OFFSHORE TECHNOLOGY REPORT 2000/108Secure Site · reported to the Offshore Division (OSD) in the Hazardous Installations Directorate of the Health and Safety Executive. Over the

HSEHealth & Safety

Executive

Multivariate analysis of injuries data

Prepared by the University of Liverpoolfor the Health and Safety Executive

OFFSHORE TECHNOLOGY REPORT

2000/108

© Crown copyright 2001Applications for reproduction should be made in writing to:Copyright Unit, Her Majesty’s Stationery Office,St Clements House, 2-16 Colegate, Norwich NR3 1BQ

First published 2001

ISBN 0 7176 2131 6

All rights reserved. No part of this publication may bereproduced, stored in a retrieval system, or transmittedin any form or by any means (electronic, mechanical,photocopying, recording or otherwise) without the priorwritten permission of the copyright owner.

This report is made available by the Health and SafetyExecutive as part of a series of reports of work which hasbeen supported by funds provided by the Executive.Neither the Executive, nor the contractors concernedassume any liability for the reports nor do theynecessarily reflect the views or policy of the Executive.

HSEHealth & Safety

Executive

Multivariate analysis of injuries data

The University of LiverpoolDepartment of Mathematical Sciences

M&O BuildingLiverpool

L69 3BXUnited Kingdom

HSE BOOKS

2

Executive Summary Incidents that cause, or could cause, injury and that satisfy the RIDDOR 95 1 criteria, are reported to the Offshore Division (OSD) in the Hazardous Installations Directorate of the Health and Safety Executive. Over the three years − 1st April, 1996, to 31st March, 1999 − 1045 injuries to personnel were reported on offshore installations. The records of these 1045 injuries are analysed using statistical methods. For one or two variables, chi-square tests of uniformity and independence are used. For three or more variables, multivariate methods are necessary: in particular, log-linear models with Poisson errors are applied. Time series methods are used to identify possible relationships between oil prices and safety. The frequencies of injuries are compared with the daily closing prices of a barrel of Brent oil, available on the web-site of the International Petroleum Exchange. For two years the numbers of injuries in intervals of 30 days have been monitored three-monthly to assess whether there have been any changes in the rate of reported injuries. In this report monitoring is done for weekly intervals, and the two schemes are compared. Results, significant at the 5% level in the analyses, are considered to represent circumstances that cannot be described adequately by randomness, but statistical significance implies neither practical significance nor cause. The physical interpretations of significant results are presented as bullet points at the end of each sub-section. The analyses in Annex 1, and in the first part of Annex 2, address possible relationships between the activities and operations being undertaken at the time of each injury, the part of the body affected, the type of installation, the time of the day and of the year, the age and employment status of each injured person, and the severity of each injury. Employment status is dichotomised as an employee of a main company or a contracted out-worker: employee and contracted out-worker, respectively. The severity of injury is also dichotomised: Fatal or Major injuries form one category, F&S, and more than three days the other, 3-day. The main findings are as follows:

• The percentage of injuries categorised as F&S increased significantly from 14.3% in 96/97, to 21.0% and 22.6% in 97/98 and 98/99, respectively. (A major cause of the significance was the percentages of F&S injuries among contracted out -workers: 11.6% in 96/97, 17.9% in 97/98 and 22.0% in 98/99.)

• The percentage of injuries classified as F&S was greater on Mobile installations than on Fixed installations - 23.4% compared with 16.7%.

• 31.7% of injuries between 10am. and 11am. were categorised as F&S compared with 19.2% for the remaining 23 hours.

• 52.7% of all injuries occurring between 4pm. and 5pm. were categorised as Maintenance, but only 30.4% of injuries were attributed to Maintenance.

• 56.8% of all injuries between 4am. and 6am. were involved in Drilling Operations, but only 31.2% of all injuries were attributed to Drilling Operations.

1 Reporting of Injuries, Diseases and Dangerous Occurrences Regulations, 1995.

3

• 12.5% of injuries categorised as Domestic/Catering Operations occurred between 6am. and 7am., compared with 2.2% for all other injuries.

• 27.3% and 33.0% of injuries categorised as Slips/Trips/Falls and Lifting/Crane Operations, respectively, were F&S to be compared with 12.0% for the remaining categories of Broad Incident Type.

• Injuries affecting the Torso were less likely to be F&S than injuries to other parts of the body: 6.5% compared with 22.7%.

• 33.0% of injuries categorised as Falls from Height were F&S injuries compared with 17.9% for all the remaining categories of Kind of Accident.

• 17.0% of injuries to contracted out-workers were reported as F&S compared with 22.5% of injuries to the employees.

The main findings from the monitoring analyses are as follows:

• An average of 11.3 injuries occurred per week in the first three weeks of August, 1996, compared with 6.6 per week for the remaining 153 weeks. This excessive number of injuries does not repeat in August for the other two years.

• An average of 5.4 injuries occurred per week in the period from November, 1998, to March, 1999, compared with 6.9 per week in the previous two and a half years.

• There is no reason to change the monitoring interval from 30 days to 7 days . A weak relationship between safety and the price of oil is found in the time series analyses. The change in the price of oil from week to week is compared with the injury frequencies and a discernible relationship is identified 18 weeks later: the larger the price increase the fewer are the number of injuries 18 weeks later and vice versa. As the relationship was not supported by other expected results, and as the relationship is weak, it is considered a statistical artefact for the time being. In the analyses of the days of the week when injuries occur, there are two points of note:

• The data are consistent with injuries occurring at the same rate on each of the seven days of the week.

• The Poisson distribution with mean 0.95 closely fitted the numbers of injuries per day. This technical point could be helpful for future statistical analyses.

Clearly some noted points reflect the population at risk: if it is known that few contracted out -workers work on Mobile installations, one might expect the percentage of injuries that are F&S to be higher for employees than for contracted out-workers. More population data are needed to separate the effects of Type of installation from Employment status. Successful implementation of the Vantage project - Offshore Passport and Personnel Tracking System - could provide additional relevant population data, which would allow many more questions to be addressed.

4

Contents

Executive Summary 2

Main Report

1. Purpose and Approach 6

2. Injury data 7 3. Methodology and Terminology 8

4. Time Series Analyses 10

5. Conclusions

5.1 Methodology 11 5.2 Results 12

References 14

Annex 1 1. The Data and Methodology

1.1 The Database 15 1.2 Ana lyses 16 1.3 Methodology 17 1.4 Computation 18

2. Analyses and Results

2.1 Year by Season by Severity by Operation 19 2.2 Year by Installation Type by Severity 20 2.3 Hour by Operation, Part of Body, Severity and BIT 20 2.4 BIT by Operation, Age, Activity and Kind of Accident 21 2.5 Kind of Accident by Operation, Age and Activity 22 2.6 Operation by Employment Status and Activity 23 2.7 Year by BIT, Kind of Accident, Severity, Shift-times,

Employment Status and Age 23 Appendices

1. Tables of Year by Season by Severity by Operation 25 2. Tables of Year by Installation Type by Severity 31 3. Tables of Hour by Operation, Part of Body, Severity and BIT 32 4. Tables of BIT by Operation, Age, Activity and Kind of Accident 36 5. Tables of Kind of Accident by Operation, Age and Activity 38 6. Tables of Operation by Employment Status and Activity 40 7. Tables of Year by BIT, Kind of Accident, Severity, Shift-times, 41

Employment Status and Age

5

Annex 2 1. Analyses and Commentary for re-grouped variables

1.1 Introduction 44 1.2 Year by Severity by Operation 45 1.3 Hour by Operation and BIT 45 1.4 BIT by Operation, Age, Activity and Kind of Accident 47 1.5 Kind of Accident by Operation, Age and Activity 48 1.6 Operation by Employment Status and Activity 48 1.7 Year by BIT, Kind of Accident and Age 48

2. Additional Analyses and Commentary

2.1 Severity by Part of Body, Age, Operation, Activity, BIT and Kind of Accident 50

2.2 Occupation by Employment Status by Type of Installation 51 3. Time Series Analyses.

3.1. Comparison of two monitoring schemes: week and month 52 3.2. The relationship between Injuries and the Price of Oil 53 3.3. Days of the Week and Number of Injuries per Day 55

Appendices 1. Tables of Year by Severity by Operation 57 2. Tables and Figures of Hour by Operation and BIT 58 3. Tables of BIT by Operation, Age, Activity and Kind of

Accident 61 4. Tables of Kind of Accident by Operation, Age and Activity 63 5. Tables of Operation by Employment Status, Activity and Age 65 6. Tables of Year by BIT, Kind of Accident and Age 66 7. Tables of Severity by Part of Body, Age, Operation, Activity,

BIT and Kind of Accident 67 8. Tables of Occupation by Employment Status and Type of

Installation 69 9. Figures for Monitoring 70 10. Figures and graphs for the relationship between Injuries and

the Price of Oil 73 11. Tables of Days of the week and the Numbers of Injuries per day 77

6

1. Purpose and Approach Incidents that cause, or could cause, injury to personnel and that satisfy criteria specified in RIDDOR 95, are reported to the Offshore Division (OSD) in the Hazardous Installations Directorate of the Health and Safety Executive. The data recorded for each injury include descriptions of the injury and the operations being undertaken at the time of the injury, as well as some personal details about the injured person, and details about the offshore installation. This report is concerned with the statistical analysis of data for incidents in which an injury occurred. The purposes of this report are to identify questions that can be addressed with the available data and suitable analytical methods, and then to analyse the data for some of the questions identified. Some examples of the questions that can be addressed with these data are as follows: Do injuries occur more frequently at some times of the year than at other times? Do injuries occur at particular times of the day? Do injuries where the given Operation is 'Production' tend to occur at different times of the day from injuries where the given Operation is 'Maintenance'? Do different age groups have different types of injuries? Can a relationship between the price of oil and the numbers of injuries be identified? For most questions one or two variables are analysed, but other questions require the simultaneous consideration of three or more variables. The approach is similar for all analyses involving one or two variables: a hypothesis is made and tested and the results are interpreted in physical terms. For questions with more than two variables, the purpose of each analysis is to identify combinations of factors that occur more frequently than can be adequately described by chance. After identifying simple models that fit the data, the physical meaning of the models and results are interpreted. The above types of analysis are reported in Annexes 1 and 2. Where there are few recorded injuries, categories are combined to allow inferences to be made. Some combinations of categories used in analyses, reported in Annex 1, were amended to be more coherent. The analyses were re-done with the amendments, and these results are presented in Annex 2. After consideration of the questions addressed in Annex 1, a list of further questions to be addressed was prepared. The additional analyses were done using the same statistical methods and the results are reported in Annex 2.

The dates when injuries occur are considered in two different analyses, which are reported in Annex 2. Firstly the dates of injuries are monitored to assess whether injury frequencies changed in a non-random way: cusums are used. Secondly possible relationships between the price of a barrel of oil and injury frequencies are addressed using time series methods: in particular, autocorrelations and cross-correlations are calculated and interpreted. These analyses are reported in Annex 2.

7

2. Injury data In this study 1045 records of injuries that occurred over three years, 1st April 1996 to 31st March 1999, are analysed. The record of each reported injury was stored in an Excel spreadsheet by OSD. The spreadsheet with the attributable data excluded was transferred to Liverpool University for analysis. Each record of an injury refers to one person injured on one occasion: a person, injured on two separate occasions, has two injury records. The analyses use 12 of the variables in the records. Three variables describe the person: Age, Employment Status and Occupation. Four variables describe the injury: Severity of injury, Date, Time and Part of Body affected. Four variables describe the operations and activities being undertaken: Operation, Activity, Kind of Accident and Broad Incident Type (or BIT). The type of installation − Fixed or Mobile − is the twelfth variable included in the analyses. In the description of injuries, Severity of injury is categorised as 'fatal', 'major', or 'over-three-days'. Throughout this report, the first two categories are combined to form F&S and the third category is called 3-day. Injury records are excluded from analyses in which they have missing values, but these records are included in other analyses. The variables describing the operations and activities being undertaken have few missing values: Operation and Activity have two missing values, BIT has three, and Kind of Accident has 20 missing values, among the 1045 records. More values are missing for Age and Employment Status: the largest number of missing values from an analysis with these two variables is 33. Occupation has over 400 missing values. Analyses that include Occupation must be treated with caution, but the missing values are not a problem for the other analyses. Because of the inevitable subjectivity, a reportable injury to one person might not be seen as reportable by another person. Knowledge, and consistent interpretation, of the guidance documents are important to reduce subjectivity. As every management is likely to prefer a lower injury rate than the average, some installations might succumb to downgrading the severity of some injuries and not reporting others. Injuries might be more likely to be reported at some installations than at others: somebody with a sprained ankle could be moved to other duties for the remainder of a two weeks tour, and the injury is not reported even though according to the guidance documents it should be. Even if management policy is in accord with the legal requirements, forms might not be completed for a 3-day injury, because the responsible person is ‘saving time’: correct management policy is not properly implemented, and RIDDOR 95 legislation is not being complied with. Differences in the interpretation of the threshold levels can occur at the management level, at the level of recording and inadvertently. These differences introduce errors and biases that cannot be adequately identified: for example, a significant result in an analysis could be caused by differences in the interpretation of the guidance notes. It follows that the results must be interpreted with great care.

8

3. Methodology and Terminology Statistics is concerned with making deductions, often called inferences, from samples of data. Statistical analysis can be considered a formal, or objective, way of weighing the evidence. The analytical methods and the terminology used in this report are described briefly in this section.

It is necessary at the start of any analysis of data to distinguish between two types of variable, qualitative and quantitative . Qualitative, sometimes called categorical or nominal, variables cannot be characterized by a numerical quantity. Most variables in the analyses in this report are qualitative. A two-way contingency table is a classification of two qualitative variables, and can be used to assess what, if any, is the relationship between the two variables. The data for many analyses consist of frequencies with which one of a list of categories occurs, or consist of the frequencies in a two-way contingency table. Before each analysis involving one or two categorical variables, a null hypothesis is formulated: the word null is often omitted. To test whether injuries are equally likely in any hour, the hypothesis of uniformity is formulated. To assess whether there is a relationship between the categories and the hour of injury, the hypothesis would be that the category and the hour of injury are independent. In these analyses the expected frequency for each category, or pair of categories, is then calculated on the assumption that the hypothesis is true.

A measure of the discrepancy between the observed and expected frequencies for each entry in the one- or two-way contingency tables is calculated, and the overall discrepancy is accumulated in the chi-squared statistic. The larger the value of this statistic, the larger the discrepancies and the less likely is the hypothesis to be true. The significance level is the probability (or chance) of obtaining the calculated value, or a more extreme value, of the calculated statistic, which was usually a chi-squared statistic in these analyses. A small significance level would imply that the hypothesis is unlikely to be true. The significance levels, p say, are presented in this report in four bands: *** represents highly significant, p<0.001; ** represents very significant 0.001≤p<0.01; * represents significant, 0.01≤p<0.05; and NS represents Not Significant, p≥0.05. For those analyses where three or more variables are considered simultaneously, the approach is different. Firstly a model, which is a set of assumptions, is defined. Models are compared with the data until one or more models are found that fit the data. The adequacy of any fit is based upon the deviance , which has a similar role as the chi-squared statistic above. Significance levels for the variables, and for the interactions or relationships between them, can be obtained. They are presented succinctly in deviance tables. The fits of the models are then interpreted physically. Log-linear models are used for the analyses of three or more variables. As well as being appropriate for these types of analysis, the software is available for their

9

implementation. The Genmod procedure in the statistical system SAS has been used here. Options in this procedure give the deviances, the observed and expected frequencies, and the discrepancies which are needed to interpret the sources of significance in the results. After performing each statistical analysis, the discrepancies between observed and expected frequencies that cannot be explained adequately by randomness, are identified and interpreted physically. The frequencies of injuries in fixed width intervals of time are obtained. Instead of treating each frequency separately, successive values are combined in cusums, which facilitate the identification of dates when there could have been a change. The cusums are plotted in cusum charts and the results of these procedures are summarised in Manhattan diagrams .

Time series methods can be applied to assess the possibility of relat ionships within and between data sets such as the injury frequencies and the price of oil. Autocorrelations are calculated to identify possible relationships within a single time series: observed values that are a fixed width time interval apart are compared in a formal way using standard methods. Cross-correlations are used to identify relationships between pairs of two or more time series. Plots of autocorrelations and cross-correlations can be easily and usefully interpreted. When analysing data, a parameter can be estimated. The confidence interval provides an interval in which the 'true' value of the parameter is likely to lie. The idea is extended in the time series analysis in sub-section 3.2 of Annex 2. If there are no relationships within a time series, then the autocorrelation would be close to zero: confidence intervals can be formed so that autocorrelations outside the limits of the interval indicate when there are possible relationships. Confidence intervals are applied similarly for cross-correlations.

10

4. Time Series Analyses Data that are collected, or recorded, sequentially in time form a time series. The frequencies of injuries in consecutive weeks or months, and the price of a barrel of Brent oil at the close of trading on each day, are time series. As such series occur regularly in widely differing situations, many analytical methods have been developed. In this report two methods are applied: cusums in sub-section 3.1 of Annex 2 and autocorrelations and cross-correlations in sub-section 3.2 of Annex 2. The methods used in the third and final sub-section of Annex 2 are not specifically for time series, but one of the results could have implications if time series are applied again in the future. The frequencies in fixed width time intervals, obtained from the injury records, can be used to assess whether there has been a change in the rate, and type, of recorded injuries. The cusum approach can be applied to identify firstly possible changes in the frequencies, and secondly intervals when a frequency is so different from others that it cannot be explained adequately by randomness. Every three months for two years this approach has been used to monitor the frequencies of reported incidents on offshore installations. The interva ls were of thirty days width. In sub-section 3.1 of Annex 2 the frequencies in weekly intervals are used, and the results are compared with those for intervals of 30 days. Possible relationships between the price of oil and the injury frequencies are addressed in sub-section 3.2 of Annex 2. The oil prices and the numbers of injuries per week form two time series. Autocorrelations are formed and plotted to assess whether or not there is any relationship between frequencies a fixed number of weeks away, the so-called lag: this is done for lags of 1, 2, 3…30. Autocorrelations are applied to detect possible seasonal effects. The plot of the autocorrelations together with the confidence interval indicates that there is no 'seasonality' in the frequencies of injuries, and that the approach using cross-correlations can proceed. The plot of the autocorrelations for the prices time series indicates that the cross-correlations approach cannot be applied directly to these data. However, the plot of the autocorrelations for the time series of the price differences from week to week, often called price changes, shows that the series of price changes is amenable to analysis using cross-correlations. In sub-section 3.2 of Annex 2, the injury frequencies and the price changes a fixed number of weeks previously, again called the lag, are compared using cross-correlations. Lags of 1, 2, 3...30 are considered. A weak relationship is identified. Because the relationship is weak and is not accompanied by other expected results, the relationship is considered a statistical artefact for the time being.

11

5. Conclusions 5.1. Methodology For one variable the analyses are tests of uniformity or of goodness-of-fit of the Poisson distribution. The significance levels are obtained from a chi-squared statistic: see sub-section 1.3 of Annex 1 for the details. For two variables the purpose of each analysis is to assess whether the observed frequencies in a two-way contingency table, such as Table 1 in sub-section 1.2, indicate a relationship. The significance level for the test of independence is obtained from a chi-squared statistic. For three or more variables, tests of independence for two variables can be extended, but the results are usually not helpful. The purpose of each analysis is to find one or more models that fit the data and to interpret the fitted models. Log-linear models with Poisson errors have been found to be appropriate and informative.

Well-established time series methods − in particular, moving averages, autocorrelations and cross-correlations − are applied. The changes in oil prices from one week to another were found to be suitable for these analyses.

Cusums have been used for monitoring safety on offshore installations for two years. The frequencies of injuries in intervals of 30 days have been monitored. The possibility of using weekly intervals is addressed here. The variables recorded affect the questions that can be addressed. The analyses in this report address questions about the hour and time of year when injuries occurred; the types of personnel injured; the operations and activities being undertaken when injuries occurred; the severity of injuries; the part of the body injured. Population data are not used in the analyses here. If the Vantage programme - Offshore Passport and Personal Tracking System - is successful, then population data will become available allowing more questions to be addressed. Epidemiological methods can then be applied. The statisticians in this investigation had little knowledge of the practices on offshore installations. Standard statistical methods were used to analyse the injury data. Some hypotheses are well-defined prior to analysis: for example, the frequencies of injuries are homogeneous around the clock. The identification, and subsequent interpretation, of models that fit the data is more contentious, because some patterns can be expected to arise by chance and because the statistical methods that are applied were developed to test whether data conformed with specified hypotheses. This approach is sometimes called data trawling. However, results are reported here if the significance level is small or if the results are repeated. If the statistical approach is continued, the next stages could be as follows:

a) to continue the three-monthly monitoring process and to include trend analyses, b) to analyse the data with questions posed by the oil industry: for example, analyses

concentrating on the effects of age might be considered, and

12

c) to incorporate population data into the analyses. 5.2. Results The results given in the Executive Summary are explained in this sub-section. The explanations are given in the order of the bullet points in the Executive Summary. The yearly frequencies for the two categories of Severity of injury are given in Table 1.1 of Appendix 1 in Annex 1. From the summary of the analysis, Table 1.1a, the independence hypothesis is rejected at the * level indicating that the Severity of injuries changes over the years. This is explained by noting that the percentage of injuries that are F&S increases from 14.3% in 96/97 to 21.0% in 97/98 and 22.6% in 98/99. The yearly frequencies for the analysis of Severity of injuries and the type of installation are given in Table 2.1 in Appendix 2 of Annex 1. The Severity/Type interaction is significant at the ** level. Table 2.2 is presented to show clearly the reason for the significance: 23.4% of injuries on mobile installations are F&S compared with 16.7% on fixed installations. (The Year/Severity interaction is again significant at the * level.) The hourly frequencies for the dichotomy of Severity of injury are given in Table 3.3 of Appendix 3 in Annex 1. The test of independence is rejected at the *** level mainly due to 31.7% of injuries between 10am and 11am classified as F&S compared with only 18.1% at other times. There could be a physiological explanation or perhaps more dangerous tasks are undertaken during this hour. The hourly frequencies of injuries for grouped categories of Operation are given in Table 2.1 in Appendix 2 of Annex 2. The test of independence is significant at the *** level: the hourly distributions of injuries attributed are different for 'Domestic/Catering', 'Drilling', and 'Maintenance'. The periods of great differences are 4pm to 5pm when many injuries are categorised as 'Maintenance', 4am to 6am when many injuries are categorised as 'Drilling' and 6am to 7am when many injuries are categorised as 'Domestic/Catering'. To a great extent these observations reflect work practice. 52.7% of the injuries between 4pm and 5pm were classified as 'Maintenance', but only 30.4% of all injuries are categorised as 'Maintenance'; alternatively, 9.2% of injuries categorised as 'Maintenance' occurred between 4pm and 5pm compared with 3.6% for all other categories of Operation. The percentages for 'Drilling' and 'Catering/Domestic' given in the Executive Summary are calculated similarly using Table 2.1. Injuries are classified by BIT and Severity in Table 7.5 of Appendix 7 of Annex 2. The test of independence is significant at the *** level mainly due to 27.3% and 33.0% of injuries categorised as Slips/Trips/Falls' and 'Lifting/Crane Operations', respectively, being classified as F&S. Only 12.0% of the injuries for the other categories of BIT combined are categorised as F&S.

Injuries are classified by Part of body and Severity in Table 7.1 of Appendix 7 of Annex 2. The test of independence is significant at the *** level mainly due to 6.5% of injuries categorised as 'Torso' be ing classified as F&S: 22.7% of injuries affecting other Parts of the body are classified as F&S.

13

Table 7.6 of Appendix 7 of Annex 2 consists of injuries classified by Kind of Accident and Severity. The test of independence is significant at the *** level mainly due to 5.3% of injuries categorised as 'Handling/Lifting/Carrying' and 33.0% of injuries categorised as 'Fall from height' being classified as F&S. Injuries are classified by Employment Status, Year and Severity in Table 7.4 of Appendix 7 of Annex 1 andsummarised in Table 7.4a. To fit the data a model must include a Year/Severity interaction, or an Employment Status/Severity interaction: these interactions are significant at the * level. The first interaction is consistent with other results. The second interaction can be explained as follows: 17.0% of injuries to out-workers and 22.5% to 'employees' are classified as F&S. Injuries have been monitored three-monthly for two years. The purposes of monitoring were to identify dates when the frequencies changed in a non-random way and to identify individual frequencies that differed substantially from the rest. Frequencies in intervals of 30 days were used in the analyses. In sub-section 3.1 of Annex 2, analyses with intervals of 30 days and of one week are described: figures are given in Appendix 9 of Annex 2. There are three main results. There was a change point in the frequencies of reported injuries in November, 1998: the injury frequencies reduced from 6.9 per week to 5.4 per week. There was a blip in the injury frequencies for three weeks in August, 1996: the injury rate was 11.3 in each of these three weeks to be compared with 6.6 for the remaining 153 weeks. The numbers of injuries were not large in either of the other two Augusts so no general inference can be made. There is no reason to change the monitoring interval from 30 days to 7 days. The closing price of oil on each day of trading is available on the web-site of the International Petroleum Exchange. The prices for Brent oil on the last day of trading of each week were used in the analyses. After preliminary analyses the differences between consecutive weeks were used together with the frequencies of injuries in the analyses. The analyses are described in sub-section 3.2 of Annex 2, and the figures are given in Appendix 10 of Annex 2. There was a week relationship between the changes in the weekly oil prices and the injury frequencies 18 weeks later: a reduction in the price of oil is related weakly to an increase in the injury frequencies 18 weeks later. As the relationship is not accompanied by results that might be expected, it should be treated as a statistical artifact. The final analyses consider the days of the week when injuries occurred and the numbers of injuries per day. The analyses are reported in sub-section 3.3 of Annex 2 and the tables are given in Appendix 11 of Appendix 2. There is insufficient reason to reject the hypothesis that injuries are equally likely to occur on each day of the week, but it is perhaps worth noting that the smallest number of injuries is on Sunday. The Poisson distribution and the numbers of injuries occurring per day fit closely. Both these results could be relevant in future analyses.

14

References The references for the body of the report and for both Annexes are listed below. Agresti, A. 1996. An introduction to Categorical data Analysis. New York, Wiley. Bland, M. 1996. An introduction to Medical Statistics. New York, Oxford. Downham, D.Y. 2000. Case Control Study. HSE Report OTO 2000/082. SAS. SAS Institute Inc. SAS Campus Drive, Cary, North Carolina27153, USA. UK Office: SAS, Withington House, Henley Road, Marlow, Bucks SL72EB

15

Annex 1

1. The Data and Methodology 1.1. The Database The records of all reported injuries occurring over the three-year period, 1st April 1996 to 31st March 1999, were transferred in non-attributable form to Liverpool University for the statistical analyses. A total of 1045 recorded injuries, which were stored in three Excel spreadsheets, were analysed. Each record included the age of the injured person, the activities being undertaken at the time of the injury, the part of body injured, the occupation and employment status of the injured employee, together with the date and time of the injury and whether the installation was fixed or mobile. The following data were recorded for each injury: Severity of Injury Incident Number Date of Incident Time of Incident Number Injured Nature of Injury Part of Body Sex Age Employment Status Occupation Operation Activity Kind of Accident Broad Incident Type (BIT) Fixed or Mobile Installation An investigation of the data revealed a few minor problems. (i) There were duplicates with more in the last year - April 1998 to March 1999 - than in the previous two years. These duplicates were identified within Excel, confirmed with OSD as duplicates and removed from the files at Liverpool University. (ii) A few missing values were found. If one or more variables in an analysis were not recorded for an injury, then that injury was excluded from the analysis; of course, that injury could be included in other analyses. (iii) Occupation was missing for many injuries so a disk with more complete information for Occupation was prepared by OSD. The data were corrected accordingly for the analyses. Criteria that must be satisfied for some statistical tests to be valid are given in sub-section 1.3 of this annex. To meet these criteria, categories were combined, sometimes on an ad hoc basis without consultation by Liverpool University. Some combinations were unsatisfactory and OSD suggested other combinations: the amended analyses are reported in Section 1 of Annex 2. For some analyses it was necessary to group ages and dates and to define seasons. Age groupings conform to other OSD reports. Dates are grouped into three years: 96/97 represents the period 1stApril, 1996, to 31stMarch, 1997: 97/98 represents 1stApril, 1997, to 31stMarch, 1998: 98/99 represents 1stApril, 1998, to 31stMarch, 1999.

16

April, May and June, is season 1: July, August and September, season 2: October, November and December, season 3; January, February and March, season 4. Seasons 1, 2, 3 and 4 correspond roughly to Spring, Summer, Autumn and Winter, respectively.

1.2. Analyses The analyses in this annex are concerned with both methodology and interpretation of the results. After discussions between OSD and Liverpool University personnel, it was agreed to address the following general questions: • Do injuries occur more frequently at certain times of the year? • Do injuries occur more frequently at certain hours of the day? • What types of injury occur at what time of year? • What types of personnel are injured? • Are some types of work more prone to injury than others? To address each question, one or more contingency tables were formed. Table 1 is a two-way contingency table, or a cross-tabulation, for Operation and Year. Each entry in the table is a frequency: for example, in 19 of the 364 injuries in year 96/97 the worker was involved in Production. Sometimes cross-tabulations of three or more variables were needed. Contingency tables such as these provide the input for the statistical analyses. Table 1. Contingency Table of Operation by Year Year Operation

96/97 97/98 98/99 Total

Production 19 20 24 63 Drilling 114 117 93 324 Maintenance 99 85 71 255 Diving 7 8 4 19 Construction 12 11 10 33 Deck Operations 44 69 55 168 Domestic 25 24 23 72 Modification of Plant 16 9 4 29 Transport 2 4 9 15 Other 26 27 12 65 Null or missing 0 2 0 2 Total 364 376 305 1045 Contingency tables needed to address the above questions are specified in the following list. Each row in the list may correspond to several analyses for different subsets of the variables. The sub-sections that include the analyses, and the appendices with additional information, are specified in the list overleaf.

17

Number Analysis Sub-section Appendix 1 Year by Season by Severity by Operation 2.1 1 2 Year by Installation Type by Severity 2.2 2 3 Hour by Operation 2.3 3 4 Hour by Part of body 2.3 3 5 Hour by Severity 2.3 3 6 Hour by BIT 2.3 3 7 BIT by Operation 2.4 4 8 BIT by Age 2.4 4 9 BIT by Activity 2.4 4 10 BIT by Kind of Accident 2.4 4 11 Kind of Accident by Operation 2.5 5 12 Kind of Accident by Age 2.5 5 13 Kind of Accident by Activity 2.5 5 14 Operation by Employment Status 2.6 6 15 Operation by Activity 2.6 6 16 Year by BIT 2.7 7 17 Year by Kind of Accident 2.7 7 18 Year by Severity by Shift times 2.7 7 19 Year by Severity by Employment Status 2.7 7 20 Year by Severity by Age 2.7 7

1.3. Methodology Table 1 is a sub-table of Analysis 1 and is used to illustrate the analysis of a two-way contingency table. A chi-squared test is applied to test the hypothesis that there is no association, or relationship, between the recorded Operations and the year: that is, the test assesses whether the pattern of Operations among the injured employees changes from year to year. Expected frequencies are formed that are based on no association between these two variables. If for a particular entry O and E represent the observed and expected frequencies, respectively, then the statistic

E

EO −

measures their discrepancy. The sum of the squares of this expression yields the chi-squared statistic. If two variables are unrelated, then the chi-squared statistic should be small. The significance level (see, Section 3 of the main report) is obtained from the value of the chi-squared statistic. If there is found to be a relationship, then the values of the above expression are examined to identify the source of the relationship.

18

The logic of the chi-squared is based upon large samples and so the methods are not suitable for small samples. The following criteria for the chi-squared test to be valid have been adopted in this report: at least 80% of the expected frequencies exceed 5 and all the expected frequencies exceed 1. These criteria are sometimes called Cochran’s criteria (see, for example, Bland, p.230, 1996). Two injured employees do not have Operation recorded and are omitted from the calculations: the modified total is 1043. The number of Production injuries in 96/97 is 19; the corresponding expected frequency is calculated by (63x364)/1043=21.9866. The measure of the discrepancy is

E

EO − = (19-21.9866)/4.689

= –0.6369 to four decimal figures. The expected frequencies comply with Cochran’s criteria, so it is unnecessary to combine any categories. The chi-squared value is 27.79, which is not significant at the usually reported levels for this size of table: the numbers of rows and columns in the table determine the so-called ‘degrees of freedom’, which can be considered to be the statistician's representation of the size of the table. No overall difference in the pattern of Operations is identified from year to year, even though the 9 injuries in the ‘Transport’ category of Operation in Year 98/99 cannot be explained wholly by randomness. If the chi-squared value had been significant, then the discrepancies for the individual entries would have been examined. 1.4. Computation The data were manipulated firstly within Microsoft Excel, and the statistical analyses were done using the statistical package, SAS. In SAS, “driver” programs are written that utilise statistical and computational procedures: proc Freq was used to form the contingency tables and proc Genmod was used for the multivariate analyses. Although proc Freq includes useful statistical procedures − in particular, for the chi-squared statistic and tests of independence − the output of proc Genmod includes additional information that is useful for explaining the analyses. Proc Genmod is used in the analysis of the two-way contingency tables in analyses 3, 4, 5,….,17 in the list. The analysis is similar, but more complicated, for three- and four-way tables. The multivariate analyses, using proc Genmod , were based on the log-linear model with Poisson errors (Agresti, 1996). The methods of analysis for such tables are the same as for unmatched Case-Control studies (Downham, 2000). Deviance is used for testing the fit of a model and follows approximately a chi-squared distribution. Tables of the deviances are formed to aid analysis; further details are given in Appendices 7 and 8 of Downham (2000). The significance levels of the tests are summarised in the deviance tables.

19

2. Analyses and Results

Tables were formed using proc Freq and analysed using proc Genmod . When Cochran’s criteria were not satisfied, the categories were combined and the adopted categorical combinations are given. Deviance tables are presented for three-way tables. The contingency tables and the deviance tables are given in the appendices. 2.1. Year by Season by Severity by Operation The contingency tables for Analysis 1 in the list (sub-section 1.2) are given in Appendix 1. Severity of injury is dichotomised as F&S or 3-day. In the deviance tables, Year, Season, Severity of injury and Operation are referred to as Y, S, Sev and O, respectively. The four -dimensional contingency table – Year by Season by Severity by Operation – was formed. There were many small or empty entries, and Cochran’s criteria were not satisfied. Four three-way tables and three two-way tables are given in Appendix 1. In the analysis of two of the three-way tables, Cochran’s criteria were not satisfied and so categories of Operation were combined: 'Diving', 'Construction', 'Modification of Plant' and 'Transport' were combined with the 'Other' category of Operations. In Table 1.1, 1045 injuries are classified by Year, Season and Severity of injury. The deviances are given in Table 1.1a. Even though the independence hypothesis is not rejected at the usual levels, the Year/Severity interaction is significant at the* level. The proportions of F&S compared with 3-day injuries differ significantly over the three years: in particular, there are fewer F&S injuries in Year 96/97 than might be expected. Injuries are classified by Year, Season and Operation in Tables 1.2 and 1.2a; Table 1.2b is the deviance table. In Table 1.2a the categories of Operation are grouped as described above. The distribution of Operation does not differ between seasons nor between years. The injuries classified by Year, Severity and Operation are given in Tables 1.3 and 1.3a: in Table 1.3a categories of Operation are grouped as described above. The hypothesis of independence is not rejected. As in the analysis of Table 1.1, the Year/Severity interaction is significant at the * level.

The injuries are classified by Season, Severity of injury and Operation in Tables 1.4 and 1.4a. The hypothesis of independence is not rejected. There is no evidence of any interactions. The three variables - Season, Severity and Operation - are considered in pairs. Tables 1.5, 1.6 and 1.7 in Appendix 1 are the contingency tables, Season by Severity, Season by Operation and Severity by Operation. The three chi-squared statistics were small and no significant results were found. No further details are given for brevity.

• The percentage of injuries that are F&S is significantly less in Year 96/97 than in 97/98 and 98/99: 14.3% compared with 21.0% and 22.6%, respectively.

20

2.2. Year by Installation Type by Severity Analysis 2 from the list in sub-section 1.2 is now considered. The detailed tables are given in Appendix 2. The installation type is dichotomised as 'Fixed' or 'Mobile'. In the deviance tables Year, Installation type and Severity are referred to as Y, FM and Sev, respectively. The three-way frequency table of Year by Installation Type by Severity is given in Table 2.1 and the corresponding deviances are given in Table 2.1a. As the hypothesis of independence is rejected at the ** level (p=0.004), there is strong evidence that the variables are inter-related. In addition to the relationship between Year and Severity of injury, found in the previous sub-section, there is an even stronger relationship between Severity of injury and Installation Type.

To focus on the second result, the two-way table of Installation type by Severity was formed (Table 2.2). The proportion of F&S injuries is smaller on 'Fixed' installations than on 'Mobile' installations at the ** level: 16.7% of injuries on 'Fixed' installations are recorded as F&S compared with 23.4% on 'Mobile' installations. • Injured personnel on a mobile installation are more likely to have an F&S

injury than injured personnel on a fixed installation: 23.4% of injuries on mobile installations are recorded as F&S compared with 16.7% on fixed installations.

2.3. Hour by Operation, Part of Body, Severity and BIT Classifications of the hour of injuries by the Operation being undertaken, the Part of Body affected, the Severity of injury and the Broad Incident Type (BIT) are considered in this sub-section. Four two-way tables are formed and the four tables – Tables 3.1, 3.2, 3.3 and 3.4 – are given in Appendix 3. These analyses correspond to 3, 4, 5 and 6 in the list in sub-section 1.2. In the four tables the hours are not combined: the intervals are always 0-1,1-2,…..,23-24. With so many categories for time, the tables have empty cells and so some categories of variables other than hour are combined.

From Table 3.1 the peak hour for injuries is between 10am. and 11am. when 7.9% of injuries occur and this is closely followed by 2pm. to 3pm. when 7.3% occur. In the analysis of Operation by Hour, Table 3.1, 'Drilling', 'Maintenance' and 'Deck Operations' are kept separate: the other categories of Operations are combined. There is a strong association between Operation and Hour. Between 4pm. and 5pm., the number of injuries in the 'Maintenance' category is significantly high: Operations classified as 'Maintenance' account for 50.9% of injuries between 4pm and 5pm compared with 24.4% overall. 56.8% of injuries occurring between 4am and 6am are among employees undertaking 'Drilling' Operations, compared with an overall value of 31.2%.

21

The categories of Part of Body are grouped as 'Head', 'Torso', 'Upper limb', 'Lower limb', and 'Other' (Table 3.2). No association was found between Part of Body and Hour. Scrutiny of the frequencies and the discrepancies revealed nothing noteworthy. The contingency table of Hour by Severity of injury, Table 3.3, yields a non-significant chi-squared value. Although overall there is no evidence of association, further scrutiny revealed one observed frequency differing significantly from its expectation: 31.7% of injuries between 10am. and 11am. are recorded as F&S compared with 19.2% overall, which represents 18.1% for the remaining hours. The contingency table of Hour by BIT (Table 3.4) yields a chi-squared value that is significant at the 1% level. The significance cannot be attributed to individual entries in the table and the findings are less strong than for Operation. However, the main reasons for the significance are excessive numbers injuries attributed to 'Objects falling' and 'Crane operations' between 2am. and 3am. and attributed to 'Use of tools and machines' between 7am. and 8am. and between 1pm. and 2pm. • An excessive number of injuries is categorised as 'Maintenance' between 4pm.

and 5pm.: 50.9% of injuries between 4pm. and 5pm. are attributed to 'Maintenance' but only 24.4% of all injuries are classified as 'Maintenance'.

• Over half the injuries (56.8%) between 4am. and 6am. are among employees undertaking 'Drilling' Operatio ns, compared with 31.2% overall.

• There is a greater proportion of F&S injuries between 10am. and 11am. (31.7%) than at other times (18.1%).

• An excessive numbers of injuries attributed to ‘Objects falling’ and ‘Crane operations’ occur between 2am. and 3am.: 6.6% of such injuries occur between 2am. and 3am. compared with 2.9% for all other injuries.

• 9.1% of injuries attributed to ‘Use of tools and machines’ occur between 7am. and 8am. and 9.9% between 1pm. and 2pm compared with 3.9% and 4.9%, respectively, for all other injuries.

2.4. BIT by Operation, Age, Activity and Kind of Accident Possible relationships between the Broad Incident Type (BIT) and Operation, Age, Activity and Kind of Accident are addressed here: analyses 7, 8, 9 and 10 in the list in sub-section 1.2. The frequencies are given in four two-way contingency tables in Appendix 4. Operation uses the original codes. The groupings for Age lead to violations of Cochran’s criteria, so wider age-groups were formed. For Activity, 'Erecting' and 'Working at Height' are combined to form ‘Height’; 'Using equipment', 'Climbing' and 'Walking' are combined to form category ‘A’; the remaining categories form category ‘B’. The categories of Kind of Accident are also combined: the given labels should be sufficient to explain their meanings.

22

Table 4.1 is a contingency table of Operation and BIT. In the test of independence the chi-squared value is significant at the 0.1% level. The main sources of the significance are within specific pairs of categories: 'Domestic' with 'Slips/Trips/Falls'; 'Drilling' with 'Objects falling' and 'Crane Operations'; 'Drilling' with 'Use of tools and Machines.' Table 4.2 is a contingency table of BIT and Age. The chi-squared value is significant at the ** level. An excessive number of injuries of employees <31 years are categorised as 'Handling Goods': 29% of injuries in this age group are reported as 'Handling Goods' compared with 23% of the total number of injuries. Personnel older than 40 years have fewer injuries than expected in the Objects falling category: the difference between 12.3% and 16.3%, could be explained by the types of job done by older personnel. Associations between BIT and Activity are addressed in Table 4.3 and those between BIT and Kind of Accident in Table 4.4. The chi-squared values are both significant at the 0.1% level, reflecting the similarity of these variables: for example, BIT has a category 'Slips/ Trips/ Falls' and Kind of Accident has a grouped category called ‘Fall’.

• 11.2% of injuries categorised as 'Slips/Trips/Falls' are categorised as 'Domestic Operations', but 'Domestic Operations' represent only 6.8% of all injuries.

• 'Drilling' Operations have high numbers of injuries from 'Objects Falling/ Crane Operations' (21.9%) and 'Use of Tools/Machines' (15.7%), compared with other injured employees (13.2% and 9.8%, respectively).

• Personnel younger than 31 years have 34.6% of injuries attributed to 'Handling goods' compared with 20.2% for the other age groups.

2.5. Kind of Accident by Operation, Age and Activity Possible relationships between Kind of Accident and three variables Operation, Age and Activity are addressed in three two-way contingency tables in Appendix 5 - Tables 5.1, 5.2 and 5.3. The categories of Kind of Accident, Operation, Age and Activity have been grouped as in sub-section 3.4. The analyses in this sub-section correspond to analyses 11, 12 and 13 in the list in sub-section 1.2. The analysis of Kind of Accident by Operation uses the frequencies in Table 5.1. The chi-squared statistic is significant at the 0.1% level. The number of injuries attributed to drilling and categorised as ‘Struck by’ is highly significant. The 'Other' category provides much of the rest of the dependency. The relationship between Kind of Accident and Age is addressed in Table 5.2. The chi-squared statistic is significant at the * level. The two sources of the significance are excess numbers of injuries among employees less than 31 years old categorised as ‘Struck by’ and for employees in the age group 51-65 years categorised as ‘Fall’. The relationship between Kind of Accident and Activity is addressed in Table 5.3. The chi-squared statistic is significant at the *** level. Not surprisingly, much of the association is explained by employees 'Working at Height' having 'Falls'. There are significantly more injuries than predicted in the Kind of Accident category ‘Struck by’

23

for grouped Activity B, which includes 'Welding/Burning', 'Handling Hazardous materials', 'Cleaning' and 'Electrical work', 'Inspection/Examination'. • 34.8% of injuries attributed to 'Drilling' Operations are categorised as 'Struck

by' compared with 21.7% for all other Operations. • 32.6% of injuries to employees <31 years old are categorised as 'Struck by'

compared with 23.5% for the other age groups. • 14.3% of injuries to employees older than 50 years are attributed to 'Falls' but

only 8.0% for the other age groups. 2.6. Operation by Employment Status and Activity Possible relationships between Operation and two variables, Employment Status and Activity, are addressed here. Two two-way tables are presented in Appendix 6. The categories for Operation are the same as coded originally. These analyses correspond to analyses 14 and 15 in the list in sub-section 1.2. Employment Status in Table 6.1 is as originally coded. For the chi-squared test the dichotomy, employee and contracted out-worker, is adopted. The chi-squared statistic is significant at the *** level. 40.5% of the injuries to employees are attributed to 'Drilling' Operations compared with 31.6% of injuries to contracted out-workers. This would be expected if 'Drilling' is contracted out less frequently than other Operations. Activity is grouped into 'Height', A and B as above. The chi-squared statistic is significant at the *** level. The main source of the significance is in the ‘Other’ category of Operation, which indicates that further redefinition is necessary.

• 40.5% of the injuries to main company employees are attributed to 'Drilling' Operations compared with 31.6% of injuries to 'Out -sourced' workers.

2.7. Year by BIT, Kind of Accident, Severity, Shift-times, Employment Status and Age Two-way and three-way tables, corresponding to analyses 16, 17, 18, 19 and 20. In the list in sub-section 1.2 are considered now. The tables are given in Appendix 7. The categories for BIT – Broad Incident Type – are those used in sub-section 2.3. The frequencies are shown in Table 7.1. The chi-squared statistic is not significant at the * level. Scrutiny of the tables and the statistics revealed nothing.

The categories for Kind of Accident are the same as those in sub-section 2.4. The frequencies are given in Table 7.2. The chi-squared statistic is significant at the ***

24

level. There are significantly high numbers of injuries in the ‘Handling’ category in 96/97 and in the 'Other' category in 98/99. In these analyses the hour when an injury occurred is dichotomised: 6pm. to 6am. represents the night-shift and 6am. to 6pm. the day-shift. Ta ble 7.3 consists of the injuries classified by Year, Shift and Severity. The chi-squared value is significant at the ** level. The significance can be attributed to a significantly lower number of F&S injuries in 96/97. In Tables 7.4 and 7.4a, Year, Se verity and Employment Status are considered. No simple models fit the data: the chi-squared values are significant at the * level. Models that include a Year/Severity interaction or a Severity/Status interaction fit the data. The first source of significance is the smaller percentage of F&S injuries in 96/97 than in the other two years: 14.3% of injuries in 96/97 are recorded as F&S compared with 21.1% in 97/98 and 22.0% in 98/99. The second source of significance is that 22.5% of injuries to employees are recorded as F&S, but only 16.8% of injuries to contracted out-workers are recorded as F&S. Year, Severity and Age are considered in Table 7.5. The test of independence yields a non-significant chi-squared value. • 27.7% of injuries are categorised as 'Handling' in 96/97 compared with 18.1%

in 97/98 and 18.0% in 98/99. • 16.8% of injuries to contracted out-workers are reported as F&S compared

with 22.5% of injuries to employees. • A major contribution to the previous bullet point is that only 11.8% of injuries

to contracted out-workers in 96/97 are categorised as F&S.

25

APPENDICES for Annex 1. Annex 1, Appendix 1. Tables of Year by Season by Severity by Operation There are four three-way tables, three two-way tables and two modified three-way tables in this Appendix. The deviance tables are given after each three-way table. The tables correspond to Analysis 1 in the list given in sub-section 1.2 and complement the discussion in sub-section 2.1.

Table 1.1. Injuries classified by Year, Season and Severity

Year 96/97 97/98 98/99 Total Severity Season

F&S 3-day F&S 3-day F&S 3-day

Apr – Jun 12 84 18 91 14 59 278 July – Sept 14 86 14 73 19 67 273 Oct – Dec 13 77 22 73 19 55 259 Jan – March 13 65 25 60 17 55 235 Total 52 312 79 297 69 236 1045

Table 1.1a. Deviance Table for Year, Season and Severity

Year Season Sev Y*S Y*Sev S*Sev Deviance df Sig. level * NS *** - - - 21.11 17 NS * NS *** NS - - 16.98 11 NS * NS *** - * - 12.03 15 NS * NS *** - - NS 15.27 14 NS * NS *** - * NS 6.20 12 NS * NS *** NS - NS 11.14 8 NS * NS *** NS * - 7.90 9 NS * NS *** NS * NS 2.27 6 NS

The categories need no further combining, as the required criteria are satisfied. Although the data always fit the model, there is some evidence that the Severity of injury changes over the three years. Relationships between Year, Season and Operation are addressed in Tables 1.2, 1.2a and 1.2b. The row totals are omitted for space considerations. The chi-squared statistical tests are inapplicable because Cochran’s criteria are not satisfied. By combining some of the categories of Operation, see sub-section 2.1, Table 1.2a is formed from Table 1.2. The deviances of the analysis are given in Table 1.2b.

26

Table 1.2. Injuries classified by Year, Season and Operation

Year 96/97 97/98 98/99 Season Operation

1 2 3 4 1 2 3 4 1 2 3 4

Production 5 5 6 3 7 2 5 6 8 8 5 3 Drilling 25 28 30 31 34 26 35 22 19 26 28 20 Maintenance 28 24 29 18 22 23 18 22 16 21 15 19 Diving 1 3 3 0 3 0 0 5 1 2 1 0 Construction 5 3 1 3 4 5 1 1 4 0 1 5 Deck Operations 12 11 10 11 14 15 20 20 13 15 16 11 Domestic 5 8 6 6 9 8 6 1 4 8 4 7 Mod of Plant 6 9 0 1 6 1 1 1 1 1 2 0 Transport 0 2 0 0 1 1 1 1 3 2 1 3 Other 9 7 5 5 9 6 6 6 4 3 1 4 Null 0 0 0 0 0 0 2 0 0 0 0 0 Total 96 100 90 78 109 87 95 85 73 86 74 72 Table 1.2a. Injuries classified by Year, Season and modified Operation

Year 96/97 97/98 98/99 Season Operation

1 2 3 4 1 2 3 4 1 2 3 4

Production 5 5 6 3 7 2 5 6 8 8 5 3 Drilling 25 28 30 31 34 26 35 22 19 26 28 20 Maintenance 28 24 29 18 22 23 18 22 16 21 15 19 Deck Operations 12 11 10 11 14 15 20 20 13 15 16 11 Domestic 5 8 6 6 9 8 6 1 4 8 4 7 Other 21 24 9 9 23 13 9 14 13 8 6 12 Null 0 0 0 0 0 0 2 0 0 0 0 0 Total 96 100 90 78 109 87 95 85 73 86 74 72

Table 1.2b. Deviance Table for Year, Season and modified Operation Year Season Operation Y*S Y*O S*O Deviance df Sig. level

* NS *** - - - 65.05 61 NS * NS *** NS - - 60.97 55 NS * NS *** - NS - 52.51 51 NS * NS *** - - NS 45.22 46 NS * NS *** - NS NS 32.69 36 NS * NS *** NS - NS 41.15 40 NS * NS *** NS NS - 48.44 45 NS * NS *** NS NS NS 28.66 30 NS

27

Relationships between Year, Severity and Operation are addressed in Tables 1.3, 1.3a and 1.3b. The row totals are omitted. The chi-squared tests are inapplicable because Cochran’s criteria are not satisfied. Combining some categories of Operation, see sub-section 2.1, Table 1.3a is formed from Table 1.3. The deviances of the analysis are given in Table 1.3b. The interaction of Year and Severity is again significant at the 5% level. Table 1.3. Injuries classified by Year, Severity and Operation

Year 96/97 97/98 98/99 Total Severity Operation


Production 3 16 2 18 6 18 63 Drilling 19 95 27 90 19 74 324 Maintenance 10 89 19 66 20 51 255 Diving 1 6 2 6 0 4 19 Construction 0 12 3 8 2 8 33 Deck Operations 7 37 15 54 9 46 168 Domestic 4 21 3 21 6 17 72 Mod of Plant 3 13 1 8 1 3 29 Transport 0 2 0 4 4 5 15 Other 5 21 6 21 2 10 65 Null 0 0 1 1 0 0 2 Total 52 312 79 297 69 236 1045 Table 1.3a. Injuries classified by Year, Severity and modified Operation



Production 3 16 2 18 6 18 63 Drilling 19 95 27 90 19 74 324 Maintenance 10 89 19 66 20 51 255 Deck Operations 7 37 15 54 9 46 168 Domestic 4 21 3 21 6 17 72 Other 9 54 12 47 9 30 161 Null 0 0 1 1 0 0 2 Total 52 312 79 297 69 236 1045

28

Table 1.3b. Deviance Table for Year, Severity and modified Operation

Year Sev Operation Y*Sev Y*O Sev*O Deviance df Sig. level * *** *** - - - 30.05 27 NS * *** *** * - - 21.39 25 NS * *** *** - NS - 17.42 17 NS * *** *** - - NS 29.66 22 NS * *** *** - NS NS 17.08 12 NS * *** *** * - NS 21.05 20 NS * *** *** * NS - 8.81 15 NS * *** *** * NS NS 8.25 10 NS

Relationships between Season, Severity and Operation are addressed in Tables 1.4 and 1.4a. Chi-squared tests are inapplicable because Cochran’s criteria are not satisfied. Table 1.4. Injuries classified by Season, Severity and Operation

Severity F&S 3-day Total Season Operation

1 2 3 4 1 2 3 4

Production 4 1 5 1 16 14 11 11 63 Drilling 14 13 20 18 64 67 73 55 324 Maintenance 9 14 8 18 57 54 54 41 255 Diving 1 0 1 1 4 5 3 4 19 Construction 2 1 0 2 11 7 3 7 33 Deck Operations 7 8 10 6 32 33 36 36 168 Domestic 2 4 4 3 16 20 12 11 72 Mod of Plant 1 3 1 0 12 8 2 2 29 Transport 0 1 1 2 4 4 1 2 15 Other 4 2 3 4 18 14 9 11 65 Null 0 0 1 0 0 0 1 0 2 Total 44 47 54 55 234 226 205 180 1045 Table 1.4a. Deviance Table for Season, Severity and Operation Sev Season Operation Sev*S Sev*O S*O Deviance df Sig. level *** NS *** - - - 59.27 66 NS *** NS *** NS - - 53.56 63 NS *** NS *** - NS - 57.76 57 NS *** NS *** - - NS 29.10 39 NS *** NS *** - NS NS 27.59 30 NS *** NS *** NS - NS 23.38 36 NS *** NS *** NS NS - 52.41 54 NS *** NS *** NS NS NS 22.01 27 NS

29

The independence of each pair of the variables Season, Severity and Operation is addressed in Tables 1.5, 1.6 and 1.7. Each of the three values of the chi-squared statistic was small and no non-random relationships were identified. Table 1.5. Injuries classified by Season and Severity Severity Season

F&S 3-day Total

Apr – Jun 44 234 278 July – Sept 47 226 273 Oct – Dec 54 205 259 Jan – March 55 180 235 Total 200 845 1045 Table 1.6. Injuries classified by Season and Operation Season Operation

1 2 3 4 Total

Production 20 15 16 12 63 Drilling 78 80 93 73 324 Maintenance 66 68 62 59 255 Diving 5 5 4 5 19 Construction 13 8 3 9 33 Deck Operations 39 41 46 42 168 Domestic 18 24 16 14 72 Mod of Plant 13 11 3 2 29 Transport 4 5 2 4 15 Other 22 16 12 15 65 Null 0 0 2 0 2 Total 278 273 259 235 1045

30

Table 1.7. Injuries classified by Severity and Operation Severity Operation

F&S 3-day Total

Production 11 52 63 Drilling 65 259 324 Maintenance 49 206 255 Diving 3 16 19 Construction 5 28 33 Deck Operations 31 137 168 Domestic 13 59 72 Mod of Plant 5 24 29 Transport 4 11 15 Other 13 52 65 Null 1 1 2 Total 200 845 1045

31

Annex 1, Appendix 2. Tables of Year by Installation Type by Severity The tables in this appendix correspond to Analysis 2 in the list in sub-section 1.2. and complement the discussion in sub-section 2.2. FM means 'Fixed' or 'Mobile' installation and so represents the installation type. Table 2.1. Injuries classified by Year, Severity and Installation Type Year 96/97 97/98 98/99 Severity Installation

F&S 3-day F&S 3-day F&S 3-day Total

Fixed 28 206 40 204 43 145 666 Mobile 24 106 39 93 26 91 379 Total 52 312 79 297 69 236 1045 Table 2.1a. Deviance Table corresponding to Table 2.1 Year Sev FM Y*Sev Y*FM Sev*FM Deviance df Sig. level

* *** *** - - - 21.38 7 ** * *** *** * - - 12.31 5 * * *** *** - NS - 20.54 5 ** * *** *** - - ** 14.28 6 * * *** *** - NS ** 13.44 4 ** * *** *** * - ** 5.20 4 NS * *** *** * NS - 11.47 3 ** * *** *** * NS ** 4.50 2 NS

Table 2.2. Contingency Table of Severity and Installation Type

Severity Structure

F&S 3-day Total

Fixed 111 555 666 Mobile 89 290 379 Total 200 845 1045

32

Annex 1, Appendix 3. Tables of Hour by Operation, Part of Body, Severity and BIT The tables in this appendix correspond to Analyses 3, 4, 5 and 6 in sub-section 1.2. This appendix complements the discussion in sub-section 2.3. Cochran's criteria are not satisfied for the two tables involving Operation and BIT. Their categories are combined as described in sub-section 2.2 to ensure that Cochran's criteria are satisfied.

Table 3.1. Contingency Table of Hour by Operation Operation Hour

Drilling Maintenance Deck Operations

Other Total

0-1 9 1 9 3 22 1-2 9 5 6 4 24 2-3 15 4 7 10 36 3-4 4 6 5 5 20 4-5 12 2 4 2 20 5-6 13 5 2 4 24 6-7 5 5 6 14 30 7-8 13 8 7 19 47 8-9 14 18 8 15 55 9-10 11 11 6 20 48 10-11 21 22 18 21 82 11-12 17 12 13 19 61 12-13 22 5 10 8 45 13-14 23 10 8 15 56 14-15 17 27 8 24 76 15-16 13 18 12 22 65 16-17 12 28 6 9 55 17-18 10 19 8 14 51 18-19 8 9 1 14 32 19-20 9 5 4 17 35 20-21 20 11 4 11 46 21-22 13 6 7 11 37 22-23 17 10 7 6 40 23-24 17 7 2 6 32 Total 324 254 168 293 1039 There are 6 missing values.

33

Table 3.2. Contingency Table of Hour by Part of Body Part of Body Hour

Head Torso Upper Limb

Lower Limb

Other Total

0-1 0 2 13 6 1 22 1-2 1 7 8 7 1 24 2-3 2 5 18 11 0 36 3-4 2 3 7 7 0 19 4-5 0 7 9 4 0 20 5-6 2 3 12 7 0 24 6-7 2 8 7 11 2 30 7-8 1 13 26 7 0 47 8-9 5 17 21 7 4 54 9-10 4 7 25 11 1 48 10-11 4 21 40 13 3 81 11-12 5 15 28 12 1 61 12-13 3 7 19 12 4 45 13-14 2 12 27 12 4 57 14-15 5 11 31 23 7 77 15-16 5 15 25 18 2 65 16-17 5 15 18 17 0 55 17-18 5 13 17 12 3 50 18-19 3 7 9 12 1 32 19-20 2 7 14 9 3 35 20-21 5 12 21 7 1 46 21-22 5 7 16 4 5 37 22-23 3 7 15 11 4 40 23-24 4 6 14 8 0 32 Total 75 227 440 248 47 1037 There are 8 missing values.

34

Table 3.3. Contingency Table of Hour by Severity

Severity Hour

F&S 3-day Total

0-1 3 19 22 1-2 7 17 24 2-3 6 30 36 3-4 2 18 20 4-5 2 18 20 5-6 5 19 24 6-7 3 27 30 7-8 7 40 47 8-9 4 51 55 9-10 9 39 48 10-11 26 56 82 11-12 14 47 61 12-13 5 40 45 13-14 11 46 57 14-15 18 59 77 15-16 15 50 65 16-17 10 45 55 17-18 9 42 51 18-19 7 25 32 19-20 7 28 35 20-21 9 37 46 21-22 5 32 37 22-23 8 32 40 23-24 8 24 32 Total 200 841 1041

There are 4 missing values.

35

Table 3.4. Contingency Table of Hour by Broad Incident Type BIT Hour

Slips Trips Falls

Handling Goods

Objects Falling, Crane

Use of Tools and machines

Other Total

0-1 8 4 4 0 6 22 1-2 10 5 3 2 4 24 2-3 5 13 11 3 4 36 3-4 8 6 4 0 2 20 4-5 9 5 5 0 1 20 5-6 4 6 7 3 4 24 6-7 15 3 8 1 3 30 7-8 11 14 6 11 5 47 8-9 13 15 7 7 12 54 9-10 16 9 4 10 9 48 10-11 29 26 11 10 6 82 11-12 17 15 12 9 8 61 12-13 17 13 4 6 5 45 13-14 18 11 10 12 6 57 14-15 32 15 7 10 12 76 15-16 23 12 10 6 14 65 16-17 27 11 4 5 8 55 17-18 16 11 10 6 8 51 18-19 10 6 8 0 8 32 19-20 18 6 4 2 5 35 20-21 15 6 9 7 9 46 21-22 10 8 7 2 9 36 22-23 15 8 5 3 9 40 23-24 9 6 6 6 5 32 Total 355 234 166 121 162 1038 There are 7 missing values.

36

Annex 1, Appendix 4. Tables of BIT by Operation, Age, Activity and Kind of Accident The tables in this appendix correspond to Analyses 7, 8, 9 and 10 in sub-section 1.2. This appendix complements the discussion in sub-section 2.4. It is necessary to combine categories to ensure that Cochran's criteria are satisfied: the grouping of categories is done as elsewhere in Annex 1. Table 4.1. Contingency Table for Broad Incident Type by Operation BIT Operation

Slips/ Trips/ Falls

Handling Goods

Objects falling, Crane

Use of Tools and Machines

Other Total

Production 27 9 2 12 13 63 Drilling 76 89 71 51 37 324 Maintenance 95 56 26 37 41 255 Diving 1 3 2 0 13 19 Construction 12 7 7 3 4 33 Deck Operations 50 49 48 6 15 168 Domestic 40 6 2 6 17 71 Mod of Plant 12 5 6 4 2 29 Transport 7 3 1 0 4 15 Other 35 8 1 2 18 64 Total 355 235 166 121 164 1041 There are 4 missing values in Table 4.1. Table 4.2. Contingency Table for Broad Incident Type by Age BIT Age

Slips/Trips/ Falls

Handling Goods



Other Total

<31 76 79 55 28 38 276 31-40 133 75 67 46 65 386 41-50 92 64 28 31 37 252 51-65 44 10 15 15 14 98 Total 345 228 165 120 154 1012 There are 33 missing values in Table 4.2.

37

Table 4.3. Contingency Table for Broad Incident Type by Activity BIT Activity

Slips/ Trips/ Falls

Handling Goods



Other Total

Height 24 7 14 2 5 52 A 299 209 124 113 85 830 B 32 19 29 6 74 160 Total 355 235 167 121 164 1042 There are 3 missing values in Table 4.3. Table 4.4 Contingency Table for Broad Incident Type by Kind of Accident BIT Kind

Slips/ Trips/ Falls

Handling Goods



Other Total

Fall 77 1 4 2 4 88 Machinery 2 10 3 23 4 42 Struck by 21 48 114 39 44 266 Transport 0 1 1 0 1 3 Strike 16 4 6 6 8 40 Handling 19 131 25 24 25 224 Trips 182 5 1 2 6 196 Other 32 29 13 24 66 164 Total 349 229 167 120 158 1023 There are 22 missing values in Table 4.4.

38

Annex 1, Appendix 5. Tables of Kind of Accident by Operation, Age and Activity The tables in this appendix correspond to Analyses 11, 12 and 13 in sub-section 1.2, and complement the discussion in sub-section 2.5. There are 21, 30 and 21 missing values in Tables 5.1, 5.2 and 5.3, respectively. Categories are combined to ensure that Cochran's criteria are satisfied: the grouping of categories is done as elsewhere in this annex. Table 5.1. Contingency Table for Kind of Accident by Operation Operation Kind

Drilling Maintenance Deck Operations Other Total

Fall 23 24 8 33 88 Machinery 18 14 4 6 42 Struck by 113 57 51 44 265 Transport 2 0 0 1 3 Strike 11 12 4 13 40 Handling 79 59 35 52 225 Trips 38 48 30 81 197 Other 40 40 32 52 164 Total 324 254 164 282 1024 Table 5.2. Contingency Table for Kind of Accident by Age Age Kind

<31 31-40 41-50 51-65 Total

Fall 22 31 19 14 86 Machinery 10 19 9 4 42 Struck by 90 105 46 23 264 Transport 1 1 1 0 3 Strike 12 15 12 1 40 Handling 66 73 69 15 223 Trips 38 82 53 23 196 Other 37 62 44 18 161 Total 276 388 253 98 1015

39

Table 5.3. Contingency Table for Kind of Accident and Activity

Activity Kind

Height A B Total

Fall 13 66 9 88 Machinery 1 36 5 42 Struck by 15 201 50 266 Transport 0 3 0 3 Strike 3 32 5 40 Handling 7 196 21 224 Trips 9 175 13 197 Other 3 113 48 164 Total 51 822 151 1024

40

Annex 1, Appendix 6. Tables of Operation by Employment Status and Activity The tables in this appendix correspond to Analyses 14 and 15 in sub-section 1.2. They complement the discussion in sub-section 2.6. Categories of Operation and Activity are combined to ensure that Cochran's criteria are satisfied. Table 6.1. Contingency Table for Operation and Employment Status

Employment Status Ope ration

Employee Self-employed Other Total

Production 26 0 37 63 Drilling 154 3 166 323 Maintenance 76 0 178 254 Diving 7 2 1 10 Construction 10 0 22 32 Deck Operations 62 0 102 164 Domestic 17 0 55 72 Modification of Plant 3 0 26 29 Transport 2 0 13 15 Other 23 0 38 61 Total 380 5 638 1023 There are 22 missing values in Table 6.1. In the calculations of the Chi-squared statistics, 'Self-employed' was combined with the 'Other' category to form contracted ou-workers. Table 6.2. Contingency Table for Operation and Activity Activity Operation

Height A B Total

Drilling 22 276 26 324 Maintenance 16 197 42 255 Deck Operations 1 148 20 169 Other 13 209 72 294 Totals 52 830 160 1042 There are 3 missing values in Table 6.2.

41

Annex 1, Appendix 7. Tables of Year by BIT, Kind of Accident, Severity, Shift- times, Employment Status and Age The tables in this appendix correspond to Analyses 16, 17, 18, 19 and 20 in sub-section 1.2. This appendix complements the discussion in sub-section 2.7. Table 7.1. Contingency Table for Year and Broad Incident Type.

Year BIT

96/97 97/98 98/99 Total

Slips, Trips and Falls 127 117 111 355 Handling Goods 92 74 69 235 Objects falling, Crane 48 67 51 166 Use of tools and machines 43 41 37 121 Other 54 77 37 168 Total 364 376 305 1045 Table 7.2. Contingency Table for Year and Kind of Accident

Year Kind

96/97 97/98 98/99 Total

Fall 37 27 24 88 Machinery 9 19 14 42 Struck by 88 112 66 266 Transport 1 1 1 3 Strike 15 16 9 40 Handling 101 68 55 224 Trip 70 70 57 197 Other 43 63 79 185 Total 364 376 305 1045

Table 7.3. Injuries classified by Year, Severity and Shift Times. Year 96/97 97/98 98/99 Total Severity Shift


6am - 6pm 28 195 49 190 54 157 673 6pm - 6am 24 115 30 107 15 77 368 Total 52 310 79 297 69 234 1041 There are 4 missing values in Table 7.3.

42

Table 7.3a. Deviance Table for Year, Severity and Shift -times Year Time Sev Y*T Y*Sev T*Sev Deviance df Sig. level

* *** *** - - - 18.96 7 ** * *** *** NS - - 13.93 5 * * *** *** - * - 9.93 5 NS * *** *** - - NS 18.88 6 ** * *** *** - * NS 9.86 4 NS * *** *** NS - NS 13.85 4 ** * *** *** NS * - 4.91 3 NS * *** *** NS * NS 4.90 2 NS

Table 7.4. Injuries classified by Year, Severity and Employment Status

Year 96/97 97/98 98/99 Total Severity Status


Employee 25 106 36 100 25 90 382 Self-employed 0 3 0 2 0 0 5 Other 27 202 41 186 40 141 637 Total 52 311 77 288 65 231 1024 There are 21 missing values in Table 7.4. Table 7.4a. Deviance Table for Year, Severity and Employment Status Year Sev Status Y*Sev Y*S Sev*S Deviance df Sig. level

** *** *** - - - 16.06 7 * ** *** *** * - - 7.90 5 NS ** *** *** - NS - 15.53 5 ** ** *** *** - - * 11.09 6 NS ** *** *** - NS * 10.56 4 * ** *** *** * - * 2.93 4 NS ** *** *** * NS - 7.37 3 NS ** *** *** * NS * 2.65 2 NS

43

Table 7.5. Injuries classified by Year, Severity and Age

Year 96/97 97/98 98/99 Total Severity Age


<31 15 91 20 75 13 62 276 31-40 15 117 27 114 26 88 387 41-50 14 78 20 74 14 53 253 51-65 7 22 8 24 11 26 98 Null 1 4 3 11 5 7 31 Total 52 312 78 298 69 236 1045

Age is missing from 31 records of injuries. These records are included in Table 7.5 in the Null category. Table 7.5a. Deviance Table for Year, Severity and Age Year Sev Age Y*Sev Y*A Sev*A Deviance df Sig. level

** *** *** - - - 18.59 16 NS ** *** *** * - - 10.57 14 NS ** *** *** - NS - 10.14 9 NS ** *** *** - - NS 13.10 4 NS ** *** *** - NS NS 9.19 6 NS ** *** *** * - NS 3.85 10 NS ** *** *** * NS - 4.06 7 NS ** *** *** * NS NS 1.56 4 NS

44

Annex 2 1. Analyses and Commentary for regrouped variables 1.1. Introduction The structure for this section is similar to that for Section 2 of Annex 1. Commentary is given in the body of the section, and contingency and deviance tables are given in the appendices. The categories of some variables in some analyses are combined to ensure that Cochran's criteria are satisfied (sub-section 1.3 of Annex 1). The categories of a variable after grouping are intended to have physical meaning. The categories of three variables − Operation, BIT and Activity − in some analyses of Section 2 of Annex 1 needed amendment. Analyses in which the criteria were not satisfied were repeated; these analyses are reported in this section. The following six categories were formed from the original categories of Operation: Production Drilling Maintenance + Construction + Modifications to plant Deck Operations Domestic/Catering Other which includes Diving and Transport When grouping BIT, the intention had been to keep 'Loss of containment' as a separate category; but this was not possible, as only 9 injuries were attributed to this category. The following seven categories were formed from the original categories of BIT: Slips/Trips/Falls Falling objects Handling goods/materials Lifting/craning operations Hand tools Use of machinery Other The following seven categories were formed from the original categories of Activity: Portable tools Manual lifting/handling Operating plant/machinery Erecting/dismantling scaffolding + Working at height Climbing/descending Walking on same level Other

45

Kind of Accident has the same groupings as in Annex 1. In this Annex the 'Other' category of Kind of Accident includes the missing values; in each table that includes Kind of Accident, the frequencies in the 'Other' category, the total and the number of missing values differ from those given in the equivalent table in Annex 1. These groupings are adopted throughout this section.

1.2. Year by Severity by Operation

The analysis in this sub-section forms part of Analysis 1 in the list in sub-section 2.1 of Annex 1. Season is not included in this sub-section because in the analyses in Annex 1 no interaction between Season and any other variable was identified in the fits of model and data. Table 1.1 in Appendix 1 is the three-dimensional contingency table with the amended grouping of the categories of Operation. The deviances are given in Table 1.1a: Year, Severity and Operation are referred to as Y, Sev and O, respectively. As in the analysis in sub-section 2.1 of Annex 1, there is a Year/Severity interaction at the * significance level. As the bullet point is unchanged, it is not repeated here. No additional insight was obtained from these analyses. 1.3. Hour by Operation and BIT The relationships between Hour of injury and the Operation being undertaken, and the BIT, are considered here. Tables 2.1 and 2.2 in Appendix 2 are the two two-way contingency tables. These analyses correspond to 3 and 7 in the list in sub-section 1.2 of Annex 1. The distributions of the times of the different categories of these variables are compared in tests of independence. Chi-squared tests were done on Tables 2.1 and 2.2 even though the required criteria were not satisfied. There was a strong relationship between Hour and Operation, significant at the *** level, but no discernible relationship between Hour and BIT. The tests were repeated with 3-hourly intervals so that the criteria were satisfied: the meaning of the analysis was the same, although the results were different in detail. The Operation being undertaken at the time of an injury depends strongly upon the time of day, but the Broad Incident Type does not depend upon the time of day. Although the observed and expected frequencies are not given for brevity, they do reveal the sources of the statistical significance. The analyses were done blind to the meaning of the categories and interpreted afterwards. Some sources could have been predicted before the analyses, because some categories of Operation are known to be undertaken more frequently at some times than at other times. Drilling - Between 4am. and 6am. and between 11pm. and midnight, there are more injuries than expected; between 3pm. and 6pm. there are fewer injuries than expected. Maintenance - There are more injuries than expected between 4pm. and 5pm.

46

Deck Operations - There are many more injuries than expected between midnight and 1am and many fewer between 6pm. and 9pm. Domestic/catering - There are many more injuries than expected between 6am. and 7am. Although Hour and BIT were not related, the observed and expected frequencies were examined. The observed and expected frequencies do not correspond on two occasions: there are more injuries than expected for 'Lifting/Craning' between 2am. and 3am. and for 'Use of Machinery' between 1pm. and 2pm. Two categories of Operation, 'Drilling' and 'Maintenance', account for 61.6% of injuries; two categories of BIT, 'Slips/Trips/Falls' and 'Handling goods/materials', account for 56.7% of injuries. Therefore the tests comparing the temporal distributions of injuries for Operation and BIT are dominated by these pairs of categories. If 'Maintenance' is generally undertaken during the day shifts, then few such injuries are likely to occur at night. If 'Drilling' Operations are undertaken more evenly around the clock, then injuries are likely to occur more evenly around the clock. It follows that the test of independence is likely to give a significant result. The significant result could reflect this situation. To address the problem of assessing the risks of injury at different times, the numbers of employees undertaking different Operations at different hours of the day must be known or estimated. The number of individuals undertaking the Operations in the two shifts could be a first approximation. The risk for the different categories of BIT can be assessed if estimates of the numbers of employees working at any given time are available. If the Vantage project shows that SMART cards are a workable proposition, then the data become available and such estimates can be made. The frequencies of injuries categorised as 'Drilling' and 'Maintenance' were tested for uniformity over the 24 hours. The frequencies of injuries categorised as 'Slips/Trips/Falls' and 'Handling goods/material' were also tested for uniformity. If more employees are working between 6am and 6pm than between 6pm and 6am, one might expect the uniformity hypothesis to be rejected, which it is at the ** level for the four categories tested. The Severity of injury for 'Drilling' and 'Maintenance' Operations was also considered, as was the Severity of injury for 'Slips/Trips/Falls' and 'Handling goods/materials': details are not included for conciseness. The only significant result ( *** ) is that an injury categorised as 'Slips/Trips/Falls' is more likely to be F&S than an injury categorised as 'Handling goods/materials': 13.2% of 'Handling goods/materials' injuries are F&S compared with 27.3% of injuries categorised as 'Slips/Trips/Falls'. The frequencies of injuries for 'Drilling' and 'Maintenance' are plotted in Figure 2.1 in Appendix 2. For 'Drilling', the peak periods are from 10am to 3pm and from 8pm to midnight; for 'Maintenance' the peak periods are from 8am to 9am, from 10am to noon, and from 2pm to 6pm. The frequencies of injuries for 'Slips/Trips/Falls' and 'Handling goods/materials' are plotted in Figure 2.2. For 'Slips/Trips/Falls' the peak periods are

47

from 9am to 6pm and for 'Handling goods/materials' the peak period is from 10am to 11am. More information about the population is needed before making proper assessment of the risk. Although the analyses here are slightly more extensive, essentially the results endorse the bullet points in sub-section 2.3 of Annex 1. • 12.5% of injuries categorised as 'Domestic/Catering' occur between 6am and

7am, compared with 2.2% for all other Operations. • 7.7% of injuries categorised as 'Drilling' occur between 4am and 6am,

compared with 2.7% for all other Operations. • 5.2% of injuries categorised as 'Drilling' occur between 11pm and midnight,

compared with 2.1% for all other Operations. • 9.2% of injuries categorised as 'Maintenance' occur between 4pm and 5pm,

compared with 3.6% for all other Operations. • 13.2% of 'Handling goods/materials' injuries are F&S compared with 27.3% of

injuries categorised as 'Slips/Trips/Falls'. 1.4. BIT by Operation, Age, Activity and Kind of Accident Possible relationships between the Broad Incident Type (BIT) and Operation, Age, Activity and Kind of Accident are addressed: analyses 7, 8, 9 and 10 in the list. The grouped categories of Activity in these analyses are substantially different from those in the equivalent analyses in sub-section 2.4 of Annex 1. In addition, 20 injuries for which Kind of Accident is not recorded are included in the 'Other' category in this Annex. The injury frequencies are given in four two-way contingency tables in Appendix 3. Before analysis one might predict that the categories of BIT are related to those of Operation, Activity and Kind of Accident. It is not surprising, therefore, that the independence hypothesis is rejected at the *** significance level for the three tests. The independence hypothesis is rejected at the *** level for BIT and Age: this is more significant than for the original grouping of the categories of BIT. All the high frequencies in Table 3.3 of Appendix 3 arise from combinations of categories that could have been predicted beforehand: for example, injuries incurred while 'Climbing' or 'Descending' (categories of Activity) are likely to result in 'Slips/Trips/Falls' (categories of BIT). These analyses provide no bullet points additional to those given in sub-section 2.4 of Annex 1.

48

1.5. Kind of Accident by Operation, Age and Activity Possible relationships between Kind of Accident and Operation, Age and Activity are addressed in three two-way contingency tables in Appendix 4 - Tables 4.1, 4.2 and 4.3. The analyses in this sub-section correspond to analyses 11, 12 and 13 in the list in sub-section 1.2 of Annex 1. The significance levels are in the same bands as for the tests in Annex 1: the independence tests of Kind of Accident and Age are significant at the ** level and the other two tests are significant at the *** level. These analyses provide no bullet points additional to those given in sub-section 2.5 of Annex 1 and so no bullet points are given here. 1.6. Operation by Employment Status and Activity Possible relationships between Operation and two variables, Employment Status and Activity, are addressed here. Both two-way tables are presented in Appendix 5. These analyses correspond to analyses 14 and 15 in the list. Employment Status in Table 5.1 retains the original coding. For the chi-squared test the dichotomy employee and contracted out-worker, is adopted. The chi-squared statistic is significant at the *** level. More injuries occurred than were predicted for 'Drilling' Operations among employees, possibly because 'Drilling' is contracted out less frequently than other Operations. More injuries than predicted among contracted out-workers were categorised as 'Maintenance': the likely cause is that in this annex the definition of 'Maintenance' now includes 'Construction' and 'Modifications to plant'. The contingency table of Operation by Activity is given in Table 5.2. The test of independence is rejected at the *** significance level: the categories of Operation and Activity are closely related. Main sources of the dependence are more injuries than predicted for the following pairings of the Operation and Activity categories: 'Drilling' and 'Manual handling' 'Drilling' and 'Operating plant' 'Maintenance' and 'Portable tools' 'Maintenance' and 'Erecting' 'Deck Operations' and 'Manual handling' 'Domestic/Catering' and 'Walking' These relationships could have been predicted prior to analysis. As no extra insight is obtained from these results, it is unnecessary to include them as bullet points. 1.7. Year by BIT, Kind of Accident, and Age Parts of analyses 16, 17 and 20 in the list, are considered here. The three contingency tables analysed are given in Appendix 6. Each of the three tests of independence is a test of whether or not the pattern of response in each of the three years can be considered the same. The chi-squared statistics for Year and BIT, Table 6.1, and of Year and Age, Table 6.3, are not significant at the usual

49

levels: in fact, the significance levels exceed 15%. There is no evidence that the pattern of frequencies in the BIT categories changes from year to year, nor is there any evidence that the age structure of the injured personnel changes from year to year. For Year and Kind of Accident the chi-squared statistic is significant at the ** level. This implies that the pattern of Kind of Accident changes over the three years. Most of this significance is explained by 101 injuries recorded as 'Handling/Lifting/Carrying' in 96/97 compared with a predicted value of 78. • There is an excess of injuries described as 'Handling/Lifting/Carrying' in year

96/97, compared with 97/98 and 98/99: the percentages for the three years are 27.7% of all injuries in 96/97, 18.4% in 97/98 and 18.0% in 98/99.

50

2. Additional Analyses and Commentary 2.1. Severity by Part of Body, Age, Operation, Activity, BIT and Kind of Accident The Severity of injury is compared in turn with Part of Body, Age, Operation, Activity, BIT and Kind of Accident in this sub-section. The six contingency tables are given in Appendix 7. The groupings for these variables are the same as elsewhere in this Annex. Table 7.1 is the contingency table of Severity of injury by Part of Body. The chi-squared value for the test of independence is significant at the ** level. Injuries affecting the 'Torso' are less likely to be F&S than injuries affecting other parts of the body: 6.5% of 'Torso' injuries are F&S compared with 22.7% for all other injuries. When the contingency table is partitioned to exclude 'Torso' injuries, there are no other relationships. Table 7.2 is the contingency table of Severity of injury by Age group. The test of independence yields a chi-squared value that is not significant at the levels considered in this report. It should be noted however that the observed number of F&S injuries in the oldest age group exceeds the expected value: among the 51-65 years age group, 26.5% of the injuries are categorised as F&S compared with 18% for the other age groups combined. Table 7.3 is the contingency table of Severity of injury by Operation. The test of independence yields a chi-squared value that is not significant at the levels considered in this report. The observed and expected values are always very close. Table 7.4 is the contingency table of Severity of injury by Activity. The test of independence yields a chi-squared value that is significant at the * level. The significance has three sources: the 'Manual' category has fewer F&S injuries than expected and the 'Erecting' and 'Climb/Descend' categories have more F&S injuries than expected. However, none of the individual discrepancies (see sub-section 1.3 of Annex 1) of these three sources is substantial. Table 7.5 is the contingency table of Severity of injury by Broad Incident Type (BIT). The test of independence yields a chi-squared value that is significant at the *** level. The significance has three sources: the 'Handling goods/materials' category has fewer F&S injuries than expected and the 'Slips/Trips/Falls' and 'Lifting/Crane Operations' have more F&S injuries than expected. Here the discrepancies are large. Table 7.6 is the contingency table of Severity of injury by Kind of Accident. The test of independence yields a chi-squared value that is significant at the *** level. Most of the significance is explained by the category 'Handling, Lifting or Carrying': there are fewer F&S injuries in this category than expected. When the remaining categories of Kind of Accident are tested, there is significance mainly due to more F&S injuries in the category, 'Fall from Height'.

51

• Injuries affecting the 'Torso' are less likely to be F&S than injuries of other parts of the body: 6.5% of injuries of the Torso compared with 22.7% for injuries to other parts of the body.

• Injuries categorised as 'Slips/Trips/Falls' or 'Lifting/Crane Operations' are more likely to be F&S than for other categories of BIT: 27.3% and 33.0%, respectively, compared with 12.0%.

• Injuries categorised as 'Handling, Lifting or Carrying' have a smaller percentage of F&S injuries than the remaining categories, and 'Fall from Height' have a higher percentage: 33.0% of 'Fall from Height' injuries are F&S but only 17.9% for the remaining categories of Kind of Accident are F&S.

2.2. Occupation by Employment Status by Type of Installation Possible relationships between the Occupation and the Employment Status of injured personnel, and the Type of Installations are considered in this section. The Occupation is not recorded for 415 of the 1045 injuries (39.7%), and so results must be treated with caution. The categories of Occupation are reduced to the following seven for the analyses: 'Professional', Medical', 'Skilled Technicians', 'General Operators', 'Semi-skilled Operators', 'Catering/Domestic' and 'Labourers'. The three-way table, Table 8.1 in Appendix 8, is analysed. The number of Employees with injuries on 'Mobile' installations is much greater than predicted. Clearly the significance is likely to be a reflection of the Employment Status of the personnel being different on 'Fixed' and 'Mobile' installations: more 'Employees' are employed on 'Mobile' Installations than 'Out-sourced' and vice versa on 'Fixed' Installations. Before starting the multivariate analysis, it was clear that an interaction existed between Status and Type of Installation and would be needed for a model to fit the data. Scrutiny of the deviances in Table 8.2 reveals that this interaction is the most important. However, none of the models fit the data closely. The largest discrepancies, sub-section 1.3 of Annex 1, were for 'Professional' and 'Medical' employees on 'Fixed' installations: the values 14 and 9 were both much greater than the expected values of 6.8 and 3.8, respectively. The model was fitted again with the observed frequencies, 14 and 9, replaced by 7 and 4. The test of the goodness of the fit was significant at the * level. Close scrutiny of the observed and expected values and of the discrepancies reveals that the significance was an accumulation of discrepancies: significance was not due to one or two sources. If the proportion of 'Professional' and 'Medical' personnel among the employees of the main companies on the 'Fixed' installations exceeds the equivalent proportion among the 'Contracted' personnel, then the numbers of injuries are likely to exceed the expected value.

52

3. Time Series Analyses 3.1. Comparison of two monitoring schemes: week and month Short reports have been prepared every three months for the last two years. They describe the results of the statistical analysis of the numbers of injuries on offshore installations, in addition to monitoring other reportable incidents. An important part of this monitoring process is the use of cusums. The adopted cusum approach and a modification are compared in this sub-section, and the associated figures are given in Appendix 9. All injuries, F&S and 3-day combined, are used in this brief comparative study. The main purpose of monitoring is to identify dates when the frequencies changed in a non-random way, so-called change-points. A secondary purpose is to identify individual frequencies that differ substantially from the rest, so-called outliers. Cusum are used to identify possible change -points and outliers. A brief outline of the method is given in the next paragraph. The numbers, or frequencies, of injuries in fixed length time intervals are counted and their mean and variance calculated. On technical grounds, the mean and variance are likely to be similar if there have been no radical changes: if the variance is much bigger than the mean, then the overall variability might disguise change -points. The cusums are formed and plotted. The so-called V-mask is then applied to identify possible change- points and outliers. Finally, the changes are superimposed on a scatter diagram of the frequencies of injuries plotted against date: this is often called a Manhattan diagram. In the currently adopted monitoring process the intervals are of 30 days length: calendar months have different numbers of days and so are not appropriate. In this sub-section the adopted monitoring procedure is compared with one in which the time interval is one week. Injury frequencies for the weekly and 30-day intervals are plotted against the date in two scatter diagrams, Figures 9.1 and 9.2, respectively. The cusums for weekly and 30-day intervals are plotted in cusum charts, Figures 9.3 and 9.4. The Manhattan diagram for the weekly frequencies seemed to confuse rather than enlighten and so is not included in Appendix 9, but the Manhattan diagram for the analysis of 30-day intervals is presented in Figure 9.5. In the analysis of the frequencies of injuries in weekly intervals, the mean and variance are 6.74 and 7.13, which are close. The study period of 1095 days is not divisible by 7, and so injuries on the first three days are omitted: the injury frequencies for 156 weeks are included in the analyses. The cusum analysis revealed the following points, which are discussed later in this sub-section: a) three weeks starting 2ndAugust, 1996, have high frequencies; b) the final three weeks of March, 1999, have low frequencies; c) three weeks starting 1stNovember, 1996, 28thMarch, 1998, and 26thSeptember,

1998, have high frequencies.

53

In the analysis of the frequencies of injuries in 30-day intervals, the mean and variance are close, 28.56 and 27.39, respectively. The study period covers 1095 days, which is not divisible by 30. The 17 injuries occurring on the first 15 days are not included in the analysis. The analysis covers 1080 day on which 1028 injuries are recorded and so 36 30-day periods are included in the analyses. The months given in Figures 9.2 and 9.4 approximate to the date. The cusum analysis revealed the following points of interest: d) the periods when the excess of injuries cannot be explained by randomness are July 96, and April and May 97; e) in three periods fewer injuries are reported than can be explained by randomness − December 96 / January 97, May 98/ June 98 and November 98 to March 99. Firstly there is no obvious seasonal pattern in either of these analyses. Some types of seasonal pattern could have been identified in these analyses, if they had been present. If, for example, construction work often started in April, then the numbers of injuries might be expected to increase in April or May each year. Alternatively, if the workload tended to reduce around Christmas, then there could be fewer injuries at this time. No such repeating seasonal patterns have been identified. The high frequencies of injuries in three weeks of August 96, mentioned in a), are split between two 30-day intervals in the 30-day analysis. The number of injuries attributed to July 96 includes some of the injuries actually occurring in August, 96,because the 30-day intervals do not correspond exactly to calendar months. The low frequency of injuries identified in the last three weeks of March 99, mentioned in b), is part of the paucity of injuries in the four months period, November 98 to March 99 period, mentioned in e). By combining the weeks into 30-day periods the longer period of reduced frequencies is identified. This suggests that the 30-day period is to be preferred as the period for the monitoring of injuries. The three separate weeks, mentioned in c) are not identified as unusual in the 30-day analysis nor are the injury frequencies in the surrounding weeks unduly large. Consequently the frequencies are considered to be outliers. • 11.3 injuries occur per week in the first three weeks of August, 1996, compared

with 6.6 per week for the remaining 153 weeks: this difference is excessive. • 5.4 injuries occur per week in the period from November, 1998, to March, 1999,

compared with 6.9 per week in the remaining two and a half years. • There is no reason to change the monitoring interval from 30 days to 7 days. 3.2. The relationship between Injuries and Price of Oil The closing price of a barrel of oil for every day of trading is available on the web-site of the International Petroleum Exchange, IPE. The closing price of oil on each day of trading from 1stJuly, 1988, until 31stJuly, 2000, is plotted in Figure 10.1 of Appendix 10.

54

The possibility of relationships between the oil prices and the frequencies of recorded injuries are addressed in this sub-section. The weekly frequencies of injuries occurring between 1stApril, 1996, and 31st March, 1999, are compared with the closing price of oil on the same day of each week; the last day of each week is chosen here. In Figure 10.2, the weekly frequencies of injuries are superimposed onto the plot of the end-of-week prices: the ordinate represents the prices, the weekly numbers of injuries are scaled to have the same average as the prices, and the time period covered is 1stApril, 1995, to 30th September, 1999. No relationships are apparent immediately: in fact, the figure is confusing. Moving averages, based on four weeks, for the injury frequencies are superimposed on to the end-of-week oil prices in Figure 10.3. The frequencies are not scaled for this figure. Again no relationship is immediately apparent. Time series methods were developed to address problems of identifying relationships in data sets such as these. The methods adopted here are well-established and depend upon few underlying assumptions. Autocorrelations and cross-correlations are used: the former are calculated to identify possible relationships within a time series and the latter to identify relationships between pairs of two or more in a so-called stationary time series. Autocorrelations and cross-correlations are calculated and their plots can be easily and usefully interpreted. A time series is stationary if there is no systematic change in the mean (no trend), no systematic change in the variation and no cyclic variation (for example, no seasonal effect). The stationarity, or otherwise, of a series can under certain circumstances be assessed using autocorrelations, which measure the correlation between lagged values in a series of data: for each autocorrelation each value is compared with the value a fixed time interval (the lag) away from it. The autocorrelations of the prices were calculated for lags 1, 2, 3,…, 30 and are plotted in Figure 10.4. The autocorrelations of the prices decrease steadily as the lag increases. Such behaviour displays non-stationarity, and is similar to that of many reported economic time series. Under these circumstances, it is often effective to cons ider the time series of price changes: the differences between prices in consecutive weeks are now considered. Their autocorrelations are plotted in Figure 10.5. If such price changes are serially uncorrelated over time, then the autocorrelations should lie in the 95% confidence interval (−0.128, 0.128). All but the first value lie within the interval. However, this does not invalidate the hypothesis that weekly price changes behave like a stationary time series. The behaviour of the autocorrelations is consistent with a standard time series model with a moving average process of order 1. The autocorrelations of the weekly frequencies of injuries for lags 1, 2, 3,….., 30 were calculated and are shown in Figure 10.6. If injuries occur randomly over time, then they can be expected to be serially uncorrelated and the autocorrelations should lie in the 95% confidence interval (− 0.157, 0.157). With the exception of lag = 26, where the frequencies of injuries six-months apart tend to be positively and weakly related, the autocorrelations lie within the range. The frequencies of injuries are treated as though they are purely random because the only identified relationship is so weak.

55

In view of the above findings, the relationship between price changes and the frequencies of injuries are investigated by examining the cross-correlations. The following relationship can now be addressed: a large change in price in consecutive weeks might have a greater effect on safety some time in the future than a small cha nge. The relationship between two time series can be assessed using cross-correlations. The values from one series are compared with the values a fixed time interval away in the other series: the interval is called the lag. The cross-correlations between price differences and injury frequencies, calculated for different lags, are plotted against the lag in Figure 10.7. If the cross -correlations lie within the 95% confidence interval, (−0.157, 0.157), then there is insufficient evidence to infer a relationship between the two time series. Except for the cross-correlation at lag 18, all cross-correlations lie in the range. At lag 18 a small negative cross-correlation is indicated − the greater the price change in a barrel of oil, then the fewer injuries there are 18 weeks later, and the less the price of a oil, then the more injuries there are 18 weeks later. However the possibility that the significant value is due to chance cannot be ruled out since the 95% confidence interval is only valid for very long series. If there is a relationship between injury frequencies and price changes lagged by 18 weeks, then one might reasonably expect similar relationships to be apparent for adjacent lags: for example, lags of 16, 17, 19 and 20 weeks. This is not the case. Furthermore, the length of the series of injury frequencies is 156, which is somewhat small. Because of a lack of supporting evidence and because of the weakness of the relationship, this result may be considered a statistical artifact. However, the possibility of such a relationship between price changes and injury frequencies should be investigated in the future as more data become available. • A weak relationship between oil prices and injury frequencies has been found,

but at this stage it should be considered a statistical artifact: the frequency of injuries increases marginally 18 weeks after a reduction in the price of oil.

3.3. Days of the Week and Number of Injuries per Day Two topics are addressed in this sub-section. The question was asked whether there were more injuries on some days of the week than on others. Secondly, the number of injuries per day is addressed, because it could have important implications for future analyses. The two tables referred to in this sub-section are presented in Appendix 11. The numbers of injuries recorded on each of the seven days of the week are given in Table 11.1. There are fewer injuries on Sunday than on any other day. However, the test of the hypothesis that the numbers of injury are the same on each day of the week yields a non-significant result: the chi-squared value of 6.12 on 6 degrees of freedom is not significant. Thus, we cannot reject the hypothesis that injuries occur with the equal chance on each of the 7 days of the week.

56

An alternative approach to monitoring the injuries data could be to consider the time between successive injuries. If the injuries occur at random at a constant rate, then the times between successive injuries follow a negative exponential distribution. The assumption of a constant rate is untenable, because the numbers of injuries at night are substantially less than during the day (see Figures 2.1 and 2.2 in Appendix 2 of this Annex). It follows that the numbers of hours between successive injuries would be difficult to model, and so the numbers of days between successive injuries are considered. Based upon a well-known set of assumptions in Statistics, the number of days between successive injuries should follow a form of negative exponential distribut ion. (This model has been applied to many problems in engineering: for example, in communications and in reliability.) Equivalently, if the numbers of days on which 0, 1, 2, 3 … injuries occur are considered, then the Poisson distribution is appropriate. These frequencies are given in Table 11.2. The mean and variance are 0.9543 and 0.8992, respectively: these values should be close if the Poisson distribution is appropriate. The expected numbers of days, calculated for a Poisson distribution with mean 0.9543, are also given in Table 11.2 and are seen to be very close to the observed frequencies. More formally, the chi-squared value of 1.47, tested on 4 degrees of freedom, is not significant. The hypothesis that the Poisson distribution fits the data is not rejected. The closeness of the fit of the exponential distribution to the days between successive injuries was also tested using probability plots and formal tests, and found to fit very closely. As the results added nothing to the argument they are omitted. • The data are consistent with injuries occurring at the same rate on each of the

seven days of the week. • The Poisson distribution with mean 0.95 closely fits the numbers of injuries per

day.

57

APPENDICES for Annex 2.

Annex 2, Appendix 1. Tables for Year by Severity by Operation The commentary for the tables in this appendix is given in sub-section 1.2. Table 1.1. Injuries classified by Year, Severity and Operation



Production 3 16 2 18 6 18 63 Drilling 19 95 27 90 19 74 324 Maintenance 13 114 23 82 23 62 317 Deck Operations 7 37 15 54 9 46 168 Domestic 4 21 3 21 6 17 72 Other 6 29 8 31 6 19 99 Total 52 312 78 296 69 236 1043 There are 2 missing values in Table 1.1. The above table is almost the same as Table 1.3a of Annex 1. The effect of the new grouping of the categories of Operation is to classify some injuries as 'Maintenance' that were previously classified as 'Other'. The deviances summarising the analysis of the Table 1.1 are given in Table 1.1a. The results are very similar to those in Table 1.3b of Annex 1: because the frequencies in only two categories have been changed, the similarity is to be expected.

Table 1.1a. Deviance Table for Year, Severity and Operation Year Sev Operation Y*Sev Y*O Sev*O Deviance df Sig. level

* *** *** - - - 30.81 27 NS * *** *** * - - 21.87 25 NS * *** *** - NS - 17.74 17 NS * *** *** - - NS 30.29 22 NS * *** *** - NS NS 17.22 12 NS * *** *** * - NS 21.35 20 NS * *** *** * NS - 8.80 15 NS * *** *** * NS NS 8.08 10 NS

The significance of the Year-Severity interaction approaches the ** level implying a strong interaction.

58

Annex 2, Appendix 2. Tables and Figures of Hour by Operation and BIT The commentary for the contingency tables and figures in this appendix is in sub-section 1.3. The grouping of the categories of Operation and BIT are given in sub-section 1.1. The frequencies of injuries classified by Hour and Operation and by Hour and BIT are given in Tables 2.1 and 2.2, respectively. 'Drilling' and 'Maintenance' are the two grouped categories of Operation that occur most frequently. 'Slips/Trips/Falls' and 'Handling goods/materials' are the categories of BIT that occur most frequently. The frequencies for these categories are plotted in two radar graphs, Figures 2.1 and 2.2. Such graphs represent the distribution of injuries in the 24-hour day so that the peak hours for injuries are easily identified. Table 2.1. Contingency Table of Hour by Operation Operation Hour

Production Drilling Maintenance Deck Ops

Domestic/Catering

Other Total

0-1 1 9 1 9 1 1 22 1-2 1 9 5 6 2 1 24 2-3 3 15 7 7 1 3 36 3-4 1 4 7 5 2 1 20 4-5 0 12 2 4 2 0 20 5-6 1 13 6 2 0 2 24 6-7 0 5 5 6 9 5 30 7-8 5 13 12 7 5 5 47 8-9 2 14 21 8 2 8 55 9-10 4 11 15 6 5 7 48 10-11 4 21 29 18 6 4 82 11-12 3 17 21 13 2 5 61 12-13 1 22 7 10 2 3 45 13-14 5 23 12 8 4 4 56 14-15 6 17 34 8 7 4 76 15-16 5 13 21 12 7 7 65 16-17 2 12 29 6 3 3 55 17-18 4 10 24 8 0 5 51 18-19 3 8 12 1 2 6 32 19-20 2 9 7 4 3 10 35 20-21 3 20 11 4 2 6 46 21-22 4 13 8 7 2 3 37 22-23 3 17 10 7 1 3 40 23-24 0 17 10 2 2 1 32 Total 62 324 316 168 72 97 1039 There are 6 missing values.

59

Table 2.2. Contingency Table of Hour by BIT BIT Hour

Slips/Trips/Falls

Falling Objects

Handling goods/ materials

Lifting/Crane Ops

Hand Tools

Use of Machinery

Other Total

0-1 8 1 4 3 0 0 6 22 1-2 10 1 5 2 2 0 4 24 2-3 5 3 13 8 1 2 4 36 3-4 8 1 6 3 0 0 2 20 4-5 9 3 5 2 0 0 1 20 5-6 4 4 6 3 1 2 4 24 6-7 15 6 3 2 0 1 3 30 7-8 11 3 14 3 6 5 5 47 8-9 13 3 15 5 1 6 12 54 9-10 16 3 9 2 6 4 9 48 10-11 29 9 26 2 6 4 6 82 11-12 17 5 15 7 4 5 8 61 12-13 17 1 13 3 2 4 5 45 13-14 18 5 11 5 4 8 6 57 14-15 32 1 15 6 5 5 12 76 15-16 23 2 12 8 3 3 14 65 16-17 27 2 11 2 4 1 8 55 17-18 16 7 11 3 3 3 8 51 18-19 10 2 6 6 0 0 8 32 19-20 18 2 6 2 1 1 5 35 20-21 15 3 6 6 5 2 9 46 21-22 10 3 8 4 1 1 9 36 22-23 15 2 8 3 2 1 9 40 23-24 9 2 6 4 2 4 5 32 Total 355 72 234 94 59 62 153 1038 There are 7 missing values.

60

Figure 2.1. Drilling and Maintenance for 24 Hours

Drilling

Maintenance

mean0

6

12

18

Figure 2.2. Slip/Trips/Falls and Handling goods for 24 Hours.

mean

Slips/Trips/Falls

Handling0

6

12

18

61

Annex 2, Appendix 3. Tables of BIT by Operation, Age, Activity and Kind of Accident The commentary for the four contingency tables given in this appendix is given in sub-section 1.4. For brevity some of the names of the categories of BIT have been changed marginally in the following tables: for example, 'Use of Machinery' is referred to as 'Machinery'. The tables are easy to read with the seven categories of BIT listed vertically: in the equivalent tables in Annex 1 these categories are listed horizontally.

Table 3.1. Contingency Table of BIT by Operation

Operation BIT


Domestic/Catering

Other Total

Slip/Trip/Fall 27 76 119 50 40 43 355 Falling Objects 2 32 23 12 2 1 72 Handling goods 9 89 68 49 6 14 235 Lifting/Craning 0 39 16 36 0 3 94 Hand Tool 4 18 25 4 6 2 59 Machinery 8 33 19 2 0 0 62 Other 13 37 47 15 17 35 164 Total 63 324 317 168 71 98 1041 There are 4 missing values. Table 3.2. Contingency Table of BIT by Age-group

Age-group BIT

<31 31 - 40 41 - 50 51 - 65 Total

Slip/Trip/Fall 76 133 92 44 345 Falling Objects 24 24 16 7 71 Handling goods 79 75 64 10 228 Lifting/Craning 31 43 12 8 94 Hand Tool 8 28 13 10 59 Machinery 20 18 18 5 61 Other 38 65 37 14 154 Total 276 386 252 98 1012 There are 33 missing values.

62

Table 3.3. Contingency Table of BIT by Activity

Activity BIT

Portable tools

Manual Operating plant

Erect Climb/ descend

Walk Other Total

Slip/Trip/Fall 18 40 15 24 110 116 32 355 Falling Objects. 10 18 20 9 0 1 15 73 Handling goods 14 174 17 7 0 4 19 235 Lifting/Craning 3 38 33 5 1 0 14 94 Hand Tool 47 5 5 1 0 0 1 59 Machinery 11 6 38 1 0 1 5 62 Other 7 26 25 5 11 16 74 164 Total 110 307 153 52 122 138 160 1042

There are 3 missing values. Table 3.4. Contingency Table of BIT by Kind of Accident

Kind BIT

Contact with Machinery

Struck by object

Handling Lifting or Carrying

Slip Trip Fall on same level

Fall from Height

Other Total

Slip/Trip/Fall 2 21 19 182 77 54 355 Falling Objects 0 60 8 0 1 4 73 Handling goods 10 48 131 5 1 40 235 Lifting/Craning 3 54 17 1 3 16 9 Hand Tool 3 18 14 2 1 21 59 Machinery 20 21 10 0 1 10 62 Other 4 44 25 6 4 81 164 Total 42 266 224 196 88 226 1042 There are 3 missing values.

63

Annex 2, Appendix 4. Tables of Kind of Accident by Operation, Age and Activity The commentary for the four contingency tables given in this appendix can be found in sub-section 1.5. Table 4.1. Contingency Table of Kind of Accident by Operation

Operation Kind


Domestic/Catering

Other Total


4 18 15 4 0 1 42

Struck by Object 8 113 73 51 10 10 265 Handling/Lift/ Carrying goods

15 79 70 35 12 14 225

Slip/Trip/Fall on same level

18 38 63 30 21 27 197

Fall from height 1 23 35 8 13 8 88 Other 17 53 61 40 16 39 226 Total 63 324 317 168 72 99 1043 There are 2 missing values. Table 4.2. Contingency Table for Kind of Accident by Age Age Kind

<31 31-40 41-50 51-65 Total

Contact with Machinery 10 19 9 4 42 Struck by Object 90 105 46 23 264 Handling/Lifting/Carrying goods

66 73 69 15 223

Slip/Trip/Fall same level 38 82 53 23 196 Fall from height 22 31 19 14 86 Other 50 78 57 19 204 Total 276 388 253 98 1015 There are 30 missing values.

64

Table 4.3. Contingency Table of Kind of Accident by Activity

Activity Kind

Portable tools



Walk Other Total


7 9 20 1 0 0 5 42

Struck by Object 38 77 68 15 3 15 50 266 Hand/Lift/Carry goods

21 140 23 7 4 8 21 224


10 22 10 9 46 87 13 197

Fall from height 6 4 1 13 45 10 9 88 Other 28 55 31 7 24 19 62 226 Total 110 307 153 52 122 139 160 1043

There are 2 missing values.

65

Annex 2, Appendix 5. Tables of Operation by Employment Status and Activity The commentary for the four contingency tables given in this appendix can be found in sub-section 1.6. Table 5.1. Contingency Table for Operation and Employment Status

Employment Status Operation

Employee Self-employed Other Total

Production 26 0 37 63 Drilling 154 3 166 323 Maintenance 89 0 226 315 Deck Operations 62 0 102 164 Domestic/Catering 17 0 55 72 Other 32 2 52 86 Total 380 5 638 1023 There are 22 missing values. In the calculations of the Chi-squared statistics, 'Self-employed' is combined with the ‘Other’ category to form the category 'contracted out-worker'. Table 5.2. Contingency Table for Operation and Activity

Activity Operation

Portable tools



Walk Other Total

Production 2 12 15 0 9 13 12 63 Drilling 35 114 80 22 20 27 26 324 Maintenance 57 81 31 27 36 39 46 317 Deck Operations

8 74 25 1 23 18 19 168

Domestic/ Catering

6 10 1 0 13 24 18 72

Other 2 15 1 2 21 18 39 98 Total 110 306 153 52 122 139 160 1042 There are 3 missing values.

66

Annex 2, Appendix 6. Tables of Year by BIT, Kind of Accident and Age The commentary for the four contingency tables given in this appendix can be found in sub-section 1.7. Table 6.1. Contingency Table for Year and BIT BIT Year

Slips/Trips/Falls

Falling Objects


Lifting/Crane Ops

Hand Tools

Use of Machinery

Other Total

96/97 127 24 92 24 20 23 54 364 97/98 117 33 74 35 20 21 73 373 998/99 111 16 69 35 19 18 37 305 Total 355 73 235 94 59 62 164 1042 There are 3 missing values. Table 6.2. Contingency Table of Year by Kind of Accident

Kind Year


Struck by object


Slip Trip Fall on same level

Fall from Height

Other Total

96/97 9 88 101 70 37 59 364 97/98 19 112 69 70 27 79 376 98/99 14 69 55 57 24 89 305 Total 42 266 225 197 88 227 1045 Table 6.3. Contingency Table of Year by Age

Age-group Year

<31 31 - 40 41 - 50 51 - 65 Total

96/97 106 132 92 29 359 97/98 95 142 94 32 363 98/99 75 114 67 37 293 Total 276 388 253 98 1015 There are 30 missing values.

67

Annex 2, Appendix 7. Tables of Severity by Part of Body, Age, Operation, Activity, BIT and Kind of Accident The commentary for the four contingency tables given in this appendix can be found in sub-section 2.1. Table 7.1. Part of Body classified by Severity of Injury Part of Body Severity

Head Torso Upper Limb

Lower Limb

Other Total

F&S 13 15 100 58 13 199 3-day 63 215 340 190 34 842 Total 76 230 440 248 47 1041 There are 4 missing values. Table 7.2. Age classified by Severity of Injury Age Severity

<31 31-40 41-50 51-65 Total

F&S 48 69 48 26 191 3-day 228 319 205 72 824 Total 276 388 253 98 1015 There are 30 missing values. Table 7.3. Operation classified by Severity of Injury

Operation Severity


Domestic/Catering

Other Total

F&S 11 65 59 31 13 20 199 3-day 52 259 258 137 59 79 844 Total 63 324 317 168 72 99 1043 There are 2 missing values.

68

Table 7.4. Activity classified by Severity of Injury

Activity Severity

Portable tools



Walk Other Total

F&S 16 46 31 15 31 31 29 199 3-day 94 261 122 37 91 108 131 844 Total 110 307 153 52 122 139 160 1043

There are 2 missing values. Table 7.5. BIT classified by Severity of Injury BIT Severity

Slips/Trips/Falls

Falling Objects


Lifting/Crane Ops

Hand Tools

Use of Machinery

Other Total

F&S 97 14 23 31 5 11 18 199 3-day 258 59 212 63 54 51 146 843 Total 355 73 235 94 59 62 164 1042 There are 3 missing values. Table 7.6. Kind of Accident classified by Severity of Injury

Kind Severity


Struck by Object



Fall from Height

Other Total

F&S 10 56 13 54 29 38 200 3-day 32 210 212 143 59 189 845 Total 42 266 225 197 88 227 1045

69

Annex 2, Appendix 8. Tables of Occupation by Employment Status and Type of Installation The commentary for the table given in this appendix is given in sub-section 2.2. Table 8.1. Occupation classified by Employment Status and Type of Installation Type of Installation Fixed Mobile Status Occupation

Employee Contracted out-worker

Employee Contracted out-worker

Total

Professional 14 24 7 6 51 Medical 9 12 1 0 22 Skilled Technician 5 61 8 8 77 General Operators 1 6 1 2 3 Semi-skilled 12 31 7 2 52 Catering/Domestic 1 23 21 1 46 Labouring 38 208 93 32 371 Null 39 181 124 51 395 Total 119 541 262 102 1024 One or more of the three variables is missing for 21 injuries and 415 of the occupations are recorded as 'Null'. Table 8.2. Deviance Table for Occupation, Employment Status and Type of Installation Occupation Status Type O*S O*T S*T Deviance df Sig. level

*** *** *** - - - 394.8 22 *** *** *** *** *** - - 368.7 15 *** *** *** *** - *** - 345.0 15 *** *** *** *** - - *** 95.0 21 *** *** *** *** - *** *** 45.2 14 *** *** *** *** *** - *** 68.9 14 *** *** *** *** *** *** - 318.9 8 *** *** *** *** *** *** *** 20.8 7 **

70

Annex 2, Appendix 9. Figures for Monitoring The figures in this Appendix are referred to in sub-section 3.1. There are two scatter diagrams, two cusum charts and one Manhattan diagram.

Figure 9.1. Weekly frequencies of injuries plotted against date

0

2

4

6

8

10

12

14

16

5-F

eb-9

6

23-A

ug

-96

11-M

ar-9

7

27-S

ep-9

7

15-A

pr-

98

1-N

ov-

98

20-M

ay-9

9Date

Fre

qu

ency

Figure 9.2. 30-day frequencies of injuries plotted against date

05

1015202530354045

Feb

-96

Au

g-9

6

Mar

-97

Sep

-97

Ap

r-98

No

v-98

May

-99

Date

Fre

qu

ency

71

Figure 9.3. The cusum chart for weekly frequencies of injuries

0

10

20

30

40

50

60

5-F

eb-9

6

5-A

ug

-96

5-F

eb-9

7

5-A

ug

-97

5-F

eb-9

8

5-A

ug

-98

5-F

eb-9

9

5-A

ug

-99

Date

Cus

um

Figur 9.4. The cusum chart for 30-day frequencies of injuries

0

5

10

15

20

25

30

35

40

45

50

Feb-

96

Aug

-96

Feb-

97

Aug

-97

Feb-

98

Aug

-98

Feb-

99

Aug

-99

Date

Cu

sum

72

Figure 9.5. Manhattan diagram for 30-day frequencies of injuries

05

1015202530354045

Feb

-96

Au

g-9

6

Mar

-97

Sep

-97

Ap

r-98

No

v-98

May

-99

Date

Fre

qu

ency

73

Annex 2, Appendix 10. Figures and graphs for the relationship between Injuries and the Price of Oil The figures in this Appendix are referred to in sub-section 3.2. Weekly prices for the period 31stJuly, 1988, to 31stJuly, 2000, are plotted in Figure 10.1 and for a shorter period in Figures 10.2 and 10.3. The weekly frequencies of injuries occurring between 1stApril, 1996, and 31stMarch, 1999, are superimposed upon the plots of the weekly prices in Figures 10.2 and 10.3. The autocorrelations of weekly oil prices, of differences in the prices from week to week and the weekly frequencies of injuries, are plotted in Figures 10.4, 10.5 and 10.6 respectively. The cross-correlations of price differences and injury frequencies are plotted in Figure 10.7.

Figure 10.1. Price of Brent Oil at close of day

0

5

10

15

2025

3035

4045

27-F

eb-8

8

11-J

ul-

89

23-N

ov-

90

6-A

pr-

92

19-A

ug

-93

1-Ja

n-9

5

15-M

ay-9

6

27-S

ep-9

7

9-F

eb-9

9

23-J

un

-00

Date

$ P

rice

of

Bar

rel

74

Figure 10.2. Price of Brent Oil and the weekly numbers of Injuries

0

5

10

15

20

25

30

35

40

1-Ja

n-9

5

20-J

ul-

95

5-F

eb-9

6

23-A

ug

-96

11-M

ar-9

7

27-S

ep-9

7

15-A

pr-

98

1-N

ov-

98

20-M

ay-9

9

6-D

ec-9

9

Date

$ P

rice

of

Bar

rel

Brent price

Injuries

Figure 10.3. Price of Brent Oil and 4-weekly moving averages of the numbers of injuries

0

5

10

15

20

25

30

1-Ja

n-9

5

20-J

ul-

95

5-F

eb-9

6

23-A

ug

-96

11-M

ar-9

7

27-S

ep-9

7

15-A

pr-

98

1-N

ov-

98

20-M

ay-9

9

6-D

ec-9

9

Date

$ P

rice

of

Bar

rel

Brent price

Injuries

75

Figure 10.4. Autocorrelations of Brent Oil prices

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0 5 10 15 20 25 30 35

Lag in units of one week

Au

toco

rrel

atio

ns

Figure 10.5. Autocorrelations for price differences

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0 5 10 15 20 25 30 35


Au

toco

rrel

atio

ns

76

Figure 10.6. Autocorrelations of Injury frequencies

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0 5 10 15 20 25 30 35


Aut

ocor

rela

tions

Figure 10.7. Cross-correlations of price differences and injury frequencies

-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0 5 10 15 20 25 30 35


Cro

ss-c

orr

elat

ion

s

Annex 2, Appendix 11. Tables of Days of the week and the Numbers of Injuries per day The two tables in this Appendix are discussed in sub-section 3.3. In Table 11.1, 131 of the 1045 recorded injuries occurred on a Sunday, 160 on a Monda y etc. It is seen in Table 11.2 that no injuries occurred on 418 days, one injury occurred on 413 days etc. A total of 1045 injuries were recorded on 1095 days: numbers given in the rows of totals in Tables 11.1 and 11.2. The entries in the column of expected numbers is based upon a Poisson distribution.

Table 11.1. The numbers of injuries on the days of the week

Day Frequency Sunday 131 Monday 160 Tuesday 139 Wednesday 153 Thursday 157 Friday 164 Saturday 141 Total 1045

Table 11.2. The numbers of days

Number of Injuries per day

Number of Days Expected Number of days

0 418 421.7 1 413 402.4 2 186 192.0 3 59 61.1 4 14 14.6 5 3 2.8 6 2 0.5

Total 1095 1095.1

77

Printed and published by the Health and Safety ExecutiveC30 1/98

Printed and published by the Health and Safety ExecutiveC0.35 10/01

OTO 2000/108

£25.00 9 780717 621316

ISBN 0-7176-2131-6

OFFSHORE TECHNOLOGY REPORT 2000/108Secure Site · reported to the Offshore Division (OSD) in the Hazardous Installations Directorate of the Health and Safety Executive. Over the

Documents