ANALYZING IRREGULAR WORKING HOURS LESSONS LEARNED IN THE DEVELOPMENT OF RAS 1.0 THE REPRESENTATION & ANALYSIS SOFTWARE

Transcription

1 ANALYZING IRREGULAR WORKING HOURS LESSONS LEARNED IN THE DEVELOPMENT OF RAS 1.0 THE REPRESENTATION & ANALYSIS SOFTWARE Gärtner Johannes A, Popkin Stephen B, Leitner Wolfgang A, Wahl Sabine A, Åkerstedt Torbjörn C, Folkard Simon D A XIMES GmbH, Schwedenplatz 2/ Vienna (Austria) gaertner@ximes.com Tel: , Fax B Volpe National Transportation Systems Center - 55 Broadway, Cambridge MA (USA) C Karolinska Institutet, Tomtebodavägen 11F, Solna, Stockholm (Sweden) D University of Wales Swansea, Body Rhythms and Shiftwork Centre, SA2 8PP Swansea (UK) ABSTRACT BETWEEN WORDS Actual working hours of employees vary widely, especially in the transportation industry. We developed a tool RAS (Representation and Analysis Software) to ease the assessment of such irregular hours and the transfer of existing knowledge of proper schedule design to the problem of irregular hours, This paper discusses several critical design questions that were addressed during software development in order for it to assess irregular work patterns, including: 1) the importance yet lack of established definitions of basic concepts like e.g., night shift, 2) the difficulty to model and adapt existing knowledge on proper design, and 3) the large number of analysis methods and additional data in addition to the company schedule necessary to meet the needs of various research groups. This article describes how RAS addresses these three issues by illustrating its application to the work schedule of a train driver involved in the Hinton train disaster. KEY WORDS Irregular working hours, shift-scheduling, software, information visualization of working hours INTRODUCTION Actual working hours of service industry employees, and particularly those within the transportation sector, often vary widely from day to day and week to week. These shifts may be highly irregular and unpredictable, and utilize split shifts and extended work hours. Given the complexity of some of these systems, it is often very difficult to analyze and compare their relevant schedule features. A large body of knowledge has been developed during the past 70 years on the effects of night and shiftwork, including recommendations regarding the assessment and design of shift schedules, e.g., (Knauth, 1996; Marek et al, 2000; Wedderburn, 1991). These recommendations, however cannot be applied directly to irregular and unpredictable work hours. There are several reasons for this. First, very short shifts of just a few hours and very long shift lengths (e.g., more than 10 hours), either in a continuous or mixed irregular series, may have different effects on employees than the classical (archetypical?) 8h shifts. Second, the variation in working time starting points may lead to different consequences, e.g. a morning shift starting at 4:30 may cause similar consequences as a night shift, whereas a morning shift starting at 6:30 is closer to a day shift. Third the rate at which a schedule may change and the degree of schedule predictability for employees as well as the control they have on such changes may also influence the overall impact of a schedule on worker safety and health. In order to address these scheduling issues the US Department of Transportation (DOT) funded and worked with XIMES GmbH between the years 2001 and 2003 through a US government Broad Agency Announcement (BAA) process to develop the RAS Representation & Analysis Software. The BAA process is a competitive solicitation procedure used to obtain proposals for basic and applied research and that part of development not related to the development of a specific system or hardware procurement. The Government drafts a statement of the problem or general research interest, in this case on fatigue management technologies. Each offeror proposes its own solution to this general problem statement. Proposals are considered to be stand-alone documents and therefore not compared to one another but rather whether or not it is considered appropriate, feasible and desirable by the

2 review committee. In effect the BAA is a technical competition in the "marketplace of ideas" where cost or price is rarely the deciding factor on the winning proposal. The goals of this software were to: 1) Provide a means of easily assessing and comparing different working time arrangements, and 2) Provide a vehicle for transferring and applying existing knowledge drawn from work scheduling research. Both of these goals lead to a better understanding of the impact of working time arrangements, especially those that are irregular and unpredictable, on worker safety and health. Two primary user communities were identified as potentially benefiting from this type of work schedule analysis tool, the scheduling practitioner and shiftwork and sleep researchers. The visual representations of work/rest data may facilitate the reading and interpretation of a work-schedule s impact on safety and fatigue, foster the identification of safety critical parts of that schedule (e.g., duration and timing of night hours, off time, etc.), and ultimately help the professional to develop and evaluate alternatives to it. This software was designed to work as a stand alone extension to users existing scheduling package and practices. The RAS software was developed through a series of anticipated use scenarios (e.g., a company scheduler developed a new schedule and wants to compare it to the current schedule by evaluating it for critical elements) that span a wide range of possible applications for the various modes of transportation. Responses to these scenario questions are typically written in a short story format, and include descriptions of operationally relevant issues. These responses are evaluated by considering the type and format of input data these users have available, and related technical restrictions with these data. These use scenarios and initial software prototypes were presented to a group of potential end users over the course of several design meetings. These design sketches were discussed at the 2001 Symposium on Night- and Shiftwork ((Gärtner and Popkin, 2001)). Conjointly, the software was tested in the lab by running work schedule data collected from field research iteratively through several prototypes, following modern techniques of software development (e.g., Extreme Programming (Beck, 1999)). While many improvement ideas were generated through this process, user feedback indicated that the respondents were able to easily understand how to manipulate the software and that there are only few difficulties for practical use. This paper discusses the lessons learned in designing this software tool. The next section discusses several fundamental design questions and methodological issues in developing RAS, and is followed by a brief description of the actual software and an example of its real-world application. Lastly, the benefits and limitations of the software are reviewed as well as consequences for future tool development in this field. CRITICAL DESIGN AND METHODOLOGICAL ISSUES The design of such software touches upon a number of design and methodological issues. Three of the most critical for this project were: 1. Specifying the overarching product development approach, 2. Developing statistical techniques for the analysis of irregular hours, and 3. Creating a knowledgebase on irregular work hours to be used when analysing work schedules The Overarching Approach A fundamental design decision was made at the start of the project for the software to not recommend one schedule over another. While scheduling practitioners may like to have this feature, there are two strong arguments for not including it in this product. First, since the body of research as well as the large variety of possible scheduling situations do not allow for such straightforward answers, it was felt that the software output could only be a basis to better inform schedulers and decision makers. Decision makers would still have to take into account all the specific contextual information, not all of which is considered by the software. Second, given that there is usually no simple, direct solution to a scheduling problem, there is the risk of liability claims that neither the US DOT nor XIMES could accept. The Techniques Developed to Analyse Irregular Hours Analysing irregular hours is a difficult task. While some features of irregular hours can be computed directly, e.g. sum of hours worked over a period of time, most issues are not so straightforward. One

3 difficulty is with definitions, e.g. one has to define what constitutes Night Work in order to be able to calculate the number of hours of Night Work or to check whether there are few or many night shifts in a series. Small changes in such definitions may cause substantial changes in the results: E.g., assume the following working time: Day 1 21:00 to 3:45, Day 2: 23:30 to 5:00. If we define night work as working 6 or more hours between 22:00 and 6:00, there would be no night-work. However, if we defined night work as working between the hours of 2:00 and 3:00, both shifts would be considered night-work. There are a number of such concepts to be defined prior to schedule analysis, including Evening Shifts, Morning Shifts, Weekends, etc. An even more complex issue to address is the issue of whether two periods of work should be considered as one shift (i.e., split-shift) or as two shifts. E.g., assume a person works from 5:00 11:00 and from 18:00 to 22:00. Is this one shift covering the range from 5:00 22:00 with 10h of work? Or are these two separate and distinct shifts, each very short. It is often difficult to foresee the consequences of these small changes to working definitions, which therefore necessitates iterative user evaluation meetings. As there was limited time to work on these definitions and as it is questionable whether there are definitions that can be generally applicable, we took the approach of creating a facility to change these definitions from within the software itself. Preset values were provided, though, as a s tarting point, in accordance to stakeholder preference. The Process used to Develop a Knowledgebase on Irregular Work Hours for use when Analysing Work Schedules There are many work schedule-related recommendations based on night and shiftwork research specific to settings shift starting times for shift durations of 8h (6:00, 14:00, 22:00) and 12h (6:00 and 18:00). Some research has been done in occupations such as healthcare provider on shift lengths, but schedule irregularity is barely addressed. In some research papers even exact times of the shifts investigated are not defined for a discussion of a standardized description of shift schedules see (Gärtner and Folkard, 2001; Tepas, Paley and Popkin, 1997). The transfer of knowledge out of this archetypical context to irregular hours brings up two main problems: First how to consider different shift length? Second how to consider different starting and ending times? For example, we may consider different work characteristics such as work load, working conditions, average working hours, etc. to determine that three night shifts (with 8h length, starting at 22:00) to be acceptable from a safety and performance aspect but four such shifts would increase risk of injury or accident to an unacceptable level (Folkard, 2000). What does this tell us about the following 4 shifts (1 st 6h starting at 22:00, then 9h 21:00, 5h 23:00, 8h 21:00)? Some shifts are shorter, and some times are different; would these four shifts together still be considered unacceptable? There are several possible approaches to address this issue. One approach is to develop models of e.g. risk that are able to handle such settings in general while also being able to consider workload, working conditions, etc. While there has been research on this topic since 1991 (Folkard and Åkerstedt, 1991), the models are still rudimentary and not ready for commercial application. A drawback of such a modelling approach is that it is very difficult to let managers, planners dispatchers, and members of the workers council participate in the adaptation of the model to the specifics of the situation, especially if the model is very abstract. Participation however is important in general (Hsu et al, 1993) and if properly done allows considering the experiences and specifics of the situation. Examples for software support focusing on participation do exist in general and also in the field of shift scheduling (Gärtner and Wahl, 1994; Gissel and Knauth, 1998). If models are abstract it will be difficult to let users change parameters as it will be hard for them to relate these parameters to their day to day experiences. A change in the corresponding parameters is difficult to assess by the industry user, and would generate a need to bring in a work scheduling expert, thereby adding cost and delay to the redesign, or worse, random parameter selection by the user, leading to flawed analyses and comparisons. An alternative approach might be meta-modelling, i.e., rules how to refine a model under specific circumstances. However this approach in still under development in the literature, and does not fall easily under the rubric of participative design. In order to simplify user participation we developed a different approach. In this approach users are asked to specify acceptable and critical levels of features for 8h-shifts and for 12h shifts separately (e.g., what number of days in a row should be the maximum). We assumed that such judgements

4 should be easier for users as it more closely resembles their day to day work Vector mathematics was then applied to interpolate values between and beyond these definitions. THE SOFTWARE RAS The following provides a brief description of the RAS software. RAS supports work schedule assessment for regular but also for irregular work schedules. Several features of RAS are shown in the following examples. The data set used comes from the Hinton Train Disaster. The Hinton Train Disaster occurred in 1986 close to Hinton, Alberta, Canada and involved a passenger train and a freight train carrying chemicals. These trains collided resulting in the death of more than 20 people and a large toxic chemical spill. The accident investigation led to a passionate discussion on whether excessive working hours contributed to this accident. Overall Structure and Display of Working Times The overall user interface of the RAS is similar to that of MS-Explorer. In Figure 1, on the leftmost side, users see a tree structure of software actions, e.g., define the settings, define the checks to be applied, enter the data and analyse the results. The work pattern analysis is selected in this figure. Similar analyses could be conducted for sleep and for time off. RAS provides various graphs to facilitate the assessment of irregular hours, time off and sleep. For work patterns one has the additional option of selecting which activities or combination of activities to be displayed (e.g., work and wait time). The graph may be used to show a double raster plot, i.e., for every day the following day is shown in the some row, which makes it easier to check e.g. for the distribution of night work. Additional data sheets would allow for analysis of shift frequency, duration, etc. Similar analyses are provided for sleep and time off. There are a few restrictions in distinguishing different categories of time (e.g., break, commuting, on duty, on call). Clear graphical representations substantially ease the assessment of data and allow users to accomplish their tasks quicker or even to be able to accomplish more complex tasks at all. The data of the Hinton train disaster displayed in RAS shows the high proportion of night work, a few numbers of nights without work, long stretches of work and the split shifts in one representation. When does Work Start and When does Work End? Start and End Times of entered working times are graphically displayed and statistically analysed. The graphical display is circular, the circle representing a 24-h-clock. Each Start time is displayed using a dot on the display. If more than one duty period starts at the same time the dots begin to trail towards the center of the circle to mark their frequency. The graphic in Figure 2 very clearly shows the regularity, or lack thereof, of a schedule with respect to start and end times. Dots distributed evenly over the circle indicates very irregular working times. If the dots accumulate in certain places on the circle it is an indication for schedule regularity in Start or End Times. Additionally the Start and End Times are statistically analysed. Clusters can be generated by the system using different methods (e.g., compute the clusters 2h maximal distance between nearest neighbours, 6h maximal distance between first and last element of the cluster). In analysing the data of the Hinton train disaster, the irregularity of the schedule becomes immediately visible. In addition, there are a substantial number of starts before 6:00 am. Work Shift Description The work shift description provides a concise summary of many relevant components of the work schedule. As previously discussed, this is where the majority of effort was expended in developing precise definitions and measurements. The data displayed in Figure 3 shows a large number of features of schedules. These data would also allow a comparison of different schedules. In the case of the Hinton train disaster it very clearly shows the irregularity of the schedule and the high workload. To pick just a few measures, in the case of the Hinton train disaster, minimum working hours per week are close to 40h, maximum is close to 60 and

5 the average is 48h. Even more striking is the variation in shift length from 2h30m to 15h20m. The number of split shifts per week averages at There is only one morning shift prior to rotating onto a night or evening shift. Rotations after afternoon shifts and night shifts rotate are also very fast. The employees only had single days off, never two days off in a row. Transfer and Apply Existing Knowledge The transfer of recommendations developed for the design of regular shift systems to irregular hours is done by defining acceptable, problematic and very critical levels of particular features of a schedule for 8h and for 12h shifts. RAS allows for the defining of these thresholds and interpolates for other shift length. Users together with experts should ideally do the tailoring of these definitions. As depicted on the top of Figure 4, the evaluation of a series of duties is displayed using different colours in the respective cells. The evaluation chart consists of the following attributes: Critical (=red, dark grey in print), Problematical (=orange, light grey in print), OK (=white), "Good"(=green, not used in this figure, light grey). E.g., if a series of Afternoon shifts with a length of 5 is considered Problematical the Rota Résumé will display all Series of afternoon shifts with a length of 5 or more with the attribute Problematical. When checking the data of the Hinton train disaster against those recommendations, it becomes apparent that many requirements are not met. DISCUSSION AND FUTURE ISSUES Stakeholder feedback indicates that the product is on track for meeting its intended goal as a transportation work schedule analysis tool. However, there are still a number of areas for improvement: Some of the research stakeholders consider the definitions to be less flexible than necessary while other user groups have difficulty in adapting them to field situations. As mentioned, it is difficult to foresee the consequences of definitions. Possible solutions to this problem with regard to the software may be an improved user interface (e.g. allowing for more types of definitions and providing example data that immediately show the consequences of a proposed definition). Also a discussion within the scientific community on the topic of work schedule definitions seems to be necessary. Interpolation in order to compute qualitative assessments is currently based on two points, 8h and 12h. This alone is probably insufficient for the addressing the growing heterogeneity of start time. Again better visualizations as well as a discussion of typical default values for certain occupations are needed. Perhaps a more general approach may be to allow for more points or to apply Artificial Intelligence techniques like case based reasoning. Even though there are already many data visualizations and computed statistics, researchers demand even greater flexibility in their use of the software. This demand calls for more open definitions, i.e. expert users should be able to define their own categories and calculations. Given that a number of research groups work on questions of sleep, risk, productivity etc it would be a useful approach to build interfaces to such models to be able to use them either directly or in helping to adjust the definition and recommendations to a particular schedule. Here it may also be useful to built batch processing capability to allow the analysis of a large numbers of irregular schedules at once. In order to ease use by the industry schedulers, a closer integration of RAS to their day to day business would be helpful. Currently the software is implemented as single user software. However, there would be a great advantage if RAS could more directly work with a company s scheduling software, or be useable over the Internet through a web-interface. To go to an even more fundamental level: currently researchers analyze specific irregular schedules. However it might be important to include data in the analysis that makes it possible to consider emerging concepts like:: how far ahead to employees know there schedule, how often and how strong does it change, who makes the decisions, etc. Overall the analysis and assessment of irregular hours is a large and complex issue. RAS 1.0 addresses several important questions related to this endeavour - however it is only a first step. REFERENCES Beck, K. (1999). Extreme Programming Explained: Embrace Change, Addison-Wesley Pub Co.

6 Folkard, S. (2000). Shiftwork and accidents. In T. Marek, H. Oginska, J. Pogorski, G. Costa and S. Folkard (Eds.) Shiftwork Implications for science, practice and business.. Krakow, Jagiiellonian University: Folkard, S. and Åkerstedt, T. (1991). A three process model of the regulation of alertness and sleepiness. In R. Ogilvie, R. T. Wilkinson and R. J. Broughton (Eds.) Sleep, Arousal and Performance: Problems and Promises.. Boston, Birkhäuser: Gärtner, J. and Wahl, S. (1994). Participative Organizational Planning of Shift Schedules. Participatory Design Conference - PDC 94, Chapel Hill - North Carolina, Computer Professionals for Social Responsibility. Gärtner, J. and Folkard, S. (2001). Preparing for Meta-Research. XVth International Symposium on Night and Shiftwork, Hayama, Japan. Shiftwork International Newsletter 18/1: 20. Gärtner, J. and Popkin, S. (2001). Work schedule representation & analysis software. XVth International Symposium on Night and Shiftwork, Hayama, Japan. Shiftwork International Newsletter 18/1: 25. Gissel, A. and Knauth, P. (1998). Knowledge-based support for the participatory design and implementation of shift systems. Scandinavian Journal of Work, Environment & Health 24: Hsu, W. L., P. M., Thompson, G. and Peng, S. (1993). A mixed initiative scheduling workbench - integrating AI, OR and HCI. Decision Support Systems 9: Knauth, P. (1996). Design of Shiftwork Systems. In P. W. Colquhoun, G. Costa, S. Folkard and P. Knauth (Eds.) Shiftwork - Problems and Solutions. Frankfurt am Main, Peter Lang - Europäischer Verlag der Wissenschaften. Marek, T., Oginska, H., Pogorski, J., Costa, G. and Folkard, S. (2000). Shiftwork Implications for science, practice and business. Krakow, Jagiiellonian University. Wedderburn, A.(1991). Guidelines for Shiftworkers. BEST - Bulletin of European Studies on Time. Dublin, European Foundation for the Improvement of Living and Working Conditions.

7 FIGURE LEGENDS Figure 1: On the very left the navigation tree of RAS is shown. On the right side the work pattern is double plotted and a few statistics are shown. The bottom of the screen contains tabs for further analysis sheets. Figure 2: Starting times of the example data showed in a 24h clock. Figure 3: The work shift description displaying a number of relevant features of irregular hours Figure 4: The Good, Acceptable, Problematic and Critical levels of features are defined by the users (definitions for 8h are shown).

8 FIGURES Figure 1 Figure 2

9 Figure 3 Figure 4