Newton did
Alan
> -----Original Message-----
> From: [log in to unmask] [mailto:[log in to unmask]] On
> Behalf Of Rui Carvalho
> Sent: 09 March 2005 18:00
> To: [log in to unmask]
> Subject: Re: Publically available standard test cases
>
> Dear Bill, Alan,
>
> Sorry to be stupid, but I still can't find a satisfactory methodological
> discussion (Alan, maybe this is your point) -at least not in 'Natural
> movement' which I have here. P 42 has a very short discussion, but there
> are
> no references to any other papers and (at least to my reading) the method
> is
> not detailed.
>
> The question is: can we do science without properly peer refereed
> methodology?
>
> Rui
>
> > -----Original Message-----
> > From: [log in to unmask] [mailto:[log in to unmask]] On
> > Behalf Of Alan Penn
> > Sent: 09 March 2005 17:51
> > To: [log in to unmask]
> > Subject: Re: [SPACESYNTAX] Publically available standard test cases
> >
> > Rui,
> >
> > You are right, of course, that everything I say doesn't appear in any
> one
> > place or I would just have cited a reference. The parts of the
> observation
> > method that are most relevant in any case - eg. Observation methods,
> sample
> > times and number and location of observations etc. are detailed in the
> main
> > texts (eg. P.347 of Ortuzar et al. Travel Behaviour Research: Updating
> the
> > state of play, Pergamon, 1998.), but the relationship between methods
> and
> > possible errors and the processes one puts in place to control these
> errors
> > tend to only exist in the protocols. One of the problems with
> publications
> > in our field is that unlike biology they lack extensive 'methods'
> sections.
> > There tends to be little discussion of these issues in any of the
> > traffic/movement literature. This may be an area where we know enough
> about
> > sampling of pedestrian movement and the errors involved in the process
> to
> > make a paper of its own, although I suspect that most journal editors in
> the
> > field would find the issue a bit dry and uninteresting. What do people
> > think?
> >
> > Alan
> >
> >
> > >
> > > Dear Alan,
> > >
> > > Thanks for the interesting points. To my knowledge this detailed
> > > discussion
> > > remains absent from the refereed literature -am I missing something?
> > >
> > > Rui
> > >
> > > P.S.: So that there are no further misunderstandings, by 'refereed
> > > literature', I mean journals indexed by Thomson ISI (this leaves out
> SS
> > > Symposia). By 'detailed discussion' I mean a full discussion on a
> > > methodology for observation counts, with references to previous work
> (not
> > > just SS).
> > >
> > > > -----Original Message-----
> > > > From: [log in to unmask] [mailto:[log in to unmask]]
> On
> > > > Behalf Of Alan Penn
> > > > Sent: 09 March 2005 16:55
> > > > To: [log in to unmask]
> > > > Subject: Re: [SPACESYNTAX] Publically available standard test cases
> > > >
> > > > Just to be clear on methodology, Sheep's 'lunchtime observation' at
> the
> > > last
> > > > symposium was fun, but a single 5 minute observation does not
> provide
> > > good
> > > > data on urban pedestrian flow rates, even if all the observers can
> be
> > > relied
> > > > on to observe for 5 minutes rather than some for 5.3 minutes.
> > > >
> > > > In the London studies that Bill referred to earlier in this thread
> the
> > > > procedure was to observe each segment of a whole area for 5 minutes
> > > within
> > > > each 2 hour period of the day between 8am and 6pm and re-observe the
> > > > following day. A total of 10 x 5 minute observations were carried
> out
> of
> > > > each space. Within any single time period (say 10am-midday) there
> were
> a
> > > > total of 2 x 5 minutes of observations on different days.
> > > >
> > > > Rui points to the issue of experimental error on the part of the
> > > observer
> > > > and the timing of an observation. I believe that with training and a
> > > watch
> > > > that in general observers do much better than 10 seconds at the
> start
> > > and
> > > > the end of the period, however errors will still be made. The result
> of
> > > > repeat observations however should be to reduce these errors as
> > > sometimes
> > > > they may be a second or two short and others may be a second or two
> > > long.
> > > > Overall, repeating reduces this type of timing error.
> > > >
> > > > Noah points to the issue of whether a 5 minute sample is adequate
> for
> > > low
> > > > levels of movement. The point here is that our total observation
> times
> > > are
> > > > 50 minutes (for all day averages) rather than 5 minutes. There are
> > > spaces
> > > in
> > > > observation studies that do record 0 passers-by in 50 minutes. Quite
> > > > honestly, by the time you have stood in one of these spaces for that
> > > long
> > > > you are pretty sure that 0 is a good estimate of its level of use.
> > > >
> > > > Clearly, the time period by time period data are based on only 20%
> of
> > > the
> > > > full sample, however tests of the errors involved in sampling - ie.
> how
> > > good
> > > > is any single time period as an estimate of the all day average -
> showed
> > > > that the 10 minutes observations were within 95% confidence limits.
> In
> > > > general, where one is particularly interested in flows in very low
> use
> > > areas
> > > > (housing estates for example) one increases the total length of
> > > > observations.
> > > >
> > > > There are other forms of experimental error that crop up from time
> to
> > > time.
> > > > In very high use spaces - Oxford Street at Christmas for example -
> it
> > > can
> > > > get very hard to count the number of people present - there are just
> too
> > > > many. The standard approach here is to first split the task, for
> example
> > > by
> > > > counting only one pavement at a time, and then by counting only one
> > > > direction of movement at a time. The positive side of over-crowding
> is
> > > that
> > > > people move more slowly, and that a shorter observation period (say
> 2
> > > > minutes for each pavement/direction) will be adequate to give a good
> > > > estimate. The point here is that counting errors will still be made,
> > > but,
> > > as
> > > > previously, mis-counting by a person or two is as likely to be
> reversed
> > > at
> > > > the next repeat observation. In this situation though, as a
> proportion
> > > of
> > > > the total flows involved, the errors will be very small, and the
> higher
> > > the
> > > > flow the smaller the proportional error resulting from miscounts.
> > > >
> > > > The main risk involved in this kind of sampling is that there might
> be
> > > some
> > > > kind of systematic bias - the observer who forgets his watch but
> doesn't
> > > > tell you for example. Since most possible systematic biases are
> related
> > > to
> > > > the individual observer the process we use involves training and
> quality
> > > > checking. We train and instruct observers to ensure that they know
> what
> > > to
> > > > do and that they all apply consistent criteria. Then we ensure that
> the
> > > same
> > > > spaces are observed by at least two different observers and we check
> > > their
> > > > results against each other. The presence of the check, and the
> knowledge
> > > of
> > > > its existence on the part of the observers acts as a peer pressure
> > > measure
> > > > to ensure compliance with the protocols. If you find that observers
> A
> > > and
> > > B
> > > > systematically disagree, you go on to check each against observers C
> and
> > > D.
> > > > It is pretty easy (but a very rare occurrence) to detect a rogue
> > > observer,
> > > > confirm with them what they were doing wrong and if need be discard
> > > their
> > > > data.
> > > >
> > > > Nothing in the field of human research is infallible, but I think
> that
> > > the
> > > > observation data we get this way are pretty robust. One of the
> things
> > > that
> > > > leads me to think this is that where we have re-observed the same
> > > systems
> > > > after a period of years, and where there have been no morphological
> > > changes
> > > > in the interim, the two sets of observations correlate to a very
> high
> > > degree
> > > > (r^2 ~ .9). This suggests that the amount of error or noise
> resulting
> > > from
> > > > the observation methodology is low.
> > > >
> > > > Alan Penn
> > > > Professor of Architectural and Urban Computing
> > > > The Bartlett School of Graduate Studies
> > > > University College London
> > > > Gower Street
> > > > London WC1E 6BT
> > > > +44 (0)20 7679 5919
> > > > [log in to unmask]
> > > > www.vr.ucl.ac.uk
> > > > www.spacesyntax.org
> > > >
> > > >
> > > > Noah wrote:
> > > > >
> > > > > The volume of traffic at each gate matters as well. Small
> variations
> > > in
> > > > > time will produce greater statistical variability if the traffic
> level
> > > > > is low. Higher volume gates have less variability from second to
> > > > > second, even though the total counts will of course be different.
> > > High
> > > > > volume gates have lower statistical error than low volume gates,
> in
> > > > > other words.
> > > > >
> > > > > Best,
> > > > > Noah
> > > > >
> > > > > -----Original Message-----
> > > > > From: Alan Penn
> > > > > Sent: 09 March 2005 12:38
> > > > > To: [log in to unmask]
> > > > > Subject: Re: Publically available standard test cases
> > > > >
> > > > >
> > > > > Absolutely right Rui - all observers should have a watch.
> > > > >
> > > > > Alan
> > > > >
> > > > > Rui said:
> > > > > >
> > > > > > I was one of the people who did these 'observations'.
> > > > > >
> > > > > > We were told nothing on methodology -how do you count people
> over
> 5
> > > > > > minutes?
> > > > > >
> > > > > >
> > > > > > My counts had an error of about 18%, depending on whether you
> count
> > > > > > for 5 minutes or 5 min 20 s (10 secs at beg & 10 sec at end).
> You
> > > can
> > > > > > fit almost anything you like to data with such high errors.
> > > > > >
> > > > > > Rui
> > > > > >
> > > > > > > -----Original Message-----
> > > > > > > From: [log in to unmask]
> > > [mailto:[log in to unmask]]
> > > > >
> > > > > > > On Behalf Of Nick Dalton
> > > > > > > Sent: 09 March 2005 11:04
> > > > > > > To: [log in to unmask]
> > > > > > > Subject: Re: Publically available standard test cases
> > > > > > >
> > > > > > >
> > > > > > > Don't forget the open observation exercise I held at the last
> > > space
> > > > > > > syntax conference.
> > > > > > >
> > > > > > > You can pick up the observation results at.
> > > > > > > http://www.thepurehands.org/massObs/
> > > > > > >
> > > > > > > more standard test sights and more standard observation
> > > information
> > > > > > would
> > > > > > be
> > > > > > > very useful.
> > > > > > >
> > > > > > > sheep
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > >For the Gassin maps on the VR website, I have actually spent
> a
> > > > > > > >couple
> > > > > > of
> > > > > > > >minutes to ask Julienne Hanson (who drew the maps) and CUP
> (who
> > > > > > > >published them) for permission to publish the digitised
> diagrams.
> > > > > > > >Both were happy for the maps to remain up for academic
> purposes.
> > > > > > > >
> > > > > > > >I hope this serves to encourage the use of a standard set of
> test
> > > > > > > >cases for space syntax. However, I would still like to see
> some
> > > > > > > >observation data made available. For example, within the
> > > > > > > >artificial intelligence community, the Anderson (1935) iris
> data
> > > > > > > >has become a standard that is freely distributed to test
> > > > > > > >classification algorithms.
> > > > > > > >
> > > > > > > >Alasdair
> > > > > > > >
> > > > > > > >Rui Carvalho wrote:
> > > > > > > >
> > > > > > > >>On Thu, 3 Mar 2005 11:15:53 +0000, Alasdair Turner
> > > > > > <[log in to unmask]>
> > > > > > > >>wrote:
> > > > > > > >>
> > > > > > > >>>On vector maps: the maps on
> > > > > > > >>>http://www.vr.ucl.ac.uk/research/axial
> > > > > > are
> > > > > > > >>>high definition vectorised versions of the maps in the
> Social
> > > > > > > >>>Logic
> > > > > > of
> > > > > > > >>>Space. For Rui -- I am not sure whether these are supposed
> to
> > > be
> > > > > > open
> > > > > > > >>>access or not given the copyright of Social Logic of Space
> > > rests
> > > > > > > >>>with CUP. (Also, sorry Bin, they don't include fig 25.)
> > > > > > > >>>
> > > > > > > >>
> > > > > > > >>
> > > > > > > >>Well, of course the copyright of those images rests with
> CUP.
> > > > > > > >>Strictly speaking, it seems that the VR centre is violating
> > > CUP's
> > > > > > > >>copyright by not stating that clearly -just joking!
> > > > > > > >>
> > > > > > > >>My point is that it is your job to negotiate with CUP to
> make
> > > > > > > >>these
> > > > > > images
> > > > > > > >>open access. The readers should be spared the red tape.
> > > > > > > >>
> > > > > > > >>Rui
> > > > > > > >>
> > > > > > > >>
> > > > > > > >>>Alasdair
> > > > > > > >>>
> > > > > > > >>
> > > > > > > >>
> > > > > > > >>
> > > > > > > >
> > > > > > > >--
> > > > > > > >Alasdair Turner
> > > > > > > >Lecturer in Architectural Computing
> > > > > > > >Bartlett School of Graduate Studies
> > > > > > > >UCL Gower Street LONDON WC1E 6BT
> > > > > > > >
> > > > > > > >Course Director: MSc Virtual Environments
> > > > > > > >MSc Adaptive Architecture and Computation
|