TRANSPORT

Do thoughts travel? How could they? According to science, a thought is an epiphenomenon. It is experienced as an 'internal' voice or image that has meaning. Scientific experiments show that there is a measurable lag between the (indirectly) observed brain activity that determines actions (behaviour), and the reported experience, including thoughts. While there is still debate amongst scientists and philosophers about the interpretation of the observed lag; on the basis of scientific theory, it should hardly surprise. How could immaterial 'thoughts’ impact the material ‘brain activity’ that supposedly generates them; any more than an image on a screen can control the projector that images it? Take the case of a person (Sean) lying in bed with his partner (Tennille) who has her reading light full on. Sean has the apparent choice to get out of bed, put a pillow over his head, or ask for the light to be dimmed. And when asked, Tennille has the choice of complying or not... or so it seems. Yet, according to science, all that happens is as a result of neurons stimulated in Sean’s brain, and Tennille’s brain. In Sean’s case, the bright light causes neurones in his visual cortex to send out electro-chemical pulses to other parts of the brain that stimulate the voice box and tongue which generate pressure waves from his mouth that interact with Tennille’s ear sending electro-chemical impulses to her aural cortex, that generate still more electro-chemical impulses that cause her hand to move and turn down the light. During this sequence of events, science also says that both Sean and Tennille are (somehow) consciously aware of the bright light and that thoughts (somehow) appear. These thoughts seem to weigh alternatives, and decide to ask Tennille to turn down the light. The thoughts give the impression of 'me choosing to ask my wife'; and from her perspective of 'me choosing to turn down the light'. However, based on science, each experience is nothing more than a charade. First, because the underlying actions are already in train before the thoughts are perceived, and secondly, because there is no way for an immaterial thought to impact material neurones. On this view, consciousness is a mere effect; it is not a cause of anything. On this view too, there is no requirement for information to be transmitted. The whole process is a seamless train of fluctuating quantum fields simply obeying fundamental laws... just as the components in a computer simply switch in response to the very same laws without any need for awareness of the 'information' being processed. The next problem is how could information travel between brains. Any compression wave is just molecules moving in the air. Light is just electro-magnetic radiation. Of themselves, these phenomena are meaningless. Any sound or pattern of colours that arise inside the brain are likewise meaningless... until ideas (number, form and meaning) are associated with them. For example, the colours forming these words are meaningless, as is evident when they are observed from the perspective of a person who cannot read English. They become meaningful only as meaning is associated with them. Which raises the question: if Ideas cannot be transmitted by any sensation, where does meaning come from? Clearly, ideas are formless. They can only be known. Stranger still, there is no ‘separate person’ inside the brain looking at the image; or tasting or smelling or feeling or hearing; or thinking thoughts and recognising ‘things’ (knowing ideas). It is as though the image sees, hears, smells, tastes, feels and knows itself... Not as a ‘flux of energy-matter in the brain’, nor even as a ‘perceptual image’, but as... a ‘separate person, observing separate objects and events’.These objects and events include both: - Apparently external objects (including people); and - Apparently internal bodily sensations, thoughts and emotions, imaginings, dreams and hallucinations. As such they are better called ‘perceptual objects and events’ (to distinguish them from the idealised objects and events described by science). Even though it appears ‘I am a separate person’ looking at ‘separate objects’, it is clear that inside the brain: Seeing (the Sense) and colour (sensation) must be inseparable, as must Hearing and sound, Tasting and flavour, Smelling and odour, and Feeling and feelings (both bodily and emotional: pain, joy, hot and cold, weight, hunger, texture, wetness, desire, anger, fear, etc.). Knowing and ideas too, are clearly inseparable.On this view, the observer and observed are one... Consciousness is One... appearing to be two: ‘subject’ and ‘object’. Being One, there is no need for information to be communicated... no need for thoughts to travel.

How to Plan a journey 4.Philosophical approaches 4.1Overview Section 2 has described a number of issues with regards to OD matrix estimation from link counts, whilst Section 3 has described a number of practical applications. One of the main aims for providing these descriptions has been to demonstrate that the application can be either relatively simple or extremely complex. To talk about philosophical approaches underlying a simple application might appear unnecessarily academic to all but a few people interested in such theorising. However, it is argued here that, as the application becomes more complex, so it becomes more important for practical reasons to reach a sound understanding of the scientific philosophy underpinning any particular estimation method. The argument can be summarised as follows. As applications become more complex, they tend to include an increasing array of heterogeneous scientific models, observations, and methods for making subjective judgements (albeit typically referred to as professional judgement). Given all these different types of informational input, the method-user is required to decide how s/he should fit them together, and in particular is required to assess the relative importance to attach to each item of information. In simple terms, s/he has to decide (quantitatively) which type of inputis more believable. The conclusion the method-user reaches on this issue will lead to significantly different quantitative results and potentially to differing verdicts on the likely benefits (or otherwise) of particular transport schemes. The complications described above apply to the use of any one particular method. However, there is a further potential source of confusion between methods. This is because methods with different philosophical approaches often resemble each other in that they have mathematically coincidental objective functions. The temptation on the part of the user is to consider that all such methods are 'the same'. If the data input to the method were independent of the philosophy of the method, such an approach would be extremely practical. However, it is contended here that the input data 12 (especially with regard to degrees of belief in data) will vary widely depending upon what philosophical approach underlies the method. Thus, coincidental objective functions will often yield differing results, simply because they are fed with differing data. The purpose of this philosophical overview on methods is to help the method-user reach conclusions on how to deal with such issues. Two significant axes of philosophical approach can be identified towards the methods discussed below in Section 5. These can be expressed in the form of dualities: • Rationalist versus empiricist• Realist versus subjective 4.2 Rationalist versus empiricist 4.2.1Statistics or models? It could be argued that a purely statistical manipulation of observed data is 'model-free' and can be carried out according to well-proven deductive statistical approaches. However, this argument is nearly always mistaken, since there will almost always be an implicit model underlying the statistical manipulation. Consider the OD matrix estimation problems being discussed in this paper. It could be argued that if the 'prior'information on the trip matrix comes from direct observation of interzonal flows, then the combination of such data with observed link flow data in a statistical framework would be 'model-free'. However, most of the issues discussed in Section 2 concerned either explicit or implicit models which needed to be employed in all but the simplest applications. For example, most complex applications explicitly use assignment models whilst spatial and long-term dynamic issues are typically dealt with by using implicit models. Furthermore, if probability distributional assumptions are made about any of the items being observed, such assumptions are (in themselves) models. It is thus assumed for the remainder of this paper that all the methods being considered contain models to some degree. 4.2.2 Definitions A pure rationalist model can be deduced from the basic first principles of a theory or set of theories underpinning an academic discipline. For example, rationalist models can be deduced from economic, psychological, physics or engineering principles. Empiricism draws conclusions based upon observed data. Two (philosophically) separate types of conclusions can be identified here:(i)If data is observed with respect to a phenomenon, statistical conclusions can be made (about the phenomenon) which are restricted to the time and place that the data was collected. This form of empiricism will be termed statistical empiricism for the reminder of this paper. As described in (4.2.1), such an 13 approach will typically contain explicit or implicit models. However, in order to avoid confusion with (ii) below, statistical empiricism is considered to be a method rather than a model.(ii)Attempts can be made to transfer the conclusions arising from the observed data to another time or place. This process involves the construction of a transferability model. If this model is based solely upon the observed data, the model can be termed a pure empiricist model. We thus have two types of models, rationalist models and empiricist models, which are constructed by diametrically opposed methods. However, it is not difficult to see that the pure forms of both rationalist and empiricist models are unattainable in practice. On the one hand, a rationalist model needs data for calibration. On the other hand, the construction of a theory of transferability for an empiricist model must be made according to rationalist principles. Given these qualifications, though, we can usually identify emphases upon either rationalism or empiricism within any particular model. However, the main focus of this paper is upon the combination of, on the one hand, statistical methods for manipulating observed data and, on the other hand, models 'in general'. The issue as to whether these models are rationalist or empiricist in nature is of secondary importance and will not be considered further. 4.2.3 Combining models with observed data All the trip matrix estimation methods to be described in Section 5 can be classified according to how they treat the combination of observed data and model information. In general we can identify two main approaches, which can be termed the rationalist approach (which puts emphasis on the model) and the empiricist approach (which puts emphasis upon the observations). Consider firstly a rationalist approach. An example of such an approach could be to use observed data (link counts) to calibrate a gravity model by finding suitable values for the model parameters. The important in this example is that the fundamental model structure is not changed by the observed data: the gravity model stays as a recognisable gravity model whatever values are put on its parameters. An example that helps demonstrate a limitation of the rationalist approach concerns an application in which an old (out-of-date) matrix is being updated by the use of link count data. Suppose, for the sake of argument, that we have chosen a uniform growth model to make an estimate of the current day matrix. This model has one parameter that needs to be estimated: the value of uniform growth. We can make an estimate of this parameter reasonably straightforwardly by comparing the sum of current day link counts with the sum of (equivalent) link flows obtained by assigning the old matrix to the network. Using this uniform growth factor, we can make our present day estimate of the trip matrix by growthing up all the cells in the old matrix by this factor.However, in doing so, we waste much spatial information that is contained in the link counts. 14 Of course, it can be argued that a uniform growth factor model in such circumstances is 'not a very good model', and it is almost certain that it could be improved. However, the quality of otherwise of a model is irrelevant to the central argument that a rationalist approach can waste observed data. To put this another way, the observed data has 'less importance' than the 'structure of the model' and, as a result, some of the data can be lost by the necessity to conform to this structure. This essential problem will be shared, to some degree, by all rationalist models, however sophisticated they are. On the other hand, the empiricist approach is to give data precedence over any model structure. In a sense, such an approach can be more attractive since it can appear to be 'more concerned with fact than theory'. However, an essential problem with this approach concerns measurement error in the data. Such error can arise from two main sources. Firstly, with respect to the problem being considered in this paper, it can be due to: 'mistakes in counting'; transcription errors; or any other mistakes by the person or machine involved with collecting data. On a more complex level, error can arise because the observed phenomenon has not being properly defined (this problem relates to the issue of 'embedded models' within equations such as (2.1) and (2.2), as discussed in (4.2.1)). Measurement errors can occur with link counts and even more so with direct observation of matrix cells. However, if the transport planner (responsible for making matrix estimates) is not aware of any error, how does s/he react to 'odd' results? The basis of this question presupposes that the transport planner has a theory on how to distinguish between 'odd' and 'not odd' results. This theory can be classified as rationalist. The fact that the 'model' might be personal to the planner, rather than existin an explicit mathematical form, makes no difference to this argument. It followsthat the initial attraction of the empiricist approach is somewhat weakened. 4.2.4 Balanced approaches As said in (4.2.3), all (practical) methods for estimating matrices from link counts can be classified in accordance to the relative weights that they put on the hand on models, and on the other hand on observed data. The generic types of method given in (4.2.3) concerned cases when one element of this combination was supremely dominant. However, a third approach can be identified which we can call the balanced approach, which seeks to combine models and observed data without giving intrinsic preference to either type of information. 15 Admittedly, the concept of deterministic subjectivity has been championed for physical science rather than social science, and it might seem to be an easy target to criticise it for the trip matrix estimation process. On the other hand, it is surprising to see that 'maximum entropy with no prior' (thus implicitly accepting deterministic subjectivity) is still occasionally used in practical applications. 4.3.5 Collective subjectivity The third notion of subjectivity, collective subjectivity, is the only one that makes an explicit recognition of society (or alternatively a specific social group as discussed above). Under this approach, the transport planner uses a formal subjective framework (as with the other subjective approaches) but represents her/his belief in the trip matrix estimation process on the basis of 'what would be expected by the society or social group on whose behalf the planning is being carried out'. No pretence is made here that the use of collective subjectivity is a simple process, especially when the needs of society in general are being considered as opposed to the needs of a specific social group. Certainly the concept of 'what is expected by society' is wide open to many different interpretations in both political and scientific dimensions. However, it is argued here that basic assumptions about politics and science are made anyway in everyday transport planning practice, and it is in society’s interests to make such assumptions as transparent (and hence as accountable) as possible. It follows that the adoption of the notion of collective subjectivity should lead to healthy transport planning practice. On the other hand, the burying of underlying political and scientific issues (in effect the pretence that they are non-existent in transport planning practice) makes transport planning opaque and brings it into disrepute. 4.4 Summary A summary can be made here of various approaches that arise from the discussion in(4.2) and (4.3) above. In (4.2), we identify a classification of three method types: empiricist, rationalist and balanced. In (4.3), we identify another classification into five method types. Interestingly, in the latter classification, the approaches discarded as not being useful for trip matrix estimation (realism, individualistic subjectivity and deterministic subjectivity) are all 'better defined' that the remaining approaches (neo-realism and collective subjectivity) which have a certain fuzziness about them. This, in fact, is in accord with the real-life observation that transport planning is often a fuzzy activity. By combining the two different types of classification, and discarding combinations that are not useful (such as empiricist subjectivity), we can list the following approaches to be used for the review in Section 5:•Rationalist•Empiricist neo-realist•Balanced neo-realist 19•Balanced subjective (where subjectivity is understood as collective subjectivity) 4.4.1 Rationalist The rationalist approach concerns using link counts to calibrate a model, which keeps its structure after the calibration process. The approach could involve the transport planner using either a neo-realist or subjective attitude towards the model concerned. In the neo-realist case, s/he would be 'pretending' that it was true (in a realist sense), whilst in the subjective case s/he would use it because that is what society (or the relevant social group) would expect her/him to do. Whilst the latter position appears more sound, the actual numerical results produced would be the same in both cases. However, the actual interpretation of the results would be different. 4.4.2 Empiricist neo-realist Empiricist neo-realist methods essentially involve statistical manipulation of link count observations and trip matrix observations, using standard statistical sampling methods. As pointed out above (in (4.3.2)), such methods are 'nearer to realism' than balanced or rationalist neo-realist approaches. Whilst they are in a sense more scientifically justifiable, they rely heavily upon the availability of large amounts of observed data. 4.4.3 Balanced neo-realist Balanced neo-realist approaches make a matrix estimation by combining a matrix estimation model with link count observations. Unlike rationalist methods, the prior model structure is liable to disappear from the final estimated matrix, due to the influence of the link count data. Two cases can be identified. If the link count observations are assumed to be completely accurate, the problem is constrained in the sense that the solution must satisfy the constraints imposed by the link counts. Practically, this assumption of complete accuracy only needs to be made relative to the quality of information from the model. However, this type of relativist assumption might be awkward for some method-users (especially if they know for certain that the link counts are inaccurate), since it contains an element of subjectivity which does not fit comfortably within the philosophy behind neo-realist models.On the other hand, if no such assumption of accuracy is made about the link counts , the problem is unconstrained. In such a case, the method-user is required to provide weights to show the relative levels of importance attached to the prior information compared to the information from the link count observations. Weights for the link count information are typically derived from standard statistical sampling methods.However, a major difficulty arises on how to attach weights to the model information. This is a serious difficulty in the everyday practical use of such a method: the user has no explicit guidance on how to obtain such weights and often believes that there exists a 'magic' formula which could solve the problem if only someone would pass it on to them. While waiting for this magic formula they can try pretending that the 20 model information is actually collected from direct observations, thus leading to misleading results, particularly in the sense of exaggerating the importance of the prior matrix. Sooner or later, they will recognise that 'professional judgement' is needed to solve the problem, thus effectively turning the method into a subjective one. A further problem that often arises in practice (and is implicit in some methods given in Section 5) is that there is a confusion as to whether the prior matrix distribution refers to the variation in the trip matrix cells or to uncertainty about the mean trip rate. 4.4.4 Balanced subjective As with the balanced neo-realist methods, we can identify two different types of balanced subjective method, depending upon whether we assume the link counts are completely accurate or not. An important difference between the two approaches (neo-realist and subjective) is that, in the subjective case, it is completely within the philosophy of the method to make a relativist assumption that the link counts are only accurate in comparison with the prior (model) information. In the constrained case, the subjective methods considered in this paper are Maximum Entropy (ME) methods. In such methods, a ma trix is estimated which is as near as possible to a prior matrix (which is created from the professional judgement of the transport planner) and which conforms precisely to the link count observations. Methods have also been developed which extend the basic ME methods to unconstrained problems. In such extensions, the user must attach weights to the model information and to the observed information. However, such a process undermines the conceptual simplicity and coherence of ME which relies upon absolute precedence of observed information over subjective information. This basic philosophical problem translates into a practical problem in that the process is not obvious by which the user decides upon the relative weights to attach to modelled and observed information. Arguably, the extended ME approach is an ad-hoc approach which does not fit into any formally subjective method, and has more in common with the balanced neo-realist approaches described above. However, since this might be disputed and since it is logically sensible to describe such methods after describing the basic ME method, they are included (in Section 5) pragmatically in the sub-section on subjective methods. On the other hand, the Bayesian approach is an internally coherent method for representing the trade-off between personal belief (as expressed by the prior distribution) and observations, in the knowledge that the observations have various uncertainties attached to them. Whilst the decision about the values of the weights to put on modelled information is inevitably still problematic, the process by which this is done is extremely transparent for the method-user. In simple terms, it represents the process experienced in our everyday lives of updating our views on the world in response to new information. Cited from Timms P.M.(2001)

You are missing home when on the road