UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Unexpected events while manually falling trees DeMille, Gregory John 2013

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2013_spring_demille_greg.pdf [ 1.76MB ]
Metadata
JSON: 24-1.0073689.json
JSON-LD: 24-1.0073689-ld.json
RDF/XML (Pretty): 24-1.0073689-rdf.xml
RDF/JSON: 24-1.0073689-rdf.json
Turtle: 24-1.0073689-turtle.txt
N-Triples: 24-1.0073689-rdf-ntriples.txt
Original Record: 24-1.0073689-source.json
Full Text
24-1.0073689-fulltext.txt
Citation
24-1.0073689.ris

Full Text

Unexpected Events while Manually Falling Trees  by  Gregory John DeMille  THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF  Master of Science  in  The Faculty of Graduate Studies (Forestry)  University of British Columbia (Vancouver)  April 2013  © Gregory John DeMille 2013  Abstract To date, the forest industry has relied on incident data in the form of fatality and serious injury statistics to improve safety in manual tree falling. These data are limited in their ability to improve falling safety. Thus, it is necessary to record another class of incident data to help gain insight into the difficulty (or danger) of falling trees. This study used a system of conditions (management requiring conditions) that are reported by the faller before a tree was felled and an assessment by the faller of whether the tree deviated from the intended plan (unexpected event) to determine if the frequency of unexpected events was correlated with site specific factors or the frequency of management requiring conditions encountered. In total, 1292 falling observations were collected during 86 one-hour observation periods. In this study it was found that there were significant differences between fallers in the frequency of management requiring conditions reported; however, other than the presence of an adjacent standing tree with falling cuts present (cut-up tree), the management requiring conditions were not good predictors of whether an unexpected event would occur. The likelihood of an unexpected event occurring was found to be correlated with terrain type, ground slope, stump diameter, and the presence of a cut-up tree. Overall, 6.9 % of the falling observations had an unexpected event occur. Of particular note, 1.2% of the observations had an unexpected event occur with a severity code 2 or 3, which indicates it was more than normal variation in falling. Given the uncertainty that a faller is exposed to when cutting into a tree, a major focus on faller safety must consider how to help the faller to be mentally and physically ready to adapt to changing conditions while working on a tree. The results indicating that all fallers who participated in this study had unexpected events, and that management requiring conditions were  ii  not good predictors of unexpected events, demonstrates that data collected at the harvest planning phase (often years before falling) will not be very effective at predicting unexpected events during falling.  iii  Table of Contents Abstract ......................................................................................................................................................... ii Table of Contents ......................................................................................................................................... iv List of Tables ................................................................................................................................................ v List of Figures .............................................................................................................................................. vi Acknowledgements ..................................................................................................................................... vii 1.  Introduction ........................................................................................................................................... 1 1.1  2.  3.  4.  5.  Objectives ..................................................................................................................................... 6  Methods................................................................................................................................................. 7 2.1  Falling Difficulty (Danger) Metrics .............................................................................................. 7  2.2  Block and Participant Selection .................................................................................................... 9  2.3  Data Collection ........................................................................................................................... 11  2.4  MRC Frequency Analysis ........................................................................................................... 13  2.5  UE Analysis ................................................................................................................................ 16  Results ................................................................................................................................................. 17 3.1  Description of Participating Subjects and General Observations ............................................... 17  3.2  Linear Mixed Effects Regression of MRC.................................................................................. 19  3.3  Logistic Regression of UEs......................................................................................................... 24  Discussion ........................................................................................................................................... 32 4.1  MRC Frequency .......................................................................................................................... 32  4.2  Unexpected Events...................................................................................................................... 33  4.3  Study Limitations and Further Research ..................................................................................... 37  Conclusion .......................................................................................................................................... 39  References ................................................................................................................................................... 42 Appendix A: Management Requiring Condition (MRC) and Unexpected Event (UE) Code Definition ... 47 Appendix B: Subject Data Collection Information ..................................................................................... 52 Appendix C: Equations Used in Preliminary Analysis ............................................................................... 53  iv  List of Tables Table 1.1  Fatalities by occupation in the forest industry (2002-2006) (BC Coroners Service 2007) ........ 2  Table 1.2  Average injury rates for the forestry subsector classified by WorkSafeBC CU for the years  2003-2007(WorkSafeBC 2010) .................................................................................................................... 3 Table 1.3  Faller serious injuries and fatalities reviewed(WorkSafeBC, 2009) ......................................... 6  Table 2.1  Description of blocks observed in the study ........................................................................... 10  Table 2.2  Covariate information collected for each observation ............................................................ 12  Table 3.1  Individual faller observation summaries ................................................................................. 18  Table 3.2  Summary of UE classified by severity code ........................................................................... 18  Table 3.3  Estimates for the fixed effects for Equation 7......................................................................... 22  Table 3.4  Estimates for the random effects for Equation 7..................................................................... 23  Table 3.5  Fixed effects for Equation 12 .................................................................................................. 30  Table 3.6  Odds ratios for Equation 12 ................................................................................................... 31  v  List of Figures Figure 3.1  Plot of the conditional residuals compared to the linear predicted values ............................. 21  Figure 3.2  Plot of the predicted number of MRC‟s as a function of DSH .............................................. 23  Figure 3.3  Area under the Receiver Operating Curve for Equation 12................................................... 28  Figure 3.4  Predicted probabilities for each observation classified by observed UEs ............................. 29  Figure 3.5  Predicted probabilities for DSH and Slope for each observation........................................... 31  vi  Acknowledgements I would like to sincerely thank my supervisor, Dr. Kevin Lyons who provided support and guidance throughout this project. As well, I am grateful to the members of my thesis committee, Drs. Peter Marshall and Mihai Pavel, for their insight and contributions to this thesis. I would also like to acknowledge the funding support supplied by WorkSafeBC, grant RS2010-OG05, through the Focus on Tomorrow program.  vii  1.  Introduction Manual tree falling is one of the most dangerous operations in forestry. It involves  numerous hazards that are difficult to predict (Peters 1991). Table 1.1 summarizes forestryrelated deaths from 2002 to 2006 by occupation. Though Table 1.1 does not provide information on fatalities per man-day worked for each occupation, it does indicate that manual tree falling had numerous fatalities over the 2002-2006 period. The number of fatality benefits accepted by WorkSafeBC in the Forestry sector in the period from 2002-2011 was 124, while in Agriculture it was 33, and Oil and Gas or Mineral Resources it was 68 (WorkSafeBC 2012a), which illustrates that forestry is more dangerous then the other resource-based industries. With the initiation of the faller training program in 2006, the number of fatalities in manual tree falling dropped; however, the serious injury rate remained relatively unchanged from 2006 to 2008 (WorkSafeBC 2010). In 2008 and 2009, there were eight and three fatalities, respectively, while manually falling trees in British Columbia (BC Forest Safety Council 2012). The risk of injury in manual tree falling compared to other occupations within the forestry sector is also high. The average non-HCO* injury rate reported by WorksafeBC(2010) for the years 2003 to 2007 was 33 claims/100 person years for the Manual Tree Falling and Bucking classification unit compared to 2 claims/100 person years for both the marine log salvage and log processing classification units (Table 1.2).  *  Non-HCO injuries: The number of claims with costs related to at least one of the following benefits types: shortterm disability benefits (STD), long-term disability benefits (LTD), or survivor (Fatal) benefits and where the first STD/LTD/Fatal payment date is within the year of injury or the three months following the year of injury (WorkSafeBC 2012b).  1  Table 1.1 Fatalities by occupation in the forest industry (2002-2006) (BC Coroners Service 2007) Occupation Manual Tree Faller Residential forestry Log hauling/ trucking Mechanical harvesting Ground skidding/ cable yarding Helicopter logging (air crew) Helicopter logging (ground crew) Helicopter operation (other than logging) Landing /log sorting Forestry road construction/ maintenance Forest fire fighting (helicopter) Travel to from work (motor vehicle) Travel to from work (aircraft or boat) Sawmilling (mechanical) Sawmilling (vehicle) Sawmilling (other) Silviculture Other Total  2002 6 0 5 0 1 2 0 1  2003 2 1 4 0 1 0 0 0  2004 1 3 4 0 1 0 1 0  2005 8 3 7 1 6 2 0 0  2006 0 1 5 2 0 0 0 0  Total 17 8 25 3 9 4 1 1  0 0  0 1  1 1  2 0  0 0  3 2  0 1 1 2 0 1 0 0 29  1 1 0 3 0 1 0 0 15  0 0 1 0 0 1 0 0 14  0 1 6 0 2 2 0 5 45  0 0 0 1 0 0 0 5 14  1 3 8 6 2 5 0 10 108  The greatest risk of injury to fallers is being struck by falling trees and other debris (Peters 1991). This is evident in WorkSafeBC claims for the years 2005-2009, of which 37% of all accident types for manual tree falling and bucking (the highest in the category) involve being struck by an object (BC Forest Safety Council 2011). Ground conditions and terrain have also been noted as contributing to injuries for manual tree fallers. A survey of injured loggers reported that two-thirds of the respondents said that one or more natural conditions contributed to their injuries (Anon. 1985). The natural conditions included heavy brush or ground cover which was considered responsible for injuries by 20% of the respondents, while 10% blamed steep  2  terrain. Paulozzi (1987) suggested that the steepness of the terrain in the Western United States and large tree size resulted in higher logging mortality rates in Washington State compared to Ontario. Falling trees that have been damaged by snow or wind also represents a considerable hazard to manual tree fallers. Gaskin et al. (1989) reported that operations that occurred in areas damaged by a cyclone in New Zealand had a higher rate of injury than those in non-cyclone areas. The tension in trees resulting from snow/wind damage have also been linked by Kubiak (1985) to an increase in injury rates. A large part of the faller‟s job involves the assessment of conditions and the likely risk each tree or stand of trees represents (Bentley et al. 2005). Poor assessments of the ambient and physical environment, as well as the condition of the tree being felled, can contribute to the likelihood of an injury occurring. Table 1.2 Average injury rates for the forestry subsector classified by WorkSafeBC CU for the years 2003-2007(WorkSafeBC 2010) Classification Unit Description Manual Tree Falling and Bucking Shake Block Cutting Cable or Hi-Lead Logging Helicopter Logging Brushing and Weeding or Tree Thinning or Spacing (not elsewhere specified) Forest Fire Fighting Tree Planting or Cone Picking Log Booming Dry Land Sort Ground Skidding, Horse Logging, or Log Loading Integrated Forest Management Logging Road Construction or Maintenance Mechanized Tree Falling Log Processing Marine Log Salvage  Average Injury Rate (non-HCO injuries per 100 person years 33 22 18 14 13 11 10 9 6 6 5 4 3 2 2 3  Laughery and Wogalter (1997) define a hazard as “a set of circumstances that can result in injury or property damage". They further define danger as a “product of hazard and likelihood”, while risk perception involves “the overall awareness of hazards and potential outcomes of a set of circumstances”. Limited research has been conducted into hazard awareness and risk perception among manual tree fallers (Slappendal et al. 1993). The studies of risk perception that have been completed have reached contradictory results. A high level of concordance between actual risk and those perceived by fallers was found by Tap et al. (1990) and Osteberg (1980). The results of studies by Dunn (1972) and Klen (1988) indicated a poor level of perception by fallers. The methodology used in these studies predominately focused on the presence of perceived impacts of individual factors in isolation (Parker 2010). Assessing risk and identifying conditions that require mitigating actions in isolation does not necessarily transfer to the operational setting (Parker 2010) due to a failure to account for factors such as environmental conditions, worker fatigue, organizational and social pressures, and other factors. Parker and Kirk (1993) and Parker (2010) attempted to quantify the hazard types and frequency experienced in an operational setting. While both studies are an important step in recognizing the frequency and types of hazards to which a faller is exposed, further research is required to provide the data necessary to investigate possible links between the frequency of hazards encountered, risk perception, and accident frequency. Recent reports by Western Forest Products (2006), WorkSafeBC (2009) and the BC Coroners Service (2010) noted that one of the most frequent causative factors in serious injuries and fatalities when manually falling trees is the lack of hazard and risk assessment at the harvest planning level. Risk assessment at the planning stage was identified as a causative factor in 88%  4  of the incidents reviewed by WorkSafeBC (2009), which was second only to risk assessment by the faller. These reports recommended developing a Hazard Matrix to help identify hazards at the harvest planning stage with the objective of matching the qualifications of individual fallers to the work difficulties that are likely to be encountered. The development of such a matrix requires the data necessary to link the frequency of hazards and risk of injury. An expert panel from WorkSafeBC reviewed 32 serious injury and fatal accidents that occurred from 2000 to 2008 (WorkSafeBC 2009). Table 1.3 summarizes the incidents reviewed by the panel. The panel notes implementation of the British Columbia Faller Training Standard (BCFTS) and the Integrated Forestry Compliance Strategy (IFCS) coincide with zero fatalities for certified manual tree fallers in British Columbia in 2006 and 2007. In response to the 7 fatalities that occurred in 2008, the report prepared by the panel states: “It appears that the old way of falling trees (i.e., prior to the introduction of the falling standard) re-emerged as the way to do the job". However, other years considered in this review also had low manual tree faller fatality rates: two fatalities in 2002 and one fatality in 2004. This highlights one of the key limitations of the incident data collected by WorkSafeBC - it does not provide information for workers who are not involved in an incident, and detailed investigations are usually only performed for fatal incidents. Thus, it is not clear why the fatal accident rate was low in 2006 and 2007. Additionally, the data in Table 1.3 indicate that fatalities while manual tree falling are relatively infrequent, although each fatality is tragic and carries a high social cost. Due to the relatively infrequent occurrence of fatalities and serious injuries, it is necessary to record another class of incident in order to measure the difficulty (or danger) of falling trees in a particular area.  5  Table 1.3  Faller serious injuries and fatalities reviewed(WorkSafeBC, 2009) Year  2000  Number of Incidents *  6  2001 2  2002 4  2003 3  2004 2  *  2005 6  2006 0  2007  2008  **  2  7  One incident a serious injury Both incidents were serious injuries  **  It is assumed that a faller does not intentionally cause an injury; therefore, serious injuries and fatalities must result from incidents that the faller was not expecting. Though most unexpected events do not result in serious injuries or fatalities, some do, and it is assumed that the higher the frequency of unexpected events, the higher the likelihood that the faller will be injured. Thus, the frequency of unexpected events that occur while falling trees could be a useful metric for measuring the difficulty (or danger) of falling trees in a particular area.  1.1  Objectives  The objectives of this thesis are: 1) to develop a metric that describes the conditions that have to be managed when falling a tree and that the faller can see before cutting into the tree; 2) to determine if the frequency of the conditions that have to be managed when falling a tree are correlated with site specific factors and whether the frequency varies between fallers; 3) to develop a metric that indicates when a tree deviated from the faller's initial plan for the tree; and 4) to determine if the metric that indicates a tree deviated from the faller's initial plan is correlated with the metric that describes the conditions the faller saw before cutting into the tree.  6  2.  Methods  2.1  Falling Difficulty (Danger) Metrics The reliance on fatality data to understand the conditions faced by fallers is inadequate  because: 1) there are only a few observations per year; 2) the information is second hand since the fallers usually work at least two tree lengths apart; and 3) there is no data on the work practices of fallers who are not injured. Thus, this study uses a system of conditions (management requiring conditions) that are reported by the faller before the tree is felled and an assessment by the faller of whether the tree deviated from the intended plan (unexpected events). Management requiring conditions and unexpected events are the metrics that will be used as outlined in objectives 1 and 3, respectively. Some fallers will not report a management requiring condition (MRC) because they believe there is a reason that the condition will not impact on them; however, this assessment is subjective and varies between fallers. Thus, a severity code was attached to the MRC code to allow the faller to differentiate between low and high perceived threats. A similar problem occurs when asking a faller to report unexpected events (UE). How far a tree has to deviate from a faller's initial plan before it is called an unexpected event also varies between fallers. Again, a severity code was attached to the UE code to allow the faller to differentiate between normal variation in the falling process and an event that posed a significant threat. A complete list of the MRC and UE codes used in this thesis is presented in Appendix A. The definitions used for MRC and UE in data collection, and examples follow.  Management Requiring Condition (MRC): Is a condition that requires either an action or decision by the faller before a tree can be felled.  7  Severity Code: 1 = Not an immediate threat, 2 = Immediate threat but have existing cover or an escape route, and 3 = Immediate threat requiring alternate falling method. The following is an example of an MRC : MRC: A dead tree (snag) is adjacent to the tree being felled. Often a faller has to open up a face to remove a particular snag and in doing this has to work adjacent to a snag. MRC Severity 1. level 1 the snag is stable and very unlikely to be hit by the tree being felled, 2. level 2 the snag may be unstable or may be situated so it can be disturbed by the tree being felled, but the faller has cover or an escape route, 3. level 3 the snag is unstable and directed towards the faller, the faller cannot work adjacent to this snag.  Unexpected Event (UE): an event that has the potential to severely injure the faller and either the faller was unaware of the possible occurrence or a planned event did not go as planned. Severity Code: 1 = within normal variation from intended plan, 2 = significant variation from intended plan but safety measures (i.e., escape route) ensure faller safety, and 3 = did not consider the event a possibility. The following is an example of a UE: UE: An unseen object falls out of the canopy while a tree is being felled. Often the faller cannot see objects hanging in the canopy, and these can be dislodged and fall to the ground. UE Severity 1. Level 1: the object fell in an area that the faller would not normally occupy (e.g. to the side of the tree in the direction of fall),  8  2. Level 2: the object fell in an area that the faller might occupy, but existing safety measures (e.g. cover or escape route) protected the faller, 3. Level 3: the object fell in an area that the faller might occupy and existing safety measures would not have protected the faller. 2.2  Block and Participant Selection The data for this study were collected from three different areas in the South Coast  Region of British Columbia and Vancouver Island. To select fallers to participate in this study it was first necessary to find logging companies that were willing to donate lost productive falling time and who were willing to have data collected on their fallers. Second, it was necessary to find fallers in these companies who were willing to participate in this study. Finally it was necessary to find cutblocks that were available to this study and that had fallers who were willing to participate in this study working in them. Numeric codes have been assigned to the fallers and the cutblocks to protect the identity of the fallers. In total, data was sampled from 7 different cutblocks and 11 fallers. All cutblocks, identified as blocks 1 to 7, sampled in this study were harvested using a clear-cut silviculture system. Blocks 1,2,3,6, and 7 were located on Vancouver Island and were required to be completely manually felled, with conventional grapple yarding. Blocks 4 and 5 were located on the lower mainland and were harvested with a combination of manual and machine falling. Hoe-chucking (loader forwarding) was used as the yarding (extraction) method in these blocks. Blocks 1, 5, and 7 occupied lower slope positions. Block 3, 4, and 6 occupied mid slope positions, while Block 2 was located higher up the slope. The blocks were comprised mainly of uniform terrain with small sections of rock outcrops and gullied drainage areas. Part of the area in Block 6 contained a large gullied drainage area that limited the direction that a tree 9  could be felled. The range of slopes in the areas sampled in each block varied from 0 to 103%; Block 6 comprised of the steepest slopes with an average of 74%. The tree species felled, by percentage, in each block can be found in Table 2.1. The fallers that participated in this study did not work in all of the blocks sampled, although some fallers did work in multiple blocks. The range in age and coastal falling experience of the 11 fallers that participated were respectively 31-59, and 7-34 years, respectively. All of the fallers who participated in this study were certified fallers registered with the British Columbia Forestry Safety Council. All of the fallers received their training before the establishment of the BC Faller Training Standard (BCFTS). The number of observation periods that each faller participated in depended on their availability.  Table 2.1  Description of blocks observed in the study % of Trees Species Felled*  Block 1 2 3 4 5 6 7  Ba 21 43 45 0 0 22 50  Cw 21 0 6 2 40 0 2  Fd 0 0 0 85 47 0 4  Hw 40 17 49 13 13 54 44  Yc 18 40 0 0 0 24 0  Stand Type Old Growth Old Growth Old Growth Second Growth Second Growth Old Growth Old Growth  Yarding Method Grapple Grapple Grapple Hoe-Chuck Hoe-Chuck Grapple Grapple  Average Slope (%) 27 65 54 43 60 74 30  * Ba = ambalis (balsam) fir (Abies amabilis), Cw = western redcedar (Thuja plicata), Fd = Douglas-fir (Pseudotsuga menziesii), Hw = western hemlock (Tsuga heterophylla), Yc = yellow cedar (Chamaecyparis nootkatensis).  10  2.3  Data Collection  Typically more than one faller was present in a block at a time. To increase coverage of the fallers and the block, data were collected from each faller for a one-hour period. This cycling was continued for the duration of the work day. Two levels of data were collected during the one-hour observation period. The first was site specific covariate data that were collected from the area that the faller would work in for the one hour period. This included six possible explanatory variables: Terrain Type (Tr), Undergrowth (Ug), Slope Percentage (Sl), Roughness Scale (Rc), Weather (Wt), Wind Scale (Ws). See Table 2.2 for a description of the covariate data collected. This information was collected at the beginning of the one-hour observation period and assumed to remain constant throughout the period. The second level of data was specific to the individual trees that were felled during the period. Data collected at this level included: species (Sp), tree diameter at stump height (DSH), MRC, and UE. This information was reported by the faller. A falling observation was defined as cutting down a tree (either dead or alive) with DSH greater than 25cm and total height greater than 3m. Trees that did not meet the minimum DSH and height requirements were considered brush and the cutting of these trees was recorded as an MRC. Cutting the roots off wind thrown trees was considered a falling observation if the tree was leaning in such a manner that a falling cut was required to bring the tree safely to the ground. To begin a falling observation, the faller would report by radio to the researcher the species, DSH, and MRCs that he was aware of. The faller would then fall the tree. Once the tree was on the ground the faller would report any UE that occurred. Each faller completed a training period before participating in data collection. The length of the training period for each faller depended on their comfort level with the protocols being 11  used in the study, and ranged from 1 to 3 hours. During the training period, the researcher would meet the faller at the base of a tree that was to be felled. The faller and researcher would discuss the tree and MRC data that were present. The researcher would then move into the clear until the tree had been felled, and then meet the faller at the stump to discuss any UE that had occurred. Table 2.2  Covariate information collected for each observation  Variable Terrain Type (Tr)  Slope (Sl)  Variable Type Class  Description Description of terrain type defined by the following four categories:  Even (E), Broken (B), Rolling (R), and Gullied (G).  Continuous  The average slope percentage for the observation period.  Roughness Scale (Rc)  Class  The number of obstacles greater than 0.25m in height that will impede falling. Obstacles may include: boulders, root balls, up-rooted trees, and old sound stumps.  Undergrowth (Ug)  Class  A description of the vegetation undergrowth based on the following four categories:  Does not affect walking (1)  Walking is affected (2)  Difficult walking and affects faller‟s work (3)  Extremely difficult walking and significantly impacts faller‟s work (4)  Weather (Wt)  Class  A description of the weather for the duration of the observation period based on the following categories:  Sunny (S)  Rain (impact on faller safety) (R)  Light rain (no impact on faller safety) (L)  Foggy (F)  Wind Scale (Ws)  Class  The effect of the wind on the falling of trees based on the following categories:  No effect (1)  Constant effect (2)  High effect (gusty winds) (3)  Tree Species (Sp)  Class  Two-letter tree species abbreviation code (BC Government)  Tree diameter at stump or bucking location (DSH)  Continuous  The diameter of a tree at the stump or the diameter at the location of the bucking cut. The size was visually estimated by the faller.  12  2.4  MRC Frequency Analysis It was found that MRCs with severity codes 2 and 3 represented only a small fraction of  the total MRCs reported and thus did not allow for a separate analysis using the frequency by severity code; therefore, only the frequency of the MRC type was used for further analysis. During some of the observations, MRC codes were reported multiple times in a single observation. This occurred if the MRC was present at different locations within the faller‟s field of awareness. An example of this would be the presence of several snag trees surrounding the tree to be felled. A contingency table analysis was performed (Hosmer and Lemeshow 2000) to determine if recording multiple similar MRC codes represented a difference from recording only the presence of the condition regardless of the number of times it was present in an observation. This analysis determined that all of the MRCs that were present multiple times within an observation could be represented as binary variables, where 1 would indicate that the condition was present and 0 would indicate that the condition was not present. Thus, in the following analysis the MRC codes are represented by binary variables, where 1 indicates the condition was present for the observation. The MRC frequency data in this study are expected to be correlated with single tree observations occurring within specific observation periods, with each observation period being specific to an individual faller. This study included site specific covariate data collected at the observation period level. Therefore, the observation period level effect was assumed to be accounted for by the measured covariates. This leaves the possibility that the data will exhibit heteroscedasticity and that the falling observations from a particular faller will be correlated. A model where the dependent variable is not normal and the data originates from subjects or clusters where the linear regression model on each subject or cluster can be assumed to be a  13  random deviation from some overall population regression model is a Generalized Linear Mixed Model (GLMM) (Littell et al. 2006). GLMM assume that the random deviations are normally distributed and that the deviation across the intercepts and slopes might be correlated (Littell et al. 2006).The MRC data collected in this study is hierarchical, with observation periods nested within fallers; however, this produces a very complex model when using random intercepts. Thus, the covariate data describing the working conditions for an observation period was used to account for correlation within an observation period, and a random intercept was used to account for the correlation within a faller. To analyze the frequency of MRCs and account for the intrafaller correlation present in the data the following random intercept was used: (  where  )  (1)  is the mean count of the number of MRCs types occurring for the ith individual and the  jth observation, conditional on the predictor variables and the random effects, i = 1,…, 11 individual fallers, and j = 1,…ni observations, µ is the population average effect, β is a row vector of unknown constants of size 1 × q, where q is the number of explanatory vectors, Xij is a q × 1 column of known values of explanatory variables, and individual fallers. The response variable  is the random effect for the  is discrete and thus requires the use of the log  function to link the linear model to the mean of . The frequency of MRCs is assumed to follow a Poisson distribution with a mean and variance equal to . A backwards stepwise approach was used to develop a parsimonious model. The likelihood ratio test (LRT) statistic was used to compare between a full model and a reduced model, where the reduced model was nested within the full model:  14  (2)  where the subscripts R and F correspond to the reduced and full models, respectively, and logL is the natural log of the likelihood function. The analysis of the random intercept model and the development of the parsimonious model were conducted using the Proc Glimmix procedure in SAS/STAT software Version 9.2 TS2M3. The Laplace estimation method was used to approximate the model in Proc Glimmix. The Laplace method is an integral approximation method that approximates the log likelihood and submits the approximated function to numerical optimization (SAS Institute 2008). The advantage of the Laplace method is that it provides an actual objective function for optimization. This allows likelihood ratio tests to be performed between nested models. The inability of the Laplace estimation method to accommodate residual covariance structures including overdispersion (SAS Institute 2008) was not a factor in this analysis.  15  2.5  UE Analysis The UE response variables are binary, where 1 indicates that the condition was present  for the observation, and 0 indicating its absence. It was discovered that a low number of UEs with a severity code of 2 or 3 were observed. Therefore the severity code for each UE reported in an observation was ignored and only the frequency of the UE type was used for further analysis. Similar to the MRC data, it is expected that the UE data will exhibit heteroscedasticity and that observations from a particular faller will be correlated. Again, this required the use of GLMM. To analyze the occurrence of UEs and to account for the intra-faller correlation present in the data, the following random intercept model was used:  (  )  Equation 4 is similar to Equation 2 except that  (3)  is the probability of UE occurring for the ith  individual on the jth observation and the occurrence of a UE is predicted by the natural log of the odds ratio. The explanatory variables considered in the UE analysis included MRCs present in the observation and the measured covariate data. A backwards stepwise procedure was used in combination with a LRT to develop a parsimonious model. Again, to compare between nested random intercept models it was necessary to use Proc Glimmix and the Laplace estimation method. When the random intercept variable was not significant in predicting the occurrence of a UE, the model defaulted to a logistic model. Logistic regression using the Proc Logistic procedure in SAS/STAT Version 9.2 TS2M3 was used to develop the parsimonious model.  16  3.  Results  3.1  Description of Participating Subjects and General Observations  Data were collected over 38 days, yielding 1292 falling observations from 86 one-hour observation periods. One minor strain injury was report during the study which occurred while a faller was moving along his escape trail during the falling of a tree. A total of 99 UEs were recorded. Multiple UEs were recorded 10 times. The multiple UEs experienced were not independent of each other and therefore the UE rates were based only on the occurrence of a UE during an observation and not on the total number of UEs that occurred within that observation. A total of 89 observations had at least 1 UE occur. Of the 1292 falling observations taken during the study, 6.9% had a UE and 1.2% had a UE with a severity code of 2 or 3. Table 3.1 summarizes the range of UEs per falling observation for each faller. Although all fallers experienced at least one UE, only 6 of the 11 fallers reported UEs with a severity code 2 or 3. The percentage of falling observations by each faller in which a UE with a severity code 1, 2, or 3 occurred; ranged from 1.0 – 17.5 %, while the range was 0.0 – 5.7% for falling observations with a UE of severity code 2 or 3. The rate of occurrence of UEs per one hour observation period was 1.03 for severity code 1, 2 or 3, and 0.18 for severity code 2 or 3. Table 3.2 presents the frequency of UEs, classified by severity code. Falling direction change due to other reasons (UET4) was the most frequent UE for severity code 1, and accounted for 28% of all UEs. UET4 was also the most frequent UE when considering events with a severity code greater than 1, with 3 events reported with a severity code of 2 and 2 events with a severity code of 3. Objects falling out of the canopy (UET1) represented the second most frequent UE with a severity code greater than 1, with 4 events reported with a severity code 2. In 17  situations where severity codes of 2 or 3 were reported, adjacent cut-up tree (CT2) was the most frequent MRC, with 4 UEs with severity code 2 or 3 having a CT2. The second most frequent MRC for UEs with severity code 2 or 3 was an adjacent snag (EA2) and the tree being felled was a snag (EF2), with 3 UEs having a EA2 and 3 having a EF2. Table 3.1  Individual faller observation summaries  FallerID  Falling observations  Number of observation periods  537 208 135 564 164 412 108 861 386 617 692  51 53 11 45 50 91 120 273 16 478 104  6 7 3 2 5 12 5 20 2 21 3  Table 3.2 UE UET4 UET1 UET7 UET8 UET14 UET9 UET15 UET6 UET2 UEO4 UEO2 UET3 UEB5 Total  Total UE Total UE (severity (severity 2, 3) 1,2,3) 9 7 1 4 4 7 9 18 1 28 1  0 3 0 0 1 1 2 3 0 6 0  Percent of trees felled with an UE (severity 1,2,3) 17.6% 13.2% 9.1% 8.9% 8.0% 7.7% 7.5% 6.6% 6.3% 5.9% 1.0%  Percent of trees felled with an UE (severity 2,3) 0.0% 5.7% 0.0% 0.0% 2.0% 1.1% 1.7% 1.1% 0.0% 1.3% 0.0%  Summary of UE classified by severity code Severity 1 28 8 11 11 8 7 3 0 2 2 0 0 1 81  Severity 2 3 4 1 0 1 0 1 2 0 0 1 0 0 13  Severity 3 2 0 0 0 0 0 2 0 0 0 0 1 0 5  Total 33 12 12 11 9 7 6 2 2 2 1 1 1 99 18  3.2  Linear Mixed Effects Regression of MRC  Understanding the factors that lead to increasing the frequency of MRCs is important. In this study the MRC data are likely correlated, since the MRC observations for a tree are nested within each faller. In addition, site conditions such as tree size, tree species, slope, and terrain type can vary between fallers. A preliminary analysis was initially conducted on all the site specific covariates collected in this study (See Appendix C). The covariates that represented the greatest likelihood of being correlated with MRCs were presented in Equation 4 for further analysis. The  following two equations were used to determine whether intra-faller correlation was significant:  (  ) (4)  (  ) (5)  where  is the mean count of the number of MRCs occurring for the ith individual and the jth  observation , conditional on the predictor variables, and in the case of Equation 5, the random effect; i = 1,…,11 individual fallers, and j = 1,…n observations; µ is the population average effect; β1 to β12 are the fixed effects for the respective variables. The reference categories for each class variable are Tr(R), Sp(YC), and Wt(S). Equation 5 differs from Equation 4 since it is a generalize linear mixed model with following a  introduced as a random effect for the individual fallers,  distribution (i.e., Equation 4 is nested within Equation 5). Therefore the  significance of the random effect can be tested using likelihood ratio testing. Setting Equation 5 19  as the full model and Equation 4 as the restricted model, the LRT test statistic is -2λ = 3131.25 – 3042.47 = 88.77; therefore the random  effect is significant  .  The variables Tr, Sl, and Wt are site specific data. It may be more efficient to include only variables that were measured at the tree level. To determine if including only the variables measured at the individual tree level and the random effect offers a more efficient model consider Equation 5 and the following: (  )  (6)  where the variables in Equation 6 are the same as Equation 5 except that the Tr, Sl, and Wt have been removed. Equation 6 is nested within Equation 5. Setting Equation 5 as the full model and Equation 6 as the restricted model, the LRT test statistic is -2λ = 3049.19 – 3042.47 = 6.72; therefore, restricting the model to include only the tree level effects DSH and Sl, and the random effect is justified  . The estimates for the Type III fixed effects for  Equation 6 indicate that the variable Sp may not significantly contribute to the prediction of the count of MRCs  . A more efficient model including DSH and the  random intercept can be compared to Equation 6: (  )  (7)  Equation 6 and 7 are nested. Setting Equation 6 as the full model and Equation 7 as the restricted model, the LRT test statistic is -2λ = 3054.05 - (3049.20) = 4.85. Thus, restricting the model to include only DSH and the random effect is justified  . This  results in Equation 7 representing the most efficient model for this data set.  20  Before evaluating the fixed and random effects it is necessary to check the model for over-dispersion. Examining the plot of the conditional residuals versus the predicted values indicated the possibility of over-dispersion in the model, as the conditional residuals appeared not to be distributed equally around the zero line (Figure 3.1). Further evaluation using the Pearson statistic for the conditional distribution reveals a value of 0.72 for Equation 7, which is not considerably smaller or larger than 1.00 (Littell et al. 2006). This indicates the dispersion of the data is as expected under a Poisson distribution. Further attempts at fitting the model using a scale parameter and negative binominal distribution did not improve the fit. Therefore, the model was considered to fit the data reasonably well and no further adjustments were made.  Figure 3.1  Plot of the conditional residuals compared to the linear predicted values  21  The fixed effects for Equation 7 can be found in Table 3.3. The „estimates‟ are defined on the linked scale; in this case a log link function is used. The estimate for the fixed effect DSH indicated that the marginal predicted count value of the number of MRCs increased with increasing DSH. The conditional predicted value, which adjusts for the random effect, also increased with increasing DSH. The rate of increase and range of values is a function of the particular faller (Figure 3.2). The estimated variance between fallers (FallerID) is 0.214 with a standard error of 0.102, indicating that there was an individual faller effect on the number of MRCs that were observed. The individual estimates of the random effect for each faller for Equation 7 can be seen in Table 3.4. Of the 11 fallers that participated in the study, five had significant random effects (α =0.10) estimates while the remaining fallers effects were not different from zero. This indicates the rates that MRCs were reported varied between the fallers participating in this study.  Table 3.3 Effect Intercept DSH  Estimates for the fixed effects for Equation 7 Estimate -0.209 0.005  Std. Error 0.157 0.001  t Value -1.333 4.745  Prob.t 0.212 2.321E-6  C.I.(Lower) α =0.1 -0.492 0.003  C.I.(Upper) α =0.1 0.075 0.007  22  Predicted Number of MRCs  Figure 3.2  Table 3.4 Subject FallerID 108 FallerID 135 FallerID 164 FallerID 208 FallerID 386 FallerID 412 FallerID 537 FallerID 564 FallerID 617 FallerID 692 FallerID 861  Plot of the predicted number of MRC’s as a function of DSH  Estimates for the random effects for Equation 7 Estimate  Std. Error  t Value  Prob. t  -0.069 0.285 0.592 0.308 0.058 0.376 0.020 -0.012 -0.454 -1.125 0.105  0.169 0.247 0.171 0.177 0.235 0.164 0.186 0.196 0.155 0.217 0.154  -0.405 1.153 3.455 1.743 0.246 2.294 0.107 -0.062 -2.926 -5.179 0.681  0.685 0.249 0.001 0.082 0.806 0.022 0.914 0.950 0.003 <0.001 0.496  23  3.3  Logistic Regression of UEs  As before, the structure of the data collected leads to the possibility of correlation among observations for each faller. The following two equations were used to consider intra-faller correlation in the UE data. A preliminary analysis was initially conducted on all the site specific covariates collected in this study (See Appendix C). The covariates that represented the greatest likelihood of being correlated with UEs were presented in Equation 8 for further analysis.  (  ) (8)  (  ) (9)  where  is the probability of UE occurring for the ith individual of the jth observation; i =  1,…11 individual fallers; j = 1,…ni observations, µ is the population average effect; and β1 to β10 are the fixed effects for the respective variables The reference categories for each class variable are as follows: Tr(B), Sp(Bf), EF2(0), and CT2(0) . Equation 8 represents a generalized linear model (GLM) where the hierarchical structure of the data is ignored and the data are treated as if they were measured at the single tree level. Equation 9 is similar to Equation 8 except that it is a generalized linear mixed model with  as the random effect for the individual fallers. During the  24  initial testing of Equation 9 the „G‟† matrix was found to be „not positive definite‟. This result was caused by the variance of the faller (FallerID) effect being close to zero. Unlike the results for the MRC frequency analysis, the individual faller effect was not a significant factor in predicting the occurrence of UE. Although the individual faller was not significant in the occurrence of UEs, it is possible that certain descriptive characteristics of fallers such as years of experience (Fallexp) were significant. To consider the role years of experience may play consider the following equation:  (  )  (10)  The definitions of the variables in Equation 10 are identical to Equation 8 except that Fallexp has been added to adjust for the years of experience of each faller. Equation 8 is nested within Equation 10. If Equation 10 is set as the full model and Equation 8 as the reduced model, the LRT test statistic was -2λ = 614.38– 613.07 = 1.31. Consequently, adding the variable Fallexp was not justified  .  To evaluate if the number of MRCs identified played a significant role in predicting the occurrence of a UE, consider Equation 8 and the following: (  ) (11)  †  The G matrix is a q × q covariance structure matrix, where q is the number of random effect parameters.  25  Equation 11 is similar to Equation 8 except that the variable NumMRC has been removed. To test the significance of NumMRC, Equation 11 is set as the reduced model and Equation 8 as the full model. The LRT test statistic was -2λ =614.73 – 613.07 = 0.35. Therefore, the total number of MRCs did not play a significant role in the occurrence of UE  .  The estimates for the Type III fixed effects for Equation 11 indicate that the variables Sp and EF2  do not significantly contribute to  the prediction of UEs. To determine if removing the variables Sp and EF2 offers a more efficient model, the following equation was fit:  (  )  (12)  The variables in Equation 12 are the same as Equation 11, except that EF2 and Sp have been removed. Setting Equation 11 as the full model and Equation 12 as the reduced model, the LRT test statistic was -2λ =622.81 - 614.73 = 8.08. Therefore, removing Sp and EF2 from Equation 12 is statistically justified  .  Interactions among the fixed variables should be considered before examining the fixed effect. The logistic model with the interactions between the fixed variables is:  (  )  (13)  26  Equation 13 is similar to Equation 12 except for the addition of the interaction variables. The interaction variable Sl*Trjk was a linear combination of the Tr variable and was not included in Equation 13 to avoid singularity. Setting Equation 13 as the full model and Equation 12 as the restricted model, the LRT test statistic was -2λ = 622.80 – 618.89 = 3.91. Thus, the addition of the interaction variables was not statistically justified  and therefore  Equation 12 represents the most efficient model to predict the occurrence of UEs for this data set. Further analysis of the scale of the continuous variables DSH and Sl using both a univariate smoothed logit function and design variables based on the quartiles (Hosmer and Lemeshow 2000), revealed that both variables are best represented as linear in the logit function and no further modification of Equation 12 was considered. The ability of Equation 12 to predict the occurrence of a UE can be analyzed using a number of metrics. The Hosmer and Lemeshow goodness-of-fit test indicate that there is not a lack of fit in the model  . The area under the Receiver Operator  Characteristic Curve (ROC) is a measure of the model‟s ability to discriminate between subjects who experience the outcome of interest versus those who do not (Hosmer and Lemeshow 2000). The area under the ROC Curve for Equation 12 is 0.6603 (Figure 3.3). Hosmer and Lemeshow (2000) indicated that values for areas under an ROC Curve of 0.5, as a general rule, indicate a poor ability to discriminate, while values between 0.7 - 0.8 are considered acceptable levels of discrimination. Using these guidelines, Equation 12 has only a moderate ability to discriminate between events. The moderate ROC value is attributable to the numerous observations that had high predicted probabilities but did not actually have an observed UE (Figure 3.4). The moderate ability of the model to discriminate between events indicates that there is a lack of strong predictor variables related to the occurrence of UEs.  27  Figure 3.3  Area under the Receiver Operating Curve for Equation 12  28  Figure 3.4  Predicted probabilities for each observation classified by observed UEs  The estimates for the fixed effects for Equation 12 are summarized in Table 3.5. The estimates are defined on a linked scale; in this case the logit link function was used. Exponentiation of the parameter estimates yields the odds ratio which allows for a more intuitive understanding of the relationship between the dependant and independent variables (Karp 2000). Table 3.6 provides the estimated odds ratios with 90% confidence limits. For the continuous variables Sl and DSH, the odds ratio gives the change in odds for an increase of “1” unit in each respective variable. For DSH, the odds of encountering a UE increase with increasing tree size  29  (Figure 3.5). The extent to which MRCs affect larger trees is difficult for a faller to perceive and likely contributes to an increase in UEs. The odds of encountering a UE decrease with increasing slope percentage (Figure 3.5). For the dichotomous class variables CT2 and Terrain, the odds ratio represents a comparison of the various class levels. For CT2 the odds of encountering a UE are twice as high when a cut-up tree is present. When considering the type of terrain the faller is working in, gullied terrain (G) compared to even terrain (E) represents the only significant difference from the pairwise comparisons. The remaining terrain type comparisons have confidence intervals that span “1.00”, therefore it cannot be ruled out that there is no difference between the remaining pairwise types. A UE is 2.5 times more likely to occur in gullied terrain than it is in even terrain.  Table 3.5 Variable Intercept DSH Slope Terrain Terrain Terrain CT2  Fixed effects for Equation 12 Class Value  R G E 1  Estimate -1.780 0.012 -0.011 0.678 0.224 -0.693 0.351  Std. Error 0.490 0.004 0.005 0.635 0.324 0.251 0.180  Wald ChiSq 13.205 9.015 3.949 1.139 0.480 7.602 3.788  Prob. ChiSq 0.000 0.003 0.047 0.286 0.488 0.006 0.052  30  Table 3.6  Odds ratios for Equation 12  Effect DSH CT2 1 vs 0 Slope Terrain R vs G Terrain R vs E Terrain R vs B Terrain G vs E Terrain G vs B Terrain E vs B  Figure 3.5  Odds Ratio 1.012 2.019 0.990 1.573 3.938 2.426 2.503 1.542 0.616  C.I (Lower) α =0.1 1.005 1.115 0.981 0.361 0.992 0.571 1.431 0.788 0.366  C.I (Upper) α =0.1 1.018 3.655 0.998 6.864 15.628 10.302 4.380 3.017 1.035  Predicted probabilities for DSH and Slope for each observation  31  4. 4.1  Discussion MRC Frequency  One of the objectives of this study was to determine if the frequency of the conditions that have to be managed when falling a tree are correlated with site specific factors and whether the frequency varies among fallers. The results of this study indicate a positive relationship between the frequency of MRCs and DSH and that there was significant intra-faller correlation in the data. The positive linear relationship between the frequency of MRCs reported and DSH is likely attributed to both the size and age of a tree as well as stand type. Tree age was not measured in this study due to operational constraints. However, larger trees tend to be older in age and older trees are more likely to have rot and other structural defects. This likely increased the number of MRCs that were present and visible to a faller. In this study the majority of large trees were located in old growth stands which generally had a higher rate of snags, coarse woody debris, and undergrowth. Thus, the positive relationship between MRC frequency and DSH could be a combined effect of tree size and stand conditions. The significance of the individual faller effect suggests that fallers perceive and identify the conditions they have to manage differently. The literature supports differences among the perception of hazards (Blignaut 1979; Parker and Kirk 1993; Duffy 2003; Parker 2010); however, these studies only considered the difference between new and experienced employees. In this study all the fallers were considered to be experienced, and received their training prior to the establishment of the BCFTS; therefore, it was not possible to compare inexperienced to experienced fallers. It is possible that the faller effect is being confounded with other variables; however, this was not clear since most of the variables describing the site were not significant in  32  explaining the variation in this data set. Thus, it might just be personal differences in how the fallers view the threat that various conditions pose. It was common throughout the study for a faller to acknowledge that another faller may have recognized and perceived the MRCs differently for a given situation. Situational awareness for a manual tree faller is thought of as a bubble around the tree that a faller is working on; ideally the radius of the bubble is two tree lengths. In this study, terrain and other features made it extremely difficult, if not impossible, for the faller to be aware of everything that was going on within a two tree length area. Therefore, the size of the situational awareness bubble is subjective and depends on the conditions the faller is working in and the personal characteristics of the individual faller. Possible differences in the size of a faller‟s situational bubble could account for the variation among faller‟s MRC frequency. Fallers with a larger situational awareness bubble likely report more MRCs than those with a smaller bubble.  4.2  Unexpected Events  Initially, it was hypothesized that areas with certain terrain features such as a high degree of slope and broken or gullied terrain would represent areas of higher risk to fallers. This initial hypothesis was supported in literature (Anon 1985; Paulozzi 1987) and in a study of 32 falling fatalities where WorkSafeBC specifically concluded “the steeper the terrain, the greater the risk to the faller” (WorkSafeBC 2009). The results from this study indicate that fallers working in gullied terrain were 2.5 times more likely to experience a UE than fallers working in even terrain. Gullied terrain is typically characterized as terrain having steep-sided slopes that limit the choice of safe falling directions. When fallers are working in or around gullied terrain they will 33  attempt to reduce breakage and keep the stream channel clear by falling trees away from the main part of the gully, despite the fact that into the gully may be the natural falling direction of the tree. As well, some gullies contain active flowing streams and falling directly into these streams must be avoided when possible. Having to fall timber against its natural falling direction, combined with limited options for safely felling a tree, make falling trees in gullied terrain more difficult and likely led to an increase in the occurrence of a UE. Broken terrain was not found to represent an area of increased risk of experiencing a UE. Two possible reasons for this include: 1) only a small portion of the samples occurred on broken terrain, and 2) the broken terrain generally occurred along open falling faces in clearcut settings where the terrain did not severely constrain falling choices. The more options from which a faller can choose, the more likely they are to pick an option with a predictable and safe outcome. The odds of encountering a UE were found to decrease with increasing slope percentage. These findings disagree with the previous cited literature by Anon (1985) and Paulozzi (1987). However, Slope was not a strong predictor of UEs; the Slope coefficient was weakly negative with an odds ratio value of 0.99. With an odds ratio value close to 1.00 it is possible that further sampling in other areas may result in different magnitudes and directionalities of the slope coefficient. In this study, the largest trees were sampled on slopes between 0 – 20%. The lack of falling observations of larger trees on higher slopes does not allow for a conclusive determination of the effect of Slope as it is possible that other unmeasured factors attributed to DSH and Slope may be influencing the result. Trees with a larger DSH were found to significantly contribute to increasing the odds of encountering a UE. The degree to which rot, lean, tension, loading, and other structural defects affect larger trees can be difficult to determine. In large trees with a dense canopy, broken  34  branches and other debris can be obscured from a faller‟s view. These factors likely led to an increase in UEs for trees with larger DSH. The significance of cut-up trees (CT2) suggests that efforts to identify hazards present at the planning stage as recommended in the 2010 BC Coroners report may be overlooking the significant role that faller-made hazards play in UEs. In this study, a faller was twice as likely to experience a UE when a cut-up tree was present. Falling situations where a cut-up tree is present usually represents a scenario where the faller is attempting to push a tree down with another tree. The cut-up tree is either hung-up or it is unsafe for the faller to be at the base of the cut-up tree when it is felled. Often multiple attempts are required to push down the cut-up tree; this results in multiple UEs being recorded and increases the odds of encountering a UE. In situations where the outcome likely had the potential to cause serious injuries (UE with severity codes 2 or 3) cutup trees were found to be present in 25% of these situations. Gaskin (1988a) noted that the third greatest cause of faller fatalities from 1968 to 1987 was the use of “driving” or pusher trees. A study by Parker (1993) reported “driving” a tree as the most frequent hazard recorded. The use of a pusher tree adds a level of complexity and uncertainty to the falling situation which increases the likelihood of a UE occurring. Focusing effort on risk management at the planning stage (as recommended in the coroner‟s report) will not address this issue. The falling of a snag tree is generally thought to represent a higher risk to fallers due to the difficulty in predicting the degree to which rot has affected the holding wood and the stability of the tree. The fact that this study did not show snag trees to be significant is interesting and may be related to the abundance of snag trees in this study. Fourteen percent of the trees that were fallen in this study were snag trees and adjacent snags (EA2) were the second most frequent MRC recorded. The constant presence of snag trees and the relatively low occurrence of UEs  35  (6.9% of the falling observations) mean that many trees that did not have UEs also had snags. So although snags may present more problems, they do not account for the variation in the data. However, this does not mean that snag trees are not injuring fallers. Two out of the three incidents where a UE with a severity code of 3 was observed involved a snag tree. In both situations, adjacent snag trees were knocked over due to a loss of control of the falling tree. In the first case, the top of adjacent snag landed nearly two tree lengths away and close to another faller. In the second incident, the adjacent snag tree fell back over towards the faller after being knocked by a group of cut-up trees. The faller had to quickly move out of the way to avoid being seriously injured by the tree. Despite the lack of significance of snag trees it is clear from these two incidents that the presence of snag trees still represents a source of potentially serious injury to fallers. The lack of significance of factors related to the individual faller suggests that the UEs were not dependent on the age or experience of the faller. This appears to contradict many studies of forestry workers (Wolf and Dempsey 1979; Gaskin 1987; Gaskin 1988b; Klen 1988; Gaskin et al. 1989; Parker and Kirk 1993; Bentley et al. 2005; Parker 2010), that considered experience and age to be the primary factors which influenced the occurrence of serious injuries among workers. However, this apparent discrepancy may well be influenced by at least two important factors: (1) no inexperienced fallers were included in this study, which restricted the range of experience present; and (2) no serious injuries occurred. Therefore, it is possible that the occurrence of a UE does not vary with experience but serious injuries do. It may be experience that helps a faller avoid serious injuries from a UE. Falling occurs in an uncontrolled, dynamic environment where numerous complex factors make predicting outcomes difficult. Regardless of age or experience, fallers are likely not  36  physically or mentally able to perceive and plan for every hazard. A faller‟s experience in dealing with certain situations allows him to make estimated „best guesses‟ of the outcome given certain actions that are taken. However, fallers are not able to determine with certainty the outcome of every event. The lack of correlation between UEs and the number of MRCs suggests that even if a faller was able to perceive every hazard, the inability to predict the interactions between all the variables involved would still lead to an outcome that could not be known with absolute certainty.  4.3  Study Limitations and Further Research This study was constrained to a two year time period, where it was necessary to develop a  data collection protocol, collect field data, and complete the analysis of the data. This time constraint severely constrained the choice of industrial partners and the cutblocks used for data collection. Most of the fallers who participated in this study were unionized employees of major forest companies. It was acknowledged through personal conversations with fallers that had worked both as unionized employees and as contract fallers that there generally is an added sense of productivity pressure when working as a contract faller. It remains unclear if this sense of productivity pressure translates into an increase rate of UEs. Further study is needed to determine if the rate of occurrence of UEs is different in unionized falling environments compared to contract falling environments. In addition, the choice of cutblocks was not random but based on the participation of the fallers in the study and thus it was difficult to eliminate confounding effects. Examples of the confounding effects producing misleading results in this study can be seen in the negative correlation between slope and the occurrence of UEs. Further data collection  37  that specifically samples a variety of tree sizes on a variety of slopes would allow for confirmation of the slope coefficient.  38  5.  Conclusion Fatality and serious injury data is limited in its ability to improve faller safety because  there are only a few observations per year which is insufficient for identifying trends in the data, the information is second hand because there is no first hand record of what happened during the fatality, and there is no data on the work practices of workers who are not injured. In order to obtain sufficient data to look for trends in how conditions affect fallers, a methodology was developed where a faller reports the problems seen before falling a particular tree (management requiring conditions), the faller attempts to fall the tree, and after the faller has felled the tree any events that occurred that were not planned for before cutting into the tree (unexpected events) are reported. In addition, data on the actual conditions the faller was working in during the observation period were collected. In this study the diameter at stump height was the only fixed factor that was found to explain a significant portion of variation in the frequency of the MRCs reported. However, diameter at stump height is confounded with stand type in this study, where the largest trees were found in older decadent stands. It was found that a random intercept based on faller ID was also significant, indicating that intra-faller correlation is significant and that there were significant differences between fallers in the frequency of MRCs reported. It is not surprising that there are differences in the frequency of MRCs reported between fallers. Fallers commonly note that different fallers will see a particular problem differently, and in practice there will be differences in how fallers plan to fall a cutblock and individual trees. The rate that unexpected events occurred (i.e. number of trees out of 100) was independent of most of the factors that were measured. In the data the rate that UEs occurred was found to be related to ground slope in the working area, terrain type and stump diameter. The 39  data also showed that a faller was twice as likely to experience an UE when a cut-up tree was present in addition to the tree that was being felled. It is often necessary to fall limb-tied trees as a group or to push one tree with another to overcome a falling difficulty; therefore, having a cutup tree present is not an error on the faller's part. However, given the results of this study fallers need to continually remind themselves of the higher probability of unexpected events when falling trees near cut-up trees. It is not correct to say that each unexpected event was a mistake made by the faller. The fallers that participated in this study were experienced fallers, and because of the way the study was conducted (they were asked to report the MRCs they saw before cutting into the tree) their awareness of potential problems was heightened. The unexpected events recorded in this study indicate experienced fallers with a heightened situational awareness are still not able to predict with absolute certainty what will happen with the tree they are about to cut. This uncertainty is a normal part of falling trees. Given the uncertainty that a faller is exposed to when cutting into a tree, a major focus on faller safety must consider how to help the faller to be mentally and physically ready to adapt to changing conditions while cutting a tree. This study used experienced fallers to report the MRCs they saw before falling a tree and then to report how actually falling the tree differed from their initial plan. In the data collected for this study the MRCs, except for cut-up trees, were not good predictors of the likelihood of a UE occurring. This result highlights a number of important attributes of falling trees. First, it indicates that many problems are not visible to the faller until they begin to cut the tree. Second, for the problems the faller can see, they are usually able to implement a plan that mitigates these problems. Third, because many of the problems a faller faces are not apparent until he cuts into  40  the tree, planning based on data collected by non-fallers years in advance of falling will only have limited success in eliminating UEs.  41  References Anon. 1985. Analyzing logging injuries: who gets hurt and why? Forest Industries, 112(10), 32– 33.  BC Coroners Service. 2007. Statistics 2002 – 2006 Forestry-Related Deaths [online]. Available from http://www.pssg.gov.bc.ca/coroners/publications/docs/stats-forestryrelated-deaths-20022006.pdf [Accessed 20 January 2011]  BC Coroners Service. 2010. BC Coroners Service Death Review Panel Report (p. 10). Burnaby: BC Corners Service.  BC Forest Safety Council. 2011. Safety Statistics | BC Forest Safety Council [online]. Available from http://www.bcforestsafe.org/other/safety_info_and_tools/statistics/2011-0308/703013_files/frame.htm [Accessed 20 September2012]  BC Forest Safety Council. 2012. Safety Statistics | BC Forest Safety Council [online]. Available from http://www.bcforestsafe.org/safety_info/statistics.html [Accessed 15 September 2012]  Bentley, T. A., Parker, R. J., and Ashby, L. 2005. Understanding felling safety in the New Zealand forest industry. Applied Ergonomics, 36(2), 165–175. doi:10.1016/j.apergo.2004.10.009  Blignaut, C. J. H. 1979. The perception of hazard II. The contribution of signal detection to hazard perception. Ergonomics, 22(11), 1177–1183. doi:10.1080/00140137908924692 42  Duffy, V. G. 2003. Effects of training and experience on perception of hazard and risk. Ergonomics, 46(1-3), 114–125. doi:10.1080/00140130303524  Dunn, J. G. 1972. Subjective and objective risk distribution: A comparison and its implication for accident prevention. Occupational Psychology, 46(4), 183–187.  Gaskin, J. 1987. Analysis of lost-time accidents -1986 (4 No. 12) (pp. 1–6). Logging Industry Research Organisation Report, New Zealand.  Gaskin, J. 1988a. Analysis of Fatal Logging Accidents - 1968 to 1987 (No. Vol. 13 No.20). Logging Industry Research Organisation Report, New Zealand.  Gaskin, J. 1988b. Analysis of lost time accidents - 1987 (Accident Reporting Statistics) (4 No. 13) (pp. 1–4). Logging Industry Research Organisation Report, New Zealand.  Gaskin, J., Smith, B., and Wilson, P. 1989. The New Zealand logging worker: A profile (Project Report No. 44) (p. 48). New Zealand Logging Industry Research Organization Report, New Zealand.  Hosmer, D. W., and Lemeshow, S. 2000. Applied logistic regression. 2nd ed. Wiley, New York.  43  Karp, A. (2000). Getting Started with PROC Logistic. In Proceeding of the Twenty-Fifth Annual Proceedings of SAS Users Group International Conference (SUGI.25). Presented at the SAS Users Group International Conference, Indianapolis, Indiana: SAS Pub.  Kawachi, I., Cryer, C., Marshall, S., Wright, D., and Slappendal, C. 1991. The epidemiology of work related injury among forestry workers in New Zealand- Results from a data-linkage study. Unpublished manuscript.  Klen, T. 1988. Subjective and objective risk estimate in logging work. Presented at the International Conference on Ergonomics, Occupational Safety, and Health and the Environments, Beijing, October 24-26.  Kubiak, M. 1985. Pattern of forestry accidents and dynamics of their increase. Sylwan, 129(12), 13–21.  Laughery, K., and Wogalter, M. S. 1997. Warnings and risk perception. In Handbook of Human Factors and Ergonomics. 2nd ed. Wiley, New York. pp. 1174–1197  Littell, R.C., Milliken, G.A., Stroup, W.W., Wolfinger, R.D., and Schabenberger, O. 2006. SAS for Mixed Models. 2nd ed. Cary, N.C.: SAS Institute Inc.  44  Östberg, O. 1980. Risk perception and work behaviour in forestry: Implications for accident prevention policy. Accident Analysis & Prevention, 12(3): 189–200. doi:10.1016/0001 4575(80)90018-4  Parker, R. J. 2010. Technological Advances in the Analysis of Work in Dangerous Environments: Tree Felling and Rural Fire Fighting. PhD Dissertation, Massey University, New Zealand.  Parker, R. J., and Kirk, P. 1993. Felling and Delimbing Hazards . 18(22). Logging Industry Research Organisation Report, New Zealand.  Paulozzi, L. 1987. Fatal logging injuries in Washington state, 1977 to 1983. Journal of Occupational and Environmental Medicine, 29(2), 103.  Peters, P. 1991. Chainsaw felling fatal accidents. Transactions of the ASAE, 34(6), 2600–2608. SAS Institute. (2008). Sas/Stat 9.2 User‟s Guide. The Glimmix Procedure (Book Expert). Cary, N.C.: SAS Institute Inc.  Western Forest Products. 2006. Management of Dangerous Trees, Findings and Recommendations of the Snag Team [online]. Available from http://www.safer.ca/docs/alerts06-09-04-dngr_trees_wfp.pdf [accessed 5 July 2012].  Wolf, C. ., and Dempsey, G. 1979. Logging work in Appalachia. Lumberman, 75–82.  45  WorkSafeBC. 2009. Occupational Health and Safety Faller Serious Injury and Fatal Review [online]. Available from http://www2.worksafebc.com/pdfs/forestry/Faller_Review_2009.pdf [accessed 17 January 2011].  WorkSafeBC. 2010. Forestry Statistics [online], Available from http://www2.worksafebc.com/Portals/Forestry/Statistics.asp [accessed 10 September 2012]  WorkSafeBC. 2012a. Statistics2011-web- Statistic Report 2011[online]. Available from http://www.worksafebc.com/publications/reports/statistics_reports/assets/flipbook/2011/index.ht ml [accessed 12 October 2012].  WorkSafeBC. 2012b. Young Worker Statistics - Statistics: WorkSafeBC Terms of Reference [online]. Available from http://www2.worksafebc.com/topics/youngworker/statistics.asp?reportid=32331 [accessed 12 October 2012].  46  Appendix A: Management Requiring Condition (MRC) and Unexpected Event (UE) Code Definition  MRC Codes for Trees Being Felled  Snag Tree (EF2) – A standing dead tree. Structural Defects (EF5) - Any structural defect such as rot, cracks, or disturbed roots that may affect the falling of the tree. This includes the potential for the defect to affect the holding wood, the ability to wedge the tree, or the defect makes the tree sufficiently weak to be a safety concern to the faller. Major Imbalance (EF6) – A tree that contains a major imbalance resulting from a combination of lean, sweep, or asymmetric crown. The imbalance is a concern under the following conditions:      The imbalance is sufficient enough to be a barber chair concern, particularly in species such as cypress and deciduous trees. The imbalance is opposite to the direction of fall and it is likely the faller will not be able to wedge the tree over in the direction of fall. The imbalance is perpendicular to the direction of fall and it is likely the holding wood will not be able to control the direction of fall. A situation where a large limb creates an imbalance in the tree that makes it difficult for the faller to interpret which direction the tree may fall.  Deciduous (EF7) – This code is used to identify all non-conifer trees. Debris in Canopy (EF8) – This code includes a hung-up top or limb(s), loose bark or other material that is an overhead hazard. This will be called an MRC only if it is seen before falling commences. If the debris is noticed after falling commences then it will be reported as an unexpected event. Hazardous Top (EF10) - A tree is defined as having a hazardous top if it contains:    A dead top that is still attached, A single or multi-stemmed top with danger indicators of decay (conks, woodpecker cavities) or other signs of structural weakness, or  47    The faller‟s view of the upper part of the tree is severely compromised due to heavy limb growth and the faller has particular concerns about unforeseen hazards in the top of the tree. If the tree contains a broken top or limb then MRC code EF8 is used. Limb Tied (EF11) - A tree is considered limb tied if there is the potential that the faller will not be able to wedge the tree over in the direction of fall.  MRC Codes for Adjacent Trees  Snag Tree (EA2) – An adjacent standing dead tree. Structural Defects (EA5) - Adjacent tree(s) that have any structural defect such as rot, cracks, or disturbed roots that have the potential to be a danger to the faller. Major Imbalance (EA6) – An adjacent tree with a major imbalance resulting from a combination of lean, sweep, or asymmetric crown. The major imbalance is reported if:     The imbalance creates an overhead hazard for the faller while working on another tree. The imbalance lean creates an overhead hazard for the faller's escape route. The imbalance creates the possibility of a complicated chain reaction that may threaten the faller.  Adjacent Debris in Canopy (EA7) – This code includes an adjacent hung-up top or limb(s), loose bark or any other material that may represent an overhead hazard. Debris in the adjacent tree will be called an MRC only if it is seen before falling commences. If the debris on the adjacent tree is noticed after falling commences than it will be reported as an unexpected event. Adjacent Hazardous Top (EA9) - An adjacent tree is defined as having a hazardous top if it contains:    A dead top that is still attached, A single or multi-stemmed top with danger indicators of decay (conks, woodpecker cavities) or other signs of structural weakness, or  The faller‟s view of the upper part of the tree is severely compromised due to heavy limb growth and the faller has particular concerns about unforeseen hazards in the top of the tree. If the adjacent tree contains a hung-up top or limb then MRC code EA7 is used.  48  Adjacent Limb-tied (EA11) - This occurs when adjacent trees are limb-tied and the manner in which they are limb-tied creates an overhead hazard. This code is only recorded when a faller is not directly trying to remove the hazard of the adjacent limb-tied trees but is aware of the danger it represents. If the faller is directly removing the limb-tied trees then EF11 is used.  Other MRC Codes  Compromised Escape Route (EO2) - Under this definition an escape route still exists for the faller but it may not be ideal, or a significant amount of effort is required to clear an adequate escape route. Falling Uphill (EO4) – This occurs when the faller is required to fall the tree uphill due to terrain, obstacles or other hazards that prevent the faller from falling the tree along the face of the cut block or downhill. Unstable Debris (E05) – This definition includes any debris that has the potential to strike a faller while he is falling a tree. Tree in the Direction of Falling (EO6) - This is defined as the possibility that the tree being felled will come in contact with any part of an adjacent tree. Obstacle affects Falling (EO7) - An obstacle will affect falling if the faller is required to significantly adjust the preferred falling direction of the tree or the faller believes the falling tree may come in contact with the obstacle. If the obstacle is another tree, EO6 will be used. Slope will be considered under this definition if it limits the ability of the faller to make cuts on a tree or if it affects the faller‟s escape route. Inadequate Working Area (EO8) - Inadequate working area is defined when the faller is required to take action to deal with a number of hazards that are present on the ground surrounding the tree that he is falling. This includes the removal of brush, small stems (<25cm in diameter), or objects to improve visibility, remove hazards that could potentially strike the faller, or improve the general working area around the tree being felled.  Created MRC Codes  Adjacent Cut-Up Tree (CT2) – This is a standing tree with falling cuts (cut-up) that is adjacent to another tree that a faller is currently working on. 49  Fence Post (CT4) - This occurs when a faller is required to use a “fence post” falling technique to remove a tree that has fallen and become hung up.  UE Codes for Tree Being Felled  Object Falls out of Canopy (UET1) – This occurs when any object falls from the canopy and the faller had failed to notice the object prior to falling. Falling Direction Change Due to Tree Hitting another Object (UET2) – This code is used to describe any deviation of falling direction caused by the tree hitting any part of an adjacent tree or other objects including rock outcrops, boulders or other terrain features. Falling Direction Change Due to Wind (UET3) – Occurs when the wind alters the falling path of a tree in any direction other than the intended falling direction. Falling Direction Change Due to Other Reasons (UET4) – This code is used to describe any unexpected change in the intended direction of fall caused by reasons not mentioned in UET2 and UET3. Barber Chair (UET6) – This results when a tree unexpectedly splits vertically before the hinge is cut thin enough to bend. Tree Hangs -Up (UET7) - This code is used when a tree does not fall to the ground due to becoming hung-up in another tree. This definition includes both unexpected hang-ups and situations where the faller was unaware that the tree was limb-tied. Tree can’t be Wedged Over (UET8) - This code is used when the faller cannot wedge the tree over and did not call a major imbalance before commencing falling. Tree in Group Falls Early (UET9) - To record an event under this code the faller must identify that he is falling the trees in a group. The unexplained event occurs when one of the cut up trees falls before the faller intends it to. If more than one tree falls early than each tree is treated distinctly and multiple UET9 codes are recorded. Unexpected Rot Resulting in the Loss of Control of the Falling Tree (UET14) – Occurs when the faller was not aware that the tree contained rot until cutting began and the faller was not able to compensate for the rot prior to the tree falling. The faller is no longer in control of the direction of fall for the tree. Falling Tree knocks Over another Tree (UET15) - This occurs when the faller did not foresee the potential for the falling tree to knock over another tree when it came into contact with it. 50  Saw Pinched (UEB5) – This occurs if a faller unexpectedly has to adjust his bucking or falling technique to avoid saw becoming pinched. A severity ranking of 1 will be used if the faller is able to adjust cuts to avoid the saw being pinched. A severity ranking of 2 will be used if the saw has become pinched and the faller has difficulty dislodging it and other methods such as using wedge(s), axe or another chainsaw must be used to free the pinched saw. A severity 3 is not applicable in this case.  Other UE Codes  Root Dislodged (UEO2) - This occurs when a falling tree dislodges its roots and stump while falling to the ground. Fall or Trip (UEO4) – This occurs when a faller trips or falls while either falling or bucking a tree.  51  Appendix B: Subject Data Collection Information Subject Data Form Identity Code: Years of manual tree falling experience Coastal BC: Total years of manual tree falling experience: Age: Sex: Are you a certified manual tree faller (BCForestSafe): Did you complete the BCForestSafe faller training program:  52  Appendix C: Equations Used in Preliminary Analysis  MRC Frequency Analysis:  (  )  is the mean count of the number of MRCs occurring for the ith individual and the jth  where  observation; i = 1,…11 individual fallers; j = 1,…ni observations, µ is the population average effect; and β1 to β18 are the fixed effects for the respective variables and  is the random effect  for the individual fallers. The variables are defined as follows: diameter at stump height (DSH), slope (Sl), terrain (Tr), species (Sp), weather (Wt), roughness scale (Rc), undergrowth (Ug), and wind scale (Ws). The reference categories for each class variable are Tr(R), Sp(bf), Wt(S), Ug(1), and Ws(1).  UE Analysis:  (  )  53  where  is the probability of UE occurring for the ith individual of the jth observation; i =  1,…11 individual fallers; j = 1,…ni observations, µ is the population average effect; and β1 to β22 are the fixed effects for the respective variables and  is the random effect for the individual  fallers. The variables are defined as follows: diameter at stump height (DSH), slope (Sl), terrain (Tr), species (Sp), number of MRCs (NumMRC), snag tree (EF2), cut-up tree (CT2), faller experience (Fallexp), weather (Wt), roughness scale (Rc), undergrowth (Ug), and wind scale (Ws). The reference categories for each class variable are Tr(R), Sp(bf), EF2(0), CT2(0), Wt(S), Ug(1), and Ws(1).  54  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0073689/manifest

Comment

Related Items