News Article 31/08/2009

Categorising data -the real problem
Tanker Operator, August/September 2009, page 24-25

In the information age, data is abundant and statistics are everywhere. But why don’t they help us make decisions better? Why do we keep blaming the data?

For years computers have been helping us collect data and draw statistics. But how often do we see a graph and are unable to make a decision because we know that the data capture fails to tell us the whole story about any of the data points?

The problem with decisions is that people’s minds do not balance all the competing influences at any one time. We need to get a balanced view of all the influences to make good decisions. While the need is to integrate the many conflicting goals affecting our decisions, depending on our state of mind and the circumstances, typically, one or other of the goals prevails, to the detriment of a well rounded and informed decision.

In business we take more time and apply methods to balance goal conflicts and evaluate obstacles. Tough ones like how to assess the market in chartering, or whether to postpone a stern tube repair on a tanker are made with every intention of balancing the conflicting elements and overcoming the obstacles. We could even look at statistics to help make these decisions more balanced. For example, we could use statistical counts of seasonal changes in the charter-market or deterioration rates of stern tube seals.

Amalgamate information
Statistics are used to amalgamate information so as to balance the influences and make decision making easier. However, statistics rarely if ever include enough contextual information and therefore cannot be weighted properly. Take for instance the charter market; it is affected by so many different influences, some of them hugely relevant to the charter levels being faced and some that are normally relevant, but may be insignificant to the current circumstances.

On top of this, statistics are often used to discover cause and effect models rather than verify them. People actually often quote statistics that seem to violate common sense and feel that this is the more educated approach. This works well in physical sciences because we often cannot understand cause and effect at all and begin to piece it together by observing trends. There is rarely any other way to understand the physical world than to gather data and compare it until some pattern emerges.

However, we are not condemned to live with inferior or misleading statistics. In most domains where people are involved, we can ask the people what their considerations are when they work and why they act in a certain way. For example, charter market analysts correlate charter rates to ships utilisation to predict rates and then estimate how closely the two correlate when the better way around is to understand the cause and effect of charter rates and then use the statistics of income rate to utilisation correlation to verify the cause and effect to some degree.

More specifically, we know that rates are dependent on many short and medium term factors, besides supply of ships and demand for transport, such as storage of commodities, charter market rate trajectory, commodity price trajectory, demand trajectory of the commodity, supply trajectory of the commodity, congestion, and many more economic indicators. If we took each charter fixture and weighted these major short and medium term influences as we charter the vessel, we would have more meaningful data to make statistics and better predictions of rates.

Therefore, getting more salient data surrounding the cause and effect of each event is most important when we are observing some type of expectation failure, for example, an observation of a defect reported by an officer on board a ship, could lead to an undesired event.

Examples

  1. A non-conformance about failure to follow the right process for a replacement oxygen and gas meter seems different from a non-conformance regarding failure to performa risk assessment before shutting down a sea water pump and blanking it off. One is about a gas meter and the other about a pump. But they may seem quite similar with respect to perceptions of the importance of safety. Seen from the viewpoint of a seafarer, who has not been informed about some vitally relevant details regarding what has been done in his absence, they are even more similar: a gas meter that behaves differently than expected and a pump that has been blanked off are very different from what he would expect.
  2. A crack in the flange of a fuel pipe in the purifier room may seem quite different and also quite similar to a crack in the flange of a pipe feeding the main engine injection pumps. However, if the main engine pipe is part of the fuel pump structure, the cause of the problem can prove to be quite different from damage on a regular pipe fitted by the shipyard.
  3. The injury to a crew member due to a fall on a slippery part of the deck may seem to be quite similar but also quite different to the injury of another crew member who recently signed on and was not aware how to use the lathe in the engine room. In the first incident the injury may have resulted and be related to a number of safety precautions that need to be taken while the crew is working on deck, such as cleaning the deck to remove oily residues, wearing safety shoes, painting the deck surface with a special non-slip coating, while the second incident might be related to training issues, lack of experience and absence of written instructions on how to use the lathe.

The right approach to decision making is to understand cause and effect by examining each case ‘story’ more closely in order to determine the goal and obstacle structure, then, to use the goal obstacles structure to better qualify larger batches of information (statistics) and use these more judiciously to make explanations and predictions. So using common sense to evaluate the factors that could apply to every situation on board where a violation of expectation occurs, seems to be the first useful step you take. The statistical approach can then be applied to a much richer set of cause and effect indicators.

In the information age, access to information and data collection is important. If you are collecting data this is a good thing and the important next step is to get our information management systems to help gather the right cause and effect indicators relevant to each case.
However, to do this, the system has to track the work of your staff as they deal with, let’s say, a defect or non- conformance and resolve it. If the system does not collect salient points in the processing of daily problems you will never extract any statistics that tell you anything useful about management and improvement.

Right point at the right time
Since there could be many salient points that make an expectation failure, such as a crack in a fuel pipe important, the system has to present the right ones to the user at the right time, otherwise, it will be quicker for the user to write down comments in a common language that the computer cannot recognise and thus cannot present later in statistical form.

For example, is there a crack or a welding pore, is the pipe under external stress, is the pipe a high pressure pipe with a wall thickness limitation? Is there a maintenance problem common to these pipes? Even more urgently, is there a process that this breakdown affects and should there be a risk assessment at a variety of levels? But how would the system know it’s a fuel pipe so as to consider corrosion as of unlikely relevance, how would the system know about the pipe configuration and design limitations so as to ask relevant questions about cause and effect?

All these questions are relevant, if this observed pipe leak is to end up as a statistic, especially if the statistic is meant to assess management quality, something that in the tanker industry is very much a target of Chapter 12 of TMSA, making this abundantly clear.

Herein are the reasons why software often does not help with management decisions that are not straightforward. They are helpful where the decisions are simple, like comparing prices, but they don’t help tell us when to stop the ship for an overhaul. To do so, the system has to understand (this means have a model within its data structure) of everything important about the enterprise and everything about what the user is doing (for example in this case reporting a defect).
In the gas meter example above, how would the system know that gas meters can cause death if they are incorrectly operated? And why would it consider asking the user if the gas meters procured operate in a way that a new user can expect? Why this is not a question normally asked when buying new binoculars or chipping hammers? And when reporting the ballast pump isolation process, how would the system know that the blanking off process requires some consideration of coinciding factors, so that someone does not inadvertently flood the engine room?

The system could, of course, ask you all the questions it knows, regardless of context, but will this help the management process or\ delay it? Would senior management, like masters and engineers, who are responsible for resolving problems, tolerate answering irrelevant questions?

A failure to apply the correct change management process to a gas meter could be categorized under Change Management, which would make sense, or under Gas Meters, or under Tank Entry, or under Safety Equipment, etc. But which one would best help indicate management quality? If the assignment of failure is incorrect, what is the point of making statistical count from this categorisation? Even if it is correct and the non-conformance is assigned under failure to manage change, would this carry the same weight and should it be considered on a par with buying binoculars and chipping hammers, without going through a change management procedure?

It is not difficult to enrich incoming information with the right contextual and salient indices, if the process is well designed and is performed in steps. Most importantly, in performing the process, the enterprise benefits anyway. So the answer to better decisions is qualifying incoming information as a by product of daily work.

In the information age, having solved the information collection problem, we now need to solve the information categorization problem. Never was this more relevant than today when the obstacle of access to information is behind us.