Chapter 6 – Summary discussion, conclusions and recommendations for further work  

6.1 – Introduction 

The purpose of this chapter is to summarise the conclusions of this project and to record recommendations for further work.  

6.2 – Summary of work done 

The project started with a review of the available literature.  The review included an examination of the current state of knowledge in the following areas:  

 The AstraZeneca Focus 2002 survey was compared to previous organisational culture measurement surveys.  The comparison confirmed that analysis of the Focus 2002 survey responses would allow measurement of AstraZeneca’s organisational culture. 

AstraZeneca sites in the UK, SE and US were then selected as candidates for statistical analysis.   

The UK, SE and US 2002 AstraZeneca lagging SHE performance indicators were reviewed.  The significant-injury frequency rate (SIFR) was chosen as the preferred lagging SHE performance indicator to be correlated with AstraZeneca’s organisational culture. 

The responses to the Focus 2002 survey questions and GSHE annual reporting data were pre-treated before performing basic data analysis.  The purpose of the basic analysis was to understand the underlying structure and correlations within the data and to see if further statistical analysis was warranted.  The results of the basic analysis identified a number of statistically significant correlations between the Focus 2002 survey responses and SIFR at the AstraZeneca UK, SE and US sites.    

PCA modelling was performed on the Focus 2002 question responses at the national level.  PCA models were produced for the UK, SE and US sites.  The strong bivariate correlations derived between Focus 2002 question responses and SIFR performance coupled with the ability to produce good PCA models led to the conclusion that PLS, PLS-DA and SIMCA analysis was warranted.  PLS, PLS-DA and SIMCA modelling techniques were applied to the Focus 2002 response and SIFR data to:  

Section 6.2.1 summarises the work associated with establishing a correlation between AstraZeneca’s organisational culture and significant-injury frequency rates.  Section 6.2.2 summarises the work associated with the discrimination of the AstraZeneca UK, SE and US sites.  Section 6.2.3 summarises the organisational culture factors that discriminate SIFR performance at the UK, SE and US sites.

6.2.1 – Establishment of a correlation between organisational climate and lagging SHE performance indicators

Bivariate Pearson correlation coefficients between the Focus 2002 question responses and the significant-injury frequency rates for the AstraZeneca UK, SE and US sites were calculated in Section 3.5.4.  Statistically significant Pearson correlation coefficients were found between several Focus 2002 question responses and significant-injury frequency rates.  None of the Focus 2002 question responses (Pearson) correlating with significant-injury frequency rates at the UK and US sites were found to be directly associated with safety.  The majority of the Focus 2002 question responses that (Pearson) correlated with significant-injury frequency rates in Sweden were found not be directly associated with safety.  The predominance of non-safety related responses correlating with injury rates suggests that ‘safety culture’ does not exist as a distinct entity.  Researchers and industry looking to improve SHE performance should not therefore attempt to extricate ‘safety culture’ from ‘organisational culture’.  The results of this research indicated that those within industry wishing to improve SHE performance would do better by taking a holistic approach through addressing general management issues such as effective communication, leadership, recognition etc.  These findings support the ‘address the basics and safety will look after itself’ message that many a manager has voiced within industry.     

Only one Focus 2002 question response (55a) was found to positively (Pearson) correlate above the level of statistical significance with SIFR in more than one nation.  The lack of a common set of Focus 2002 question responses correlating with SIFR performance in the UK, SE and US was reflected in the results of the PLS modelling described in Chapter 4.  It was shown that PLS was unable to model SIFR performance in more than one nation.  There are several possible explanations for this observation, which include:

The design of this project was such that it was not possible to identify which, if any, of the above explanations might account for the observation.   The responses to several Focus 2002 questions were found to be strongly positively correlated with safety in one nation and strongly negatively correlated in another.  The fact that an organisational climate metric is positively correlated with SIFR in one nation and negatively correlated in another has significant implications for multi-national companies; for example, a company may introduce a global program to improve SHE performance in several nations only to find that it causes a decrease in SHE performance in some of them.   PLS models were built for the AstraZeneca UK, SE and US sites.  Each of the PLS principal components can be visualised as a dimension of organisational culture that is related to SIFR performance.  The work detailed in Chapter 4 represents the first time that several dimensions of organisational culture have been simultaneously correlated with a lagging SHE performance indicator.  The resultant PLS models were shown to be able to model the SIFR performance at the UK, SE and US sites.   A review of the SIFR variation accounted for by each PLS model principal component suggested that:

It is not known whether injury rates are driven by organisational culture, or that injury rates drive organisational culture.  It was speculated that the relationship between organisational culture and injury rates is reciprocal, the precise weighting and balance of one with the other continuously and dynamically changing dependent upon a complex inter-relationship with internal and external influences.  Further work to resolve the dynamics and interplay of organisational culture and SHE performance is suggested as being of value to industry and academia.   The ability to model SIFR performance without consideration of the site relative risk profile information suggests that, for AstraZeneca UK, SE and US sites, the levels of inherent risk at each site do not significantly affect SIFR performance.  It is therefore suggested that AstraZeneca high-hazard sites control their risks such that the residual risks are comparable with lower hazard sites.   Chapter 4 concluded that the use of PLS modelling of organisational culture survey data as a robust leading SHE performance indicator would be possible if the following two statements are confirmed:

·        Organisational culture is stable with time, and;

·        The relationship between organisational culture and the lagging performance indicator of interest is stable over time. 

The predictor X block of the PLS models consisted solely of the Focus 2002 question responses.  The literature review indicated that several previous researchers in the field of organisational culture have advocated the ‘triangulation approach’ to measure organisational culture.  Proponents of such an approach are of the opinion that measurement of several attributes allows a more accurate assessment of organisational culture compared to the measurement of a single attribute.  The inclusion of other factors may therefore improve the already excellent predictive ability of the PLS models.  The technique of PLS modelling is well able to deal with a vast number of predictor variables and would therefore be ideally suited to the inclusion of additional organisational culture attributes.   The ‘triangulation approach’ may also be helpful in assessing an organisation’s overall SHE performance.  Instead of using a single SHE performance metric, several SHE performance metrics could be measured simultaneously.  This approach would give a better appreciation of the overall SHE performance of the organisation or unit.  The work performed in Chapter 4 used only one lagging SHE performance indictor as a metric of site SHE performance.  Section 2.9.3.2 explained that PLS is not only able to deal with several X block (predictor) variables, it is also capable of dealing with several Y block (predicted) variables.  One can therefore imagine creating a PLS model with many predictor and predicted variables.  In such a model one could select an area in ‘n’ dimensional Y space that corresponded to desirable SHE performance.  The PLS model could then be used to identify those predictor attributes that were associated with the desirable SHE performance.  Taking this idea one stage further, the Y block could contain many business outputs such as productivity, down time, SHE performance, absenteeism, etc. to give an overall organisational performance indicator.  It is recommended that further research work be carried out in this area to fully exploit the potential of PLS modelling.  

6.2.2 – Discrimination of AstraZeneca sites from different nations

 Chapter 5 summarised the application of PLS-DA and SIMCA methodologies to the Focus 2002 question response data.  Both PLS-DA and SIMCA methodologies were shown to be able to discriminate between AstraZeneca UK, SE and US sites.   For the Focus and GSHE SIFR 2002 data, the ability of PLS-DA was found to be more consistent than SIMCA with respect to its ability to discriminate the UK, SE and US sites.  It was suggested that, as a result of the ease of model interpretation, PLS-DA is likely to be the preferred technique for discrimination when the number of classes to be discriminated is five or less, and when the number of classes exceeds five, SIMCA is the most appropriate technique.     It was hypothesised that the UK, SE and US sites are distinguishable as a result of the combination of organisational and national cultural differences.  The work associated with this thesis was unable to determine whether organisational or national culture dominated the way in which the employees responded to the Focus 2002 survey.  The ability to discriminate nations based upon the responses to an organisational climate survey has significant implications for multi-national companies.  Multi-national companies should not compare or try to interpret differences between organisational climate survey responses of sites in different nations until the relative importance of national and organisational culture are known and understood. 

6.2.3 – Discriminatory factors 

Both PLS and PLS-DA methodologies were shown to be able to identify those Focus 2002 question responses that discriminate, at a national level, AstraZeneca sites with desirable SHE performance from those with less desirable SHE performance.  This ability should allow AstraZeneca to construct targeted strategies aimed at improving SHE performance.  PLS-DA and SIMCA methodologies have also been shown to be able to discriminate AstraZeneca UK, SE and US sites based upon the responses to the Focus 2002 survey.  National cultural differences are suggested to be a potential hindrance to building PLS models that are able to predict SIFR performance in more than one nation.  The ability of PLS-DA and SIMCA methodologies to identify factors that discriminate nations may facilitate multi-national companies to design organisational culture surveys that minimise national cultural differences, and hence increase the likelihood of producing a multi-national PLS model. 

Comparison of the PLS-DA and PLS model questions that load highly on SIFR confirms expectations.  When the PLS-DA a priori classes are based upon the SIFR and the PLS model Y data are the SIFR, one would expect the questions that discriminate sites to be the same.  For example, based upon the score and loadings plots of models SIFR-PLS-DA-UK2 (Figures 5.10 and 5.11) and PLS-UK2 (Appendix 13.9 and 13.10) one can make the following observations:  The poorer SIFR performing sites UK2 and UK6 are characterised by higher than average responses to the Focus 2002 questions in the lower right of the loadings scatter plot of model PLS-UK2.  UK2 and UK6 are characterised by higher than average responses to the Focus 2002 questions in the left hand side of SIFR-PLS-DA-UK2 loadings scatter plot.  All of the Focus 2002 question responses characterising UK2 and UK6 found in model PLS-UK2 are also present in model SIFR-PLS-DA-UK2.  A similar observation is made with the SE data.  The poorer performing sites SE1 and SE9 are characterised by higher than average responses to Focus 2002 question responses in the lower left quadrant of the score scatter plot of model SIFR-PLS-DA-SE1 (Figure 5.17).  The same sites are characterised by higher than average responses to Focus 2002 questions in the upper right quadrant of the loadings scatter plot of model PLS-SE2 (Appendix 13.28).  The questions that characterise sites SE1 and SE9 in the PLS-SE2 model are also found to characterise SE1 and SE9 in model SIFR-PLS-DA-SE1.   

It is noted that fewer Focus 2002 question responses are retained in the PLS models compared to the corresponding PLS-DA models.  A possible explanation of this observation is that the PLS model was optimised to best represent the X and Y data blocks as well as the relationship between them.  During the model optimisation process, the question responses that did not account for the variance in the X block and the relationship between the X and Y block were removed.  In PLS-DA there is only one requirement in that the principal components are oriented to best discriminate the sites.  Because only one condition needed to be fulfilled, more questions are retained within the PLS-DA model compared to the PLS model. 

6.3– Contribution to knowledge

 Research into the associations between organisational culture and SHE performance is still in its infancy.  Research appears to have concentrated very much upon the social science aspects of organisational culture rather than on any practical industrial applications.    The results of this research project have significantly contributed to existing knowledge of organisational culture in the following areas. 

 Over the past three decades, many social scientists have suggested the relationships between organisational culture and SHE outcomes such as injury accidents.  Although many hypotheses have been made, few researchers have attempted to build predictive models.  The few models that have been produced by previous researchers were poor in their predictive ability.  It is suggested that the poor predictive ability of these models is due to the failure of the researchers to address the multi-dimensional nature of organisational culture.   

In addition to the novel application of PLS-DA, SIMCA and PLS techniques to organisational culture data, the work contained in this thesis reports the first time that statistical models have been built that simultaneously correlate several dimensions of organisational culture with a SHE outcome.  Because the models are able to account for, on average, over 90% of the SHE outcome variation, they should be of real practical use to industry.  The findings detailed within this project have enormous potential to proactively reduce accidents and their associated costs.  The techniques detailed within this thesis provide industry with a leading SHE performance indicator that is capable of modelling injury performance and allows the identification of those organisational climate factors that influence it.  The application of PLS-DA, PLS and SIMCA techniques to organisational culture and SHE outcome data may be considered as one of the most significant advances in proactive SHE management in recent years. 

The modelling work detailed in Chapters 4 and 5 was based upon a large combined UK, SE and US Focus data set of 23,728 responses.  Although it is clearly advantageous to possess large data sets, they are not necessary to make use of the PLS, SIMCA and PLS-DA techniques.  Using smaller data sets is possible so long as the researcher takes into consideration the possible presence of sub-cultures.  Modelling problems may be encountered if sample sizes are small (e.g. 10 individuals) or distinct sub-cultures are present within the group.  The use of the mean organisational climate responses, as used in this research, clearly would not be appropriate in these circumstances.  Alternative modelling strategies could, however, be used, and include PCA pre-treatment of the data to identify the sub-cultures that could then be analysed separately.   

9 UK, 10 SE and 5 US sites were modelled in this project.  Industrialists need not have several sites to take advantage of the PLS, SIMCA and PLS-DA techniques.  Single site companies could collect organisational climate and SHE outcome data from several different functions and/or departments.  The data could then be subjected to PLS, SIMCA and PLS-DA techniques to identify those organisational culture factors that relate to desirable and undesirable SHE outcomes.   

 Industry should not restrict the application of PLS-DA, PLS and SIMCA techniques to the linking of organisational culture with SHE outcomes.  Conceptually, the techniques could be used to establish links with other business outcomes such as productivity, absenteeism, staff turnover and quality metrics.  The number of potential applications of the techniques is vast.    

 The contribution to knowledge summarised above provides industry with the necessary confidence and information to enable it to implement local initiatives to exploit organisational climate metrics to improve SHE and potentially many other business outcomes. 

6.4 – Recommendations for further work

 Throughout this research project, questions have arisen which lie outside its scope, but which, if answered, would significantly advance the understanding, knowledge, and practical application of organisational culture metrics within industry.  The recommendations for further work are as follows:

·        Investigate the stability of organisational culture with time. 

·        Investigate whether the correlation between organisational culture and significant-injury frequency rates is stable over a period of several years. 

·        Perform research to find out the extent to which accident rates drive organisational culture and visa versa. 

·        Examine the relationship between organisational climate metrics and other business outputs such as productivity and quality metrics. 

·        Further develop the vision of a single tool that takes several different types of organisational measurements into consideration in order to predict SHE performance.

·        Investigate the integration of organisational culture metrics into a relative risk ranking methodology, which takes into account several facets associated with SHE performance. 

·        Perform PLS, PL-DA and SIMCA analysis on subsequent AstraZeneca Focus surveys in order to find a common set of question responses that correlate with safety in the United Kingdom, Sweden and the United States. 

·        Explore the usefulness of artificial neural networks in revealing the relationships between organisational climate metrics and lagging SHE performance indicators.

·        Explore whether national culture dominates over organisational culture in multi-national climate surveys. 

·        Promote the usefulness of the measurement of organisational culture throughout AstraZeneca. 

·        Collect organisational culture data from personnel who have experienced an accident and those who have not.  Perform discriminant analysis on the information in order to identify the factors that are associated with those who have accidents.

·        Investigate the relationship between the degree of risk a person is subjected to and the likelihood of that person being involved in an undesired SHE outcome.  

·        Investigate the relationship between organisational sub-cultures and the likelihood of an individual belonging to the sub-culture having an accident.

·        Incorporate the ‘behavioural’ and ‘safety management system’ elements of Cooper’s [26] model of safety culture into a PLS model.