Monday, 30 September 2013

The Tube Challenge and Underground Mathematics

The Tube Challenge is a race around the London Underground system.  The rules are that participants must pass through all 270 stations as quickly as possible, using the trains, or walking or using public transport between stations.

Harry Beck's map of the London Underground system pioneered such maps of public transport systems around the world.  It transforms the geography of London into a network - essentially,, it is a public demonstration of topology.  It is also one of the best known network diagrams in Britain, if not the world.  This year, the underground system has celebrated its 150th birthday.  One part of that celebration is the installation of 270 different labyrinths in the stations, numbered according to the sequence taken by the current holder of the record time for the Tube Challenge.

Details can be found here; there is a link from that page to the challenge, where an animated diagram shows the journey, and the stations are highlighted as they are "visited".  If you watch the route develop you will see the stages where it is optimal to leave the underground system to travel on the surface between stations.

Now, consider the O.R. problem that underlies this challenge.  It has links to both the travelling salesperson problem, and to the Chinese postperson problem, but the timetable of the public transport is also part of the complexity.  Not all edges of the network need to be traversed, one doesn't have to return to the start.

Do other transport systems have similar challenges?

One of our O.R. students at Exeter had a summer holiday job working for London Underground.  His job was to deliver new electronic devices to each of the 270 stations and install them.  He could carry six machines at a time, and that was a a day's work, so he had to plan his trips to reach the stations in groups of at most six, another interesting piece of O.R.  He wrote it up and analysed the problem as a project.

Saturday, 28 September 2013

What a silly question!

There is, inevitably, a close connection between statistics and operational research.  The data for O.R. work is collected and analysed by people with statistical expertise.  Those engaged in an O.R. study need to advise on what data sets are needed, and the sort of analysis which might help. 

So, it is hardly surprising that most O.R. training involves lecture modules and courses on elementary and advanced statistics.  Hopefully, these include warnings about what might go wrong.  I am sure that there are many O.R. projects that have horror stories about mistakes in the way that observations were made, or incomplete data sets. 

One of the warnings that I received many years ago, and which I stressed with students, concerned collecting data by means of questionnaires.  Someone said that there is a popular attitude that "Anyone can produce a questionnaire", to which one replies "Anyone can make a mess of a questionnaire".  Surveys and questionnaires need to be thoroughly tested before they are released. 

Why mention this?  Tina was sent a lifestyle questionnaire this week, for a long-term study for which she has volunteered.  Several of the questions were amazingly badly thought out, but her prize was:

"In the past six months, on average, on weekdays, how many hours have you spent out-of-doors?"

The question was intended to look at risk of skin cancer. 

But could you answer that question with any confidence of being accurate? 

What do they mean by "Out-of-doors"?  Who measures their time out of doors? Did any of the team test themselves on the questionnaire?  Did they try it out on anyone else?  A question like that should NEVER have got through to the final survey!

Good luck to whoever has to analyse the responses.

Friday, 27 September 2013

Problems with forecasts

My friend S works in management of a shop in Exeter.  It is one of a national chain, selling consumer goods.  Earlier this year, the chain opened a second branch in Exeter (500metres away, as the crow flies).  I asked him how the second shop was going, and what the effect had been on "his" branch. 

He said that like for like sales per week had fallen by about 15%, but the sales at the second shop were running at about 80% of the sales at the first.  So, total takings in the two shops combined were significantly higher than in the first store.  The expansion seemed to be successful.  But, there was a  problem.

Head office sets each branch a target for weekly sales.  It is based on the corresponding time period a year earlier, together with an underlying trend upwards.  And the first branch is failing to meet its targets, because some of their sales have been shifted to the second branch. 

It looks as if head office has a forecasting model which is overlooking the launch of the second branch.  And this suggests that the forecasting model that head office is using is a naive one - certainly S thinks so.  It is based on the assumption that tomorrow will be just like today, with - perhaps - some growth.  There are times when this model is as good as you can get, especially when you don't understand the underlying cause of the time series generating the data.  But, there is a good explanation for changes in the data about sales at the first shop.  And the evidence ("Fallen by about 15%") gives a measure of the change.  So it would be easy to adjust the targets to reflect the consequences of the decision to open the second branch.

It is a warning that forecasting models and tools should not be based on simple "off the shelf" software.  Human beings need to be involved, to make sure that what is being modelled has not been changed in a significant way.

A newspaper report today tells of another business which is suffering from forecasts which don't appear to take all the relevant information into account.  A well-known company which hires our men's formal clothes has reported drops in its sales.  And the drop is due, they say with hindsight, to the decline in the number of couples getting married.  Maybe this could have been included in the forecasting model for their work?

Friday, 13 September 2013

Identifying the decision maker

In my early days as an academic, long before email, we communicated between universities by post, or "snail mail".  One of my students chose to investigate how large the inter-university mail flow was.  She asked the question: is there enough mail to justify packing a bundle of all the day's mail from Exeter to, let us say, University B?  All the outgoing mail from Exeter went through a postroom, and University B's postroom could receive a large envelope of post and distribute its contents. Paying for one large envelope through the post would be cheaper than sending two or more separate ones. 

So the project meant collecting data (and how to do that) and discussing the strategy with the university administrator who oversaw the postroom.  The student also sat with the postroom staff as the mail from Exeter's buildings and departments arrived, was franked and then sent out in the post office van each afternoon.

The administrator's attitude was that there was no need to investigate; there was not enough inter-university mail to make it worthwhile to bundle up outgoing mail.  He was the manager, and assured us that he knew all about his department.  On the other hand, the man in the postroom was interested in what we were doing.  And, he told us, he already sometimes did bundle up outgoing mail in the way that we were suggesting.  But he didn't do it systematically, and his manager didn't show much interest in the details of the working of the postroom.

The project showed that there wasn't enough inter-university on normal days to make it worth sorting the outgoing mail regularly, but it was worth encouraging the postroom to use their common sense and to approve the ad hoc way taht they were already acting.  It wouldn't save much money, but the psychological effect would be considerable - the postroom staff wouldn't see the wasted money of six letters being sent individually to University B one day.

It reinforced one of the first lessons in the practice of OR that I learnt, and often relearnt.  Identify the decision maker, who may not be the obvious person.  Here the manager was not the real decision maker; staff in the postroom were the people who took decisions.

Another variant of this lesson is the so-called "Beer-truck accident" story.  This consists of the question: "What would happen to your production line if your scheduler was killed by a runaway beer-truck?"  Thinking of the answer to that question affects how you see how important that person would be, for decision making, for their knowledge of the system, for their salary.

I was thinking of this lesson because the latest issue of ORMS Today from INFORMS includes a reminder of the work of Gene Woolsey, who put his students onto production lines to learn what would and would not work if an OR study suggested changes.  They would work alongside those, like the postroom staff, who aimplemented management policy.

And I was thinking of it, because one of the papers that I heard at OR55 seemed to have ignored the lesson.  The speaker said that the changes proposed by the OR study had been well received by managers, but with scepticisim by those whose work pattern would be affected.  It could have been because those people were scared of change, or - rather more likely - because they knew that the study hadn't taken all relevant factors into consideration, so had aspects which would not work.

So - identify who really takes decisions, and work with them.  Identify the value of each member of the team, and honour their skills and knowledge

Thursday, 5 September 2013

OR55 day three - the last day

Nobody likes the last day of conferences; it would be optimal for most people if the last day was abolished.

We have to recover from the fine dining of the previous evening, and the exertions of dancing for those who did so, and then one is faced with having to pack, not only everything that you brought, but also the free conference bag, USB stick, books and papers that you collected, plus anything that you have bought.  And all this has to be done in a way that allows you to arrive at the conference venue in time to hide the case away in the luggage room, and get to the first session of the day.  Presenters and stream chairs hate being on the programme first thing on the last day.  But someone has to do it, and the delegates who had this did very well - though some of the audiences were smaller than usual.

We'd had a brush with university bureaucracy late on the previous evening.  The plenary speaker for day three did not come to the dinner, so - en route for our rooms - we asked the porter whether he had arrived.  "Sorry", came the reply, "to tell you would be a breach of privacy".  The following morning, the assistant at the desk cheerfully told us that he had come.

I had business outside the presentations of papers so only went to five papers today.  The first looked at the fitness of rugby players, and models that could be used to assess fitness from the team coach's regular questionnaires; the measures of fitness could not be collected directly as frequently as the questionnaires.  With some judicious modelling, regression equations could be found which had an R^2 of 79% - pretty good.  Following that, we heard about some models of training schedules for athletes; based on some ideas of the interaction between the benefit of a period of exercise and the fatigue it would lead to.

After a break, I was glad that the seminar rooms were close together as I moved between rooms at the 30 minute "bell".  First, a fascinating talk about how OR can be used to help charities be more effective.  David Pritchard said that there were several reasons why OR methods are not used more widely by charities:
  • Lack of awareness of what OR can bring
  • Culture, since the sector defines itself as being driven by concern for the objects of the charity, compassion, equity, needs etc.  The emphasis on the concern for these may be at odds with interest in efficiency, optimisation and cold analysis
  • A data-poor environment with little data for traditional OR
  • Weak mechanisms for rewarding effectiveness
And these are not just in charities!  After that I skipped along to hear about models for queues at Heathrow, where there are models to help plan the deployment of immigration staff meeting the passengers off planes.  It was good, old-fashioned queue-modelling, with several relevant twists.  The models need to relate to the times of arrival of the planes, where they come from (passenger mix), and the size of them - as well as how far the passengers have had to walk to get to immigration.

Back along to the first room to hear about "Hands-on OR in healthcare in remote rural Africa" from Andrew Dobson who had recently returned from Bwindi hospital in Uganda ("Twelve hours by bus from Kampala, the last eight on unmade roads").  He had analysed the sources of funding for the hospital, noting how donations from individuals had dropped over recent years.  Overseas visitors come to the area to see mountain gorillas, and there was scope for a regular marketing of the needs of the hospital to these tourists; another piece of work concerned making the pharmacy more efficient.

The final plenary was an informative and (in places) amusing walk through some of the work that the government's Behavioural Insights team (part of the Cabinet Office) has done and is planning.  Michael Sanders suggested that there are three ways that authority can affect behaviour - by regulation, by incentives, by information.  The team had developed the acronym EAST, for Easy, Attractive, Social and Timely for its ideas.  They have a blog, where much of the material we heard is recorded, so I won't repeat it. 

And then there were the farewells, though many people had left earlier in the day.

It has been a good conference, and lots of people gave the event 8 or 9 out of 10 on the feedback.  A common complaint was the hills on the campus, and we can't get rid of them!  I only attended one presentation which was less good than it could have been; it has been mentioned in one of the daily blogs, but I will not say which!  There was death by PowerPoint, a phenomenon which happens when a slide has too much written on it.

I crossed Exeter by bus, as planned, and got home to find that the expert had been and the drain was sorted out, after Sunday's debacle

OR55 the gala dinner

The last night of OR Society conferences is the time for the gala dinner.  We dined at Holland Hall at the University of Exeter, with a panorama of the evening sunset over the hills beyond the window.  Several years ago, the ORS did away with guests giving after-dinner speeches on these occasions, so the only speech (short) was by the ORS president, who awarded the President's medal to the paper from Ernst and Young, describing their work with New Zealand Post.  He also thanked everyone who had made the conference happen.  The two co-chairs were given goodies from an Exeter delicatessen as a "Thank-you"; fortunately, Tina was there to take these home for me, as I was planning to go home after the conference by bus;  Phil had to pack his into a suitcase for his train journey home.

After the dinner, there was a ceilidh dance, with Mrs Midnight from Exeter (Mrs Midnight was a four-person band with a dance caller).  Some of the delegates escaped being dragged onto the floor, others pretended to be deep in academic conversations, and a few dozen joined in with half a dozen dances.  We ended with a madcap dance that the caller said was the Circassian Animal dance, a new one to all of us, where some of the Circassian moves had to performed in imitation of animals.  There was considerable laughter around the circle as men and women imitated kangaroos and mice.

So the second day ended happily.

Wednesday, 4 September 2013

OR55 day two

Everyone seems to have appreciated the second plenary at the conference.  Colin Shearer spoke on ".The Analytical Revolution", speaking from his experience at IBM and other companies dealing with data in large amounts.  He gave us several bullet points to describe the current flood of data and how we deal with it.  So: the world has data which is Instrumented, Interconnected and Intelligent.  And it is characterised by Volume, Variety and Velocity.  Later in the talk he added the need to question the Veracity of the data.  After a wide-ranging talk he expressed concern about those who claim with big data that "You are sure to find something interesting in your data" and the claim that being a data scientist is the sexiest job of the 21st century is a misleading one.  The essentials for doing good analytics are (1) problem solving skills, (2) communication skills, (3) open-mindedness.

In summary we are at a crucial stage in the evolution of advanced analysis as a discipline applied to real-world problems.

I went on to hear a stream on sports modelling, and after lunch we had the three presentations for the president's medal.  British Airways, New Zealand Post, Ministry of Justice.

In the afternoon there was a choice of the Making an Impact sessions, or the "networking" trips out.  A small coach took us to Yearlstone Vineyard, for a tour of the vines and the wine-making process, and then tasting of five wines from the vineyard.  Excellent!

OR55 simulation stream on day one

Went to some of the simulation stream at the conference and enjoyed very good presentations.  Nav Mustafee spoke about the vast number of keywords used to classify OR papers and suggested how a taxonomy might be created.  Herb Daly spoke on multi-method approaches, with the catchy title "models versus muddles", and John Salt introduced us to the word "Tyromancy" (forecasting by cheese!) and argued that simulation is better - much better - than that!  Finally, Ray Paul asked "Why do analysts throw the model away?"

All very good, and we ended the first day with a relaxing meal and the traditional rivalry of the conference bar quiz.  I was on the team which came third ... And there were seven teams!

Tuesday, 3 September 2013

OR55 day one

With the original plenary speaker in hospital,Geoff Royston and Phil Jones shared the opening session on the general theme of behavioural O.R.

Geoff talked about the Johari window of hidden OR, to articulate the science of better.  Breaking our work into four cells, decision physics, human behaviour, problem structuring, system design, he said that people outside O.R. do not recognise our skills in problem structuring, and we underestimate our skill in system design.  Numbers 1 & 2 are recognised by others, 3&4 under-recognised.  1&3 are recognised by us, 2&4 under recognised.

The key words he chose for O.R. are science, improvement, systems.  Our science is - like engineering and medicine - an improvement science.   We improve systems which matter in the real world to real people.     He used four words to describe our work, in a cycle:  discovery, design, decision, delivery.

Focussing on tools does not get to the heart of our focus on system improvement.  Back to "our Johari   window".  We need to get behind this to look at the Why? as well.   This should affect how we teach O.R.  And it leads to a merging of that window into a fabric of a multidisciplinary topic.

Phil spoke about putting the Science of behavioural change into our discipline.  Some studies in conventional O.R. lead into the need for change in organisation.  He also used the Johari window, saying that work on human behaviour is in the hidden, under-appreciated section of the window.  His take on the O.R. Process in behaviour was:  Appreciation, analysis, assessment, action; the four "a"s of our discipline

 He drew on Zacharias' work on behavioural modelling challenges - such as making unwarranted assumptions about behaviour; people adapt.  He concluded that we need to work together as an inter-disciplinary team.

I went to hear about community O.R. .. a session about ideas for a special interest group within the O.R. society. Huw Evans led the workshop.
Community O.R. Differs in several ways from two other (related) areas - pro-bono and third sector work. But all three depend on involvement with people, and this links to the topic of behavioural research.  We set up graffiti boards about values, options and outcomes for such a special interest group.

After lunch - a buffet - we talked about a methodology for the classification of O.R. Terms.

Paul Randall spoke about community O.R.  One project in Namibia, was about whether to mine sepiolite or not.  Another was about a community project "build together", characterised as a problem were that a lot of money was put in and not many houses were built.  He reflected on what he called "The moral compass" about work in the third sector in Africa.  Talking about third sector work in the UK he drew attention to the range of data sources and easy it is to overlook some of them

Oh dear, I kept notes of the next session here and there were connection problems, so I will send this blog off and start again z

Monday, 2 September 2013

OR55 day zero

It is a law of conferences that you only remember the extraordinary events.  Soon after lunch, we heard that the plenary speaker for the first day was ill and in hospital.  Phil, the co-chair asked the ORSociety president to step in if necessary, and I made a delicate enquiry about another possible speaker.  Meanwhile there was a great deal of traffic on the motorway to Exeter, and an accident blocked it for a time.  So our 3pm start became a nearly 6pm start.  Still, we got the packing of delegate bags finished.  Just glad that we don't have the thousands of delegates that some INFORMS and EURO conferences attract!

The first evening has gone with the renewal of old friendships, over a meal and a drink, and then a social for about thirty, playing a game of horse racing.   I have started to tweet with the tag #OR55 and w have a twitter feed screen for that.

Sunday, 1 September 2013

OR55 minus 1

Yet more emails today; I have agreed to chair one stream as the previous chair has had to withdraw at the last moment.  But there are no real panics.

Sadly, the press release about OR55 that we sent to the local Exeter newspaper didn't get into the printed version, though it is online.

It was a funny day - all-age service at one church, traditional service at the second.  Then after lunch we cleared a blocked drain (we had thought there was a bad smell outside the back door) and then got a drain rod stuck.  Now what is the OR solution to that?  I can picture a model of where it is stuck and why (there is obviously a change of diameter of pipes), and can find suggested solutions online from people who have had this problem before (experience is important in OR practice), but I lack the expertise to put those solutions into practice.  Therefore - more good OR philosophy - don't try to solve a problem for which you do not have the correct equipment or expertise, and get someone in to do the work from outside.  There is nothing wrong for any OR professional to admit that they are not qualified in an area; for instance, I know a little about yield management, but would not start a project in that area.