(Not So) Bad Performance Measurement On Tour (#3)

‘Data have no meaning apart from their context’. – Donald Wheeler

Cast your mind back to Episode One of this series. No beer, no food, no chillies; just trains. The examples provided in that blog post highlighted the dire (but typical) methods used for presenting ‘performance’ information. One of my readers, (Ian Gilson) responded with an enlightening post of his own, which delved into railway performance measurement in more detail, demonstrating why the headline figures presented for timeliness can be misleading.

Anyway, I am currently visiting one of my favourite places in the world – Whitby, North Yorkshire – and whilst pottering about near the train station yesterday, I encountered another such performance board for train services. This is what I saw:

At first I noticed the same sort of pretty useless stuff as seen in my previous blog post, i.e.

  • Comparisons between different work groups. This ignores local context and those system conditions existing in different parts of the overall rail network which affect performance. Also, cue individualist behaviour and sub-optimization (see my previous post) as work units strive to out-do each other at the expense of the overall system.
  • The appearance of the inevitable asterix that relates to the definition of what ‘on time’ means in the eyes of the ‘Customer’s Charter’, along with the small print underneath. (Probably not what ‘on time’ means to the customer, but there you go).
  • The declaration that, ‘All our service groups exceed the levels required’. In other words, arbitrary numerical targets have been achieved. How they have been achieved, or whether they correspond with what matters to the customer is unclear.
  • That lost-looking pair of numbers on the second poster under ‘Customer contacts for the period’. Look – it’s this year vs last year again! A binary comparison or two is never too far away from this type of performance measurement. Anyway, what does it mean in this case? I’d have thought every person who buys a ticket would count as a ‘customer contact’. If it means something else, such as the number of complaints, letters of praise, or enquiries, why lump them all together like this? It makes no sense. Is this year’s lower figure supposed to be better or worse? No one knows.

None of this stuff actually tells us anything about performance. Furthermore, some of it actually risks generating the wrong type of behaviour. It’s not at all useful to the weary traveller who might either be sheltering from the rain and seeking inspirational reading material, or just trying to come to an informed conclusion about whether railway performance is any good. It’s data without context.

Now, on the other hand, I have to take my hat off to one aspect of what the railway people have done here. Look at the first poster. There’s something there that was completely absent from the performance document examined in Episode One – a bit of context. Some narrative. Some information about why performance data is what it is. Look closely – at the top right hand side of the first poster, there are reasons given why things happened:

  • A flood.
  • A vehicle hit a railway bridge.
  • Signal failure.

There’s even dates! This is great to see, as it puts a bit of meat on the bones of the raw data, and highlights the fact that it is not always possible to achieve perfection; most importantly, it acknowledges that any shortfall is usually partly due to a range of external factors that affect performance. If a specific line’s performance for the month was adversely affected because of severe flooding on a particular day (and trains therefore failed to achieve their timeliness target), it would be wrong to castigate that work unit for ending up at the bottom of some league table. Conversely, if by looking beyond the raw data (in context of course) it becomes apparent that there is a repeated problem with the same set of signalling equipment, then the train company needs to take action to resolve that recurrent issue.

In my ideal world, the data would be presented using control charts (which give context within the data) alongside the narrative (which gives context around the data). Russ Ackoff would say that useful narrative accompanying badly-presented data (as with the overall presentation of these railway posters) is just doing ‘less of the wrong thing’, but hey, I’m trying to accentuate the positive.

The same applies to crime rates and other police data. There is little currency in presenting rows of numbers detailing total arrests or detected offences per team, often supplemented by worthless ‘this month vs last month’ binary comparisons. Measures should be used to help understand the performance and capabilities of the system, with a view to identifying opportunities for improvement. Simply presenting data in the traditional format does not achieve this, be it for train services or policing.

Conversely, if the right measures are presented in a useful format and with context, this enables managers to understand why things happen and what to do to improve performance. For example, if police managers were pondering an apparent change in crime rates or reported incidents, it would be useful to understand the narrative and context behind the numbers. This approach could indicate a description of factors such as:

  • “That was when we did a proactive policing operation and caught loads of people committing crime, thereby increasing the volume of reported crime”.
  • “Incidents were unusually high that week – as it happens we had fewer officers on duty available to police the town centre”.
  • “Criminal Damage offences went off the scale for that area because one individual went on a tyre-slashing spree one night and damaged 30 cars”.

By understanding the context of the data (along with being statistically literate when it comes to interpreting the numbers in the first place), managers can begin to unearth the evidence base that either provides a mandate to change tactics, or not to knee-jerk. It also means more to the public, as opposed to “we had ‘X’ amount of things this year compared to ‘Y’ amount last year, and by the way, the target was achieved”. I’ve always found the public tend to be quite reasonable when they understand why something has happened – this applies as much to when a train is cancelled because a third party has crashed into a railway bridge, as it does when poorly-presented data invites either hysteria or complacency over crime rates.

At least some of the content of the board on the wall of Whitby train station suggests a step in the right direction. This is data with context. It makes a nice change.

About these ads

About InspGuilfoyle

I am a serving Police Inspector and systems thinker. I am passionate about doing the right thing in policing. I have a big problem with numerical targets, unnecessary bureaucracy, and anything else that stops police officers from providing the best possible service. I believe that by adopting a systems approach, policing can be transformed beyond the wildest expectations of many.
This entry was posted in Systems thinking and tagged , , , , , , , , , , , . Bookmark the permalink.

20 Responses to (Not So) Bad Performance Measurement On Tour (#3)

  1. tactical says:

    Finally, something that makes you happy, apart from chillies!

  2. ThinkPurpose says:

    Here’s a question. Whenever you see weather named in a performance report, it’s always as a hindrance. Rain stopping council grass cutting or snow and ice increasing potholes.
    How come “a nice sunny day” is never listed as AIDING a performance measure? I think it’s because whenever context is added, it’s when “reasons” for under performance is asked for, not “over” performance. And living in UK provides a handy scapegoat with the weather. Also, any “over” performance is clearly down to the keen business acumen of the managers.

    • yo mo says:

      Not a comment as such, but ‘a nice sunny day’ makes trains run slower, particularly on the ‘up line’. A very sunny period can stop them completely.

  3. Ian Gilson says:

    Good post Simon. I agree entirely with the thrust of it, but find myself with a similarly cynical view to Think Purpose. I think the context is there as a “it wasn’t me, it was them what did it!” piece of PR. If their performance had been top of the shop, would they have bothered to add context?

    Anyway, damn you; now I’m wondering whether South and East Yorkshire really is any better than the rest, or whether they just have more temporary timetable adjustments ;-)

    Mind you, is their performance within normal variation compared to the others? Nurse! Fetch the Control Charts!

    • Oh you cynical pair! I did say I was trying to accentuate the positive! You’re right though, its just not me is it?

      On a sensible note, its absolutely proper that we capture the reasons for both ‘good’ and ‘bad’ performance. It helps establish an organisational memory that is linked to the evidence base…vital for fostering true continuous improvement. I don’t need to tell you guys that though, do I? Converted / Preaching. ;)

  4. yo mo says:

    The customer contact data is a bit suspect, remember that at lot of people are splitting tickets, ie. buying a ticket from A to B and another from B-C to save money, when their journey is from A-C. The TOC’s don’t mind this as it inflates the number of passengers apparently carried, but it can screw up the planning of station resources, for ‘phantom’ passengers that neither board or alight.

    Ian’s charts are interesting, no time to get into them today.
    [I left my tin hat and flak jacket at work. ;-) ]

  5. Blue Eyes says:

    I thought data was by definition contextless. Data + interpretation = information. The zeros and ones on my hard drive are data, the computer is able to use them to display information.

    I’m worried how much effort is going into production of this data which is trading fraudulently as information. There are people in offices earning salaries (and therefore costing the fare-payer) thinking about what proportions various statistics make up the Performance Management Unit. It’s bonkers.

    Someone was trying to explain to me the difference between an input contract and an output contract. If I have got it the right way around, an input contract is where a train company (for example) arranges for each carriage to be cleaned at a particular frequency, for example every time it arrives at the end of the line. An output contract is where a train company arranges for the carriages to always be within a particular range of cleanliness.

    It seems to me that in some circumstances an output contract makes sense. You don’t want your electricity meter to be serviced once a year, you want it to be working 100% of the time. However you don’t mind especially if the grass on the motorway verge grows faster in spring than in autumn, as long as it is mown sufficiently often to not be a problem.

  6. yo mo says:

    As my annual trip to the Midlands has been suddenly, and unceremoniously binned,so I have the time to return to this topic. And can already feel it’s going to get a bit long.

    “Whom the gods would destroy, they first make mad .” That’s you lot, determined to mash my swede. At first I thought “at last a process, a proper one with goals and a metric and everything”. So I went over to look at Ian’s post, and I was left holding my head in my hands again.

    Lets clear up a couple of definitions first.

    Target
    Targets or goals, are the basis on which the degree of success of failure is determined. They come in various forms, some are spacial (think archery), some temporal, but most are numeric (cue howls of anguish from the audience). If one is cooking for example a nice curry, the ingredients are measured in accordance with a recipe, be it written down in a book or simply remembered. There is a certain amount of each spice required to give the best balance of flavours. There therefore exists a number of targets for the right amount of each spice. Nobody throws in a random amount of chillies, and hopes the end result will still be edible. (OK someone might, though I’m not looking at anyone in particular, of course.)

    A target can ‘appear’ to be many things, difficult, demoralising, stupid, or dangerous. These are not attributes of the target in and of itself, they represent they feasibility or practability of reaching the target, in the current situation.

    Let’s say I commanded you build me a pyramid, a proper old school, Egyptian job, with frescos and everything. You might say various things, (some of them rude probably), and declare the job impossible. Now if I issued the same command and placed you at the head of a mindbogglingly vast army of architects, stonecutters, masons, plasterers and painters, then it’s easy, “No problem, should have it finished Wednesday week”. The target hasn’t changed at all, only it’s context.

    A target of it’s self is not arbitrary, (oh it’s that word again), there is a rationale behind each target, and it is that rationale that needs to be carefully examined, before, during and after any operation.

    Measurement
    Measurement, verb and noun, the common link? They only take place in the past. You cannot measure the future, it hasn’t happened yet. Nor does past measurement have any mystical connection with future outcomes. (don’t believe what the bods at CERN say, the effect is very small, just leave them to play with their whizzy, smashy machine, and they’ll be quite happy).
    As a process runs there maybe hundreds of measurements being made, but each measurement is a discrete point on a specific continuum. And unless you know precisely how to weight each factor, then ‘meta-measurements’ like overall customer satisfaction are most likely to be of no practical value.

    Now lets turn to Ian’s charts.
    Scenario, two trains arrive at their destination at various times, which is the most reliable?

    It’s a process, “shifting large hunks of metal along a track”, and the target is to move said hunks from point A to point B in a given amount of time, X and because it is a real world situation there will be chance events that introduce variability. So there is a tolerance allowed, in Ian’s example it’s 5 minutes.

    At first glance Ian’s charts and looking just at the data points, (after checking that the axis are the same, not accusing Ian of any graphical legerdemain, it’s a reflex action of mine, whenever charts are presented separately). They are classic archetypes of a process that is stable and one in chaos, respectively. However on closer examination there is something odd.

    To recap target X with allowable deviation of + 5 minutes, (I didn’t pluck these numbers out of the air it’s in the rubric).

    First Chart
    Plot a chart with X at origin and a UCL of 5.
    Plot Data points.

    Results
    All points above UCL.
    Points appear in a narrow band.

    Process stable, but may not be capable.

    Second Chart
    Plot as above

    Results
    42 of 125 of data points above UCL (33.6%)
    Wide scatter of points

    Process capable (76.4% of the time), but unstable.

    But using the “data wrapping” technique, tells you very little useful about the process. In the first example all data points appear within limits, even though the results are 100% failure. The second example has data point appearing outside (the expanded) limits even though the results are 76% success (76% figure based on X+5 measurement). This would tend to suggest that the technique is either not statistically valid or more likely being misapplied.

    It doesn’t even seems to adequately describe the variation in the process, as a single ‘figure of merit’, (which could be derived from the interval between UCL and LCL).

    Which train to choose? It depends, if you want to get somewhere before a rival, the second train will see you win 3 times out of 4. If you want to guarantee meeting a connecting train, good luck, past performance is no guarantee of what is going to happen in the future. A train could have been on time to the second for 10 years, but when you really, really need it, it’ll breakdown.

    If that’s not put you to sleep yet, have a peek at the solution to the ‘German Tank Problem’.

    http://en.wikipedia.org/wiki/German_tank_problem

    • I think ThinkPurpose has covered Ian’s charts, so I’ll focus on a couple of quick principles. Firstly, I argue there’s a difference between numerical targets and broad goals (e.g.a numerical target of achieving 93% timeliness, or 88.5% satisfaction IS arbitrary, whilst a goal of aiming to get as many trains on time as possible is in line with the customer’s nominal value, and therefore represents the company aiming for perfection, or what I consider to be the only acceptable ‘target’ of 100%). A system has to have a goal (or ‘purpose’) – this is different to arbitrary targets.

      On that note, if you aren’t already familiar with the Taguchi Loss Function, it is a good graphical representation of the difference between aiming for nominal value from the customer’s perspective vs arbitrary specifications. It’s a lot simpler than the inpenetrable algebra surrounding that German tank thingy.

      Secondly, there’s no argument against measures, as long as they are the right ones (i.e. derived from purpose).

      Thirdly, no numerical target is immune from causing dysfunctional behaviour, which is why they are so pernicious.

      Last point is that there’s no conspiracy to mash anyone’s swede. Catch you next time! ;-)

      • yo mo says:

        Time is a bit short so I only have time to reply to your comments, (I’ll get to Ian and Thinkster later).

        {Edit to add, I seem to have gone on a bit again, sorry folks}

        Sorry, Boss but you’re mixing up target and measurement again. If we accept that even in the best constructed system there will be variation, either natural/ random or so small that the cost of removing it is not economic.

        Therefore even if we strive for 100% we will never ever achieve it, except at a ‘cost’ that is greater than the ‘value’. (I can demonstrate this if you like).

        Thus we have to arrive at a figure that appears feasible, and given most real world systems are too complicated to model accurately, or the cost in time and resources required is prohibitively expensive, it will be the best available estimate. This will be the first iteration, and after the system has been allowed to run for a sufficiently long period to gather enough decent data, the figure will be reviewed, and one of three things will happen depending on the results, the performance might be acceptable (no change), the performance might be lacking (requires investment in process improvements) or performance may be better than predicted, in the last case the benchmark can be redrawn a bit higher, to give the second iteration. Rinse and repeat, until there comes a point that any further improvement is impossible, normally because the causative effect is either not controllable or it’s uneconomic to do so.

        Now you may still be going, “But, but, but” and in danger of spilling the well deserved contents of that glass of beer. There is an important bit still to come.

        Given that the future performance of a system is ‘un-knowable’ in absolute terms, but if we remember the laws of thermodynamics, systems tend to perform less well as time passes, can’t beat entropy. To produce the best outcome (however you want to define that!) the optimum strategy is to run the system as close to 100% as it is feasible to do so from the start, because you can not tell with any reasonable accuracy what is going to happen in the future. No rational person would run a process based on the benchmark level, because any natural variation will inevitably take the result below it.

        As I described in my earlier comment, it’s important to note the distinction between ‘target’ and ‘measurement’. The numbers in your photograph are just measurements, they tell you only what has happened, they don’t tell you what the target was at all.

        Taguchi Loss Function, well being an honourable person I did go and look it up, (I was at a loss as what the loss of very annoying, bleeping, needy, little virtual pets had to do with anything but I did)
        He assumes that loss follows a quadratic curve, I can’t decide if that is too general for everyday application, or aimed at a specific subset of processes. (I can demonstrate at least two cases including the train example, where it is not accurate).

        To your second point, a measurement is not derived from anything, it is just a measurement. It just exists only in the past as a dimension with a magnitude and an associated reference point in time.

        To point three, may I gently suggest that you re-read the last three paragraphs on targets, “wot I wrote” putting your ‘ST’ head aside for just a few minutes, focusing on ‘appear’, ‘context’ and ‘rationale’ . And then consider how you would feel if the payroll department suddenly decided that the amount of wages due to you each month was a ‘arbitary’ target and they could decide on a whim how much to pay you themselves?

        (I still can’t believe officers of your rank don’t get paid O/T, in the light of Winsor, I’d be breaking down the Fed doors to get that put back on the table, especially now that the pension is going to be based on average earnings.)

        Don’t get wet. :-D

      • Sorry, beer glass upended by the end of your first sentence. “Don’t confuse targets with measures” is MY phrase. It is not for hire. Also, I haven’t… (There are no targets in this set of railway posters; just measures, as you point out).

        My other phrase about the only acceptable ‘target’ being 100%, or perfection, is meant to highlight the arbitrary nature of defining a target that is anything less. I agree, 100% will rarely be attainable, but that isn’t the point. ‘Do your very best at all times’ should be the mantra – systems conditions, economic conditions and exogenous factors will affect how close you get to that magic 100%, but striving toward the ultimate goal is better than aiming to achieve a 13.7% vehicle crime detection target, or whatever. (Who puts 13.7% effort into chasing a car thief after all?)

        Further on, your efforts at defining a figure (presumably a ‘target’) is a good example of single loop learning (see Chris Argyris’s work) and actually demonstrates to a point how measures can be used to identify the performance of the system, as well as opportunities to improve it via successive iterations. This is great, except for the bit where you then drop a cheeky little benchmark in – benchmarks tend to be ignored by the data.

        I’m afraid your method of defining this benchmark / target confirms my darkest fears – i.e. that the ‘science’ behind it is little more than to see how we got on last time and set the benchmark a bit higher for the future. You seem to acknowledge that no rational person would run the process at the benchmark level, which kind of defeats the need for a benchmark. (This is good, by the way. The benchmark is arbitrary and can be dispensed with). Stick at aiming for 100%, use the right measures to ascertain actual performance and then tweak the system based on the data to improve future performance, if it is viable to do so.

        I sense we are not actually at opposite poles on this point when you get down to the detail – I hope this prospect doesn’t offend you too much…

        As for your bleepy little friend, the model is not necessarily quadratic in real life – it’s based on the parabolic loss model and is simply intended to demonstrate that loss is incremental, rather than a sudden step from ‘good’ to ‘bad’. Analysis of different datasets will yield different results (same as with bell curves not always being symmetrical). Not a biggie.

        Re: wages, Winsor et al, I won’t go there thanks, except to say that a wages are as much a ‘target’ as a tree or a bottle of milk. Haha!

        Did you get my email by the way? Don’t pretend you didn’t…

      • yo mo says:

        You last reply didn’t have a “click here” to reply tab on the end so I’ll plonk this here and hope it comes out somewhere near the correct place.

        I also don’t think we are poles apart either, if I did I would have given up ages ago. Re the ‘benchmark’ perhaps I picked a word that has a pre-existing definition for you, in which case ‘my bad’. Warning analogy coming up.

        Think of being a donkey, the target is the carrot that drives us forward, the ‘benchmark’ is the stick that will hit us if we fall behind, and the hoofprints we leave in our wake are the measurements of where we were in the past.

        (At the risk of confusing things further, I would suggest that what I have termed (perhaps wrongly) a ‘benchmark’ is a sort of ‘anti-target’ something we seek to avoid with the just as much effort as we chase the true 100% target . As a leader you will know your better people will move forward naturally, while your worst will have to be driven forward.)

        My first thought on the tamgochi curve was, it’s too broad for general use, and the first process I ran through my head exhibited a slow curve toward the midline and a step loss just past it. I’m not saying it’s never valid, just it’s not universally valid.

        Did you get my real email address? Don’t pretend you did. I hope you didn’t say anything rude in your email, because if you didn’t get it bounched back as undeliverable, you might have some serious explaining to do to the random stranger, that actually owns that address. 8O

        I’m trying to work out if it’s feasible to have a Skype-up but I’ve got a few tech. problems to fix before that can happen.

        Apologies for refering to Winsor, I completely forgot that as a serving officer, you have to be circumspect in what you say, and even if you did think he was ‘a clueless ginger t*sser, that has only been put in place to carry out the political ambitions of his paymasters’, you would be wise enough to keep such an opinion to youself.

        Not sure about milk, your payroll crew have a set of targets viz: to deliver 100% of monies due to 100% of staff, on the correct day 100% of the time. ;-)

        PS. have you got tags turned on for comments? I do intend to reply to Ian & Thinkster but feel I might need to put some graphs on. (They have gone a bit quiet, you don’t think they were only ever here just to pimp their blogs do you? :evil: )

    • Ian Gilson says:

      Thanks for taking the time to read the blog Yo Mo. I would just like to add a further couple of points to both Simon and Think Purpose’s replies, which have already addressed most of the issues.

      Regarding targets and the archery/cooking analogies. I see what you are trying to say, but I think you are confusing ‘target’ with ‘purpose’.

      The purpose of archery is something like ‘to hit the target as close to the bullseye as possible. An arbitrary target would be to hit the bullseye 90% of the time.
      Similarly with cooking a recipe; the purpose is to make it taste as good as possible. An arbitrary target would be to get the ingredients right 87% of the time.

      The purpose of a rail service might be to provide trains that run consistently on time and are clean, comfortable and fairly priced. An arbitrary target is to ensure that 95% of trains are less than 5 minutes late.

      In all these examples, why would you ever set a target at anything less than 100%?

      The original point in Simon’s blog, is that any target to make trains run on time for a certain percentage of journies cannot be anything other than an arbitrary figure. The quality of the rolling stock, track and original estimation of the timetable will all affect how often trains are on time, as well as the system’s ability to deal with adverse weather and vandalism. Targets are set in an attempt to influence performance for the better. However, if you set a target above the current capabilities of the system (above the Upper Control Limit), then one of three things has to happen to meet target;
      1. Improve the system
      2. Fudge the system
      3. Fudge the figures

      Guess which one is the hardest to achieve and, therefore, which is least likely to occur when financial gain is directly linked to performance?
      The rail network is attemtping to do 2 & 3 by setting ‘temporary trimetables to avoid trains being deemed ‘late’ and deciding that up to five minutes late is actually ‘on time’. Go figure…

      Regarding the charts in my blog, I apologise if I have misunderstood your post Yo Mo, but I’m not sure you understand what the charts are telling you. In particular, you seem to be suggesting that I set the control limits at particular levels of tolerance? The control limits are not calculated by me, but come from the data itself; they are 3 standard deviations either side of the mean. The greater the variety in the data points, the larger the standard deviation is and the wider apart the limits go. You can’t decide where the limits are set; it’s the data that triggers the calculation. Any point outside of the control limits is a ‘special cause’ and indicates that something unexpected has happened. Any point within the limits represents ‘normal variation’ and is a predicatble consequence of the current system.

      My main point from the data was to back up Simon’s assertion that averages tell us nothing in isolation. If you look at averages without understanding variation, you would not know which train was the better option. Seeing the data in the form of a control chart allows you to make a far more informed decision than any average will ever give you. One train is later, but very predicable, the other is slightly earlier on average, but wildly unpredictable. Regardless of which you decide to pick, the presentation of data in any time series allows you far greater understanding than just knowing the average.

      As a last point, you mistakenly seem to think that I set a ‘tolerance’ of five minutes. I didn’t, the reference to five minutes came from the rail companies who specify that five minutes late counts as ‘on time’.

      I hope that clears up some of the confusion for you Yo Mo. I didn’t seek to explain statistical process control charts in my blog, so I can understand your reaction to them if you have not encountered them before.

      Cheers,

      Ian

  7. ThinkPurpose says:

    In the first train chart you mention “plot a UCL of 5″, after talking about the “allowable deviation of + 5 minutes”. These are two different things. On the perfectflow control chart the UCL is, correctly, calculated from the datapoints and is 7.8.
    The “allowable deviation of + 5 minutes” or “target” should never make an appearance on a Shewhart control chart. They are a tool to distinguish between special and common cause in a process. Targets are irrelevant as this is used to understand what the process is doing, not what you would LIKE it to be doing.

    Targets. In the case of the trains, if there is a stated departure time of, for example , 11:27am is it more useful to think of THAT as the purpose of the train. not to be within 5 minutes of 11.27am
    That 5 minutes is as arbitary as my arse. Why 5 and not 6? or 4? 11.27am is not arbitary, probably. But the 5 minutes tolerance is. Look at the control charts. Neither train is arriving at the right time, adding a 5 minute tolerance into the mix isn’t going to help.

    As a user of public transport, buses mainly, the WORST thing for me is unpredictability. Standing at a bus stop in the rain i don’t care about averages. I care about knowing how long I have to wait, this is my “what matters” . The first train IS predictably on time. Just not on time according to the timetable. It is always pretty much 7 minutes late, knowing that as a regular user I would know when it left and arrived. Thats the beauty of control charts, they tell you what is predictable. Not what is “on target”.

    In this case purpose is not arbitrary (arrive at 11.27 am) but targets are (arrive within 5 minutes of 11.27am)

    Having been involved professionally for years around people who were setting targets in a large public sector organisation, I never saw any target that wasn’t arbitrary. Move it up by increments of 5 or 10. Add the same increase as last year but rounded to the nearest even number. Make it the same as the level as upper quartile performance from similar organisations reported last financial year. “Don’t like that, make it 3 more, 2 looks silly”. God the permutations of “arbitrary” are like human stupidity, as Einstein said.

    “Only two things are infinite, the universe and human stupidity, and I’m not sure about the former”

  8. ThinkPurpose says:

    “You can’t measure the future”, correct, but you can make attempts at predicting it, and the best thing for that, in the case of these trains, is a control chart of past performance.
    I predict that the first train will be around 7 minutes late the next time it runs.
    Bet you a fictional £5 that that first fictional train will be around 7 minutes late from the stated train timetable the next time it runs, and the time after that.

    “If you want to guarantee meeting a connecting train, good luck, past performance is no guarantee of what is going to happen in the future. A train could have been on time to the second for 10 years, but when you really, really need it, it’ll breakdown.”
    no guarantee, correct. But a good leading indicator of it. If the train you want has been on time to the second for 10 years, it is highly unlikely to have a breakdown when you need it.

  9. Ian Gilson says:

    Interesting debate. I’ll post a fuller response after I’ve despaired over QPR (predictably) struggling to overcome Reading.
    In general, Think Purpose has already said most of what I would say, but one further point; the UCL in my charts are calculated automatically from the actual data; you can’t choose where to place them and then add the data. Sorry if you already know this Yo Mo, but the more predictable the data, the closer the UCL and LCL are to the mean and the more predictable future data is.
    Control charts DO help predict the future. We successfully use them to decide what to stock in warehouses and on the vans of tradespeople. I’m happy to provide some real life examples, but that’ll be later!

  10. yo mo says:

    Dear MrG, Think, and Ian.
    Thank you all for taking the time to (a) read my comment and (b) write such fulsome replies.
    They have given me a number of interesting avenues to explore, and if you will allow me a few days to consider your ideas, I will come back and respond to them.

    (A note for just Think and Ian, MrG and I have corresponded before and he has correctly divined my comments contain a certain amount of banter, that may not be immediately obvious.)

    Ian, may I offer my very sincere condolences, if the ‘SuperHoops’ can only scrape a draw against ‘Waitrose FC’ then something is very,very wrong.

    Think, I’m sorry to hear you have been lumbered with an ‘arbitary arse’, if I were you I’d take it back, and get one that has defined boundries, [and looks good in jeans, like mine ;-) ]

  11. ThinkPurpose says:

    “I do intend to reply to Ian & Thinkster but feel I might need to put some graphs on. (They have gone a bit quiet, you don’t think they were only ever here just to pimp their blogs do you? )”
    No, I’ve never won or lost an argument on the internet. Nobody has, they just stop at some point. I stopped at some point. Your words smell a bit like a trolls, so clearly i should have stopped earlier.

    • Ian Gilson says:

      Indeed Mr Purpose.
      All a bit unnecessary Yo Mo. I thought the clue was in your line: “I do intend to reply to Ian and Thinkster”. If you ever deem me worthy of a response, I’ll get back to you.
      Have fun now.

      • yo mo says:

        I&T the intention was sincerely meant, but I knew that having to go away for a few days to sort out a remote site would prevent me following up in a timely fashion. I’ve only just got back.

        The reference to graphs was to ask our host if he had allowed the posting of images in a comment, not all blogs do so by default, and I did drop a rather explicit clue that my comments contain a certain amount of banter, and should not be taken as a personal affront.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s