associated with favourable scores linked to good safety performance.  A definition of Spearman rank correlation coefficients is given in Section 2.9.3.1.   

Table 2.4 – Correlation between 1998 climate factors and SHE lagging indicators [88]

Table 2.5 – Correlation between 1999 climate factors and SHE lagging indicators [88]

It is interesting to note that, in Table 2.5, the relationship between management commitment and dangerous and RIDDOR occurrences is converse to intuitive expectations.  In their conclusions the HSE [88] state: 

“Assuming that the Offshore Safety Questionnaire does measure safety climate reliably, it would appear that dimensions of climate predictive of safety outcome in one time period do not retain their predictive power either between years or between accident types”.

Silvia et al [154] examined the correlation between safety climate and safety outcomes.  In their study an organisational safety climate questionnaire was administered to 930 individuals, representing 40% of the population, in fifteen Portuguese organisations in different sectors, including the chemical industry, public administration, electricity and health.  The responses to the survey were used to provide measures of the following five safety climate dimensions:

·        Strength of organisational climate index (OCSI) (sic).

·        Strength of safety climate index

·        Strength of safety as an organisational value index

·        Strength of organisational safety practices index.

·        Strength of personal involvement with safety index.

Attempts to obtain accident rate, accident frequency rate and severity rate data for each of the fifteen organisations were made.  Not wholly clear definitions for the three rate criteria used by Silvia et al [154] are as follows.

Accident rate:  An instantaneous bodily defect so that the individual is physically or mentally, as determined by a competent medical authority, incapable to work on a scheduled day or shift, resulting in at least three days off the job [18] .   It is noted that accident rate, as defined by Silvia et al, is actually the number of accidents that have occurred and therefore is not a rate.  

 The amount of time lost due to injuries per million working hours.

 Severity rate:  The number of workdays lost per million hours (it is noted that Silvia et al did not state in their paper that the severity rate refers to the number of workdays lost per million working hours).

After examination of the collected data, Silvia et al [154] found that not all of the 15 organisations were able to provide information that followed the above definitions.  Seven organisations were able to provide accident rate data.  Six organisations were able to provide accident frequency rate data and five organisations were able to provide severity rate data.  Silvia et al [154] then went on to correlate the metrics of safety climate with the above three types of accident rate data.  The resultant Spearman correlations are reproduced in Table 2.6.  Numbers in parentheses indicate those results above the 95% significance level.  Further information regarding Spearman correlations and their significance is given in Section 2.9.3.1. In their conclusions Silvia et al [154] write: 

“…. these results suggest that OSCI has some capacity to predict and discriminate organisations with different accident levels”.

  Table 2.6 - Silvia et al [154] Spearman correlation between safety climate metrics and accident data 

Table 2.6 indicates statistically significant Spearman correlations between all five safety climate metrics and the accident rates.  Silvia et al do not provide information regarding how many personnel were in each of the organisations taking part in the study.  The author of this thesis is of the opinion that the use of accident rate as defined above, cannot be used in the above correlational exercise unless the number of personnel in each organisation is the same.  This opinion is reinforced by examining the Spearman correlations between the safety climate metrics with the accident frequency rates.  Column 3 of Table 2.6 indicates that only the ‘strength of organisational safety practices’ correlates with accident frequency rates above the 95% confidence level.   Zohar [177] administered an organisational climate question survey to 534 production workers, divided into 53 work groups, in a metal-processing plant.  He performed principal component analysis (PCA) on the question responses.  The results of the analysis indicated two principal components which Zohar labelled as ‘Supervisory Action’ and ‘Supervisor Expectation’.  The subunit risk of each of the 53 workgroups was subjectively assessed and scored by each of the workgroup supervisors.  The numbers of lost days due to injury and micro-accidents (minor accidents requiring first aid) were recorded for a period of five months following the administration of the survey.  Zohar then went on to correlate subunit risk, supervisory action, supervisory expectation, injury rate (based upon micro-accidents divided by group size) and accidents (expressed as the number of lost days due to injury).  The results of his inter-correlations are given in Table 2.7.  In his paper, Zohar does not state what type of correlation is used.  Zohar’s results indicate:    

·        There is a marginally statistically significant relationship between micro-accidents and the number of lost days due to injury.

·        There was no significant relationship between supervisory action or expectation and the number of micro-accidents or number of lost days due to injury.   

Table 2.7 – Zohar’s [177] correlational results

 In the final part of Zohar’s paper he performed least squares regression of subunit risk, supervisory action, and supervisory expectation with micro-accidents as the outcome variable.  Insufficient detail is provided within the paper to understand precisely what analysis was done.  Zohar reported that his model accounted for 16% of the micro-accident variation seen within the responses.  He went on to write:

‘It is evident that both climate subscales provided significant prediction of the micro-accident rate’.

For Zohar to write the above is inappropriate as his model fails to account for 84% of the observed micro-accident variation.

Zohar’s use of PCA followed by regression of the resultant principal components with micro-accidents is arguably also fundamentally flawed.  When PCA is applied to data, the principal components are arranged to maximise the amount of explained variance.  PCA is not therefore able to take into account the correlation between the principal components and the outcome variable of interest, in this case, micro-accidents.  The application of suitable statistical techniques to establish relationships between multivariate and univariate data is further explored within Section 2.9.3. All of the research surveyed during the literature review that has attempted to correlate organisational culture with SHE outcomes used the mean survey responses as predictor variables.  No research was found that examined the relationship between metrics associated with the distribution of the predictor variables and SHE outcomes.

 According to Sorensen [158] :   

“No performance indicators to gauge safety culture and its impact on safety of operations appear to have been identified and validated”. 

After reviewing organisational safety culture and climate research over the proceeding twenty years, Guldenmund [68] concluded:

“ .. the measurement of safety climate could be considered an alternative safety performance indicator … research should not be undertaken to develop ‘new’ safety climate measurement instruments, but should rather focus on the validity of the construct and whether it indeed yields a robust indication of an organisation’s safety performance”.

2.7 - Modelling of accident causation factors

 In an attempt to better understand the factors that cause accidents, several researchers have hypothesised the sequence and relationship of events leading up to accidents.   A survey of the literature indicates that the causation factors featuring in accident causation models differ from researcher to researcher.  Factors such as organisational issues, lack of personal attention, attitudes, perceptions, stress and peer pressure have all featured within accident models.  According to Guldenmund [68] there are two different types of cultural model. 

·         “Normative or prescriptive models which seek to describe and specify safety climate or culture per se”, and;

·         “Descriptive or empirical models, which attempt to summarise findings from one or several organisations studied”.

According to Tomas et al [164] , although models of accident causation are often hypothesised, they are not usually validated by the use of techniques such as structural equation modelling (SEM).  Further information regarding SEM is given in Section 2.9.2.   Thompson et al [163] modelled safety climate and perceptions and demonstrated that management have an important role in establishing an organisational climate that affects workplace self-reported accidents.   Cheyne et al [20] modelled employees’ attitudes toward safety by relation to their appraisals of commitment to safety within their organisation.    They found that:

“ … the architecture of attitudes to safety is, at least in part, dependent on the industrial context, or work environment.  … The models showed that perceptions of management actions and safety training were related to appraisals of the organisation’s commitment to safety as well as … to personal actions for safety”. 

Tomas et al [164] hypothesised and tested, using structural equation modelling, a seven  component model of the factors that cause accidents (Figure 2.6).  As well as the factors being identified, the model represents the inter-relationships between the factors.  In their paper, Tomas et al also, by the use of structural equation modelling, calculated the relative strength of relationships between the factors.  The model was tested on three Spanish sample groups.  The number of personnel in the groups was 123, 182 and 124.  The testing indicated that the model fitted two of the three samples well.  The third sample was not well modelled as indicated by poor model fit indices. 

Figure 2.6 – Tomas et al [164] model of accident causation

Cox and Cox [28] suggested a model of attitudes to safety (Figure 2.7) based upon employees’ attitudes toward  four levels of objects,  namely,  hardware (ie safety hardware and physical hazards), software (ie rules and procedures, legislation, safety management and policy), people (ie all classes of people involved, such as workers, supervisors, management, safety committees, specialists, authorities, unions) and risks (ie risky behaviour and its regulation).  According to Cox and Cox, when talking about attitudes to safety, the objects of these attitudes could always be classified into one of the four levels of objects.   

Figure 2.7 – Cox and Cox [28] model of safety attitudes

After performing a review of various cultural models, Guldenmund [68] has summarised his conclusions within a three-layer model that consists of an outer layer, middle layer and core.  An outline of each layer, together with examples, is provided in Figure 2.8.    All of the accident models surveyed during the literature review had one common thread in that they all consider workers’ attitudes and perceptions as precursors to accidents.  Margolis [111] found that workers’ attitudes toward safety are directly related to managerial attitudes towards safety.  The majority of the models reviewed tended to focus upon those factors directly associated with safety.  It appears that research to date has paid little attention to non-safety related organisational factors such as job satisfaction, job security, and others contained within the AstraZeneca Focus 2002 survey.   

Figure 2.8 – Guldenmund’s [68] summary of cultural levels

The HSE [88] developed a seven-component three-layer model of safety climate.  The first layer of the model is labelled the information exchange level and includes three factors, namely, policy awareness, involvement and communication.  The three factors in the first level measure employees’ attitudes regarding health and safety information, their involvement in planning for health and safety issues and the level of direct communications about safety.  The second layer of the model is labelled central affective level and contains two factors, namely, perceived supervisor competence and perceived management commitment.  The two factors in the second level measure the attitudes of the workforce regarding the perceived competence and commitment of management.  The third layer of the model is labelled manifest and peripheral variable level and includes two factors, namely, general safety behaviour and job satisfaction.  The behavioural factor measures workers’ attitudes toward issues such as procedural transgression, rule bending and taking chances.  The job satisfaction factor measures the workers’ attitudes regarding how satisfied they are with their job.  By application of structural equation modelling, the HSE went on to establish the relative strength of the relationships between the factors within the model.   

2.8 – AstraZeneca metrics

2.8.1 - Introduction

AstraZeneca regularly measures the attitudes and satisfactions of all its employees and the SHE performance of all of its sites.  Section 2.8.2 summarises the way in which AstraZeneca measures the attitudes and satisfactions of its employees.  Section 2.8.3 summarises the way in which it measures and monitors site SHE performance.

2.8.2 - AstraZeneca focus surveys

AstraZeneca recognises that employees’ attitudes and satisfactions can strongly influence its ability to meet business objectives.  In an attempt to measure attitudes and satisfactions, AstraZeneca has undertaken three ‘Focus’ surveys, conducted at two-yearly intervals commencing in 2002.  The surveys were developed and administered in conjunction with an external consultancy, International Survey Research (ISR) [44] .  This research uses the data generated by the 2002 survey but there was no involvement in either the design or administration of the survey itself.  Previous to the research contained within this thesis, only the Focus 2000 responses have been subjected to formal analysis; in this case structural equation modelling was used to model innovative organisational climate [74] .  Although the AstraZeneca Focus 2000 and 2002 [6] surveys contain some common questions, the majority of them differed.  At the time of writing this thesis the Focus 2004 question set and survey responses were not available to incorporate into this project.  The 2002 Focus survey measured employees’ attitudes, satisfactions and perceptions in ten factor areas.  The mechanism by which the questions were assigned to the climate factors is unknown to AstraZeneca personnel.  It is not known if factor analysis has been performed to confirm the validity of a question being allocated to a specific climate factor.  The lack of information regarding assignment of Focus 2002 survey questions to factors is unimportant, as the factors are not used in the work detailed in Chapters 4, 5 and 6.  The factor areas toge